Abstract: In order to explore the performance of novice drivers’ fatigue characteristics in visual characteristics, this paper designs a driving simulation experiment based on driving simulators and Dikablis Glass 3 eye tracker to collect drivers’ visual data. Combined with the video expert method, drivers’ fatigue levels are divided into awakening, mild fatigue, moderate fatigue and severe fatigue. First of all, the data are cleaned by using Rayda criterion and Kalman filter. Secondly, according to the results of one-way ANOVA and post multiple comparison, mean value of blink duration, mean value of saccade duration, total saccade duration, mean value of pupil area, mean value of pupil variation coefficient, mean value of fixation time and other visual characteristics are selected as the evaluation indexes of drivers’ fatigue driving. Finally, a grey wolf optimized support vector machine (GWO-SVM) fatigue driving recognition model is constructed based on the visual features of novice drivers. The results show that, with the accumulation of driving fatigue of novice drivers, the duration of blinking increases significantly, the saccade duration and total saccade time decrease significantly, the pupil area decreases, and the pupil variation coefficient increases. The SVM identification results show that the fatigue state of novice drivers can be effectively identified by eye movement indicators, and the GWO-SVM model further improves the recognition accuracy, proving that eye movement features have a good applicability in the fatigue detection of novice drivers.