Abstract: Fatigue detection of construction machinery operators is a complicated and difficult issue in construction safety management. There are various problems with the existing fatigue detection methods. Traditional fatigue questionnaire is difficult to quantify and poor timeliness. At the same time, the physiological based examination is highly invasive and costly. Computer Vision (CV) technology is a valuable technique for fatigue detection based on facial information collection and analysis. This paper proposes an approach for detecting fatigue operations of construction equipment operators based on CV. This approach uses the dlib model to mark 68 face feature points, and then calculates the real-time Eye Aspect Ratio (EAR) and Mouth Aspect Ratio (MAR) values, and takes the first 30 seconds of video as a sample to calculate the corresponding threshold, and then calculates the four indicators of blink frequency, average blink duration, Percentage of Eyelid Closure over the Pupil (PERCLOS) and yawn frequency. After that, normalized methods are employed to integrate indicators, and different fatigue countermeasures can be taken according to the value and duration of the comprehensive fatigue index. Finally, the accuracy of the method is verified through experiments. The results show that the comprehensive fatigue index proposed in this paper can reflect the trend of the fatigue state of construction equipment operators under different situations, and the correct rate of blink judgment is about 95%.