关键词: 点云立体粗配准, 点云立体精配准, 飞行时间相机, 三维点云重建

Abstract: Against the problem of low accuracy, slow speed of point cloud registration in 3D point cloud modeling on the surface of the object based on the principle of time of flight(TOF) camera, a fast, easy method is proposed to realize the scattered points clouds registration, this method realize the point cloud initial position registration through extracting feature point from distance image of target objects, obtaining the initial transform parameter using an iterative solving process. Based on this, using gradient value of intensity image of TOF camera and K-nearest neighbor algorithm based on the local decomposition of 3D space to find the closest point between point cloud as the matching point pair. According to the iterative process of primitive iterative closest point(ICP) algorithm to iterate the matching point pair and obtain the best transform parameter in point cloud, at the same time, improving the precision of iterative point clouds registration combined with the error matching point removing rule, to realize the point cloud precise location registration. To verify the algorithm combined with the actual space objects, the experimental results show that this point cloud registration improved speed and the precision of registration compared with the traditional registration method, this will help to improve the precision of object surface reconstruction, have a high practical value.

Key words: 3D point clouds reconstruction, Point clouds stereo coarse registration, Point clouds stereo fine registration, Time of flight camera