Research On Parts Recognition And Pose Estimation Technology Based On Three-Dimensional Point Cloud

In the industrial automation assembly process,the robot needs to grab different types of parts.The traditional way is to fix the parts and then program the robot by manual teaching.This method is less efficient because the teaching needs to be redone after the change of working position.In this case,a method of parts recognition and pose estimation based on three-dimensional point cloud is proposed.Besides,the three-dimensional point cloud vision software is developed,which improves the intelligence level of parts grabbing in assembly process and reduces the cost of enterprises.In this paper,the depth camera is selected at first,and then an image acquisition system is built.After the camera is calibrated,the acquired depth image of the parts can be converted into a point cloud image.Due to the existence of noise and interference,it is necessary to preprocess the collected point cloud.The pre-processing process and parameters are determined,the conditional filtering,voxelgird filtering and radius filtering are used to filter and finally the preprocessed point cloud is obtained.The preprocessed point cloud data needs to be segmented.This paper determines the segmentation algorithm and parameters.First,the RANSC algorithm is used to separate the point cloud of parts and platform,and then use the point cloud boundary based european clustering algorithm clusters the segmented point clouds to obtain the clustering point cloud of each part.After each part cluster point cloud was segmented,this paper analyzes the point cloud feature extraction method,selects the FPFH feature of the part,and then gives the recognition algorithm based on the implicit shape model by using the CAD model of the part.For the pose estimation of parts,the point cloud of model and the true parts are registered to obtain the transformation matrix.This paper has studied the registration errors and running time after the coarse registration of SAC-IA or Super4 PCS and the precise registration of ICP.After analysis,the algorithm using Super4 PCS coarse registration and ICP precise registration is more accurate and less time consuming.Finally,the application software of point cloud recognition and pose estimation was designed,and the experiments of component recognition and pose estimation were carried out.The feasibility of the scheme and the functionality of the software were verified.It has been proved that the software meets the application requirements of actual industry in recognition accuracy,pose estimation accuracy and real-time performance.