Research On Chessboard And Circle Images Feature Points Detection Algorithm

Machine vision is widely used in the fields such as3D reconstruction, visualmonitoring, industrial inspection.The image geometric feature points detection isnecessary for the majority of visual measuring systems. The overall performance ofthe system will be directly influenced by the detection precision and efficiency ofimage feature points, and it is an important part of image visual detection. Chessboardand circle images are taken as the research object of image feature pointsdetection.The detection algorithms and evaluation methods for chessbord corners andthe center of circles are researched by theoretical analysis, simulation studies andactual detection experiments.The main studies and results of this thesis are asfollows:(1)Creation of simulation image. Get chessboard and circle images through linearcamera perspective projection model and get simulation image similar to what theactual camera shoot by Gauss function convolution. The model can provide referencefor the detection experiments and evalution of image feature points detecitonalgorithm.(2)Detection algorithm study of chessboard image feature points. Pixel andsub-pixel detection algorithms of chessboard image feature points are introduced andcompared and analysed by experiments. To solve the corner detection problem ofalgorithm complexity, calculated quantities and stability, a method of chessboardsub-pixel corner detection algorithm based on gray symmetry factor is presented. Thealgorithm take square closed lop templates to achieve corner detection that withoutGaussian filtering or gradient computation, and it can effectively overcome theproblems such as large algorithm data amount and low operation speed. Sub-pixellevel corner coordinates are obtained by taking the gray symmetry factor weightingmethod in the neighborhood of pixel level corner coordinates which can effectivelydecrease noise and other uncertain factors influenced on the precision of cornersub-pixel detection.(3)Detection algorithm study of circle image feature points. Several circle imagefeature points detection algorithms are introduced. Here proposes a new algorithmbased on gray symmetry of ellipse geometric center to reduce the calculated quantitiesand complexity. Firstly using sobel operator to achieve image edge detection, then center of gravity are used to get elliptic center point coordinate, lastly using graysymmetry factor weighting method to obtain feature points coordinates of ellipseimage. The algoritnm has advantages such as smaller calculated quantities and highdetection efficiency and virtual precision and relatively small influence of noise.(4)Detection algorithm evalution of image feature points. Simulation standardimages with accurately known feature points coordinates are obtained by linearcamera perspective projection model to offer reference for detection algorithmprecision, and the precision is furtherly tested when target is in stationary state andCoordinate measuring machine (CMM) in micro displacement state. The experimentalresults show that both of the algorithms have the precision around0.2pixels, and theyare stability to the rotation of image and noise and illumination and can quicklyachieve the coordinate location of image feature points.The two algorithms caneffectively apply to machine vision3D four-wheel alignment and camera calibration.