An Accurate and Light Poolformer‐Based Model for Autonomous Driving Road Object Detection
IET Intelligent Transport Systems · 2026-01 · 20 (1)
DOI: 10.1049/itr2.70202 ↗Kurz gefasst
Fachartikel in der Zeitschrift IET Intelligent Transport Systems, online erschienen am 19.05.2026. Verfasst von Yue Zhou und 3 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 44 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Key Research and Development Program of China.
Abstract (Originalsprache)
Computer vision algorithms are increasingly crucial for perception in automated driving systems, particularly in the context of object detection and recognition. Nonetheless, the limited inference speed and accuracy in object detection models pose significant obstacles to the progress of autonomous driving technology. Furthermore, these algorithms encounter various challenges, including multi‐scale variations, dense scenes, and the detection of tiny‐sized objects. In this article, we present an efficient deep learning model, PF‐YOLO, based on YOLOv5 and a lightweighted Poolformer. Firstly, new feature fusion layers and a prediction head were added for small object detection because they can maximally preserve the feature information. Then, a Poolform block was inserted into the backbone to extract representative global features. Finally, we integrate a triplet attention model to select the critical information and suppress irrelevant information. The performance of our proposed model is evaluated and tested on the BDD100K dataset and KITTI dataset for pedestrians, vehicles, and traffic lights under various conditions. The results demonstrate that the improved model can achieve higher mean average precision and better detection ability, especially for small targets. Conducting an ablation study on the autonomous driving dataset BDD100K, our algorithm achieves a mean average precision of 57.96%, which is 3.69% better than the YOLOv5l algorithm. Meanwhile, these results demonstrate the effectiveness and superiority of our algorithm for automatic driving object detection.
Förderung
- National Key Research and Development Program of China
Zitieren
APA
Zhou, Y., Ma, J., Dong, H., & Li, W. (2026). An Accurate and Light Poolformer‐Based Model for Autonomous Driving Road Object Detection. IET Intelligent Transport Systems, 20(1). https://doi.org/10.1049/itr2.70202
Vancouver
Zhou Y, Ma J, Dong H, Li W. An Accurate and Light Poolformer‐Based Model for Autonomous Driving Road Object Detection. IET Intelligent Transport Systems. 2026;20(1). doi:10.1049/itr2.70202
BibTeX
@article{Zhou2026,
author = {Zhou, Yue and Ma, Jihui and Dong, Honghui and Li, Wei},
title = {An Accurate and Light Poolformer‐Based Model for Autonomous Driving Road Object Detection},
journal = {IET Intelligent Transport Systems},
year = {2026},
volume = {20},
number = {1},
doi = {10.1049/itr2.70202}
}
EUVIMED