Research On Application Of Object Recognition And Robot Grasp Based On Vision

The continuous progress of robot technology has greatly expanded the application scenarios of intelligent robots in real life.Robot visual perception ability is an important embodiment of robot automation and intelligence,which has a broad development prospect.Visual sensors and other devices can make the robot perceive the environment and pave the way for subsequent autonomous and intelligent decision-making.However,the existing algorithms can not meet the requirements of practical application in terms of detection accuracy and speed when in complex environment or multi-object accumulation.This paper mainly studies the robot visual perception algorithm.Firstly,it studies the special object recognition,and on this basis,we integrate the object recognition algorithm with the grasp detection algorithm.The main content is as follows:1.Aiming at the problems of incomplete types of special object and simple background of images in the existing dataset,a new dataset of special object is made,which includes five kinds of common special object in total.The background of pictures includes various real environments,indoors and outdoors.M-YOLO model is used to identify and detect datasets,and lightweight network is used to extract features,and features of different levels are fused to improve the detection ability of objects of different sizes.2.Aiming at the defects of existing robot grasping algorithms that cannot recognize objects at the same time and low detection accuracy,using a multi-task grasping detection network to realize the fusion of object recognition and grasping detection.The regional proposal network is used to separate objects from input.In the grasp detection branch,we use the channel attention mechanism to enhance the feature extraction ability,optimize the decoding mode of the grasp rotation boxes,and the regression loss function is improved to enhance the detection ability of the network in multi-object environment.3.In the ROS environment,based on Kinova Gen3 robotic arm,Kinect V2 camera and other devices,the robot grasping interactive system is built.Combined with the speech recognition and hand detection technology,the robot arm could automatically grasp the target and place the object into the hand when the voice command was input,so as to improve the intelligent level of the robot.