Research On Target Tracking Based-on Particle Filter Algorithm

Automatic sequential image target tracking is of vital importance in both civil and military use. Demand of researching and developing tracking algorithms and systems with high reliability and high real time have been more and more in these years. Kalman Filter and Extended Kalman Filter are the most typical filter algorithms in target tracking domain. Although the above two methods own pretty good filtering performance when system noise and observation noise are non-Gaussian, their filtering performance will descend or even diverge when non-Gaussian distribution occurs. Because vision procedure is actually nonlinear/ non-Gaussian, using the corresponding techniques become a kind of important research tend in image fields. Particle filter is one of theories as description above, and has distinguishing features. Particle filter algorithm is main content of this paper, which can be implemented by a Bayesian recursion process though a Monte Carlo simulation method.In this paper key issues of applying the theory to target tracking are studied. The core content in realizing target tracking with this theory, which is used in this paper is to construct a target tracking model and to form a corresponding automatic process. Two important components of this particle filter are state transition model and observation model, which are built according to the results of target detection[1]. Based-on the target detection, the picked-up characteristic information of colour and texture is the sequential observation datum. Given that the position speed and acceleration of the moving target is the unknown motion state and that observation datum, a particle filter based on motion detection is built, and automatic target tracking turns into reality with it at the last.The approach developed in this paper is experimented to a variety of video sequences. By experimental[11], this method is proved much effectively, and the accuracy of tracking is improved greatly.