Research On Recognition And Grasping System Of Industrial Robot Based On 2D And 3D Vision

With the transformation and upgrading of traditional manufacturing industry and the strategy of "made in China 2025",industrial robots are widely applied in industry.But the traditional industrial robot still stays in the simple replacement stage.When the object is changed,the robot needs to be reprogrammed,which reduces the efficiency of the robot.In recent years,the integration of vision and industrial robots is becoming more and more popular.Visual guidance has enhanced the ability of robots to cope with environmental changes.However,traditional object 2D recognition can only locate objects and lack 3D posture information.The traditional 3D point cloud vision needs to deal with a large number of point clouds,with complex algorithm and long computing time.Therefore,this paper proposes a method of object recognition and attitude estimation based on 2D-3D combined vision,and uses the advantages of 2D and 3D vision to complete the guidance of industrial robots and realize the recognition and grasping of target objects.This paper first calibrated the hand eye of the RGBD camera.According to the imaging principle of the camera,the mathematical model of the camera is analyzed,and the alignment between the color map and the depth map is completed.According to the checkerboard calibration method,the camera's internal reference and external parameter matrix are measured,and the transformation between pixel coordinates and robot world coordinates is realized.Then,2D and 3D vision are studied in detail.The whole scheme of finding object area with 2D vision and estimating the pose of 3D point cloud in this area is proposed.In 2D vision,we extract template features and SURF feature points of scene graph respectively,and complete matching.We compute a homography matrix based on matching points.Then the matrix is applied to do the perspective transformation of the four vertices of the template,and the envelop box of the object in the scene map is obtained.In 3D vision,the region in the envelope of depth map is transformed into point cloud,which contains the posture information of objects.Then,registration of template point cloud and scene area point cloud is carried out.Because of the small number and low accuracy of point cloud,ICP algorithm is directly used to register it.Finally,the rotation matrix in the results of the point cloud registration is represented by four elements,and the four elements indicate that the object has the advantages of simple and no universal joint lock,and it can be butted seamlessly with the end of the manipulator.Finally,an industrial robot recognition and grasping system is built based on robot operation system ROS.The URDF model of the scene is created,and the Move It UR5 motion planning configuration of the robot is completed.RRT algorithm is selected as motion planning algorithm.Then the visual module,UR5 motion planning module and Reflex three dexterous hand are respectively packaged into ROS Node,and the whole system is built.Experimental results show that 2D-3D combined visual guided UR5 can accomplish target recognition and grasping very well.