Research On Robot Classification Grasping System Based On Deep Learning

Currently in an era of intelligence,the basic research of intelligent robots and their use in life and production have been rapidly improved in recent years.The classification and grasping of robots has always been a key research object at home and abroad,and the powerful feature expression ability of deep learning can provide effective support for the development of target detection.has important meaning.This paper discusses the current popular convolutional neural network algorithms and selects the best performance algorithm for improvement.Taking the REALSENCE D435 i camera as the main visual sensor,the recognition and positioning of target objects are studied,and the conversion between twodimensional images and three-dimensional space is found.relationship,and establish a robot classification and grasping system.The specific research contents of this paper are as follows:(1)In-depth study of the R-CNN,Faster R-CNN,Mask R-CNN and Efficient Det network models,improved the Efficient Det algorithm,and established a data set for training.The experiments show that the improved Efficient Det algorithm has better performance,and finally select the improved Efficient Det algorithm to complete the detection of the target object.(2)For the sake of obtain the transformational relation with the plane position information of the image and the actual spatial position information,this paper aimed at the camera to complete the corresponding calibration test,calculates the corresponding between the pixel and camera coordinate system transformation matrix,and completes the color chart with the depth map matching,again through the camera and robot hand-eye calibration to acquire the coordinate system transformation matrix between the two.(3)An experimental platform was built by Fanuc robot and REALSENCE D435 i camera,the kinematics of the robot was analyzed,and the classification and grasping experiments of different objects were completed for many times,and the results of the experiments were analyzed to verify the system scheme and feasibility of the method.