On the use of joint estimation in particle filters for object tracking in video

Object tracking is an important problem, whose effective solution is crucial to lot of applications. Though several methods have been proposed in the literature, they fail when the object of interest does not conform to a specific process model. Also, current methods have to be tuned for different videos and objects (say, for example, using training data). One solution for such a problem is to estimate the parameters of the process model while estimating the state. In this paper, we propose such joint estimation of particle filters for object tracking and show that for the same model for state estimation, particle filtering with joint estimation performs better (in terms of the tracking error) than conventional particle filtering.