SafeDrive-V2V: A Real-Time Driver Distraction Detection Framework Using CIRNet and V2V Safety Communication

Driver distraction constitutes a predominant factor which is the main reason for road accidents globally, representing a considerable increase in the percentage of traffic-related injuries and death rates. The timely identification of distracted drivers is imperative for the prevention of collisions and the assurance of road safety, particularly in increasing complexity of contemporary driving environments. This research work introduces an innovative framework named SafeDrive-V2V, which aims to detect distracted drivers in real-time and send vehicle-to-vehicle (V2V) alerts via Dedicated Short-Range Communication (DSRC). A hybrid deep learning architecture, referred to as CIRNet (Capsule-Infused ResNet), has been developed by augmenting the ResNet50 model with capsule layers, which serve to maintain spatial hierarchies and enhance the recognition of distracted behavioral indicators. The model undergoes training by utilizing the State Farm distracted driver dataset and is validated for its high accuracy in classifying diverse distracted states. For real-time application, a Raspberry Pi 4 equipped with a Pi camera is installed within the vehicle to perpetually capture images of the driver. These inputs are subsequently processed by the pre-trained CIRNet model for the detection of distraction in real time. Upon the identification of a distracted state, the system generates a Basic Safety Message (BSM), which is transmitted to proximate vehicles utilizing DSRC. This anticipatory communication is designed to caution other drivers regarding potential hazards, thereby mitigating the likelihood of collision. The proposed system presents a holistic solution that integrates deep learning-based monitoring with V2V communication to augment road safety. The model's effectiveness in real-time scenarios is confirmed by experimental results, confirming its feasibility for use in intelligent transportation systems.