An Improved UNet Architecture With Convolutional Block Attention Module (CBAM) for Road Segmentation in Autonomous Driving
IET Image Processing · 2026-01 · 20 (1)
DOI: 10.1049/ipr2.70410 ↗Kurz gefasst
Fachartikel in der Zeitschrift IET Image Processing, online erschienen am 03.07.2026. Verfasst von Pooya Bazrafshan und 2 weiteren Autor:innen. Die Arbeit nennt 40 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).
Abstract (Originalsprache)
Accurate road segmentation is an essential component of autonomous driving systems, as it supports reliable navigation and decision‐making in diverse environments. This study presents a UNet‐based architecture enhanced with the Convolutional Block Attention Module (CBAM) to improve feature selection in complex and noisy road scenes. The proposed UNet‐CBAM model was trained and evaluated on the KITTI dataset, achieving a validation IoU of 0.9238 and an accuracy of 0.9864. Additional experiments were conducted on the CamVid dataset to further evaluate the segmentation performance of the model, where it achieved an mIoU of 0.9405 and a precision of 0.9572 on the test set. Quantitative analyses indicated stable convergence behaviour throughout training, and qualitative inspections showed that the model was able to delineate road regions with reasonable accuracy across various real‐world scenarios. Experiments under different illumination conditions as well as varying noise and blur levels provided further insights into the model's behaviour in visually challenging environments. Nevertheless, sharp road curves and extreme illumination variations remained difficult cases for the model, suggesting directions for future improvement. Overall, the findings suggest that integrating attention mechanisms such as CBAM into UNet architectures may help improve feature representation for road segmentation tasks.
Zitieren
APA
Bazrafshan, P., Alitappeh, R. J., & Danesh, M. (2026). An Improved UNet Architecture With Convolutional Block Attention Module (CBAM) for Road Segmentation in Autonomous Driving. IET Image Processing, 20(1). https://doi.org/10.1049/ipr2.70410
Vancouver
Bazrafshan P, Alitappeh RJ, Danesh M. An Improved UNet Architecture With Convolutional Block Attention Module (CBAM) for Road Segmentation in Autonomous Driving. IET Image Processing. 2026;20(1). doi:10.1049/ipr2.70410
BibTeX
@article{Bazrafshan2026,
author = {Bazrafshan, Pooya and Alitappeh, Reza Javanmard and Danesh, Malihe},
title = {An Improved UNet Architecture With Convolutional Block Attention Module (CBAM) for Road Segmentation in Autonomous Driving},
journal = {IET Image Processing},
year = {2026},
volume = {20},
number = {1},
doi = {10.1049/ipr2.70410}
}
EUVIMED