![[ISPRS home page]](../../../../images/home.gif)
Commission III Symposium General Informations
Photogrammetric Computer VisionISPRS Commission III, Symposium 2002 September 9 - 13, 2002, Graz, Austria |
During the 90s important progess has been made in the area of structure-from-motion. From a series of closely spaced images a 3D model of the observed scene can now be reconstructed, without knowledge about the subsequent camera positions or settings. From nothing but a video, the camera trajectory and scene shape are extracted. Progress has also been important in the area of structured light techniques. Rather than having to use slow and/or bulky laser scanners, compact one-shot systems have been developed. Upon projection of a pattern onto the scene, its 3D shape and texture can be extracted from a single image. This paper presents recent extensions on both strands, that have a common theme: how to cope with large baseline conditions. In the case of shape-from-video we discuss ways to find correspondences and, hence, extract 3D shapes even when the images are taken far apart. In the case of structured light, the problem solved is how to combine partial 3D patches into complete models, without a good initialisation of their relative poses.
Full article in PDFFactorization algorithms for recovering structure and motion from an image stream have many advantages, but traditionally requires a set of well tracked feature points. This limits the usability since, correctly tracked feature points are not available in general. There is thus a need to make factorization algorithms deal successfully with incorrectly tracked feature points.
We propose a new computationally efficient algorithm for applying an arbitrary error function in the factorization scheme, and thereby enable the use of robust statistical techniques and arbitrary noise models for individual feature points. These techniques and models effectively deal with feature point noise as well as feature mismatch and missing features. Furthermore, the algorithm includes a new method for Euclidean reconstruction that experimentally shows a significant improvement in convergence of the factorization algorithms. The proposed algorithm has been implemented in the Christy-Horaud factorization scheme and the results clearly illustrate a considerable increase in error tolerance. Full article in PDFCAMIS is a multi-band airborne remote sensing instrument and is designed to utilize modern solid-state imaging and data acquisition technology. It is composed of four CCD cameras with band pass optical filters to obtain four band images. In this paper, we summarize the geometric calibration procedure and results of the CAMIS sensor. We modified the conventional calibration procedure especially for this sensor to make the process more efficient. A network bundle adjustment program was developed and used to adjust the laboratory measurements and locate the targets. Images of the target field were then taken by each of the four cameras of the CAMIS sensor. Two matching techniques were used to determine and refine the target locations in the image space. We modified the matching algorithm to overcome certain radiometric effects and thereby found the location of the target centers in image space.
A full math model was used to recover the most significant camera parameters. The unified least squares approach was used iteratively to solve this nonlinear overdetermined system. In order to determine the lens distortion behaviour, the radial and decentering components were estimated. Then the radial distortion curve was equalized and the corresponding changes to the sensor parameters were recorded. Finally, we present four sets of adjusted parameters, one per camera. For simplicity, the graphical user interface feature in MATLAB was used to create a small user-friendly window with an executable file to adjust the image measurements for the four images based on their parameters. Full article in PDFThe need for a fast, efficient and low cost algorithm for extracting 3D features in urban areas is increasing. Consequently, research in feature extraction has intensified. In this paper we present a new technique to reconstruct buildings and other 3D features in urban areas using LIDAR data only. We have tried to show that dense LIDAR (Light detection and ranging) data is very suitable for 3D reconstruction of urban features such as buildings. This concept is based on local statistical interpretations of fitting surfaces over small windows of LIDAR derived points. The consistency of the data with surfaces determines how they will be modeled. Initially, the data has been filtered to remove extraneous objects such as trees and undesired small features. Then, boundaries will be extracted for each facet using statistical information from the surface fitting procedure, and using inferences about the dominant direction. Building features extracted from actual dense LIDAR collected over the Purdue campus are presented in the paper.
Full article in PDFBayesian object recognition is applied to the analysis of complex forest object configurations measured in high-density airborne laser scanning (LIDAR) data. With the emergence of high-resolution active remote sensing technologies, highly detailed, spatially explicit forest measurement information can be extracted through the application of statistical object recognition algorithms. A Bayesian approach to object recognition incorporates a probabilistic model of the active sensing process and places a prior probability model on object configurations. LIDAR sensing geometry is explicitly modelled in the domain of scan space, a three- dimensional analogue to two-dimensional image space. Prior models for object configurations take the form of Markov marked point processes, where pair-wise object interactions depend upon object attributes. Inferences are based upon the posterior distribution of the object configuration given the observed LIDAR. Given the complexity of the posterior distribution, inferences are based upon dependent samples generated via Markov chain Monte Carlo simulation. This algorithm was applied to a 0.21 ha area within Capitol State Forest, WA, USA. Algorithm-based estimates are compared to photogrammetric crown measurements and field inventory data.
Full article in PDFIn this paper we present a new method for single image orientation using an orthographic drawing or map of the scene. Environments which are dominated by man made objects, such as industrial facilities or urban scenes, are very rich of vertical and horizontal structures. These scene constraints reflect in symbols in an associated drawing. For example, vertical lines in the scene are usually marked as points in a drawing. The resulting orientation may be used in augmented reality systems or for initiating a subsequent bundle adjustment of all available images.
In this paper we propose to use such scene constraints taken from a drawing to estimate the camera orientation. We use observed vertical lines, horizontal lines, and points to estimate the projection matrix P of the image. We describe the constraints in terms of projective geometry which makes them straightforward and very transparent. In contrast to the work of (Bondyfalat et al., 2001), we give a direct solution for P without using the fundamental matrix between image and map as we do not need parallelity constraints between lines in a vertical plane other than for horizontal lines, nor observed perpendicular lines. We present both a direct solution for P and a statistically optimal, iterative solution, which takes the uncertainties of the contraints and the observations in the image and the drawing into account. It is a simplifying modification of the eigenvalue method of (Matei and Meer, 1997). The method allows to evaluate the results statistically, namely to verify the used projection model and the assumed statistical properties of the measured image and map quantities and to validate the achieved accuracy of the estimated projection matrix P. To demonstrate the feasibility of the approach, we present results of the application of our method to both synthetic data and real scenes in industrial environment. Statistical tests show the performance and prove the rigour of the new method. Full article in PDFObject recognition and extraction have been of considerable research interest in digital photogrammetry for many years. As a result, many conventional tasks have been successfully automated but, despite some advances, the automatic extraction of buildings remains an open research question. Machine learning techniques have received little attention from the photogrammetric community in their search for methods of object extraction. While these techniques cannot provide all the answers, they do offer some potential benefits in the early stages of visual processing. This paper presents the results of an investigation into the use of machine learning in the form of a support vector machine. The images are characterized using wavelet analysis to provide multi-resolution data for the machine learning phase.
Full article in PDFDifferent data sources for the determination of a digital terrain model (DTM) are available. New measurement techniques like laser scanning (LS) and interferometric synthetic aperture radar (InSAR) allow a high degree of automation and dense sensing. They pushed the development of new methods for data modelling. This paper presents robust methods for the automatic determination of a DTM from point cloud data. These robust methods operate on the original data points and allow the elimination of off-terrain points and the modelling of the terrain surface within one process. After a short description of the algorithms the paper focuses on the results of this technique applied to datasets from different sensors (airborne LS, terrestrial LS, satellite LS, tacheometry and photogrammetry).
Full article in PDFThe interpretation of remote sensing data is a complex task because of the high variability of the image material under investigation. At the Institute of Communication Theory and Signal Processing procedures for maximizing the degree of automation of the interpretation of remote sensing data are developed. In this paper the system GEOAIDA is presented, which allows an intelligent, concise and flexible control of a scene interpretation by utilizing a semantic scene description. The system produces a hierarchic, pictorial description of the results as well as the structural context of the identified objects including the associated attributes. The output of GEOAIDA can be used for update of geographic information systems and for map generation.
Full article in PDFLaser scanning is dependent on georeferencing by satellite positioning and inertial navigation to give orientation of each laser shot. Orientation errors are at the same time one of the main contributors to the laser data error budget. Satellite positioning error like atmospheric delay, cycle slips and loss-of-lock together with drifts in accelerometers and gyros in the inertial system results in orientation errors which often are of a systematic nature. Some errors can be corrected for by making overlapping laser strips coincide and by making laser strips coincide with ground truth. In this purpose a laser strip adjustment program, TerraMatch, was developed. This paper presents the mathematical model used, the main features of the program and results from practical tests.
Full article in PDFIn this paper we present a new classification technique for segmenting remotely sensed images, based on cluster analysis and machine learning. Traditional segmentation techniques which use clustering require human interaction to fine-tune the clustering algorithm parameters and select good clusters. Our technique applies inductive learning techniques using C4.5 to learn the parameters and pick good clusters automatically. The techniques are demonstrated on level 1 of RAIL, a hierarchical road recognition system we have developed.
Full article in PDFWithin the last years extensive tests were done to investigate the accuracy performance of integrated GPS/inertial systems for direct georeferencing in airborne photogrammetric environments. Based on commercially available high performance GPS/inertial systems direct georeferencing was shown to be a serious alternative to standard indirect image orientation using classical or GPS-supported aerial triangulation. Nonetheless, correct overall system calibration including the GPS/inertial component as well as the imaging sensor itself is the limiting factor in this approach. Since direct georeferencing without ground control relies on an extrapolation process only, remaining errors in the system calibration will significantly decrease the quality of object point determination. Therefore, special focus has to be laid on the overall system calibration procedure. Within this context the stability of system calibration over longer time periods and the influence of additional self-calibration on the calibration parameter estimation are of special interest. The investigations presented in this paper are based on test material from a real flight test, where as one part of a big project a calibration field was flown several times within a two month period using the same GPS/inertial-camera system installation. From this test data first statements on the long term stability of system calibration are feasible, which are important especially from a practical point of view when applying direct georeferencing in photogrammetric production processes.
Full article in PDFThe new Dutch national digital elevation model (DEM), acquired with laser altimetry, is almost complete. The Dutch Survey Department fulfils an intermediary function between laser data suppliers and final users of the DEM. One of the most important tasks of the Survey Department is to guarantee the quality of the delivered laser DEMs to the users. For this purpose, a new height error description scheme is developed, allowing to quantify error effects at different scales. Such a new error description scheme was necessary because the former quality description of DEMs was insufficient. The error behaviour of laser altimetry data, acquired by a complex system of different sensors, cannot be expressed by solely two parameters: a bias and a standard deviation. The former and the new error description scheme will be addressed together with methods to quantify the differently scaled error components. Among them are cross correlation techniques, empirical covariance function analysis from geostatistics and 1d strip adjustment. The contractual demands for maximal allowed error amplitudes, derived from real data, will be presented. The benefit of the error description scheme for DEM users will be illustrated by propagating the error components to the height precision of derived products. Summarizing, it can be stated that this paper deals with assessing the height precision of laser altimetry DEMs and quantifying the effects of the different error components on the measured heights. This paper does however not aspire to give methods for eliminating or minimizing these errors.
Full article in PDFBuilding extraction in urban areas is one of the difficult problems in image understanding and photogrammetry. Building delineations are needed in cartographic analysis, urban area planning, and visualization. Although one pair of images is adequate to find the 3D position of two visibly corresponding image features it is not sufficient to extract the entire building due to hidden features that are not projected into the image pair. This paper presents a new technique to detect and delineate buildings with complex rooftops by extracting roof polygons and matching them using multiple images.
The algorithm discussed in this paper starts by segmenting the images into regions. Regions are then classified into roof regions and non-roof regions using a two-layered Neural Network. A rule-based system is then used to convert the roof boundaries to polygons. Polygon correspondence is established geometrically, all possible polygon correspondent sets are considered and the optimal set is selected. Polygon vertices are then refined using the known geometric properties of urban buildings to generate the building wireframes. The algorithm is tested on a number of buildings and the results are evaluated. The RMS error for the extracted building vertices is 0.25m using 1:4000 scale aerial photographs. The results show the completeness and accuracy that this method can provide for extracting complex urban buildings. Full article in PDFThe use of non-metric cameras in photogrammetric applications is considered under very strict constraints due to their instability and lack of fiducial coordinate system. Architectural building documentation, monuments registration, and monitoring structure deformations are very essential close-range photogrammetric applications that require high accuracy and quick data acquisition. Using metric cameras, in such situations, is quite uneconomic and non-metric cameras are in favor. In order to accelerate the processing time of analyzing non-metric cameras digital techniques are preferred.
The aim of this research is to investigate the use of two inexpensive techniques for object reconstruction using digital images produced by non-metric cameras. The first technique employs an inexpensive 35mm camera and a cheap scanner, while a low-cost digital camera is used in the second technique. Both techniques are thoroughly evaluated and the RMS errors are investigated. Results show that the 6-paramter transformation model is the best model to handle geometric errors introduced by scanners. The object reconstruction process results show that sub millimeter accuracy, in object coordinates, can be achieved if systematic errors are considered. Full article in PDFAccurate 3D surface models in urban areas are essential for a variety of applications, such as visualization, GIS, and mobile communications. Since manual surface reconstruction is very costly and time consuming, the development of automated algorithms is of great importance. On the other hand LIDAR data is a relatively new technology for obtaining Digital Surface Models (DSM) of the earth´s surface. It is a fast method for sampling the earth´s surface with a high density and high point accuracy. In this paper a new approach for building extraction from LIDAR data is presented. The approach utilizes the geometric properties of urban buildings for the reconstruction of the building wire-frames from the LIDAR data. We start by finding the candidate building points that are used to populate a plane parameter space. After filling the plane parameter space, we find the planes that can represent the building roof surfaces. Roof regions are then extracted and the plane parameters are refined using a robust estimation technique and the geometric constraint between adjacent roof facets. The region boundaries are extracted and used to form the building wireframes. The algorithm is tested on two buildings from a locally acquired LIDAR data sets. The test results show some success in extracting urban area buildings.
Full article in PDFThis paper deals with the automatic reconstruction of 3D object lines from airborne three-line scanner data. These novel sensor systems contain three linear CCD arrays which are oriented perpendicular to the flight direction of the aircraft. In a first processing step, raw image data is projected onto a plane at mean terrain height, resulting in three socalled rectified images with different viewing angles. Although these images are visually equal to frame images, after rectification they still contain small projection errors due to aircraft movements in combination with the line perspective geometry of the sensor. We present a new two-step feature-based approach to extract 3D information from three-line scanner data. In the first step 2D extracted lines are matched in three rectified images. Due to errors in the rectified images and in the 2D line extraction, the matching algorithm only provides approximate 3D information. After the approximate alignment of the object line by the matching algorithm, more precise estimates of pose parameters can be obtained by a least squares fitting algorithm in the second step. In this step the geometry of the recording situation is used in order to remove projection errors. Results of the two-step approach are shown for a complex building model, proving the high potential of airborne digital line scanners for 3D feature extraction purposes.
Full article in PDFVarious filtering techniques exist to obtain the ground from laser radar data to use when building digital terrain models. This paper develops an active shape model approach to estimate the ground surface from laser radar data. The active shape model acts like a rubber cloth with elasticity and rigidity. With constraint forces the model is formed to an estimate of the ground surface. The model is glued against the measured points from underneath, forming the envelope of the point cloud. Even in a thick forest as much as 25 per cent of the data points represent the ground. The stiffness of the shape model stretches it out to a continuous surface in between the ground points. The algorithm implemented in this paper is suited to use on very dense data sets, it has been designed for data sets of more than 10 points per square meter. We propose suggestions of changes to the algorithm to adjust it to work on more sparse data sets.
Full article in PDFThis paper presents an algorithm for the extraction of surface clusters from airborne laser data. Surface structure analysis is fundamental to almost any application involving LIDAR data, yet most algorithms focus only on identifying planar segments. The proposed algorithm is more general insofar as it aims at extracting surface segments that exhibit an homogeneous behavior, without restriction to one specific pattern. The algorithm adopts a data clustering methodology for this purpose, which offers a very general and flexible way to identify homogeneous patterns in the data.
Full article in PDFThe extraction of spatial data from satellite imagery requires that precise sensor models are used to orientate images. This is particularly important with images acquired by linear sensors, such as SPOT, frequently pointed with large incidence angles. Image orientation with pixel accuracy requires that accurate ground control points are used. The acquisition of ground control is especially difficult in remote areas, where satellite images are important data sources for the production and updating of topographic maps. This paper describes a method for the orientation of SPOT panchromatic images making use of tie points with a SAR image. SAR and SPOT images compose stereo-pairs with a good stereo intersection, from which heights can be derived. For the SAR-SPOT tiepoints, approximate heights are derived from parallaxes and can be corrected using altimetric ground control points. The method described makes use of the fact that the SAR image orientation can be derived from the orbit and SAR processing parameters, provided with SAR images, and does not require ground control points.
The paper reports on a study carried out with SPOT and Radarsat images from Portugal. Accuracy assessments were done with digital cartography and field surveyed GPS data. It was possible to conclude that using a SAR image, accurate SPOT image orientation can be achieved requiring only very few altimetric control points.
Full article in PDFThis paper presents a matching procedure for automatic DSM generation from the Three-Line-Scanner (TLS) raw images. It can provide dense, precise and reliable results. The proposed method combines matching procedures based on grid point matching and feature point matching. Modified Multiphoto Geometrically Constrained Matching (MPGC) and Geometrically Constrained Multipoint Matching (GCMM) are used to refine the relaxation matching results in order to achieving sub-pixel accuracy on the grid DSM. We match three TLS images and provide the pixel and object coordinates for grid points simultaneously. In order to compensate the disadvantages of terrain modeling by grid points, an additional feature-point matching procedure is performed. The feature points are extracted by using an interest operator such as Moravec´s. Then we activate the modified MPGC, using three TLS images simultaneously, and achieve potentially sub-pixel accuracy. The sensor model used for the geometric constraints derivation is based on the collinearity equations appended by some trajectory models.
The algorithms proposed in this paper have been applied to different areas with varying textures and terrain types. The accuracy test is based on the comparison between well-distributed semi-automatically measured feature points and the automatic extracted DSMs, and on visual inspection of the results.
Full article in PDFWithin the paper the integration of 3D building models in the context of location aware services is presented by two exemplary applications. First the intuitive access to object related information within an urban environment is by telepointing is described. For that purpose terrestrial images are mapped against a 3D model of the environment based on direct georeferencing and a image based localisation of depicted buildings. The oriented image then allows for the access to information related to the depicted objects by pointing to the respective image sections. Secondly, the paper presents an approach for the generalisation of 3D building models with regard to the visualisation of urban landscapes. Especially for applications such as web-based 3D GIS and the presentation of virtual walk-throughs on mobile computing devices, these generated representations of the 3D models are a prerequisite in order to generate level of detail structures for real-time rendering.
Full article in PDFRecent developments of digital cameras in terms of size of Charged Coupled Device (CCD) arrays and reduced costs are leading to their applications to traditional as well as new photogrammetric surveying and mapping functions. Digital cameras, intended to replace conventional film based mapping cameras, are becoming available along with many smaller format digital cameras capable of precise measurement applications. All such cameras will require careful assessment to determine their metric characteristics, which are essential to carry out photogrammetric activities. This paper introduces a new approach for incorporating straight lines in bundle adjustment for calibrating off-the-shelf digital cameras. The optimal configuration for successfully deriving the distortion parameters is considered when establishing the required test field. The developed calibration procedure can be used as an effective tool for investigating the most appropriate model that compensates for various distortions associated with the camera being calibrated. Performed experiments to compare line-based with traditional point-based self-calibration methods proved the feasibility of the suggested approach.
Full article in PDFAutomatic Single Photo Resection (SPR) remains to be one of the challenging problems in digital photogrammetry. Visibility and uniqueness of distinct control points in the input imagery limit robust automation of the pose estimation procedure. Recent advances in digital photogrammetry mandate adopting higher-level primitives such as free-form control linear features for replacing traditional control points. Linear features can be automatically extracted from the image space. On the other hand, object space control linear features can be obtained from an existing GIS layer containing 3-D vector data such as road network, or from terrestrial Mobile Mapping Systems (MMS). In this paper, we present a new approach for simultaneously determining the position and attitude of the involved imagery as well as the correspondence between image and object space features. This approach does not necessitate having one to one correspondences between image and object space primitives, which makes it robust against changes and/or discrepancies between them. This characteristic will be helpful in detecting changes between object and image space linear features (e.g. due to temporal effects). The parameter estimation and matching follow an optimal sequential procedure that depends on the magnitude and direction of image space displacements resulting from incremental changes to the Exterior Orientation Parameters (EOP). Experimental results using real data proved the feasibility and robustness of our approach, especially when compared to those obtained through traditional manual procedures. Changes and/or discrepancies between the data sets are detected and highlighted through consistency analysis of the resulting correspondences.
Full article in PDFThe paper describes the mathematical model used for the triangulation of the ADS40. The ADS40 Airborne Digital Sensor is LH Systems´ multi-line scanner (Röser, 2000; Sandau, 2000). It is based on the push broom principle and therefore requires more complex mathematical modelling than the collinearity equations used in classical bundle triangulation. Particularly significant is the use of high precision GPS and Inertial Measuring Unit (IMU) technologies, which constrain the bundle adjustment process.
Full article in PDFThis paper focuses on internal quality measures for automatic road extraction from aerial images taken over urban areas. The motivation of this work is twofold: Firstly, any automatic system should provide the user with a small number of values indicating the reliability of the obtained results. This is often referred to as ″self-diagnosis″ and is in particular a crucial part of automatic image understanding systems. Secondly, and more important in the scope of our research, a system designed for the extraction of man-made objects in complex environments (like roads in urban areas) inherently implies many decisions during the extraction process. Such decisions are highly facilitated when both low level features and high level objects are attached with confidence values indicating their relevance for further processing. The basic idea for defining evaluation criteria from which the confidence values can be calculated is to split the components of a semantic object model into two different types. The model components of the first type are used for extracting features, i.e., parts of the object, and the components of the other type serve as criteria for evaluating the quality of the extracted features. For guaranteeing an unbiased evaluation one has to ensure that model components belonging to different types are independent from each other (at least theoretically). We illustrate this concept by our system for road extraction in urban areas. Examples are given for both low level features like lines and ribbons as well as higher level features like lanes and road segments.
Full article in PDFLaser scanner data is being increasingly used to obtain topographical and object height information for mapping and GIS tasks. Valuable information can be derived of the terrain and objects of a region. Different methods have been published to segment laser scanner data in order to extract information. This paper aims to discuss the extraction of buildings to transfer 2D building data into 3D building data of a higher accuracy. For this study, only laser scanner data and scanned topographical maps are to be used. The study can be divided into two parts: segmentation and detection. Firstly, the region-based segmentation method is used to delimit objects. Then a scanned topographic map, filtered and converted into a vector format, is utilised to locate already recorded houses. Buildings not contained in the map could be found by combination of segment attributes. The study shows promising results - more than 95% of the buildings were detected.
Full article in PDFAn approach for parallel-sequential binocular stereo matching is presented. It is based on discrete dynamical models which can be implemented in neural multi-layer networks. It is based on the idea that some features (edges) in the left image exert forces on similar features in the right image in order to attract them. Each feature point (i,j) of the right image is described by a coordinate x(i,j). The coordinates obey a system of time discrete Newtonian equations of motion, which allow the recursive updating of the coordinates until they match the corresponding points in the left image. That model is very flexible. It allows shift, expansion and compression of image regions of the right image, and it takes into account occlusion to a certain amount. To obtain good results a robust and efficient edge detection filter is necessary. It relies on a non-linear averaging algorithm which also can be implemented using discrete dynamical models. Both networks use processing elements (neurons) of different kind, i.e. the processing function is not given a priori but derived from the models. This is justified by the fact that in the visual system of mammals (humans) a variety of different neurons adapted to specific tasks exist. A few examples show that the problem of edge preserving smoothing can be solved with a quality which is sufficient for many applications (various images not shown here have been processed with good success). A certain success was also achieved in the main problem of stereo matching but further improvements are necessary.
Full article in PDFThis paper presents a fully automated approach for area detection based on multi-spectral images and features from a topographic database. The vectors residing in the database are refined using active contours (snakes) according to updated information provided by the multi-spectral images. The conventional methods of defining the external energy of the snake based on statistical measures or gradient-based boundary finding are often corrupted by poor image quality. Here a method to integrate the two approaches is proposed using an estimation of the maximum a posteriori (MAP) segmentation in an effort to form a unified approach that is robust to noise and poor edges. We further propose to improve the accuracy of the resulting boundary location an update of the snake topology. A number of experiments are performed on both synthetic and LANDSAT 7 images to evaluate the approach.
Full article in PDFIn this paper we report on a new system to enhance a 3D block model of a real city with terrestrial measured data of the facades. The terrestrial data consists of photographs taken by a hand-held digital consumer camera using short baselines. The relative orientation of the photographs is calculated automatically and fitted towards the 3D block model with minimized human input using vanishing points. The introduced city documentation system delivers a fully 3D geographic information data set and is called MetropoGIS.
Full article in PDFPerceptual organization is proposed as a promising intermediate process toward object recognition and reconstruction from 3D surface points, which can be derived from aerial stereo-images, LIDAR data or InSAR data. Here, perceptual organization is to group sensory primitives originating from the same object and has been emphasized as a robust intermediate-level grouping process toward object recognition in human and computer vision. Despite intensive research on 2D data, perceptual organization of 3D entities is still in its infancy, however. Therefore, the purpose of this research is to develop a robust approach for constructing perceptual organization particularly with irregularly distributed 3D surface points. The scope of perceptual organization presented in this paper is limited to signal, primitive and structural levels. At the signal level, we organize raw 3D points into spatially coherent patches. Then, at the primitive level, we merge the patches into co-parametric surfaces. Finally, at the structural level, we group the surfaces into perceptually meaningful surface clusters. We establish a novel approach and implement the approach as an autonomous system. The system is evaluated with real LIDAR data by inspecting the quality of organized output. The evaluation substantiates a promising performance of the system. The organized output serves as a valuable input to higher order perceptual processes, including the generation and validation of hypotheses in object recognition tasks.
Full article in PDFWe present a new evaluation methodology and a feature extraction scheme for segmentation algorithms in the context of photo-interpretation. The novelty of the proposed methodology is that subjective evaluation marks are involved in the determination of the feature subspace. In fact, our aim is to determine features in alignment with the perception of photo-interpreters, alternatively called psychovisual features. The proposed methodology was applied to the detection of building targets in aerial images. More specifically we considered the delineation of polygonal buildings in semiurban areas on IKONOS images (1 meter resolution). We determined from the images, concurrently, various objective performance measures and collected votes of a jury of evaluators. The methodology to find the concordance between objective features and subjective marks was the canonical analysis of tables.
Full article in PDFThis paper describes a method for blind estimation of sharpness and resolving power from a single image. These measures can be used to characterize images in the context of the performance of image analysis procedures. The method assumes the point spread function (PSF) can be approximated by an anisotropic Gaussian. The width of the PSF is determined by the ratio sigma g / sigma g' of the standard deviations of the intensity and of its derivative at edges. The contrast sensitivity function (CSF) is based on an optimal model for detecting straight edges between homogeneous regions in noisy images. It depends on the signal to noise ratio and is linear in the frequency. The method is applied to artificial and real images proving that it gives valuable results.
Full article in PDFIn this paper we propose a robust hierarchical approach for the estimation of the trifocal tensor. It makes use of pyramids, the sub-pixel Förstner point operator, least squares matching, RANSAC, and the Carlsson-Weinshall duality. We also show how the trifocal tensor can be utilized for an efficient view synthesis which we have optimized by parameterizing it according to the epipolar lines.
Full article in PDFThis paper first introduces a canonical representation for cylinders. The canonical representation introduced here is closely related to the Plücker line respresentation. It then drives the general equation of a cylinder using this representation. In this paper, we also show that this representation is an appropriate one for computer vision applications. In particular, it allows us to easily develop a series of mathematical methods for pose estimation, 3D reconstruction and motion estimation. Finally, one of the main objectives of this paper is to introduce the main equations dominating the three view geometry of cylinders. We show the relationship between cylinders´ three-view geometry and that of lines (Spetsakis and Aloimonos, 1990, Weng et al., 1993) and points (Shashua, 1995) defined by the trilinear tensor (Hartley, 1997), and propose a linear method, which uses the correspondences between six cylinders over three views in order to recover the motion and structure.
Full article in PDFMapvision 4D measurement machine uses video cameras and a light projector to quickly make a precise 3-D point cloud from a given small-scale 3-D object. The use of more than just two cameras makes it possible to control the quality of the measured data, which makes Mapvision 4D more flexible than conventional 3-D measurement machines. The calibration of this kind of photogrammetric machine requires that all possible effects due to the optical, electronic, and mathematical transformations from the 3-D space to the video images have to be taken into account. The calibration is made using a free-network bundle adjustment, which is constrained using known distances and points on precise planes. With the first prototype, an operational measurement accuracy of ±5 μm is achieved.
Full article in PDFThis paper presents a new type of stereo matching algorithm called ″Stereo Plane Matching″. Stereo plane matching adopts least square method under the constraint that all points in an area specified by a polygon should lie on a common plane. This technique utilizes the fact that correspondence between stereo images becomes 2-D projective transformation (homography) within an area where a common plane is projected. Three corresponding point pairs on stereo image pair are used for parameterization of geometry of a plane, which allows computation of homography within the polygon. The most powerful feature of this technique is that it can impose geometrical constraint in planar direction or position. For example, the target plane can be fixed in horizontal or vertical direction, as well as in other specified direction in 3-D space. Another feature is that it can be applied to unrectified stereo pair, so far as its orientation parameters are known. Experimental results show that this algorithm can measure oblique roof or vertical wall of a building.
Full article in PDFThis paper describes a procedure for an automatic multitemporal interpretation of vegetation areas, which uses both structural features and temporal knowledge. For the interpretation of vegetation areas the concept of manual interpretation by using interpretation keys was transformed into the automatic interpretation system. For interpretation of temporal changes an approach was used, which discretely describes temporal conditions of regions, and which transfers the most probable temporal changes of the given conditions as temporal knowledge into a state transition diagram, then using it for multitemporal interpretation. Based on these approaches a procedure for automatic multitemporal interpretation of industrially used moorland was successfully developed. Proceeding from an initial segmentation based on Geo-Data a resegmentation and an interpretation of the segments is carried out for each investigated epoch. By using temporal knowledge it is possible to separate moor classes, which can only be detected in temporal order. The application of temporal knowledge and structural features enables the exclusive use of grey scale images for interpretation of vegetation areas. The results show that the presented procedure is suitable for multitemporal interpretation of moorland, and that it is able to distinguish additional moor classes compared to the approaches used so far. It is further applicable for a more robust multitemporal interpretation, and does not depend on colour images.
Full article in PDFThis paper deals with the simulation of very high spatial resolution images in the thermal infrared range, from 3 to 14 μm. It recalls the physical processes occurring in this spectral range. The complexity of the simulation of such processes is explained. The specifications of the simulator and a methodology for the simulation of infrared imagery are proposed. This methodology enables a very accurate simulation of the signal coming from each object constituting the landscape. The interactions between the radiations and objects and the convective and conductive interactions between objects themselves are considered. Their variations in time, and the recent past of the temperature and the humidity for each object, are taken into account. To reproduce these physical phenomena, the computation is performed on elements, which are defined as homogeneous entities with respect to the physical processes. The architecture of the simulator OSIRIS using the elements is presented. Finally, examples are shown, showing the efficiency of the simulator and the methodology used.
Full article in PDFMulti-line array sensors, carried on airborne or satellite, acquire images with along or across track stereo viewing and are used for photogrammetric mapping at different scales. The main characteristic of the imagery provided by this kind of sensors is that each image line is independently acquired with a different sensor external orientation (position, attitude). If positioning instruments (GPS/INS) carried on board provide the sensor external orientation of each line, the ground coordinates of the observed points can be estimated with direct georeferencing. Anyway the positional and angular displacements of the GPS/INS instruments with respect to the image frame with origin in the sensor perspective centre must be estimated, together with additional measurement errors contained in the observations. Therefore a triangulation integrated with the sensor external orientation modelling (indirect georeferencing) has been implemented. The algorithms have been tested on a simulated testfield, supposing an airborne three-line sensor with optical system consisting of one lens. After simulating the sensor trajectory and the coordinates of 40 object points, the image coordinates of each point in the three images were calculated with back projection. In order to test the indirect georeferencing model, some perturbations and constant offsets in the correct sensor external orientation were introduced and afterwards estimated with the proposed integrated triangulation. The RMS obtained on the checkpoints using different Ground Control Points (GCPs) and Tie Points (TPs) distributions are presented.
Full article in PDFIn this contribution we intend to present a complete system that takes a video sequence of a static scene as input and generates visual 3D model. The system can deal with images acquired by an uncalibrated hand-held camera, with intrinsic camera parameters possibly varying during the acquisition. In a first stage features are extracted and tracked throughout the sequence. Using robust statistics and multiple view relations the 3D structure of the observed features and the camera motion and calibration are computed. In a second stage stereo matching is used to obtain a detailed estimate of the geometry of the observed scene. The presented approach integrates state-of-the-art algorithms developed in computer vision, computer graphics and photogrammetry. The resulting models are suited for both measurement and visualization purposes.
Full article in PDFThe goal of the paper is to describe one method of software resolution enhancement of records obtained by CCD cameras or other devices able to view and digitalize the image. The paper is divided into two parts. The first one explains the theoretical principle of the random shift method. The second part is dedicated to application HimRes which tests these method and creates resulti images with enhanced resolution.
Full article in PDFIn the reconstruction process of geometric objects from several three-dimensional images (clouds of measurement points) it is crucial to align the point sets of the different views, such that errors in the overlapping regions are minimized. We present an iterative algorithm which simultaneously registers all 3D image views. It can also be used for the solution of related positioning problems such as the registration of one or several measurement point clouds of an object to a CAD model of that object. Our method is based on a first order kinematical analysis and on local quadratic approximants of the squared distance function to geometric objects.
Full article in PDFGiven a cloud of measurement points from the surface of a 3D object, we address the problem of recognizing and reconstructing special surface types. We survey our work on this problem, which is based on approximation in the space of lines and in the space of planes. Moreover, we discuss new generalizations which also use a recently developed technique for parametric surface fitting with an active contour model.
Full article in PDFThe topic of this paper is the so-called trifocal tensor (TFT), which describes the relative orientation of three uncalibrated images. The TFT is made up of 27 homogenous elements but only has 18 DOF. Therefore, its elements have to fulfil 8 constraints - a new form for these constraints is presented in this paper. Furthermore, a new minimal parameterization for the TFT is presented having exactly 18 DOF and which is generally applicable for any arrangement of the three images - provided not all three projection centers coincide. Constraints and parameterization are found using the so-called correlation slices.
Full article in PDFIf an aerial-triangulation (AT) is performed in the National Projection System (NPS), then the conformal map projection (which is the base for the NPS) effects the results of the AT by raising three problems: P1) the effect of the Earth curvature, P2) the different scales in plane and height, and P3) the variation of the scale in plane across large areas of interest. Problem P1 may be solved by the so-called Earth curvature correction. Whereas P2 and P3 have negligible effects in plane and height (at least for non-mountainous areas) when performing a conventional AT using control- and tie-points (so-called indirect georeferencing), their effect in height is not negligible (even for plane surfaces) when performing direct georeferencing (using GPS/INS). Simple solutions for solving these problems are presented - by considering the varying planar scale with the point heights or the principal distance.
Full article in PDFThe paper considers the problem of object segmentation and shape recognition in discrete noisy data. Two different algorithms combine region growing techniques with principal component analysis. The proposed algorithms are applied to a data set from airborne laser scanners.
Full article in PDFIn this paper, a new method for the automated generation of 3D building models from directly observed point clouds generated by LIDAR sensors is presented. By a hierarchic application of robust interpolation using a skew error distribution function, the LIDAR points being on the terrain are separated from points on buildings and other object classes, and a digital terrain model (DTM) can be computed. Points on buildings have to be separated from other points classified as off-terrain points, which is accomplished by an analysis of the height differences of a digital surface model passing through the original LIDAR points and a digital terrain model. Thus, a building mask is derived, and polyhedral building models are created in these candidate regions in a bottom-up procedure by applying curvature-based segmentation techniques. Intermediate results will be presented for a test site located in the City of Vienna.
Full article in PDFThis paper deals with forest inventory methods based on laser scanning and satellite remote sensing. It will be demonstrated to what extent forest inventories can benefit from the synergistic use of both sensor types. The forest inventory parameters to be assessed are: tree height, timber volume, tree species, tree age, stand boundary, and basal area. The results presented are derived from the ″HIGHSCAN″ project (Assessing forest stand attributes by integrated use of high-resolution satellite imagery and laser scanner) which is coordinated by the Helsinki University of Technology and financed by the EU, DG XII. Developments have been carried out in close co-operation with forest management authorities, in particular with private forest owners. In this paper the results derived from the Austrian test sites will be presented.
Full article in PDFPhotogrammetry is the traditional method of surface reconstruction such as the generation of DTMs. Recently, LIDAR emerged as a new technology for rapidly capturing data on physical surfaces. The high accuracy and automation potential results in a quick delivery of DEMs/DTMs derived from the raw laser data. The two methods deliver complementary surface information. Thus it makes sense to combine data from the two sensors to arrive at a more robust and complete surface reconstruction. This paper describes two aspects of merging aerial imagery and LIDAR data. The establishment of a common reference frame is an absolute prerequisite. We solve this alignment problem by utilizing sensor-invariant features. Such features correspond to the same object space phenomena, for example to breaklines and surface patches. Matched sensor invariant features lend themselves to establishing a common reference frame. Feature-level fusion is performed with sensor specific features that are related to surface characteristics. We show the synergism between these features resulting in a richer and more abstract surface description.
Full article in PDFAccurate, robust, and automatic image registration is a critical task in many typical applications that employ multi-sensor and/or multi-date imagery information. In this paper we present a new approach to automatic image registration, which obviates the need for feature matching and solves for the registration parameters in a Hough-like approach. The basic idea underpinning GIPSC methodology is to pair each data element belonging to two overlapping images, with all other data in each image, through a mathematical transformation. The results of pairing are encoded and exploited in histogram-like arrays as clusters of votes. Geometrically invariant features are adopted in this approach to reduce the computational complexity generated by the high dimensionality of the mathematical transformation. In this way, the problem of image registration is characterized, not by spatial or radiometric properties, but by the mathematical transformation that describes the geometrical relationship between the two images or more. While this approach does not require feature matching, it does permit recovery of matched features (e.g., points) as a useful by-product. The developed methodology incorporates uncertainty modeling using a least squares solution. Successful and promising experimental results of multi-date automatic image registration are reported in this paper.
Full article in PDFAlgorithmic variations to the glacier motion estimation from differential SAR interferometry are discussed in the present paper. Two efficient albeit relatively simple algorithms for modelling glacier dynamics using spaceborne INSAR data have been devised and tested as alternatives to the conventional DINSAR approach. Neither of the algorithms involves the procedure of interferometric phase unwrapping, thus excluding the areal error propagation and improving the modelling accuracy. In general, they remain feasible even under significant phase noise. An original gradient approach (GINSAR) to differential processing of repeat-pass SAR interferograms based on the calculation of interferometric phase gradients, the generation of glacier slope maps and the analysis of differences between multitemporal slope maps provides global and fast solutions to unsupervised glacier change detection and ice motion estimation. A transferential approach is based on the interferometric measurement of the fast-ice translation forced by the glacier flow and provides good reference values on the glacier frontal velocity and velocity gradients for the GINSAR technique. A comparative analysis of the results obtained by different techniques was performed and algorithmic singularities were discussed. The revealed differences of up to 40% between the GINSAR velocities and those surveyed in the field are explained.
Full article in PDFThe filtering of a laser scanner point-cloud to abstract the bald earth has been an ongoing research topic in laser altimetry. To date a number of filters have been devised for extracting DEMs from laser point-clouds. The measure of the performance of these filters is often based on tests against some reference data (rms, ratio of misclassifications vs. correct classifications, etc.,) obtained by photogrammetric measurement or other means. However, measures based on such tests are only global indicators of how the filters may perform. Therefore, when applied to real life applications, based on such measures it is not possible to say with certainty how well a filter has performed. This uncertainty suggests that a method be devised to identify in a point-cloud those regions where a filter may have difficulty classifying points. This done other sources of information can be gathered to clarify the status of points in (difficult) regions. This fits in with the thinking that external sources of data, such as imagery, maps have be used in the filtering of laser scanner point-clouds. However, devising a method as suggested above requires that the reasons for the misclassification of points be first identified. When filtering a point-cloud based on spatial information alone, misclassification arises from three sources, (1) the nature and arrangement of objects and the terrain in a landscape (e.g., terrain, buildings, vegetation, etc.,) (2) the characteristics of the data (resolution, outliers, data gaps, etc.,) and (3) the implementation of filters. In this paper, the first two reasons for misclassification are outlined because they are common to all filtering problems, and an initial attempt at developing a method for identifying regions in a point-cloud where a filter may have trouble in classifying points is described.
Full article in PDFA LIDAR filtering technique is used to differentiate on-terrain points and off-terrain points from a cloud of 3-D point data collected by a LIDAR system. A major issue of concern in this low-level filter is to design a methodology to have a continual adaptation to terrain surface variations. To this end, several essential observations are discussed in this paper: i) the terrain surface can be fragmented into a set of piecewise ″homogeneous″ plane surfaces, in which terrain surface variations are smoothed out, ii) a criterion for differentiating on- and off-terrain point from plane terrain surface can be equivalently applied to these terrain segments assumed as being plane, and iii) an inter- and intra-relationship of on- and off-terrain points can be as verifying the a priori taken assumption of the plane terrain surface. The main strategy implemented in our LIDAR filtering technique is to iteratively generate a number of terrain surface models in order to hypothesize and test a plane terrain surface over a local area. Finally, the most reliable plane terrain surface model is selected as an optimised solution and thus the terrain surface model is refined. To this end, we devise a two-step divide-and-conquer triangulation in terms of downward and upward model refinement; in this framework, a tetrahedron is used in order to hypothesize a plane terrain surface and the Minimum Description Length (MDL) criterion is employed for the selection of an optimized plane terrain surface model. The useful characteristics of this method are discussed with results derived from real LIDAR data.
Full article in PDFAn object recognition system for industrial inspection that recognizes objects under similarity transformations in real time is proposed. It uses novel similarity measures that are inherently robust against occlusion, clutter, and nonlinear illumination changes. They can be extended to be robust to global as well as local contrast reversals. The matching is performed based on the maxima of the similarity measure in the transformation space. For normal applications, subpixel-accurate poses are obtained by extrapolating the maxima of the similarity measure from discrete samples in the transformation space. For applications with very high accuracy requirements, least-squares adjustment is used to further refine the extracted pose.
Full article in PDFIn this paper we describe our recent work on the automatic extraction of trees from high resolution aerial images. In order to be more independent of color information we have investigated textural properties of trees and buildings. The aim is to be able to differentiate between object classes based on textural information. Texture is a characteristic feature of trees, and if color information is not available it is an important cue to differentiate between trees and buildings. The Gabor filter bank of the standardized MPEG-7 Homogeneous Texture Descriptor (HTD) was used for the extraction of the textural properties. The qualification of the HTD for the extraction and classification of trees is evaluated. The evaluation is based on first experimental results, which are presented in the paper.
Full article in PDFThis paper presents a knowledge based approach for 3D reconstruction of buildings from aerial images. The aerial images are combined with information from 2D GIS database and specific knowledge of the buildings. This paper describes the generation of building hypotheses in case of large variations in the terrain height. First the possible locations of a building in the images are determined by using the ground plans of the building defined in the map and lines extracted from images. These possible locations are verified by the 3D building model generation process. The generated building hypotheses are improved by fitting them to the image data. An evaluation function based on information theory principles is used to select the best model.
Experiments are presented that demonstrate the approach by reconstructing 3D building models.
Full article in PDFIn this paper a new GPS/INS/CCD integrated system for precise monitoring of highway center and edge lines is presented. The system has been developed at The Ohio State University (OSU) for the Ohio Department of Transportation (ODOT). The positioning component of the system is based on tightly integrated GPS/INS (dual frequency GPS receiver and a high-accuracy strapdown INS), and the imaging component comprises a fast, color digital camera from Pulnix (TMC-6700, based on 644 by 482 CCD, with the acquisition rate up to 30 Hz), installed in a down-looking position. The high image rate provides sufficient overlap of the subsequent images at reduced highway speed. The stereo image data processing is supported in near real-time by on-the-fly navigation solution. In this paper, we discuss the design, algorithmic solution and operational aspects, as well as the calibration and performance analysis of the developed system. Feasibility of the application of real-time navigation data to on-the-fly image processing is also presented. In particular, a performance analysis of the integrated system, based on reference ground control, is discussed.
Full article in PDFWe propose an empirical performance evaluation of five different 2D object recognition techniques. For this purpose, two novel recognition methods that we have developed with the aim to fulfill increasing industrial demands are compared to the normalized cross correlation and the Hausdorff distance as two standard similarity measures in industrial applications, as well as to PatMax R - an object recognition tool developed by Cognex. Additionally, a new method for refining the object´s pose based on a least-squares adjustment is included in our analysis. After a description of the respective methods, several criteria that allow an objective evaluation of object recognition approaches are introduced. Experiments on real images are used to apply the proposed criteria. The experimental set-up used for the evaluation measurements is explained in detail. The results are illustrated and analyzed extensively.
Full article in PDFOffsets between overlapping strips of laser altimetry data serve as the input for strip adjustment procedures that estimate and eliminate systematic errors in laser altimetry datasets. For a three-dimensional strip adjustment offsets are to be measured in three dimensions. Height offsets can be determined straightforward by comparing the heights of horizontal planes. Planimetric offsets are more difficult to determine. This paper shows that the usage of standard least squares matching algorithms on height data as well as on reflectance data may lead to significant biases in the estimation of planimetric offsets. For height data, a model based estimation of linear features is proposed since the number of locations in strip overlaps that are suitable for the estimation of offsets in three dimensions may not be sufficient to estimate all error parameters of a strip adjustment. To improve both the offset estimation and the offset variance estimation using reflectance data an edge response function is introduced. This function takes into account the difference in size of a laser beam's footprint and the distance between successive laser points.
Full article in PDFWe present a set of algorithms that recovers detailed building surface structures from multiple images taken under normal urban conditions, where severe occlusions and lighting variations occur and are difficult to be modeled effectively. An iterative weighted-average algorithm is designed to recover high-quality consensus texture of the wall facades. A generic algorithm is developed to extract the 2D microstructures. Depth is estimated and combined with 2D structures to obtain 3D structures, facilitating urban site model refinement and visualization.
Full article in PDFWe introduce herein the concept of the ill-configured object (ICO). An ICO is a geometrical object having a stable (unique) name but varying configurations (shape, size, components, and component layout). In addition, we introduce the concept of the neighbour set representation (NSR) of an object, and show that the NSR is well-fitted to the ICO. Moreover, we show that any object, either non- ICO or ICO, can be characterized as a solution of a set theoretic equation defined on its NSR. An algorithm is thus designed to detect ICOs in images. Two applications of the proposed theory are then presented. The first is ICO recognition in aerial images, and the second is automatic matching of highly deviated landmark-less images. The latter provides a foundation for automatic land cover change analysis using satellite/aerial images obtained under different conditions (time, height, and direction).
Full article in PDFOne of the main obstacles in image registration is the precise estimation of a mapping function that determines geometric transformation between two image coordinate systems. For conventional image registration methods, their registration results are not the global optimal, and accuracy is low because only a few local control points are used for the estimation. In this paper, we develop a global optimal method in order to get a registration approach with high accuracy. In our method, an energy function that is directly related to the parameters of the mapping function is defined in the whole image. Thus, estimation of the global optimal mapping function can be solved through energy optimization. In defining the energy function, we choose a strength measure that is based on contour edge points. It is demonstrated that the strength measure is insensitive to image radiometric distortion. Therefore, our method is applicable for various kinds of images, even for different sensors images. In order to solve the energy optimization, we design a pipelining hybrid framework that combines genetic algorithms (GAs) and a simplex method (SM). The GAs are applied firstly to look for a few initial guesses from some sub-images, and then the SM is employed to get the optima of the energy function near these initial guesses. It is found that the pipelining hybrid framework is not trapped in a local optimum, and converges fast. Hence, one of the advantages of our algorithm is that it successfully avoids advanced feature extraction and feature matching in the image registration. Its characteristics are of automatic and robust. Experimental results have shown that our method can provide better accuracy than the manual registration.
Full article in PDFThe extraction of road networks from aerial images is one of the current challenges in digital photogrammetry and computer vision. In this paper, we present our system for 3D road network reconstruction from aerial images using knowledge-based image analysis. In contrast to other approaches, the developed system integrates processing of color image data and information from digital spatial databases, extracts and fuses multiple object cues, takes into account context information, employs existing knowledge, rules and models, and treats each road subclass accordingly. The key of the system is the use of knowledge as much as possible to increase success rate and reliability of the results, working in 2D images and 3D object space, and use of 2D and 3D interaction when needed. Another advantage of the developed system is that it can correctly and reliably handle problematic areas caused by shadows and occlusions. This work is part of a project to improve and update the 1:25,000 vector maps of Switzerland. The system has been implemented as a standalone software package. We tested the system on a number of images in different landscapes. In this paper we present the results of our system in recent benchmark tests conducted independently by our project partner in Switzerland, and test results with black and white images in a test site in Belgium.
Full article in PDFIn this research, a method is proposed to create and/or update road maps in urban/suburban area using high-resolution satellite images. ″Road mask″ is defined in this research as a mask of road pixels, which are discriminated from others using a commercial remote sensing software. ″Road seed″ is defined in this research as a directional point, indicating that a road is passing through the point along the direction. Road seeds are extracted from edge pixels. Road line extraction is conducted in a semi-automatic way by fusing both road mask and road seeds. Experiments are conducted using an IKONOS image of nearby KAWAGOE city, Japan, with a ground resolution of 1 meter, and four bands, i.e. red, green, blue, and near infrared. Experimental results show that the method is valid in extracting main roads in high dense building area and all roads in countryside efficiently.
Full article in PDFIn this paper, a method is presented to generate surface model of urban out-door environment using vehicle-borne laser range scanners, which have been developed in Konno et al.[11].A classification is conducted first, where range points are divided into the groups of vertical building surface, road surface, other surface, window, tree and others, unknown objects. Erroneous measurement are corrected, e.g. window data or discarded, e.g. irregular data. Volumetric modelling and marching cube method are exploited in this research to model both surface-structured objects, e.g. building and road surface, and volume-structured objects, e.g. tree. Estimates for signed distance are proposed.Through an experiment, it is demonstrated that urban out-door environment can be reconstructed with high automation and efficiency using our method.
Full article in PDFWe consider the problem of resampling for the data obtained by laser scanning on an irregular grid. The data obtained by airborne laser scanning are scattered, and the resampling on a regular grid is needed for the generation of digital surface models (DSM). Some well-known methods are considered in this article: triangle-based linear interpolation, trianglebased nearest neighbor interpolation and kriging. An energy minimization approach is proposed which allows to avoid the drawbacks relevant to these methods. This approach allows to impose a model of surface corresponding to urban areas. The energy function is adapted to deal with the scattered data, and its parameters are chosen to fit the model of a urban area. A correlation coefficient is used to compare the results. The methods are tested on real data - irregularly spaced 3D points - laser scanner data of Brussels. Visual and numerical experimentation results are presented.
Full article in PDFAn automatic tie-point (TP) detection algorithm for multi-image triangulation and registration is described. The algorithm uses a hierarchical approach that leads from the initial extraction of local image patches to the final TP detection on multiple imagery. The distribution of the TP detection is automatically adjusted to meet the needs of the triangulation or registration. Candidate point features are extracted based on the information of local image space only. A relational-based matching scheme using consistent labelling is designed for matching conjugate point features over multiple imagery. The final TPs are refined using the traditional area-based matching. The algorithm was applied successfully to the in-flight georectification of global imagery from the nine pushbroom cameras of the Multiangle Imaging SpectroRadiometor (MISR) instrument. TP match is accurate at 0.2 pixels and is closely successful on all cameras regardless a diverse range of geometric and radiometric distortions. The software was initially tested with simulated MISR data resampled from LandSat imagery and later applied to the production operation of the MISR in-flight georectification. The software was also applied to airborn MISR imagery. The results indicated the system could be adaptive to and effective on TP matching of space-born as well small scale air-born imagery.
Full article in PDFThis work presents a general and formal solution to the problem of fusion of a multi?spectral (MS) image with a higher?resolution panchromatic (P) observation. The method relies on the generalised Laplacian pyramid, which is an over?sampled structure obtained by subtracting from an image its low?pass version. The goal is to selectively performs spatial?frequencies spectrum substitution from an image to another with the constraint of thoroughly retaining the spectral information of the coarser MS data. To this end, a vector injection model has been defined: at each pixel, the detail vector to be added is always parallel to the MS approximation. Furthermore, its components are scaled by factors measuring the ratio of local gains between the MS and P data. Quantitative results are presented and discussed on simulated SPOT 5 data of an urban area (2.5m P, 10m XS) obtained from the MIVIS airborne imaging spectrometer.
Full article in PDFThe approach of using an airborne laser scanning for DTM generation has made great progress recently. However, it is not always clear the reliability and the quality of the geometry that characterise the generated DTM. Thus, a special test was conducted to assess the quality and the reliability of the generated DTM via an airborne laser scanning system. To perform this test, an airborne laser scanner aligned with an airborne digital camera was mounted on a Helicopter (TopEye AB system). The test was performed in Italy and the flying height was about 200 meter. An overlap percentage of 50% were maintained between two successive images. The digital camera is a Hasselblad camera with 2Kx3K resolution. An INS/GPS system is integrated and used for position and orientation determination.
A series of tests were done to check the DTM quality. The main approach used is based on comparing it to the DTM generated from the digital aerial photogrammetry (automatic and semi-automatic matching approaches). Additional comparison is conducted between the Orthophoto created by stereoscopic model and the one created by the laser data and digital images. The test results will be presented and discussed. Full article in PDFFIFEDOM (Frequent-Image-Frames Enhanced Digital Ortho-rectified Mapping) is an ongoing, multidisciplinary research project aiming to provide low-cost, information-rich digital mapping through the development of an intelligent digital airborne image acquisition strategy, and novel image processing techniques. In particular, FIFEDOM is designed to exploit the high degree of overlapping in digital aerial imagery to generate accurate map products. FIFEDOM utilizes high-overlap digital image acquisition to provide multiple-look-angle reflectance for each pixel in a scene. The FIFEDOM project concentrates on four technology areas: Image Acquisition System and Sensor Calibration; Radiometric Balancing; Digital Surface Model (DSM) generation; and Bidirectional Reflectance Distribution Function (BRDF) model extraction.
This paper emphasizes the FIFEDOM DSM module. It introduces an innovative, breakthrough methodology for automated DSM generation. This is achieved through utilization of multi-view, multi-frame highly-overlapped digital images. This capability significantly improves the result significantly by generating a very dense, high-quality, reliable Digital Surface Model. Full article in PDFThe aim of the investigation is to point out common elements of reasoning in to two domains which appear to belong to very different disciplines: image analysis and philosophy. It will be shown that on the iconic, symbolic and semantic level of analysis similar or even identical tools are applied and equivalent results are elaborated. This circumstance is explained by the fact that in both cases it is the human individual who creates and controls the respective tools. Thus they emerge from the same source which unfortunately can only be classified as ″unreliable″ in the light of rigorous investigation.
Full article in PDFThe paper presents a road junction operator, which was developed for medium resolution black-and-white orthoimages. The operator uses a feed-forward neural network applied for a running window to decide whether it contains a 3- or 4-arm road junction or not. The training set was created by a data analysis based feature selection. The best features took part in the training of 3-layer neural networks. The features are coming from the central kernel of the window (raster data) and from edge detection (vector data). The vectorized edges are only kept for training, if they are going through the central circle, which represents the junction central in a rotation invariant way. The edges fulfilling the circle criterion are applied to derive features, like edge length and direction measures. A set of identically structured networks with varied parameters was generated and trained by an efficient optimization algorithm. The evaluation of the networks was carried out in in-sample tests, where the main traditional methods are compared to the neural solution. The out-of-sample test was performed by real image chips with different rotations. The obtained results demonstrate the principal feasibility of the developed method.
Full article in PDFIn this paper we report on a system to extract 3D information from oriented digital facade images using various feature extraction/matching methods. We emphasize on 2D feature extraction from digital images including contour chains, line segments and vanishing points. The extraction of 3D primitives is based on line matching over multiple oriented views and 3D plane detection by a line grouping process. The MetropoGIS system is designed for the automatic extraction of 3D data from facade images.
Full article in PDFIn this paper, we present a prototype system for semi-automatic extraction of road axes and a study on its efficiency for operational use. The core of our system is a road tracker based on profile matching. This road tracker is enhanced with a graphical user interface that guides the operator through the whole data acquisition process. For typical tasks and problems that occur during the extraction process the system provides road specific options for interaction. I.e., the system monitors the behavior of the tracking module, identifies problems, reports information to the operator, and if necessary asks the operator for his decision or action. In order to evaluate the efficiency of our system we compared it to manual plotting of road axes.
Full article in PDFSeveral methods exist to combine a panchromatic image of high spatial resolution with lower resolution multispectral imagery. Of particular interest are those methods designed to simulate real multispectral images having the spatial resolution of a panchromatic image. To help justify an algorithm over another one, quantitative evaluation of the quality of a fused image is necessary. In most cases, the evaluation is performed with the original high- and low-spatial resolution images degraded to a coarser resolution by pixel-block averaging. The multispectral image of the highest resolution serves as a reference image (real image). Most approaches proposed for quality assessment are based on statistical measures computed over the whole image; typical measures are the correlation coefficient and the root-mean-square deviation. However, these measures make no reference to the spatial domain. In this paper, we suggest measures based on local variance computed over a three-by-three pixel window as complementary measures to evaluate the quality of the fused images. The rationale is that an ideal fused image must replicate the variance of the reference image when estimated locally. To help discriminate between local variance induced by real details as opposed to artefacts, the variance is partitioned into two terms. Each term takes into consideration the expected direction of the added details over the multispectral image oversampled by pixel replication. The method is illustrated with different fusion models applied to an Ikonos image.
Full article in PDFSurveying complex shapes or very large entities by laser scanners often requires the registration of a sufficient number of partial 3-D range images in order to completely reproduce the model of the real object. If redundancy exists among the partial models composing the measured entity, a global adjustment of the model components improves the final accuracy with respect to a simple pairwise registration.
To this regard, a new solution for the optimal least squares registration of range images, based on the Generalised Procrustes Analysis techniques, has been recently developed by the authors and can be found in the literature. The method, using the classical tie point correspondence, has proven to be very efficient since it does not require any prior information of the geometrical relationship existing among the particular reference frames in which the different partial 3-D models are expressed. Considering its computational advantages, it does not involve linearisation of equation systems nor matrix inversions, the only requirement is the singular value decomposition (SVD) of matrices of order 3 x 3. In this paper a significant analytical enhancement of the Procrustean method is presented, to manage the stochastic properties of the tie point coordinates in a more complete and exhaustive way. In the previous formulation the possibility to assign a different isotropic weighting factor to the single tie points, according to their specific accuracy, was considered. With the new proposed method, also the positional components, i.e. each coordinate, can be weighted separately. In this way a complete anisotropic and inhomogeneous factored stochastic model can be introduced in the Procrustes procedure. The generalisation of the stochastic model is recommended for certain practical applications, particularly for joining aerial laser scanners strips produced with low sampling density. In these cases, matching correspondence points of low resolution range images generates poorly accurate tie point coordinate estimation. Indeed, this event introduces an uncertainty in the 3-D position that must be considered anisotropic, i.e. not affecting the three components of the same amount. In fact, considering one tie-point laying on a surface perpendicular to the laser beam, the effective position of the laser footprint on the correspondence element affects the planimetric position more than the related altimetric component. In these situations, the different quality of the tie points position components must be correctly and advantageously preserved, performing the global registration adopting the anisotropic model here presented. A suitable application is discussed in the paper to illustrate the registration problem and the expected advantages of the method proposed. Full article in PDFThe recent introduction of laser scanning devices has led to a set of new surveying products for the field of civil engineering and environmental analysis. The laser scanner is an instrument that permits one to acquire irregular point clouds of land areas, rivers and infrastructures in a fast and cheap way. The use of raw laser scanner data requires orientation and filtering procedures to generate a 3D model of the surveyed object., It is possible to automatically derive a set of geometric information from this 3D model that is useful for a variety of particular applications (engineering and environmental): dense DTM generation, sections and profiles, contour maps, volumes and so on.
This paper describes an original, fully automatic software that has been implemented by the authors and which can supply a set of procedures that makes use of laser scanner data for engineering and environmental applications. Some pratical examples have been shown: the Lys stream in Pont Saint Martin in the Aosta Valley, a stone quarry in Bagnolo Piemonte and a landslide in St. Rhémy en Bosses in the Aosta Valley (Italy).
Full article in PDFAs in the foreseeable future a human will be expected to remain as part of the object extraction system, semi-automated road extraction methods are still important. Following this trend, this paper presents a semi-automated method for extracting road segments from medium- and high-resolution images based on active testing and edge analysis. The method is based on two sequential and independent stages. Firstly, an active testing method is used to extract an approximated road centreline, which is based on a sequential and local exploitation of the image. Secondly, an iterative strategy based on edge analysis and the approximated centreline is used to measure precisely the road centreline. Based on results obtained using a medium-resolution test image, the method seems to be very promising. In general, the method prove to be very accurate whenever the roads are characterised by two well-defined anti-parallel edges and robust even in the presence of larger obstacles, as e.g. trees and shadows.
Full article in PDFPhotogrammetry has been introduced as an alternative to direct measurements to obtain facial distances for a variety of anthropometric applications. We apply photogrammetry to assist in the screening of children for Fetal Alcohol Syndrome (FAS). We have developed a stereo-photogrammetric image acquisition platform to obtain digital photographs of 6 to 7 year-old children, and image processing algorithms to extract the eye features required for FAS screening from these photographs. Images are calibrated using the Direct Linear Transform. Eye features are extracted by examining peak and valley maps to locate the eyes and extract the iris contour, and applying a genetic algorithm to match cubic splines to eye contours.
Full article in PDFIn digital photogrammetry applications such as digital elevation model generation (DEM), demanding highly detailed images, it is often not feasible or sometimes possible to acquire images of such high resolution by just using hardware (high precision optics and charged-coupled devices). Instead, image processing methods may be used to construct a high-resolution image from multiple, degraded, low-resolution images. This paper presents an algorithm which is device independent and can increase the spatial resolution of a sub-sampled, and thus aliased, image sequence. This algorithm is illustrated with applications which show its implementation using harmonic theory. The proposed method is of moderate computational complexity and has proved to be robust under noisy circumstances. In order to validate the algorithm´s effectiveness as a photogrammetric tool, it was used in a series of three-dimensional tests using images of objects of known geometry. Stereoscopy sets of left and right images were taken of these objects, DEMs were created using both the original images and images enhanced by the algorithm, and these indicated its potential use for photogrammetric surface modelling.
Full article in PDFThe availability of Shuttle Radar Topography Mission (SRTM) data opens the possibility to characterize urban areas by means of this sensor all over the world. In particular, the global digital elevation model (DEM) can provide a three-dimensional atlas of urban agglomerates. However, the coarse ground resolution and the problems due to radar imaging of the complex urban environments do not allow a precise characterization of urban features.
This work is aimed to a first assessment of building detection in urban areas using SRTM DEMs. We considered data over Los Angeles and Salt Lake City, and showed that, despite the above mentioned problems, it is still possible to detect tall structures and identify the major buildings in the area. Due to the coarseness of the data, classical algorithms for building detection and recognition are not immediately useful, but a comparison with a ground survey for the Wilshire Corridor in Los Angeles provides a first evaluation of the usefulness of SRTM data for urban applications. Full article in PDFIn this paper, a fusion method is proposed, which provides an integrated map of an urban area starting from co-registered optical (panchromatic and multispectral) and SAR images. Classical Intensity-Hue-Saturation (IHS) transform based methods, or Principal Component Substitution (PCS) approaches do not take into account the contextual spatial information, and do not exploit conveniently the complementary characteristics of passive optical and radar data: they may provide interesting visual representations, but they do not guarantee denite advantages to a subsequent classication phase. The proposed method aims to generate an integrated map which selects specic information from SAR data to be injected into the optical data. This choice is application-dependent: in this paper, information related to buildings from the SAR image has been selected, since SAR point-targets strongly characterize urban areas.
Full article in PDFIn this contribution the way algorithms for object detection in urban areas are integrated into the knowledge-based image interpretation system GeoAIDA is described. Generic scene models are used for object detection in settlement areas, whereas the implementation of the respective algorithms is a collection of stand-alone-operators. With GeoAIDA a system is available which uses these operators in the first phase (model-driven, top-down) in order to generate hypotheses for objects in the scene. In the second phase (data-driven, bottom-up) the hypotheses are further processed using structural knowledge about the scene. Here, detected buildings are grouped using the Relative Neighborhood Graph. An example shows that the combination of low-level image operators and high-level grouping operators leads to enhanced scene analysis results.
Full article in PDFThe urban growth and the environmental relationship are several times joint as to the of the urban soil occupation as to the urban area development. The transformation of the several kinds of soil use and the human being activities related to them happen in all sections of the urban space. Therefore the development of the cities is not always followed by the substructure wanted which is going to roduce during a time environmental problems. The flood occurence is one of the environmental problems which has caused many damages to the inhabiting people from the cities. These problems are due to most of the time to the use and occupation of the soil without anandequated planning, besides the lack of state government management of pluvial water. In the city of Itajubá - MG, the flood has been drastic lately and because of it this study had been done with the objective of analysing the structure and the occupation of the soil integrated to the physical features of this landscape. At this rate, contributing to a more efficient planning, based on the environmental management which provides the citizen rights related to their worth life conditions.
Full article in PDFThis research work is concerned with the implementation of a shape from shading (SFS) technique for automatically generating a digital terrain model (DTM) using a single digitized aerial photograph of a terrain area with low signal (information) content. The mathematical model for the SFS is established based on the fact that the pixel´s gray level variations in image space are proportional to the shading intensity variations of the terrain morphology. The terrain shades in its turn is the function of the illumination intensity and the direction of the incident light with respect to the local surface orientation as well as the incident light direction and the terrain albedo. In this project the Lambertian model is utilized for modeling the terrain reflectivity property. The implemented SFS algorithm is then tested with the simulated and the real data. The simulated data was generated using a predefined bilinear surface and for the real data, an aerial photograph of a smoothed hilly terrain with low information content was chosen. The rmse for the reconstructed surface for the simulated data and the real data were estimated to be equal to ± 3 cm and ± 4 m respectively. This means that the gray shade values are also significantly influenced by the other factors such as the non-uniform terrain albedo, atmosphere, etc., which have not been included into our SFS functional model. It should also be mentioned that our implemented SFS model for the real data was accompanied by a preprocessing stage by which the influential noise was significantly reduced.
Full article in PDFPresently, a data fusion contest is conducted to compare the potential of airborne SAR with optical sensors for mapping applications. The goal of the test is to answer two questions: (1) Can state-of-the-art airborne SAR compete with optical sensors in the mapping domain? (2) What can be gained when SAR and optical images are used in combination, i.e., when methods for information fusion are applied? The test is organized in the framework of the IEEE GRSS data fusion technical committee (DFC), ISPRS working group III/6 ″Multi-Source Vision″, which both have strong relations with scientists, active in research on sensor fusion and automation in mapping, and - as the provider of the main organizational framework - the European Organization for Experimental Photogrammetric Research (OEEPE), which is the European research platform of national mapping agencies and other institutions, regarding technology developments to optimize the use of core data in a geoinformation infrastructure context. In the preparatory phase of the test, which has been started on the occasion of IGARSS 2001, test data has been collected and the scope of object extraction for mapping has been defined. The outcome of this phase, i.e. the test imagery to be used in the contest and its potential for mapping is presented in this paper.
Full article in PDFPreviously, we presented a method for automatic registration of real time multi-view range image obtained under the condition free from any artificial restriction. This paper describes a method for real time and automatic reconstruction of thus registered multi-view range images by means of geodesic dome type data structure. It ensures isotropy of sampling region and so may faithfully represent 3-D shape data. It, however, takes a lot of time to store 3-D shape data into it because of non-systematic data structure and difficulty in defining and searching cells neighbouring to each cell in it. On the other hand, data structure based on spherical coordinate system is simple and enables us to store 3-D shape data in real time because of 2-D array type structure. But it is not isotropic and does not ensure to reconstruct 3-D shape data faithfully. So we try to overcome above-mentioned disadvantages of both data structure by storing 3-D shape data into geodesic dome type data structure through 2-D array type data structure based on spherical coordinate system. The validity of the proposed method was proved by experiments on processing time for reconstruction and accuracy of reconstructed data.
Full article in PDFAn inevitable task before direct georeferencing can be performed is the calibration of the GPS/IMU system and the imaging sensor. The permanent calibration fields for large- and medium-scale (1:3 300 - 1:16 000) system calibration constructed by Finnish Geodetic Institute are described in this article. Theoretical results concerning the effect of block structure on the accuracy and determinability of various parameters indicate that the block structures with cross-strips give the most stable results both with and without ground control points. The preliminary results of an extensive investigation concerning the calibration of two Applanix POS AVTM 510 GPS/IMU systems recently purchased by National Land Survey of Finland are given. After boresight calibration, the RMS values between rotations obtained by aerial triangulation and GPS/IMU integration varied between 0.003 and 0.007 gon.
Full article in PDFFor 3-D building reconstructions of urban areas, we present a fully automatic shape recovery method that uses 3-D points acquired from aerial image sequences. This paper focuses on shape recovery of flat rooftops that are parallel to the ground. We recover each rooftop from a set of 3-D points located at nearly the same height. Such 3-D point sets are made by merging point sets under the MDL (Minimum Description Length) principle, which finds suitably concise point sets for 3-D building models. Often, only parts of rooftop shapes can be recovered because of the 3-D position errors being generated in the points. To refine the recovered shapes, we merge the parts under a heuristic condition in which shapes will have a pair of orthogonally oriented edges. To optimize parameters and estimate the viability of our method, we defined a success rate, called the cover ratio, as the area in which the recovered shape and a correct shape (given as reference data) overlap to the combined area of the recovered and correct shapes. Experimental results showed that our method achieved a cover ratio of 75.25%, and through improved cover ratio we also confirmed effectiveness of shape refinement. We also found that even if only one-ninth of the reference data could be used in the optimization of parameters, the cover ratio was 70.96%. The experimental results we obtained showed that our point-based method was effective in enabling the recovery of buildings in urban areas.
Full article in PDFThis paper describes an algorithm which aims at deriving digital terrain models (DTM) from dense digital elevation models (DEM). This algorithm takes place in a processing chain for building detection and modelling in urban areas from high resolution aerial imagery. DEM either come from stereophotogrammetry, or from direct measures such as airborne laser data. DTM are computed as low order frequency models, which exclude ″high frequencies″ such as buildings and other man-made structures. Above-ground regions are then detected by differences between DEM and DTM. Our processing chain ends with the 3D building modelling.
Full article in PDFThis paper describes a novel technique to extract 3D elementary linear elements from multiple views, often called edgels, that are 3D points with tangent direction. Techniques based on stereopair feature, edgel or contour, matching have been intensively developedin the 80s. The relatively poor quality of images used at the time and the ill-posed stereo context contributed to focus the research around segment matching, which provided fairly good results on objects with rectilinear boundaries. Here, we take advantage of the multiplication and of the high quality of images provided by a digital frame camera to revisit this feature matching technique and to enhance the accuracy, the detection and the robustness of a 3D elementary edgel feature estimator. 3D edgels introduce the lowest implicit modelisation as possible, thus allowing the characterisation and reconstruction of curved and straight linear structures which are widespread in our man-made world. The 3D points with tangent directions generated by our estimator can be directly injected in a surface reconstruction framework based on tensor voting (Tang and Medioni, 1998). In a more practical and photogrammetric context, e.g. HR aerial images of urban landscapes, these low-level features can be used to build 3D lines that can be injected as constraint lines in addition to other features, such as 3D points and 3D segments, in a triangulation process (Paparoditis et al., 2001) to improve the surface reconstruction of manmade objects (buildings, pavements, roads, etc.). These features can also be used to construct higher level features such as planar facets in aerial images of urban areas for building reconstruction.
Full article in PDFFor rescue management after natural desasters in urban areas, information about severe building damages is very important. The building damages could be detected automatically by means of image analysis. As final goal, the information could directly flow to a GIS-based desaster management system.
An object-oriented spectral segmentation (software 'eCogniton'), based on previous explicit knowledge, is used to classify facade elements. This is more robust against shadows and foreground objects than approaches based on primary point and line segmentation. Between the found elements, represented e.g. by their central point or line, geometric relations can be examined. The results can be used for several purposes:
Assuming that an intact facade has some geometric properties like flatness and rect-angularity, one single oblique image can already be rectified, and scattered deviations from the constraints can be determined.
With a couple of images, the elements found in each of them can be matched to give identical points for a general threedimensional reconstruction of the scene. Using such coarse object structures promises to be more robust in finding suitable initial values than using image edges and corners, which may be used to refine the first result.
Furthermore, if applied to endangered areas, some parameters can be derived that are of interest for robustness assessment, like relative window size and distance.
Full article in PDFRock glaciers are striking periglacial landforms, which can be found in many cold mountain areas on earth. Rock glaciers are composed of rocks and ice, and must not be confused with debris-covered glaciers. These ice/rock mixtures are slowly creeping downhill by force of gravity. Due to the physical properties of ice the creep process of active rock glaciers is mostly steady-state with annual creep/flow velocities in the cm to m range. Cumulative deformation causes a lava-like surface topography with furrows and ridges. In 1999 the International Permafrost Association (IPA) and the International Commission on Snow and Ice (ICSI) established a Task Force on 'Rock Glacier Dynamics and Permafrost Creep' in order to better understand the complex dynamics of creeping permafrost and, especially, to study the influence of thermal conditions, e.g. global warming, on rock glaciers. In this paper we introduce our in-house developed software package ADVM (Automatic Displacement Vector Measurement) with which surface flow velocities and surface height change of rock glaciers can be measured semi-automatically by means of digital photogrammetric methods applied to digitized multi-temporal aerial photographs. The ADMV software has been tested within the framework of two case studies which comprised the spatio-temporal analysis of the dynamic behavior of three active rock glaciers in the Austrian Alps, i.e., the Inneres and Aeusseres Hochebenkar rock glaciers in the Oetztal Alps and the Hinteres Langtalkar rock glacier located in the Schober group, Hohe Tauern range. Multi-year aerial photographs taken in the years between 1953 and 1999 were evaluated. Some results of both case studies are presented numerically and graphically.
Full article in PDFThe rational polynomial coefficients (RPC) model is a generalized sensor model that is used as an alternative solution for the physical sensor model for IKONOS of the Space Imaging. As the number of sensors increases along with greater complexity, and the standard sensor model is needed, the applicability of the RPC model is increasing. The RPC model has the advantages in being able to substitute for all sensor models, such as the projective, the linear pushbroom and the SAR.
This report aimed to generate a RPC model from the physical sensor model of the KOMPSAT-1(Korean Multi-Purpose Satellite) and aerial photography. The KOMSAT-1 collects 510 ˜ 730 nm panchromatic imagery with a ground sample distance (GSD) of 6.6 m and a swath width of 17 km by pushbroom scanning. The iterative least square solution was used to estimate the RPC. In addition, data normalization and regularization were applied to improve the accuracy and minimize noise.
This study found that the RPC model is suitable for both KOMSAT-1 and aerial photography.
Full article in PDFUsing a set of colour-infrared aerial photos, we compare a newly developed neural net based clustering method with a method based on the classical ISODATA algorithm. The primary focus is on the detection of buildings and it shows that while the traditional method has an advantage in splitting background from foreground, the neural net based method results in a much more uniform segmentation of the input images.
Full article in PDFWith the growing use of digital map data, the requirements for frequent and efficient revision of digital map databases is equally growing. Revision of map databases can be divided into 3 steps: Change detection, classification of changes and registration/updating of the thematic layers of the database. Often the change detection step is carried out in an entirely manual workflow, which calls for automation since it is tedious, labour intensive, and hence very costly. In this paper, we concentrate on automated change detection for the ″building″ layer in a fully 3D geo-spatial database. We present a method based on the use of unsupervised classification as supervising input to a Mahalanobis classification algorithm. The method is evaluated in test cases based on building registrations from the Danish TOP10DK map database, in combination with a plain RGB colour aerial photo and a colour infra red (CIR) aerial photo covering the same area. The test case presented here is from a suburban residential area. At the present stage, plain RGB aerial photos are not enough to significantly reduce the update task. CIR photos show more promise, as the change algorithm, in general, detects all new buildings, although it still needs refinements to reduce the number of false alarms. Buildings with flat asphalt roofs are an entirely separate problem, since, in the lack of height information, they are extremely difficult to discern from roads.
Full article in PDFThe principal function of any automated stereophotogrammetry algorithm is to match corresponding points in a stereo pair. A lot of investigations in this area have been made but problem of corresponding points in considerable occlusions hadn´t solved yet. As a rule we meet essential occlusions in a city area photos, where usual correlation methods are inapplicable. This work introduces new approach to stereo matching problem solution that autonomously constructs the net of corresponding points. The algorithm is based on comparison of intensity profiles extracted from left and right frames. Each profile is presented by pixel sequence along line between two nodes of the net. We construct function which close to zero when profiles are projections of the same line segment from spatial scene. The technique takes account of intensity errors appeared because of perspective photos digitalizing where applying of merely pixel correlation is incorrect. We suggest iterative optimization algorithm which moves nodes of the net. After stabilization of the algorithm every node gives pair of corresponding points. Presence of relationship between nodes gives more useful information in contrast to disjoint spatial points set obtained by ordinary methods. Considerable occlusions appear usually on projections of spatial object edges. That´s why we modify initial net configuration accordingly to detected edges on images before iterative process has started. Obtained experimental results demonstrate applicability of the method.
Full article in PDFThis paper presents an approach for enabling a mobile system to estimate position and orientation of users walking around a city. In an urban environment, localization methods based on the cellular phone network or GPS are not accurate or not reliable enough. This is the reason why we propose to use information out of a 3D model representing the surroundings where the users are staying. A camera represents the line of sight of the users. Frames out of the video stream can then be used to be compared with views of the 3D model. To do so, image processing techniques are applied to the system to calculate the absolute orientation of the camera which the user is carrying. Reconstructions of ancient buildings can then be presented by their exact position. By using specific displays that can be hold in front of the eyes the users can watch the reality and at the same time observe the replaced buildings. This proposed method will be applied in the GEIST research project. Here, a learning game for students is being developed to transmit historical facts of the history of a town.
Full article in PDFLarge-scale laser projection systems used in industrial environment, e.g. for the projection of CAD drawings onto large work pieces, require external calibration to ensure constant precision and hence quality. Basically, each different projected pattern has to be calibrated at least once, to account for temperature-induced variations in the laser scanning systems, relative displacements due to variable distances and shifts between projector and projection target.
The calibration system consists of 4 to 10 standard machine vision B/W cameras or alternatively less expensive modified USB or 1394 cameras, depending on the specific requirements of the factory. Specific calibration patterns are projected at given positions, acquired by spatially calibrated cameras, image processed and compared with the default coordinates. Evaluation algorithms perform error minimisations between the actual data and the set point data, involving different parameter models accounting for offset, rotation, scaling and perspective distortion. The resulting correction routines are fed back into the laser projection control software, thus facilitating an automated re-calibration and error correction. Difficulties with the nature of the projection surface that result in parts of the patterns to be indistinct or entirely missing are encountered by projecting dedicated, stable calibration patterns and using suitable image processing algorithms, in particular sub-pixel methods, for the laser line detection and subsequent computation of calibration data points. This procedure ascertains, even in difficult environmental conditions, a projection accuracy of 2 x 10-4 relative to the total projection area, both in position and scaling even with simultaneous or sequential use of several projectors. Full article in PDFThe 30 m high cylindrical monument ″Dhamek Stupa″ is located at Sarnath in India. The Stupa was build about 2200 years ago in the district of Uttar Pradesh, where Buddha is said to have preached his First Sermon. On October 11-13th 2002 the famous Buddhist ritual ″Kalachakra″ will take place at Graz. On the occasion of this event the idea of building a virtual 3D-model of the Stupa using modern photogrammetric and computer vision methods was born. While on expedition in India in 1982, R. Kostka took photographs of the Stupa with a non-calibrated Hasselblad camera. Two images have been taken in approximated canonical stereo configuration, a third photo in a convergent direction (from the same position). Besides the nominal focal length (100mm), only the length of the baseline and some approximate dimensions of the Stupa are known. In order to reconstruct the Stupa´s surface the stereo images have been oriented on a digital photogrammetric station (DPS). Point measurements - taken either manually or by digital point transfer - were subsequently used for self-calibration of the camera (focal length only). Once all of the Stupa´s parts had been mod-elled geometrically using different methods (geometric primitives, point and edge measurements), the surface patches were triangu-lated manually or by delaunay triangulation. The non visible parts on the backside have been modelled by cylindrical and conical patches, which are continued symmetrically from the front. The different parts - which had been stored using the GNU Triangulated Surface (GTS) format - have been converted into the VRML 2.0 format and assembled into a single scene for viewing with any stan-dard VRML viewer.
Full article in PDFThis paper reports on the development of an algorithm that can renovate the RPC(Rational Polynomial Coefficients) models using GCPs (Ground Control Points) and Pseudo_GCPs (Pseudo_Ground Control Points) in order to improve the positional accuracy of the 3D ground coordinates extracted from IKONOS-2 satellite images. When 3D ground coordinates are extracted from IKONOS-2 satellite images, their positional accuracy depends on the RPC models. Therefore, in the case where RPC models are renovated by GCPs, the positional accuracy of the extracted ground coordinates can be improved. However, in actual cases, the GCPs are neither distributed evenly on the area of the satellite images nor can be obtained sufficiently. To solve these problems, Pseudo_GCPs, which were calculated from the original RPC models, were used. The RPC models were then renovated using both the GCPs and the Pseudo_GCPs together, and the 3D ground coordinates were extracted using the renovated RPC models. Finally, the validity of the algorithm was demonstrated on the basis of an accuracy evaluation using the check points obtained from the GPS(Global Positioning System) survey and triangulation. In addition, it was demonstrated that the vertical accuracy of the ground coordinates could be improved by using this algorithm.
Full article in PDFPose estimation, which determines the position and the attitude of the camera, is a prerequisite for a variety of photogrammetric tasks such as image matching, surface reconstruction, ortho-photo generation, and object recognition. As digital imagery is becoming more popular in photogrammetric applications, features other than points are increasingly employed in pose estimation. For frame imagery, prior research has incorporated geometric constraints into the bundle adjustment to take advantage of high-level features, such as straight lines. This study introduces high-level feature constraints into the pose estimation problem for aerial line scanners. Traditionally, it has been assumed that robust pose estimation for a dynamic (non-frame) sensor cannot be obtained without direct measurements using Global Positioning System (GPS) and Inertial Navigation system (INS). This study attempts to challenge, in part, this argument by showing that the unknown orientation parameters for all scan lines, within a scene captured by line scanner, can be estimated by including linear features in an indirect orientation procedure that is less dependent on GPS/INS data. This indirect orientation is achieved by incorporating straight-line constraints in the bundle adjustment for aerial linear array scanners. A scene captured by an aerial line scanner is composed of sequence of scan lines, each of which may be slightly shifted against each other due to changes in the system´s trajectory. As a result, straight lines in object space do not appear as straight lines in image space. The straight-line constraint is proposed in this study to cope with this phenomenon. This constraint makes use of straightline features in object space to aid independent recovery of exterior orientation parameters of the entire scan lines as well as to increase geometric strength of the bundle adjustment. Straight lines have been adopted since imagery of manmade environments is rich with straight lines. Moreover, free-form linear features can be represented with sufficient accuracy by a sequence of straight lines (poly-lines). Experiments conducted in this study show the effectiveness of using free-form linear as well as straight-line features in the recovery of the exterior orientation parameters of aerial line scanners.
Full article in PDFThe idea and concept of Seamless Stereo orthoimage Database is firstly put forward around the world. The meanings and basic principle of Seamless Stereo Orthoimage Database are deeply analyzed in this paper. Based on stereovision of human´s eyes, the main idea of Seamless Stereo Orthoimage Database is to generate a seamless Digital Orthophoto Quadrangles (DOQ) database from DEM and odd photos and generate additionally a Digital Stereo Orthophoto Partner (DSP) from DEM and even photos, and then a large area 3D virtual landscape environment can be formed without y-parallax. The database is called Seamless Stereo Orthoimage Database formed by DOQ and DSP together. In such environment, the 3D measurement and analysis can be done under the interface of normal GIS or CAD systems without complex DPW. The 3D objects, such as houses, trees, cloverleaf junctions, geologic ruptures, and so on, which are not acquired during data acquisition at DPW, can be measured by end user himself. The principle, algorithm and excellent measurement accuracy analysis are presented in detail in this Paper.
Full article in PDFThe method of aircraft navigation system correction using position of buildings on a surface is considered. It is supposed that the aircraft flies at low altitude and the surface region is probed by an onboard laser locator under flat angles. The analysis of the selected flat vertical surfaces directions on plane image of considered region and it´s etalon is the base of this method. To do this distance image is transformed from a central projection to the Cartesian coordinate system and then it is reduced to a fixed point. Then the reduced distance image is segmented to select flat vertical surfaces. After selection the plane image of scene with projections of vertical surfaces is created on the basis of the segmented image. The similar plane image is created using digital model of scene. Generally there is no one-to-one correspondence between these two images. To solve this problem Hough transform is made for both plane images. Comparison of straight line images by increments of an inclination angle of their normals relative to an axis x is made. Some measure of distinction is introduced. Finding the minimum value of measure of distinction means finding the best correspondence of straight lines on images. Determining of correspondence between straight lines allows to determine an angle of turn of one coordinate system concerning another. At last the displacement vector of coordinate systems is determined. For this purpose appropriate pairs of straight lines are selected. The displacement vector can be determined by the difference of coordinates of cross points of corresponding lines.
Full article in PDFThis paper proposes a method for outlier detection in recovery of projective shape and motion from multiple images by factorization of a matrix containing the images of all scene points. Compared to previous methods, this method can handle perspective views, occlusions, and outliers in image correspondences jointly. The main novelty of this paper is the method for outlier detection whereas the proper reconstruction was described in (Martinec and Pajdla, 2002). In this work we assume that the amount of inliers is significantly larger than the amount of outliers. The main idea is that minimal configurations of points in triples of images are sufficient to validate inliers reliably. The RANSAC paradigm is used. Trifocal tensors are computed from randomly selected minimal n-tuples of points in triples of images. After the tensor estimation, the number of points consistent with the tensor is counted. If there are sufficiently enough consistent points, those not used to estimate the trif. tensor receive one positive vote. The voting is repeated until points in the measurement matrix are sufficiently sampled. The points that obtain zero or a very small number of votes are rejected as outliers. Inliers are used by the method described in (Martinec and Pajdla, 2002) to obtain a projective reconstruction. The set of inliers can be further enlarged by an iterative process. The new method is demonstrated here by experiments with laboratory and outdoor image sets.
Full article in PDFA new filtering method of airborne laser scanner data has been developed that is suitable for complex terrain, that is flat but with artificial depression such as openings of underground tunnels, typical in highly developed urban areas. The method firstly selects lowest points in small patches from laser scanner data. Then inappropriate points for ground estimation is removed based on statistics within a buffer around the point. The method was tested and gave fairly good result with the laser scanner data of Shinjuku, Tokyo where many skyscrapers and openings to underground tunnel exist.
Full article in PDFAutomatic extraction of built-up areas from a multitemporal interferometric ERS-1/2 Tandem SAR dataset was studied. The image data were segmented into homogeneous regions and the regions were classified using their mean intensity and coherence values and additional contextual information. According to comparison with a set of reference points, an overall classification accuracy of 97% was achieved when classifying the dataset into three classes: built-up area, forest and open area. Reference points in densely built-up urban areas were recognized as built-up with 100% accuracy. In small-house areas the percentage of reference points correctly classified as built-up ranged from 66% to 94%, depending on the channel combination and classification rules used. Use of texture or rules related to classes of neighbouring objects improved the accuracy. The classification process was highly automatic; training areas covered only about 0.12% of the study area and did not have any overlap with the reference points used in accuracy estimation. Built-up areas could be recognized clearly better than in some previous studies with interferometric ERS data. Possibility to classify built-up areas further into subclasses was investigated using digital map data. The results suggest that the built-up classes of Finnish 1:50 000 topographic maps (small-house areas, apartment house areas and industrial areas) are difficult to distinguish reliably from each other. On the other hand, a correlation was found between the percentage of an area covered with buildings and the mean intensity and coherence of the area in the imagery. This information was used to classify built-up areas into subclasses.
Full article in PDFRecent increase in number of pixels of images acquired by a non-metric digital camera encourages an amateur to utilize it for 3D measurement. Most of current camera calibration methods are inconvenient and expensive for an amateur to calibrate his digital camera. Therefore we have developed a method for an amateur to calibrate a non-metric digital camera easily. Our proposed method needs a lot of objects (calibration points) recognizable in an image instead of targets. Eight convergent images are acquired from eight different directions with four different camera frame rotation angles of 0°, +90°, +180° and -90°. Measurement of image coordinates of calibration points are in principle executed by template matching. A set of camera parameters is determined by the bundle adjustment.
A field experiment and a numerical simulation were conducted to investigate performance of our proposed method. The results of the field experiment and the numerical simulation indicate that the proposed method using a set of calibration points distributed on the 2D plane can provide a reliable set of camera parameters. Error of an image distortion model estimated by the proposed method is expected to be up to approximately ±3 pixels or ±12 μm on the image.
Since our proposed method needs neither special equipment nor ground survey of control points, we believe that the proposed method is useful for an amateur to calibrate his digital camera.
Full article in PDFPreviously, we developed a real time 3-D measurement system based on slit-ray projection method, which enables us to obtain serial range images at 1/30 seconds per scene with an error within ±0.3%. It, however, uses laser beam as light source, so it has some disadvantages in safety and applied field of it is somewhat limited. In this regards, the aim of this paper is to develop a new system for 3-D measurement using incoherent light source, which is not dangerous to surroundings and enables us to obtain serial high-resolution range images in real time.
Our new system is based on intensity ratio method. In this method, the target space is encoded by intensity ratio of two light patterns projected alternately onto the target space. In this research, a flat light pattern and a linearly increasing one along horizontal direction are used as two light patterns and projected from a DLP projector on the market. Since one range image is obtainable from two successive video images corresponding to two light patterns and procedure for range image calculation is very simple and carried out at high speed, serial range images are obtained at high speed from serial video images, which are captured by a 3CCD video camera on the market synchronizing with alternate projection of two light patterns. Thus our system realizes to obtain serial range images in real time. In addition, since one range data is obtained from one pixel of the image plane of the video camera, spatial resolution of our system is same to that of the video camera. Moreover, since each component of our system is available on the market, total cost of it is very cheap. To improve accuracy of measurement by our system, optical characteristics of video camera and DLP projector are investigated and taken account into processing video images. Moreover, since accuracy is badly affected by the fact that reflectance on the surface of the target depends on the frequency of the light projected, monochromatic light source is used in our system.
As a result, our trial system enables us to obtain serial range images at 30 scenes per second with spatial resolution of 640 × 480 and an error within 1%. Thus, it may be applicable to the field of robot vision, gesture recognition, motion analyzer, recognition of moving target and so on.
Full article in PDFPhotogrammetric Computer Vision algorithms often deal with complex numerical calculations and complex data structures. For modelling the numerical calculations usually a mathematical model is used. For the data structures an Object-Oriented model is appropriate. But there is still a lack in modelling the combination of both kinds of notation.
In this paper we present a new modelling technique for presenting mathematical and object-oriented aspects in one model. This modelling technique is based on the Unified Modelling Language (UML) and optimally suited for algorithms with complex numerical models and data models. For the semantics of class members in the UML model we use a formal description of constraints, which is similar to the Object Constraint Language (OCL) included in the UML specification. Since OCL is not capable of complex mathematical expressions, we enhance this notation to handle mathematics, and we name it OCL+M. We show, that typical problems in photogrammetry can be modelled by using an object-oriented model with OCL+M.
The advantages of this modelling technique are: 1. A clear presentation of the mathematical model within the object-oriented model, 2. The methods of the classes are easy to implement, 3. The implementation is straightforward and maintainable and 4. The model can be tested with little efforts.
Full article in PDFIn thermal transformation processes for the production of metal parts of complex shape, shaping tools are frequently used. In die forging as an example, the tools are exposed to high mechanical and thermal loads and therefore subject to heavy wear. For fast feedback directly in the production process the accuracy to size of metal parts produced in such a way is at present frequently checked with groping measuring tools. This is unsatisfactory because of the complex forms, the broad product range and the thermal conditions. A 3D - measurement system for highest possible flexibility and industrial environment, conceived for this application and based on a 3D - sensor attached to an industrial robot for positioning is described here. The Structured - Light approach is used, i.e. the sensor consists of one or several CCD - cameras, combined with a programmable line projector. Two different calibration procedures of the line projector system are compared, on the one hand based on a polynomial model with a large number of coefficients, on the other hand based on a model targeted towards the actual optical / physical conditions with a significantly smaller number of coefficients. Finally a procedure for the solution of the Sensor to Hand calibration of the measuring robot is shown.
Full article in PDFThe central perspective model is generally used in photogrammetry because of the high reliability, but it has a weak point that the initial values of orientation parameters are necessary due to the non-linearity. On the other hand, the model based on projective geometry, which is widely used in computer vision, can be treated as linear forms, but many indefinitenesses in the model leads to the lower precision and the lower stability. The authors developed an alternative model called the orthogonal projection model, which is as reliable as the central perspective model and does not need initial values of orientation parameters. The orthogonal projection model is derived from affine projection model with a constraint of orthogonality. The model is therefore appropriate to long distance observation like affine projection model, and it is also applicable to close range photogrammetry with high accuracy. This paper describes the derivation of the orthogonal projection model and the geometric characteristics, and also verifies the effectiveness by various simulations and a field test.
Full article in PDFThis paper presents a basic theory of intelligent photogrammetry and corresponding physically independent agent systems named Photogrammetron. Intelligent photogrammetry is a fusion of ideas, techniques and applications from digital photogrammetry, active vision and intelligent agents, aiming at developing highly automatic and intelligent photogrametric systems. The tenet of intelligent photogrammetry is that photogrammetry is a nonlinear process for which a full automation, if indeed required, is only possible if and only if the system is an autonomous and intelligent agent system. In its basic form, photogrammetron refers to a stereo photogrammetric system which is designed to possess an architecture of intelligent agent, a physical structure of qualitative and active vision, and a complete functionality of quantitative photogrammetry. In a broader sense, photogrammetron may have three physically different forms: I - coherent stereo photogrammetron, II - separated stereo photogramemtron and III - multi-camera network photogrammetron. Principal applications of photogrammetron include: photogrammetry-enabled robots, intelligent closerange photogrammetry, intelligent video surveillance, real-time digital videogrammetry. The agent architecture of photogrammetry consists mainly of five components: imaging perception, scene beliefs and models, missions and desires, intentions and planning, action and control. The photogrammetric visual information processing comprises seven major parts: motion camera calibration and relative orientation, optical flow computation and object tracking, stereo focus of attention and vergence control, stereo motion image sequence matching and correspondence maintenance, object structure from motion, multi-object tracking and recognition. While a complete realization of photogrammetron design will take time to appear, this paper only intends to outline a basic theory and address fundamental theoretical feasibility and challenging problems. A real prototype of Photogrammetron-I is under development.
Full article in PDFDigital cameras have started to replace analog film cameras for many applications. Considering the huge consumer market of small format cameras and compact cameras or the professional medium format or large format photography, digital senors are on the advance.
The comparison between digital and film-based data is done taking into account photographic and photogrammetric issues. Not only radiometric quality, but also the geometric performance of digital and analog cameras are investigated. When trying to compare the quality and the information content of panchromatic digital images versus analog film, only a small amount of literature is available to brush up one´s mind. Encouraged by this situation, we have investigated different film types and a digital sensor and their output in order to better understand how to compare and to judge image data.
Full article in PDFSystem calibration is required in integrated IMU/GPS systems to account for the spatial offset and misalignment between IMU, GPS and camera frames; synchronization is to be maintained, to predict IMU/GPS position and orientation data at the mid-exposure time of the images. To this aim, measurement on the ground, complemented by a calibration flight over a test field, are performed. Depending on the mathematical model, a two steps or a single step procedure may be used to recover the calibration parameters. Within the OEEPE test ″Integrated Sensor Orientation″ the authors proposed a simple but effective two steps procedure, where calibration parameters are computed as a weighted average of the discrepancies between the EO parameters of the images derived from block adjustment and those computed from the IMU/GPS data. Here a new single step calibration procedure is presented, where the calibration parameters are explicitely inserted in the collinearity equations and the IMU/GPS data are considered as pseudo-observed quantities, replacing EO parameters as unknowns in the block adjustment. Applied to the same OEEPE data set , the new procedure yields the same results of the previous one, with less ground control points.
Full article in PDFOur approach to image region approximation offers a complete scheme consisting from few steps. The original image is segmented using an unsupervised colour-texture regions segmentation method. Following polygonal approximation of created regions causes the region boundaries degradation. This approximation has not any important influence to the image quality and enables increasing of code efficiency. Especially for very low bit rates image coding we achieved better values of objective criteria (PSNR). For texture approximation we use the 2D shape independent orthogonal transform (DCT II.). The encoding and decoding of polygons is very efficient, because we need to store only the endpoints of them. The texture is coded with modified code similar to JPEG arithmetic code.
Full article in PDFThe objective of this study was to carry out reconstruction of single tree crowns from laser scanner data to use the obtained vector model for feature extraction. The reconstruction was implemented in several stages. First, pulses which have reflected from each tree were marked off from the original point cloud. Ground points were then separated from all points using digital terrain model and analysing the histogram of terrain height values. In the next stage canopy was described with vector polygons, and the location of the trunk was estimated. With respect to the location of the trunk, tree points were transferred from (xyz)-co-ordinate system to the polarco- ordinate system (a,r,h), and features were estimated from the vector model. Evaluation of the reconstruction was performed choosing a test area and processing 50 single trees, and comparing results to the field measurements. In the study it was found that dense laser scanner data detail describes the upper canopy of forest and therefore is suitable for tree height information extraction. The lower crown was found less detail measured with laser scanner and parameters extracted from that part were less accurate, but trend setting. Obtained distance profile seemed to give tendency for the tree specie.
Full article in PDFFor the work programme BIOTOPMONITORING VIENNA of the Austrian Health Institute, multi-spectral air-borne scanner data have been recorded in the intention of automation supported identification and classification of forests and green urban areas in Vienna. Before multi-spectral and textural image analysis can be performed, the scanner data have to be geometrically rectified (georeferenced) with respect to a map projection system. The geo-referencing of the GPS/IMU-supported scanner data of the surveying flight Vienna 2000 was carried out at the Institute of Photogrammetry and Remote Sensing of the Vienna University of Technology. In the course of the geo-referencing process 'drift-phenomena' within the GPS/IMU-data were detected, which decrease significantly the quality of the geo-referenced scanner data. In this paper principles of geo-referencing of GPS/IMU-supported multi-spectral air-borne scanner images are outlined. Furthermore it is presented, how possibly occurring drift-phenomena within GPS/IMU-data can be corrected or at least dampened by a hybrid adjustment applying an extended mathematical model and by providing additional control- and tie-information.
Full article in PDFThis paper describes the use of computer vision techniques to obtain cartography of materials and damages, in a semi-automatic way, of indoor façades of historical buildings, as a support to planify the restoration works. For this purpose, a cost-effective equipment is designed to acquire digital images from the visible and near infrared regions of the spectrum, and spectral and texture classification techniques are used to convert those images into informational classes. The equipment is composed of a monochromatic CCD camera, sensitive to the visible and NIR, with a peak of sensitivity at about 760 nm. The camera sends the signal to a frame grabber connected to a computer, where the image is processed. After calibration of the camera, 5 images centred at different spectral wavelengths (from 400 to 900 nm.) are acquired using optical filters. The scene is illuminated using diffuse light to avoid shadows on materials. Radiometric adjustment and geometrical registration are made, and then a mosaic of the images is composed using ground control points.
In addition to the spectral images, spatial information is obtained from textural features extracted by mean of the co-occurrence matrix, energy filters and multi-resolution analysis using a wavelet transform. A statistical selection process of the most relevant features is made, and the final classification is evaluated attending to the information provided by archaeologists and architects that carry out the restoration works. The methodology is applied indoors to a monument and the results are discussed, as well as some improvements that should be made in the future to integrate the equipment in order to be operative in the field. Full article in PDFThis paper investigates into the model-based reconstruction of complex polyhedral building roofs. A set of 3D line segments, obtained from multiview correspondence analysis of high resolution colour imagery, is used as input data. The 3D line segments are grouped into planes by means of a Bayesian model selection procedure. In the resulting planes, models for polygonal patches are then instantiated. Driven by the Maximum Expected Utility principle, the algorithm chooses the optimal patch and plane configuration non-deterministically. Roof reconstruction is completed by further reasoning steps which are guided by the semantic interpretation of the intermediate patch configuration. Several successfully reconstructed complex roof structures corroborate the potential of the approach.
Full article in PDFTracking of multiple objects from ′real world′ is one of the most complicated problems in computer vision since the behaviour/motion of these objects is unpredictable and cannot be assumed. Tracking is also ″application depended″ task; there is no one general tracking methodology for solving all tracking problems. Therefore, different algorithms and procedures are needed and used for different applications. However, it is possible to rank tracking applications in two general categories: the ones that need a moving camera and the ones where a stable camera is enough. This paper deals with the second category and the technique of the ″closed-world″ for object tracking. It defines the lower level problem, the possible circumstances of the objects movements and the algorithms to solve some of the tracking situations. The obtained experimental results are pretty promising, approaching 100% hit rate, even in cases of two-object collision. In three-object collision occasional re-initialisation is required.
Full article in PDFBy relating the projective camera model to the perspective one, using homogenous coordinates representation, the interior orientation parameters constitute what is called the calibration matrix. This paper presents two new algorithms to retrieve the calibration matrix from the projective camera model. In both algorithms, a collective approach was adopted, using matrix factorization. The calibration matrix was retrieved from a quadratic matrix term. The two algorithms were framed around a correct utilization of Cholesky factorization to decompose the quadratic matrix term. The first algorithm used an iterative Cholesky factorization to retrieve the calibration matrix from the quadratic matrix term. The second algorithm used Cholesky factorization to factor the quadratic matrix term but after its inversion. The basic argument behind the two algorithms is that: the direct use of Cholesky factorization does not reveal the correct decomposition due to the missing matrix structure in terms of lower-upper order. In both algorithms, a successful retrieval of the calibration matrix was achieved. This paper explains the key ideas behind the two algorithms, accommodated with a simulated example to demonstrate their validity.
Full article in PDFThe Direct Linear Transformation (DLT) model requires 3-D object space control points to estimate the full set of its parameters that can be used to recover exterior orientation parameters. The use of DLT in a planar object space leads to a rank deficient model. This rank deficient model leaves the DLT defined up to a 2-D projective transformation, which makes the explicit solution of the exterior orientation parameters or a single photo-resection represented by the camera position and orientations, a non-trivial task. This paper presents a new closed form solution to a single photo-resection in a planar object space based on homogenous coordinate representation and matrix factorization. Homogenous coordinate representation offers a direct matrix correspondence between the parameters of the 2-D projective transformation and the collinearity model. This correspondence lends itself to a direct matrix factorization to solve the photo-resection problem. The matrix factorization starts by recovering the elements of the rotation matrix and then solving for the camera position. It will be shown that an incomplete representation of the rotation matrix is captured by the 2-D projective parameters but the actual physical parameters of the rotation matrix still can be recovered explicitly and without any ambiguity. These elements were used to build a complete rotation matrix, which in turn facilitates the correct solution of the camera position. The developed solution can serve as a complementary companion to the classical DLT model. In this paper, a detailed derivation of the proposed solution will be presented, accommodated with two simulated examples to demonstrate its validity. The first example simulates aerial photography and the second one simulates close range photogrammetric application.
Full article in PDFIn this paper a new approach to the problem of edge preserving smoothing is proposed and evaluated. The new algorithm is based on the combined forward and backward anisotropic diffusion with incorporated time dependent cooling process. This method is able to efficiently remove image noise while preserving and enhancing image edges, lines and corners. The new method has been applied for the denoising of typical airborne images. Obtained results confirm good properties of the new noise reduction technique and its usefulness for the enhancement of various kinds of remotely sensed data in different detection and recognition systems [1,2].
Full article in PDFThe increasing resolution of SAR data opens the possibility to utilise this data for scene interpretation in urban areas. Different SAR specific phenomena like foreshortening, layover, shadow and multipath-propagation burden the interpretation. For many urban areas detailed additional information is available which can support both the mission planning and the analysis of the acquired SAR data afterwards. In this paper a high resolution LIDAR DEM is incorporated to investigate the impact of the phenomena on the visibility of scene objects by a SAR measurement from a given sensor trajectory and orientation in an urban environment. This kind of ground truth is well suited for this task because it contains elevation information of every object in the scene. Shadow and layover areas are simulated by incoherent sampling of the DEM. By a variation of viewing and aspect angles a large number of such simulations are carried out. From this set of segmentations the n best are determined according to two example tasks, namely the analysis of objects on roads and the detection and reconstruction of buildings. The locations of the roads and the buildings are extracted from a GIS.
Full article in PDFLiDAR (Light Detection and Ranging) is widely used in such fields as Digital Surface Model (DSM) production. It provides intensity data that reflect the material characteristics of objects, so it is possible that intensity data could be used for land-cover classification.
In this study, we assessed the possibility of land-cover classification using LiDAR intensity data instead of the multi-spectral data that has commonly been used for classification. We converted LiDAR point data to a grid and assessed the separability of intensity data on some classes, including asphalt road, grass, house roofs, and trees. However, the grid data was very noisy because of errors during data acquisition or from the resampling processes. To solve this problem, we examined some resampling and filtering methods that can remove noise effectively while the original information is preserved as much as possible. From this study, we concluded that LiDAR intensity data could be used for land-cover classification. Full article in PDF3D object reconstruction that is based on an overdetermined set of control points is considered in this contribution. The crucial difficulty of quality assessment of 3D reconstruction is nonlinearity of the fundamental equations that describes relationship between coordinates of corresponding points in image and reality. Solution of this difficulty is the main objective of this contribution. Probability distribution of measurement errors is not approximated by normal distribution, but by a special distribution that allows exact and efficient Bayesian solution of the original nonlinear problem. Bayesian approach to nonlinear estimation results in true probability distribution of estimated parameters that enables rigorous quality assessment of position of the reconstructed object.
Full article in PDFIn this paper, a new method for reconstruction of 3D segments from multiple images in urban areas environment is presented. Compared to previous algorithms, this one performs the matching of 2D segments in the Object Space through a sweep plane technique, thus avoiding the combinatorial exploration of all possible correspondences and handling images in a symmetric way. Furthermore, a method for reconstruction of 3D line from 2D lines, which takes into account the uncertainty on the parameters that define the 2D lines is also presented. It enables to get normalized residuals, which is used as a geometric criterion usable whatever the number of images is, to assess or reject potential correspondences. This criterion along with an unicity criterion is at the heart of the algorithm to prune the set of possible correspondences and to keep only reliable matches. Promising results are presented on simulated and real data. They show the ability of the algorithm to overcome detection errors in images and its robustness to occlusions in some images.
Full article in PDFA new approach for hybrid matching based on polynomial models is proposed. Polynomial plays a key role in matching process; while the Efficiency Index measures the accuracy of the matching result. Airborne laser and aerial photograph are two data sources for this study. Even those sources perform in different projections, one can be transformed to another projection based on purposed transformation model and finally represented in the same projection as parallel. Both manual and automatic functions were employed to assist and improve modes of gathering and increasing number of tie points. Apart from edges and corners, height constraint and buildings configuration are employed. By result of this study, it was found that an accuracy of matching at checked points, when polynomial transformation was executed, was 0.99 of EI, and in the meantime the correlation (R2) of resampling was about 0.99. Furthermore, height constraint was very important when various height of buildings were involved. It was found that dense buildings area with homogeneous height gave the highest accuracy of matching when polynomial transformation was applied.
Full article in PDFA supervised category classifier for satellite images by using the Modified Counter-Propagation(MCP) is proposed. The MCP is a neural network which consists of three layers: the input layer, the competition layer and the output layer[1,2,3]. The input and the competition layers form the Self-Organizing Map(SOM)[4,5,6]. The connections of Counter-Propagation[1,2,3] from the competition layer are extended to the output layer. The Landsat image data are adopted as the input data of the MCP, and the output layer consists of the pixel values, which represent categories to be classified. Our result shows that the MCP can classify more accurate and precise than that of the SOM only, especially for the classification of vegetation, farm and wood.
Full article in PDFTo realize a better representation of complete three-dimensional (3D) city models in Taiwan, this paper proposes an alternative concept and algorithm for 3D geometrical city surface modelling based on wavelets and least squares adjustment. The reasons why wavelets are adopted as an alternative module are stated. Firstly in this algorithm, a wavelets-based module for surface modelling makes it convenient to depict different types of city surfaces, where it enables an operator to choose a proper father wavelet in an interactive manner. It also can describe a fractal surface if fractal wavelets are adopted. Secondly, observations are acquired manually or (semi-)automatically. Simultaneously, any observation is recorded with a predefined code that defines a specific topology relationship with others. Then, the well-known least-squares adjustment is utilized to let the adopted wavelet surface function fit all observations, where the surface parameters, namely wavelet coefficients, in a local small stereo model (image window) are directly estimated and observation errors are filtered out. The surface function and observation equations are linear so that both estimation of initial values for unknown parameters and Gauss-Newton iteration are not needed. Current personal computer (PC) makes it possible to complete all computations in a short duration of time. It also provides quality figures, namely a covariance matrix, and enables a (near) real-time visual check on a DPW (Digital Photogrammetry Workstation). The afore-mentioned processes are done in each small window. Finally, a 3D city surface in a large area can be reconstructed by collecting the results in all windows, where neighbouring windows have a proper overlap. Some preliminary test results are also shown.
Full article in PDFThis paper deals with the purpose of a methodology integrating different surveying techniques to efficiently produce a large-scale digital mapping of urban areas.
At first, a comparison among methods for reuse/updating of existing maps, aerial modern techniques and terrestrial ones is carried out, highlighting the pros/cons characterizing each one and in the attempt to find the ″best technique″ as well a suitable integration/synergism among them.
The integration herewith adopted considers the availability of a 3D-vector numerical mapping and the photogrammetric exploiting of images acquired either (dynamically) by a low-cost MMS or (statically) by a non-metric high-resolution CCD. MMS allows the survey of 3D-control points for the successive processing of the CCD images.
Afterwards for such techniques, the global work strategy and the different analytical model applied are explained. The main features of the algorithm for MMS data processing are: a direct/indirect orientation model, the stochastic prediction of point coordinates, and a pseudo-dynamic solution approach. Moreover, the analytical model of photogrammetric use of CCD images takes into account the inauspicious errors due to MMS survey by means of suitable techniques for stochastic prediction of unknowns.
At the paper end, the promising results obtained from an application on real data of this integrated method are shown and evaluated.
Full article in PDFSince Airborne laser scanning sensors are operational and the data capture, including the calculation of the exterior orientation by using GPS and INS, has reached a high level of automation, the focus has turned on the development of algorithms to extract information from the 3D point cloud. The main tasks are the derivation of terrain information, forest parameters or the extraction of buildings. Since terrain information also affects the calculation of forest parameters and gives an input to building extraction, many different approaches were developed in the past years to derive highly accurate digital terrain models. Most of these approaches, like mathematical morphology, weight iteration or triangulation, work with the 3D data points itself. This paper presents an approach that is based on the rasterization of data points which allows the usage of fast digital image processing methods for the calculation of DTM´s. The algorithm consists of a hierarchical approach in combination with a weighing function for the detection of raster elements that contain no ground data. The weighing function considers the terrain shape as well as the distribution of the data points within a raster element.
Full article in PDFThe extraction of road crossings is often neglected by systems for automatic road extraction. Frequently, crossings arise as by-products of the grouping step. This leads to geometrical inaccuracies and visually and cartographically unsatisfactory results. In this paper, an approach for the improvement of extracted road crossings is presented as well as a method for the external evaluation of the extraction results.
Full article in PDFThis paper presents the results of a modified calibration method for the calculation of the interior orientation of digital consumer cameras. With a two-dimensional calibration field and the use of the CameraCalibrator 4.0 and PhotoModeler 4.0 software from EOS Systems Inc. an efficient and fast calculation method has been applied. The calibration results of the digital cameras Olympus E-10 and Nikon Coolpix 990 have been verified by repeated calibrations and the additional adjustment with the bundle adjustment software CAP. To show the important effect of full field calibration the results of a project are presented before and after calibration.
Full article in PDFRemote sensing in urban areas has been a challenge for quite some time due to complexity and fragmentation of objects and the combination of man-made features and natural features. Airborne laser altimetry data offer possibilities for feature extraction and spatial modelling in urban areas. There are many approaches of deriving buildings and other features currently available in literatures. However, there are many cases, which are still difficult for particular features to be extracted by using these approaches. For instance, in an urban area where many roads are raised above ground level with special characters similar to buildings. Building extraction in such a complicated urban context is still a difficult task for these available approaches. The proposed approach was developed to solve this type of cases. It tries to extract buildings through reasoning in a layer space in general. In the proposed approach, airborne laser altimetry data in raster format was segmented by using several thresholds with 1-meter interval of altitude. These image segments were then labelled and assigned with unique label values, which are treated as image objects. Hence, a number of properties can be derived based on labelled segments (image objects) such as size, shape, orientation etc. These properties are used for reasoning in the layer space. The layer space is defined as such that use altitude with 1-meter interval as a variable in X-axis and use these properties as functions of altitude in Y-axis. Vertically segmented image objects are linked and inferred vertically as well. A tree structure was created using links between different layers of segments vertically. Reasoning is based on patterns of these properties on the paths of each branch of searching tree in the layer space. Sequential experiments have been tested in a study area, southeast of Amsterdam based on the proposed approach. The test results show that the proposed approach is a robust and reliable approach in building extraction. It has potential capacities in extraction of other features. The theoretical framework, detailed description of major steps as well as experimental results are presented in this paper.
Full article in PDFThe successful lunch of IKONOS and QUICKBIRD, and a serious of plans for acquiring the remote sensing images with high resolution let it become possible that the basic spatial information is obtained from the remote sensing images in low costing and sort period. The first processing step is calculating the parameters of image position and orientation. Because of the very strong relativity of the traditional parameters of remote image with high resolution, the parameter calculation is not solved perfectly up to now. A strict geometric model, proposed in this paper, adopts a new method with three steps of transformations based on parallel ray projection. The first step is reducing the three dimensional space to the image space by the similar transformation. Then, the small space is projected to the level plane, which passes the center of the image plane, by parallel rays (Affine transformation). Finally, the level image is transformed to the original declining image. Every step of the model is strict, and the map function of each transformation is the first order polynomials and other simple function. The final calculation of the parameters is for the linear equations with good status. As a result, the problem of the relativity of image parameter calculation is solved completely. Some experiments are carried on for a lot of images with 10, 3 and 1-meters resolution. All of the results are quite perfect. Thus, the validity of the strict geometric model has been verified.
Full article in PDFIn the context of a proposed project for a video-based choreography tool, we carry out video acquisition experiments with a Sony DCR-VX2000 camcorder that operates at 25 frames/sec and a high-speed HCC1000 video camera with adjustable frame rates. We perform tests with frame rates up to 462 frames/sec and observe that the performance of the Kanade-Lucas-Tomasi tracking algorithm improves significantly as the frame rate increases. Furthermore, we demonstrate the application of an available dance notation, the Labanotation, to the video-captured ballet scene.
Full article in PDF