Research On Circle Detection Algorithm And Its Application In The Measurement Of Concentricity

In the pattern recognition and the computer vision domain, the detection of different circle forms such as a single circle, multi-circle and concentric circle is common and important. The key technique to do the detection is to obtain the information of the circles in a certain circumstance. Usually we collect the data of the parameters (radius, the coordinates of centers, etc.) with the help of detection algorithms. This study has important applications in the industry such as automatic testing, recognition of target position, cell analysis in the medical industry, etc.Hough transformation is an important mathematical tool in the study of detection algorithm. In the case that the space is two-dimension space, it works very well. But when the number of the dimensions of the space is more than 2, the complexity of the computation and space requirement for the storage increase rapidly that would force the program run very slowly, bringing down the efficiency to the very low level --- this is the difficulty that we would face when the space is 3-dimensional. In order to improve the efficiency, people in the area have founded randomized Hough transformation (RHT). RHT would decrease the complexity and the storage in some cases. But when dealing with complex image, the random sampling will bring lots of invalid accumulation and calculation. For a better application in the industry, in this paper, we propose two new circle detection algorithms based on our study on the current development of RHT in the area. Theoretical analysis and experimental results have proved that the proposed methods are feasible and effective. We also have applied our method in the practical detection of the nested fiber concentricity. The contents of this paper will be presented through the following five aspects.(1) The noise elimination of the image.In the process of data collection, obtaining, coding and transmission of images, there would be different noise disturbances, visible or invisible, on different degrees. These noises would affect the results of the detection. Based on the principal of vacuum wave filter, we have improved the Gaussen Module, a new noise elimination module has been constructed, this noise eliminator would reduce the noise effectively.(2) Edge detection.As we all know that usually we obtain the images in the concentricity detection system by using video cameras that take pictures of the circles. The pictures are affected by unevenness of the lighting, and the intensity of the lighting. To decrease these effects, among many methods, through experiments, we have chosen Canny operator as our edge detection operator.(3) Fast algorithm for circle detection using randomized Hough transform.RHT is a well-known method for circle detection. But when dealing with complex image, the random sampling will bring lots of invalid accumulation and calculation. This paper presents a fast algorithm for circle detection using RHT, and researches on evidence accumulation from three aspects. (i) Fast eliminating the points which are not on the circle. (ii) A simplified algorithm with which we can easily determine whether the point is on the edge of the circle. (iii) To determine a certain circle, a fewer number of edge points are needed.(4) Fast multi-circle detection algorithm based on randomized Hough transform. To solve the problem of low efficiency of random sampling, we propose a fast multi-circle detection algorithm based on RHT. Three kinds of noises are deleted. To determine a candidate circle, when randomly pick a sample point and search the other two points according to certain rules. The original image affirms the possible circle for true circle.(5) The concentricity detection system based on circle detection.At the present, the concentricity detection technique of large scale components is more advanced. But still, its precision rate is low, and the cost of the product is high. It still can not be used directly on the micro-scale components. The research of this technique is still at the beginning stage in the country, we still depend on the techniques imported from abroad. Throughout a series of researches on the detection of the nested fiber concentricity, we have found a new method of sampling that controls the evenness of the data collection, guarantees the precision rate of the detection and by using the algorithms we have proposed, we have increases the speed of computation, obtained better results.To summarize our achievements:(1) The noise elimination algorithm is comparably fast. It can efficiently exclude the background noise of the images, works well.(2) Our fast randomized Hough transformation detection algorithm reduces the complexity of computation. In many practical cases, it takes only 1/5 of the amount of time that the current RHT take. Furthermore, our algorithm is compatible with current algorithms so that by combining them together, we can expect better applications.(3) Our multi-circle detection algorithm is considerably faster in the case that there are multiple circles in a complex image, even in the case that some of the circles are not integrated. Regarding the existence many kinds of noise multi-circle complex image, the algorithm superiority can be more obvious.(4) The concentricity detection system based on circle detection has higher precision rate, better stability and better reliability.