Drowsiness in drivers is a common phenomenon often experienced by commuters. It is also a major contributor to the number of deaths in motor vehicle traffic incidents, causing around 1.6% deaths on US highways in 2021. To mitigate this, the development of a detection and alert system for drowsy drivers has received a steady influx of studies. A common approach to detect drowsiness in these systems utilizes visual cues that are then processed by computer vision technology. Most of these studies use PERCLOS (percentage closure of eyes), whose accuracy depends on categorizing eye states to determine the drowsiness level of a driver. This paper seeks to study the performance of 3 well-known deep learning models that could perform this task: MobileNetV2, EfficientNetB0, and NASNet Mobile. These models are then trained utilizing a subset of the MRL Eye dataset. Our team found that EfficientNet B0 performed the best out of all the other models in this task, achieving an average accuracy of 98.5%.
Driver Drowsiness Detection Using NasNet Mobile, MobileNetV2, and EfficientNetB0
2024-02-21
496504 byte
Conference paper
Electronic Resource
English
IEEE | 2020
|Springer Verlag | 2017
|Overview of Research on Driver Drowsiness Definition and Driver Drowsiness Detection
British Library Conference Proceedings | 1994
|