关键词: 粒子滤波, 快速Metropolis-Hastings, 遗传算法, 红外弱小目标, 跟踪与检测

Abstract:

Crucial issues in a particle filter(PF) are to remove the degeneracy phenomenon and alleviate the sample impoverishment problem. In this paper, by using techniques from a genetic algorithm we propose some modifications to solve these problems simultaneously. A genetic algorithm with arithmetic crossover, fast Metropolis-Hastings mutation and the roulette wheel selection improves resampling procedures for the standard particle filter. The new particle filter is developed for IR dim target tracking and detection. By using multi-frame target amplitude and motion features some values of the filter's output are used to approximately construct the likelihood ratio for hypothesis test in the detection stage. Simulation results show that genetic resampling based on fast Metropolis-Hastings can produce various particles and remove the degeneracy phenomenon. For some actual image sequences with an SNR of 2.0, successful detection probability reaches 98.5％ with 10^-3 false alarm probability. Experimental results show that the performance of the proposed algorithm is better than that of traditional resampling particle filter algorithms.

Key words: particle filter, fast Metropolis-Hastings, genetic resampling, IR dim target, tracking and detection