Probability hypothesis density filter for multitarget multisensor tracking

Multiple target tracking techniques require data association that operates in conjunction with filtering. When multiple targets are closely spaced, the conventional approach (MHT/assignment) may not give satisfactory results, mainly due to the difficulty in deciding the number of targets. Recently, the first moment of the "multi-target posterior density", called the probability hypothesis density (PHD), has been proposed to address the multi-target tracking problem. Particle filtering techniques have been applied to implement the PHD based tracking. In this paper, we explain our interpretation of the PHD, then investigate its performance on the problem of tracking unresolved targets from multiple sensors. In the set-up, there are two different radars which monitor the targets, and the PHD is fed sequentially by these scans. In the scenario, we investigate 3 different levels of complexity in terms of measurement extraction methodologies of sensors when there are unresolved targets 1. Sensor model reports a measurement with variance sigma(2)(mono). (Sensor is not capable of sensing any abnormality in radar return). Sensor model gives a single measurement with sigma(2)(azi) >= sigma(2)(mono) a larger variance sigma(2)(azi) >= sigma(2)(mono) 3. Sensor model uses a multi-target measurement extractor Unresolved targets 2 create separate measurements with variance sigma(2)(mono), Simulation results for two-dimensional scenario are given to show the performance of the approach. Based on our simulation results, we also discuss difficulties the PHD algorithm seems to encounter especially as is reflected in the target "death" event.