Akademik Veri Yönetim Sistemi
2025 Innovations in Intelligent Systems and Applications Conference (ASYU), Bursa, Türkiye, 10 - 12 Eylül 2025, ss.1-4, (Tam Metin Bildiri)
Accurate driver pose estimation is essential for building intelligent, safety-focused monitoring systems. In this study, we propose a multimodal approach that combines RGB and depth imaging, advanced keypoint detection via Vision Transformer Pose Estimation (ViTPose), and steering wheel interaction classification using a Random Forest classifier. We categorize driver hand positions relative to the steering wheel into three classes: both hands, one hand, or no hands at all. By leveraging detailed wrist position analysis and steering wheel segmentation via You Look Only Once (YOLO), our approach achieves robust performance in constrained vehicular environments. Comprehensive testing with different states proves the capability of our framework in handling occlusion, lighting, and body positions.