Towards the generation of 3D OpenStreetMap building models from single contributed photographs

3D city models are valuable and useful for many experts using this information for a wide array of environmental and sustainable analyses and services. To produce 3D city models, conventional production processes are typically conducted by authoritative mapping agencies, which mostly rely on aerial photogrammetry and laser scanning. Consequently, obtaining public domain 3D city models is challenging and limited, where the majority of open data is collected and mapped by participatory mapping driven communities. These are still limited to 2D data collection proficiencies due to the used mapping infrastructures and technological limitations. Thus, the 3rd (height) dimension is mostly missing from these maps and models, resulting in the fact that public domain 3D city models are still limited, and only scarcely used for environmental applications. Perhaps one of the most important features in 3D city models are the buildings, since they serve as a major geospatial element in many environmental applications. Our objective is to use a single contributed photograph and OpenStreetMap vector data to precisely calculate the photographed building height, and add this data to the OpenStreetMap map to enable the creation of open source Level of Detail 1 (LoD1) city models. To this end, we have developed a Newton's-based method in optimization to accurately calculate building heights from single contributed photographs taken by citizens using smartphones or tablets. An Android app, OpenStreetHeight, is developed to carry out the experiments. Based on the various medium-height buildings that were photographed using the app and processed using the developed algorithms we received accurate building height values. When compared to reliable reference field measurements, the average height mean absolute error was 30 cm. Combined with the OpenStreetMap footprint vector data, we were able to produce an average LoD1 volume mean absolute error of less than 5%, satisfying the CityGML standard quality. This research presents a framework for a semi-automatic crowdsourced user-generated content calculation of OpenStreetMap building heights, and the creation of reliable and accurate LoD1 building models, as a first step to enhance the already established 2D OpenStreetMap map infrastructure to the 3D domain. This enables expanding the use of OpenStreetMap as a comprehensive and detailed representation of our urban environment for various environmental and sustainable applications and analyses.