Particle filter based on the lifting scheme of observations

Recently, particle filter (PF) has been introduced as an effective means to solve the state estimation problems of non-linear and non-Gaussian system. If the observation sensor accuracy is lower, the measurement of importance weights according to the observation likelihood degree will result in the performance degeneration of general PF. Aiming at this problem, the authors propose a novel scheme of observations to enhance the reliability and stability of importance weights. In realisation of algorithm, firstly, a set of virtual observations are constructed on the basis of the current observation and the priori information of observation noise, also known as the accuracy of sensor. Secondly, combining with the distribution characteristics of virtual observations, the importance weights of particle are calculated by the weight fusion. From the derivation, it is easy to know that the variance of particles importance weights is decreased, and the adverse effect on importance weights from the randomness of observation noise is improved. The theoretical analysis and experimental results show that the proposed method outperforms the general PF.