Geodesic Active Regions for Tracking

In this paper we propose some new ideas for tracking multiple moving objects by the propagation of curves. We assume a static observer, as well as the existence of a background reference frame. The tracking is performed using an improved Geodesic Active Contour model that incorporates boundary-based and region-based motion information. This model is called a Geodesic Active Region model. Initially, a statistical analysis is performed which provides a measurement that distinguishes between the static and mobile regions in terms of conditional probabilities. Based on this analysis, the boundary-based information is expressed via discontinuities in the associated probability space, while the region-based motion information is expressed directly from the conditional probabilities. An objective function is deened that integrates boundary-based and region-based tracking modules by seeking curves that provide high boundary-based information and maximum a posteriori segmentation probability with respect to the mobile hypothesis on the interior curve regions. This function is minimized using a gradient descent method, where boundary-based and region-based motion forces move the initial curves towards their global minima. Additionally, a regularity constraint is imposed for the propagation of the curves which is manifested as an intrinsic term of the partial diierential equation that deforms these curves. This PDE is implemented using a Level-Set approach, where a very fast front propagation algorithm evolves the initials curve toward their global minima. Very promising experimental results are provided using real video sequences.