Research On 6D Pose Estimation Of Scene Target Based On Structured Light 3D Vision

Random bin picking(RBP)is a robot that uses 3D vision technology to grab scattered and stacked objects.With the application of 3D vision technology,3D vision robot grabbing is a popular direction.3D vision has one more dimension than 2D vision.The environment information is more abundant,and the robot grasping based on 3D vision can handle the problems in more complex environments,which solves the limitations of 2D vision robot grasping.The pose of an object can be represented by 3 rotations and 3 translations in a reference coordinate system in three-dimensional space.Obtaining the 6D pose information of the object to be grasped is the premise of robot grasping and computer-controlled robot motion.In order to obtain the 6D pose information of the target,this paper builds a structured light 3D reconstruction system to first obtain the 3D point cloud information of the object,and use the point cloud processing algorithm to estimate the 6D pose of the scene target,including the construction of the mechanical platform and the selection of equipment.Obtain the 3D point cloud information of the object.Use surface structured light projection technology,including the problem of grating resolution,camera calibration,and structure cursor calibration in the structured light projection technology.After the entire system is calibrated,the 3D points of the object can be obtained.Cloud information.Aiming at the problem of phase jump error in the traditional phase-resolving algorithm,this paper proposes an improved phase-resolving algorithm,which is more accurate than the existing algorithm for phase-resolving and measurement results.Simulation analysis shows the effect of the improved algorithm on error suppression.Experimental verification shows the suppression effect of phase error in real environment.And the 3D reconstruction accuracy is verified by the experiment of measuring standard blocks,which shows the reliability of the robot grasping.6D pose estimation includes two stages: online and offline.The point cloud processing algorithm is used to calculate the point cloud model and the characteristics of the scene point cloud.The offline stage is about the creation of the reference point cloud model and the calculation of the point cloud model pose.The last stage is about the processing and calculation of the scene point cloud,using the characteristics of the model and the scene point cloud to perform target matching to obtain the pose transformation matrix of the scene target in the camera coordinate system.Aiming at the problem of multi-target matching,this paper improves the multi-target matching algorithm,so that only the missed matching of the target caused by the clustering algorithm can be matched,which is more suitable for robot grasping.Finally,a robot grasping platform was built on the basis of the structured light 3D reconstruction system to verify the accuracy of 6D pose estimation.After multiple sets of robot grasping experiments,it shows that the 6D pose estimation of the scene target and the accuracy of robot grasping,finally show that the research content of this paper is feasible.