The invention discloses a pedestrian trajectory prediction system and method based on global information extraction and a Transform model. The system is composed of a global information extraction module, a Transform backbone network and a CVAE network. The global information extraction module is composed of an own vehicle speed extraction network and a space motion conversion model and is responsible for estimating the own vehicle speed and obtaining the space motion relation between the own vehicle and a target pedestrian, and then extraction of global information such as the own motion characteristics of the vehicle and the motion relation between the pedestrian historical motion trail and the pedestrian is completed. The Transform backbone network module is responsible for learning pedestrian motion trails, by introducing a sparse attention mechanism, optimization of the attention mechanism is achieved, memory consumption of the network is reduced, prediction efficiency of the network is improved, and the problems that a pedestrian trajectory prediction algorithm is low in operation efficiency and large in memory consumption are solved. The CVAE network is responsible for generating a plurality of reasonable prediction trajectory samples, so that the over-fitting of training data is avoided, and the robustness and universality of the algorithm are improved.

    本发明公开了一种基于全局信息提取和Transformer模型的行人轨迹预测系统及方法,系统由全局信息提取模块、Transformer主干网络和CVAE网络分组成。全局信息提取模块由自车车速提取网络与空间运动转换模型组成,负责对自车车速的估计以及获取自车与目标行人的空间运动关系,进而完成车辆自身运动特性以及行人历史运动轨迹与人车间的运动关系等全局信息的提取。Transformer主干网络模块负责学习行人运动轨迹,通过引入稀疏注意力机制,实现对注意力机制的优化,减少网络的内存消耗并提高网络的预测效率,解决了行人轨迹预测算法运行效率低,内存消耗大的问题。CVAE网络负责生成多个合理的预测轨迹样本,避免了对训练数据的过拟合,提高了算法的鲁棒性与普适性。


    Access

    Download


    Export, share and cite



    Title :

    Pedestrian trajectory prediction system and method based on global information extraction and Transform model


    Additional title:

    一种基于全局信息提取和Transformer模型的行人轨迹预测系统及方法


    Contributors:
    HE YOUGUO (author) / YANG YONGXIN (author) / YUAN CHAOCHUN (author) / CAI YINGFENG (author) / TIAN LIWEI (author)

    Publication date :

    2024-05-17


    Type of media :

    Patent


    Type of material :

    Electronic Resource


    Language :

    Chinese


    Classification :

    IPC:    G06T Bilddatenverarbeitung oder Bilddatenerzeugung allgemein , IMAGE DATA PROCESSING OR GENERATION, IN GENERAL / G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen / G06V / G08G Anlagen zur Steuerung, Regelung oder Überwachung des Verkehrs , TRAFFIC CONTROL SYSTEMS



    Enhancing Pedestrian Trajectory Prediction with Crowd Trip Information

    Tamaru, Rei / Li, Pei / Ran, Bin | ArXiv | 2024

    Free access

    Pedestrian trajectory prediction method and device

    ZHENG WEN / LIU CHUANG / XU GUANGYU et al. | European Patent Office | 2023

    Free access


    A Pedestrian Trajectory Prediction Model Based on Interactive Attention Mechanism

    Dou, Wanqing / Lu, Lili / Huang, Yamei | Springer Verlag | 2024


    Crossing pedestrian trajectory prediction method and device

    WANG QING / TAN XIUQUAN | European Patent Office | 2022

    Free access