Improving particle filter with a new sampling strategy

Particle filter has many variations, one of which is the unscented particle filter. The unscented particle filter uses the unscented Kalman filter to generate particles in the particle filtering framework. This method can give better performance than the standard particle filter in some practical problems that are raised in computer vision field. But one critical issue in the unscented particle filter is that it has very high computational complexity which constrains its broader application. In this paper, we give an improvement strategy aiming at reducing the computational complexity of the algorithm. This strategy combines the general framework of particle filtering with the transition prior and the unscented Kalman filter, taking advantage of the low computational complexity of the standard particle filter and the high estimation accuracy of the unscented particle filter. The experimental results show that this strategy can reduce the running time cost of the unscented particle filter greatly without loss of accuracy.