Scholarlink

<p>Predicting how pedestrians behave around vehicles is a major challenge in Advanced Driving Assistance Systems (ADAS) and Autonomous Vehicles (AV) in the real world. One particular problem involves predicting pedestrians crossing intention, which is influenced by various factors such as arrival time, waiting duration, the presence of a crosswalk, and the personal characteristics of both pedestrians and drivers. These factors profoundly affect how pedestrians interact with vehicles, making accurate predictions of their intentions a complex and crucial aspect of the ADAS functionality. However, these factors have not been fully explored in existing benchmarking datasets in the literature, such as Joint Attention in Autonomous Driving (JAAD) and Pedestrian Intention Estimation (PIE). In this research, we used a recent dataset named Distributed Simulation Study (DSS) for two major tasks: firstly, to analyze the interaction factors between pedestrians and drivers and secondly, to predict pedestrian crossing attention. The latter task is treated as a classification problem for which we employed machine learning models, such as Random Forest (RF), and deep learning models, such as Graph Neural Networks (GNN). The proposed models use information from only a single frame for prediction, making it well-suited for real-time applications. This approach eliminates the need for sequential data processing while maintaining high accuracy. Our method outperforms the baseline implementations of the DSS dataset by 8.1% in accuracy. The implementation is available at <a href="https://github.com/YoussefTaki/CrossOrWait">https://github.com/YoussefTaki/CrossOrWait</a>.</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

Interaction-based models for cross or wait scenarios: understanding driver-pedestrian dynamics

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Taki Youssef,
  • Elmoukhtar Zemmouri

摘要

Predicting how pedestrians behave around vehicles is a major challenge in Advanced Driving Assistance Systems (ADAS) and Autonomous Vehicles (AV) in the real world. One particular problem involves predicting pedestrians crossing intention, which is influenced by various factors such as arrival time, waiting duration, the presence of a crosswalk, and the personal characteristics of both pedestrians and drivers. These factors profoundly affect how pedestrians interact with vehicles, making accurate predictions of their intentions a complex and crucial aspect of the ADAS functionality. However, these factors have not been fully explored in existing benchmarking datasets in the literature, such as Joint Attention in Autonomous Driving (JAAD) and Pedestrian Intention Estimation (PIE). In this research, we used a recent dataset named Distributed Simulation Study (DSS) for two major tasks: firstly, to analyze the interaction factors between pedestrians and drivers and secondly, to predict pedestrian crossing attention. The latter task is treated as a classification problem for which we employed machine learning models, such as Random Forest (RF), and deep learning models, such as Graph Neural Networks (GNN). The proposed models use information from only a single frame for prediction, making it well-suited for real-time applications. This approach eliminates the need for sequential data processing while maintaining high accuracy. Our method outperforms the baseline implementations of the DSS dataset by 8.1% in accuracy. The implementation is available at https://github.com/YoussefTaki/CrossOrWait.