Research And System Development Of Target Recognition And Grasping Technology Of Visual Palletizing Robot

Since the 21st century,the global artificial intelligence industry has developed faster and faster,so the development of automation and intelligent robots has become more and more rapid.With the development of computer technology and deep learning algorithm,the visual industrial robot has more advantages than the traditional industrial palletizing robot.It can accurately locate the unknown target object,dynamically adjust its own joint rotation,and control the mechanical gripper according to the position and attitude of the object,so as to accurately identify and grasp the unknown object in complex environment.Therefore,exploring the visual grasping technology combining robot and human eye is of great significance to improve the automation and intelligence of palletizing robot.This thesis mainly discusses the palletizing robot with vision,the accurate recognition and grasping pose prediction of various types of irregular objects.Based on previous academic research,with the help of advanced deep learning algorithm,this thesis carries out the research content of this subject.Aiming at the problem of robot autonomous recognition,the mathematical model of binocular vision system is established,and the camera is calibrated by Zhang Zhengyou calibration method to correct the distortion;The target recognition method in complex environment is studied,the image detection data set is created,and the boundary filling method is used for image enhancement and gray preprocessing;The application of image detection based on deep neural network is studied,SSD separation convolution technology and YOLOv3 target recognition method are adopted.Aiming at the problem of low accuracy of recognition and detection task,YOLOv3-ours algorithm is proposed,and the decoder(PDM)module is introduced to complete the optimal design of target recognition algorithm.Compared with the original yolov3 algorithm,the target detection accuracy is improved by 3.7%;The method of grasping attitude prediction of various types of unknown target objects is studied.The three-dimensional grasping prediction method is combined with RGB to label the data set and predict the grasping rectangular frame;Aiming at the accuracy of grasping attitude prediction,a five-dimensional prediction method based on Dark Net-53 is established to effectively predict the grasping rectangular frame in the face of unknown irregular objects of different types and sizes;Aiming at the problem that it is difficult to realize high-precision control of visual robot,the controller is optimized by RBF network,and the simulation analysis is carried out based on double manipulator system.In this thesis,a palletizing robot system platform with vision is built,the coordinate transformation is completed by hand eye calibration,the connection between modules is realized through the software platform,and the experiment of palletizing robot identifying and grasping unknown irregular objects is carried out.The research results show that the stacking robot based on binocular vision proposed in this thesis can effectively improve the accuracy and real-time of robot recognition and grasping,and provide a theoretical basis for the follow-up intelligent algorithm research of robot.