基于KD-ICP算法的机器人拆解目标三维点云配准方法  

3D Point Cloud Registration Method for Robot Dismantling Targets Based on KD-ICP Algorithm

作　　者：沈晨 薛龙[1] 李万祺 郑黎明 吴成龙 邹勇[1] SHEN Chen;XUE Long;LI Wan-qi;ZHENG Li-ming;WU Cheng-long;ZOU Yong(Mechanical Engineering College,Beijing Institute of Petrochemical Technology,Beijing 102617,China)

机构地区：[1]北京石油化工学院机械工程学院,北京102617

出　　处：《科学技术与工程》2024年第21期9025-9032,共8页Science Technology and Engineering

基　　金：国家重点研发计划(2022YFB4701101,2019YFB1310805)。

摘　　要：随着中国核能产业发展,大量在役涉核装置运行过程中出现的事故和隐患需要通过涉核机器人进行远程处置,部分事故处置需对出险装置进行精确拆解,基于三维重建视觉图像定位是其关键技术之一。针对核应急事故中待拆解目标的精确重建,开展三维点云配准方法研究,在配准算法总体流程基础上设计出三维点云配准算法,利用高斯滤波进行点云数据去噪,提出一种兼顾点云信息完整性和算法效率的K维迭代最近点(K-dimensional iterative closest points,KD-ICP)算法,利用点快速特征直方图(fast point feature histograms,FPFH)和主成分分析(principal component analysis,PCA)实现特征点提取并降低点云数据量,使用全等四点集(4-points congruent sets,4PCS)进行点云粗配准,采用KD-ICP算法实现点云精配准。在所搭建的拆解机器人试验平台上,针对管、板及涉核装置模拟体等测试对象,对所提出的配准算法进行试验分析与评价,与其他3种常规算法的对比试验表明,所提出的点云算法在相同的条件下均方根误差最小,运行时间最短,具有更高的点云配准效率及较好的鲁棒性,为复杂拆解目标的三维重建及精确定位奠定了基础。With the development of China s nuclear energy industry,numerous accidents and hidden dangers that occur during the operation of nuclear equipment need to be remotely disposed of using nuclear robots.Precise dismantling of affected equipment is required for some accident disposals,and 3D reconstruction and visual image positioning are key technologies for this purpose.A method for 3D point cloud registration was presented to achieve precise reconstruction of targets to be dismantled in nuclear emergency accidents.A 3D point cloud registration algorithm was proposed based on the overall process design of the registration algorithm,with Gaussian filtering used for denoising the point cloud data.The K-dimensional iterative closest points(KD-ICP)algorithm was introduced to balance the integrity of point cloud information and algorithm efficiency.Feature point extraction and point cloud data reduction were achieved using fast point feature histograms(FPFH)and principal component analysis(PCA)principal component analysis.Coarse registration of point clouds was performed using 4-points congruent sets(4PCS),while fine registration was accomplished using the KD-ICP algorithm.The proposed algorithm was evaluated on a dismantling robot test platform,targeting objects such as pipes,plates,and simulated nuclear equipment.Comparative experiments with three other conventional algorithms showed that the proposed algorithm achieved the smallest root mean square error and shortest running time under identical conditions,exhibiting higher point cloud registration efficiency and better robustness.The study lays a foundation for 3D reconstruction and precise positioning of complex dismantling targets for nuclear robots.

关 键 词：三维点云 点云配准 KD-ICP算法 拆解机器人 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]