Drowsiness Detection System Using Computer Vision for Performance Characterization on a Raspberry Pi

Abstract

There is a need to monitor individuals performing critical tasks who are affected by drowsiness or fatigue using computer vision detection processes on embedded devices, but this presents a challenge due to the use of complex processing hardware. Previous research describes how embedded devices and microcontrollers perform drowsiness detection tasks using algorithms optimized for limited resources. Additionally, facial recognition methods, such as Haar Cascades or convolutional neural networks, are used to identify fatigue based on facial features. This research implements a computer vision-based drowsiness detection system that calculates facial metrics on a Raspberry Pi 4, using MediaPipe and OpenCV tools, integrating an ESP32 module for MQTT alerts, and a remote monitoring platform with Node-RED. The system recorded an average processing time per frame of 129.08 ms, with increases during drowsiness events (eye closure at 270 ms and yawning at 230 ms). Furthermore, the Raspberry Pi analysis showed an average CPU usage of 51%, indicating its stability despite variations in computational load. The results indicate that it is possible to detect drowsiness and yawning states, characterizing their computational performance, although it has limitations due to the camera resolution and processing time.