Research On The Key Algorithms Of Glass Completeness On-line Inspection And System Implementation

Glass stacking is the last procedure in the process of flat glass manufacturing. Usually,before a stacker snatches glass cut plates, the completeness of these plates should be inspected.Glass completeness inspection is to inspect the surface quality of glass cut plates, whichincludes size, shape, edge breach, level mark and so on. Presently, most native glassmanufacturers still adopt traditional manual inspection pattern, that is one critical factorwhich causes the low quality of native glass. Recently, the rapid development of machinevision makes it one of the most important products’ quality inspection methods. This paperstudies two key glass completeness inspection algorithms based on machine vision——edgedetection and ink mark recognition, and implements the glass completeness on-line inspectionsystem.Firstly, through an analysis of the function requirements of the glass completenesson-line inspection system, a total scheme is proposed. The function of each module of thesystem is expatiate. According to the design requirements of the image processing algorithm,an image processing scheme based on contour is adopted, and the basic procedures of thealgorithm are proposed.Then, this paper completes the detailed design of the glass edge detection algorithm.Glass images are binarized through threshold segmentation, and contours in these images areextracted. In order to detect straight lines in these contours, the characteristics of straight linesin digital images are analyzed, and an improved straight line detection algorithm is proposed.A glass contour is initially divided into several subsections, then each subsection is detectedby the straight line detection algorithm, and finally the subsections which are recognized asstraight lines are combined to form the straight edge of glass.The ink mark recognition algorithm is designed. The algorithm firstly extracts the localregions of ink marks, then accomplishes pre-processing and threshold segmentation of theseregions, and extracts the contours of ink marks. A region filling algorithm based on Freemanchain code is used to fill the contours of ink marks. Through an analysis of the geometricaland gray characteristics, six features are selected to recognize ink points. Finally, the detailed design of the glass completeness on-line inspection system wascompleted, and all modules of the system are implemented. An application instance of thesystem is showed.