Nighttime pedestrian detection by selecting strong near-infrared parts and enhanced spatially local model

We propose a nighttime pedestrian detection method for a moving vehicle equipped with a camera and the near-infrared lighting. The objects in the nighttime environment will reflect the infrared projected. In some cases, however, the clothes absorb most of the infrared and make the pedestrian partially invisible in that part. To deal with this, a part-based pedestrian detection method according to the feature points marked on parts is used. Due to high computation load, selection of effective parts becomes imperative. In this research work, we analyze the relations between the detection rate/processing time and different numbers/types of parts. Besides, traditional training of the part detector normally requires a large number of occlusion samples. To overcome this problem, we learn the spatial relationship between every pair of two parts. The confidence of the detected parts can be enhanced even if some parts are occluded. While trying to refine pedestrians after detection, we use two filters and segmentation method to verify their bounding boxes. The proposed system is verified by experiments and appealing results have been demonstrated.