Research On Techniques Of Vision Measurement Based On Sub-pixel Detection Algorithms

With the enhancement of the level of social development, strict standard for all kinds of products is increasingly required. And as an important measuring method applied widely in actual manufacture and ordinary life, the technology of machine vision has the prominent benefits of non-contact, high precision, high speed and good flexibility. Therefore, its research value has been paid more and more attention at home and abroad.Edge detection, the important step of vision measurement technology, aims at accurately detection and extraction of edge position in order to achieve high-precision measurement. To meet the requirements of precision in the actual measurement, sub-pixel edge detection, which detection accuracy is less than one pixel, has become one of the most active issue in the field of vision measurement technology. In this dissertation, the sub-pixel edge detection and its associated vision measurement techniques were studied.Firstly, the theory of edge detection in image pre-processing, which includes image noise, wave filter algorithm, Threshold segmentation and traditional pixel edge detection, is studied. Three types of sub-pixel edge detection method is summarized. These methods commonly used, including Zernike moment method, polynomial fitting and bilinear interpolation is simulated in MATLAB. These features, which include precision, speed and noise immunity, are compared about the advantages and disadvantages of each method.Secondly, a trigonometric edge model is proposed, which use the period [-1.5,1.5] trigonometric curve to simulate sub-pixel edge. The disadvantage of existing linear edge model, that is not generalized, is overcome. And compared with quadratic curve edge model, the new method is ensured that the window is always through from left to right. As for the sub-pixel edge detection algorithm based on partial area integration, the new methods of side procession in the image is proposed. And the problem of horizontal single pixel edges is discussed in detail. In the end, the simulation and experiments results indicate that the trigonometric model algorithm proposed could detect accurately the subpixel edge position, which the accuracy as far as is 0.03 pixel. And obviously, the results is best than the other theories.Thirdly, in order to ensure the improved detection algorithm could be applied in the high-noise images, a method is improved about the nonlinear diffusion filter equations in this dissertation. The standard deviation of the Gaussian equation is changed from constant to variable functions. With the increase of iterations, the standard deviation of the Gaussian function decreases gradually. Thus it both ensures the filtering effect, and reduces the loss of edge details. The results about NSNR of the high noise image is respectively decreased by 6.6 % and the improvement is good.At last, a vision measurement system based on the lab is set up and the calibration is completed. Images of a razor blade which has clear edges is captured and processed. The results simulated by MATLAB show that high-precision could be achieved by the method of trigonometric edge model detection, compared with the original algorithm and Zernike moment method.