Improved Particle Filter Based on Kernel Density Estimation

Particle filter (PF) algorithm can be used for nonlinear and non-gaussian system state estimation, the core lies in how to accurately describe the posterior density function with limited particles. In recent years, there have been many improved particle filter algorithms, but the effect is not good in the strong nonlinear/non-Gaussian environment. The improved method proposed in this paper can overcome this difficulty. This method uses kernel density estimation to fit the priori PDF, then evenly sample new particles in a reasonable area and calculate their priori weights. Then these particles' weights are updated according to the likelihood function, and then normalize the weights, and the state is estimated by using the particles and their normalized weights. This method is called KDE-PF. Simulation results show that this method is better than the general particle filter algorithm.