Optimized instrumental density for particle filter in Track-Before-Detect

In this paper, we address the problem of the detection and tracking of a single target in Track-Before-Detect with a particle filter from radar measures. The studied particle filter solves the detection problem by introducing a Markovian variable that stands for the presence or absence of the target. We claim that the usual prior density used as instrumental density is not suitable here, and we propose an efficient particle filter based on a relevant proposal density based on detection and estimation considerations that aims at extracting all the available information from the measurements. This filter leads to dramatically improved performance compared to the particle filter based on the classic instrumental distribution, both in terms of detection and estimation.