Research On Automatic Recognition And Positioning Of Parts In Intelligent Coordinate Measurement

The coordinate measuring machine(CMM)has the absolute advantage over the detection precision and the measurement space range,so it is widely used in the industrial detection environment.The non-contact detection of products with CMM and visual inspection system has become a new trend of intelligent detection.Considering that machine vision for the three coordinate measuring machine detection system is expensive,complex structure.A three-dimensional vision detection solution based on Kinect sensor is proposed.The visual system is used to quickly identify the shape of the part and its position on the measurement platform,and the conversion among the machine coordinate system,the part coordinate system,and the camera coordinate system is completed,which establishes the foundation for programming the inspection path.First of all,aiming at the scene 3D point cloud collected by Kinect,the points beyond the user's setting range are removed by the Pass-Through filter,and the point cloud density is reduced by the Voxel-Grid filter.Then the filtered scene cloud is segmented,in terms of segmentation integrity and real-time performance,the segmentation effects and parameter settings of the different segmentation algorithms are compared.Experiments show that the target point cloud can be effectively segmented by the random sampling consistency segmentation algorithm combined with the Euclidean cluster segmentation algorithm.Secondly,the VFH feature is extracted from the target point cloud after the segmentation.In the same way,the VFH features of the model point clouds from different perspectives are extracted,and the off-line database is established.The segmented target point cloud is used for searching for off-line database by K-nearest neighbor search,and the chi square distance between the point cloud and the target point cloud in the neighborhood is calculated and sorted.The model point cloud with the smallest chi-square distance is the most similar model.The matching accuracy between the most similar model and the target point cloud is further improved by 3D-NDT registration,combined with the external parameter calibration,then the precise position and pose estimation of the parts under the CMM can be achieved.In the end,the position and pose estimation of the parts system based on Qt is developed,The functions in this system such as point cloud data acquisition andvisualization,preprocessing,model training,feature calculation,segmentation,object recognition and location are implemented.In order to verify the performance of the system,recognition and location experiments for different parts have been carried out.The experimental results show that the parts can be recognized and located accurately in a short time under the coordinate measuring machine.