Nonnegative coding based ensemble tracking

摘要: Abstract We describes a novel ensemble learning framework for tracking single visual object that, unlike existing approaches, relies on the modified nonnegative coding to select optimal subset of classifiers and determinate corresponding weights. The obtained classifier makes tracker be more robust. iteration update proof convergence solving objective function based are provided. For tracking, we use predicted labels generated by each selected individual compute correct classification rate, thence it identify occlusion, which is critical minimize drift. Evaluation performed fifty challenging benchmark sequences, shows our approach achieving or exceeding state art.