基于时空特征融合与候选策略的智能汽车多模态轨迹预测

doi:10.3778/j.issn.1002-8331.2404-0090

摘要/Abstract

摘要： 针对现有轨迹预测模型在捕捉复杂时空动态方面的局限性，以及部分预测轨迹不符合实际场景约束等问题，提出了一种基于时空特征融合和候选策略的智能汽车多模态轨迹预测模型。在场景编码和特征融合阶段，设计了非对称双向门控循环单元以捕获历史轨迹序列之间的双向依赖性；引入一种基于交叉注意力的混合特征注意力方法，以建模车道与交通参与者间的隐式交互，并在车道图节点中深度融合车道空间特征和轨迹的时序特征。在解码器前引入直接使用车道拓扑结构的候选策略，该策略将利用先验知识指导预测过程，并通过覆盖目标车辆可能的未来轨迹，确保解码器能够输出可靠的多模态轨迹。该模型在公开数据集nuScenes上进行验证，实验结果表明，在预测5条和10条轨迹时，minADE和MR分别较最佳对比模型提高了7.5%、11.5%和5.5%、21.4%。可视化结果展现出更强的稳健性和解释性。

关键词: 智能驾驶, 轨迹预测, 时空特征融合, 注意力机制, 多模态预测

Abstract: To address the limitations of existing trajectory prediction models in capturing complex spatiotemporal dynamics and predicting some trajectories that meet real-world constraints, this paper proposes a multimodal trajectory prediction model for intelligent vehicles based on spatiotemporal feature fusion and candidate strategies. First, an asymmetric bidirectional gated recurrent unit is designed in the scene encoding and feature fusion stage to capture bidirectional dependencies within historical trajectory sequences. Next, a hybrid feature attention method based on cross-attention is introduced to model implicit interactions between lanes and traffic participants, deeply integrating lane spatial features and trajectory temporal features in lane graph nodes. Finally, a candidate strategy directly utilizing lane topology is introduced before the decoder, guiding the prediction process with prior knowledge and covering potential future trajectories of the target vehicle, ensuring the decoder outputs reliable multimodal trajectories. The model is validated on the public nuScenes dataset. Experimental results show that for predicting 5 and 10 trajectories, the minADE and MR improved by 7.5%, 11.5%, and 5.5%, 21.4% respectively, compared to the best benchmark models. Visualization results demonstrate the model’s enhanced robustness and interpretability.

Key words: intelligent driving, trajectory prediction, spatio-temporal feature fusion, attention mechanisms, multimodal prediction

杨智勇, 杨俊, 许沁欣. 基于时空特征融合与候选策略的智能汽车多模态轨迹预测[J]. 计算机工程与应用, 2025, 61(13): 217-226.

YANG Zhiyong, YANG Jun, XU Qinxin. Multimodal Trajectory Prediction for Intelligent Vehicles Based on Spatio-Temporal Feature Fusion and Candidate Strategies[J]. Computer Engineering and Applications, 2025, 61(13): 217-226.

参考文献

[1] 中华人民共和国公安部. 全国机动车保有量达4.35亿辆驾驶人达5.23亿人新能源汽车保有量超过2000万辆[EB/OL]. [2024-01-11]. https://www.mps.gov.cn/n2254098/n4904352/c9384864/ content.html.
Ministry of Public Security, People's Republic of China. The number of motor vehicles in China reached 435 million, with 523 million drivers, and the number of new energy vehicles exceeded 20 million[EB/OL]. [2024-01-11]. https://www.mps.gov.cn/n2254098/n4904352/c9384864/content.html.
[2] XIE G T, GAO H B, HUANG B, et al. A driving behavior awareness model based on a dynamic Bayesian network and distributed genetic algorithm[J]. International Journal of Computational Intelligence Systems, 2018, 11(1): 469-482.
[3] MIN K, KIM D, PARK J, et al. RNN-based path prediction of obstacle vehicles with deep ensemble[J]. IEEE Transactions on Vehicular Technology, 2019, 68(10): 10252-10256.
[4] DEO N, TRIVEDI M M. Convolutional social pooling for vehicle trajectory prediction[C]//Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. Piscataway: IEEE, 2018: 1468-1476.
[5] 孟宪伟, 唐进君, 王喆. 考虑换道意图的LSTM-AdaBoost车辆轨迹预测模型[J]. 计算机工程与应用, 2022, 58(13): 280-287.
MENG X W, TANG J J, WANG Z. Trajectory prediction of vehicles based on LSTM-AdaBoost model considering lane-changing intention[J]. Computer Engineering and Applications, 2022, 58(13): 280-287.
[6] PHAN-MINH T, GRIGORE E C, BOULTON F A, et al. CoverNet: multimodal behavior prediction using trajectory sets[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020: 14062-14071.
[7] CHAI Y N, SAPP B, BANSAL M, et al. MultiPath: multiple probabilistic anchor trajectory hypotheses for behavior prediction[J]. arXiv:1910.05449, 2019.
[8] DEO N, TRIVEDI M M. Trajectory forecasts in unknown environments conditioned on grid-based plans[J]. arXiv:2001. 00735, 2020.
[9] GAO J Y, SUN C, ZHAO H, et al. VectorNet: encoding HD maps and agent dynamics from vectorized representation[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020: 11522-11530.
[10] LIANG M, YANG B, HU R, et al. Learning lane graph representations for motion forecasting[C]//Proceedings of the 16th European Conference on Computer Vision. Cham: Springer, 2020: 541-556.
[11] 胡启慧, 蔡英凤, 王海, 等. 基于层次图注意的异构多目标轨迹预测方法[J]. 汽车工程, 2023, 45(8): 1448-1456.
HU Q H, CAI Y F, WANG H, et al. Heterogeneous multi-object trajectory prediction method based on hierarchical graph attention[J]. Automotive Engineering, 2023, 45(8): 1448-1456.
[12] 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 Y W, XIA M, ZHU B. Multimodal vehicle trajectory prediction methods considering multiple traffic participants in urban road scenarios[J]. Automotive Engineering, 2024, 46(3): 396-406.
[13] JIA X S, WU P H, CHEN L, et al. HDGT: heterogeneous driving graph transformer for multi-agent trajectory prediction via scene encoding[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(11): 13860-13875.
[14] LIU M M, CHENG H, CHEN L, et al. LAformer: trajectory prediction for autonomous driving with lane-aware scene constraints[J]. arXiv:2302.13933, 2023.
[15] NAYAKANTI N, AL-RFOU R, ZHOU A, et al. Wayformer: motion forecasting via simple & efficient attention networks[C]//Proceedings of the 2023 IEEE International Conference on Robotics and Automation. Piscataway: IEEE, 2023: 2980-2987.
[16] ZHANG Z J, LINIGER A, SAKARIDIS C, et al. Real-time motion prediction via heterogeneous polyline transformer with relative pose encoding[C]//Advances in Neural Information Processing Systems, 2024.
[17] FENG C, ZHOU H N, LIN H D, et al. MacFormer: map-agent coupled transformer for real-time and robust trajectory prediction[J]. IEEE Robotics and Automation Letters, 2023, 8(10): 6795-6802.
[18] VARADARAJAN B, HEFNY A, SRIVASTAVA A, et al. MultiPath++: efficient information fusion and trajectory aggregation for behavior prediction[C]//Proceedings of the 2022 International Conference on Robotics and Automation. Piscataway: IEEE, 2022: 7814-7821.
[19] ZHAO H, GAO J, LAN T, et al. TNT: target-driven trajectory prediction[C]//Proceedings of the 2020 Conference on Robot Learning, 2021: 895-904.
[20] GU J R, SUN C, ZHAO H. DenseTNT: end-to-end trajectory prediction from dense goal sets[C]//Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2021: 15283-15292.
[21] GILLES T, SABATINI S, TSISHKOU D, et al. GOHOME: graph-oriented heatmap output for future motion estimation[C]//Proceedings of the 2022 International Conference on Robotics and Automation. Piscataway: IEEE, 2022: 9107-9114.
[22] GILLES T, SABATINI S, TSISHKOU D, et al. THOMAS: trajectory heatmap output with learned multi-agent sampling[J]. arXiv:2110.06607, 2021.
[23] ROCKLAGE E, KRAFT H, KARATAS A, et al. Automated scenario generation for regression testing of autonomous vehicles[C]//Proceedings of the 2017 IEEE 20th International Conference on Intelligent Transportation Systems. Piscataway: IEEE, 2017: 476-483.
[24] GRAVES A, GRAVES A. Long short-term memory[M]//Supervised sequence labelling with recurrent neural networks. Berlin, Heidelberg: Springer, 2012: 37-45.
[25] RASOULI A, KOTSERUBA I, TSOTSOS J K. Pedestrian action anticipation using contextual feature fusion in stacked RNNs[J]. arXiv:2005.06582, 2020.
[26] DEO N, WOLFF E M, BEIJBOM O. Multimodal trajectory prediction conditioned on lane-graph traversals[C]//Proceedings of the Conference on Robot Learning, 2022: 203-212.
[27] VELI?KOVI? P, CUCURULL G, CASANOVA A, et al. Graph attention networks[C]//Proceedings of the International Conference on Learning Representations, 2018: 1-12.
[28] LI Z N, YU H. Trajectory prediction for autonomous driving using a transformer network[J]. arXiv:2402.16501, 2024.
[29] SUN R, LINGRAND D, PRECIOSO F. Exploring the road graph in trajectory forecasting for autonomous driving[C]//Proceedings of the 2023 IEEE/CVF International Conference on Computer Vision Workshops. Piscataway: IEEE, 2023: 71-80.
[30] WANG C H, WANG Y C, XU M Z, et al. Stepwise goal-driven networks for trajectory prediction[J]. IEEE Robotics and Automation Letters, 2022, 7(2): 2716-2723.
[31] SALZMANN T, IVANOVIC B, CHAKRAVARTY P, et al. Trajectron++: dynamically-feasible trajectory forecasting with heterogeneous data[C]//Proceedings of the 16th European Conference on Computer Vision. Cham: Springer, 2020: 683-700.
[32] KIM B, PARK S H, LEE S, et al. LaPred: lane-aware prediction of multi-modal future trajectories of dynamic agents[C]//Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2021: 14631-14640.
[33] WANG Z B, GUO J Y, HU Z M, et al. Lane transformer: a high-efficiency trajectory prediction model[J]. IEEE Open Journal of Intelligent Transportation Systems, 2023, 4: 2-13.