With the continuous increase in the number of motor vehicles, fatigue driving has become one of the important factors leading to traffic accidents. With the rapid development of deep learning and computer vision technology, fatigue driving detection systems based on image and video analysis have gradually become a research hotspot. This article aims to achieve efficient detection and warning of fatigue driving behavior through deep learning technology combined with computer vision methods, and improve road safety. This study reviewed the relevant work in the field of fatigue driving detection and proposed a computer vision fatigue driving detection model based on deep learning. The model combines multiple information sources such as facial blink frequency, mouth yawning, and head nodding. The Yaw DD dataset and YOLOv7 algorithm were used to optimize feature fusion, and 2D and 3D facial features were used for precise localization and pose analysis for model training and optimization. The experimental results show that the multi-source feature fusion model proposed in this paper has achieved significant results in fatigue driving detection. In the ablation experiment, the model that integrates facial features and human head features improved accuracy by 6.969%, precision and recall by 6.969% and 2.626%, respectively. This indicates that the detection system optimized by deep learning can effectively improve the accuracy of fatigue state recognition. The innovation of this article lies in proposing a deep learning model based on multi-source feature fusion, which overcomes the limitations of traditional physiological signal monitoring methods by combining facial and head feature.