Research On Track-before-detect Algorithms For Multiple-target Detection And Tracking

The detection and tracking of low observable targets (e.g. stealth targets, targetsburied in strong clutter) in complex environment are great challenges for modern radarsystems. Different from the traditional detect-before-track (DBT) methods,Track-before-detect (TBD) is a novel and efficient signal processing method which isproposed in recent years to detect low observable targets, and has got much attentioninternationally in radar research area. By jointly processing several data frames, TBD isable to produce more reliable detection and tracking results. As a developing newtechnique, TBD has its own problems and challenges. For example, existing studies onTBD are focusing on the single target tracking problem. Because of the curse ofdimensionality and the performance degradation due to the interference of closelyspaced targets, the multi-target tracking problem is always a big challenge for TBDmethods. Besides, its computational expense and algorithm complexity are muchheavier than DBT methods.Regarding the challenges mentioned above, in this thesis, the multi-target TBDalgorithm and its application in surveilence radar systems are studied. The maincontributions of the thesis are as follows:1. The mathematical formulation of multi-target tracking problem is given basedon the Bayesian estimation theory. Then the traditional DBT method and TBD methodare united and discussed under the framework of Bayesian estimation theory. Thischapter makes the theoretical foundation of the subsequent chapters of this thesis.2. According to the curse of dimensionality and the performance degradation dueto the target interference when targets are in proximity, a novel multi-target Dynamicprogramming (DP) based TBD algorithm is proposed in this chapter. By comparisonwith the existing SP-STC-VTA algorithm, we show that the proposed algorithm has lessfalse tracks, less computational complexity but has better tracking performance.3. In order to tracking unknown and time-varying number of targets, a novelparticle filtering (PF) based TBD algorithm which uses independent and joint optimalimportance density (IJOID) sampling method and two-layer structure is proposed. This PF tracking algorithm has lower computational complexity, better detectionperformance and can deal with the target interference when targets are near each other.4. With the purpose of reducing the computational complexity of the DP-TBDalgorithm, a thresholding process based DP-TBD algorithm is proposed. Besides, theapplications of the DP-TBD on the surveillance radar system are also studied.The above tracking methods and corresponding engineering implementationscheme are tested using both simulated data and real data collected from surveillanceradar. The results demonstrate the efficacy of the proposed algorithm andimplementation scheme.