Research On Robot Grasping System Based On Faster R-CNN Object Detection Technology

In order to solve the problem of identification of various kinds of objects and stacked objects in industrial production,this thesis combines machine vision technology and applies the Faster R-CNN object detection model in the robotic grasping system.According to the color image of the object collected by the Kinect2 camera in different positions and poses,these images are used to train their own Faster R-CNN object detection model,and then the border of the object in the image is obtained according to the Faster R-CNN model,according to the position of the border is detected to obtain the pixel coordinate value and the depth value of the object.Then,according to the calibration result of Kinect2 and the robot system and the homogeneous transformation matrix between the joints of the robot,the three-dimensional coordinates of the object can be obtained.Finally,the robot performs the inverse kinematics solution according to the three-dimensional coordinates of the object to realize the task of grasping the object.This thesis mainly completes the following work: 1.Applying the Faster R-CNN object detection model to the industrial production classification task,it solves the identification problem of various kinds of objects and stacked objects in industrial production,and achieves better recognition result,achieving accurate positioning of the object.The object detection algorithm has better generalization ability.2.Establish the hand-eye calibration system between the Kinect2 camera and the robot and the motion control system of the robot,which can well complete the positioning and grasping tasks of the object.3.Establish a robotic grasp system that integrates the Faster R-CNN object detection system,camera calibration system,and robot motion control system.The collaborative work of the three subsystems can complete the color image and depth image from the Kinect2 camera,identify the object by the Faster R-CNN model and obtain the pixel coordinate value and depth value,and then obtain the three-dimensional coordinates of the object according to the Kinect2 camera calibration system result.Finally,the robot complete the object grasp task.