Objectives Aiming at the defects of the traditional random sample consensus (RANSAC) algorithm for complex roof segmentation, is paper proposes an optimal RANSAC method for segmentation of complex building roof planes.

Methods First, the normal vector of the point cloud is estimated based on the k-nearest neighbor of the point cloud, and the seed selection process is optimized by using the point normal equation to improve the effectiveness of the initial plane generation; Second, the weight function based on the point normal and its distance to the initial plane is used to suppress the false plane generation. And inner points of the plane are modified iteratively by the weight function to improve the correctness of the segmented plane. Finally, the patch competition method is used to optimize the segmentation results and achieve roof point cloud segmentation.

Results Segmentation experiments results of multiple groups of buildings show that the proposed method can effectively suppress false plane, and the precision, recall and overall accuracy of roof plane segmentation of complex buildings are 96.8%, 98.2% and 95.1% respectively.

Conclusions Compared with the traditional method, the proposed method has obvious advantages in the accuracy and time-consuming of point cloud segmentation results.