Research On Recognition And Orientation Of Object Based On Binocular Vision And Robot Grasping

In recent years,with the development of intelligent manufacturing technology,robots are being applied to various in fields.In order to improve the autonomy and intelligence of the robot,machine vision is used to endow their environmental perception.With the increase of demand,identifying and locating objects with arbitrary pose under complicated environment rapidly and accurately is becoming a hot issue.In this paper,a way of combining stereo vision with robot technology is adopted to solve this problem.Firstly,the imaging model and the distortion model of the camera are analyzed by establishing coordinate system of the visual system.A binocular stereo vision system was built and calibrated using the Zhang calibration method and the result was used to correct the images.An experiment was designed to verify the accuracy of the calibration.Template matching method based on descriptor was used to for recognition,for the poor real-time performance of traditional SIFT based matching,Harris corner detection algorithm was combined with the random tree classification algorithm to transform the matching into classification by describe feature point with categories.Through experiments,the performance of these two algorithms are evaluated,which proved that the method adopted in this paper has better stability and real-time performance.Disparity of two images was calculated after stereo matching,which can be used to reconstruct the feature points with the result of calibration as well.According to the position information of 2D-3D feature points,poses of the objects can be obtained by iterating by Levenberg-Marquardt algorithm with minimum error of re-projection as a standard.The orientation of the object under different conditions was studied and the accuracy was verified through experiments.Finally,a vision-based target recognition and orientation experiment system was built.The orientation result was transformed into robot coordinate system with hand-eye calibration,and joint angles of the robot was calculated with analysis of kinematics.The object was grasped be robot accurately.In this paper,algorithm for recognition and orientation of object with arbitrary pose under complicated background is studied,the corresponding experiment is designed to verify the algorithm.The results show that the proposed method can accurately,quickly and stably achieve the recognition and orientation of objects,which can meet the real-time performance of robot grasping.