Road safety remains a pressing national concern, with seat belt non-compliance continuing to be one of the leading contributors to severe injuries and fatalities in traffic accidents. This issue is particularly evident in developing regions and countries, where enforcement mechanisms and monitoring systems are often limited. To address this challenge, this paper introduces a deep learning-based framework for automatic seat belt violation detection, specifically tailored to Moroccan road conditions. A custom dataset was developed, capturing a wide range of vehicles and environmental scenarios representative of local traffic. We employed the YOLOv7 and YOLOv7-Tiny object detection algorithms to accurately classify drivers into two categories: person_seatbelt and person_noseatbelt. The model was trained and rigorously evaluated on this dataset, achieving a mean average precision (mAP) of 83.63%. The proposed approach not only enhances region-specific violation detection but also lays an essential foundation for future automated driver behavior analysis and traffic law enforcement technology systems.