The popularity of COVID‐19 has rapidly affected our daily life. One of the solutions to prevention is to wear masks. At present, wearing masks has become a new norm. However, when a motor vehicle driver wears a mask, all the parts below his eyes are blocked. Most face detection algorithms can't extract effective information, which leads to low accuracy and completely unreliable precision, and can't achieve a fatigue detection effect. Therefore, in order to reduce the traffic accidents caused by this kind of fatigue driving, this paper first detects the face from the image, then accurately locates the pupil in the detected face area, compares some non‐deep learning algorithm face detectors, using the Residual Network (ResNet) to connect the SSD network to detect the face key points. The front structure of the network uses the residual module to construct the deep learning face detection algorithm of ResNet10SSD. The pupil circle detection algorithm of Hough transform is adopted to analyze the pupil opening of the eyes, and the driver's eye opening and closing state is effectively judged. Finally, the blink frequency and PERCLOS value are obtained. The experimental results show that this method can effectively detect the eye indicators under the condition of face occlusion, and it has good universality and effectiveness, thus further realizing fatigue monitoring.