Research On Detection System Of Flat Glass Defects Based On Monocular Vision

In production and life,the defects of glass not only affect the appearance of product,but also reduce the quality of the mechanical strength or other indexes,which brings potential safety hazards.Therefore,it’s an essential part that inspecting the quality of glass products.Traditional method that manual inspection is labor-intensive.Its judgment standards are subjective,and inspection effects are unstable.Additionally,defect data is difficult to record,which impedes persons improving production and manufacturing processes by using the original data.In other hand,efficient defect detection systems are not yet popular,and currently there are many shortcomings in the detection systems.In this paper,on the basis of that consulting domestic and foreign related documents on glass defect detection and identification,and that comparing the detection systems that are currently used in the industry,defect like bubbles,stains,white spots,chipping,scratches,and cracks are detected by machine vision.The related preprocessing and defect recognition and classification algorithms have been studied deeply.The design of the glass defect detection and grasping system based on monocular vision has been completed.The main research contents of this paper are as follows:Firstly,according to the national standards and the actual needs of industrial production,the overall functional requirements and schemes of the system are drawn up,and specific judgment standards are given for the various defects detected in this article,so as to complete the design and selection of hardware system including industrial cameras,lighting schemes,etc.Secondly,in the preprocessing stage,in order to reduce the dependence on manual experience and improve the level of automation,the traditional Canny detection algorithm was modified by the Otsu method.Because that the contrast between the light glass and the background is small and weak,and that affected quite by fine dust,there are still some problems such as redundant contour of dust and some edge disconnection when extracting edges.A morphological operator is proposed to process edge images to further improve the quality of image preprocessing.Thirdly,the feature vector composed of geometric feature parameters and global gray information is used as the basis for defect recognition and classification.By calculating the circularity,aspect ratio,elongation,centroid position,Euler number,gray mean value,variance,etc.and analyzing the commonality and characteristics of various defects,a high-efficiency and strong-reliability classification algorithm is designed;The transformation matrix obtained by hand-eye calibration realizes the coordinate transformation from the pixel coordinate system to the robot coordinate system,which realizes the accurate grasping of defective workpieces.Finally,an experimental platform according to the existing conditions and functional requirements is built.The design and programming of the control software is completed by using MFC.Defect identification and grasping experiments of glass workpieces were carried out through the experimental platform,which verified the effectiveness and practicability of the system.