The Research Of Multi-sensor Target Tracking Algorithm Based On Information Fusion

The target-tracking problem has attracted lots of atentions for its wide applications in the military and civil affairs, and the maneuvering target tracking studies have especially become a hotspot to study in this field. At present, the researches of the target tracking are mainly concentrated on the maneuvering target tracking and multisensor information fusion. The aims of this paper is to research the maneuvering target tracking algorithms,but the algorithms exist some problems. Some improved algorithms with information fusion technology is introduced. the performance of the algorithm is verified through many simulations.Kalman filter has optimal estimation for linear Gaussian system. The maneuvering target tracking system is always non-linear. Extended Kalman filter is the most commonly used algorithm to solve the nonlinear system. Idea of the algorithm is a nonlinear system for the linear treatment, and then go on with recursive filtering based on Kalman filtering framework. When non-linear degree of system is lager, the system linearization error will also be larger, sometimes even divergent.To solve the problem, UKF algorithm becomes by the U transform used in the Kalman filter algorithm. And 2N Sigma-point used to describe the system status (N is the dimension of the status),it is not necessary for the linearization treatment. Compared with the EKF, the UKF is easy to achieve unified, stable performance and so on. It is entirely possible to replace the EKF to become a commonly used non-linear filtering algorithm.The particle filter is one of the most popular maneuvering target tracking algorithms.PF has no limits in non-linear/non-Gaussian so that PF has broader application compared with the EKF and UKF. However, the PF algorithm is easy to appears particle degradation phenomenon, the importance selection function plays a vital role in the particle degradation. To overcome the problem, the EPF and UPF algorithms are improved based on PF by using EKF and UKF algorithms to producethe the importance function. The EPF and UPF. algorithms are better than PF. Through simulation experiments, we can see that the performance of UPF and EPF are better than PF, UPF is the best of three algorithms, but it uses the longest time so less real-time.Finally, many models of multi-sensor maneuvering target tracking algorithm is studied.At present the IMM algorithm is the mainstream of multi-model algorithm. In this paper, a fusion algorithm - interactive multi-model unscented particle algorithm (IMM-UPF) is proposed by combinating the advantages of UPF algorithm and IMM algorithm. The Model Set contains of three non-linear models according the characteristics of the filters.The models work parallelly based on Markov transition probability. through many simulations,the results prove the effectiveness of the and the performance of the new algorithm is even better than the UPF.