Towards Automatic 3D Change Detection inside Urban Areas by Combining Height and Shape Information

Monitoring of urban areas using remote sensing data requires reliable change detection techniques. While most of the changes are optically visible and easily detectable by an expert user, automatic processes that remain valid even when different kinds of input data are considered, are quite difficult to develop. This paper provides new solutions for semi-automatic 3D change detection of buildings based on the joint use of height and spatial information. It is an attempt to build a reliable scheme for change detection able to process high as well as lower quality Digital Surface Models (DSMs). The subtraction of DSM, computed from stereo pairs acquired at different epochs, provide valuable information about 3D urban change. However, when at least one of the DSMs presents some artifacts, a simple DSM subtraction may result also in the detection of virtual changes. Several post-processing steps are proposed in this paper and adapted to different DSM qualities in order to quantify real changes. Shape features are introduced to describe the geometry of the detected changes and a Support Vector Machine (SVM) classifier is used to differentiate real from virtual changes. Evaluation of the proposed approach on object and pixel level in terms of completeness, correctness, overall accuracy, etc is performed, proving its efficiency and relatively high accuracy for different kind of stereo images and consequently different DSM qualities.