基于灰色关联分析的三维点云识别算法  

3D point cloud recognition algorithm based on grey relational analysis

作　　者：郑震 查冰婷 谭钰然 张合[1] Zheng Zhen;Zha Bingting;Tan Yuran;Zhang He(School of Mechanical Engineering,Nanjing University of Science and Technology,Nanjing 210094,China;Science and Technology on Electromechanical Dynamic Control Laboratory,Xi’an 710000,China)

机构地区：[1]南京理工大学机械工程学院,江苏南京210094 [2]机电动态控制重点实验室,陕西西安710000

出　　处：《南京理工大学学报》2022年第6期679-687,共9页Journal of Nanjing University of Science and Technology

基　　金：国家自然科学基金(52201399);中国博士后科学基金(2021M701713);军委科技委技术领域基金(2022-JCJQ-JJ-0394);机电动态控制重点实验室开放课题基金(2021-JCJQ-LB-039);江苏省卓越博士后计划(2022ZB245)。

摘　　要：为了适应激光引信识别战场目标时高动态、信息量少与干扰复杂的环境,该文提出一种基于灰色关联分析(GRA)的三维点云识别算法。构建由目标尺寸信息、目标轮廓信息组成的26维目标特征描述子,分析描述子中各元素的信息量、信息重要性、抗噪声能力和对遮挡的鲁棒性。提出基于层次分析法(AHP)的特征权重分配方法,建立目标特征空间转换和基于GRA的三维点云识别模型,实现对典型目标的识别。完成了对典型目标的仿真。实验结果表明:各种遮挡情况下的尺寸描述子的综合关联度最小为0.811,截面轮廓描述子的综合关联度在Ⅰ级和Ⅱ级的非严重遮挡下最低达到了0.875;对于Ⅲ级遮挡和Ⅳ级遮挡,通过降低特征融合时截面轮廓描述子边缘元素部分的分配权重,F最小为0.69;在各种遮挡情况下的平均识别率达到了0.862;最大识别时间为12 ms。In order to adapt to the high dynamic,less information and complex interference environment when laser fuzes recognize battlefield targets,a 3D point cloud recognition algorithm based on gray relational analysis(GRA)is proposed here.A 26 dimensional target feature descriptor consisting of target size information and target contour information is constructed,and the information content,information importance,anti noise ability and robustness to occlusion of each element in the descriptor are analyzed.A feature weight assignment method based on analytic hierarchy process(AHP)is proposed,and target feature space transformation and 3D point cloud recognition models are established based on GRA to realize the recognition of typical targets.The simulation of typical targets is completed.The experimental results show that the minimum comprehensive correlation degree of the size descriptor under various occlusion conditions is 0.811,and that of the profile descriptor under non severe occlusion conditions as level Ⅰ and Ⅱ is 0.875;for level Ⅲ and Ⅳ occlusion,the minimum F is 0.69 by reducing the assigned weight of sub edge elements of profile descriptor during feature fusion;the average recognition rate under various occlusion conditions reaches 0.862;the maximum recognition time is 12 ms.

关 键 词：灰色关联分析 点云处理 目标识别 激光引信 战场目标 三维点云 层次分析法 目标特征 

分 类 号：TJ439.2[兵器科学与技术—火炮、自动武器与弹药工程]