EUVIMEDEUVIMED entdecken
Open Access · CC BY-SA 4.0

Real-Time Semantic Segmentation for Energy-Efficient Autonomous Vehicle Navigation

Mustafa M. Zayer, Zaid Ghanim Ali, Yasser Abdulrahim Ahmed, Salima Baji Abdullah, Matai Nagi Saeed

JOIV : International Journal on Informatics Visualization · 30.03.2026 · 10 (2) · 768

DOI: 10.62527/joiv.10.2.5330 ↗

Kurz gefasst

Fachartikel in der Zeitschrift JOIV : International Journal on Informatics Visualization, online erschienen am 30.03.2026. Verfasst von Mustafa M. Zayer und 4 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-SA 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Semantic segmentation with the ability to interpret complex driving conditions is highly effective and real-time, which is essential in autonomous driving. Nevertheless, the high segmentation accuracy at energy efficiency on embedded systems is still difficult to attain. The article presents a low-complexity convolutional neural network (CNN) model to real-time semantic segmentation, as implemented with energy-constrained inference on the edge platform, the NVIDIA Jetson Xavier NX and the Intel Neural Compute Stick 2. The design of the model adds the depth-wise separable convolutions and residual bottlenecks and uses Atrous Spatial Pyramid Pooling (ASPP) to provide the multiscale spatial context without compromising the computational power. Channel Recalibration: Squeeze-and-Excitation Inspired Mechanisms Channel Calibration Squeeze-and-Excitation mechanisms provide a good method to enhance feature selectivity at zero additional cost. Other post-training optimization techniques are structured pruning, 8-bit quantization and knowledge distillation, which again shrink the model size and power usage. The accuracy of segmentation, energy consumption, and inference robustness were tested on the Cityscapes dataset in different conditions as nighttime, rain. It also performed better in terms of mean Intersection over Union (mIoU), energy per inference, and latency than state-of-the-art lightweight networks, ENet and Fast-SCNN. It was suggested that a new performance index should assess the trade-off between accuracy, speed and power efficiency in a combined fashion. These findings show that with proper architecture and post-dynamic processes, robust semantic segmentation can be implemented which can be implemented on real-time autonomous systems. The model suggested provides a viable route to scalable and viable intelligent transport systems.

Zitieren

APA

Zayer, M. M., Ali, Z. G., Ahmed, Y. A., Abdullah, S. B., & Saeed, M. N. (2026). Real-Time Semantic Segmentation for Energy-Efficient Autonomous Vehicle Navigation. JOIV : International Journal on Informatics Visualization, 10(2), 768. https://doi.org/10.62527/joiv.10.2.5330

Vancouver

Zayer MM, Ali ZG, Ahmed YA, Abdullah SB, Saeed MN. Real-Time Semantic Segmentation for Energy-Efficient Autonomous Vehicle Navigation. JOIV : International Journal on Informatics Visualization. 2026;10(2):768. doi:10.62527/joiv.10.2.5330

BibTeX
@article{Zayer2026,
  author = {Zayer, Mustafa M. and Ali, Zaid Ghanim and Ahmed, Yasser Abdulrahim and Abdullah, Salima Baji and Saeed, Matai Nagi},
  title = {Real-Time Semantic Segmentation for Energy-Efficient Autonomous Vehicle Navigation},
  journal = {JOIV : International Journal on Informatics Visualization},
  year = {2026},
  volume = {10},
  number = {2},
  pages = {768},
  doi = {10.62527/joiv.10.2.5330}
}
Kostenlos starten →