The necessary yet complex evaluation of 3D city models: a semantic approach

International audienceThe automatic modeling of urban scenes in 3D fromgeospatial data has been studied for more than thirty years.However, the output models still have to undergo a tedious taskof correction at city scale. In this work, we propose an approachfor automatically evaluating the quality of 3D building models.A taxonomy of potential errors is first proposed. Handcraftedfeatures are computed, based on the geometric properties ofbuildings and, when available, Very High Resolution images anddepth data. They are fed into a Random Forest classifier for theprediction of the quality of the models. We tested our frameworkon three distinct urban areas in France. We can satisfactorilydetect, on average 96% of the most frequent error