A method of maneuvering target prediction is presented for vehicle-mounted image tracking system. The target motion is divided into two parts, namely global motion and local motion, which are separately predicted according to their intrinsic dynamic behavior. The local motion prediction is described emphatically in this paper. A high-performance filtering tracker, which was then extended into a predictor to estimate the future moving state of target, was built up through the research on the maneuvering target tracking (MTT) technique, thereby the “interacting multiple-model(IMM) prediction algorithm” is put forward. As the simulation shows, this algorithm has high precision for predicton and good self-adaptive capability. Besides, the method of the prediction gate computation is given in this paper.