A new plane segmentation method of point cloud based on mean shift and RANSAC

Abstract

Three dimensional laser scanning technology has been widely used in machine vision and reverse engineering. Plane segmentation is an important step for object recognition in the point cloud obtained by laser scanner. Traditional plane segmentation method cannot obtain a specific plane accurately when normal is unknown. This paper proposes a new method based on Mean Shift normal clustering and RANSAC with constraints and initial point to segment the specific plane whose the normal is unknown. Firstly, the point cloud is down sampled using Voxel Grid method. Secondly, the algorithm uses Mean Shift clustering method on the normal sphere to obtain the actual normal of the plane to be segmented. Thirdly, with stopping point as initial condition and actual normal as constraint, RANSAC algorithm is used to segment the specific plane. Finally this algorithm is experimentally validated in point cloud data of actual scene.