Research on Feature Point Cloud Extraction and Registration Based on SIFT-SRBICP Algorithm

Abstract

In view of the low efficiency and poor stability of image feature point extraction, as well as the inaccurate matching problem caused by scale transformation and the poor registration effect caused by too large rotation angle change in the actual situation of point cloud, this paper compares and analyzes three algorithms of SIFT, SURF and ORB, respectively, and verifies that the SIFT algorithm is better than the other two algorithms in solving the problem of image feature point extraction. In this paper, the SIFT algorithm is used to extract feature points, and an improved SRBICP algorithm is proposed to register the point cloud. This method not only increases the randomness of initial pheromone and considers the boundary of scale matrix, but also adds some factors such as rotation angle constraint matrix, dynamic iteration coefficient and annealing coefficient to construct point cloud registration model. Finally, the Ubuntu operating system based on the open source GNU/Linux system is used to perform simulation experiments on the Ubuntu system platform. The experimental results show that the improved algorithm improves the registration accuracy by about 50%, and the registration speed by about 40% compared with the traditional ICP algorithm..