Fast Circle Object Detection Using Gradient-Orientation based Clustering

摘要

This paper presents a fast, highly reliable and accurate algorithm to detect and accurately locate the circular objects in digital images. The algorithm includes four steps. First the Canny edge pixels are classified into four types based on their gradient orientation. Then a circular fit algorithm is hired to calculate radii and curve center information for each edge pixel cluster of one of the types (a potential arc or a segment of a circle). In third step, a robust criterion is developed to separate the valid arcs from invalid arcs. Finally, those valid arcs which fall on the same circle are regrouped to calculate the final circle radius and the center information for that circle. The experiment shows the algorithm has three advantages. Its speed is much faster than Hough Transform-based circle detection algorithms; it is able to reliably detect partial and full circles even in the noisy environments; the achieved accuracy for radius and center position detection has reached sub-pixel level on average.