单车智能条件下综合车辆运动状态的轨迹预测方法

doi:10.13229/j.cnki.jdxbgxb.20250495

摘要/Abstract

摘要：

为实现自动驾驶汽车在单车智能条件下对周车未来轨迹进行准确预测的功能，提出了综合车辆感知检测框与运动状态的轨迹预测方法。通过融合车载摄像头图像、激光雷达以及自车运动信息，广泛提取目标车运动特征及自车运动趋势。首先，对目标车轨迹和位移进行编码并建立视觉特征融合的跨模态注意力机制。其次，设计自车位移差-角度差编码器，结合光流特征准确捕捉自车的运动趋势，并采用Transformer架构实现目标车运动特征与自身运动特征的深度融合。最后，将融合后的特征与目标车的光流特征输入解码器，生成目标车的预测轨迹。本文提出的轨迹预测方法在NuScenes数据集上进行了训练与测试，并且部署于东风AX7改装智能车辆进行实车实验，结果表明本文方法在预测精度和部署能力上具备优势。

关键词: 车辆轨迹预测, 激光雷达, 视觉特征, 光流信息, 多模态融合

Abstract:

To achieve accurate prediction of surrounding vehicles' future trajectories for autonomous vehicles under single-vehicle intelligence， a trajectory prediction method integrating vehicle perception bounding boxes and motion states is proposed. By fusing data from onboard camera images， LiDAR， and ego-vehicle motion information， the proposed method extensively extracts the motion features of target vehicles and the motion trends of the ego vehicle. First， the target vehicle's trajectory and displacement are encoded， and a cross-modal attention mechanism with visual feature fusion is established. Second， an ego-vehicle displacement-difference/angle-difference encoder is designed； combined with optical flow features， it accurately captures the ego vehicle's motion trend， and a Transformer architecture is employed to achieve deep fusion of the target vehicle's motion features with the ego vehicle's motion features. Finally， the fused features， together with the target vehicle's optical flow features， are fed into a decoder to generate the predicted trajectory of the target vehicle. The proposed trajectory prediction model was trained and tested on the NuScenes dataset and deployed on a modified Dongfeng AX7 intelligent vehicle for real-vehicle experiments. The results demonstrate that the proposed method has advantages in both prediction accuracy and deployment capability.

Key words: vehicle trajectory prediction, LiDAR, visual features, optical flow information, multimodal fusion

中图分类号:

U495

郭洪艳,谢鸿楠,孟庆瑜. 单车智能条件下综合车辆运动状态的轨迹预测方法[J]. 吉林大学学报(工学版), 2026, 56(10): 2535-2545.

Hong-yan GUO,Hong-nan XIE,Qing-yu MENG. Trajectory prediction method of integrated vehicle motion states under single-vehicle intelligence conditions[J]. Journal of Jilin University(Engineering and Technology Edition), 2026, 56(10): 2535-2545.

图/表 11

图1

表1

表2

表3

图2

图3

表4

图4

图5

表5

表6

参考文献 24

[1]	黄玲, 崔躜, 游峰, 等. 适用于多车交互场景的车辆轨迹预测模型[J]. 吉林大学学报: 工学版, 2024, 54(5): 1188-1195.
	Huang Ling, Cui Zan, You Feng, et al. Vehicle trajectory prediction model for multi-vehicle interaction scenarios[J]. Journal of Jilin University(Engineering and Technology Edition), 2024, 54(5): 1188-1195.
[2]	Xing Y, Lü C, Cao D. Personalized vehicle trajectory prediction based on joint time-series modeling for connected vehicles[J]. IEEE Transactions on Vehicular Technology, 2019, 69(2): 1341-1352.
[3]	张新锋, 赵娟, 刘国华, 等. 基于图时空注意力的多车交互轨迹预测模型[J]. 汽车技术, 2025(3): 30-38.
	Zhang Xin-feng, Zhao Juan, Liu Guo-hua, et al. Multi-vehicle interactive trajectory prediction model based on graph spatio-temporal attention[J]. Automobile Technology, 2025(3): 30-38.
[4]	Malla S, Dwivedi I, Dariush B, et al. Nemo: Future object localization using noisy ego priors[C]∥IEEE 25th International Conference on Intelligent Transportation Systems(ITSC), Macau, China, 2022: 3704-3711.
[5]	Dasgupta K, Das A, Das S, et al. Spatio-contextual deep network-based multimodal pedestrian detection for autonomous driving[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(9): 15940-15950.
[6]	Monjurul Karim M, Qin R, Wang Y. Fusion-GRU: a deep learning model for future bounding box prediction of traffic agents in risky driving videos[J]. Transportation Research Record, 2024, 2678(9): 699-709.
[7]	Gu J, Hu C, Zhang T, et al. Vip3D: end-to-end visual trajectory prediction via 3D agent queries[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Vancouver, Canada, 2023: 5496-5506.
[8]	周亦威, 劳逸男, 赵靖, 等. 考虑多类型交通参与者的图遍历多模态车辆轨迹预测[J]. 公路交通科技, 2025, 42(3): 1-10.
	Zhou Yi-wei, Lao Yi-nan, Zhao Jing, et al. Graph-traversal based multimodal vehicle trajectory prediction considering multiple types of traffic participants[J]. Journal of Highway and Transportation Research and Development, 2025, 42(3): 1-10.
[9]	Hu Y, Yang J, Chen L, et al. Planning-oriented autonomous driving[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Vancouver, Canada,2023: 17853-17862.
[10]	Liang Ming, Yang Bin, Zeng Wen-yuan, et al. Pnpnet: end-to-end perception and prediction with tracking in the loop[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Online,2020:11553-11562.
[11]	Cheng H, Liu M, Chen L, et al. Gatraj: a graph-and attention-based multi-agent trajectory prediction model[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2023, 205: 163-175.
[12]	Ren Y, Lan Z, Liu L, et al. EMSIN: enhanced multistream interaction network for vehicle trajectory prediction[J]. IEEE Transactions on Fuzzy Systems, 2024, 33(1): 54-68.
[13]	袁静, 夏英. 基于时空图注意力卷积神经网络车辆轨迹预测[J]. 计算机科学, 2024, 51(12): 157-165.
	Yuan Jing, Xia Ying. Vehicle trajectory prediction based on spatio-temporal graph attention convolutional neural network[J]. Computer Science, 2024, 51(12): 157-165.
[14]	Schmidt J, Jordan J, Gritschneder F, et al. Crat-pred: Vehicle trajectory prediction with crystal graph convolutional neural networks and multi-head self-attention[C]∥International Conference on Robotics and Automation(ICRA), Philadelphia, USA, 2022: 7799-7805.
[15]	陈文强, 王东丹, 朱文英, 等. 基于时空图注意力网络的车辆多模态轨迹预测模型[J]. 浙江大学学报:工学版, 2025, 59(3): 443-450.
	Chen Wen-qiang, Wang Dong-dan, Zhu Wen-ying, et al. Multimodal vehicle trajectory prediction model based on spatio-temporal graph attention network[J]. Journal of Zhejiang University(Engineering Science), 2025, 59(3): 443-450.
[16]	Sun Z, Wang Z, Halilaj L, et al. Semanticformer: holistic and semantic traffic scene representation for trajectory prediction using knowledge graphs[J]. IEEE Robotics and Automation Letters, 2024,9(9): 7381-7388.
[17]	Li L, Wang X, Yang D, et al. Real-time heterogeneous road-agents trajectory prediction using hierarchical convolutional networks and multi-task learning[J]. IEEE Transactions on Intelligent Vehicles, 2023, 9(2): 4055-4069.
[18]	周亦威, 夏莫, 朱冰. 城市道路场景下考虑多类交通参与者的多模态车辆轨迹预测方法研究[J]. 汽车工程, 2024, 46(3): 396-406.
	Zhou Yi-wei, Xia Mo, Zhu Bing. Research on multimodal vehicle trajectory prediction method considering multiple types of traffic participants in urban road scenarios[J]. Automotive Engineering, 2024, 46(3): 396-406.
[19]	Teed Z, Deng J. Raft: Recurrent all-pairs field transforms for optical flow[C]∥The 16th European Conference on Computer Vision(ECCV), Glasgow, UK, 2020: 402-419.
[20]	Caesar H, Bankiti V, Lang A H, et al. nuscenes: A multimodal dataset for autonomous driving[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Online, 2020: 11621-11631.
[21]	Mohamed A, Qian K, Elhoseiny M, et al. Social-stgcnn: a social spatio-temporal graph convolutional neural network for human trajectory prediction[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Online, 2020: 14424-14432.
[22]	Lu Y, Wang W, Hu X, et al. Vehicle trajectory prediction in connected environments via heterogeneous context-aware graph convolutional networks[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 24(8): 8452-8464.
[23]	Yuan Y, Weng X, Ou Y, et al. Agentformer: Agent-aware transformers for socio-temporal multi-agent forecasting[C]∥Proceedings of the IEEE/CVF International Conference on Computer Vision, Montreal, Canada, 2021: 9813-9823.
[24]	Wang J, Guo J, Feng M, et al. Trajectory grid diffusion for multimodal trajectory prediction in autonomous vehicles[J]. IEEE Transactions on Intelligent Vehicles, 2024: 3495037.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

模型	minADE₅	minADE₁₀		minFDE₁₀
本文方法	2.02	1.75	3.86	3.12
DSF?AF	2.06	1.66	4.67	3.71
DLow?AF	2.11	1.78	4.70	3.58
MID	2.15	1.85	4.70	3.90

硬件名称	参数介绍
车载计算单元	型号为Nuvo?6108GC；CPU为Intel Xeon E3?1275 v5，4 核 8 线程，主频 3.6 GHz；GPU为NVIDIA GeForce RTX 2080，8 GB GDDR6；内存为32 GB；操作系统为Ubuntu 18.04 LTS
摄像头	5 000 万像素；Sony IMX800 / 光影猎人900，f/1.6，等效 23 mm，OIS，8K@24 fps，4K@60/30 fps；支持 OIS + EIS
激光雷达	32 线机械旋转式；激光波长为905 nm；激光等级为Class 1；测距精度为±3 cm；测距范围为0.4 m~200 m，目标反射率为20 %；视场角为360°× 40°；输出三维空间坐标、反射强度

工况	平均位移误差（ADE）/m
工况	0.5 s	1.0 s	1.5 s	2.0 s	2.5 s	3.0 s
直行	0.29	0.40	0.67	0.82	0.84	0.80
转向	0.64	0.89	1.07	1.20	1.27	1.28

工况	最终位移误差（FDE）/m
工况	0.5 s	1.0 s	1.5 s	2.0 s	2.5 s	3.0 s
直行	0.29	0.51	1.21	1.28	0.94	0.56
转向	0.64	1.14	1.42	1.58	1.57	1.31