Research On Robot Stacking Target Recognition And Location Grasping System

With the increasing modernization of manufacturing process,the popularity and intelligence of industrial robots will be one of the developing trends of manufacturing industry in the future.The target stacking is usually found in the industrial site,and the use of monocular vision technology could not solve the problem of location and capture of stacking targets well.Aiming at the problem of stacking target recognition and location,this paper develops the research of robot stacking target recognition and location.The main research contents are as follows:(1)The robot binocular vision system and hand-eye calibration with robots are researched.A hand-eye relationship calibration method based on binocular stereo vision measurement is proposed.The method controls manipulator with three translational motion and a rotation motion based on binocular measurement and obtaining linear transformation matrix.Establishing hand-eye calibration model and robot binocular vision system measurement model,and the calibration precision and accuracy of binocular vision measurement are analyzed.The optimal parameters of the binocular vision system are obtained through experiments.(2)Comparing the edge detection results of several common edge detection operators,the Sobel operator is selected as the stacking object edge detection method.An improved ellipse fitting algorithm based on random Hough transform is proposed for edge detection,and the edge contour is elliptically fitted by the improved random Hough transform elliptic fitting algorithm.The edge fitting effect is guaranteed,and the real-time performance of the algorithm is improved.(3)Aiming at the problem of localizable and unlocatable target recognition in stacking targets,a classification method of global feature and local feature combination is proposed,which is based on the single feature recognition result and distributes the combined feature weight coefficient.By extracting the global Hu geometric invariant moment feature vector and the local direction histogram gradient feature vector and combining the support vector machine,the classification and recognition of localizable and unlocatable grasping targets are carried out.The results show that the combined features have higher recognition accuracy and better stability.(4)For stacking target pose acquisition,the fitting edge of localizable grab target is reconstructed by binocular vision,the stacking object pose is obtained by spatial circle fitting of spatial point,and the position and pose error is analyzed.The targets and image plane inclination angle are used as the evaluation criterion of grasping priority.(5)A robot stacking target recognition and location grasping system is developed,and an experimental platform for robot stacking target location and capture is built.The experiment platform is used to verify the positioning accuracy of binocular vision system and the robot stacking target location experiment.It is verified that the system can satisfy the recognition and capture of stacked targets.