Stabilizing Object Tracking Trajectory on the Basis of Kalman Filter

Object tracking is an important and classic problem in the computer vision area, which has widespread applications in many fields. Currently, most of state-of-the-art object tracking approaches pay more attention on improving the localization precision by dealing with some complex transformations such as illumination change, occlusion and deformation. However, smoothness of object trajectory doesn't attract enough attention. In some application scenarios like weapon-aiming systems, the object tracking systems are generally associated with mechanical control systems, and frequent vibrations will lead to violent mechanical shaking of controllers and result in uncontrollable consequences. In this paper, we focus mainly on improving the smoothness of object trajectory. In particular, we first propose a new criterion to evaluate the level of the vibrations. Then, a simple but effective approach, which is based on Kalman filter, is proposed to remarkably reduce the vibrations in both horizontal and vertical directions. Extensive experiments show that introducing the proposed smoothing method into existing object trackers can significantly improve the smoothness of tracking trajectories.