URBAN BUILDING COLLAPSE DETECTION USING VERY HIGH RESOLUTION IMAGERY AND AIRBORNE LIDAR DATA
Keywords: Urban building collapse detection, very high resolution image, airborne LiDAR data, multi-level, One-Class Support Vector Machines (OCSVM), image segmentation
Abstract. The increasing availability of very high resolution (VHR) remotely sensed images makes it possible to detect and assess urban building damages in the aftermath of earthquake disasters by using these data. However, the accuracy obtained using spectral features from VHR data alone is comparatively low, since both undamaged and collapsed buildings are spectrally similar. The height information provided by airborne LiDAR (Light Detection And Ranging) data is complementary to VHR imagery. Thus, combination of these two datasets will be beneficial to the automatic and accurate extraction of building collapse. In this study, a hierarchical multi-level method of building collapse detection using bi-temporal (pre- and post-earthquake) VHR images and postevent airborne LiDAR data was proposed. First, buildings, bare ground, vegetation and shadows were extracted using post-event image and LiDAR data and masked out. Then building collapse was extracted using the bi-temporal VHR images of the remaining area with a one-class classifier. The proposed method was evaluated using bi-temporal VHR images and LiDAR data of Port au Prince, Haiti, which was heavily hit by an earthquake in January 2010. The method was also compared with some existing methods. The results showed that the method proposed in this study significantly outperformed the existing methods, with improvement range of 47.6% in kappa coefficient. The proposed method provided a fast and reliable way of detecting urban building collapse, which can also be applied to relevant applications.