Poisson Multi-Bernoulli Mixture Filter for Target Tracking Using a Gimballed Camera

In this paper we present a multi-target tracker (MTT) for an aerial gimballed camera operating in an area much larger than can be seen within the field of view of the camera. The MTT is based on the Poisson Multi-Bernoulli Mixture filter (PMBM) derived from a random finite set (RFS) approach to tracking. PMBM uses Poisson distributions for undetected targets and a Bernoulli representation for known target tracks. We define a new update strategy that accounts for the typical error and FOV constraints of a gimballed camera. The update strategy uses a sampling approach to estimate the probability of detection of a given target if we were to assume it could have been observed. Probability of detection is parameterized by the attitude noise of the gimbal and extrinsic camera model. Using the probability of detection, target tracks outside the field of view can be maintained and not prematurely pruned. We demonstrate that the proposed MTT, enables larger operating areas and longer times between observation through probabilistic updates of target tracks that have not been observed for some time.