Joint multi-sensor kinematic and attribute tracking using Bayesian belief networks
M.L. Krieg · 2003
Bayesian belief networks provide a convenient method for modelling the uncertain dependencies between variables. Here, they are used to formulate the multi-sensor joint kinematic and attribute tracking problem, in particu- lar; modelling the relationship between target kinematics and class. The resulting belief network is used to develop a joint kinematic and class tracking algorithm, in which the kinematic and attribute measurements are used to estimate both the kinematic state and the class of the target. The solution presents multiple Kalman filters, each represent- ing the kinematic behaviour of one of the possible target classes, with the$nal kinematic state being the sum of the filter outputs weighted by the appropriate target class prob- abili9. The target class probabilities are estimatedfrom the class measurements and the appropriate filter innovations. Simulation results illustrate the effect of both position mea- surements on class probabilities and class measurements