Fatigue and distracted driving are major causes of road traffic accidents, and real-time, accurate monitoring is crucial for ensuring driving safety. This study aims to develop a driving behavior monitoring solution based on the YOLO (You Only Look Once) object detection system, utilizing in-car cameras to analyze driver physical demeanor and posture characteristics in real-time. We focus on detecting key physical indicators highly correlated with abnormal driving behavior, including fatigue characteristics (such as head tilting or nodding, frequent blinking, yawning) and distraction characteristics (such as using a mobile phone, turning the head to interact with passengers, eating, etc.). By constructing a targeted dataset and optimizing the YOLO model architecture, this system achieves high-precision identification of these subtle or obvious behavioral changes in the driver with extremely low latency.