Point Cloud Registration Algorithm Based on Improved Rotational Projection Statistical Feature Descriptor

Abstract

Aiming at the problems of poor registration accuracy and low computational efficiency of point clouds with low resolution and uneven density distribution, an automatic point cloud registration algorithm based on improved rotational projection statistical features is proposed. Firstly, the feature points are extracted using the Intrinsic Shape Signatures (ISS) algorithm, then the feature points are described using the Improved Rotation Projection Statistics (IRoPS) algorithm, and then use Random Sampling Consistency (RANSAC) to eliminate false matches and calculate the transformation matrix, and finally complete the fine registration based on the improved Iterative Closest Point (ICP) algorithm. Experiments on three data sets show that the algorithm has the advantages of strong anti-interference ability, high registration accuracy and fast calculation speed, and can meet the registration requirements in practical engineering.