Track Before Detect Algorithm For Infrared Dim Target Based On Improved Particle Filter

In modern warfare, the key to win is who can detect the enemy threat earlier. Comparedto the active tracking mode of early radar systems, the infrared detection system’s passivetracking mode has a broader application in the military field. But gray images obtained by theinfrared detection system, which contain small targets often exist problems of non-linearobservations and low SNR. These bring serious challenges to small target detection andtracking. The main purpose of this paper is to provide an effective method to solve smallinfrared target detection and tracking problems in the complex background.Track before detect (TBD) algorithm first tracks all suspicious targets which containednoise, then get rid of false targets by continuous filtering, finally reach the purpose of smalltarget detection and tracking. This paper uses particle filter (PF) which applies to nonlinearsystems to achieve small infared target detection and tracking. In order to reduce thecomplexity of the algorithm and improve the tracking accuracy, PF improved algorithm ispresented according to characteristics of the target tracking model.This paper first introduces the basic theory of particle filter, and then discusses itsexistting problems and what improved methods can solve these problems. Simulation resultsshow that the particle filter algorithm have the superior performance in the case of nonlinear,non-Gaussian. Based on these achievements, marginalized particle filter algorithm isresearched because target tracking models often contain both linear and nonlinear state vector.This algorithm respectively estimates linear and nonlinear state by means of optimal Kalmanfilter and particle filter. At the same time, introduce MCMC steps to alleviate the problem ofparticle degradation. Besides, this dissertation proposed an effective method by adjusting thenumber of particles adaptively to reduce the algorithm’s computation. Simulation results aboutpure angle target tracking verify the proposed algorithm is effective. And then TBD algorithmbased on particle filter is studied. Create a virtual infrared simulation scene, compare thetracking performances of PF-TBD algorithm, MPF-TBD algorithm, MPF-MCMC-TBDalgorithm and AMPF-MCMC-TBD algorithm in different conditions. Finally differentalgorithms are applied into32real infrared image sequences. Before compare different algorithms’ tracking precision and real-time, grayscale morphology filtering is used to inhibitmuch background noise in infrared images. Experimental results show that theAMPF-MCMC-TBD algorithm has better comprehensive properties.