Research On Point Cloud Data Processing Method In Industrial Porting Process

With the continuous transformation and upgrading of the manufacturing industry,the application of intelligent industrial robots has become more extensive,and the technical research in related fields has been paid close attention by more and more domestic and foreign scientific research institutes and high-tech enterprises.As one of the main means of intelligent robot,the combination of visual recognition and detection system and robot technology is also the core technology to realize intelligent sorting and grasping of robot.In this paper,the robot hand-eye system in the process of disordered sorting operation is taken as the research object,and the key technical problems of disordered sorting of workpieces in logistics enterprises are studied in depth.A three-dimensional visual identification and detection system and a software platform are built,and the experimental verification of the whole scheme is completed.Firstly,the hand-eye system based on Kinect 2.0 depth camera and ABB manipulator arm was constructed,and the calibration method of the hand-eye system of the manipulator arm with the eyes outside the hand was studied.The calibration of Kinect 2.0 camera and the hand-eye system was realized,and the internal and external parameters of the camera were obtained,and the transformation relationship between the camera coordinate system and the robot arm base coordinate system was pushed down.Secondly,an improved point cloud pretreatment method is proposed.The depth image of scattered stacked workpiece scene in unstructured environment was obtained by Kinect2.0 camera,and the up-sampling filtering processing was carried out on the depth image by combining the joint bilateral filtering algorithm.The holes in the depth image were filled,and the cloud data of scenic spots in unstructured environment was recovered according to the imaging principle of depth camera.By sampling the vertices of the triangular facets of the CAD model in STL format,the model point cloud data is obtained.K-D tree is used to construct spatial topological relationship of disordered point cloud data.The straight-through filtering algorithm is used to remove the background plane,the statistical filtering algorithm is used to remove the outlier noise in the point cloud data,and the voxel grid filtering algorithm is used to complete the de-sampling processing of the point cloud data.Based on the principal component analysis algorithm,the normal vector information of point cloud data surface is solved,and the obtained normal vector is redirected.Thirdly,a point cloud recognition and location algorithm combining global feature description and local feature description is proposed.The point pair feature descriptor(PPF)of the point cloud data of the model was calculated and stored in the Hash table,and the scene local coordinate system was established.The candidate pose of the workpiece was obtained by using the voting scheme based on the generalized Hough voting principle.DBSCAN clustering algorithm was used for pose clustering for many candidate poses,and the initial pose transformation relationship was obtained.Finally,the iterative nearest point algorithm was used to optimize the pose transformation matrix,and the final pose of the target point cloud was deduced.Finally,based on the Visual Studio 2017 compiler and the configuration of PCL1.8.1and Open CV4.2.0 open source package,the application software system under the Windows10 system is developed,and the human-computer interaction interface is built based on the Qt plug-in,and the experimental verification of the method proposed in this paper is completed.The experimental results show that the algorithm in this paper is effective in solving the problem of workpiece identification and location in unstructured environment,and it can solve the problem of actual workpiece sorting.