Generation of Synthetic Datasets of Highway Driving Scenarios
IEEE Transactions on Intelligent Transportation Systems · 2026-11 · 27 (11) · 15005-15021
DOI: 10.1109/tits.2026.3730710 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Intelligent Transportation Systems, 2026-11. Verfasst von Marianna Cossu und 13 weiteren Autor:innen. Die Arbeit nennt 69 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: European Union’s Horizon 2020 Research and Innovation Program.
Abstract (Originalsprache)
Driving scenario recognition is crucial for vehicle context adaptivity, decision-making, and automated driving function validation. However, research in this area is limited by a critical shortage of publicly available, scenario-labeled video datasets. To bridge this gap, we propose a system for generating scenario-labeled synthetic video clips from statistical distributions of traffic datapoints. Leveraging the large-scale, real-world, proprietary MOOVE collection, we generated and publicly released a 9-class dataset featuring varied highway conditions. Based on our experience, key insights for efficient synthetic dataset creation include the necessity of a rigorous scenario-modeling methodology, for which OpenSCENARIO 2.0 was effective, and the need to enrich a simulator like CARLA with behaviors for enhanced realism and scalability to produce thousands of samples for each class. Exploiting the synthesized dataset, we developed a real-time classification framework based on a 3D-CNN architecture and attention-enhanced variants. To our knowledge, the study provides one of the first dataset-oriented evaluations of real-time video-based recognition across a broad set of highway driving scenarios. Preliminary cross-domain experiments further show that models trained on the synthetic videos retain meaningful performance on real-world footage without fine-tuning. The proposed method combines real-world parameter distributions with logical scenario models and simulator-specific behavioral mechanisms to produce a shareable synthetic dataset suitable for model pre-training and subsequent domain adaptation.
Förderung
- European Union’s Horizon 2020 Research and Innovation Program
Zitieren
APA
Cossu, M., Berta, R., Lazzaroni, L., Forneris, L., Fresta, M., Pighetti, A., Pilla, A., Sauvaget, J., Touil, H., Durville, L., Griffon, T., Nejma, G. B., Selem, F. H., & Bellotti, F. (2026). Generation of Synthetic Datasets of Highway Driving Scenarios. IEEE Transactions on Intelligent Transportation Systems, 27(11), 15005-15021. https://doi.org/10.1109/tits.2026.3730710
Vancouver
Cossu M, Berta R, Lazzaroni L, Forneris L, Fresta M, Pighetti A, et al. Generation of Synthetic Datasets of Highway Driving Scenarios. IEEE Transactions on Intelligent Transportation Systems. 2026;27(11):15005-15021. doi:10.1109/tits.2026.3730710
BibTeX
@article{Cossu2026,
author = {Cossu, Marianna and Berta, Riccardo and Lazzaroni, Luca and Forneris, Luca and Fresta, Matteo and Pighetti, Alessandro and Pilla, Alessandro and Sauvaget, Jean-Louis and Touil, Hind and Durville, Laurent and Griffon, Thibault and Nejma, Ghada Ben and Selem, Fouad Hadj and Bellotti, Francesco},
title = {Generation of Synthetic Datasets of Highway Driving Scenarios},
journal = {IEEE Transactions on Intelligent Transportation Systems},
year = {2026},
volume = {27},
number = {11},
pages = {15005--15021},
doi = {10.1109/tits.2026.3730710}
}
EUVIMED