The treelike assembly classifier for pedestrian detection

Abstract

Until now, classification is a primary technology in Pedestrian Detection. However, most existing single-classifiers and cascaded classifiers can hardly satisfy practical needs (e.g. false negative rate, false positive rate and detection speed). In this paper, we proposed an assembly classifier which was specifically designed for pedestrian detection in order to get higher detection rate and lower false positive rate at high speed. The assembly classifier is trained to select out the best single-classifiers, all of which will be arranged in a proper structure; finally, a treelike classifier is obtained. The experimental results have validated that the proposed assembly classifier generates better results than most of the existing single-classifiers and cascaded classifiers. © Springer-Verlag Berlin Heidelberg 2007.