Research Of Magnetic Material Detection Methods Based On Machine Vision

As a new technology of potential development, machine vision had wide applications in security systems, manufacturing detection systems, medical image analysis, unmanned aerial vehicle and many other fields. Vision algorithms were starting to mature, more powerful cameras, widespread of computers and so on, had brought the rapidly growing of this technology. To solve the automatic measurement problems of magnetic material's perimeter and area, this thesis studied some real-time, high-precision detection algorithms, compiled special detection software and experimental verification was also carried out by the use of rectangular type, hollow cylinder type and small tile-shaped magnetic materials. At the same time the right hardware was selected according to the requirement, based on machine vision theory, the magnetic material detection system was build, thus promote the surface quality detection technology of magnetic material.Combined with detection requirement of magnetic material geometry parameters, the mechanical structural and control part of the overall detection system was designed. Based on a single camera, the detection system designed a complete mathematical model for dimension parameters measurement. A method to measure the distance by the use of a single camera plus a laser pointer, an algorithm which could quickly convert color image to gray image based on GPU acceleration, better binary effect was obtained by Otsu threshold algorithm, also a Gaussian and Laplacian pyramid edge detection algorithm was used. A circularity indicator to determine the shape of the magnetic material, a statistics method to measure the area based on region was proposed, the edge pixels were labeled, and then the perimeter was obtained by accumulating the number of the labeled pixels.According to the camera perspective projection model, lens radial distortion and tangential distortion were fully considered. By the use of Intel open source computer vision library OpenCV and an KLT corner detection algorithm of sub-pixel precision, an improved camera calibration algorithm was designed, in the mean time, stereo calibration was achieved using a Matlab stereo camera calibration toolbox, the calibration results of the two methods were compared and analyzed, precise intrinsic and extrinsic parameters of the camera were obtained, made a good foundation for subsequent binocular vision research.In view of monocular vision had many limitations on measurement of complex targets and stereo vision, the thesis also studied binocular vision measurement method. The detection principle, epipolar geometry, stereo match and disparity calculation of binocular stereo vision were analyzed in detail, the existing stereo match algorithms were classified by deep study, the stereo match algorithms of graph cuts were analyzed in-depth, a combination of a pyramid algorithm and a block matching algorithm with sub-pixel estimation were proposed, meanwhile, the results was reconstructed and system error was analyzed.