基于动态进化图的行人轨迹预测  

Pedestrian trajectory prediction based on dynamic evolving graph

作　　者：芈菁 张旭秀[1] 闫涵 MI Jing;ZHANG Xu-xiu;YAN Han(College of Automation&Electrical Engineering,Dalian Jiaotong University,Dalian 116021,China;College of Control Science and Engineering,Dalian University of Technology,Dalian 116024,China)

机构地区：[1]大连交通大学自动化与电气工程学院,辽宁大连116021 [2]大连理工大学控制科学与工程学院,辽宁大连116024

出　　处：《控制与决策》2024年第7期2345-2353,共9页Control and Decision

基　　金：国家自然科学基金项目(62103074);辽宁省重点研发计划项目(2022020594-JH1/108)。

摘　　要：行人轨迹预测在自动驾驶和社交机器人等领域有着广泛的应用.对行人间复杂的交互关系进行有效建模是提高轨迹预测准确性的关键问题.然而,基于图神经网络的方法建模行人间的复杂交互时,存在行人间交互关系不会随着时间推移而改变,并且图模型无法自适应地调整网络参数,导致预测轨迹与真实轨迹偏差较大.为此,提出基于动态进化图的行人轨迹预测方法,设计动态特征更新(DFU)以定义行人间的动态特性,对行人间动态交互进行建模以构建时间域的网络动态性,提升对行人间复杂交互关系建模的能力.采用进化图卷积单元优化编码器,灵活进化图模型网络参数,增强图模型的自适应能力.研究结果表明,在预测8个时间步长下,与STGAT模型相比,所提出模型在两个公开数据集(ETH和UCY)上取得了更好的性能,平均位移误差降低12.26%,最终位移误差降低14.10%.Pedestrian trajectory prediction has a wide range of applications in areas such as autonomous driving and social robotics.Effective modeling of complex interactions between pedestrians is a key issue to improve trajectory prediction accuracy.However,when modeling the complex interactions between pedestrians based on graph neural networks,there are problems that the interactions between pedestrians do not change over time and the graph model cannot adjust the network parameters adaptively,resulting in large deviations between predicted and real trajectories.Therefore,this paper proposes a pedestrian trajectory prediction method based on dynamic evolving graphs,and designs dynamic feature update(DFU)to define the dynamic characteristics between pedestrians and model the dynamic interactions between pedestrians to build the network dynamics in the time domain,which improves the ability to model the complex interactions between pedestrians.An evolving graph convolution unit optimization encoder is used to flexibly evolve the graph model network parameters and enhance the adaptive capability of the graph model.The results show that the proposed model achieves better performance on two publicly available datasets(ETH and UCY)with 12.26%reduction in average displacement error and 14.10%reduction in final displacement error compared with the STGAT model at predicted 8 time steps.

关 键 词：动态图 进化图 图卷积 轨迹预测 行人轨迹 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]