Advanced computer vision for extracting georeferenced vehicle trajectories from drone imagery

Fónod, Róbert; Cho, Haechan; Yeo, Hwasoo; Geroliminis, Nikolas · 2025 · OpenAlex-citations

DOI: 10.1016/j.trc.2025.105205

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper presents a comprehensive framework for extracting georeferenced vehicle trajectories from high-altitude drone imagery, addressing the limitations of traditional ground-based traffic monitoring systems such as loop detectors and stationary cameras. The authors aim to provide a scalable, cost-effective solution for urban traffic analysis by integrating advanced computer vision techniques with unmanned aerial vehicle (UAV) data. The research is motivated by the need for flexible, high-resolution traffic data that can capture complex urban dynamics without the constraints of fixed infrastructure or the coverage gaps of satellite navigation systems. The methodology was validated through a large-scale multi-drone experiment conducted in the Songdo International Business District, South Korea, from October 4 to 7, 2022. A fleet of 10 commercial drones captured 4K ultra-high-definition video at altitudes of 140–150 meters, monitoring 20 intersections and generating approximately 12TB of data. The extraction pipeline includes several novel components: a tailored YOLOv8-based object detector optimized for bird’s-eye view perspectives; a unique track stabilization method that applies homographic transformations to extracted tracks using vehicle bounding boxes as exclusion masks to reduce computational overhead; and a master frame-based georeferencing strategy using a high-resolution orthophoto and 13 real-time kinematic (RTK)-calibrated ground control points. Additionally, the framework incorporates robust vehicle dimension estimation and detailed road segmentation. The study produced two major datasets: the Songdo Traffic dataset, containing approximately 700,000 unique vehicle trajectories with metadata including speed, acceleration, and vehicle type, and the Songdo Vision dataset, comprising over 5,000 human-annotated images with nearly 300,000 vehicle instances across four classes. Validation against high-precision sensor data from an instrumented autonomous vehicle demonstrated high consistency in speed, acceleration, and dimension estimates, confirming the accuracy of the extraction pipeline in dense urban environments. The results showed close agreement with known vehicle dimensions and realistic traffic dynamics, even during complex events like congestion and post-accident disruptions. The significance of this work lies in its contribution to open science and intelligent transportation systems. By publicly releasing the Songdo Traffic and Songdo Vision datasets alongside the complete source code, the authors establish new benchmarks for data quality, reproducibility, and scalability in traffic research. The framework demonstrates that integrating drone technology with advanced computer vision enables precise, adaptable, and cost-effective urban traffic monitoring. This approach supports the development of responsive traffic management strategies and provides valuable resources for studying complex traffic phenomena, such as network modeling and emission estimation, in smart cities.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• distraction detection algorithms