Scholarlink

Accurate prediction of vehicle trajectories is crucial for bypassing traffic congestion and enhancing road safety, which are key components of realizing Intelligent Transportation Systems (ITS). This capability supports the development of advanced driver assistance, autonomous vehicles, and smart traffic management strategies, especially in mixed traffic conditions. The real-time prediction of the future motion of the vehicle brings about advanced driver assistance, the development of autonomous cars, and improvements in smart traffic management. This paper presents a new approach based on Liquid Neural Networks (LNN) together with the pNEUMA dataset acquired by Unmanned Aerial Vehicles (UAVs) to model and predict vehicle trajectories over the complex urban environment of Athens, Greece. In order to further increase the predictive accuracy, the minimum redundancy maximum relevance (mRMR) technique for feature selection and dimensionality reduction was used. The selected features are taken and incorporated within the LNN architecture to perform better than generic machine learning models. The introduced model ensures a reliable and robust trajectory prediction solution under challenging urban traffic conditions. This research is essential for building effective ITS by proposing scalable approaches to improve traffic management and safety in dense urban areas, ultimately making transportation systems smarter and more efficient.

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

Vehicle Trajectory Prediction in Congested Urban Traffic Leveraging Liquid Neural Network and UAV Data

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Mohammad Reza Mohebbi,
  • Maryam Tavasoli,
  • Mario Döller,
  • Javad Mohebbi Najm Abad,
  • Mohammad Javad Hassan Zada,
  • Iuliia Yamnenko

摘要

Accurate prediction of vehicle trajectories is crucial for bypassing traffic congestion and enhancing road safety, which are key components of realizing Intelligent Transportation Systems (ITS). This capability supports the development of advanced driver assistance, autonomous vehicles, and smart traffic management strategies, especially in mixed traffic conditions. The real-time prediction of the future motion of the vehicle brings about advanced driver assistance, the development of autonomous cars, and improvements in smart traffic management. This paper presents a new approach based on Liquid Neural Networks (LNN) together with the pNEUMA dataset acquired by Unmanned Aerial Vehicles (UAVs) to model and predict vehicle trajectories over the complex urban environment of Athens, Greece. In order to further increase the predictive accuracy, the minimum redundancy maximum relevance (mRMR) technique for feature selection and dimensionality reduction was used. The selected features are taken and incorporated within the LNN architecture to perform better than generic machine learning models. The introduced model ensures a reliable and robust trajectory prediction solution under challenging urban traffic conditions. This research is essential for building effective ITS by proposing scalable approaches to improve traffic management and safety in dense urban areas, ultimately making transportation systems smarter and more efficient.