This paper presents an online feature selection algorithm for video object tracking. Using the object and background pixels from the previous fr a me as training samples, we model the feature selection problem as finding a good subset of features