基于Transformer和多关系图卷积网络的行人轨迹预测  

Pedestrian Trajectory Prediction Based on Transformer and Multi-relation Graph Convolutional Networks

作　　者：刘桂红[1] 周宗润 孟祥福[1] LIU Guihong;ZHOU Zongrun;MENG Xiangfu(Liaoning Technical University,Huludao,Liaoning 125105,China)

机构地区：[1]辽宁工程技术大学,辽宁葫芦岛125105

出　　处：《计算机科学与探索》2025年第5期1353-1364,共12页Journal of Frontiers of Computer Science and Technology

基　　金：国家自然科学基金(61772249)。

摘　　要：在自动导航应用领域,行人轨迹相对复杂,有效且合理地预测行人未来轨迹对自动驾驶和出行安全至关重要。行人轨迹随机性和动态性极高且与交通环境有着复杂相互作用,因此需要对行人的时间依赖性和空间相互作用进行合理建模。为了解决该问题,提出了一种基于Transformer和多关系图卷积网络(GCN)的行人轨迹预测模型。该模型由交互捕获模块、锚点控制模块和轨迹修正补全模块构成。交互捕获模块由T-Transformer和多关系图卷积网络组成,分别提取每个行人在时间序列和空间序列上的运动特征,并结合锚点控制模块推断行人的中间目的地以减少递归累计误差,由修正补全模块进行最终轨迹细化。在提取特征时添加逆关系可得到更为优化的结果,使用高斯剪枝减少虚假路径的生成也可提高模型效率。在ETH与UCY数据集上的实验结果表明,在平均位移误差(ADE)和最终位移误差(FDE)方面,该模型具有比现有大部分主流模型更好的性能。由于该模型在行人轨迹预测上的出色性能,可避免不必要的轨迹变更和碰撞风险,为行人轨迹预测应用提供了更为可能的解决方案。In the field of autonomous navigation,pedestrian trajectories are relatively complex,and accurately pre-dicting pedestrian trajectories is crucial for ensuring safe travel and autonomous driving.Pedestrian trajectories are highly random,dynamic,and influenced by their surroundings,necessitating the effective modeling of their temporal and spa-tial interactions.To address this,a pedestrian trajectory prediction model combining Transformer and multi-relation graph convolutional network(GCN)is proposed.The model is composed of interaction capture module,anchor control module,and trajectory refinement module.The interaction capture module extracts motion features of each pedestrian on temporal and spatial sequences by using T-Transformer and GCN,while the anchor control module reduces errors by inferring inter-mediate destinations.The trajectory refinement module enhances predictions.Adding the inverse relationship when extract-ing features can obtain more optimized results,and using Gaussian pruning to reduce the generation of false paths can also improve the efficiency of the model.Experimental results on ETH and UCY datasets show superior performance in av-erage displacement error(ADE)and final displacement error(FDE)compared with mainstream models.The excellent per-formance of the model on pedestrian trajectory prediction minimizes unnecessary trajectory changes and collision risks,offer-ing a promising solution for pedestrian trajectory prediction applications.

关 键 词：T-Transformer 图卷积网络(GCN) 锚点控制 行人轨迹预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]