图像分割引导的散堆工件结构光三维位姿估计

Structured Light Based 3D Pose Estimation of Piled Workpieces Guided by Image Segmentation

针对散堆工件场景中点云生成耗时久、位姿估计困难、多类工件混合情形难处理等问题,提出图像分割引导的散堆工件结构光三维位姿估计方法,并基于JAkA Zu36 DoF机器人开发散堆工件抓取实验系统。采用YOLACT模型获取散堆工件图像中的工件信息,通过自适应阈值筛选待抓取工件,利用双目结构光生成目标工件所在区域的局部点云,并基于投票匹配算法和迭代最近邻算法估计工件位姿。通过搭建的实验系统对文中方法进行测试。实验结果表明:系统完成目标工件位姿估计时间约为3.641 s,其中,点云计算需0.536 s,点云配准需0.446 s;与其他方法相比,文中方法平均可缩小点云规模44%,点云生成时间平均缩减24%,配准成功率提升至100%。

Aiming at the problems of point cloud generation time consuming,pose estimation difficulty,and multi class workpiece mixing difficult handling in the scene of piled workpieces,a structured light based 3D pose estimation of piled workpieces guided by image segmentation is proposed,and an experimental system of piled workpiece picking is developed based on the JAkA Zu36-DoF robot.The YOLACT model is used to extract workpiece information from the piled workpieces images.The workpiece to be grabbed is filtered through adaptive threshold,the local point clouds in the area where the target workpiece is generated using binocular structured light.The workpiece pose based on the voting matching algorithm and the iterative nearest neighbor algorithm is estimated.The proposed method is tested by the constructed experimental system.The experimental results show that the system takes approximately 3.641 s to complete the target workpiece pose esti-mation.Among them,point cloud computing takes 0.536 s and point cloud registration takes 0.446 s.Compared with other methods,the proposed method can reduce the size of point clouds by an average of 44%,reduce the time of the point clouds generation by an average of 24%,and improve the registration success rate to 100%.