Scholarlink

<p>The increasing frequency of alcohol-related accidents highlights the need for efficient methods to identify intoxicated individuals. Traditional techniques like breathalyzers and direct cooperation face challenges, especially in large-scale or public safety scenarios. In response, we present a novel, markerless approach for detecting drunkenness through gait analysis, utilizing standard cameras. Our method applies MobileNetV2, a convolutional neural network, to analyze gait energy images and classify individuals as either drunken or sober. By employing background removal, the neural network focuses solely on the human body, improving detection accuracy. The dataset consists of 60 walking videos, with equal representation of drunken and non-drunken states. Through the use of image augmentation, we achieved an 85.64% accuracy in classification, highlighting the potential of vision-based gait analysis for real-time public safety applications aimed at preventing alcohol-related accidents.</p> Graphical abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

MobileNetV2-driven gait analysis for drunkenness detection: a scalable vision-based approach for enhancing public safety

  • 期刊论文

  • 发表日期：

    2026年1月23日

查看全文
  • S. Abirami,
  • D. Amuthaguka

摘要

The increasing frequency of alcohol-related accidents highlights the need for efficient methods to identify intoxicated individuals. Traditional techniques like breathalyzers and direct cooperation face challenges, especially in large-scale or public safety scenarios. In response, we present a novel, markerless approach for detecting drunkenness through gait analysis, utilizing standard cameras. Our method applies MobileNetV2, a convolutional neural network, to analyze gait energy images and classify individuals as either drunken or sober. By employing background removal, the neural network focuses solely on the human body, improving detection accuracy. The dataset consists of 60 walking videos, with equal representation of drunken and non-drunken states. Through the use of image augmentation, we achieved an 85.64% accuracy in classification, highlighting the potential of vision-based gait analysis for real-time public safety applications aimed at preventing alcohol-related accidents.

Graphical abstract