Building change detection with RGB-D map generated from UAV images

Automatic change detection for urban buildings is very important for disaster assessment, map updating, etc. Height and color information is commonly used for change detection and existing methods use height information from 3D geometry model (e.g. Digital Surface Model, Geographic Information System) and color information from radiometric images captured by satellites or special aircrafts. However, they are either costly for timely change detection or sensitive to large illumination changes. With the rapid development of UAV technique, capturing the urban building images with high resolution camera at a low altitude becomes easier. In order to utilize these easily acquired aerial images, we propose a novel change detection framework with RGB-D map generated by 3D reconstruction, which can bear the large illumination change. Firstly, an image-based 3D reconstruction is applied to retrieve two point clouds and their related camera poses from two aerial image sets captured at different periods. Then, a RGB-D map could be generated from each 3D model, followed by a coarse-to-fine registration procedure to align the two reconstructed 3D point clouds together. At last, depth difference map and grayscale difference map could be generated from which we can use random forest classification and component connectivity analysis techniques to segment the changed building areas out. Experimental results have illustrated the effectiveness and applicability of the proposed framework.