Human fatigue detection based on the machine vision methods is non-invasive, fast, and accurate and is unhindered by weather conditions. Owing to these advantages, this technique has gradually become a hot research topic worldwide. However, it is easily affected by complicated illumination and changes in the pilot position. To solve this problem, on the basis of previous studies on driver fatigue detection under complicated illumination conditions and postural changes, we propose a fatigue detection method based on the real-time enhanced constraint local model. First, the collected images are subjected to real-time high-dynamic-range enhancement. Then, the enhanced image is used to model the driver′s face in order to extract his/her vision and percentage of eye closure characteristics. Finally, the fatigue state is detected and an identification method based on Bayesian confidence networks is established. Our experimental findings show that the proposed method robustly detects the fatigue states of drivers under complex illumination and change in position.