Traffic intersection safety is a growing concern in urbanizing areas with dense traffic and complex road-user interactions. Real-time conflict estimation is essential for reducing risks, yet existing cloud-based methods face challenges like high transmission costs, impractical data handling, and privacy issues due to centralized storage. To address these challenges, we propose an Edge-AI-based conflict estimation system for busy intersections. By processing data locally on edge devices, our approach reduces bandwidth usage, lowers reliance on expensive cloud infrastructure, minimizes centralized server loads, and enhances privacy by avoiding raw data storage. The system utilizes a roadside camera with an edge device running a lightweight, modified YOLOv8 model for real-time detection and the ByteTrack algorithm for object tracking. Camera calibration converts image coordinates into real-world coordinates, enabling the estimation of vehicle speed, heading, and future trajectory. Using this data, the system predicts potential conflicts by calculating speed, heading, distance, and time-to-collision (TTC) in real time, ensuring timely interventions with low latency.
VDO Real-time Edge-AI dashboard for intersection safety, showcasing vehicle detection, tracking, and conflict estimation using YOLOv8 and ByteTrack.