Multi-target tracking with background discrimination using PHD filters

In this paper, we propose a new double PHD filter for simultaneous multi-target tracking and background discrimination for airborne radar applications. Both the foreground and the background processes are modeled as Poisson point processes, which gives a symmetric formulation of the coupled filters. The differences between foreground and background lie in the assumed target dynamics, and in the sensor detection probabilities. Although there are proposals for PHD filter with adaptive background models in the literature, our filter appears to be novel and also the simplest possible. To implement the filter we use a Gaussian mixture approximation of the intensities, which enables simple and effective ways to extract tracks. For the evaluations we use a simulated target tracking scenario with an airborne radar tracking a number of flying targets over a background of road objects. First, the performance of the Gaussian mixture PHD filter with track extraction is illustrated. Second, the superior ability of the foreground-background PHD filter to suppress clutter and disturbing road traffic is illustrated.