Seatbelt compliance is essential for road safety, yet traditional detection systems often fail under poor lighting, occlusion, motion blur, and high-speed driving conditions. To overcome these challenges, we propose a Transformer-Augmented YOLOv8s framework for real-time seatbelt detection, optimized for edge-based automotive systems. By integrating a lightweight transformer block into YOLOv8s, the model combines efficient local feature extraction with global context reasoning, enabling robust detection in harsh environments. Experiments on a custom dataset show that the proposed model achieves 90.3% mAP@0.5, operates at 100 FPS, and maintains a compact size of 10.4M parameters. The results confirm that the framework outperforms the baseline detectors in both accuracy and robustness, making it a practical solution for intelligent vehicle safety monitoring.