A Data Fusion Algorithm for Multisensor Systems

Data fusion techniques are used in many tracking and surveillance systems as well as in applications where reliability is of a main concern. One solution for design of such systems is to employ a number of sensors (maybe of different types) and to fuse the information obtained from all these sensors on a central processor. Past attempts to solve this problem required an organization of a feedback from the central processor to local processor units (each of which includes a sensor and a local processor). Local estimations are then generated from the global estimation obtained from the previous step [1]. However, this causes computational bottleneck problems when data is transmitted. This problem was solved later in [2] for both cases: decentralised estimation and decentralized LQG control. The algorithms based on parallelization of the Kalman filter equations, as proposed in [3], extend the previous results allowing one to obtain the global estimation using only local estimates without transmission of information between sensors. Another method for data fusion is based on the so-called Federated Filter (squareroot version of which is given in [4]). The Bayesian method based and linear sensor fusion algorithms are developed in [5] for both configurations: with a feedback from the central processor to local processing units and without such a feedback.