A Methodological Framework for Automated Extraction and Analysis of Traffic Parameters From High-Resolution Unmanned Aerial Vehicle Video Data

Recent advances in unmanned aerial vehicles (UAVs) platforms and computer vision techniques have created new opportunities for collecting high-resolution vehicle trajectories; however, a unified and operational framework for converting raw UAV footage into analytically usable traffic parameters is largely lacking. This study develops a comprehensive, end-to-end methodological framework that transforms UAV video into structured multiscale traffic information, including vehicle trajectories, lane configurations, stop-line locations, signal timing plans, queue profiles, and spillback characteristics. The framework integrates YOLO-based detection, DeepSORT tracking, DBSCAN-based signal phase inference, and a spatiotemporal queue-identification algorithm that jointly captures queue formation, dissipation, and spillback propagation. Its robustness and generality are validated through extensive experiments across three representative settings—road-segment, single-intersection, and multi-intersection environments—using both self-collected UAV footage and the large-scale pNEUMA dataset. Results demonstrate accurate trajectory reconstruction, reliable lane and stop-line estimation, precise signal timing recovery, and finegrained queue length identification even under complex congestion patterns. To support reproducibility and future research, the complete implementation of the proposed framework is publicly released.