Centralized sensor data fusion is really more powerful than track fusion

Abstract

Systematic sensor errors complicate centralized data fusion. [1] sensor model specific alignment procedures, which allow the compensation of essential bias values like azimuth deviation and range offset, do not remove systematic errors completely. The residual errors, like time differences between different sources and varying spatial deviations, lead to a reduction of the tracking performance in the centralized architecture. A detailed statistic of deviations between predicted track states and observations, evaluated for each track and each sensor, provides additional model independent bias information. It turns out, that this information is sufficient for a further decisive reduction of the residual errors. A feedback control mechanism can be established, which allows a continuous compensation of sensor and track specific bias values down to a level appropriate for the application of advanced recursive tracking methods. The centralized tracking, supported by bias control, finally gains its superiority over track fusion mainly from the following capabilities: Maneuver recognition based on bias controlled, multiple source measurements; Recursive filter selection according to the maneuver condition; Performance enhancement in group tracking, 2D data processing, track initiation and correlation gate determination. The processing of the complete set of aligned data, recursively adapted to the target maneuver, allows an optimal use of information from all contributing sensors.