Saved in:
| Main Author: | |
|---|---|
| Format: | Recurso digital |
| Language: | |
| Published: |
Zenodo
2025
|
| Online Access: | https://doi.org/10.5281/zenodo.15604536 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Table of Contents:
- <p>Smart city technology development demands new approaches for improving road safety by detecting the various causes of driver distraction which leads to accidents. The research presents a deep learning combination that unites Convolutional Neural Networks (CNN), Vision Transformer (ViT), and SigLIP (Sigmoid Labeled Image Processor) for handling this issue. The three components including CNN (ResNet18) for spatial feature extraction and ViT for contextual relationship processing and SigLIP for feature representation produce an accurate behavior classification framework. The State Farm Distracted Driver Detection dataset\cite{b11} was used and the hybrid model accomplishes 97.35\% validation accuracy through its 25th training epoch which surpasses conventional single-method approaches. ViT reaches high success rates through its ability to perform multi-modal feature fusion and backbone-freezing techniques alongside dropout regularization for overfitting prevention. The paper proposes a prototype for smart city integration architecture which also includes algorithmic innovations for practical implementation. The system distributes processing tasks across edge devices which perform camera feeds analysis while the cloud system collects excessive information for complex analysis and alerts traffic officials about identified distractions through automation. The proposed architecture enables system resilience and compatibility with current smart city infrastructure. The project integrates current computer vision innovations with IoT deployment realities to show that artificial intelligence solutions boost traffic safety and accident reduction and enable real-time traffic control.</p>