Recursive displacement estimation for use in multiple object tracking

The authors develop a recursive model-based maximum a posteriori (MAP) estimator that estimates the displacement vector field (DVF) from a noisy image sequence. To model the DVF, they develop a nonstationary vector field model called the vector coupled Gauss-Markov (VCGM) model. The VCGM model consists of two levels, an upper level, which is made up of several submodels with various characteristics, and a lower level or line process which governs the transitions between the submodels. By estimating the line process the authors segment the DVF along its motion boundaries, which is a critical step in the tracking of objects through a video sequence.<
>