Scholarlink

<p>The emergence of autonomous vehicles has revolutionized road transportation, with lane-changing operations being crucial for cooperative driving with other road users. This study focuses on improving lane-change intention recognition and risk prediction, key aspects in ensuring the safety and efficiency of autonomous driving. Traditional research often treats these areas separately and lacks proper modeling of two-lane, bidirectional roads, reducing the robustness of Advanced Driver Assistance Systems (ADAS). To address these gaps, this study used virtual simulation tools to create a realistic two-lane, bidirectional road environment, collecting driver eye-tracking and vehicle interaction data from 80 participants. Approximately 900 lane-change events and over 9,000 data points were gathered. A Transformer-based intention recognition model using a novel Vehicle feature and Eye-tracking feature Labelling (VEL) method was developed, achieving a 96.6% accuracy 4.55&#xa0;s before a lane change. The VEL method enhanced model flexibility and robustness. Meanwhile, in the risk prediction domain, a CatBoost-based model, combined with fault tree analysis and fuzzy clustering, outperformed alternatives like Light, achieving 92.9% accuracy in classifying lane-change risk into four levels. The Shapley Additive exPlanations (SHAP) model was used to identify key risk factors, offering deeper understanding, and contributing to the intelligent development of ADAS. The integration of intention recognition and risk prediction provides a comprehensive solution for safer lane-changing operations in autonomous vehicles.</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

An active Safety Framework for Intelligent Vehicles Considering Lane-change Intention Recognition and Risk Prediction in Oncoming Traffic

  • 期刊论文

  • 发表日期：

    2025年11月18日

查看全文
  • Hao Liu,
  • Hui Wang,
  • Ming Chi,
  • Tao Wang,
  • Xinqiang Liu

摘要

The emergence of autonomous vehicles has revolutionized road transportation, with lane-changing operations being crucial for cooperative driving with other road users. This study focuses on improving lane-change intention recognition and risk prediction, key aspects in ensuring the safety and efficiency of autonomous driving. Traditional research often treats these areas separately and lacks proper modeling of two-lane, bidirectional roads, reducing the robustness of Advanced Driver Assistance Systems (ADAS). To address these gaps, this study used virtual simulation tools to create a realistic two-lane, bidirectional road environment, collecting driver eye-tracking and vehicle interaction data from 80 participants. Approximately 900 lane-change events and over 9,000 data points were gathered. A Transformer-based intention recognition model using a novel Vehicle feature and Eye-tracking feature Labelling (VEL) method was developed, achieving a 96.6% accuracy 4.55 s before a lane change. The VEL method enhanced model flexibility and robustness. Meanwhile, in the risk prediction domain, a CatBoost-based model, combined with fault tree analysis and fuzzy clustering, outperformed alternatives like Light, achieving 92.9% accuracy in classifying lane-change risk into four levels. The Shapley Additive exPlanations (SHAP) model was used to identify key risk factors, offering deeper understanding, and contributing to the intelligent development of ADAS. The integration of intention recognition and risk prediction provides a comprehensive solution for safer lane-changing operations in autonomous vehicles.