A comparative analysis of YOLOv5, YOLOv8, and YOLOv10 for balancing accuracy and speed for real-time vehicle detection
Nature Journal of Emerging Sciences Technologies and Innovations · 28.11.2025 · 3 (2) · 117-130
DOI: 10.65752/wb2sep51 ↗Kurz gefasst
Fachartikel in der Zeitschrift Nature Journal of Emerging Sciences Technologies and Innovations, online erschienen am 28.11.2025. Verfasst von Oluwaseyi Ezekiel Olorunshola und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
The need for real-time and reliable vehicle detection in dynamic traffic environments is increasingly critical for applications such as autonomous driving, smart traffic monitoring, and surveillance systems. While many existing object detection models excel either in accuracy or in speed, few maintain a consistent balance between the two. This study proposes a lightweight, GPU-efficient system for vehicle detection and conducts a comparative analysis of three popular YOLO-based deep learning models, YOLOv5s, YOLOv8s, and YOLOv10s, using the UA-DETRAC dataset to determine which best balances detection accuracy and inference speed. The dataset was preprocessed to include only relevant vehicle classes (car, van, and bus), and each model was trained using consistent hyperparameter settings. Evaluation metrics included mean average precision, mAP@0.5, mAP@0.5:0.95, precision, recall, and frames per second (FPS). YOLOv5s demonstrated superior inference speed (~92 FPS), making it suitable for latency-critical applications, though with slightly reduced precision. YOLOv8s offered a balanced trade-off with improved localization accuracy and moderate speed (~83 FPS). YOLOv10s achieved the highest mAP scores and robustness across all classes, maintaining acceptable speed (~66 FPS), and was selected as the core model for the proposed system in high-accuracy real-time deployments. Overall, this work presents amicable insights on the tradeoffs offered by the different YOLO variants for vehicle detection system to ensure dependable performance across varying computational constraints.
Zitieren
APA
Olorunshola, O. E., Iheagwara, S. E., Lakan, N. E., & Akanbi, A. O. (2025). A comparative analysis of YOLOv5, YOLOv8, and YOLOv10 for balancing accuracy and speed for real-time vehicle detection. Nature Journal of Emerging Sciences Technologies and Innovations, 3(2), 117-130. https://doi.org/10.65752/wb2sep51
Vancouver
Olorunshola OE, Iheagwara SE, Lakan NE, Akanbi AO. A comparative analysis of YOLOv5, YOLOv8, and YOLOv10 for balancing accuracy and speed for real-time vehicle detection. Nature Journal of Emerging Sciences Technologies and Innovations. 2025;3(2):117-130. doi:10.65752/wb2sep51
BibTeX
@article{Olorunshola2025,
author = {Olorunshola, Oluwaseyi Ezekiel and Iheagwara, Stephen Ebuka and Lakan, Nanji Emmanuella and Akanbi, Aisha Olamide},
title = {A comparative analysis of YOLOv5, YOLOv8, and YOLOv10 for balancing accuracy and speed for real-time vehicle detection},
journal = {Nature Journal of Emerging Sciences Technologies and Innovations},
year = {2025},
volume = {3},
number = {2},
pages = {117--130},
doi = {10.65752/wb2sep51}
}
EUVIMED