Circle detection is one of the most basic and important tasks in machine vision. In order to accurately determine the circle location in complex background images, a new joint algorithm that combines the model of support vector regression with the three-point fitting circle detection algorithm is proposed. The different types of circular samples are trained by the support vector regression model in the algorithm. So the hyperplane equation f(x) can be obtained. Taking the f(x) as the center line, one similar circular ring with the width of 2 can be constructed. The points in this interval are considered as the circular boundary points. Then, the center and radius can be calculated based on the three-point fitting circular geometry algorithm, so as to achieve the purpose of identifying the circle. The experimental results show that the circular boundary information can be obtained from the relatively noisy background images by learning the training samples thereby determining the location of the circle, which has some advantages over using only a certain circular recognition algorithm. In the field of machine vision positioning with circles, this joint algorithm has important theoretical research value and practical significance.