Robotics: Science and Systems II

A new inlier identification scheme for robust estimation problems

W. Zhang, J. Kosecka

Abstract: Common goal of many computer vision and robotics algorithms is to extract geometric information from the sensory data. Due to the presence of the sensor noise and errors in matching or segmentation, the available data are often corrupted with outliers. In such instances, the problem of estimation of parametric models needs to be tackled by robust estimation methods. In the presence of large fraction of outliers sampling based methods are often employed to tackle the task. When the fraction of the outliers is significant and the parametric model is complex, the traditionally used RANSAC algorithm requires large number of samples, prior knowledge of the outlier ratio and additional, difficult to obtain, inlier threshold for hypothesis evaluation. To tackle these problems we propose a novel and efficient sampling based method for robust estimation of model parameters from redundant data. The method is based on the observation that for each data point, the properties of the distribution of the residuals with respect to the generated hypotheses reveal whether the point is an outlier or inlier. The problem of inlier/outlier identification can then be formulated as a classification problem. The proposed method is demonstrated on motion estimation problems from with large percentage of outliers ( 70%) on both synthetic and real data and estimation of planar models from range data. The method is shown to be of an order of magnitude more efficient than currently existing methods and does not require prior knowledge of an outlier ratio and inlier threshold.



    AUTHOR    = {W. Zhang and J. Kosecka},
    TITLE     = {A new inlier identification scheme for robust estimation problems},
    BOOKTITLE = {Proceedings of Robotics: Science and Systems},
    YEAR      = {2006},
    ADDRESS   = {Philadelphia, USA},
    MONTH     = {August},
    DOI       = {10.15607/RSS.2006.II.018}