Research Of Twisted-pair Pitch Measurement Method Based On Image Detection

With the speed of data transmission improving, internal cross-talk noise of the twisted-pair will enhance significantly. In order to keep high data transmission speed of the twisted-pair and maintain internal cross-talk noise within a controllable scope, we need to measure the pitch of the twisted-pair accurately. However, the traditional methods are inefficient to measure the pitch of twisted-pair. The traditional methods are likely to lead a big error by the effect of subjective vision. To improve the efficiency of twisted-pair pitch measurement and reduce artificial errors in the process of pitch measurement, the thesis puts forward a pitch measurement method based on image detection. The proposed pitch measurement method substitutes machine vision for human vision. It can overcome the subjective error of human vision effectively. And the whole process of measurement can be completed independently by industrial PC. It will improve the efficiency of pitch measurement effectively. In the thesis, the main achievements are included as follows:Firstly, the thesis outlines the whole framework of twisted-pair pitch measurement system and the effect of each modules in the system. It expounds the position of the parallel light source, twisted-pair and CCD camera in the measured cavity. In the captured images, the difference between the foreground and the background is more obvious by this way. From the above it is good for subsequent image segmentation.Secondly, the gray difference between the foreground and the background is obvious in the captured images. To utilize this characteristic, an improved two-dimensional Otsu algorithm is proposed. In the first place, the improved algorithm constitutes a new two-dimensional histogram by a new local template. The improved algorithm can include more than 99.99% of probability density points by using two parallels. The two parallels are parallel to the main diagonal. The optimal scope of new two-dimensional histogram is about 30.63% of the traditional two-dimensional histogram. Using new two-dimensional histogram will reduce the optimal scope of algorithm and accelerate the speed of optimization. In order to avoid the problem of repetitive computation in the traditional two-dimensional Otsu algorithm, the new algorithm sets up a look-up table of probability density and two other look-up table of gray vector mean. Through the above three look-up table, the new algorithm can reduce repetitive computation effectively. At last, the foreground and the background of twisted-pair image can separate from each other rapidly and accurately through the improved two-dimensional Otsu algorithm.Thirdly, for the two kinds of different colors of wires in the foreground, their gray difference are not obvious. To separate from the two kinds of wires, an improved fuzzy c-means clustering algorithm is put forward. First of all, the improved algorithm initializes clustering center by the idea of the optimal binary tree. The initializing clustering center can make the improved algorithm has fast convergence. Using the feature distance to replace the Euclidean distance, the universality of the improved algorithm can be enhanced effectively. The improved algorithm can obtain the optimal fuzzy weighting parameter by analysis evaluation coefficients of segmented images. At last, the image segmentation of the foreground can be completed, the binary image of wire 1 and 2 can be acquired.Fourthly, due to winding between the two wires in the foreground, this situation will lead to a lack ofpixels in the parts. To deal with the problem of the missing pixels in the parts, a specific area pixels filling algorithm is proposed. The proposed algorithm is based on edge detection. Then the center line coordinates of two wires can be acquired. Next the center line coordinates of two wires can be fitted by Fourier fitting model. Through the fitting equation, the feasible solution can be acquired. The distance between two feasible solutions is the pitch of twisted-pair in the image. Through the conversion of image scale, the actual pitch can be obtained. The thesis compares the twist-pair pitch measured by image detection with the twist-pair pitch measured by micrometer. Through the comparison with two pitch measurement methods,the accuracy of the proposed pitch measurement method can be acquired.