Study Of Tracking Before Detection Based On Random Finite Set Theory

Multitarget detecting and tracking has important application in both nationaldefense and civilian areas. But it becomes serious challenge in study since the numberof targets varied randomly in time, easily impacted by clutter, noise and otherbackground factors. Since the dim small target is difficult to come from demarcation ofbackground noise, it becomes a subject worthy of further research how to detect andtrack the multitarget in low signal to noise ratio environment. This paper focuses on themultitarget tracking algorithms based on Random Finite Set(RFS) theory and theirapplication to tracking before detection in the infrared dim small targets.First, this paper introduces the random set filter model, PHD filter and its twoimplementations, and studies the smoothing algorithm of GM-PHD filter, and clarifiesits inhibition effect of clutter by theoretical analysis and simulations, and proposed animproved GM-PHD filter against the defects of traditional algorithm.Second, this paper researches the application of PHD filter to multitarget TrackBefore Detect(TBD) in low signal to noise ratio environment. For the problems such aslow accuracy in tracking, high complexity and the sample impoverishment suffered bythe current Sequential Monte Carlo Probability Hypothesis Density (SMC-PHD) basedTBD algorithm, this paper replaces the Particle Filter with the Quasi Monte CarloGaussian Particle Filter(QMC-GPF) and proposes a new TBD algorithm based on PHD.Simulation results show that the proposed algorithm has solved the defect of theconventional methods. The new algorithm is more suitable for the engineeringapplicationFinally, this paper studies the multitarget tracking algorithm based on BernoulliRFS, focuses on the multitarget joint detecting and tracking algorithm based onMulti-Bernoulli filter and its two implementations, the particle filter implementationand the Gaussian particle implementation. Simulation show the pros and cons of the twoimplementations. For the problem that the current TBD based on Multi-Bernoulli filterfalsely merges the Bernoulli component from different targets, an improved algorithmis proposed. Simulation shows the improved algorithm perform well in the target crossscene.