Target Tracking Algorithm Research Based On Particle Filter

Object location and tracking is the typical dynamic system state estimation problems. Many filtering algorithms can get perfect tracking results within the linear and Gaussian model; But when the objects are in the background of higher maneuvering, multi-model, non-Gaussian, strong noise, the algorithms of kalman filter and extended kalman filter within the Gaussian background will appearance the filter precision decrease and divergence phenomenon. As a nonlinear filter algorithm based on Bayesian estimation, particle filter has original advantage at treating the parameter estimation and state filtering aspects of nonlinear non-Gaussian time-varying systems. Thus a great development is obtained. This paper proposes several improved particle filter algorithms, and main research work is as following:1. The paper firstly introduces some filter algorithms about target tracking, such as extended kalman filter (EKF), unscented kalman filter (UKF) etc, which obtain well tracking results in the field of object tracking. But they all have the shortcomings of the low precision and the inadaptation to models.2. And then, the paper mainly introduces the particle filters, and proposed several improved filtering algorithms.Firstly the paper presents a new algorithm, which combines the particle filter algorithm with EKF filter algorithm. When the new algorithm calculates the proposed probability density distribution, the sampling particles can utilize the system current measures. That gets the particles distribution more approach to the station posterior distribution.Then this paper combines the particle filter algorithm with UKF filter algorithm. Owing to the transition prior does not take into account the current observation, so the effect of lots of particles could be negligible. Instead of using transition prior as proposal distribution, UKF is used to generate the proposal distributions, which takes the current measurement into account so as to improve the tracking performance greatly with fewer particles.The last, based on the two improved algorithms introduced above. The paper presents a new algorithm combined with Markov Chain Monte Carlo (MCMC).In the field of target tracking ,the MCMC filter algorithm is used as the following two aspects:1.Finding the Maximum a posteriori estimate; 2.Based on the generic particle filter frame, it can make the samples more various.