MULTI-TARGET, MULTI-SENSOR TRACKING BASED ON QUALITY-OF-INFORMATION AND FORMAL BAYESIAN FRAMEWORKS

We consider a multi-target tracking probl em that aims to simultaneously determine the number and state of mobile targets in the field. Co nventional paradigms tend to report only the existence and state of targets according to central ized detection and data fusion. On the contrary, we investigate a multi-target, multi-sensor scenari o in which (a) both the number and the state of the targets are unknown a priori; and (b) the detec tion with respect to targets is employed in a distributed manner. Toward this end, we exploit random set theory, a statistical tool based on Bayesian framework, for establishing generalized likelihood and Markov density functions to yield an iterative filtering procedure. We conduct a stud y regarding how the design of distributed detection has impact on the result of system level information fusion. The sources of analyzed data include (a) simulation-based sensor readings through bi-directional sensing/communication; and (b) practical images taken by multiple cameras thro ugh uni-directional sensing/communication. The formulation of Bayesian filtering suggests that a design of a tracking system be adaptive to change of detection performance.