An Efficient Data Collecting Method for Enhanced Real-Time Drowsiness Detection Systems

In the past few years, there have been many traffic accidents, of which the leading cause was the decrease in driver’s focus while driving. To help alleviate road crashes owing to the reason mentioned above, an assistance system that can detect driver’s drowsiness based on eye-blink information is proposed in this paper. Our proposal system encompasses three main stages. To begin with, we mount a camera on the car dash to capture the driver’s upper body image. The next stage is face and eye detection. More concretely, when a face is detected, the eye regions will be extracted using Facial Landmark Detection. Additionally, we introduce a dataset to improve the robust performance of a deep learning-based drowsiness detection method. To this end, the eye-blink frequency will be collected by computing the Eye Aspect Ratio (EAR) from the eye regions and then passed to a lightweight Long Short-Term Memory (LSTM) model to determine the driver’s awareness level. Last but not least, if the driver is drowsy, the system will produce a sound to warn participants to regain focus. The system’s testing hardware is a Raspberry Pi board, a camera, and an alarm speaker. The Raspberry kit is a versatile, low-power consumption processor, combined with its affordable price, demonstrated that it is ideally suited to be designed as our system’s central processing unit. The drowsiness detection result with compelling accuracy suggests that this research has the potential for practical applications.