Research On Target Tracking Algorithm Based On Nonlinear Filters

Target tracking has been found widely applicated in both military and civil areas, thus many countries and researchers have paid much more attention. Recently, there is more research on passive detection system in order to alleviate the challenge and threat of radar. However, the passive detection devices mostly provide the information of angle or gray, which will bring great difficulty with target tracking. Therefore, how to realize target tracking under the background refered above is a challengeable research topic.The dissertation mainly deals with the nonlinear filtering problem of multiple passive sensors target tracking and the track-before detect algorithms of weak target in the environment with low signal to noise ratio.The main contributions are as follows.Firstly, on the basis of the research on the newly proposed quadrature kalman filter and cubature kalman filter, combined with multi-sensor centralized fusion rule, the corresponding algorithms are proposed for multiple passive sensors, which solve nonlinear filtering problem effectively. The detailed approach is given to resolve the quadrature problem of the product of the certain function and Gaussian distribution.Secondly, combined with multi-sensor centralized fusion rule, a target tracking algorithm based on Quasi-Monte Carlo Gaussian particle filter is proposed for multiple passive sensors. The algorithm employs Quasi-Monte Carlo sampling to produce less but more regular distributed point set to replace conventional Monte Carlo sampling randomly generated point set, which solves the problem of performance degradation caused by the formation of Monte Carlo sampling"gaps and clusters". It not only reduces the computational complexity, but also improves the accuracy and stability of the passive tracking algorithms, thus has fast convergence speed.Thirdly, aiming at the problems of the unobserved radial distance and the unknown process noise intensity in a multiple passive sensor tracking system, combined with the multi-sensor optimal information fusion rule, a target tracking algorithm based on hierarchical particle filter is proposed for multiple passive sensor. The algorithm constructs two sub-states by rewriting the system equation under the logarithmic polar coordinates and adding the auxiliary parameter which is defined as the ratio of noise intensity to radial distance. The sub-states and the auxiliary parameter are estimated by the improved particle filter, avoiding the error resulting from using the maximum of noise intensity.Lastly, according to the detection and track of weak target in the environment with low signal to noise ratio, a recursive track-before-detect algorithm is proposed combining with marginalized particle filter and cusum joint detection. The method of cusum joint detection can improve signal to noise ratio through cumulating multi-frame measurements. Calculation of the disappearance statistic is started the moment a target is detected, which ensures the continuous detection of target appearance and the shortest delay for the detection of target disappearance.