Night Time Object Detection Using Deep Learning for Enhanced Vehicle Safety

Abstract— Night time road accidents contribute a high percentage of all traffic fatalities because of the low visibility, glare, and non-uniform lighting conditions. The failure to spot pedestrians and vehicles under dim light conditions rises the risk of collisions. Conventional methods of object detection are unable to work with ease in such conditions. In this research, the use of You Only Look Once (YOLO) and Region-Based Convolutional Neural Networks (RCNN) is investigated to improve detection accuracy in nighttime scenarios. Preprocessing operations such as noise removal and brightness are incorporated to enhance the quality of images taken prior to detection, Performance analysis on benchmark datasets illustrates a high level of vehicles and pedestrian detection with a priority given to real-time processing. The results identify the strengths and weaknesses of these deep learning methods, giving insight into future development in nighttime object detection. Keywords—YOLO, RCNN, Night-Vision, Real-time, Night Time Object Detection Time, Noise Reduction, Low light Enhancement, pedestrian and vehicles detection.