Abstract: Particle filters have a high computational complexity when using quasi-Monte Carlo (QMC) methods, and are subject to the sample impoverishment caused by resampling step. To solve these problems, a new particle filter algorithm based on QMC method is proposed. It generates the randomized QMC points after sample importance, and then transforms them into some independent sub-spaces, whose kernels are the particles with heavy weights, to avoid predicting the sampling space and preserve the diversity of samples. The simulation results suggest that the algorithm can escape successfully from the sample impoverishment, provide more accurate estimators than the Monte Carlo (MC) method, meanwhile has a computational cost similar to the general particle filter.