Research On Techniques Of Detection And Tracking Of Multiple Dim Moving Targets From IR Multispectral Image

Detection techniques of multiple targets including decoys become more and more important due to the development of target stealth and decoy interfere. However, the current broadband infrared detection technique can not effectively solve the problem of detection and tracking of low signal-to-noise ratio targets immersed complicated background. Multispectral images are used to detect and track multiple dim moving targets with low signal-to-noise ratio in this dissertation. The main contributions of the thesis are as follows:The multiscale decomposition of infrared multispectral images were performed by the bidimensional empirical mode decomposition(BEMD) due to its ability of dealing with nonlinear and non-stationary data. On this basis, a novel anomaly detection algorithm was proposed based on BEMD to detect IR dim targets. Because the multiscale information of infrared multispectral image was used, this algorithm can improve detection performance via suppressing background clutter and subtracting high-frequency noise. Experimental results showed that compared with the traditional anomaly detection algorithm, the presented algorithm has a better ability to detect dim targets. Meanwhile, the proposed algorithm in this chapter can provide the priori for detection and tracking of dim targets from sequence image.The detection and tracking of single dim moving target in infrared multispectral image sequence was investigated. The dynamics model of single target and measurement model were proposed. A new particle filter tracking algorithm was proposed based on the local trajectory which was produced by the distributed estimation fusion strategy. The sequential measurement fusion strategy was utilized to propose a novel particle filter detection algoritm. The sequential measurement fusion strategy can improve detection performance because of using the raw measurement. The proposal density fuction of the above two algoritms was designed according to the test statistics of BEMD. This proposal density can reduce tracking area of particl and enhance the computing efficiency. The proposed two algorithm can detect and track dim moving target under low signal-to-noise ratio via multispectral data fusion.In order to detect multiple dim moving targets in infrared multispectral image sequence, the multi-target dynamics model and measurement model were presented based on single target dynamics model and measurement model. The target existence was modeled by Markov chain. The joint multi-target detection and tracking can be formulated by the hybrid filtering. The hybrid filtering was used to derive the probability of target existence and state estimation. Then the multiple targets track-before-detect algorithm were proposed according to Bayes filtering framework. This algorithm gave the analytic solution to the posterior probability of target number. The algorithm was implemented by multiple interactive mixing particle filtering. Simulation results demonstrated that the proposed algorithm can detect multiple dim moving targets in infrared multispectral imagery sequence.The tracking of multiple dim moving target in infrared multispectral image sequence was investigated. Because the probability hypothesis density filtering can solve the problem of multi-target tracking in single target state space without data association and reduce computation, the random finite set theory and PHD filtering were firstly studied. Secondly the implementation of probability hypothesis density filtering based on sequential Monte Carlo was derived. On this basis, the probability hypothesis density filtering track before detect algorithm was proposed. Then adaptive multispectral probability hypothesis density filtering track before detect algorithm was proposed by using centralized fusion strategy. The detection probability can be calculated according to image data. The simulation results showed that the presented algorithm can track multiple dim moving targets in infrared multispectral image sequence.