Research On Bin-picking System Of Industrial Robot Based On Binocular Vision For Stamping Parts

In recent years,with the popularization of the concept of "industry 4.0" in the world,a series of policies are put forward in China,such as "intelligent manufacturing industry development plan",to improve the level of industrialization.It means that traditional industrial robots will face unprecedented intelligent upgrading.And vision guidance is an important method to improve the intelligent level of robot.Comparing with the current technology,binocular vision guidance is the system closest to human,which is very prominent in the location and pick-up of workpieces.However,the main technical problems are the stereo matching of left and right images and the identification and positioning of workpieces in the process of picking up by the robot.In order to solve these problems,this thesis mainly focuses on the research of realtime stereo matching,identification and positioning algorithms.Also,a bin-picking system of industrial robot is developed based on binocular vision for scattered stamping parts.The main work is described as follows.1.Preprocessing of calibration images is studied via the experiment.By using calibration method of Zhang Zhengyou,the two calibration programs are compared,which are implemented with Matlab and opencv.And the optimal calibration results are obtained.Finally,the relationship of robot coordinate system and camera coordinate system is calibrated.2.A binocular vision system is established to meet the demands of the project.Through the analysis of the principle of binocular vision,the placement of the camera and the principle of calculating depth and mathematical model are determined.Then,infrared LED light and laser are utilized as the light sources,which is verified by the experiment.Combined with the characteristics of the laser,a new algorithm combining SGBM and improved RANSAC is proposed to realize a two-step matching scheme,which is verified by comparative experiments.Finally,the point cloud of the workpiece is carried out according to the parallax method.3.The scheme of workpiece identification and location system is constructed based on template matching method,which establishes the process of offline target workpiece template library and online matching and picking.Some preprocessing methods for initial point cloud are studied,including direct-through filter,statistical filter and radius filter.Then the Euclidean distance clustering algorithm is used to segment the point cloud stack.Compared with ISS,SIFT 3D and Harris 3D,the ISS algorithm has the best effect for extracting the key points.The FPFH feature is used to describe them.According to the FPFH feature description of the key points,SAC-IA and ICP algorithms are used to complete the rough pose estimation and the accurate pose estimation respectively.Finally,a reasonable matching calculation method is set up to finish the recognition and positioning work.4.The overall framework of this bin-picking system is designed by the actual needs and the characteristics of workpiece.The hardware and software of this system is also designed.Finally,the intensive experiments are conducted in the actual scenes,which demonstrate the feasibility of the system.Also,the shortcomings of the system are analyzed in the end of the thesis.