Curvature aided Hough transform for circle detection

• Curvature radius is adopted to improve the H-transform for circle detection.
• Curvature pre-estimation avoids senseless accumulation operation, work faster.
• The CACD is capable to detect circles of different radius in complex scene.
• “Statistic deviation” is defined to measure the saliency of circle center.

Conventional Hough based circle detection methods are robust, but for computers in last century, it is to slow and memory demanding. With the rapid development of computer hardware, Hough transform is acceptable now. Improvement on Hough based circle detection is valuable. In this paper, we present a novel curvature aided Hough transform for circle detection (CACD) algorithm, which estimates the circle radius from curvature. Curvature pre-estimation is capable to avoid both accumulating operations of all the points and interruption between different scales, which result in faster and more precise circle detection. Compared to the conventional Hough-based algorithm for circle detection, the algorithm is more practical and less time consuming. Its time taking is about 1/8 of that of conventional algorithm. Test results on traffic sign images shown that The CACD gets an AUC (Area Under Curve) of 0.9125. The CACD is capable to detect circles of different radius in complex scene.