Dempster-Shafer fusion based building change detection from satellite stereo imagery

Building change detection is one of the most essential processing steps for urban monitoring using remote sensing data. As 2D automatic change detection from satellite images is a very challenging task, 3D information, derived e.g. from Digital Surface Models (DSMs) can provide valuable additional information to detect and analyze such changes. In the first step of the proposed approach, a radiometric co-registration workflow is proposed by automatically selecting pseudoinvariant feature (PIF) pixels using DSMs and multispectral satellite images. Then, an improved Dempster-Shafer (DS) fusion model is proposed for building change detection. In this fusion model, five indicators are adopted to classify five kinds of land cover change/non-change situations. The proposed method is applied to industrial areas using IKONOS stereo imagery of two dates. The evaluation result shows the improvement in true-detection-rate compared to the previous approach.