H. Yan, T. Tjahjadi

Proceedings of 2004 IEEE International Conference on Systems, Man, and Cybernetics, October 2004, 3082-3087.

Abstract

This paper addresses optical flow estimation and motion segmentation through a highly robust estimator vbQMDPE. A couarse-to-fine model based optical flow estimation technique through the robust local regression is first presented. Then, the estimated optical flow is further optimised in each hierarchical level based on piecewise spatial coherence constraint using a robust M-estimator. This technique can produce an accurate optical flow estimate for image sequences containing multiple motion and large motion. In the motion segmentation, we present a model-based motion segmentation technique through the mean shift based robust estimator vbQMDPE. The experimental results have demonstrated that our approach achieved good performance in most synthetic and real video sequences.