Study Of Grinding Wheel Topography’s Feature Points Base On Binocular Image

In the detection of grinding wheel topography, the exposed height is a criticalparameter which will directly influence the type of grinding. So, grasp the exposed heightand distribution information of grit quickly and accurately is very essential forexploring the grinding mechanism of grinding wheel. The conventional detectionmethods of topography of grinding wheel have defects such as low speed of inspection,low precision and lack of height information of grits. With the fast development ofcomputer vision techniques, there has a solid theoretic researching about the feasibilityof detection techniques of topography of grinding wheel base on binocular vision. It hashas advantages of high calculation efficiency and strong adaptability which based onbinocular vision theory.Feature point extraction and matching is the binocular stereo reconstructiontechnology core and difficulty. The positioning accuracy of corners and matchingprecision will directly influence the effect of3D reconstruction of topography,especially the binocular vision image of topography of grinding wheel. Aimed at thedifficulties of this technology, an in-depth research on the extraction and matchingalgorithm of the binocular vision image of topography of grinding was performed, andexcellent results were gained which provide theoretical basis for the following work.The major work of this paper summarized as follows:(1) The device of shooting binocular images such as the grinding wheel, camera,light source, line move device and pesersonal computer have been designed using theoptical axes boresight method.(2) Four commonly used algorithm of corners detection were studied. Thesealgorithm include Moravec corner detection algorithm, SUSAN corner detectionalgorithm, Harris corner detection algorithm and SIFT corner extraction algorithm,the merit and demerit of which were analyzed carefully. Harris corner detectionalgorithm and SIFT corner extraction algorithm were testified by programming.(3) The ZNCC and NN feature point matching algorithms were studied. Harriscorner and SIFT corner matching using ZNCC and NN algorithm were conducted andsatisfactory results were obtained by Simulation Experiments. (4) Aimed at the foreseeable mismatch, this paper used RANSAC algorithm toeliminate the mismatched points. Simulation Experiments showed that RANSACalgorithm can eliminate the most majority of mismatched points.(5) Bring forward the evaluation index according to the characteristic of topographyof grinding wheel, and then valuate the feature points of Harris-ZNCC and SIFT-NNalgorithms.