Automatic building change detection in wide area surveillance

We present an automated mechanism that can detect and characterize the building changes by analyzing airborne or satellite imagery. The proposed framework can be categorized into three stages: building detection, boundary extraction and change identification. To detect the buildings, we utilize local phase and local amplitude from monogenic signal to extract building features for addressing issues of varying illumination. Then a support vector machine with Radial basis kernel is used for classification. In the boundary extraction stage, a level-set function with self-organizing map based segmentation method is used to find the building boundary and compute physical area of the building segments. In the last stage, the change of the detected building is identified by computing the area differences of the same building that captured at different times. The experiments are conducted on a set of real-life aerial imagery to show the effectiveness of the proposed method.