Performance Evaluation of RANSAC Family

Random Sample Consensus (RANSAC) [3] has been popular in regression problem with samples contaminated with outliers. M-estimator, Hough transform, and others had been utilized before RANSAC. However, RANSAC does not use complex optimization as like M-estimator. It does not need huge amounts of memory as like Hough transform to keep parameter space. RANSAC is simple iteration of two steps: hypothesis generation and hypothesis verification. It is now widely applied to many vision problem such as epipolar geometry estimation, motion estimation, structure from motion. Many researches on robust estimation have followed after RANSAC, but there are a few and old survey and performance evaluation [4, 8, 9]. An insightful view of the RANSAC family is described in this paper. The view categorizes them into their research objectives: being accurate, being fast, and being robust (Figure 1). It can be useful to analyze the previous works and develop the new method. Each viewpoint are also examined according to tactics to achieve the objectives. For example, guided sampling and partial evaluation have been tactics to accelerate RANSAC. Computing time of RANSAC is