Pedestrian Detection Based on HOG-LBP Feature

In this paper, we propose a new framework in pedestrian detection by combining the HOG and uniform LBP feature on blocks. Contrast experiment result shows that detector using combined features is more powerful than one single feature. To further improve the detection performance, we make a contrast experiment that the HOG-LBP features are calculated at variable-size blocks to find the most efficient feature vector. The linear SVM is used to train the pedestrian classifier. Results presented on the INRIA dataset show that our detector is more discriminative and robust than the state-of-the-art algorithms.