Bayesian multiple target tracking in forward scan sonar images using the PHD filter

A multiple target tracking algorithm for forward-looking sonar images is presented. The algorithm will track a variable number of targets estimating both the number of targets and their locations. Targets are tracked from range and bearing measurements by estimating the first-order statistical moment of the multitarget probability distribution called the probability hypothesis density (PHD). The recursive estimation of the PHD is much less computationally expensive than estimating the joint multitarget probability distribution. Results are presented showing a variable number of targets being tracked with targets entering and leaving the field of view. An initial implementation is shown to work on a simulated sonar trajectory and an example is shown working on real data with clutter. © IEE, 2005.