A probabilistic framework for tracking in wide-area environments

摘要

Surveillance in wide-area spatial environments is characterised bycomplex spatial layouts, large state space, and the use of multiplecameras/sensors. To solve this problem, there is a need for representingthe dynamic and noisy data in the tracking tasks, and dealing with themat different levels of detail. This requirement is particularly suitedto the layered dynamic probabilistic network (LDPN), a special type ofdynamic probabilistic network. In this paper, we propose the use of LDPNas the integrated framework for tracking in wide-area environments. Weillustrate, with the help of a synthetic tracking scenario, how theparameters of the LDPN can be estimated from training data, and thenused to draw predictions and answer queries about unseen tracks atvarious levels of detail