Research On Algorithm Of Passive Sensors Multi-target Tracking Based On Particle Filter

Multi-target tracking has broad application prospects in whether civilian or military areas, and has been regarded as a research focus and difficulty by scholars and researchers. In military applications, since the low survival rate of traditional active detection equipment such as radars and the scarcity of single sensor's information, passive multi-sensor multi-target tracking plays more and more important practical significance in constructing a solid national defense system. Generally, there is only angle information obtained by passive sensors, which means that nonlinear relation is existence between targets and measurements, so a variety of non-linear filtering methods need to be researched.To solve the above problems, this thesis mainly researches on passive sensors multi-target tracking based on particle filter, the main achievements are as follows:1. To deal with the problem of passive multi-sensor multi-target tracking with for invariant target number, iMC-JPDA and jMC-JPDA filter algorithms are researched, in which particle filter is introduced on the basis of joint probability data association(JPDA), employing independent sampling and joint sampling technologies and utilizing multi-sensor centralized fusion strategy. A passive sensors multi-target tracking algorithm based on hybrid sampling particle filter is proposed in view of the poor performance of jMC-JPDA in weak correlation targets tracking, in which samples distribution is optimized by introducing particle swarm optimization and decomposing joint sample weight before sampling the target samples independently. The proposed algorithm has a good performance in tracking accuracy.2. Taking into account the passive multi-sensor multi-target tracking for variant target number, the particle PHD filter algorithm is researched combing multi-sensor centralized fusion strategy. Whereas, the need of non-null measurement noise and analytical measurement likelihood restricts its application domain. To solve this problem, a convolution kernel particle PHD filter algorithm is researched by introducing the convolution kernel density estimation theory and rewriting the measurement likelihood in sample weight update process of PHD. Furthermore, considering the advance of data association technologies in accurate partition the targets and measurements in traditional multi-target tracking methods, an improved convolution kernel particle PHD algorithm is proposed for passive muti-target tracking. In the proposed method, the corresponding relation of estimated targets and measurements is confirmed by a reasonably minimized cost, so that the estimated states can be corrected furtherly. Simulation results show that the proposed method for multi-target tracking has better performance for the case of small measurement noise.