For 2021 competition, the Council has approved the seven selected projects for funding. Council would like to take thank all applicants who responded this Call for Proposals. The following provides a brief summary of the above-awarded projects together with the information of their principle investigator(s) and co-investigator(s):

UAV-based Multi-sensor Dataset for Geospatial Research - UseGEO

PIs: F. Nex, University of Twente, Netherlands & M. Weinmann (Karlsruhe Institute of Technology, Germany)

CoIs: M. Y. Yang (University of Twente, Netherlands), E. Ozdemir (FBK Trento, Italy), F. Remondino (FBK Trento, Italy), B. Jutzi (Karlsruhe Institute of Technology, Germany)

This proposed Scientific Initiative aims at delivering a new and unique benchmark for the rigorous assessment of 3D reconstruction algorithms applied to datasets acquired with UAV platforms. Simultaneous acquisitions are foreseen using images, video sequences and LiDAR in different areas. While LiDAR will be used (primarily but not necessarily) as reference, the first and the second data sources will be used for the training and testing of multi-view and monocular 3D reconstruction algorithms. The acquired datasets will be accurately co-registered together to guarantee the best assessment. Different areas characterized by different landscapes will be considered in the acquisition, to deliver relatively heterogeneous scenes. The collected data will be then sufficiently large to allow the training and testing of all the typologies of algorithms, with specific regard to deep learning ones. To this purpose, the SI consortium will implement state-of-the-art algorithms to test the benchmark and provide a few baseline models for the scientific community. A dedicated website and specific dissemination activities will boost the visibility of this initiative inside and outside the ISPRS community.

Final Report »

ISPRS Benchmark Dataset on Object Detection in High-Resolution Satellite Images

PI: X. Sun, Chinese Academy of Science, China

CoIs: C. Wang (Xiamen University, China), M. Weinmann (Karlsruhe Institute of Technology, Germany)

Object detection in high-resolution satellite images aims to identify objects in an image and localize them either through bounding boxes, heat maps or segmentation. It provides an efficient way to monitor the earth and has been a demanding technology serving the civil life. ISPRS benchmark on object detection in high-resolution satellite images provides an effective way for the evaluation and comparison of object recognition and detection in high-resolution satellite images, which focuses on automated interpretation for optical and synthetic aperture radar (SAR) imagery. To stimulate and promote academic research, this benchmark also provides standard and large-scale datasets foundation for applying advanced deep learning technology to the field of remote sensing. The benchmark provides more than 15,000 satellite images, including more than 1000,000 labeled instances with bounding boxes and pixel-wise segmentation. The categories of datasets are more than 30 in total and the sensors cover optical and SAR. Comprehensive fine-grained types, large range of sizes and orientations, multiple sensors, tremendous object instances, and complex scenes with crowded objects make this benchmark more challenging. Benchmark datasets will be available from the dedicated webpage on the ISPRS website and mirror website, and interested participants can test their methods and submit their results for evaluation. The evaluation results are calculated by the well-known metrics (e.g., mAP, mIoU, Accuracy) and will be updated on the leaderboard webpage of the mirror website.

Final Report »

H3D - Hessigheim 3D - Benchmark on Semantic Segmentation of High-Resolution 3D Point Clouds and Meshes from Airborne LiDAR and Multi-View-Stereo-Image-Matching

PI: N. Haala, University of Stuttgart, Germany

CoIs: F. Rottensteiner (Leibniz Universität Hannover, Germany); J. D. Wegner (ETH Zurich, Switzerland); M. Kölle (University of Stuttgart, Germany); D. Laupheimer (University of Stuttgart, Germany), H. Ledoux (TU Delft, Netherlands)

Automated extraction of geographic objects from airborne data is an important research topic in photogrammetry and remote sensing since decades. In addition to the analysis of images, developed approaches frequently apply 3D point clouds from airborne LiDAR and Multi-View-Stereo-Image-Matching as a basic data source. In order to solve tasks like semantic segmentation and object detection, supervised machine learning systems such as Convolutional Neural Networks are becoming state-of-the-art. However, these techniques require large corpora of annotated training data, which are quite scarce in the geospatial domain. The Hessigheim 3D benchmark alleviates this issue by providing annotated 3D data sets to the community, which can be used by interested researchers to test own methods and algorithm.  The dataset consists of both LiDAR data and imagery captured from a recent UAV platform and processed by up-to-date software systems. The LiDAR point cloud incorporates a mean point density of about 800 pts./m2 while the imagery used for Multi-View-Stereo-Image-Matching and 3D mesh texturing have a Ground Sampling Distance of about 2-3 cm. This enables detection of fine-grained structures and represents the state-of-the-art in UAV based mapping. In the current version of the benchmark, the point clouds are manually labeled into 11 classes. The provided data set was flown in March 2018. Additional data sets captured in November 2018 and March 2019 will follow to additionally enable multi-temporal analysis. Further information on the underlying data, the evaluation process and how to participate can be found at the benchmark web-page https://ifpwww.ifp.uni-stuttgart.de/benchmark/hessigheim/default.aspx.

Final Report »

Urban Underground Space 3D Semantic Mapping

PI: B. Yang, Wuhan University, China

CoIs: Y. Chen, Xiamen University, China

Urban underground space plays a very important role in the global sustainable development and urban management due to the huge worldwide demand of infrastructure to accommodate the growing urbanized populations and the numerous renewals of aged facilities. 3D semantic mapping is of vital importance and urgency for exploring urban underground space. Manual mapping and semantic perception on underground space is laborious, dangerous and expensive task. However, there is little knowledge on automated 3D semantic mapping with accurate geometry and rich semantics and topological relations for underground space because of the unique characteristics of low-light, unknown layout, and dynamic environments of urban underground space, posing great challenges for positioning, perception, and 3D mapping. To fill this gap, this scientific initiative proposes 3D semantic mapping and benchmark of underground space to evaluate 3D mapping and object detection methods from multiple sensors (e.g., laser scanning, infrared camera), which will promote the exploration and research of ISPRS on underground space and enhance the reputation of ISPRS and cross-exploration. The expected outcome of the proposed scientific initiative is to facilitate interactions between researchers, developers with interests related to urban underground space mapping, and to promote the development of cross-cutting techniques of robot navigation, 3D SLAM, and 3D dynamic mapping.

Final Report »

ISPRS Open-Data Collector for Supporting Remote Sensing Analyses over Agriculture and Natural Ecosystems by Sharing Sampled Ground Truth (Shy)

PI: F. Pirotti, University of Padova, Italy

CoIs: M. Yoshimura (The University of Tokyo, Japan), J. Hernandez (Universidad de Chile, Chile), B. Leblon (University of New Brunswick, Canada), M. A. Brovelli (Politecnico di Milano, Italy), M. Yamashita (Tokyo University of Agriculture and Technology, Japan)

Sampling and measuring data in the field is costly and time-consuming and all too often data are used for specific projects and get easily lost shortly after the end of activities. Sampling data over areas covered with vegetation can be particularly complex (thick canopy/crop, wet areas, limited accessibility), but it is necessary for many remote sensing applications e.g. biomass estimation, pest monitoring, yield prediction, analysis of climate change effects. The volume of remote sensing data, from an increasing number of sensors and platforms, has been growing exponentially in the past years, making the availability of ground-truth information increasingly crucial for calibrating, correcting, training, validating and testing models. Shy will consist in a web-platform that will provide a user-friendly and guided interface for uploading spatial datasets with position, collected measurements and metadata regarding timestamp, description, authors, license, links and references to related literature (DOIs). Both metadata and spatial information will be searchable to allow users to find data that might support their research. Contributors can also choose to upload only the metadata with an external link to the data or with a contact for requesting the full dataset. Users can either download the data directly or send an email to authors asking for the full dataset, in case only metadata is provided. It is expected that the effort of the data owners be rewarded through dissemination of scientific products and public engagement through Shy. Geodata availability can foster new collaborations between research groups. The findability of geodata can increase efficiency of future surveying campaigns (e.g. using the same study area to add multi-temporal information). Last, but not least, ISPRS geotagged products, e.g. articles, benchmarks, can benefit from a centralized mapping and archiving service.

Final Report »

Modelling interrelationships among air quality, crop residue burning, soil moisture and crop yield using geospatial technology

PI: Raj Setia, Punjab Remote Sensing Centre, India

CoIs: S.K. Gupta (Punjab Remote Sensing Centre, India), H. Sembhi (University of Leicester, UK)

Rice-wheat cropping system is widespread in Indian Punjab because of the availability of high-yielding rice and wheat varieties, favorable soil and climate conditions, and subsidized inputs together with price support. The major constraint in rice-wheat cropping system is the small time window available between rice harvesting and preparing fields for the winter crop (wheat). In such situations, burning leftover rice stubble is a quick fix, easy and affordable solution. The burning of crop residues from the annual rice paddy harvest is a practice that dates back decades. Crop residue burning emits particulate matter and greenhouse gases, which aggravates poor air quality. Besides, the residue burning causes the nutritional loss (such as organic carbon, nitrogen, phosphorus, sulfur, and potassium) from the topsoil layer, thus reducing soil fertility and viability in the long run. The retention of crop residues helps in improving water infiltration and aeration within the soil profile. Many studies have evaluated the effects of residue burning on air quality. However, very few studies considered the interrelationship amongst air quality, crop residue burning, soil moisture, and crop yield. Hence, in this research, we will study the effect of initial soil moisture on the concentration of emitted gases due to crop residue burning but also the effect of crop residue burning on wheat yield in the parts of central Punjab (India) using satellite remote sensing. This study will help in the trade-off among food security, resource depletion, and environmental quality for achieving sustainable development in India, particularly Indo-Gangetic plains.

Pushing forward the development of software tools for IndoorGML

PI: A.A. Diakite, UNSW Sydney, Australia

CoIs: L. Díaz-Vilariño (University of Vigo, Spain), F. Biljecki (National University of Singapore, Singapore), S. Zlatanova (UNSW Sydney, Australia), K. Li (Pusan National University, South Korea), Ü. Işıkdağ (Mimar Sinan Fine Arts University, Turkey), B. Lathouwer (OGCE iVZW, Belgium)

In recent years, the interest in 3D indoor models is increasing. Often made available as BIM models (IFC), they are generally complex and detailed, raising thereby privacy concerns. IndoorGML is a standard for describing 3D indoor space to support Location Based Services (LBS). It contains a relatively simple geometry (space-based) and semantic and provides mechanism for aggregation, allowing to protect sensitive property information. IndoorGML relies on solid scientific concepts and offers a high flexibility with extension mechanisms. It provides a geometric, topological, and semantic description of the indoor, which facilitates specifically applications like indoor navigation or facility management. Accepted as an Open Geospatial Consortium (OGC) standard since January 2015, it is actively developed and extended by several universities. A new version, IndoorGML 2.0 is currently under development to enhance and comply with user requirements. However, despite its solid conceptual basis, IndoorGML is suffering from a lack of practical tools and remains largely an academic development. The goal of this project is to bring together all the actors of the IndoorGML ecosystem around the development of a basic and strong software tool for supporting IndoorGML production. The intended tool will focus on producing IndoorGML models from common 3D building standards such as IFC. This will heavily leverage currently existing developments. We expect more 3D IndoorGML models to be created and made freely available for research and development within the ISPRS community but also as examples to industry developers and end users.

Final Report »