Research On Target Recognition And Grasping Strategy Based On Machine Vision

In today's society,human beings are more and more inseparable from intelligent robots,and the enthusiasm for researching the grasping methods of robotic arms with higher intelligence has been high.In industrial production,it is inevitable to encounter situations such as random placement of workpieces,overlapping,disorganization and so on.Those traditional visual recognition technologies cannot accurately identify workpieces in these environments,so a visual recognition method in complex environments has been developed.It is extremely important.This paper first studies the advantages and disadvantages of current hot target detection algorithms in visual recognition;and proposes a new object recognition method based on the combination of feature point algorithm and RGB color space;finally,combines the robotic arm to design a robot grasping system.It can judge the position of the object to be grasped by itself,and realize the grasping and placing action.The main work includes: target detection experiments based on deep learning,kinematics modeling and simulation experiments of robotic arms,target pose detection based on monocular vision,platform construction and practical inspection of the grasping robot system,etc.Target detection experiments based on deep learning include: choosing two widely used target detection algorithms.This article chooses YOLOv3 and Faster R-CNN to train the target model,and compares the training and prediction experiments on the same data set.,Compare the detection success rate and training speed of the two algorithms,analyze their respective pros and cons,and make improvements,Choose the target detection algorithm suitable for industrial production.The kinematic modeling and simulation experiments of the robotic arm include:drawing a six-degree-of-freedom robot model through 3D drawing software,importing it into the simulation software for kinematic modeling and forward and inverse kinematics solution,using 3D visualization tools to draw the robot The change curve of each joint angle under straight-line trajectory planning and arc trajectory planning,and its advantages and disadvantages in the two motion states are analyzed.The posture detection experiment of monocular vision includes: comparing the advantages and disadvantages of the two calibration methods of eye-on-hand and eye-outof-hand,selecting the camera calibration method with eye on the hand;simultaneously calculating the conversion relationship of various coordinate systems;adopting Tasi's RAC calibration method performs camera calibration;experiments have verified the accuracy of the pose calculation based on monocular vision.The construction and practical inspection of the grasping robot system platform include:building the robotic arm hardware equipment and software operating platform,and completing the practical verification of its effectiveness in object recognition,positioning,grasping,and placement compared with traditional grasping methods.This experiment provides a theoretical basis for the application of the target detection algorithm in industrial production.The robot arm grasping based on the target detection algorithm consumes less cost and has strong environmental adaptability.At the same time,the success rate of the robot arm grasping is also higher.