An occlusion-resistant circle detector using inscribed triangles. (January 2021)

• Record Type: Journal Article
• Authors: Zhao, Mingyang; Jia, Xiaohong; Yan, Dong-Ming
• Is Part Of: Pattern recognition. Volume 109(2021)
• Publication Date: 2021-01
• Subjects: Circle detection -- Inscribed triangle -- Parameter estimation -- Hough transform; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2020.107588
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: A novel circle detection method based on inscribed triangles, which is resistant to occlusion and robust to noise. A new arc grouping strategy using the relative position constraint and the inscibed triangle constraint.. A total geometry-based parameter estimation method by inscribed triangles without the dependence of least-square fitting but with the equivalent accuracy. A new collected real-world dataset with the sufficient examination for circle detection methods. Extensive experiments compared with representative state-of-the-art methods and better results have been obtained both in F-measure and execution speed. Abstract: Circle detection is a critical issue in pattern recognition and image analysis. Conventional geometry-based methods such as tangent or symmetry are sensitive to noise or occlusion. Area computation is more robust against noise, because it avoids differential calculations. Inspired by this characteristic, we present a novel method for fast circle detection using inscribed triangles. The proposed algorithm, which is robust to noise and resistant to occlusion, first extracts circular arcs by approximating line segments and identifying inflection points and sharp corners. To speed up the computation, irrelevant segments are filtered out through the triangle inequality. Arcs that belong to the same circle are then combined according to the position constraint and the inscribed triangle constraint. The circle parameters are further estimated by inscribed triangles based upon the Theil-Sen estimator and linear error refinement without the dependence of least-square fitting but still with the equivalent accuracy. Finally, candidate circles are verified to prune false positives through an inlier ratio rule, which jointly considers both distance and angle deviations. Extensive experiments are conducted on synthetic images including overlapping circles, and real images from four diverse datasets (three publicly available and one we built). Results are compared with those of representative state-of-the-art methods, and the proposed method is demonstrated to embraces several advantages: resistant to occlusion, more robust to noise, and better performance and efficiency.