Real-time Driver Drowsiness Detection for Android Application Using Deep Neural Networks Techniques (1811.01627v1)

Abstract: Road crashes and related forms of accidents are a common cause of injury and death among the human population. According to 2015 data from the World Health Organization, road traffic injuries resulted in approximately 1.25 million deaths worldwide, i.e. approximately every 25 seconds an individual will experience a fatal crash. While the cost of traffic accidents in Europe is estimated at around 160 billion Euros, driver drowsiness accounts for approximately 100,000 accidents per year in the United States alone as reported by The American National Highway Traffic Safety Administration (NHTSA). In this paper, a novel approach towards real-time drowsiness detection is proposed. This approach is based on a deep learning method that can be implemented on Android applications with high accuracy. The main contribution of this work is the compression of heavy baseline model to a lightweight model. Moreover, minimal network structure is designed based on facial landmark key point detection to recognize whether the driver is drowsy. The proposed model is able to achieve an accuracy of more than 80%. Keywords: Driver Monitoring System; Drowsiness Detection; Deep Learning; Real-time Deep Neural Network; Android.

• The paper presents a lightweight drowsiness detection model that compresses a heavy baseline into a 100KB solution for Android systems.
• It employs a Multilayer Perceptron with facial landmark analysis extracted from video frames to achieve an accuracy rate of approximately 81%.
• The study highlights real-time performance and reduced computational demands, facilitating practical integration into cost-effective embedded applications.

Real-time Driver Drowsiness Detection Utilizing Lightweight Deep Learning Models on Android Systems

This paper presents a method for real-time driver drowsiness detection using deep neural network techniques suitable for deployment within Android applications. The authors propose a solution aimed at enhancing the accessibility and practicality of drowsiness detection systems, which are traditionally limited by hardware requirements and are often available only in luxury vehicles. The paper leverages advancements in deep learning and mobile technology to provide a system that is broadly applicable, cost-effective, and able to mitigate one of the leading causes of road accidents globally.

Methodology Overview

The research utilizes a Multilayer Perceptron (MLP) classifier to detect facial landmarks and assess driver drowsiness. The core innovation within this paper is the compression of a heavyweight baseline model into a lightweight architecture. This enables seamless integration into Android systems, which typically face limitations in computational power and storage capacity. The model achieves an accuracy rate of approximately 81%, demonstrating sufficient efficacy for practical applications.

The system is based on five consecutive stages, beginning with video extraction from the NTHU Driver Drowsiness Detection Dataset. Images are then generated from video frames, followed by facial landmark detection using the Dlib library. Model training is conducted with these landmarks to create a predictive system which identifies driver drowsiness through real-time analysis. The trained model is ultimately applied within an Android framework, ensuring responsiveness by alerting drivers both visually and audibly.

Data and Computational Findings

The experimental analysis incorporates 200 videos from the NTHU dataset, encompassing diverse driving scenarios with varying lighting conditions and eyewear. The accuracy rates were highest in scenarios without sunglasses, highlighting limitations in facial landmark detection under obstructed view conditions. Computational testing on conventional hardware further confirmed the model's efficiency, achieving execution times of 43 milliseconds with GTX 1080, characterizing an optimal blend of precision and computational speed.

Implications and Future Work

The lightweight model proposed is notably advantageous for embedded systems and represents a significant reduction in model size compared to conventional CNN-based models, which often require hundreds of megabytes of storage. The proposed solution occupies only 100 KB making it feasible for integration into large-scale production environments and mobile applications, thereby enhancing the practicality of driver-monitoring systems.

A future direction for this research could encompass further refinement for performance improvement by addressing distractions and yawning detection alongside current fatigue metrics. Enhancing model robustness under conditions of varying lighting and eyewear would also be beneficial for broadening application capabilities and improving accuracy rates.

While this paper contributes valuable insights into the deployment of deep learning models on mobile systems for driver safety, ongoing research is warranted to further reduce computational requirements and to increase detection precision, ensuring greater applicability across automotive platforms and widespread adoption.

The results herein are supported by funding from the Qatar National Research Fund, underscoring the potential regional and international impact of this paper in advancing public safety through innovative technological solutions.