Research On Disorderly Grasping System Of Robot Arm Based On 3D Vision

With the widespread application of machine vision in the field of industrial robots,industrial robots with visual perception are also entering the era of intelligence,gradually replacing manual labor to become a trend in various industries.In order to improve the problem For the recognition and grasping accuracy of stacked yarn bobbins,a robot arm disorderly grasping system based on 3D vision is proposed.Through 3D vision,the target information of the yarn bobbin in the work scene of the robot arm is obtained,so as to guide the robot arm to complete the grasping of the yarn bobbin.Take-in operation,instead of manual completion of the operation of picking up and transferring the yarn bobbin,it improves the automation and flexibility of the textile process.The main research content of this article is(1)A disorderly grasping system for industrial robotic arms based on 3D vision is designed.The system design plan is proposed,the hardware equipment is analyzed and selected according to the design plan,and the system software plan is designed with the idea of modularization.The MATLAB calibration toolbox is used to complete the calibration of the robotic arm vision system,and the accurate Kinect V2 camera internal and external parameters and the conversion relationship matrix between the camera and the robotic arm are obtained through calibration experiments.(2)The method of acquiring and preprocessing the 3D point cloud data of the bobbin target is studied.According to the depth image of the yarn bobbin obtained by the Kinect V2 depth camera,it is converted into point cloud data with three-dimensional information according to the principle of small hole imaging;the scene point cloud and the box point cloud are eliminated through straight-through filtering and MSAC algorithm fitting plane.Keep only the point cloud data of the yarn bobbin;by comparing the denoising effects of different filtering methods,statistical filtering is selected to denoise the yarn bobbin target.According to the processing results,the method can obtain high-quality point cloud data of the bobbin.(3)Aiming at the problem that the European clustering segmentation algorithm cannot effectively segment the target of disorderly stacked bobbins,a point cloud segmentation algorithm based on improved European clustering is proposed.According to the normal vector characteristics of the target point cloud,the edge points of the contact area between the bobbin targets are eliminated,the Euclidean clustering is used to segment the bobbin point cloud after the edge points are eliminated,and finally the edge points are complemented.According to the segmentation results,the algorithm can accurately segment a single bobbin target in a disorderly stacking scene.(4)The detection problem of the bobbin target is studied.Based on the ISS algorithm,extract and down-sample the key points of the point cloud of the bobbin template and the bobbin to be detected,and then describe the extracted key points through the SHOT feature descriptor;use the RANSAC algorithm to complete the bobbin template and the bobbin to be matched Coarse registration of the point cloud,determine the initial position,and then realize the precise registration of the two through the ICP algorithm,combined with the hand-eye calibration result to determine the precise pose of the identified yarn bobbin.(5)On the basis of hardware,an experimental platform for 3D vision guided robotic arm to grab the target of the bobbin was built,and the experimental verification of the grabbing of the bobbin under different specifications and different scenarios was carried out in the actual environment.Experimental results show that the average working time of the system is within 10 s,and the crawling success rate is above 80%,which is in line with the expected results.