Research On Grasping Method Of Robot Arm Based On Monocular Vision

China is a big manufacturing country.With the development of technology,modern industry demands more intelligence and adaptability.Robot technology is an important part of intelligent manufacturing.The development of visual technology has enabled robots to have more powerful perception capabilities and expanded the scope of their applications.Driven by big data and fast-growing computer technology,deep learning has made great progress.At present,it has made remarkable achievements in target detection.The method of deep learning applying to the control of industrial robots,which is of great importance to the realization of intelligent production and practical production practice.The existing robot vision system is based on the calibration of the coordinate system to identify and locate the position and posture of the object.This topic aims to use the deep learning method to achieve the position recognition of objects in the area of target detection.Then according to the location of the object in the image,we get the grasping information needed by the manipulator.That is,the end-to-end capture of the physical coordinate from the image to the robot arm is achieved.First,the robot arm grabs the data information of the object and forms a data set.Then process the data set;Locating and recognizing objects;then the network model of grasping information needed by the manipulator is designed and trained,and the training parameters of the network model are modified at the same time.F,inally,we test and evaluate the location,recognition and grasping of the object grabbed by the manipulator.In the end,we make the test results visualization and complete the GUI design of detection system.