Multi-resolution mesh-based motion estimation using a "backward in forward" tracking method

This paper introduces a mesh-based motion estimation scheme for image sequences and nodal motion vectors optimization by using a multi-resolution differential method. Because our final aim is mesh tracking throughout a video sequence, neither backward tracking nor forward tracking is well suited. The backward tracking provides good results when simply applied to two successive images, but it is damaged by the initial mesh which can not generally be considered as known on the last frame. Likewise, forward tracking suffers from an analysis/synthesis problem but it allows long time trackings. The motivation of our work is to take advantage of both forward tracking (which enables tracking) and backward tracking (for its efficiency) in a "backward in forward" method. To illustrate the proposed method, first results concerning the motion compensation of a mesh are shown and compared to the usual forward approach for one single mesh level, with or without multiresolution.