An Adaptive Fatigue Driving Detection Method Based on BP-SVM

This study proposes a Back Propagation algorithm into mechanism-Support Vector Machine model (BP-SVM), which dynamically adjusts the parameters of the model. In this paper, we used SILAB driving simulation software to simulate the driver’s driving state under different weather conditions (i.e., sunny and foggy) and to detect driver fatigue by monitoring vehicle behavior (i.e., acceleration, steering wheel turns, standard deviation of speed, and longitudinal acceleration). The effectiveness of the BP-SVM model was measured by comparing the performance of RF, VGG, and SVM models in terms of accuracy and ROC curve. The results show that the average accuracy of BP-SVM, RF, SVM, and VGG are 93.97%, 89.53%, 83.88%, and 91.95%, respectively, all of which are lower than that of BP-SVM. The AUC areas of RF, SVM, VGG, and BP-SVM are 0.94, 0.92, 0.83, and 0.99, respectively, which shows that the BP-SVM model outperforms other classification models, particularly the VGG model.