Detection Algorithm Based On Bayesian Filtering Tracking

Long distance dim small target detection and tracking is a key technology in such systems as infrared surveillance and tracking system, precision guidance system, wide field-of-view target surveillance system, and satellite remote sensing system. Target is so small and target signal is so weak relative to background clutter and noise that image signal-to-noise ratio (SNR) is very low. Study on real-time detection and tracking algorithms for moving small target in low SNR is very prospective.In this thesis, in-depth research work has been done on Track before Detect (TBD) for dim small moving targets in low SNR environment, and the main contributions are as follows:1. For dim small moving target detection problems, the mainstream algorithms of TBD are reviewed in details, including main ideas, research development, merits, shortcomings and conditions.2. In different consideration of target birth or death, Baysian TBD framework can be divided into Salmond-TBD framework and Rutton-TBD framework. In two frameworks, UPF based Salmond-TBD algorithm and UPF based Rutton-TBD algorithm are proposed. Instead of particle filter, Baysian TBD framework is implemented by UPF. Simulation results indicate that UPF can effectively solve TBD problems and detect moving small target in low SNR.3. After providing performance evaluation, four implemental algorithms in Baysian TBD framework are compared. Performances of these algorithms under different SNR conditions and effect of particle number are analyzed. Simulation analysis shows that UPF based TBD greatly improves tracking precision and detection performance, and Salmond-TBD framework is more reliable and has better real time ability than Rutton-TBD framework.