Comparison of variance based fusion and a model of centralised Kalman filter in target tracking

Multi sensor target tracking is well known to have advantages over single sensor target tracking in many applications. Observational data from sensors, may be fused at different levels, ranging from the raw data level to feature level, or at decision level. This paper presents an experimental comparison of an averaging model of Kalman filter fusion and variance based fusion in estimating the position of a moving target. The target is assumed to follow a uniform velocity model. The performance of the two fusion techniques have been experimented in detail for various cases of noise variances and initial estimate of target's position. The sensors are assumed to be sensor suites, having autonomy and sufficient computational power. Accordingly, the variances of state estimate are assumed to be available from an EKF, where one EKF is running independently for each sensor suites. The mean square error in estimation for both techniques have been simulated in matlab and compared.