Research On Target Tracking Algorithms Based On Particle Filter Method

Video-tracking technology is very important in both military and civil use. Researching and developing tracking algorithms and systems with high reliability and high real time have been more and more useful in these years. The theory of filter has been widely used in tracking area. Particle filter realizes recursive Bayesian filter via Monte Carlo simulation. The method is suitable for any non-linear system that could be represented with state space model. It is more practical than conventional Kalman filter and its precision could approach optimal estimation. Particle filter also has parallel and open structure. Because it has so many advantages, Particle filter has been more and more used in tracking algorithms.Particle filter algorithm and derivation of non-linear Bayesian filter theory are researched in this thesis. The tracking framework is formed according to Monte Carlo simulation. The target is denoted with grey feature, color feature and SIFT feature. And the particle filter algorithms based on region-match tracking, color-feature tracking and SIFT-feature tracking methods are discussed. The tracking result and the data analysis show that the three tracking methods can achieve good effect in both translation space and affine space. Because the particle filter has the feature of'multi-summit', the three methods in this thesis can also achieve good tracking result when the target is blocked.A set of target auto-tracking device using camera and pan/tilt is constructed in this thesis. The three methods proposed in this thesis and the CamShift algorithms are transplanted to the device. The system software and hardware structures are given and the tracking result is studied. The result shows the three methods can achieve a stable target tracking on this device.