Double-Directional Bernoulli Track-Before-Detect Filter With Particle Flow

Track-before-detect (TBD) aims to detect and track a weak target by integrating measurements over time and position. The key prerequisite to the integration is the accurate representation of the target measurement. However, the TBD methods implemented by the particle filter rely too much on the proposal density, and thus the representation of the measurement is not accurate enough. Therefore, this paper proposes a novel TBD method, which incorporates the newly developed particle flow filter into the Bernoulli TBD filter to improve the measurement representation. Moreover, the poor initial state estimation is also refined by a double-directional filtering solution. The effectiveness of this method is demonstrated on the simulation results of a fluctuating weak target tracking in K-distributed clutter.