ABSTRACT
Many advanced estimation and target tracking systems, including aerospace, defense, robotics and automation systems, and the monitoring and control of generation plants, often involve multiple homogeneous or heterogeneous sensors that are spatially distributed to provide a large coverage, diverse viewing angles, or complementary information. An important practical problem in these systems is to find an optimal state estimator given the sensor observations. Kalman filtering is the best known recursive linear MSE algorithm to optimally estimate the unknown state of a dynamic system. If a central processor receives all measurement data from sensors directly and processes them in real time, the corresponding processing of sensor data is known as the centralized Kalman filtering fusion, whose state estimates, since using all sensor observations, are clearly globally optimal in the MSE sense. However, this approach has several drawbacks, including poor survivability, reliability, heavy communication and computational burdens, along with the number of sensors increasing. An alternative approach is the so-called distributed or decentralized approach. In this approach, also known as sensor-level estimation, each sensor maintains its own estimation file based only on its own data and messages received. These local estimates are then transmitted to and fused in a central processor to form a fused estimate that is superior to the local estimates in terms of some optimality criteria, such as the MSE. In addition to better survivability and reliability and usually a lower communication load, this approach has the advantage of distributing the computational load. This distributed approach has two major components: sensor-level estimation and estimation fusion. Estimation fusion, or data fusion for estimation, is the problem of how to best utilize useful information contained in multiple sets of data for the purpose of estimating an unknown quantity—a parameter θ or a state process x k . These data sets are usually obtained from multiple sources (e.g., multiple sensors). Like most other work on distributed estimation, this chapter deals only with the second component: optimal distributed Kalman filtering fusion.
