Adaptive hysteresis thresholding based pedestrian detection in nighttime using a normal camera

This paper presents a novel approach for pedestrian detection in nighttime with a normal camera. Generally, there is a real-time requirement in the pedestrian detection system which limits the computational complexity of algorithms. Thus most systems utilize the thresholding method for pedestrian detection or segmentation. However, the single adaptive threshold based approach doesn't always work well and sometimes gives poor results. In this paper, we propose adaptive hysteresis based segmentation algorithm which holds two adaptive thresholds. This inspiration comes from the Canny operator which uses hysteresis for edge thresholding. Furthermore, we expand the hysteresis to be proper for region segmentation in pedestrian detection. Experiments prove that the proposed method performs well most of the time and improves the ability of the pedestrian detection system.