A Bayesian analysis of surveillance attribute data

Abstract

Surveillance system sensors generally provide information on location and on other attributes of the object detected. This additional attribute data can be employed in associating a given report with a set of previous reports in the data base (track) thought to represent a single object. The present Bayesian analysis of the association probabilities, arising from such attribute data goes beyond previous treatments in three ways. First, explicit allowance is made for four different types of attribute parameters encountered in many multi-sensor systems. The second distinguishing feature of this scheme is the explicit consideration of uncertainties in report parameters due to errors and deception, and of uncertainties in track parameters due both to these causes and to association probabilities less than unity. Finally, an inference procedure, based on conditional prior probabilities, is developed to treat cases where there is limited overlap between report and track attribute sets. This situation is frequently encountered in multi-sensor systems.