3D change detection using low cost aerial imagery

We present a method to register point clouds obtained from aerial images through Structure from motion (SFM) techniques with data from airborne LiDAR systems. The data was obtained by the United States Geological Survey (USGS) over a 800 sq km stretch in California using airborne LiDAR. The images were obtained by a downward looking camera on an autonomous helicopter along the San Andreas fault [9]. A 3D point cloud is built by fusing GPS information with the aerial images. Our approach to detect changes is to compare the LiDAR data with 3D point cloud derived from aerial images. This comparison necessitates the two point clouds to be in the same co-ordinate frame. We adopt a registration approach to bring the point clouds to the same co-ordinate frame. We highlight the challenges involved in registering aerial point clouds and propose a semi automated way for registration. We also present a simulation of a change detection scenario by introducing displacement fields in the source point cloud and obtaining a target point cloud by additionally simulating the GPS offsets. We recover the displacement vectors in two steps (1) globally registering the source and target point clouds using the method described in this paper (2) using our change detection module [5] for computing the displacement fields. We present results for global registration and change detection.