Mobile Application Development for Drowsy Driving Alert using Convolutional Neural Networks

Drowsiness is one of the main causes of traffic accidents that can result in fatalities. Therefore, a system is needed that can directly detect driver drowsiness and provide early warnings. This study aims to design and build a drowsiness detection system using image processing and machine learning with the Convolutional Neural Network (CNN) method. The system uses two CNN models, namely the eye CNN to detect open or closed conditions, and the mouth CNN to detect yawning activity. As an additional filter, the Mouth Aspect Ratio (MAR) calculation is used to distinguish between yawning and talking. The training results show that the eye detection model achieves an accuracy of 99.82%, while the mouth detection model achieves an accuracy of 98.02%. The integration of MAR logic has proven to be effective in reducing detection errors when the user is talking. Application testing on 10 respondents in yawning, speaking, and blinking scenarios resulted in a 100% success rate in detecting drowsiness in real-time under normal lighting (300-350 lux). However, in tests with low light intensity (10-25 lux), the accuracy dropped to around 87-90%, with some detection errors (false negatives) and slower response times. This system is implemented in the form of an Android application with an easy-to-use interface, capable of providing alerts through sound, vibration, flashing lights, and notifications. With these results, the application has proven to be reliable in detecting driver drowsiness, although it still has limitations in low light conditions. This application is expected to help reduce the risk of traffic accidents caused by drowsy driving.