With nighttime economic and traffic activity growth, nighttime pedestrian detection technology has become particularly critical in intelligent transportation systems and autonomous driving. However, color distortion and edge information loss exist in the image at night. The pedestrian detection model has problems with computing resource limitations and too many parameters on the deployed embedded platform. Focusing on the challenges posed by low-light conditions prevalent in urban road environments during nighttime, this research endeavors to enhance the visibility of pedestrian detection by incorporating the URetinex-Net algorithm. Meanwhile, this study designs the network structure of YOLOv8 lightweight, selects MobileNetV3 as the backbone of YOLOv8, and introduces the hybrid attention mechanism to enhance the adaptability to the complex environment at night. Compared with the original YOLOv8, the parameters are reduced by 22.8%, and the GFLOPs are reduced by 33.3%.