Detection Method for 3D Lanes Based on Graph Relationship Optimization Integrating Point and Lane Features

doi:10.16182/j.issn1004731x.joss.25-0591

Abstract

Abstract:

Under complex road conditions, the thin and elongated structure and small proportion of lanes lead to blurred visual features and insufficient positioning accuracy, which in turn threatens the road safety of autonomous driving. To address these issues, a 3D lane detection method or graph-based point and lane optimization network (GPLNet), based on graph relationship optimization integrating point and lane features, was proposed. Preliminary feature extraction was completed by the backbone network. 3D spatial positional coding with geometric constraints was obtained through a joint query embedding generation module. A graph relationship optimization network was utilized to perform graph relationship calculation and optimization modeling on point- and lane-level features of lanes to enhance the context awareness capability of lanes. Lane prediction and loss function calculation were achieved with a 3D prediction head. Experimental results indicate that the overall performance of the proposed method is superior to existing mainstream 3D lane detection algorithms, and the detection accuracy is higher in low-visibility lane scenarios, which verifies the effectiveness of the method.

Key words: autonomous driving, 3D lane detection, positional coding, query embedding, graph relationship optimization

CLC Number:

TP391.9

Jiang Yanji, Xiao Xingyi, Dong Hao, Yu Miao, Huang Jinshan, Liu Daqian, Fei Bowen. Detection Method for 3D Lanes Based on Graph Relationship Optimization Integrating Point and Lane Features[J]. Journal of System Simulation, 2026, 38(5): 1303-1319.

Figures/Tables 17

Fig. 1

Fig. 2

Fig. 3

Fig. 4

Fig. 5

Fig. 6

Fig. 7

Table 1

Table 2

Fig. 8

Table 3

Table 4

Table 5

Table 6

Ablation study of each module

联合查询嵌入	图关系优化	F1值/%	C.ACC/%	$E x$ /(C/m)	$E x$ /(F/m)	$E z$ /(C/m)	$E z$ /(F/m)
×	×	67.19	89.17	0.276	0.343	0.099	0.132
√	×	69.34	91.49	0.260	0.319	0.098	0.128
×	√	68.60	90.60	0.259	0.320	0.098	0.131
√	√	71.30	93.03	0.247	0.307	0.097	0.125

Table 6

Table 7

Ablation study of graph relationship optimization

结构关系	类别关系	F1值/%	C.ACC/%	$E x$ /(C/m)	$E x$ /(F/m)	$E z$ /(C/m)	$E z$ /(F/m)
×	×	69.34	91.49	0.260	0.319	0.098	0.128
√	×	71.13	92.94	0.248	0.311	0.097	0.126
×	√	69.45	91.97	0.255	0.324	0.097	0.130
√	√	71.30	93.03	0.247	0.307	0.097	0.125

Table 7

Fig. 9

Table 8

Ablation experiment of distance threshold ϵ

$ϵ$	F1值/ %	C.ACC/ %	$E x$ /(C/m)	$E x$ / (F/m)	$E z$ / (C/m)	$E z$ / (F/m)
0	69.36	91.54	0.256	0.322	0.128	0.154
10	70.77	91.96	0.251	0.316	0.098	0.130
15	70.93	92.87	0.252	0.311	0.097	0.125
20	70.96	92.72	0.245	0.293	0.095	0.119
25	71.15	92.93	0.249	0.310	0.097	0.126
30	70.93	92.63	0.234	0.315	0.095	0.129
35	70.24	92.25	0.245	0.319	0.096	0.127

Table 8

References 33

[1]	侯志明. 复杂场景下的车道线检测算法研究[D]. 北京: 北京交通大学, 2024.
	Hou Zhiming. Research on Lane Line Detection Algorithm in Complex Scenes[D]. Beijing: Beijing Jiaotong University, 2024.
[2]	崔辰. 5G通信技术下的汽车无人驾驶技术之研究[J]. 时代汽车, 2022(3): 16-18.
	Cui Chen. Research on Vehicle Driverless Technology Under 5G Communication Technology[J]. Auto Time, 2022(3): 16-18.
[3]	罗为明, 应朝阳, 孙巍, 等. 辅助驾驶汽车安全技术检验需求研究[J]. 中国标准化, 2023(21): 219-223.
	Luo Weiming, Ying Zhaoyang, Sun Wei, et al. Research on Safety Technological Inspection Requirements of Vehicles with Driving Assist[J]. China Standardization, 2023(21): 219-223.
[4]	Krizhevsky A, Sutskever I, Hinton G E. ImageNet Classification with Deep Convolutional Neural Networks[J]. Communications of the ACM, 2017, 60(6): 84-90.
[5]	谢春丽, 梁梓涵. 复杂环境下基于实例分割的车道线检测[J]. 重庆交通大学学报(自然科学版), 2025, 44(4): 79-86, 120.
	Xie Chunli, Liang Zihan. Lane Detection Based on Instance Segmentation in Complex Environment[J]. Journal of Chongqing Jiaotong University(Natural Science), 2025, 44(4): 79-86, 120.
[6]	Zheng Tu, Huang Yifei, Liu Yang, et al. CLRNet: Cross Layer Refinement Network for Lane Detection[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2022: 888-897.
[7]	Wang Jinsheng, Ma Yinchao, Huang Shaofei, et al. A Keypoint-based Global Association Network for Lane Detection[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2022: 1382-1391.
[8]	Feng Zhengyang, Guo Shaohua, Tan Xin, et al. Rethinking Efficient Lane Detection via Curve Modeling[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2022: 17041-17049.
[9]	葛泽坤, 陶发展, 付主木, 等. 改进多头注意力机制的车道检测方法[J]. 计算机工程与应用, 2024, 60(2): 264-271.
	Ge Zekun, Tao Fazhan, Fu Zhumu, et al. Lane Detection Method Based on Improved Multi-head Self-attention[J]. Computer Engineering and Applications, 2024, 60(2): 264-271.
[10]	邹斌, 刘晶. 基于3DLaneTR的三维车道线检测算法[J]. 武汉理工大学学报, 2024, 46(7): 146-151.
	Zou Bin, Liu Jing. 3D Lane Detection Algorithm Based on 3DLaneTR[J]. Journal of Wuhan University of Technology, 2024, 46(7): 146-151.
[11]	Guo Yuliang, Chen Guang, Zhao Peitao, et al. Gen-LaneNet: A Generalized and Scalable Approach for 3D Lane Detection[C]//Computer Vision-ECCV 2020. Cham: Springer International Publishing, 2020: 666-681.
[12]	Garnett Noa, Cohen Rafi, Pe'er Tomer, et al. 3D-LaneNet: End-to-end 3D Multiple Lane Detection[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2019: 2921-2930.
[13]	Chen Li, Sima Chonghao, Li Yang, et al. PersFormer: 3D Lane Detection via Perspective Transformer and the OpenLane Benchmark[C]//Computer Vision-ECCV 2022. Cham: Springer Nature Switzerland, 2022: 550-567.
[14]	Huang Shaofei, Shen Zhenwei, Huang Zehao, et al. Anchor3DLane: Learning to Regress 3D Anchors for Monocular 3D Lane Detection[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2023: 17451-17460.
[15]	Luo Yueru, Zheng Chaoda, Yan Xu, et al. LATR: 3D Lane Detection from Monocular Images with Transformer[C]//2023 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2023: 7907-7918.
[16]	王世乾. 基于深度学习的车道线检测算法研究[D]. 杭州: 浙江科技大学, 2024.
	Wang Shiqian. Research on Lane Detection Algorithm Based on Deep Learning[D]. Hangzhou: Zhejiang University of Science and Technology, 2024.
[17]	胡杰, 陈楠, 徐文才, 等. 基于自适应门控的双路激光雷达三维车道线检测[J]. 中国激光, 2023, 50(22): 132-146.
	Hu Jie, Chen Nan, Xu Wencai, et al. Three-dimensional Lane Detection Algorithm of Lidar Based on Adaptive Gating and Dual Pathways[J]. Chinese Journal of Lasers, 2023, 50(22): 132-146.
[18]	Xu Keyulu, Hu Weihua, Leskovec J, et al. How Powerful are Graph Neural Networks[EB/OL]. (2019-02-22) [2025-04-25]. .
[19]	Shi Weijing, Rajkumar R. Point-GNN: Graph Neural Network for 3D Object Detection in a Point Cloud[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2020: 1708-1716.
[20]	钟凯, 陈颖, 颜承志, 等. 基于图神经网络与动态采样的三维目标检测算法[J]. 激光与光电子学进展, 2025, 62(14): 191-202.
	Zhong Kai, Chen Ying, Yan Chengzhi, et al. 3D Object Detection Algorithm Based on Graph Neural Network and Dynamic Sampling[J]. Laser & Optoelectronics Progress, 2025, 62(14): 191-202.
[21]	Büchner Martin, Zürn Jannik, Todoran Ion-George, et al. Learning and Aggregating Lane Graphs for Urban Automated Driving[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2023: 13415-13424.
[22]	刘凯, 汪佳琴, 李汉涛. 基于深度学习的车辆轨迹预测研究综述[J]. 郑州大学学报(工学版), 2025, 46(5): 77-89.
	Liu Kai, Wang Jiaqin, Li Hantao. A Review of Vehicle Trajectory Prediction Based on Deep Learning[J]. Journal of Zhengzhou University(Engineering Science), 2025, 46(5): 77-89.
[23]	Wang Yue, Solomon J. Object DGCNN: 3D Object Detection Using Dynamic Graphs[C]//Proceedings of the 35th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2021: 20745-20758.
[24]	Yang Chenhongi, Lin Tianwei, Huang Lichao, et al. WidthFormer: Toward Efficient Transformer-based BEV View Transformation[C]//2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Piscataway: IEEE, 2024: 8457-8464.
[25]	Brody S, Alon U, Yahav E, et al. How Attentive are Graph Attention Networks[EB/OL]. (2022-01-31) [2025-04-25]. .
[26]	Yan Fan, Nie Ming, Cai Xinyue, et al. ONCE-3DLanes: Building Monocular 3D Lane Detection[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2022: 17122-17131.
[27]	Pan Xingang, Shi Jianping, Luo Ping, et al. Spatial as Deep: Spatial CNN for Traffic Scene Understanding[C]//Proceeding of the Thirty-Second AAAI Conference on Artificial Intelligence and the Thirtieth Innovative Applications of Artificial Intelligence Conference and the Eighth Symposium on Educational Advances in Artificial Intelligence. Palo Alto: AAAI Press, 2018: 7276-7283.
[28]	Wang Ze, Ren Weiqiang, Qiu Qiang, et al. LaneNet: Real-time Lane Detection Network[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. Piscataway: IEEE, 2018: 1-5.
[29]	Efrat N, Bluvstein M, Oron S, et al. 3D-LaneNet+: Anchor Free Lane Detection Using a Semi-local Representation[EB/OL].(2020-11-04) [2025-04-25]. .
[30]	Bai Yifeng, Chen Zhirong, Liang Pengpeng, et al. CurveFormer++: 3D Lane Detection by Curve Propagation with Temporal Curve Queries and Attention[EB/OL]. (2025- 03-16) [2025-04-25]. .
[31]	Pittner Maximilian, Janai Joel, Condurache Alexandru P. LaneCPP: Continuous 3D Lane Detection Using Physical Priors[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2024: 10639-10648.
[32]	Han Wencheng, Shen Jianbing. Decoupling the Curve Modeling and Pavement Regression for Lane Detection[EB/OL]. (2023-09-19) [2025-04-25]. .
[33]	Bai Yifeng, Chen Zhirong, Fu Zhangjie, et al. CurveFormer: 3D Lane Detection by Curve Propagation with Curve Queries and Attention[C]//2023 IEEE International Conference on Robotics and Automation (ICRA). Piscataway: IEEE, 2023: 7062-7068.

方法	F1值/%	C.ACC/%	E_x /(C/m)	E_x /(F/m)	E_z /(C/m)	E_z /(F/m)
3D-LaneNet^[29]	44.1	‒	0.479	0.572	0.367	0.443
Performer^[13]	50.5	92.3	0.485	0.553	0.364	0.431
CurveFormer++^[30]	52.7	88.1	0.337	0.801	0.198	0.676
Anchor3DLane^[14]	53.1	90.0	0.300	0.311	0.103	0.139
LaneCPP^[31]	60.3	‒	0.264	0.310	0.077	0.117
LATR^[15]	61.9	91.9	0.219	0.259	0.073	0.104
GPLNet	63.1	92.4	0.195	0.230	0.070	0.100

方法	全部	上下坡	弯道	极端天气	夜间	交叉路口	合并与分流
3D-LaneNet	44.1	40.8	46.5	47.5	41.5	32.1	41.7
Performer	50.5	42.4	55.6	48.6	46.6	40.0	50.7
CurveFormer++	52.7	48.3	59.4	50.6	48.4	45.0	48.1
Anchor3DLane	53.1	45.5	56.2	51.9	47.2	44.2	50.5
LaneCPP	60.3	53.6	64.4	56.7	54.9	52.0	58.7
LATR	61.9	55.2	68.2	57.1	55.4	52.3	61.5
GPLNet	63.1	53.3	68.5	58.5	55.7	53.8	62.0

方法	参数量×10⁶	帧率/(帧/s)	F1值/%
Performer	63.11	11.67	50.5
LATR	44.47	13.34	61.9
GPLNet	43.55	13.75	63.1

方法	F1值/ %	精确率/%	召回率/%	距离误差/m
3D-LaneNet	44.73	35.16	61.46	0.127
Performer	74.33	80.30	69.18	0.074
Anchor3DLane	74.44	80.50	69.23	0.064
DecoupleLane^[32]	75.07	81.19	69.26	0.062
CurveFormer++	77.85	82.22	72.79	0.081
LATR	80.59	86.12	75.73	0.052
GPLNet	80.62	83.92	77.57	0.040

场景	评估指标	3D-LaneNet	CLGO	PersFormer	CurveFormer^[33]	Anchor3DLane	LATR	GPLNet
平衡场景	F1值/%	86.4	91.9	92.9	95.8	95.6	96.8	96.9
	E_x /(C/m)	0.068	0.061	0.054	0.078	0.052	0.022	0.019
	E_x /(F/m)	0.477	0.361	0.356	0.326	0.306	0.253	0.243
	E_z /(C/m)	0.015	0.029	0.01	0.018	0.015	0.007	0.005
	E_z /(F/m)	0.202	0.25	0.234	0.219	0.223	0.202	0.199
稀有场景	F1值/%	72.0	86.1	87.5	95.6	94.4	96.1	96.3
	E_x /(C/m)	0.166	0.147	0.107	0.182	0.094	0.050	0.040
	E_x /(F/m)	0.855	0.735	0.782	0.737	0.693	0.600	0.580
	E_z /(C/m)	0.039	0.071	0.024	0.039	0.027	0.015	0.011
	E_z /(F/m)	0.521	0.609	0.602	0.561	0.579	0.532	0.527
视觉变体	F1值/%	72.5	87.3	89.6	90.8	91.4	95.1	95.2
	E_x /(C/m)	0.115	0.084	0.074	0.125	0.068	0.045	0.036
	E_x /(F/m)	0.601	0.464	0.430	0.410	0.367	0.315	0.302
	E_z /(C/m)	0.032	0.045	0.015	0.028	0.020	0.016	0.015
	E_z /(F/m)	0.23	0.312	0.266	0.254	0.232	0.228	0.220