DL-RANSAC: An Improved RANSAC with Modified Sampling Strategy Based on the Likelihood

This paper intends to improve the RANSAC by introducing new sampling method which provides prior knowledge of random population before selecting the hypothesis set. In traditional RANSAC algorithm, a minimal set of samples are chosen randomly from the population containing uneven noise and iteration continues until the desired model is found. Ambiguity remains to find the desired result within a short time because of the random sampling technique. The proposed method, DL-RANSAC (Descendant Likelihood sampling RANSAC), reduces the randomness by introducing descending likelihood based minimal set selection which converges to the desired result faster than the conventional RANSAC. Experiments shows the superiority of this proposed method in line fitting problem, correspondence points matching between a pair of images and loop closing of ORB-SLAM2. Less computational time and easy implementation ability make it beneficial to use over other methods.