Target Pose Recognition And Grasping Based On 3D Point Cloud Information

With the continuous advancement of industrial intelligence,vision-guided robotic grabbing systems have been widely used in assembly-line-based product assembly,sorting,packaging,and palletizing.Vision technology is the core of this type of capture system,and 3D vision has received more and more attention and research because of the more detailed information such as the shape,position,and posture of the target it expresses.Aiming at the problem of automatic grabbing of scattered workpieces,this subject has studied target pose recognition and grabbing technology based on 3D point cloud information,and designed and implemented a prototype of a 3D vision-guided robotic grabbing system.The main results are as follows:(1)Research on point cloud preprocessing technology.The depth image of the Realsense D435 depth camera was used to reconstruct the point cloud of the three-dimensional scene,and the irrelevant background was removed by pass-through filtering and a plane detection algorithm based on RANSAC.Aiming at the interference problem of outlier point noise,a fast search algorithm for point cloud neighbors based on kd tree is proposed.By reducing the search range,the point cloud neighborhood query speed is improved,and the point cloud noise removal efficiency is improved.Aiming at the huge amount of point cloud data,a downsampling algorithm based on voxel lattice was used to streamline the point cloud data.(2)Research on point cloud segmentation technology.Aiming at the selection of distance threshold based on Euclidean distance segmentation algorithm,an adaptive threshold calculation method is proposed.This method automatically adjusts the distance threshold based on whether the query point is the boundary point of the point cloud,which effectively improves the point cloud segmentation Efficiency and accuracy.In order to make up for the shortcomings of the Euclidean distance that can't segment the adhesion target,a segmentation strategy based on the edge cloud point information is proposed.This strategy can find the edge line according to the degree of change of the normal direction of the point cloud surface.Use Euclidean distance segmentation algorithm to achieve target segmentation.(3)Research on point cloud target recognition and pose registration technology.On the basis of the FPFH local point feature descriptor,an angle element between the surface normal vector at each point and the plane normal vector of the stage is added to construct a new global feature descriptor GFH.Obtain point cloud information of workpieces in multiple perspectives and different poses and extract their global features GFH to create a point cloud template library,and use this template library for feature matching to realize point cloud type recognition and rough pose estimation,and then use this The ICP registration algorithm performs target pose registration to obtain the precise pose of the target.(4)Research on target grabbing technology based on 3D vision guidance.The standard D-H kinematic model of the UR3 manipulator was established,and the inverse kinematics solution was solved by the analytical method.Use the calibration board to perform hand-eye calibration on the robot arm and camera,calculate the pose of the target in the base coordinate system of the robot,and then determine the gripping attitude of the grippers at the end of the robot arm based on the gripping points.Calculate the rotation angle of each joint axis of the robot arm and send it to the robot arm control cabinet via TCP / IP to realize the grasping of the target by the robot arm.A large number of experimental results show that the target pose recognition algorithm based on point cloud information designed in this paper has good adaptability to relatively complicated situations such as target sticking and stacking,and the robot based on 3D vision guidance can accurately capture various targets.