Wang, HanziSuter, David2010-01-222010-01-222003http://hdl.handle.net/2440/55768Estimating information from data with multiple structures has obtained more and more attention in computer vision community. When data include multiple structures, two major steps should be taken: i) robustly estimate the parameters of a model, and ii) differentiate inliers from outliers. In this paper, we propose two new robust techniques — robust Two-Step Scale estimator (TSSE) and robust Adaptive Scale Sample Consensus (ASSC) estimator. The first estimator (TSSE) applies nonparametric density estimation and density gradient estimation techniques, to robustly estimate the scale of inliers for heavily contaminated data. The second estimator (ASSC) is a complete robust fitting estimator. ASSC is based on both Random Sample Consensus (RANSAC) and TSSE. The ASSC estimator can tolerate more than 80% outliers. The main advantage of the ASSC estimator over RANSAC is that prior knowledge about the scale of inliers is not needed. The ASSC estimator can simultaneously estimate the parameters of a model and the scale of inliers belonging to that model. Comparative experiments show that the ASSC estimator has better robustness to heavily corrupted data with multiple structures than other robust methods: such as Least Median Squares (LMedS), Residual Consensus (RESC), and Adaptive Least Kth order Squares (ALKS).enReport0020094194