Error analysis of modeling in a low-resolution color-based visual tracker

If the Kalman filter is effective to enhance the CAMSHIFT tracker in the low-resolution image sequence, the state-space model (SSM) that the Kalman filter is based on needs satisfying three stochastic conditions: (1) the observation noise is relatively small; (2) the state noise and observation noise are independent to each other; (3) the distribution of the observation noise is near to Gaussian distribution. In this paper, given the maximum likelihood (ML) estimator to exactly incorporate the quantization information of the tracking results into the observation matrix and observation noise matrix of the SSM, these conditions can be acquired.