An algorithm of Building Extraction by Fusing Spectral and Geometrical Features

LiDAR system can obtain 3D information rapidly and conveniently. It has been widely applied in object detection and extraction. However, because of the lack of spectral feature, error detections often occur at dense canopy in building extraction. Aiming at the problem, an algorithm of building extraction fusing spectral and geometrical features is proposed. Firstly, the spectral information is extracted by registration between LiDAR point data and aerial images. Secondly, spectral and geometrical features are fused by improved tensor voting, the new fusion feature is created. Finally, extract building points by random forest algorithm. Simulation test dataset are provided by ISPRS, the simulation results show that building points extraction accuracy is improved a lot after fusing spectral feature. The extraction accuracy reaches 94.26%. The importance of spectral feature has been proved.