Research on branch and bound for pedestrian detection

Object detection and localization is an important research subject in computer vision. In pedestrian detection, the task of classification is to decide whether an object is present or not in current scanning window, while location focusing on more difficult problem. Sliding window approach is still the main approach now, but its computational cost strongly increases with the image size. In order to perform location as soon as possible, this paper introduce a method for pedestrian detection that relied on branch and bound search proposed by Lampert et al. Compare to sliding window, it can find a globally optimal classifier functions over all candidate subwindows in linear time. For feature vectors, we used HIK(histogram intersection kernel) to calculate the the similarity, and voting according to their values. Compare to others, the approach used few trainging samples, experimental showing the result.