Drowsiness Classification for Internal Driving Situation Awareness on Mobile Platform

Nine, Julkar; Ahmed, Naeem; Mathavan, Rahul · 2021 · Crossref

DOI: 10.14464/ess.v8i2.491

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the critical safety issue of driver drowsiness, which contributes significantly to road accidents, fatalities, and economic losses. Motivated by the limitations of existing detection methods—such as subjective self-reporting, invasive physiological monitoring, and vehicle-based metrics—the authors propose a lightweight, non-invasive solution for real-time drowsiness classification on mobile platforms. The primary objective is to develop a system that utilizes facial landmarks and head pose estimation via Convolutional Neural Networks (CNNs) to detect sleepiness accurately while remaining computationally efficient enough for embedded devices and Android smartphones. The methodology involves a computer vision-based approach implemented on an Android application. The system captures video input from the device’s front-facing camera and processes it using the DLIB library to detect facial landmarks. Two key features are extracted: eye aspect ratio (EAR) to monitor blinking patterns and head pose estimation using the `cv2.solvePnP` algorithm to track head orientation. These features are fed into a CNN model trained to classify the driver’s state as active, sleepy, or drowsy. The software architecture leverages the Android Native Development Kit (NDK) and Java Native Interface (JNI) to integrate C++-based DLIB processing with the Android framework. The system was tested on various Android devices, including the Samsung S8 and Xiaomi Redmi 9, using a dataset of 40 training videos and 20 evaluation videos from over 10 subjects, covering scenarios with and without glasses and under varying lighting conditions. The results demonstrate that the proposed system achieves an overall identification accuracy of 86% across all tested classes. Performance varied by condition: accuracy reached 95% in daytime scenarios with or without white glasses, but dropped to 78% at night without glasses and 76% at night with white glasses. The system processed frames at an average speed of 240.3 fps on a Samsung S8, with a drowsiness detection time of 50.6 ms. The error rate was reported to be less than 0.4%. The study found that eye visibility is a significant factor; accuracy diminished when eyes were obscured by black glasses or when uneven illumination reduced facial feature clarity. However, the integration of head pose estimation helped strengthen reliability in these challenging conditions. The significance of this work lies in providing a cost-effective, accessible, and real-time drowsiness monitoring framework that does not require specialized hardware. By achieving high accuracy on standard mobile devices, the system offers a practical alternative to heavier, less accessible solutions. The authors conclude that this technology can be integrated into in-car dashboard cameras to provide timely warnings, potentially reducing accidents caused by driver fatigue. Future work aims to improve detection efficiency through extensive dataset training and further optimization for diverse environmental conditions.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• drowsiness detection algorithms
• drowsiness
• distraction detection algorithms
• microsleep
• gaze based attention detection
• dms validation

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: physiological data
• Methodological Resource: tool software, validation psychometrics