High Precision Photogrammetric Data Set for Building Reconstruction and Terrain Modelling with Ground Truth

This test data set was prepared for use in the project AMOBE (Automation of Digital Terrain Model Generation and Man-Made Object Extraction from Aerial Images) being conducted at ETH Zurich between the photogrammetric and vision groups and in cooperation with SRI. We are distributing this data with the purpose of providing a basis for comparison and evaluation of image understanding techniques, in particular those dealing with the reconstruction of man-made objects. Features of this data set include provision of the full photogrammetric information, four-way image overlap and ground truth.
 

The Avenches area

The set includes the following data:

1. High resolution, large-scale colour aerial images RGB images, image scale 1:5,000, images scanned with a geometrically precise photogrammetric scanner with 15 microns pixel size. Four images, forming a 2 x 2 block, with 60% forward and sidewards overlap were used. To reduce the amount of data only two image subsets (an industrial and a residential area) will be distributed. These areas are 1800 x 1800 pixels and are imaged in all 4 aerial images.
   Grey level images can be derived from the RGB images by making an RGB to HSV transformation and using the V channel as a grey level image. These images have not been included in this data set to reduce data size.
2. Full photogrammetric and other information on the images.
   • Transformation parameters from scanner (pixel) to photo (photogrammetric) coordinate system.
   • Interior orientation parameters (camera constant, coordinates of principal point).
   • Exterior orientation parameters They were determined by a bundle adjustment and include the position of perspective centers, rotation matrices and variance-covariance matrices (i.e. precision estimates) for the parameters of each image.
   • Time information, latitude and longitude This information can be used for shape-from-shadow computations.
3. Digital Terrain Models (DTMs) Data measured manually with a high accuracy photogrammetric instrument (expected accuracy = 0.1 m). The DTM is in ASCII format with a 10 m grid spacing. Contour information is contained in DXF- and IGES-format CAD models.
4. Buildings Separate DXF and IGES files for the residential and industrial area contain building roof models. These models are represented in terms of line entities manually measured by an operator to an expected accuracy of 0.1 m.
5. Orthophotos The residential and industrial areas in all 4 images were orthorectified (terrain-corrected) by using the DTMs. The ground resolution is approx. equal to the resolution of the original images. The map coordinates of the orthophotos and the grid spacing are given. These images are not included in this data set to reduce data size. They can be made available upon request from {mason,manos}@p.igp.ethz.ch
6. Digital Surface Model (DSM) (only for residential area) The difference to DTM is that DSM contains the visible surface, i.e. also 3-D objects not included in the DTM like buildings, trees etc. The DSM was generated with a commercial photogrammetric system (the Helava DPW 770) and has a grid spacing of 0.25 m. The DSM is distributed both in numeric form and as a grey level image.

All images are in Sun Raster format.

Description file

dataset_descr.pdf
The files contain all the description of the data set and the images. Thus, the images can be viewed first before deciding to transfer the data.

Get the data

You can download the data here: dataset.tar.gz

The size of the uncompressed data is 85 Mbytes. The top directory of the uncompressed data contains a description of the data set in the file dataset_descr.ps. NOTE that this file includes the description of all data, although some of them (grey level images and orthophotos) have not been included in dataset.tar

 
©Institute of Geodesy and Photogrammetry, ETH Zurich