基于驾驶人认知决策空间的换道意图预测

doi:10.19562/j.chinasae.qcgc.2025.08.004

摘要/Abstract

摘要：

智能车辆人机协作的关键是以人为核心，换道作为最基本的驾驶任务之一，准确高效预测驾驶人换道意图对人机协作拟人化发展至关重要。本文基于驾驶人认知决策空间的理论，设计了驾驶人换道意图预测试验，分析了车辆操纵数据、驾驶人视觉特性与驾驶场景之间的关系，生成了驾驶人注视区与驾驶场景拓扑关系图，构建了不同时间窗口的驾驶人换道意图预测模型数据集，基于ConvNeXt（convolutional network）模型的逆残差深度可分离卷积，结合注意力机制ECA（efficient channel attention）、ConvLSTM（convolutional long short term memory）网络以及GCN（graph convolutional networks）图神经网络等结构，构建了基于注意力机制的驾驶人换道意图预测模型。结果表明，数据集时间宽度为3 s时模型的预测准确率表现最佳，为91.15%，通过对比试验、消融试验充分验证了所提出的基于注意力机制的驾驶人换道意图预测模型的优越性能。

关键词: 智能车辆, 换道意图, 认知决策空间, 视觉特性, 图像检测, 深度学习

Abstract:

The key to human-machine collaboration in intelligent vehicles is to focus on people. Lane changing is one of the most basic driving tasks. Accurately and efficiently predicting the driver's lane changing intention is crucial to the development of humanized human-machine collaboration. Based on the theory of driver cognitive-making space， in this paper the driver's lane changing intention prediction experiment is designed. The relationship between vehicle operation data， driver's visual characteristics and driving scenes is analyzed， and a topological relationship diagram between the driver's gaze area and the driving scene is generated. Then the driver's lane changing intention prediction model dataset with different time windows is constructed. Based on the inverse residual deep separable convolution of the ConvNeXt （convolutional network） model， combined with the attention mechanism ECA （efficient channel attention）， ConvLSTM （convolutional long short term the memory） network and GCN （graph convolutional networks） figure structure， a predictive model of driver intention lane changing based on attention mechanism is built. The results show that when the time width of the data set is 3 s， the prediction accuracy of the model is the best， which is 91.15%， and the superior performance of the proposed driver lane change intention prediction model based on attention mechanism is fully verified by comparison and ablation experiments.

Key words: intelligent vehicle, lane change intention, cognitive-making space, visual properties, image detection, deep learning

孙秦豫,周航,付锐,王畅,黄涛,杨骏锋,王芸豪. 基于驾驶人认知决策空间的换道意图预测[J]. 汽车工程, 2025, 47(8): 1468-1478.

Qinyu Sun,Hang Zhou,Rui Fu,Chang Wang,Tao Huang,Junfeng Yang,Yunhao Wang. Prediction of Lane Change Intention Based on Driver's Cognitive-Making Space[J]. Automotive Engineering, 2025, 47(8): 1468-1478.

图/表 28

图1

图2

图3

图4

图5

图6

表1

图7

图8

图9

图10

图11

图12

图13

图14

图15

图16

图17

图18

图19

图20

图21

表2

表3

表4

表5

表6

表7

参考文献 26

[1]	LI L， LI P. Analysis of driver's steering behavior for lane change prediction［C］.2019 11th International Conference on Intelligent Human-Machine Systems and Cybernetics （IHMSC）. IEEE， 2019： 71-75.
[2]	SCHMIDT K， BEGGIATO M， HOFFMANN K H， et al. A mathematical model for predicting lane changes using the steering wheel angle［J］. Journal of Safety Research， 2014， 49： 85-90.
[3]	DIEHL F， BRUNNER T， LE M T， et al. Graph neural networks for modelling traffic participant interaction［C］. 2019 IEEE Intelligent Vehicles Symposium （IV）. IEEE， 2019： 695-701.
[4]	LIU S， TAN D， HONG S， et al. Study on the prediction of lane change intention of intelligent vehicles in the network environment［J］. World Electric Vehicle Journal， 2021， 12（1）： 27.
[5]	宋晓琳，曾艳兵，曹昊天，等. 基于长短期记忆网络的换道意图识别方法［J］. 中国公路学报， 2021， 34（11）： 236-245.
	SONG X L， ZENG Y B， CAO H T， et al. Lane-changing intention recognition method based on long short-term memory network［J］. China Journal Of Highway And Transport， 2021， 34（11）： 236-245.
[6]	BHATTACHARYA S， BERNADIN S. Eye-glance frequency as a function of driver's intent to change lanes［C］. 2018 IEEE 88th Vehicular Technology Conference （VTC-Fall）. IEEE， 2018： 1-5.
[7]	GUO Y， ZHANG H， WANG C， et al. Driver lane change intention recognition in the connected environment［J］. Physica A： Statistical Mechanics and its Applications， 2021， 575： 126057.
[8]	YI D， SU J， LIU C， et al. Trajectory clustering aided personalized driver intention prediction for intelligent vehicles［J］. IEEE Transactions on Industrial Informatics， 2018， 15（6）： 3693-3702.
[9]	LIU Z Q， PENG M C， SUN Y C. Estimation of driver lane change intention based on the LSTM and dempster-shafer evidence theory［J］. Journal of Advanced Transportation， 2021.
[10]	LIU Y， WANG X， LI L， et al. A novel lane change decision-making model of autonomous vehicle based on support vector machine［J］. IEEE Access， 2019， 7： 26543-26550.
[11]	LI X， WANG W， ROETTING M. Estimating driver’s lane-change intent considering driving style and contextual traffic［J］. IEEE Transactions on Intelligent Transportation Systems， 2018， 20（9）： 3258-3271.
[12]	LI L， ZHAO W， XU C， et al. Lane-change intention inference based on rnn for autonomous driving on highways［J］. IEEE Transactions on Vehicular Technology， 2021， 70（6）： 5499-5510.
[13]	XING Y， LV C， WANG H， et al. An ensemble deep learning approach for driver lane change intention inference［J］. Transportation Research Part C： Emerging Technologies， 2020， 115： 102615.
[14]	CHEN S， JIAN Z， HUANG Y， et al. Autonomous driving： cognitive construction and situation understanding ［J］. Science China Information Sciences， 2019， 62： 1-27.
[15]	DOLAN R J， DAYAN P. Goals and habits in the brain ［J］. Neuron， 2013， 80（2）： 312-325.
[16]	CHEN X， TREIBER M， KANAGARAJ V， et al. Social force models for pedestrian traffic-state of the art ［J］. Transport Reviews， 2018， 38（5）： 625-653.
[17]	HO M K， ABEL D， CORREA C G，et al. People construct simplified mental representations to plan［J］. Nature， 2022， 606（7912）： 129-136.
[18]	PAN Y， ZHANG Q， ZHANG Y， et al. Lane-change intention prediction using eye-tracking technology： a systematic review［J］. Applied Ergonomics， 2022， 103： 103775.
[19]	LIU H， WANG T， LI W， et al. Lane-change intention recognition considering oncoming traffic： novel insights revealed by advances in deep learning［J］. Accident Analysis & Prevention， 2024， 198： 107476.
[20]	HIMBERGER K D， CHIEN H Y， HONEY C J. Principles of temporal processing across the cortical hierarchy ［J］. Neuroscience， 2018， 389： 161-174.
[21]	LOWE D G. Distinctive image features from scale-invariant keypoints［J］. International Journal of Computer Vision， 2004， 60（2）： 91-110.
[22]	LIN T Y， GOYAL P， GIRSHICK R， et al. Focal loss for dense object detection［C］. Proceedings of the IEEE International Conference on Computer Vision， 2017： 2980-2988.
[23]	PAN Y. Heading toward artificial intelligence 2.0［J］. Engineering， 2016， 2（4）： 409-413.
[24]	SIMONYAN K， ZISSERMAN A. Very deep convolutional networks for large-scale image recognition［J］. arXiv preprint arXiv：， 2014.
[25]	VASWANI A， SHAZEER N， PARMAR N， et al. Attention is all you need［J］. Advances in Neural Information Processing Systems， 2017， 30.
[26]	DOSOVITSKIY A， BEYER L， KOLESNIKOV A， et al. An image is worth 16x16 words： transformers for image recognition at scale［J］. arXiv preprint arXiv：， 2020.

类别	左换道意图时窗/s	右换道意图时窗/s
均值	2.96	3.09
标准差	0.56	0.43

项目	版本型号
CPU	i9-10700CPU
GPU	NVIDIA RTX 3090
内存	64 GB
环境	Keras2.2.4和Tensorflow1.15

参数名称	数值
Epoch	100
Batch	32
初始学习率	0.001
权值衰减系数	0.000 5
动量系数	0.9

输入	操作	c	n	k	s	C-S
3×256×192×3	TB（Z-P-2D）			（3，3）
3×262×198×3	TB（C-A-B-M）	64		（7，7）	2
3×128×96×64	TB（M-P-2D）			（3，3）	2
3×64×48×64	DSC-ECA	［64，64，128］	1		1	是
3×64×48×128	DSC-ECA	［64，64，128］	3			否
3×64×48×128	DSC-ECA	［128，128，256］	1		2	是
3×32×24×256	DSC-ECA	［128，128，256］	2			否
3×32×24×256	DSC-ECA	［256，256，512］	1		2	是
3×16×12×512	DSC-ECA	［256，256，512］	2			否
3×16×12×512	ConvLSTM2D	512		（3，3）	2
16×12×512	GCN	512
16×12×512	C-E-B-M	512		（3，3）	2
8×6×512	GAP
512	FC	256
256	FC	64
64	FC	3

模型	训练损失	准确率/%	t/s
ResNet^［17］	0.192	79.67	0.28
AlexNet^［18］	0.126	81.42	0.30
Transformer^［7］	0.178	85.56	0.26
本模型	0.021	91.15	0.31