Automatic building rooftop extraction using a digital surface model derived from aerial stereo images

Automatic building rooftop extraction is of great importance to many applications including building reconstruction, solar energy supply, and disaster management. This study proposes a building rooftop extraction method using DSM data generated from aerial stereo images and vegetation cover vector data. The method consists of five steps: noise filtering, dilation reconstruction, vegetation and terrain region removal, region growing and merging, and post-processing. We applied the proposed method to the centre of Shanghai, China, a typical urban area. Experimental results show that the proposed method can successfully extract building rooftops, with an approximately 82.6% quality percentage and 96.2% matched overlay.