Abstract:Concerning the uncertainty of video target state transition model and random noise distribution in actual environment, an algorithm used for video target tracking in complicated movement was proposed. The algorithm adopted the advantage of Kalman Filter (KF) which was of excellent real-time quality, and the advantage of Partical Filter (PF) which could deal with non-linear and non-Gaussian filtering at the same time. By analyzing the performance of KF and making its performance parameters as a determinstic term, KF and PF could be switched adaptively. By means of test, it is suggested that the method proposed in this paper can carry out steady tracking when the target motion state changes significantly, with high tracking accuracy.