Enhanced Road Safety in IoV Using Deep Learning for Real-Time Accident Prediction and Prevention

The increasing rate of traffic accidents due to the growing volume of vehicles on roads necessitates the integration of advanced technologies for road safety. This paper presents an enhanced framework leveraging the Internet of Vehicles (IoV) to predict and prevent accidents in real-time using a deep learning-based approach. The proposed system integrates IoTenabled devices and roadside units (RSUs) to collect and analyze spatio-temporal data, including road traffic conditions, weather patterns, and vehicle behaviors using Graph Neural Networks (GNNs). A improved Spatio-Temporal Conv-Long Short-Term Memory Autoencoder (ISTCLA) framework processes this multidimensional data on a cloud-based control server, enabling accurate prediction of potential accidents and providing timely preventive measures. Unlike conventional methods, the proposed system emphasizes high-speed data transmission, seamless integration across IoV devices, and adaptive learning to address diverse accident scenarios. The system achieved a significant improvement in prediction accuracy, with an AUROC score of 0.94, surpassing existing models. This research offers a scalable and robust solution for enhancing road safety, emphasizing the practicality of deep learning in real-time accident management within IoV ecosystems. Future extensions will focus on incorporating blockchain for secure data management and decentralized decision-making.