Affine Modeling of Targets in Video Sequences by Particle Filters

We propose an affine template matching with a statistical approach based on particle filtering for tracking objects of interest in video sequences. The widely used Kalman filter can not directly address the dynamics with affine transformation because of nonlinearity. In contrast, particle filters are capable of dealing with nonlinear and non-Gaussian state space models using Monte Carlo approximation. Decomposing affine transformation into six geometric parameters, we naturally model visual motion of targets by a state space model. Experimental results with real video sequences are shown to evaluate the performance.