Research On Pose Clustering And Defect Detection Technology Of Industrial Products Based On Machine Vision

It is a general trend for intelligent machines to replace human beings and improve production efficiency.At present,the orderly grasping system for industrial products has more mature solutions.Meanwhile,people are also studying low-cost,high-efficiency schemes for grasping workpieces disorderly.Obtaining accurate position and orientation information of workpieces is one of the key research topics in the disordered grasping work of manipulators.At the same time,the intelligent grasping and placement of workpieces can make subsequent quality inspection such as defect detection and size measurement more efficient and convenient.At present,the pose estimation of the target is more focused on using the deep neural network method,which is mostly carried out in the laboratory environment.But it is still difficult to achieve a wide range of applications in the unstable industrial environment.In addition,the position and pose of the workpiece can be solved more accurately by using 3D structured light camera and other equipment to obtain depth information.However,it is expensive to equip each pipeline with a structured light camera,and it is difficult to meet the requirements of real-time processing multi-dimensional information.Therefore,to explore a simple and low-cost method becomes the urgent need of intelligent manufacturing industry.The thesis proposes a method to build a pose template library offline by combining the point cloud and RGB images of the workpiece based on machine vision technology to solve the above problems.The method lays a foundation for obtaining the pose of the workpiece by using image registration technology in real time.At the same time,the thesis also studies the defect detection technology of related workpieces.The main work of the thesis is as follows:(1)A method of constructing position and attitude template library based on target point cloud and RGB image is proposed.Firstly,under the premise that the images presented by different perspectives of the workpiece itself are quite different,the images with similar postures are divided into one class by using the idea of clustering.At the same time,only one3 D structured light camera is used to obtain the attitude parameters of the cluster center of the RGB image marked by the point cloud of the workpiece.According to the imaging principle,the distance between the target and the camera in the template image is used to estimate the position.Finally,the establishment of posture template library is completed.(2)In order to express the attitude information of the workpiece,a feature extraction method of centroid gray-weighted curve based on boundary marker graph is proposed in the process of attitude clustering.In order to improve the clustering efficiency,the simple clustering method based on similarity threshold and minimum distance principle is improved,and a twostage simple clustering method is proposed.In the process of calculating similarity measure between feature vectors,an algorithm optimization method from coarse to fine is proposed to effectively improve the efficiency of the algorithm.(3)A method of marking attitude by establishing spherical coordinate system of target point cloud is proposed.Firstly,a 3D structured camera is used to obtain the point cloud of the piston to establish a spherical coordinate system.Secondly,the projection of point cloud in any visual angle is automatically saved,and the parameters of each visual angle are read as attitude information.Then all the clustering centers of RGB images are contour matched with 2D projection.Finally,the pose parameters corresponding to RGB images are determined,and the construction of pose template library is completed.(4)Aiming at the defects of missing coating of ceramic filter and the defects of missing edges and protruding sides of circular ceramic filter,a detection method combining machine vision and 3σ criterion is proposed.This method uses traditional image processing algorithm to extract corresponding features by analyzing the differences between defective images and normal images,and uses the idea of outlier detection to detect defects.This method is simple and efficient,and skillfully solves the detection problem caused by the small number of defect samples.The idea of outlier detection is also applied to the detection method of tire belt edge folding defects.In this thesis,a method of automatically establishing posture template library is proposed.The established pose template library marks the distance and pose information of the template,which lays a foundation for obtaining the pose of the workpiece by using image registration technology in real time.At the same time,it solves the intelligent detection of various defects of related workpieces in the research of the research group.