Driver Drowsiness Detection System based on Ensemble deep learning model

Abstract

Driver fatigue remains one of the most critical causes of road-traffic accidents. In the context, Drowsiness could be shown as a sleepiness state when a driver must have rest, with the emergence of several symptoms that have an impact on the performance of tasks. With the rapid progress of computer vision and deep learning frameworks, intelligent models have emerged as effective solutions for drowsiness detection. Therefore, this paper introduces drowsiness detection techniques using ensemble model based on deep learning algorithms. The ensemble model integrates ResNet50 for eye-state detection, VGG16 for mouth-state detection, and EfficientNetB3 for head position detection network pipelines with weighted average ensemble. The system operates on day and nighttime video frames generated by NITYMED and Yawdd. The ensemble model was assessed with three testing datasets with different conditions. These include daytime, nighttime, and a combined dataset. The model shows superiority in comparison with several existing models. It achieves an accuracy of 98%, 96% and 98% when testing on the combined dataset, daytime, and nighttime frames, respectively.