Fine-Grained Intoxicated Gait Classification Using a Bilinear CNN.

Area of Science:

• * Computational neuroscience and machine learning applications.
• * Biomedical engineering and wearable sensor technology.
• * Public health and road safety initiatives.

Background:

• * Excessive alcohol consumption leads to impaired mobility, judgment, and significant daily injuries and fatalities from driving accidents.
• * Passive detection methods for intoxicated drivers can enable timely alerts and reduce Driving Under the Influence (DUI) incidents.
• * Smartphones possess motion sensors and processing capabilities suitable for gait analysis and machine learning model deployment.

Purpose of the Study:

• * To propose a novel method for detecting alcohol intoxication using smartphone gait data.
• * To leverage a Bi-linear Convolution Neural Network (BiCNN) for analyzing accelerometer and gyroscope data.
• * To determine if smartphone users exceed the legal driving limit (0.08 BAC) based on their gait patterns.

Main Methods:

• * Gait data segmented into steps and converted into Gramian Angular Field (GAF) images.
• * Utilization of a BiCNN for fine-grained image classification of GAF-encoded gait data.
• * Implementation of a comprehensive pipeline including step detection, data normalization, fusion, GAF generation, and BiCNN classification.

Main Results:

• * The proposed BiCNN model achieved an accuracy of 83.5% in classifying intoxicated vs. sober users.
• * The method successfully distinguished subtle differences in gait patterns indicative of intoxication.
• * Performance surpassed previous state-of-the-art methods for intoxication detection via gait analysis.

Conclusions:

• * The novel approach effectively utilizes smartphone motion sensing and BiCNN for passive alcohol intoxication detection.
• * This method demonstrates the feasibility of using gait analysis from everyday devices for road safety.
• * The findings pave the way for just-in-time alerts to prevent DUI incidents.