MECSE 96-4 Motion Estimation Using Robust Statistics

A method for deriving optic ow, using robust statistics, is derived and applied to standard test image sequences. We compare the performance of our approach against the best results derived from competing methods. The method out-performs all published methods in terms of accuracy. The method is also cheaper to implement than most of the other methods. One of the key features in the success of this method, is that we use Least Median of Squares, which is known to be robust to outliers. However, the Least Median of Squares is known to be computationally expensive to compute. We manage to keep the computational cost very low, whilst still retaining excellent performance, by using an approximate solution to the Least Median of Squares only in a rst stage that detects outliers. The approximate solution is fast and adequately robust, as our extensive experiments demonstrate. The essential features of our method should be applicable in the solution of a wide range of other problems: essentially, wherever one solves an over-determined set of linear equations.