Cloud-Enabled Pedestrian Safety and Risk Prediction in VANETs Using Hybrid CNN-LSTM Models

Abstract

As pedestrian safety in Vehicular Ad-hoc Networks (VANETs) becomes more critical, there is an increasing need for effective, real-time risk assessment systems to predict pedestrian-vehicle interactions. Typical problems that most pedestrian detection and risk assessment systems experience include high latency, low accuracy, and limitations in scale when it comes to context shifting—all of which can be effectively tackled by cloud computing as a solution to their scalability and efficiency effect on data processing for VANET real-time decision-making. This framework proposes the incorporation of pedestrian safety cloud systems with the potential for a robust dynamic risk assessment using a hybrid CNN-LSTM model, coupled with a YOLO-based real-time pedestrian identification scheme. To warn automobiles and pedestrians of possible dangers, the system makes advantage of V2X communication. The proposed approach is a reliable, flexible and effective solution for pedestrian safety in VANETs, which guarantees fast actions and minimizes accidents in dynamic situations. Tests showed that the technology demonstrated scalability and good real-time performance across different traffic and weather scenarios. Identification and assessment of pedestrian hazards had an average latency of less than 200 ms, 97% accuracy, recall, and precision above 90%.