Visual Target Tracking Using Geometrical Particle Filter and Analytic Color-Based Histogram Model

Color is an important feature to describe object in visual tracking. Color-based histogram is used to model the object properly and Bhattacharya distance is also used to measure the error between reference and candidate histogram. Particles filter estimate position of target while two-dimension affine transformation is used as state of the system. Considering geometric properties of affine transformation as affine group cause two-dimensional mapping of the object to be closer to the real three-dimensional model. Approximation of optimal importance function of particles filter is obtained from Taylor expansion of Bhattacharya distance. Experiments show the accuracy and stability of the proposed tracker for fast and complex movement of a color target versus the gray level geometric particle filtering algorithm.