Local Haar-like features in edge maps for pedestrian detection

Pedestrian detection is a significant problem in computer vision. Dollar et al. at [1] indicate that current methods are not efficient enough to solve the problem and it remains challenging. Besides, most of the proposed features are also not efficient in describing the features of pedestrians. In this paper, we concentrate on this issue and introduce the local Haar-like features in edge maps to describe the edge/contour of pedestrians efficiently. To reduce the background noise, localized normalization is proposed. Also the integral and tilted integral images are applied to fast extraction of the feature values. The experiments on pedestrian detection show both the describing and computational efficiency of the proposed features.