Uloženo v:
| Hlavní autoři: | , |
|---|---|
| Médium: | Recurso digital |
| Jazyk: | angličtina |
| Vydáno: |
Zenodo
2026
|
| Témata: | |
| On-line přístup: | https://doi.org/10.5281/zenodo.19710577 |
| Tagy: |
Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
|
Obsah:
- <p>This paper presents a comparative study of two deep learning-based object detection approaches: a Vision Transformer (ViT)-based detector and a fine-tuned YOLOv8 nano model. The ViT model is trained on the Pascal VOC 2012 dataset, while YOLOv8 is trained on a domain-specific vehicle dataset obtained from Roboflow.</p> <p>The proposed system performs real-time vehicle tracking, classification, and traffic violation detection, including stop-line violations, restricted zone intrusion, and wrong-way driving. A complete end-to-end pipeline is developed covering data preprocessing, model training, inference, and visualization.</p> <p>Experimental results demonstrate that YOLOv8 achieves efficient real-time inference suitable for deployment, while the Vision Transformer provides richer global feature representations at the cost of higher computational complexity. The study highlights key trade-offs between transformer-based and convolution-based detection architectures for real-world intelligent traffic monitoring systems.</p>