<![CDATA[Driving safety is a crucial issue significantly influenced by the driver's physical condition, where fatigue and drowsiness are major factors causing traffic accidents. This study aims to develop a real-time drowsiness detection system utilizing Edge AI technology based on the MobileNet architecture. This architecture was selected due to its efficiency in performing image classification on resource-constrained devices. The dataset used consists of 4,000 digital images balanced into open-eye and closed-eye classes. The model was trained using the TensorFlow framework and optimized through post-training quantization into the TensorFlow Lite format to reduce model size and inference latency. Performance evaluation was conducted by testing 372 new test images. The results indicate that the balanced model achieved an accuracy rate of 94%. Confusion matrix analysis showed a precision value of 1.000 for the closed-eye class and a recall of 1.000 for the open-eye class, indicating that the system is highly reliable in minimizing detection errors. With processing speeds reaching 10 to 22 Frames Per Second (FPS) on edge devices, this system is proven effective for implementation as a responsive driving safety assistant. Drowsiness detection duration indicator “Closed: 0.32s” represents part of the system logic used to trigger an alert. The system does not immediately activate an alarm during normal blinking, it measures the duration of eye closure. If the duration exceeds a predefined threshold (e.g., >0.30 seconds), an alert is triggered in the form of an audible alarm]]>
Copyrights © 2026
