Multi-view Point Cloud Registration Based on Improved 3D-NDT Combining The Feature of NARF

An accurate and efficient algorithm of multi-view registration algorithm is proposed in this paper. First，corresponding feature points in each 3D images are found by using the NARF algorithm, and the FPFH feature of the keypoints is described , in a local coordinate system with origin in the keypoint position of the corresponding range image,and then false matches are eliminated by Correspondence Estimation and Correspondence Rejector Sample Consensus based on RANSAC. According to the corresponding relation, the matching feature points are identified, the transformation matrix is calculated, and the initial registration of point clouds is completed. Then, the point clouds are divided into three-dimensional voxel grids and registered precisely by the probability distribution function based on initial registration. Furthermore, incrementally register a series of point clouds two by two in order to transform all the clouds in the first cloud’s frame. The experiment results indicate that the method proposed in this paper can accurate register the point clouds on different viewpoints acquired by Kinect V2.0,and the efficiency of point cloud registration greatly.