Nighttime Pedestrian Detection Using Local Oriented Shape Context Descriptor

Abstract

In this paper, we propose a novel feature based on local oriented shape context (LOSC) descriptor for nighttime pedestrian detection. Shape context descriptor is widely used in object recognition but not making a good performance in complex situations because it does not consider the edge's orientation. To address this limitation, our method adds orientation information to the shape context descriptor. We first compute the image gradient in nine directions and then extract shape context descriptor in each direction. Finally we put the feature vector to linear SVM for training. We tested this descriptor's performance on our nighttime pedestrian samples captured by normal night-vision camera. The experiment results show that our method achieved high detection rate and had fewer dimensions than the HOG descriptor.