Abstract

This paper presents a framework for extracting georeferenced vehicletrajectories from high-altitude drone imagery, addressing key challenges inurban traffic monitoring and the limitations of traditional ground-basedsystems. Our approach integrates several novel contributions, including atailored object detector optimized for high-altitude bird's-eye viewperspectives, a unique track stabilization method that uses detected vehiclebounding boxes as exclusion masks during image registration, and an orthophotoand master frame-based georeferencing strategy that enhances consistentalignment across multiple drone viewpoints. Additionally, our frameworkfeatures robust vehicle dimension estimation and detailed road segmentation,enabling comprehensive traffic analysis. Conducted in the Songdo InternationalBusiness District, South Korea, the study utilized a multi-drone experimentcovering 20 intersections, capturing approximately 12TB of 4K video data overfour days. The framework produced two high-quality datasets: the Songdo Trafficdataset, comprising approximately 700,000 unique vehicle trajectories, and theSongdo Vision dataset, containing over 5,000 human-annotated images with about300,000 vehicle instances in four classes. Comparisons with high-precisionsensor data from an instrumented probe vehicle highlight the accuracy andconsistency of our extraction pipeline in dense urban environments. The publicrelease of Songdo Traffic and Songdo Vision, and the complete source code forthe extraction pipeline, establishes new benchmarks in data quality,reproducibility, and scalability in traffic research. Results demonstrate thepotential of integrating drone technology with advanced computer vision forprecise and cost-effective urban traffic monitoring, providing valuableresources for developing intelligent transportation systems and enhancingtraffic management strategies.