Automatic edge extraction by lidar-optical data fusion adaptive for complex building shapes

Abstract

This paper presents a new method of automatic edge extraction by LIDAR-optical fusion adaptive for complex building shapes. Different building features are extracted respectively from the two data sources and fused to form the ultimate complete building edges. Firstly, the points of each roof patch are detected from LIDAR point cloud, which consists of four steps, namely filtering, building detection, wall point removing and roof patch detection. Secondly, the initial edges are extracted from images using the improved Canny detector which is conducted by the edge location information from LIDAR point cloud in the form of edge buffer areas. Finally, the roof patch and initial edges are integrated to form the ultimate complete edge by mathematical morphology. This method presents an innovative strategy to fuse the two data sources to get building edges of high accuracy, which don't impose any constraints or rules on building shape. So this method is fully data-driven. The experimental results demonstrate that our method can automatically extract the accurate edges of various buildings of complex shapes, which have high robustness for complex scene.