融合手部姿态的物体6D位姿估计算法  

An object 6D pose estimation algorithm integrated with hand pose

作　　者：王健[1] 郭宇[1] 黄少华[1] 汤鹏洲 郑冠冠 陆云霞 WANG Jian;GUO Yu;HUANG Shaohua;TANG Pengzhou;ZHENG Guanguan;LU Yunxia(College of Mechanical and Electrical Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 210001,China;Beijing Spacecrafts Limited Company,Beijing 100190,China)

机构地区：[1]南京航空航天大学机电学院,南京210001 [2]北京卫星制造厂有限公司,北京100190

出　　处：《现代制造工程》2025年第2期76-83,共8页Modern Manufacturing Engineering

摘　　要：在基于增强现实装配引导的复杂零/部件装配场景中,由于手部对零/部件的遮挡,导致零件位姿解算时产生较大的误差,甚至造成求解失败。目前针对手工装配零件的位姿估计算法在解决零件遮挡问题时没有考虑手部信息,使得位姿估计精度难以满足增强装配实际应用的要求。针对上述问题,提出了融合手部姿态的零件6D位姿估计算法,即HandICG算法。该算法将手部的姿态信息与迭代对应几何(Iterative Corresponding Geometry,ICG)算法进行融合,当发生手部遮挡时,将手部的姿态信息应用到零件姿态的求解中,从而显著提高手部遮挡情况下零件位姿估计的精度,实验表明,平均模型点距离(Average Distance of Model points,ADD)相关评价指标达到74.73%,是ICG算法的2.61倍。该算法显著提升了增强装配场景中零件位姿解算的准确性和鲁棒性。In complex component assembly scenarios based on augmented reality assembly guidance,the occlusion of the parts by the hands leads to significant errors and even failure in the pose calculation of the parts.At present,pose estimation algorithms for manually assembled parts do not consider the use of hand information when solving part occlusion problems,making it difficult for pose estimation accuracy to meet the requirements of augmented assembly.In response to the above issues,the article proposes a 6D pose estimation algorithm for parts that integrates hand posture,namely the HandICG algorithm.This algorithm integrates the pose information of the hand with the Iterative Corresponding Geometry(ICG)algorithm.When hand occlusion occurs,the pose information of the hand is applied to the solution of part pose,significantly improving the accuracy of object pose estimation under hand occlusion.Experiments show that the Average Distance of Model points(ADD)index reaches 74.73%,which is 2.61 times that of ICG.This algorithm significantly improves the accuracy and robustness of part pose calculation in augmented assembly scenarios.

关 键 词：位姿估计 手部姿态 遮挡 增强装配 跟踪注册 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]