Kernel Regression Based Online Boosting Tracking

Although online boosting algorithm has received an increasing amount of interest in visual tracking, it is susceptible to class-label noise. Slight inaccuracies in the tracker can result in incorrectly labeled examples, which degrade the classifier and cause drift. This paper proposes a kernel regression based online boosting method for robust visual tracking. A nonlinear recursive least square algorithm which performs linear regression in a high-dimensional feature space induced by a Mercer kernel is employed to derive weak classifiers. Online sparsification to filter samples in feature space is adopted to reduce the computational cost of the recursive least square algorithm. In our method, weak classifiers themselves can be modified adaptively to cope with scene changes. Experimental results compared with several relevant tracking methods demonstrate the good performance of the proposed algorithm under challenging conditions.