基于图模型和注意力机制的车辆轨迹预测方法

doi:10.12382/bgxb.2022.0117

摘要/Abstract

摘要：

为提高结构化道路场景下车辆轨迹预测准确度,提出一种基于图模型和注意力机制的多模态轨迹预测方法(GA-MTP)。构建车道图和车交互图,实现道路环境特征、车辆运动特征和车辆间交互特征建模;通过堆叠的注意力模块完成环境-车辆特征融合,统一交通场景静态特征和动态特征;由两分支解码网络模块得出最终轨迹预测和相应概率。在Argoverse数据集进行模型训练和测试,并进行结果分析。实验结果表明,新方法在结构化交通场景的车辆轨迹预测中取得优异的效果,预测准确程度优于当前主流方法。

关键词: 车辆轨迹预测, 图模型, 注意力机制, 多模态

Abstract:

In order to improve the accuracy of vehicle trajectory prediction in structured road scenes, a multimodal trajectory prediction method based on graph models and attention mechanism is proposed. For the purpose of modeling road environment characteristics, vehicle motion characteristics and characteristics of interaction between vehicles, the lane graph and vehicle interaction graph are constructed. Environment-vehicle feature fusion is completed by stacked attention modules, so as to realize the unification of static and dynamic features of traffic scenes. The final predicted trajectories and corresponding probabilities are obtained through a two-branch decoding module. The Argoverse dataset is used to train and validate the proposed method. The experimental results show that the proposed method achieves excellent performance of motion prediction in structured road scenes. The prediction accuracy is better than the current mainstream methods.

Key words: vehicle trajectory prediction, graph models, attention mechanism, multimodal

连静, 丁荣琪, 李琳辉, 王雪成, 周雅夫. 基于图模型和注意力机制的车辆轨迹预测方法[J]. 兵工学报, 2023, 44(7): 2162-2170.

LIAN Jing, DING Rongqi, LI Linhui, WANG Xuecheng, ZHOU Yafu. Vehicle Trajectory Prediction Method Based on Graph Models and Attention Mechanism[J]. Acta Armamentarii, 2023, 44(7): 2162-2170.

图/表 9

图1 预测模型总体结构

Fig.1 General structure of prediction model

图2 车道图及车交互图构建示意图

Fig.2 Schematic diagram of lane graph and vehicle interaction graph

图3 车辆交互特征融合结构示意图

Fig.3 Schematic diagram of fusion structure of vehicle interaction features

图4 车辆对车道注意力部分(V2L)结构示意图

Fig.4 Schematic diagram of vehicle to lane attention (V2L) structure

表1 预测结果对比

Table 1 Comparison of prediction results

方法	单模态(k=1)			多模态(k=6)
方法	min ADE	min FDE	MR	min ADE	min FDE	MR
恒定速度基准方法^[22]	2.96	6.81	0.81	2.34	5.44	0.69
加权最邻近回归基准方法^[22]	3.45	7.88	0.87	1.71	3.28	0.54
Jean^[13]	1.74	4.24	0.69	1.00	1.42	0.13
TNT^[23]	1.77	3.91	0.59	0.94	1.54	0.13
WIMP^[24]	1.82	4.03	0.63	0.90	1.42	0.17
GA-MTP(本文方法)	1.74	3.89	0.60	0.86	1.34	0.15

图5 预测结果可视化

Fig.5 Visualization of prediction results

表2 消融实验预测结果

Table 2 Predicted results of ablation experiment

模型组件				评价指标
车交互图	车道图	V2L	L2V	min ADE₁	min FDE₁	min ADE₆	min FDE₆
√				2.14	5.23	1.01	2.06
√	√	√		1.82	4.84	0.89	1.69
√	√		√	1.63	4.27	0.82	1.50
√	√	√	√	1.55	3.57	0.79	1.32

图6 多模态预测轨迹数量对预测结果的影响

Fig.6 Influence of the number of multimodal prediction trajectories on prediction results

图7 预测时长对预测结果的影响

Fig.7 Influence of prediction duration on prediction results

参考文献 24

[1]	周彬. 交通场景中多任务多模概率轨迹预测方法研究[D]. 大连: 大连理工大学, 2021.
	ZHOU B. Study on multi-task and multi-model probabilistic trajectory prediction method in traffic scenes[D]. Dalian: Dalian University of Technology, 2021. (in Chinese)
[2]	贾晓港. 交互场景下的车辆轨迹预测研究[D]. 哈尔滨: 哈尔滨工业大学, 2021.
	JIA X G. Research on vehicle trajectory prediction in interactive scenarios[D]. Harbin: Harbin Institute of Technology, 2021. (in Chinese)
[3]	PHANMINH T, GRIGORE E C, BOULTON F A, et al. CoverNet: multimodal behavior prediction using trajectory sets[J]. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2020: 14062-14071.
[4]	CHAI Y N, SAPP B, BANSAL M, et al. MultiPath: multiple probabilistic anchor trajectory hypotheses for behavior prediction:arXiv: 1910.05449[R]. Ithaca,NY,US: Cornell University, 2019:1910.05449.
[5]	GU J R, SUN C, ZHAO H. DenseTNT: end-to-end trajectory prediction from dense goal sets:arXiv:2108.09640[R]. Ithaca,NY, US: Cornell University, 2021:2108.09640.
[6]	ZENG W Y, LIANG M, LIAO R J, et al. LaneRCNN: distributed representations for graph-centric motion forecasting[C]//Proceedings of the 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems Prague. Czech Republic: IEEE, 2021: 532-539.
[7]	LIANG M, YANG B, HU R, et al. Learning lane graph representations for motion forecasting:arXiv:2007.13732[R]. Ithaca,NY, US: Cornell University, 2020: 2007.13732.
[8]	HOMAYOUNFAR N, MA W C, LIANG J, et al. DAGMapper: learning to map by discovering lane topology[C]// Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision.Seoul,Korea:IEEE, 2019: 2911-2920.
[9]	SALZMANN T, IVANOVIC B, CHAKRAVARTY P, et al. Trajectron++:dynamically-feasible trajectory forecasting with heterogeneous data:arXiv:2001. 03093[R]. Ithaca,NY, US: Cornell University, 2020: 2001.03093.
[10]	GILLES T, SABATINI S, TSISHKOU D, et al. HOME: heatmap output for future motion estimation[C]//Proceedings of the 2021 IEEE International Intelligent Transportation Systems Conference.Indianapolis, IN, US:IEEE, 2021: 500-507.
[11]	张宏鹏, 黄长强, 唐上钦, 等. 基于卷积神经网络的无人作战飞机飞行轨迹实时预测[J]. 兵工学报, 2020, 41(9): 1894-1903. doi: 10.3969/j.issn.1000-1093.2020.09.022
	ZHANG H P, HUANG C Q, TANG S Q, et al. CNN-basedreal-time prediction method of flight trajectory of unmanned combat aerial vehicle[J]. Acta Armamentarii, 2020, 41(9): 1894-1903. (in Chinese)
[12]	VASWANI A, SHAZEER N M, PARMAR N, et al. Attention is all you need: arXiv: 1706. 03762[R]. Ithaca,NY, US: Cornell University, 2017: 1706.03762.
[13]	MERCAT J, GILLES T, ZOGHBY N E, et al. Multi-head attention for multi-modal joint vehicle motion forecasting[C]// Proceedings of the 2020 IEEE International Conference on Robotics and Automation.Paris, France:IEEE, 2020: 9638-9644.
[14]	曾伟良, 陈漪皓, 姚若愚, 等. 时空图注意力网络在交叉口车辆轨迹预测的应用[J]. 计算机科学, 2021, 48(增刊1):334-341.
	ZENG W L, CHEN Y H, YAO R Y, et al. Application of spatial-temporal graph attention networks in trajectory prediction for vehicles at intersections[J]. Computer Science, 2021, 48(S1):334-341. (in Chinese)
[15]	IVANOVIC B, PAVONE M. The trajectron: probabilistic multi-agent trajectory modeling with dynamic spatiotemporal graphs[C]// Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision.Seoul, Korea:IEEE, 2019: 2375-2384.
[16]	WANG X, LÜ R R, ZHAO Y, et al. Multi-scale context aggregation network with attention-guided for crowd counting[C]//Proceedings of the 2020 15th IEEE International Conference on Signal Processing. Beijing, China: IEEE, 2020: 240-245.
[17]	JAIN A, ZAMIR A R, SAVARESE S, et al. Structural-RNN: deep learning on spatio-temporal graphs[C]// Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, NV, US:IEEE, 2016: 5308-5317.
[18]	IVANOVIC B, SCHMERLING E, LEUNG K, et al. Generative modeling of multimodal multi-human behavior[C]// Proceedings of the 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems. Madrid, Spain:IEEE, 2018: 3088-3095.
[19]	VEMULA A, MUELLING K, OH J. Social attention: modeling attention in human crowds[C]// Proceedings of the 2018 IEEE International Conference on Robotics and Automation. Brisbane, QLD, Australia:IEEE, 2018:4601-4607.
[20]	PARK S, LEE G, BHAT M, et al. Diverse and admissible trajectory forecasting through multimodal context understanding: arXiv:2003. 03212[R]. Ithaca,NY, US: Cornell University, 2020:2003.03212.
[21]	刘创, 梁军. 基于注意力机制的车辆运动轨迹预测[J]. 浙江大学学报(工学版), 2020, 54(6):1156-1163.
	LIU C, LIANG J. Vehicle motion trajectory prediction based on attention mechanism[J]. Journal of Zhejiang University (Engineering Science), 2020, 54(6):1156-1163. (in Chinese)
[22]	CHANG M F, LAMBERT J, SANGKLOY P, et al. Argoverse: 3d tracking and forecasting with rich maps[C]// Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Long Beach, CA, US:IEEE, 2019: 8740-8749.
[23]	ZHAO H, GAO J, LAN T, et al. TNT:target-driven trajectory prediction: arXiv:2008. 08294[R]. Ithaca,NY, US: Cornell University, 2020: 2008.08294.
[24]	KHANDELWAL S, QI W, SINGH J, et al. What-if motion prediction for autonomous driving:arXiv:2008.10587[R]. Ithaca,NY, US: Cornell University, 2020: 2008.10587.