Research On Robot Grasping And Positioning Based On 3D Target Point Cloud Contour

When the end effector of an industrial robot uses a vision system to capture a variety of disordered targets,inaccurate target recognition and unstable grasping positions will occur.This subject conducts research on the actual grasping of an object by an enterprise,and provides processing algorithms so that the robot end effector can obtain the accurate grasping point and pose of the target.First,the robot vision system is calibrated,and the rotation matrix of the robot coordinate system and the camera coordinate system is established.Since the original target point cloud collected by the 3D vision system has a lot of invalid information,preprocessing operations are carried out on the point cloud,including plane segmentation,voxelized grid downsampling,target segmentation and other processing processes.Research on typical target point cloud contour recognition algorithms,collect such as paper towels,pencil cases and other captured object point clouds,analyze the algorithms based on two-dimensional point set and three-dimensional point set contour recognition,and estimate the original normal,the method of extracting the target point cloud contour has been optimized.The 3D Douglas-Puck algorithm is introduced to remove internal points that do not belong to the target contour.At the same time,the 3D Douglas-Puck algorithm is improved to strengthen its applicability and make the algorithm applicable for small-scale multi-target point cloud contour recognition.Experiments show that the algorithm increases 19.75% in the removal rate of internal points compared with the normal estimation method.Research industrial robot target recognition and target posture acquisition methods,calculate the contour posture of the target at various angles,use the VFH feature histogram to establish the target model library,and mark the best capture point of the contour model,and compare and analyze the actual calculated capture point.The average calculation error of the grasping point is within 3mm,and the accuracy rate reaches 97.2%,which meets the requirements of the robot grasping accuracy.Completed the target segmentation and grasping position calculation of the two-dimensional image and three-dimensional point cloud in the robot system,and the contour extraction algorithm was experimentally verified.At the same time,in the three-dimensional visual grasping system after the robot hand-eye calibration,the three-dimensional point The algorithm for cloud edge extraction has designed a sorting and grasping experiment for multiple targets.The experiment shows that the proposed point cloud contour extraction algorithm can accurately calculate the grasping point of the target,and the robot can also accurately locate and The grasping is successful and can be used in industrial robot systems.