摘要:针对图像特征点提取效率低和稳定性差，点云集在实际情况下由于尺度变换引起的匹配不准确和旋转角变化过大引起的配准效果不佳的问题。本文对SIFT、SURF及ORB三种算法进行对比分析，验证了SIFT算法在解决图像特征点提取问题上优于其它两种算法。本文采用SIFT算法进行特征点提取，并提出基于改进型的SRBICP算法对点云进行配准，该方法不仅增加了初始信息素的随机性和考虑了尺度矩阵的边界，并加入了旋转角约束矩阵、动态迭代系数以及退火系数等因素对点云配准模型进行构建。最后采用基于开源GNU/Linux系统所搭载的的Ubuntu操作系统，在Ubuntu系统平台上进行仿真实验，实验结果表明，改进后的算法比传统ICP算法配准精度提高了约50%，同时配准速度提高了约40%。

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..