Research On Target Workpiece Recognition And Location Technology Based On Monocular Vision

Robot vision system is an important part of robot system.The recognition and location of objects through vision has always been the focus of research in the field of robots.In the flexible assembly process of the satellite-oriented mobile manipulator,visual-based workpiece identification and location can help the manipulator achieve autonomous and flexible operation,thereby improving the assembly efficiency under complex background.Because of the narrow workspace,large light and scale transformation in satellite assembly process,the current recognition and positioning system has low accuracy and poor positioning accuracy for target workpiece.In this paper,a workpiece recognition method based on improved image feature matching is proposed.On this basis,an improved Grab Cut algorithm is used to locate the workpiece,and then a vision-based workpiece recognition and location system is built.The main contents of this paper are as follows:First,based on the camera model,Zhang’s calibration method is used to calibrate the monocular camera.The eye-in-hand model is analyzed and the coordinate transformation matrix of the eye-in-hand system is deduced based on the least square method.Finally,the calibration error is analyzed through experiments.Secondly,object recognition based on improved image feature matching.To extract and generate descriptors of SURF features in images,an improved recognition algorithm based on SURF feature bag model is proposed to overcome the inefficiency of SURF feature recognition.Firstly,object feature bag model is constructed by K-means clustering algorithm,and then object binary description vector is constructed.Then object matching is carried out based on object description vector,and corresponding points are selected in the correct matching vector to realize object recognition.Finally,the recognition accuracy and real-time performance of the improved algorithm are verified by experiments.Thirdly,on the basis of correctly identifying the target workpiece,the research of workpiece location based on improved Grabcut algorithm is carried out.Aiming at the problems of inefficient execution and defective edge of current Grab Cut segmentation algorithm in foreground segmentation of target workpiece,an improved Grabcut algorithm based on multi-scale image is proposed.Firstly,the foreground is segmented on the highscale image,then the contour of the target image is mapped on the low-scale image.Then the low-scale image is segmented by Grabcut algorithm.Secondly,the contour is detected and optimized on each scale map to improve the accuracy of contour detection.The image invariant moments are used to solve the centroid position of the target workpiece.Finally,the position is converted to the coordinate system of the manipulator with the calibration parameters,and the location of the target workpiece is completed.Finally,the experiment platform consisting of monocular camera,client,control terminal and UR3 manipulator is used to validate the vision recognition and positioning system.Two different target workpieces are selected to verify the validity of the system by comparing the estimated position and the real position of the target workpiece on the basis of target recognition.