Robust estimation of optical flow

Abstract

This thesis is concerned with the robust estimation of the optical flow from time-varying image sequence. The robust statistics has recently been adopted by the computer vision community. Various robust approaches in the computer vision research have been proposed in the last decade for analyzing the image motion from the image sequence. Because of the frequent violation of the Gaussian assumption of the noise and the motion discontinuities due to multiple motions, the motion estimates based on the straightforward approaches such as the least squares estimator and the regularization often produces unsatisfactory result. Robust estimation is a promising approach to deal with these problems because it recovers the intrinsic characteristics of the original data with the reduced sensitivity to the contamination. Several previous works exist and report some isolated results, but there has been no comprehensive analysis. Multiple constraints are obtained by fusing the weighted data measured from various frequency channels based on two methods for image motion estimation, which have their respective strengths and weakness. Combined with a careful data filtering by the step edge analysis, the filtered data had a good quality of multiple constraints enough to reform data of a certain frequency channel or a certain estimation method. Moreover, the adaptive modeling technique proposed in this research is shown to balancing the numerical stability and the complexity of flow model provided by the normal flow data. The robust approaches to the optical flow estimation armed with the adaptive flow modeling technique are proposed based on three robust estimators: the maximum likelihood estimator, the least median of squares (LMS) and the least trimmed squares (LTS) estimators. Especially, to evaluate the performance of various M-estimators, comparative studies are conducted for every possible combinations of the parameters of three types of M-estimators, two methods of scale es...