Abstract
In this thesis, we propose a new algorithm that improves the efficiency and robustness of random sampling consensus (RANSAC) for robust model fitting problems. To be more general and practical, this algorithm is designed to be fully data-driven, robust to highly contaminated data, and efficient enough to pursue real-time response for practical applications. To achieve this objective, three techniques are developed in an iterative consensus framework. Firstly, we propose a consensus sampling technique to increase the probability of sampling inliers by exploiting the feedback information obtained from the evaluation procedure. Secondly, the preemptive multiple K-th order approximation (PMKA) is developed for efficient model evaluation with unknown error scale. Lastly, we propose a coarse-to-fine strategy for the robust standard deviation estimation to determine the unknown error scale. We apply the algorithm to several 3D vision problems, including fundamental matrix computation and multi-view metric structure from motion. Experimental results on both simulated and real data are shown to demonstrate the superiority of the proposed algorithm over the previous methods.