Investigation Of Key Technologies On Dimensional Detection Of Mechanical Parts Based On Computer Vision

With the rapid development of technology of computer vision and technical production technology, the requirement of metrical precision for the sized measurement of mechanical parts have increasingly heightened. Therefore, the study of how to improve the metrical accuracy of component size has become an important problem of current researches. On the basis of the theory of existing vision measurement, this paper improves the metrical accuracy of machine parts which possess geometrical figure by studying the interrelated technology of visual detection.Computer vision inspection is a kind of technology that images are collected by the camera and then transmitted to a computer. The computer is used to process the image in order to get the precise size of the edge. Digital image processing algorithm is the basic theory of visual detection. First of all the paper segments the image which has been preprocessed. In order to improve the accuracy of fringe image, the image segmentation method which is on the base of ant colony algorithm has been improved, the initial clustering centers are set by two-dimensional histogram, Laplacian operator is used to divide the clustering center into two types of background and border points, and the update strategy of pheromone is adopted to avoid premature convergence and stagnation phenomenon. Experiments show that the improved ant colony algorithm can segment the background and borders quickly and accurately and achieve the desired results.In the interest of achieving sub-pixel accuracy for the edge, the paper uses spatial– moment method to locate the edge of image in view of the characteristics of mechanical parts’s edge . Due to the errors of sub-pixel detection are very large at the neighborhood of the junction, it analyses the edge-model of the junction and advances a kind of subdivision for edge junction. The spatial– moment method uses the template of 7 * 7 to enhance the precision of edge subdivision. The Random Hough Transform Algorithm and least squares method are used to detect the parameters of lines and circles, and the paper handles the invalid round accumulation which is brought by random Hough transform algorithm .The final calibration method uses the traditional calibration methods, it gets the connection between the physical size of pixels and the real relations so that receives the geometry size of parts.