Robust visual tracking via online informative feature selection

摘要: An efficient and effective algorithm which online exploits informative features for visual tracking is presented. First, a high-dimensional multi-scale spatio-colour image feature vector developed, takes into account both appearance spatial layout information; secondly, this randomly projected onto low-dimensional space, where its projections preserve intrinsic information of the but effectively avoid curse dimensionality; finally, an selection technique to design adaptive model proposed, explores most from via maximising entropy energy. Experiments on extensive challenging sequences demonstrate superiority proposed method over some state-of-the-art algorithms.