FURI | Spring 2025

Enhancing Low-light Traffic Monitoring at Intersections Using Event-based Vision Systems

Traditional traffic monitoring systems using frame-based cameras that struggle in low-light and high-contrast environments (e.g., nighttime or glare from headlights). These limitations result in motion blur, missed objects, and delayed detection, which is particularly dangerous at intersections where vehicle and pedestrian interaction is complex. This project explores the use of event-based cameras to build a dataset across different lighting conditions and develop real-time detection and tracking models such as Recurrent Vision Transformers (RVTs) to better handle the asynchronous, high-temporal data captured by these cameras. By improving detection accuracy and reducing latency, the system has potential applications in smart city infrastructure, supporting safer and more efficient urban mobility.