Fatigue Detection System for Extracting Driver's Eye Features

Traffic safety remains one of the most concerning issues for humans, with people dying in traffic accidents every moment, and nearly half of them being related to fatigue driving. When drivers feel fatigued, the eyes undergo significant changes. In this study, eye movement characteristics were utilized to detect the fatigue state of drivers, and a fatigue detection system was developed, combining the PERCLOS algorithm and the EAR algorithm, which were validated through experiments to assess system usability. The system was developed and designed based on traditional image processing algorithms in OpenCV and the facial feature recognition capabilities of the Dlib library. By using the 68-dimensional facial landmark detection model in the Dlib library, facial feature points were extracted, and eye tracking functionality was achieved through the feature points of the eyes. Subsequently, the PERCLOS algorithm, EAR algorithm, and a combination of the PERCLOS and EAR algorithms were employed. In this experiment, thresholds were set separately for the EAR and PERCLOS algorithms to compare the accuracy of eye movements. The system sets a threshold of 0.4 for PERCLOS, classifying it as fatigue when the proportion of closed eye time exceeds 0.4, and collects 20 sets of EAR data from the subject using an average value threshold of 0.2 to determine eye closure actions. Finally, through experiments monitoring the driver's eyes, the presence of fatigue state was determined, and the advantages and disadvantages of the three algorithms were summarized based on the experimental results. The experiment demonstrated that the Combined method algorithm has a more comprehensive detection capability compared to the PERCLOS and EAR algorithms, improving the fatigue detection performance.