Research On Particle Filter Based Track Before Detect Algorithm For Maneuvering Weak Target

Detecting and tracking weak target in the strong clutter background hasattracted much more attention in the field of radar defense system for many years.Track before detect (TBD) seems to be a good approach to solve this kind of problem.In order to avoid being detected by radar and to meet other military requirements,targets are more likely to undertake short and strong maneuvers. How to detect andtrack those targets accurately in the complex background is of great significanttherefore.Particle-based TBD (TBD) is available in many weak target detecting andtracking problems whether the models are linear-Gaussian or not. As the basicknowledge, TBD models for maneuvering targets are established firstly. The staticmultiple model (SMM), interacting multiple model (IMM) and variable rate particlefilter (VRPF) are applied into TBD. The main works of this dissertation are listed asfollows:(1)PF-TBD theories and methods are introduced for the non-maneuveringtarget as well as the basic motion models and measurements models. Performance ofthe PF-TBD with different cell selections, signal to noise ratios (SNR) and particlenumbers are discussed in the simulations.(2)For the detecting and tracking of the maneuvering target, Multiple-modelparticle filter based TBD (MMPF-TBD) and the Interacting multiple-model particlefilter based TBD (IMMPF-TBD) are introduced and analyzed in details.IMMPF-TBD estimates the maneuvers that being undertaken through multiple mixingfilters. The output of every filter is verified by others and then used as the input ofitself at next measurement time. Simulations show that IMMPF-TBD can detect thetarget timely and estimate the maneuvers accurately.(3)VRPF-based TBD is present as a novel approach to detect and track theweak targets that behave short maneuvers in the strong clutter background. Thesampling rate of states is no longer the same with that of measurements. Instead, itbecomes a variable which is subjected to the gamma distribution. Fewer state pointsare used to track the line trajectory while more points are allowed to get the regions ofmaneuvers. An interpolation function is used to construct the neighborhood that isused to compute the particle weights. Simulation shows that it can use less storage forparticles than PF-TBD.