Abstract: In view of the problem of satellite video point-target tracking, a method of Bayesian classification for tracking with the constraint of motion smoothness is proposed, which named Bayesian MoST. The idea of naive Bayesian classification without relying on any prior probability of target is introduced. Under the constraint of motion smoothness, the gray level similarity feature is used to describe the likelihood of the target. And then, the simplified conditional probability correction model of classifier is created according to the independence assumption Bayes theorem. Afterwards, the tracking target position can be determined by estimating the target posterior probability on the basis of the model. Meanwhile, the Kalman filter, an assistance and optimization method, is used to enhance the robustness of tracking processing. The theoretical method proposed are validated in a number of six experiments using SkySat and JL1H video, each has two segments. The experiment results show that the BMoST method proposed have good performance, the tracking precision is about 90% and tracking trajectory is smoothing. The method could satisfy the needs of the following advanced treatment in satellite video.

Key words: satellite video, point-target tracking, Bayesian classification, motion smoothness, SkySat, JL1H