Object Detection of Autonomous Vehicles in Adverse Weather Conditions

Abstract

Autonomous vehicle technology has rapidly advanced, with object detection and classification playing a pivotal role in enabling safe and intelligent navigation. However, adverse weather conditions—such as fog, rain, glare, and low light—continue to challenge the reliability of perception systems, potentially compromising vehicle safety and performance. As the push toward full autonomy grows, the development of robust object detection systems that maintain high accuracy in diverse environmental conditions becomes essential. This research focuses on enhancing vehicle detection capabilities under harsh weather scenarios using Convolutional Neural Networks (CNNs). Our approach integrates a customized YOLO-based algorithm, optimized for real-time performance, to identify and classify vehicles despite visual distortions caused by weather. The model is trained on weather-affected datasets, enabling it to recognize objects even when visibility is reduced. Through extensive testing, our system demonstrates improved detection accuracy and stability across challenging environments, significantly reducing the risk of perception failure. By combining deep learning techniques with weather-aware adaptations, this study contributes to the creation of more resilient autonomous systems. The proposed framework ensures that autonomous vehicles can maintain operational reliability and safety, regardless of environmental uncertainty. This work represents a crucial step toward building perception systems that support full autonomy in real-world, unpredictable driving conditions.