An objective tracking method based on Kalman filter
Yung-Yao Chen, Po‐Han Chen, Shih-Che Chien · 2016
This paper presents a novel object tracking system that combines support vector machines (SVM) and Kalman filter. Objective tracking in videos is a challenging problem due to loss of information, which may be caused by varying illuminance in a scene, occlusions, similar target appearances, and so on. In this paper, we use Kalman filter to predict the dynamics of target object, so as to generate candidate blocks of the target object. Then, structured SVM is applied to classify those blocks and estimate the location of the target object. However, when the target is away from (or near to) camera, the tracking frame does not conform to a fixed size. Therefore, we add background estimation to adapt the size of a tracking frame. Experimental results demonstrate the robustness of the proposed system.