Scholarlink

<p>The increasing integration of Internet of Things (IoT) technologies into modern vehicles has enabled real-time access to driving data. However, analyzing this information in embedded environments remains challenging due to the continuous nature of the data streams and the limited computational resources available. This study addresses these constraints by proposing an evolving online clustering approach for real-time monitoring of driver behavior. The proposed method integrates Tiny Machine Learning (TinyML) techniques with soft sensors derived from physical data captured via on-board diagnostics (OBD-II). Central to the system is the MMCloud algorithm, which incrementally updates clusters without requiring retraining, enabling adaptation to concept drift and evolving driving conditions. The methodology involves the real-time acquisition of vehicle signals, including speed, RPM, engine load, and throttle position, followed by the calculation of a soft-sensor metric known as radar area. The algorithm was deployed on a Freematics One+ device and validated through case studies conducted in varied urban driving conditions. The results indicate that the system can classify driver behavior into cautious, normal, and aggressive categories, achieving Silhouette Scores up to 0.83 and maintaining execution times between 600–700 <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\mu s\)</EquationSource> </InlineEquation> in most cases. These findings suggest that the approach is suitable for embedded applications requiring adaptive, real-time behavior analysis.</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

Evolving online clustering for real-time driver behavior analysis in IoT-enabled vehicles

  • 期刊论文

  • 发表日期：

    2025年9月30日

查看全文
  • Morsinaldo Medeiros,
  • Marianne Silva,
  • Ivanovitch Silva

摘要

The increasing integration of Internet of Things (IoT) technologies into modern vehicles has enabled real-time access to driving data. However, analyzing this information in embedded environments remains challenging due to the continuous nature of the data streams and the limited computational resources available. This study addresses these constraints by proposing an evolving online clustering approach for real-time monitoring of driver behavior. The proposed method integrates Tiny Machine Learning (TinyML) techniques with soft sensors derived from physical data captured via on-board diagnostics (OBD-II). Central to the system is the MMCloud algorithm, which incrementally updates clusters without requiring retraining, enabling adaptation to concept drift and evolving driving conditions. The methodology involves the real-time acquisition of vehicle signals, including speed, RPM, engine load, and throttle position, followed by the calculation of a soft-sensor metric known as radar area. The algorithm was deployed on a Freematics One+ device and validated through case studies conducted in varied urban driving conditions. The results indicate that the system can classify driver behavior into cautious, normal, and aggressive categories, achieving Silhouette Scores up to 0.83 and maintaining execution times between 600–700 \(\mu s\) in most cases. These findings suggest that the approach is suitable for embedded applications requiring adaptive, real-time behavior analysis.