Research On Multi-target Recognition And Positioning Based On Edge Detection And Feature Combination

With the continuous in-depth reform of China’s intelligent manufacturing,industrial production has gradually developed from automation to intelligent and flexible,and the fusion of visual technology and industrial robots has become a popular research direction in intelligent transformation.At present,most of the domestic researches on specific target workpieces are placed on the same horizontal plane for gripping research.Therefore,it is extremely important to study the multiple stacking target picking and sorting problems that are common in industrial production and manufacturing processes.Binocular vision technology,as one of the important means for robots to perceive the external environment information,can obtain the three-dimensional depth information of the target,effectively solve the identification and positioning of multiple scattered stacked targets,improve the efficiency of the enterprise,simplify the production and feeding system,and improve the production line Flexibility promotes the development of industrial production towards intelligence and flexibility.Therefore,it is of great theoretical value and practical significance to research the robot based on binocular vision to grab multiple scattered and stacked workpieces.First,the robot binocular vision system modeling and hand-eye relationship calibration were studied.The nonlinear distortion correction was performed on the basis of the linear pinhole imaging model.The parallel binocular vision system was established,and the camera perspective transformation matrix calibration method was selected to solve The internal and external parameters of the camera;according to the eye-to-hand model,a fast hand-eye calibration method is proposed,and the effectiveness of the method is verified through experiments.Secondly,the edge feature detection and fitting of multiple stacked targets are studied,and a hybrid algorithm combining gamma transformation and median filtering is used to preprocess the image,which can eliminate the main noise and accurately identify the image edge feature information;On this basis,the ellipse fitting of the edge contour is carried out by the improved random Hough transform ellipse fitting algorithm,which improves the integrity of the edge fitting and the real-time performance of the algorithm.Then,for the classification and positioning of multiple stacked targets,the principle of support vector machine classification and recognition is studied,and a classification and recognition method is proposed that combines the global feature Hu geometric invariant moment and the local feature direction gradient histogram.The two features are trained separately to obtain the best recognition rate,and used as the weight coefficient of the combined classifier.Through multiple experiments,it has been shown that the recognition accuracy based on feature combination has less fluctuation and higher classification robustness.Finally,according to the actual application requirements of this article,the overall scheme design,the binocular vision system experiment platform was built,and the positioning grab test software was developed to perform experimental verification and error analysis on the positioning accuracy of the system,and to verify through the workpiece grab experiment The reliability of the system.