A Comprehensive Framework for Accident Detection and Risk Assessment Using CNN and Graph-Based Analytics

This proposed work introduces a comprehensive accident detection and risk assessment system that coordinate the strengths of Convolutional Neural Networks (CNNs) and a graph-based analytical approach. With support of YOLOv8, deep learning model for object detection, the system perform real-time identification vehicles, pedestrians, and other road entities from video feeds, such as CCTV footage and user-generated videos. This detection capability is integrated with a graph-based model that represents these entities as nodes and their interactions as edges. The proximity, trajectory, and relative speeds of the entities are analyzed to assess collision risk dynamically. The combined approach enhances the system’s ability to identify potential accidents and evaluate risk levels with high precision. The CNN component ensures accurate object detection and classification, while the graph-based analysis facilitates real-time spatial reasoning and interaction assessment. A user-friendly web interface enables real-time visualization of detections and risk analyses, supporting collaboration with traffic authorities and urban planners. This approach not only aids in accident detection but also provides actionable insights for proactive risk mitigation and traffic management.