| 258 | 2 | 131 |
| 下载次数 | 被引频次 | 阅读次数 |
时空图被广泛应用于行人轨迹预测等时间序列任务中，如何更精确地捕捉不同时间段的轨迹位置信息以及更充分地利用时空图的结构信息对于轨迹预测至关重要。传统的轨迹预测方法规则复杂、约束性强、可扩展性较差，往往只能应用于特定领域。基于学习的轨迹预测方法不依赖于专家经验的物理规则，根据观察的轨迹数据来学习不同时间段各个空间位置之间的变化规则。基于学习的方法存在一定局限性，如没有充分利用时空图的结构信息，导致轨迹预测模型的性能下降。针对上述问题，提出了一种新型基于时空图联合关系路径的行人轨迹预测框架(Spatio-Temporal Graphs with Relationship Path Trajectory Prediction Framework, STRP-TPF)。STRP-TPF主要包括EdgeRNN和NodeRNN模型。STRP-TPF基于时空图构建关系路径，基于关系路径构建因子图；构建EdgeRNN和NodeRNN模型，并将因子图作为输入；输出下一时刻行人的位置，并且预测完整的行人轨迹。STRP-TPF利用关系路径能够准确捕捉时空图的结构信息，充分学习行人在不同时间和空间点的轨迹关系。大量实验结果表明，在ETH和UCY数据集上，STRP-TPF的整体性能均优于目前最先进的方法。在平均位移误差和最终位移误差方面，STRP-TPF比目前最先进方法低32.6%和37.7%。STRP-TPF的预测轨迹能够更准确地匹配真实轨迹。
Abstract：Spatio-temporal graphs are widely used in time-series tasks such as pedestrian trajectory prediction. How to accurately capture the trajectory position information in different time periods and utilize the structure information of spatio-temporal graphs is crucial for trajectory prediction. Traditional trajectory prediction methods have complex rules, strong constraints and poor scalability, and can only be applied to specific areas. In recent years, learning-based trajectory prediction methods do not rely on the physical rules of expert experience, and learn the variation rules between various spatial locations in different time periods according to the observed trajectory data. However, learning-based methods also have certain limitations, such as not fully utilizing the structure information of the spatio-temporal graph, which leads to the degradation of the performance of the trajectory prediction model. To cope with the above problems, a new pedestrian trajectory prediction framework based on spatio-temporal graph with relationship path called Spatio-Temporal Graphs with Relationship Path Trajectory Prediction Framework(STRP-TPF) is proposed. STRP-TPF mainly includes EdgeRNN and NodeRNN models. Firstly, STRP-TPF constructs a relationship path based on the spatio-temporal graph; secondly, STRP-TPF constructs a factor graph based on the relationship path; then, the EdgeRNN and NodeRNN models are constructed, and the factor graph is used as input data; finally, the location of pedestrian at the next moment is output, and the complete pedestrian trajectory is predicted. By using relationship path, STRP-TPF can accurately capture the structure information of the spatio-temporal graph, and fully learn the trajectory relationship of pedestrians at different time and space points. The experimental results show that, the overall performance of STRP-TPF is better than the state-of-the-art methods on the ETH and UCY datasets. In terms of average displacement error and final displacement error, STRP-TPF is 32.6% and 37.7% lower than the state-of-the-art methods. The predicted trajectories of STRP-TPF can accurately match the real trajectories.
[1] CHEN G X,HU L,ZHANG Q Z,et al.ST-LSTM:Spatio-temporal Graph Based Long Short-Term Memory Network for Vehicle Trajectory Prediction[C]//2020 IEEE International Conference on Image Processing (ICIP) .Abu Dhabi:IEEE,2020:608-612.
[2] ZAMBONI S,KEFATO Z T,GIRDZIJAUSKAS S,et al.Pedestrian Trajectory Pediction with Cnvolutional Nural Networks[J/OL].(2021-09-16)[2022-03-10].https://arxiv.org/abs/2010.05796.
[3] YU C J,MA X,REN J W,ZHAO H Y,et al.Spatio-temporal Graph Transformer Networks for Pedestrian Trajectory Prediction[J/OL].(2020-07-24)[2022-03-05].https://arxiv.org/abs/2005.08514.
[4] GUPTA A,JOHNSON J,LI F F,et al.Social GAN:Socially Acceptable Trajectories with Generative Adversarial Networks[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Salt Lake City:IEEE,2018:2255-2264.
[5] HUANG Y F,BI H K,LI Z X.STGAT:Modeling Spatial-temporal Interactions for Human Trajectory Prediction[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV).Seoul:IEEE,2019:6271-6280.
[6] VINTR T,YAN Z,DUCKETT T.Spatio-temporal Representation for Long-term Anticipation of Human Presence in Service Robotics[C]//2019 International Conference on Robotics and Automation (ICRA).Montreal:IEEE,2019:2620-2626.
[7] ALEXS A,ALBERTI C,BRUNO S.A Comparative Study of Spatio-temporal U-Nets for Tissue Segmentation in Surgical Robotics[J].IEEE Transactions on Medical Robotics and Bionics,2021,3(1):53-63.
[8] JOHNM P.Spatio-temporal Deep Learning for Robotic Visuomotor Control[C]//2018 4th International Conference on Control,Automation and Robotics.Aucklan:IEEE,2018:94-103.
[9] HAN M,LIU B.Ensemble of Extreme Learning Machine for Remote Sensing Image Classification[J].Neurocomputing,2015,149:65-70.
[10] ASIMINA D,TRIANTAFYLLIDIS D,STELIOS K,et al.MLP for Spatio-temporal Traffic Volume Forecasting [C]∥2021 IEEE International IOT,Electronics and Mechatronics Conference (IEMTRONICS).Toronto:IEEE,2021:1-7.
[11] LU F,CHEN B F,ZHOU X D.STA-VPR:Spatio-temporal Alignment for Visual Place Recognition[J].IEEE Robotics and Automation Letters,2021,6(3):4297-4304.
[12] LIN J F,ZHANG L,HE M,et al.Multi-path Relationship Preserved Social Network Embedding[J].IEEE Access,2019,7:26507-26518.
[13] ZHANG Q,CHANG J L,MENG G F,et al.Spatio Temporal Graph Structure Learning for Traffic Forecasting[C]∥Proceedings of the AAAI Conference on Artificial Intelligence.New York:AAAI,2020:1177-1185.
[14] CHEN K,CHEN F,LAI B S,et al.Dynamic Spatio-temporal Graph-based CNNs for Traffic Flow Prediction[J].IEEE Access,2020,8:185136-185145.
[15] ZENG P,TAN Q P,MENG X K.Modeling Complex Relationship Paths for Knowledge Graph Completion[J].IEICE Transactions on Information and Systems,2018,101(5):1393-1400.
[16] GOLDHAMMER M,GERHARD M,ZERNETSCH S,et al.Early Prediction of a Pedestrian’s Trajectory at Intersections[C]//16th International Conference on Intelligent Transportation Systems.The Hague:IEEE,2013:237-242.
[17] RUDENKO A,PALMIERI L,HERMAN M.Human Motion Trajectory Prediction:A Survey[J/OL].(2019-12-17)[2022-03-12].https://arxiv.org/abs/1905.06113.
[18] KIM J,JEONG S,YU Z B.Multiple Timescale Recurrent Neural Network with Slow Feature Analysis for Efficient Motion Recognition[C]//Proceedings of the 20th International Conference Neural Information Processing.Daegu:[s.n.],2013:323-330.
[19] ZHANG P,OUYANG W L,ZHANG P F.SR-LSTM:State Refinement for LSTM Towards Pedestrian Trajectory Prediction[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Long Beach:IEEE,2019:12077-12086.
[20] VEMULA A,MUELLING K,OH J.Social Attention:Modeling Attention in Human Crowds[J/OL].(2018-10-29)[2022-03-20].https://arxiv.org/abs/1710.04689.
[21] HIROSE Y S,SHIMBO M,WATANABE T.Transductive Data Augmentation with Relational Path Rule Mining for Knowledge Graph Embedding [C]//2021 IEEE International Conference on Big Knowledge (ICBK).Aucklan:IEEE,2021:377-384.
[22] PELLEGRINI S,ESS A,SCHINDLER K.You’ll Never Walk Alone:Modeling Social Behavior for Multi-target Tracking [C]//2009 IEEE 12th International Conference on Computer Vision.Kyoto:IEEE,2009:261-268.
[23] LERNER A,CHRYSANTHOU Y,LISCHINSKI D.Crowds by Example[J].Computer Graphics Forum,2007,26(3):655-664.
[24] ALAHI A,GOEL K,RAMANATHAN V,et al.Social LSTM:Human Trajectory Prediction in Crowded Spaces[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).Las Vegas:IEEE,2016:961-971.
[25] ASHESH J,AMIR Z,SILVIO S,et al.STRUCTURAL-RNN:Deep Learning on Spatio-temporal Graphs[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).Las Vegas:IEEE,2016:5308-5317.
基本信息:
DOI：
中图分类号:U491
引用信息:
[1]孙科,鄢府,范勇强等.基于时空图联合关系路径的行人轨迹预测框架[J].无线电工程,2023,53(02):281-289.
基金信息:
四川省科技计划资助(2021JDJQ0021,2022YFG0186)~~