HYBRID PSO–GA PARTICLE FILTERS WITH SORTED LOOKAHEAD WEIGHTING AND MINIMALISTIC RESAMPLING FOR TRACK-BEFORE-DETECT APPLICATION

Abstract

 In the Track Before Detection (TBD) concept, the target is observed in an environment characterized by a low signal-to-noise ratio prior to its identification as a target. TBD finds application in various fields such as air traffic control, maritime surveillance, underwater autonomous vehicle (UAV) tracking, and the detection of small objects in remote sensing. A fundamental technique employed in TBD is the particle filter (PF). In nonlinear and non-Gaussian environments, the PF serves as a prominent method for target detection. However, the standard Particle Filter (PF) algorithm experiences particle diversity loss due to degradation and resampling, which hinders the particle set from accurately reflecting the true state probability density function. The Particle Swarm Optimization (PSO) algorithm can alleviate particle degradation in PF; however, its performance is significantly affected by measurement noise variance and is susceptible to becoming trapped in local optima, which restricts its filtering accuracy. To tackle these challenges, a hybrid approach that integrates the Genetic Algorithm (GA) with PSO is proposed. By merging the rapid convergence of PSO with the robust global search capabilities of GA, this method enhances particle diversity, maintains the effectiveness of high-quality particles, and improves overall accuracy. GA utilizes global search operations, including selection, crossover, and mutation. In this paper, we propose two selection mechanisms: sorted weighting lookahead and minimalistic resampling schemes. The integration of the optimization algorithms PSO and GA with these selection mechanisms will enhance particle diversity, mitigate the issue of sample impoverishment, and improve the root mean square value. The tracking accuracy of the proposed methods is evaluated in this paper using two nonlinear models.