Visual Object Tracking with Colored Measurement Noise using Kalman and UFIR Filters

Visual object tracking is commonly accompanied with large variations in the image frame size and position, as caused by the object dynamics and low frame rate. In this paper, we treat such variations as a Gauss-Markov colored measurement noise (CMN), modify the Kalman filter and unbiased finite impulse response filter using the backward Euler method and employing measurement differencing, and apply to some simulated and benchmark data. It is shown that the modified filters can suppress efficiently both the slow frame variations associated with CMN and fast variations associated with white noise. Extensive experimental investigations conducted for the "Car4" benchmark database has demonstrated a high efficiency of the modified algorithms.