The University of Sussex

Tracking models using convergence techniques

Alistair J. Bray

An algorithm is described for tracking a known model from frame to frame in a sequence of grey- level images. The model is initially located using the local constraint method described by Goad, and is then tracked into subsequent frames using Lowe's convergence algorithm. The correspondence problem is solved using standard optic flow techniques to provide the disparity vectors necessary for the convergence. The integration of this optic flow technique with the predictive capacities of model-based methods yields distinct advantages. Amongst these are the side stepping of the aperture problem and the ability to cope with occlusion.


This paper is not available online