Deep Learning Approaches for Vehicle and Pedestrian Detection in Adverse Weather

Vision-based vehicle and pedestrian identification methods to enhance road safety have become more critical during the last decade. Unfortunately, these identification approaches suffer from robustness because of enormous changes in vehicle form, crowded surroundings, different light circumstances, weather, and driving behavior. Therefore, developing automated vehicle and pedestrian detection and tracking systems is a demanding research field in Intelligent Transport Systems (ITS). In recent years, image-based, deep-learning object identification algorithms have become effective autonomous road item recognition agents. The profound learning processes for identifying road vehicles have produced remarkable achievements. However, while numerous types of research have extensively examined using different kinds of deep learning techniques, further studies still need to integrate poor weather conditions with the standard vehicle detection algorithms for profound learning objects. This paper investigates the qualitative and quantitative analyses of four recent deep-learning object-detection algorithms for vehicle and pedestrian identification, i.e., the faster F-RCN, SSD, HoG, and YOLOv7 and classifying weather conditions utilizing a real-time dataset (DAWN). The efficacy of the suggested technique, which superposes the state-of-the-art vehicle identification and tracking methodology under unfavorable and adverse circumstances, is verified by experimental findings.