Fused multi-sensor data using a Kalman filter modified with interval probability support

Multi-sensor fusion problem is mainly composed of three sub-problems: selection, fusion and estimation. Selection is choosing a representative subset of the sensors. Fusion is to take two or more separate sensors data and merge them to form a single entity. Estimation is the process of identifying the features of fused data. This paper addresses the issue of estimating an original system state from fused noisy sensor data by using a Kalman filter modified with interval probability support. The “measured noise variance” in Kalman filter is varied as the confidence in the fused measurement changes. Confidence is determined by means of interval probability (evidential reasoning), and has a net effect of increasing the filter gain as the confidence increases. The modified Kalman filter is compared to one with constant noise variance, and shows an increase in estimation performance level