Research On Grabbing Method Of Moving Objects Based On Robot Hand-eye Coordination

With the introduction of "Intelligent Manufacturing 2025",robots and machine vision have gradually become one of the hotspots of research at home and abroad.In recent years,due to the substantial improvement of computer computing capabilities,deep learning has a broad development space.How to integrate deep learning and machine vision into the field of robots,and to increase the level of industrial production intelligence are important for improving China's technological strength and accelerating China's industrial upgrading significance.This paper combines deep learning technology,robot technology,and machine vision technology to conduct research on the intelligent grasping of moving targets on industrial production lines,constructs an intelligent grasping system,proposes an improved target tracking strategy,and carries out object motion poses.It is estimated that it has certain reference value for robot dynamic crawling.The article mainly discusses the following points.1)First,the development status of target tracking and robot vision grabbing are introduced.In-depth analysis of target recognition and positioning,target tracking,robot grabbing.A set of modular dynamic target tracking and grabbing system that integrates target positioning,target tracking,robot control and robot grabbing strategy is built.2)Secondly,the coordinate conversion model of camera-robot-moving target was established,the forward and reverse kinematics of the robot arm were analyzed,and the target detection algorithm of SSD(Single Shot Multibox Detector)was used to locate the target object to verify the coordinate conversion The accuracy of the module.3)Research the target tracking technology,use stepless speed regulation to control the conveyor belt,and calibrate the conveyor belt to calculate the position of the robot grabbing.This paper proposes a target updating strategy based on ATOM(Accurate Tracking by Overlap Maximization)target tracking algorithm improved fusion time decay function.By comparing with other excellent algorithms in the data set and performance testing,the innovation of the improved algorithm is effectively verified.4)In the research of grasping attitude,an attitude estimation network(Atnet Estimation network,AEnet)is proposed,which uses the image features that have been extracted in the tracking stage to predict the grasping labels,and converts the grasping position labels into category labels for analysis.Simplified the complexity of crawling.Finally,the experiments show that the ideas and methods designed in this paper can improve the intelligence of the robot to dynamically grasp the target,and also verify the advanced nature of the algorithm and the stability of the system.