Reduced-order multisensory fusion estimation with application to object tracking

This paper investigates the track-to-track state estimation for a class of linear time-varying multisensory systems. We propose a novel low-complexity reduced-order filter (ROF) under the Kalman filtering framework. Unlike the majority of previous track-to-track strategies, the proposed fusion strategy applies only to special variables or required components that contain critical information about a target system of interest. Also, unlike existing suboptimal fusion filters such as the covariance intersection, the proposed ROF algorithm makes use of nonzero cross-covariances between local filters that greatly improve its estimation accuracy. The theoretical aspect of ROF application to multisensory systems with identical sensors is also thoroughly investigated. Finally, we show the effectiveness and accuracy of the ROF when applied to objects (including a drone) performing a two-dimensional maneuver using numerical simulations.