With the development of computer vision and other related realms, visual object tracking has become a hotspot issue with universal concern, and it has shown a broad prospect in various applications. In this thesis, we propose a novel tracking algorithm based on the survey of existing methods and techniques. The contribution of the this thesis are fourfolded: Firstly, some visual object tracking algorithms published recently, especially the branch of tracking-by-classification, are reviewed in detail. And some related technics are also discussed to some extent; Secondly, in order to mitigate model drift problems, visual single object tracking algorithm with discriminative ranking lists is introduced, which is based on the concept of ranking list, purity and multi-scales; Thirdly, further work on this topic is also explored, including weighting different elements in the ranking lists and modeling confusing background, to enhance the robustness and adaptability of the tracker; Lastly, we build up a visual object tracking platform from the discriminative ranking lists based single object tracking algorithm and its related improvements. The proposed platform, equipped with proper system and data structure design and algorithm optimization, is maintainable and scalable.
修改评论