Research On Target Tracking Based On Sampling Nonlinear Filter

With the development of technology, target tracking has been widely applied to military and civilian areas. And at the same time, the surrounding become more complex and the promotion of target tracking is required. As the core of target tracking, filtering algorithm has gradually become a research hotpot for domestic and international experts. Nonlinear filtering algorithm has better tracking performance than Kalman algorithm when system is nonlinear. Among all the filtering algorithms, the sampling nonlinear filtering algorithm has been an important part of target tracking application. In allusion to multi-sensor data fusion and the optimization of algorithm performance, some new algorithms are presented based on nonlinear filtering in this paper. And the results indicate the proposed methods have validity. The main contributions of this dissertation are summarized as follows:A new method, federated UKF target tracking algorithm, is put forward for the normal UKF filter with Gaussian noise. This algorithm is suitable for multiple motion models. With the algorithm, the noise influence on measurement was reduced by multiple sensors observation fusion, and then the tracking performance of algorithm is improved.In order to solve the problems of particle degradation or diversity depletion after resampling, a hybrid leapfrog cost assessment particle filter algorithm is proposed. By introducing the idea of hybrid leapfrog into particle filter, and using hybrid leapfrog mechanism to find the optimal particle set, the particle degradation is avoided effectively. This results show that new algorithm has made effective use of advantages of hybrid leapfrog, and filtering precision is improved at the same time.Aiming at solving particles degradation phenomena, which was caused by the different contribution of particles to state estimation, a new ensemble Kalman filtering algorithm based on the optimization of particle set has been proposed. Firstly, contribution for the state estimation of current moment is reasonably evaluated though introducing importance sampling. Secondly, overall optimization of particle set is achieved based on single particle validity to state approach, through combining with importance resampling method, and then increasing the number of high weight particles, reducing the number of low weight particles.