Abstract
We propose a novel mapping method to improve the training accuracy and effificiency of boosted classififiers for object detection. The key step of the proposed method is a nonlinear mapping on original samples by referring to the basis samples before feeding into the weak classififiers, where the basis samples correspond to the hard samples in the current training stage. We show that the basis mapping based weak classififier is an approximation of kernel weak classififiers while keeping the same computation cost as linear weak classififiers. As a result, boosting with such weak classififiers is more effective. In this paper, two different nonlinear mappings are shown to work well. We adopt the LogitBoost algorithm to train the weak classififiers based on the Histogram of Oriented Gradient descriptor (HOG). Experimental results show that the proposed approach significantly improves the detection accuracy and training effifi- ciency of the boosted classififier. It also achieves high performance on public datasets for both pedestrian detection and general object detection tasks.