Study On Robotic Dynamical Tracking And Grasping Technology Based On Visual Servoing

Target grasping is one of the most typical applications of robot.Tracking and grasping of random moving targets is the most challenging subject in robot target grasping,which can significantly improve the efficiency and flexibility of robotic operations,and ability of human-robot or multi-robot cooperation,with broad application prospects.Vision is one of the most important sensing technologies of robots.Visual servo technology combined with machine vision and robotic control provides a feasible solution for tracking and grasping of random moving targets.However,current visual servoing technology still has obvious shortcomings in the tracking and grasping of random moving targets.The core problem lies in the lack of real-time and robustness of the tracking and grasping algorithm for random moving targets.Based on visual servoing technology,this paper further studies the tracking and grasping of random moving targets under visual guidance,and improves the tracking algorithm,and the tracking and grasping control algorithm of the robot.The main research contents include:The study of visual tracking algorithm based on improved geometric particle filter improves the real-time and robustness of dynamic tracking.Firstly,the feature points are simplified by edge extraction and morphological expansion,so the edges and their surrounding pixels are obtained and used as feature points.Secondly,the original gray value feature is replaced with the HSV color feature.Finally,the target features are updated at a certain rate after each accurate tracking.Through the test of the OTB2015 data set,the tracking speed of the improved algorithm has been increased from 45.68 fps to 52.71 fps,and the position error has been increased from 0.61 to 0.74.The pose of the moving object in the three-dimensional space is reconstructed,so that the result of the visual tracking of the improved geometric particle filter can be used for the servo control of the robot.The visual tracking results are reconstructed in three dimensions using both color images and depth images of RGB-D camera.In the process of reconstruction,in order to improve the calculation speed of the target pose,the entire point cloud information is not reconstructed,but the depth information of a small number of points on the target is selected in a targeted manner,and the pose of the target is obtained by SVD decomposition.The dynamic tracking and grabbing strategy of the robot is "tracking in far and grasping in close",which improved the speed of robot tracking and grasping.The tracking process uses a position based visual servoing(PBVS).The pose of the target in the PBVS is predicted and offset,which reduces the lag of the robot motion during the tracking process,and reduces the possibility that the robot body obscures the target during the tracking process.A robotic visual tracking and grasping experimental platform was built.Three kinds of tracking and grasping experiments were carried out,which includes: one-dimensional linear motion,two-dimensional random motion and three-dimensional random motion.The experimental results show that the proposed moving target tracking and grabbing algorithm based on improved geometric particle filter has good real-time and robustness.The maximum translation speed of the three-dimensional random moving target arrived 0.25 m/s,and the maximum rotation speed arrived 70°/s.