Drunk Driving Detection Using Two-Stage Deep Neural Network

Abstract

Drunk driving accidents have been rapidly increasing in recent times. Although the statistics show a decreasing trend in recent years, reports of drunk driving accidents are often seen in the news. To assess vehicle operators for drunk driving, the police still use breath-alcohol testers as the primary method. However, a certified instrument to measure alcohol consumption is expensive, and the mouthpiece used in the instrument is a consumable. Moreover, the breath detection method used involves contact measurement, which may cause hygiene concerns. To achieve more convenient and accurate detection, many researchers have proposed methods to replace the traditional breath-type measurement instruments. The present study proposes a two-stage neural network for recognition of drunk driving: the first stage uses the simplified VGG network to determine the age range of the subject, and the second stage uses the simplified Dense-Net to identify the facial features of drunk driving. The age discrimination stage obtained an accuracy of 86.36%. In addition, in drunk driving recognition tests among various age groups (18-30, 31-50, and ≥51 years), accuracies of 94%, 83%, and 81% were obtained, respectively. The overall system also showed a high accuracy of 89.62% and 87.44%, which proves the robustness of the system while supporting its practical application.INDEX TERMS Alcohol test, artificial intelligence, convolutional neural networks, deep neural networks, drunk driving detection.

Comprehensive Assessment of Artificial Intelligence Tools for Driver Monitoring and Analyzing Safety Critical Events in Vehicles

“…Sharma and Sood [ 69 ] employed an alcohol sensor and air pressure sensor for sobriety checks, and ML algorithms for drivers’ drowsiness detection via camera. Chang et al [ 70 ] explored drunk driving detection via facial images and breath-alcohol tester from 124 subjects (ages 18–70) using simplified VGG and Dense-Net: VGG classified the age range of the subject while Dense-Net identified the facial features of drunk driving for alcohol test identification, as shown in Figure 6 C. The model achieved an accuracy of 87.44% and the results showed that (1) the ears, chin, forehead, neck, cheek, and other facial parts of subjects’ images are good characteristic areas for alcohol tests, and (2) age affects the identification results in the alcohol test.…”

Leveraging driver vehicle and environment interaction: Machine learning using driver monitoring cameras to detect drunk driving

“…Although existing work on driver state monitoring addresses safety-critical states such as drowsiness or distraction, previous attempts at detecting drunk driving with vehicle signals have not achieved sufficiently good results or were not rigorously evaluated (e.g., [17,35,47,48,52,83,98,100]). • Significance: Our system offers a viable approach based on existing technologies, allowing for a rapid implementation to prevent potential negative consequences of drunk driving after decades of stagnating high alcohol-related road crashes without any significant advancement by regulators and industry [67,90].…”

Efficient Driver Drunk Detection by Sensors: A Manifold Learning-Based Anomaly Detector

“…Recently in [25], a two-stage deep learning approach is proposed to detect drunk driving using a Convolutional Neural Network (CNN). At first, the simplified VGG (Visual Geometry Group) network, a standard CNN, is applied to estimate the driver's age, and then the simplified Dense-Net for identifying the facial features of drunk driving for alcohol test discrimination.…”