| With the development of industrial intelligence,the research of machine vision is on the rise.The vision sensor can guide the robot to complete some designated work,which can improve the work efficiency and reduce the cost.However,round feature objects such as bottle caps,liquid tank holes,bolt holes and so on can be seen everywhere.In the industrial field,it is necessary to determine the position of the circular target before automatic processing and application can be realized.The traditional circular target positioning method is too flexible and only suitable for fixed occasions.This dissertation takes the nuclear power maintenance robot to identify and locate circular bolt holes as the research background,and conducts research on the recognition and positioning of circular targets by vision-based robots.The main work is as follows:The accuracy and working environment requirements of the three positioning methods of monocular visual plane positioning,monocular visual three-dimensional positioning and binocular visual stereoscopic positioning are analyzed and compared.Combined with the needs of the subject,the monocular visual plane positioning method is selected to determine the overall route scheme of the research on the recognition and positioning of circular objects for the nuclear power robot based on vision.The advantages and disadvantages of three kinds of camera calibration methods are analyzed and compared.Zhang zhengyou calibration method with high precision and low operation requirements is adopted to carry out camera calibration experiment by MATLAB software.The reprojection error is 0.1777 and the camera internal parameter matrix meets the requirements.The Eye-in-hand calibration relationship was derived in detail,and the Tasi two-step method was used to collect calibration images by changing the robot’s pose and fixing the calibration plate.The rotation matrix R and translation vector T in the hand-eye relationship were solved,and the position transformation matrix between camera and robot end was determined.The weighted method is used for image graying,adaptive mean filtering and field mean binarization to highlight the circular target,remove the noise,and reduce the calculation amount of the target positioning algorithm.This paper focuses on the recognition and positioning algorithm of circular feature targets,and proposes a circle detection algorithm based on pixel ratio.By comparing with the traditional Hough method,it is proved that the algorithm is simple,effective and can accurately locate the center of the target circle.A vision-based robot grabbing experimental platform is built,and MFC programs that integrate some software operation interfaces such as cameras,robots and graphics are established.According to the z-value compensation,three kinds of round bottle caps were used as grasping objects,and the success rate of the experimental grasping was 92.2%,which verifies that the vision-based robot could achieve the positioning of the circular target. |
