Human Detection in UAV Thermal Imagery: Dataset Extension and Comparative Evaluation on Embedded Platforms
Journal of Imaging · 09.12.2025 · 11 (12) · 436
DOI: 10.3390/jimaging11120436 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Imaging, online erschienen am 09.12.2025. Verfasst von Andrei-Alexandru Ulmămei und 3 weiteren Autor:innen. Beteiligte Einrichtung: Department of Electronic Devices, Circuits and Architectures, National University of Science and Technology Politehnica Bucharest, 060042 Bucharest, …. Die Arbeit wurde bisher 10-mal zitiert (Crossref) und nennt 75 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Program for Research of the National 305 Association of Technical Universities, National Program for Research of the National Association of Technical Universities.
Abstract (Originalsprache)
Unmanned aerial vehicles (UAVs) equipped with thermal cameras are increasingly used in search and rescue (SAR) operations, where low visibility and small human footprints make detection a critical challenge. Existing datasets are mostly limited to urban or open-field scenarios, and our experiments show that models trained on such heterogeneous data achieve poor results. To address this gap, we collected and annotated thermal images in mountainous environments using a DJI M3T drone under clear daytime conditions. This mountain-specific set was integrated with ten existing sources to form an extensive benchmark of over 75,000 images. We then performed a comparative evaluation of object detection models (YOLOv8/9/10, RT-DETR) and semantic segmentation networks (U-Net variants), analyzing accuracy, inference speed, and energy consumption on an NVIDIA Jetson AGX Orin. Results demonstrate that human detection tasks can be accurately solved through both semantic segmentation and object detection, achieving 90% detection accuracy using segmentation models and 85% accuracy using the YOLOv8 X detection model in mountain scenarios. On the Jetson platform, segmentation achieves real-time performance with up to 27 FPS in FP16 mode. Our contributions are as follows: (i) the introduction of a new mountainous thermal image collection extending current benchmarks and (ii) a comprehensive evaluation of detection methods on embedded hardware for SAR applications.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Electronic Devices, Circuits and Architectures, National University of Science and Technology Politehnica Bucharest, 060042 Bucharest, Romania.
- Computer Science Department, National Institute of Applied Sciences of Lyon, 69621 Lyon, France.
ORCID-Profile: Andrei-Alexandru Ulmămei, Taddeo D'Adamo, Costin-Emanuel Vasile, Radu Hobincu
Förderung
- National Program for Research of the National 305 Association of Technical Universities
- National Program for Research of the National Association of Technical Universities
Zitieren
APA
Ulmămei, A. A., D'Adamo, T., Vasile, C. E., & Hobincu, R. (2025). Human Detection in UAV Thermal Imagery: Dataset Extension and Comparative Evaluation on Embedded Platforms. Journal of Imaging, 11(12), 436. https://doi.org/10.3390/jimaging11120436
Vancouver
Ulmămei AA, D'Adamo T, Vasile CE, Hobincu R. Human Detection in UAV Thermal Imagery: Dataset Extension and Comparative Evaluation on Embedded Platforms. Journal of Imaging. 2025;11(12):436. doi:10.3390/jimaging11120436
BibTeX
@article{Ulmmei2025,
author = {Ulmămei, Andrei-Alexandru and D'Adamo, Taddeo and Vasile, Costin-Emanuel and Hobincu, Radu},
title = {Human Detection in UAV Thermal Imagery: Dataset Extension and Comparative Evaluation on Embedded Platforms},
journal = {Journal of Imaging},
year = {2025},
volume = {11},
number = {12},
pages = {436},
doi = {10.3390/jimaging11120436}
}
EUVIMED