一种新的三维点云兴趣点提取算法  

A NOVEL ALGORITHM FOR INTEREST POINT EXTRACTION OF 3D POINT CLOUDS

作　　者：郭建华 吕常魁[1] Guo Jianhua;LüChangkui(College of Mechanical and Electrical Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,Jiangsu,China)

机构地区：[1]南京航空航天大学机电学院,江苏南京210016

出　　处：《计算机应用与软件》2023年第3期248-254,297,共8页Computer Applications and Software

基　　金：国家自然科学基金项目(61671240)。

摘　　要：现有的三维点云兴趣点提取算法容易漏检和误检兴趣点,针对该问题,提出一种新的三维点云兴趣点提取算法。假设锥体为三维物体边角基元特征,根据各点与其k个近邻点的差向量集,构建突出度特征值描述点的局部锥度特征。基于点云突出度特征值的全局阈值得到初始兴趣点集,按照局部最大原则获取候选兴趣点集,依据每个候选兴趣点被重复选中的次数进行投票,获取最终兴趣点。在单位圆上模拟点云的突出度相关参数特征,检验了算法的鲁棒性。以人工标注统计确定的兴趣点作为真实值评估算法的性能,结果表明,该算法能准确提取到大部分真实兴趣点,整体性能优于传统算法。The existing interest point extraction algorithms of 3D point clouds are prone to misdetection and omission.To solve this problem,we proposed a novel algorithm for interest point extraction of 3D point clouds.Under the assumption that cone was the primitive feature of the edges and corners of a 3D object,salience degree was constructed to describe the local cone feature of the object based on the difference vectors set calculated by each point and its k nearest neighbors.The initial set of interest points was obtained based on the global threshold of the salience degree of point cloud data.The set of candidate interest points was obtained based on the local maximum principle.The final interest points were selected by voting according to the number of times that each candidate interest point was repeatedly selected.The parameters related to salience degree of the point clouds were simulated on the unit circle to test the robustness of the algorithm.The performance of the proposed algorithm was evaluated using manually labeled real interest points as discriminating criterion.The results show that the proposed algorithm can accurately detect most real interest points,and outperformed traditional algorithms in terms of overall performance.

关 键 词：点云 兴趣点 锥体 突出度特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]