A new method for quadratic curve detection using K-RANSAC with acceleration techniques

摘要

In this paper, we propose several K-RANSAC algorithms for quadratic curve detection based on random sample consensus [M.A. Fischler and R.C. Bolles, Commun. ACM 24, 381–395 (1981)]. The proposed K-RANSAC algorithms are memory efficient and allow the user to project a probability of success in detection. It is shown that at the cost of a small probability of failure, running time of the proposed algorithms can be attractive. Moreover, the proposed algorithms treat the detection problem in the image space; thus, tolerance can be accurately compensated. Geometric properties of the quadratic curves are exploited to provide speed-up for the basic random sample consensus algorithms. Random pre-sampling provides further speed-up. Throughout the paper, experiments are conducted to illustrate the new algorithms.