Online-Learning Structural Appearance Model for Robust Visual Tracking

摘要

The main challenge of robust visual tracking comes from the difficulty in designing an adaptive appearance model to account for appearance variations. Existing tracking algorithms often build an representation for the tracked object, and perform self-updating of the object representation with examples from recently tracking results. Slight inaccuracies in the tracker can degrade the appearance models. In this paper, we propose a robust tracking method with an online-learning structural appearance model based on local sparse coding and online metric learning. Our appearance model employs structural feature pooling over the local sparse codes of an object region to obtain a robust object representation. Tracking is then formulated as seeking for the most similar candidate within a Bayesian inference framework where the distance metric used for similarity measurement is learned in an online manner to match the varying object appearances. Both qualitative and quantitative evaluations on various challenging image sequences demonstrate that the proposed algorithm outperforms the state-of-the-art methods.