Bag of Features with Dense Sampling for Visual Tracking ?

The bag-of-feature model has become a state-of-the-art method of visual classification. Visual codebooks can be used to capture image statistical information for object detection and classification, which is extracted from local image patches and based on the quantization of robust appearance descriptors. In this paper, more information of target objects can be captured by dense sampling rather than sparsely sampling. Then a robust visual tracking method is proposed based on dense sampling and bag of features. Firstly, local image patches are densely extracted by sliding windows and represented as invariant descriptors. Secondly, visual codebooks are generated by fast clustering algorithms such as hierarchical k-means. Therefore, the object region and candidate regions are represented by the bag-of-feature model with the learnt codebooks. After that, tracking can operate in a Bayesian inference framework. The bag-of-feature tracking method with dense sampling is adaptive and flexible. It works independently in many situations without the complement of existed tracking algorithms. The experiments on various challenging videos demonstrate that the proposed tracker outperforms several state-of-art algorithms.