Research On Visual Grasping Technology Of Flexible Gripper Based On Deep Learning

With the rapid development of national industrial strength,robotic grasping technology plays an indispensable role in industrial production.At present,grasping technology in industry is based on a set process plus the cooperation of traditional manipulator claws to realize objects.This kind of gripping technology is only a fixed and repeated gripping of a single part,and does not have the characteristics of being widely used.Therefore,the grasping technology based on deep learning has received extensive attention from domestic and foreign researchers.In order to solve many different types of object recognition and grasping problems in the field of industrial production,this paper conducts research on the flexible gripping technology based on deep learning.The main contents of this paper are as follows:First,the Faster R-CNN object detection model is introduced in detail,including the feature extraction network,the RPN region suggestion network,and the anchor frame of Anchors.In view of the large amount of calculation parameters of the network,model redundancy,and slow detection speed,this thesis improves and optimizes the backbone extraction network and NMS algorithm of the Faster R-CNN network model.This paper uses a depth camera to obtain 400 pictures of the object being grasped,and enhances and expands each picture.Use Label Img labeling tool to label pictures to create their own data set.The improved detection model is trained based on the created training data set,and finally an object detection model with higher accuracy and faster speed is obtained.Secondly,introduce the kinect depth camera.Use Matlab to perform camera calibration and hand-eye calibration to obtain the mapping relationship between the pixel coordinate system and the base coordinates of the robotic arm.Finally,the space coordinate conversion relationship between the robot arm and the object is obtained,so that the positioning of the target object is completed more accurately,and the flexible gripper can perform the subsequent grasping operation.Then,the platform control system,the end flexible gripper,and the mechanical structure of the robotic arm are introduced.Then,the DH parameter method is used to establish the linkage coordinate system of the robotic arm,and the forward kinematics model and inverse kinematics model of the robotic arm are carried out.Solve.Finally,the flexible gripper can accurately reach the position of the target object and complete the grasping task of the target object quickly and accurately.Finally,the results of the object detection experiment and the object grabbing experiment are analyzed and researched.The experimental results verify the validity and applicability of the research content in this paper.