Genetic Particle Filter(GPF) for Online Non-Linear/Non-Gaussian Bayesian Tracking

The dynamic state estimation problems can be efficiently solved by Bayesian methods. The Bayesian approach is to construct the probability density function (pdf) of the state vector based on all available information. This pdf plays an important role to approximate the current state location of knowledge about the state vector. In the state-space approach to model a non-linear dynamic system, for the target tracking problem, the particle filtering has been successfully proven. In the standard particle filter, a resampling scheme is used to reduce the degeneracy problem and improve the estimation performance. But this resampling introduces some problems which is called sample impoverishment or particle deprivation. Genetic Particcle Filter (GPF) can overcome this problem in standard particle filter by using some genetic operatiots like crossover, mutation in prediction steps and roulette-wheel samping in genetic algorithm. M. Z. Islam, C. Oh, C. Lee, Genetic Particle Filter(GPF) for Online Non-Linear/Non-Gaussian Bayesian Tracking, The 6th International Conference on Ubiquitous Robots and Ambient Intelligence (URAI 2009), pp. 554-557, Oct. 2009