Research On Moving Object Grasp By Robot Based On Machine Vision

Grasping the object is the basic task that should be fulfilled for theindustrial robot working on the product line. Currently, the most commonwork pattern of the robot is the teach-play mode. However, there are manydrawbacks of the method, such as the complexity of the teach-play modeprocess, highly structural work environments requirement and hystereticnature of response to the task changes on the production line. In this paper, themachine vision technology and robot technology were used to upgrade thetraditional industrial robot, improving the work pattern by the informationacquire from the vision system to make better adapting to the requirements ofthe modern manufacture.The theses mainly focus on the following aspects:The molding method of the system and forward&inverse kinematicsolving method are analyzed, the formula to calculate the joint angle bydirectly passing the hand pose is provided, which paves way for the following robot control research. After the analysis of the existing research achievemtnts,the end point open-loop object grasping system is built, and the workingprinciple is deduced. Two basic tasks of the object grasping job are confirmed:target location and robot control which should be fulfilled in the objectgrasping process, the two tasks clarify the ideas of primary research for thesubsequent chapters. According to the features and requirements of the task inthis thesis, the principle of visual system construction and hardware selectionare discussed. The camera calibration process and the structure of the imagingsystem are analysed in detail, conducts the comparison and experiments ofimage segmentation, object classification, image matching, corner detection,object localization etc image processing algorithm during the targetlocalization procedure. The deformable template matching method is used toderive the3D pose of the object. The results prove Sojka operator has the bestcorner detection effect, shape-based deformable template matching is the bestmethod for monocular fixed vision system, the pyramid searching operatorimproves the matching calculation speed significantly.Through the pose transform relationship between the coordinates, thepose for robot grasping was obtained. According to the feature of the task, thetrajectory planning in Cartesian space is fulfilled. Based on the shape of thetarget and the structural style of the end-effector, the selection method of theoptimal grasp point is solved. Utilizing serial interface established thecommunication between computer and robot control cabinet. Combined with the function of MOTOMAN robot, two frequently-used control methods arediscussed. Finally the individual function derives before are integrated, therobot grasping program based on machine vision is developed in visual C++6.0environment, realizing the two basic functions proposed before. Throughexperiments prove the correctness and reliability of the theory, lays a solidfoundation to improve the efficiency and flexibility of the automaticproduction line.