Researches On Particle Filter-based Track-before-detect

With the increase of military and civil demands and the rapid development of science and technology, the techniques of detecting and tracking dim targets are getting more and more important. In the field of the detecting and tracking of dim targets,traditional detect-before-track(DBT) has poor performance because of its single frame threshold decision. On the contrary, a technique named track-before-detect(TBD)without single frame threshold decision arises at the historic moment. TBD retains the maximum raw observation information and realizes the detecting and tracking of dim targets by means of the multiframe accumulation.Particle Filter(PF) is confirmed as an efficient method to realize TBD because of its recursive nature and suitableness for non-Gaussian and nonlinear application. In this dissertation, the particle filter based track-before-detect technique(PF-TBD) is studied and the main contents are summarized as follows:(1) The theory of two basic PF-TBD methods named SPF-TBD and RPF-TBD have been studied and the methods to build TBD models of radar and infrared have been introduced. The simulation comparison between SPF-TBD and RPF-TBD shows that RPF-TBD performs better than SPF-TBD both in detecting and tracking.(2) A novel BOX-PF-TBD algorithm is proposed by using BOX particle which is based on interval analysis theory and Bayes theory. The simulation comparison between RPF-TBD and BOX-PF-TBD is made and the results show that BOX-PF-TBD runs about 14 times faster than RPF-TBD when achieving the same detecting and tracking performance.(3) The PF-TBD algorithms used in multi-target scenarios are studied and the radar multi-target TBD model is built. Simulations are made to show the detecting performance of both multi-target RPF-TBD and multi-target SPF-TBD.(4) Several improvements of the PF-TBD are made in the engineering applications.The methods to estimate the noise power and the distribution of the target returning power are researched. In consideration of the broblem that the particle weights would overflow during the filtering process in high signal-to-noise ratio(SNR)scenarios, a numerical improved method is proposed. To show the advantages of the TBD technique in low signal-to-noise ratio scenarios, The improved BOX-PF-TBD andthe traditional CA-CFAR methods are used to deal with the same measured data of radar.The results show that in low signal-to-noise ratio(SNR) scenarios, BOX-PF-TBD algorithm has better detecting and tracking performance than traditional DBT methods such as CA-CFAR.