Improved mean shift target tracking based on self-organizing maps

Thanks to its simplicity and real-time processing possibility, mean shift has been widely used for video tracking. However, it often fails when the background is similar to the intended object or when the object is partially or completely occluded. To address these two problems, in this paper we propose a novel algorithm based on mean shift by exploring simultaneously the temporal and spatial information of the tracked object. A cascade classification method based on nearest neighbor and self-organizing maps is employed as a confirmation step to eliminate spurious objects through the structure information of the object. The forward and backward tracking results are further combined to improve the localization accuracy and tolerate at the same time scale variation. Experiments have shown clearly the superior performance of the proposed system in terms of accuracy, stability and robustness.