Likelihood-based Sensor Fusion in Radar/Infrared System Using Distributed Particle Filter

In this paper, the distributed data fusion problem in radar/infrared system which is composed of radar and infrared, is considered. Generally, the different dimensions of local measurements and the strong nonlinearity of infrared measurement equation are two major issues in radar/infrared system. For these issues, a parameterized likelihood-based distributed particle filter (P-L-DPF) algorithm is used, where the local likelihood function (rather than posterior or measurement) is regraded as the filtering results since the likelihood function can preserve the most original measurements information. Meantime, we approximate the likelihood function using polynomial expansion, and transmit polynomial coefficients to the fusion center, which efficiently reduces the transmission requirements. In the simulation, an example that a radar/infrared system tracks a moving target is given, the results show that the tracking performance of the P-L-DPF algorithm outperforms the posterior-based DPF (P-DPF) algorithm and is very close to the measurement-based centralized particle filter (M -CPF) algorithm.