Research On Visual Inspection System For Hole Shaft Assembly Robot

With the development of computer technology,the advantages of machine vision technology in the industrial robot assembling more and more significant.It is an important trend of intelligent development at the present stage to integrate machine vision into the robot production line of various kinds of hardware cookers,so that it can replace human to complete tasks in the process of assembly.Therefore,this paper focuses on the key technologies of robot system application of the axle hole components in a variety of hardware cookers.First of all,select appropriate hardware equipment to build EIH hole shaft assembly vision detection system.According to the imaging model of pinhole camera,the model of RGB-D depth camera is established,and the fusion principle of depth information and color information is expounded.The visual system calibration,including camera calibration based on Zhang Zhengyou method and EIH system hand eye calibration.Through matlab toolbox to complete the camera calibration error within the 0.4 pixel,confirmed that the calibration of camera parameter is effective.Secondly,a target recognition and location method based on Mask R-CNN model is proposed.Assembly components of the data set is made by a separate,to Mask R-CNN model training,combined with the depth information and color information fusion experiments,using the Mask R-CNN model Mask combined with depth information for workpiece center position.The rapidity and accuracy of the algorithm are verified by experiments.Again,in order to improve the positioning precision and efficiency of fetching,for color images using traditional image processing methods of gray level,the filtering process is used to estimate workpiece shaft neck measurement and grasping posture.An image contrast enhancement algorithm with Gamma correction combined with pixel intensity improvement is used to optimize image contrast adaptively.In order to obtain more accurate edge information,an adaptive Zernike moment sub-pixel edge detection method based on region division is proposed.After coarse location is carried out by Canny edge detection,fine location is carried out by this algorithm.Finally,the Angle and basic dimensions of the workpiece are determined by the least square linear fitting.Finally,the assembly hole location is recognized and located by ellipse detection in traditional image processing.The advantages and disadvantages of ellipse detection based on random Hough transform are analyzed.An ellipse recognition method based on arc edge clustering was proposed.A series of processes,such as image preprocessing,edge connection,arc clustering,and least square fitting,were used to realize rapid and accurate ellipse detection.Then the center of the circle was determined,and the alignment of the axis and shaft was carried out.