Robust estimation for the fundamental matrix based on LTS and bucketing

The fundamental matrix is an effective tool to analyze epipolar geometry. An accurate solution for obtaining fundamental matrices is the basic requirement in many applications of computer vision. When noises and outliers exist in the set of initial match points, the estimation of the fundamental matrix becomes to a tough mission owing to the invalidation of normal linear and iterative methods. This paper proposes a novel robust technique for estimating the fundamental matrix by combining bucketing technique and the least trimmed squares (LTS) regression into one intelligent algorithm. The new algorithm solves the problem of even distribution of sample data. Also, it eliminates limitations on the proportion of outliers and the requirement a predefined threshold. Comparing with traditional robust methods, the proposed approach is proved to be accuracy and robust by simulation and real image experiments.