A Comparative Study and State-of-the-art Evaluation for Pedestrian Detection

Abstract

Pedestrian detection has many applications in computer vision including robotics, scene understanding, person reidentification and video-surveillance system. In fact, the process of person detection aims to detect and localize each person in the images, represented via bounding boxes. Recent deep learning pedestrian detectors, which are hybrid methods that combines traditional hand-crafted features and deep convolutional features such as Fast/Faster Region based-CNN (R-CNN), have shown excellent performance for general object detection. In this context, we propose in this paper an overview of the state-of-the-art performance of current deep learning pedestrian detectors and a comparison of these detectors is provided. Evaluation criteria, popular datasets used for evaluation and a quantitative results are also described and discussed in this work.