An efficient particle filter based distributed track-before-detect algorithm for weak targets

An efficient particle filter based distributed track-before-detect (PF-DTBD) algorithm is presented in this paper. It key idea is the fusion of multi-sensor local estimated conditional probability density functions (PDFs). Firstly, the PDFs among sensors nodes are estimated by multivariate kernel density estimation (MKDE) technique based on finite particles set and fused to calculate the fused particle´s weight at fusion node. Next, according to Bayes rule, we prove that the unnormalized fused particle´ weight is actually composed of sensors´ local measurement likelihood, which makes the likelihood ratio test feasible at fusion node. Finally we introduce a detection scheme combining sequential probability ratio test (SPRT) and fixed sample size (FSS) likelihood ratio test to definitely realize TBD process for weak targets. Simulation results show our algorithm is efficient, which reduces delay of detection and improves the precision of state estimation simultaneously.