Driver distraction is one of the leading causes of road accidents, and developing reliable real-time monitoring systems is a key challenge for modern driver-assistance technologies. Real-world applications require a careful balance between accuracy and low-latency inference on resource-constrained devices, making Edge AI approaches increasingly relevant. This thesis addresses this challenge by investigating a computer-vision pipeline for detecting distracted-driving behaviours, specifically focusing on identifying the presence of a phone or a bottle in the driver’s hands. The work evaluates recent YOLO models and their suitability for deployment on mobile and embedded platforms. A comparative analysis was conducted on several YOLO versions fine-tuned on a manually annotated custom dataset using optimized training strategies and hyperparameter tuning.