一种基于法向量夹角的点云配准方法

New approach of point cloud registration algorithm based on intersection normal angle

点云配准是逆向工程、机器人导航、计算机视觉等领域中进行三维重建的关键问题.针对4PCS配准算法对点云数据密度变化强烈的情况表现不稳定,以及为了保证效率海量点云数据必须进行下采样而导致对应点对无法得到保证的情况,本文提出了基于法向量和邻近点数目的特征点提取方法对算法进行改进.选取一个半径范围内的点作为邻域,并通过总体最小二乘法拟合局部平面求解法向量,之后利用法向量夹角和邻域大小进行特征点提取,最后在特征点集上进行4PCS算法.因为点集基数大幅度减少并且特征明显,有效提高了4PCS算法的速度和精度.

Point cloud registration is a key problem of 3D reconstruction in reverse engineering, robot navigation, computer vision and other fields. Because the 4PCS registration algorithm cannot cope well with the point cloud density varying strongly, and point-to-point correspondence cannot being guaranteed caused by huge point cloud must being down-sampled in order to remain efficient. In this paper the feature points extraction method based on the normal vector angle and the number of proximate point is proposed. Points within a radius are selected as a neighborhood, the local least square method is used to fit the local plane for computing normal vector. Then feature points are extracted by using the angle between the normal vector and the neighborhood size, finally the 4PCS algorithm is applied in the feature points set. Due to the scale of the point cloud is greatly reduced, and registration is performed on a set of points with obvious feature,this method can effectively improve the speed and accuracy of the 4PCS algorithm.