An efficient tracking system using adaptive approach for online feature selection

Existing online feature selection algorithms have shown good performance in tracking objects. But, there is still issue in how to choose the optimal number of features. If there is a wide difference between object and background, the tracking system using a feature can shows a good performance. In contrast, many features are needed to increase the performance when there is uncertainty of object and background. Therefore, it is reasonable to dynamically switch the number of used feature to follow the object. In this paper, we propose an efficient tracking system using adaptive approach for online feature selection.