Article(id=1319954959790097377, tenantId=1146029695717560320, journalId=1278651732997652489, issueId=1319954930014732947, articleNumber=null, orderNo=null, doi=10.12086/oee.2026.250359, pmid=null, cstr=32245.14.oee.2026.250359, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=null, receivedDate=1763654400000, receivedDateStr=2025-11-21, revisedDate=1773331200000, revisedDateStr=2026-03-13, acceptedDate=1773936000000, acceptedDateStr=2026-03-20, onlineDate=1792631385990, onlineDateStr=2026-10-22, pubDate=1782316800000, pubDateStr=2026-06-25, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1792631385990, onlineIssueDateStr=2026-10-22, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1792631385990, creator=13701087609, updateTime=1792631385990, updator=13701087609, issue=Issue{id=1319954930014732947, tenantId=1146029695717560320, journalId=1278651732997652489, year='2026', volume='53', issue='6', pageStart='250315', pageEnd='1', issueExtLink='null', onlineDate='null', pubDate='1782316800000', pubDateStr='2026-06-25', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=-1, specialIssue=null, createTime=1792631378892, creator='13701087609', updateTime=1792631378892, updator='13701087609', preIssue=null, nextIssue=null, articleTotal=null, ext=null, issueFiles=null, downloadFileDto=null}, startPage=250359, endPage=1, ext={EN=ArticleExt(id=1319954960016589794, articleId=1319954959790097377, tenantId=1146029695717560320, journalId=1278651732997652489, language=EN, title=HA-Lane Net: Hybrid feature encoding and attention for monocular 3D lane detection, columnId=1279511634116841602, journalTitle=Opto-Electronic Engineering, columnName=Article, runingTitle=null, highlight=null, articleAbstract=
Objective

Three-dimensional lane detection provides essential geometric information for autonomous driving systems, supporting vehicle localization, trajectory planning, and lateral control. Compared with traditional two-dimensional lane detection, three-dimensional lane detection aims to recover the spatial structure of lane markings directly from monocular images. This task remains challenging because monocular imagery lacks explicit depth cues and is affected by perspective distortion, scale variation, and complex road environments. Recent studies have increasingly adopted bird's-eye-view representations that transform front-view image features into a top-down spatial layout. Such representations simplify geometric reasoning and facilitate the estimation of lane structures in three-dimensional space. However, existing convolutional neural network architectures still face difficulties in simultaneously preserving detailed lane geometry and extracting high-level semantic information. Shallow layers capture fine textures and structural boundaries, whereas deeper layers emphasize semantic abstraction and contextual understanding. Excessive dependence on deep features may introduce localization offsets after spatial projection, while shallow features alone are insufficient for reliable detection under challenging conditions such as illumination changes, occlusions, or background interference. A network architecture capable of integrating geometric detail with semantic representation is therefore important for improving monocular three-dimensional lane detection.

Methods

To address this issue, a monocular three-dimensional lane detection framework named HA-Lane Net was developed. The framework focuses on constructing balanced feature representations through a hybrid encoding strategy together with an attention-based feature refinement mechanism.

The backbone of the proposed framework is a hybrid encoder that combines complementary properties of two classical convolutional architectures. Shallow layers derived from VGG were used to capture detailed textures and structural patterns in lane regions. These features preserve spatial cues that are important for accurate geometric reconstruction. In parallel, deeper layers based on ResNet were employed to extract semantic representations and contextual information from complex road scenes. To integrate these heterogeneous features effectively, channel adaptation and hierarchical coupling mechanisms were introduced. These operations align feature dimensions and progressively fuse shallow geometric cues with deep semantic representations across multiple network stages. As a result, the encoder constructs a multi-scale feature hierarchy that maintains geometric precision while retaining strong semantic discrimination.

Results and Discussions

Feature refinement was further introduced before spatial projection through a pre-projection channel attention mechanism. Inspired by the squeeze-and-excitation principle, this module recalibrates feature responses by strengthening channels related to lane structures and suppressing background signals. Applying attention prior to the front-view-to-bird's-eye-view projection improves feature consistency across viewpoints and reduces the influence of image blur, lighting variation, and environmental clutter. This design helps preserve structural lane information during the spatial transformation process.

Training stability was improved using a cosine annealing learning-rate schedule, which facilitates smoother optimization of the multi-task loss function. The framework was trained and evaluated on the Apollo-Sim3D dataset, which contains diverse simulated driving scenes with accurate three-dimensional lane annotations. During inference, the network directly predicts spatial coordinates and shapes of lane markings from monocular images, enabling reconstruction of lane geometry in three-dimensional space without relying on additional sensors.

Results and Discussions

Experiments on the Apollo-Sim3D dataset show that HA-Lane Net improves three-dimensional lane detection performance compared with representative monocular approaches such as 3DLaneNet and Gen-Lane-Net. The hybrid encoder effectively integrates geometric detail with semantic context, enabling more accurate estimation of lane geometry.

The pre-projection attention mechanism further improves robustness by highlighting lane-related feature responses and suppressing irrelevant background information. This targeted refinement enhances feature consistency during the transformation from front-view images to bird's-eye-view representations and reduces localization errors caused by occlusion, illumination variation, or scene complexity. Consequently, the network produces more stable detection results across different driving scenarios.

Quantitative comparisons indicate improved accuracy in three-dimensional lane coordinate estimation relative to the reference methods used for comparison. Detection results also show increased stability across diverse road environments. These improvements arise from the combined effect of hybrid feature encoding and attention-based feature refinement, which strengthen lane structure representation while mitigating background noise. In addition, the proposed framework maintains a relatively efficient architecture. Performance improvements are achieved without introducing Transformer-based components or other computationally intensive modules, allowing the network to retain manageable complexity for real-time perception tasks.

Conclusions

A monocular three-dimensional lane detection framework named HA-Lane Net has been presented for improved lane perception in autonomous driving environments. The proposed approach integrates a hybrid encoder that combines shallow VGG layers with deep ResNet layers, enabling balanced extraction of geometric detail and semantic representation. A pre-projection channel attention mechanism further refines lane-related features before spatial transformation, improving robustness against environmental interference.

Experimental evaluation on the Apollo-Sim3D dataset shows that the proposed framework improves three-dimensional lane detection performance compared with representative monocular approaches such as 3DLaneNet and Gen-Lane-Net. The hybrid encoding strategy together with attention-based refinement enables more reliable estimation of lane geometry while maintaining a relatively efficient network structure. These results suggest that the proposed framework provides a practical solution for monocular three-dimensional lane perception and offers useful insights for future research on autonomous driving scene understanding.

, authors=Kui Zheng, Yingping Huang, Xin Luo, authorsList=Kui Zheng, Yingping Huang, Xin Luo, authorCompany=null, correspAuthors=Yingping Huang, authorNote=null, correspAuthorsNote=
, copyrightStatement=Copyright © 2026 Opto-Electronic Engineering. All rights reserved., copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=null, magXml=null, pdfUrl=null, pdf=null, pdfFileSize=null, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=null, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=null, mapNumber=null, fund=null), CN=ArticleExt(id=1319954996565754085, articleId=1319954959790097377, tenantId=1146029695717560320, journalId=1278651732997652489, language=CN, title=混合编码器融合注意力机制的单目三维车道线检测, columnId=1279511637858160772, journalTitle=光电工程, columnName=科研论文, runingTitle=null, highlight=null, articleAbstract=

三维车道线检测是自动驾驶领域的关键技术,旨在从二维图像中恢复车道线在三维空间中的几何形态。受三维信息的不确定性与复杂场景下表征歧义的限制,现有基于卷积神经网络的方法常面临特征分辨率与语义聚合之间的权衡:浅层特征具备更强的几何细节保真度,而深层特征语义判别性强但易出现空间量化损失与定位偏移。为此,本文提出一种新型三维车道线检测网络HA-Lane Net:1)设计了一种混合编码器,利用VGG浅层捕获细粒度纹理,结合ResNet深层提取高阶语义,通过通道适配与分级耦合构建均衡的特征层次;2)在输出层上引入投影前通道注意力机制,对车道相关通道进行自适应重标定,以增强跨视角一致性并抑制背景干扰。在Apollo-Sim3D数据集上的实验结果表明,该方法有效缓解了跨尺度与跨视角的不一致性。在不引入Transformer等复杂结构的前提下,HA-Lane Net的性能指标相较于3DLaneNet与Gen-Lane Net均有显著提升。

, authors=郑奎, 黄影平*, *, 罗鑫, authorsList=郑奎, 黄影平, 罗鑫, authorCompany=null, correspAuthors=黄影平, authorNote=

作者贡献

郑奎:研究资料的收集、分析,实验的设计与完成,论文写作;黄影平、罗鑫:指导论文选题、论文修改。

郑奎 (1999-),男,硕士研究生,主要从事计算机视觉的研究。E-mail:

黄影平 (1966-),男,教授,从事汽车电子、计算机视觉的研究。E-mail:

, correspAuthorsNote=
黄影平,
, copyrightStatement=版权所有©《光电工程》编辑部 2026, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=tPPqJpRkbids6+uyvHRpZg==, magXml=SfFNUFd+uZnh4eFyPSq91w==, pdfUrl=null, pdf=GnOYq67SOAWNDg+3Vpm/ig==, pdfFileSize=7098373, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=dxzC38NNCeraR7o9I9bepg==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=4zERQV1fHyzcWOzn44lLtA==, mapNumber=null, fund=null)}, authors=[Author(id=1319954997291368682, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=2049482458@qq.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1319954997366866156, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, authorId=1319954997291368682, language=EN, stringName=Kui Zheng, firstName=Kui, middleName=null, lastName=Zheng, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=School of Opto-Electronic and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1319954997492695277, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, authorId=1319954997291368682, language=CN, stringName=郑奎, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=上海理工大学光电信息与计算机工程学院,上海 200093, bio={"img":"kjk1n7LA69c7R4qwswQfxw==","content":"

郑奎 (1999-),男,硕士研究生,主要从事计算机视觉的研究。E-mail:

"}, bioImg=kjk1n7LA69c7R4qwswQfxw==, bioContent=

郑奎 (1999-),男,硕士研究生,主要从事计算机视觉的研究。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1319954997215871206, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, xref=null, ext=[AuthorCompanyExt(id=1319954997224259815, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, companyId=1319954997215871206, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Opto-Electronic and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China), AuthorCompanyExt(id=1319954997232648424, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, companyId=1319954997215871206, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=上海理工大学光电信息与计算机工程学院,上海 200093)])]), Author(id=1319954999212359919, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=huangyingping@usst.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1319954999313023217, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, authorId=1319954999212359919, language=EN, stringName=Yingping Huang, firstName=Yingping, middleName=null, lastName=Huang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=School of Opto-Electronic and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1319954999396909298, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, authorId=1319954999212359919, language=CN, stringName=黄影平, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, *, address=上海理工大学光电信息与计算机工程学院,上海 200093, bio={"img":"2FIfRjA/OlSmV4dXkTeDMg==","content":"

黄影平 (1966-),男,教授,从事汽车电子、计算机视觉的研究。E-mail:

"}, bioImg=2FIfRjA/OlSmV4dXkTeDMg==, bioContent=

黄影平 (1966-),男,教授,从事汽车电子、计算机视觉的研究。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1319954997215871206, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, xref=null, ext=[AuthorCompanyExt(id=1319954997224259815, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, companyId=1319954997215871206, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Opto-Electronic and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China), AuthorCompanyExt(id=1319954997232648424, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, companyId=1319954997215871206, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=上海理工大学光电信息与计算机工程学院,上海 200093)])]), Author(id=1319954999480795380, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1319954999745036534, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, authorId=1319954999480795380, language=EN, stringName=Xin Luo, firstName=Xin, middleName=null, lastName=Luo, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=School of Opto-Electronic and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1319954999824728311, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, authorId=1319954999480795380, language=CN, stringName=罗鑫, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=上海理工大学光电信息与计算机工程学院,上海 200093, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1319954997215871206, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, xref=null, ext=[AuthorCompanyExt(id=1319954997224259815, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, companyId=1319954997215871206, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Opto-Electronic and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China), AuthorCompanyExt(id=1319954997232648424, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, companyId=1319954997215871206, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=上海理工大学光电信息与计算机工程学院,上海 200093)])])], keywords=[Keyword(id=1319954999942168824, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, orderNo=1, keyword=3D lane detection), Keyword(id=1319955000021860602, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, orderNo=2, keyword=hybrid encoder), Keyword(id=1319955000156078331, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, orderNo=3, keyword=attention mechanism), Keyword(id=1319955000227381500, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, orderNo=4, keyword=autonomous driving), Keyword(id=1319955000307073277, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, orderNo=1, keyword=三维车道线检测), Keyword(id=1319955000369987838, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, orderNo=2, keyword=混合编码器), Keyword(id=1319955000445485311, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, orderNo=3, keyword=注意力机制), Keyword(id=1319955000525177088, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, orderNo=4, keyword=自动驾驶)], refs=[Reference(id=1319955008683098409, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=1, rfOrder=0, authorNames=null, journalName=null, refType=null, unstructuredReference=Luo X, Huang Y P, Cui J Y, et al. Deep learning-based lane detection for intelligent driving: a comprehensive survey of methods, datasets, challenges and outlooks[J]. Neurocomputing, 2025, 650: 130795., articleTitle=null, refAbstract=null), Reference(id=1319955009014448426, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=2, rfOrder=1, authorNames=null, journalName=null, refType=null, unstructuredReference=He X, Guo H Y, Zhu K, et al. Monocular lane detection based on deep learning: a survey[EB/OL]. arXiv: 2411.16316, 2024. https://doi.org/10.48550/arXiv.2411.16316., articleTitle=null, refAbstract=null), Reference(id=1319955009081557291, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=3, rfOrder=2, authorNames=null, journalName=null, refType=null, unstructuredReference=Garnett N, Cohen R, Pe'er T, et al. 3D-LaneNet: end-to-end 3D multiple lane detection[C]//Proceedings of 2019 IEEE/CVF International Conference on Computer Vision, Seoul, Korea (South), 2019: 2921‒2930. https://doi.org/10.1109/ICCV.2019.00301., articleTitle=null, refAbstract=null), Reference(id=1319955009152860460, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=4, rfOrder=3, authorNames=null, journalName=null, refType=null, unstructuredReference=Guo Y L, Chen G, Zhao P T, et al. Gen-LaneNet: a generalized and scalable approach for 3D lane detection[C]//Proceedings of the 16th European Conference on Computer Vision, Glasgow, UK, 2020: 666‒681. https://doi.org/10.1007/978-3-030-58589-1_40., articleTitle=null, refAbstract=null), Reference(id=1319955009232552237, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=5, rfOrder=4, authorNames=null, journalName=null, refType=null, unstructuredReference=Efrat N, Bluvstein M, Oron S, et al. 3D-LaneNet+: anchor free lane detection using a semi-local representation[EB/OL]. arXiv: 2011.01535, 2020. https://doi.org/10.48550/arXiv.2011.01535., articleTitle=null, refAbstract=null), Reference(id=1319955009391935790, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=6, rfOrder=5, authorNames=null, journalName=null, refType=null, unstructuredReference=Chen L, Sima C H, Li Y, et al. PersFormer: 3D lane detection via perspective transformer and the openlane benchmark[C]//Proceedings of the 17th European Conference on Computer Vision, Tel Aviv, Israel, 2022: 550‒567. https://doi.org/10.1007/978-3-031-19839-7_32., articleTitle=null, refAbstract=null), Reference(id=1319955009459044656, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=7, rfOrder=6, authorNames=null, journalName=null, refType=null, unstructuredReference=Wang R H, Qin J, Li K Y, et al. BEV-LaneDet: an efficient 3D lane detection based on virtual camera via key-points[C]//Proceedings of 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Vancouver, Canada, 2023: 21611‒21620. https://doi.org/10.1109/CVPR52729.2023.00103., articleTitle=null, refAbstract=null), Reference(id=1319955009542930737, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=8, rfOrder=7, authorNames=null, journalName=null, refType=null, unstructuredReference=Liu R J, Chen D P, Liu T, et al. Learning to predict 3D lane shape and camera pose from a single image via geometry constraints[C]//Proceedings of the 36th AAAI Conference on Artificial Intelligence, 2022: 1765‒1772. https://doi.org/10.1609/aaai.v36i2.20069., articleTitle=null, refAbstract=null), Reference(id=1319955009643594034, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=9, rfOrder=8, authorNames=null, journalName=null, refType=null, unstructuredReference=Chen Z Y, Smith-Miles K, Du B, et al. An efficient transformer for simultaneous learning of BEV and lane representations in 3D lane detection[EB/OL]. arXiv: 2306.04927, 2023. https://doi.org/10.48550/arXiv.2306.04927., articleTitle=null, refAbstract=null), Reference(id=1319955009735868723, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=10, rfOrder=9, authorNames=null, journalName=null, refType=null, unstructuredReference=Li C X, Hu Y H, Zheng Z W, et al. Bi2Lane: Bi-Directional temporal refinement with Bi-level feature aggregation for 3D lane detection[C]//Proceedings of 2024 IEEE International Conference on Robotics and Automation, Yokohama, Japan, 2024: 2894‒2900. https://doi.org/10.1109/ICRA57147.2024.10610794., articleTitle=null, refAbstract=null), Reference(id=1319955009807171892, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=11, rfOrder=10, authorNames=null, journalName=null, refType=null, unstructuredReference=Yan F, Nie M, Cai X Y, et al. ONCE-3DLanes: building monocular 3D lane detection[C]//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition, New Orleans, USA, 2022: 17122‒17131. https://doi.org/10.1109/CVPR52688.2022.01663., articleTitle=null, refAbstract=null), Reference(id=1319955009865892149, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=12, rfOrder=11, authorNames=null, journalName=null, refType=null, unstructuredReference=Zheng Z W, Zhang X M, Mou Y Q, et al. PVALane: prior-guided 3D lane detection with view-agnostic feature alignment[C]//Proceedings of the 38th AAAI Conference on Artificial Intelligence, Vancouver, Canada, 2024: 7597‒7604. https://doi.org/10.1609/aaai.v38i7.28592., articleTitle=null, refAbstract=null), Reference(id=1319955009924612406, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=13, rfOrder=12, authorNames=null, journalName=null, refType=null, unstructuredReference=Bai Y F, Chen Z R, Fu Z J, et al. CurveFormer: 3D lane detection by curve propagation with curve queries and attention[C]//Proceedings of 2023 IEEE International Conference on Robotics and Automation, London, United Kingdom, 2023: 7062‒7068. https://doi.org/10.1109/ICRA48891.2023.10161160., articleTitle=null, refAbstract=null), Reference(id=1319955009991721271, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=14, rfOrder=13, authorNames=null, journalName=null, refType=null, unstructuredReference=Huang S F, Shen Z W, Huang Z J, et al. Anchor3DLane: learning to regress 3D anchors for monocular 3D lane detection[C]//Proceedings of 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Vancouver, Canada, 2023: 17451‒17460. https://doi.org/10.1109/CVPR52729.2023.01674., articleTitle=null, refAbstract=null), Reference(id=1319955010063024440, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=15, rfOrder=14, authorNames=null, journalName=null, refType=null, unstructuredReference=Luo Y R, Zheng C D, Yan X, et al. LATR: 3D lane detection from monocular images with transformer[C]//Proceedings of 2023 IEEE/CVF International Conference on Computer Vision, Paris, France, 2023: 7907‒7918. https://doi.org/10.1109/ICCV51070.2023.00730., articleTitle=null, refAbstract=null), Reference(id=1319955010134327609, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=16, rfOrder=15, authorNames=null, journalName=null, refType=null, unstructuredReference=Bi J P, Song Y C, Jiang Y H, et al. Lane detection for autonomous driving: comprehensive reviews, current challenges, and future predictions[J]. IEEE Trans Intell Transp Syst, 2025, 26(5): 5710−5746., articleTitle=null, refAbstract=null), Reference(id=1319955010209825082, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=17, rfOrder=16, authorNames=null, journalName=null, refType=null, unstructuredReference=Huang J, Choudhury P K, Yin S, et al. Real-time road curb and lane detection for autonomous driving using LiDAR point clouds[J]. IEEE Access, 2021, 9: 144940−144951., articleTitle=null, refAbstract=null), Reference(id=1319955010289516859, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=18, rfOrder=17, authorNames=null, journalName=null, refType=null, unstructuredReference=Luo Y R, Yan X, Zheng C D, et al. M2-3DLaneNet: exploring multi-modal 3D lane detection[EB/OL]. arXiv: 2209.05996, 2022. https://doi.org/10.48550/arXiv.2209.05996., articleTitle=null, refAbstract=null), Reference(id=1319955010373402940, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=19, rfOrder=18, authorNames=null, journalName=null, refType=null, unstructuredReference=Zhao R K, Heng Y W, Wang H, et al. Advancements in 3D lane detection using LiDAR point clouds: from data collection to model development[C]//Proceedings of 2024 IEEE International Conference on Robotics and Automation, Yokohama, Japan, IEEE, 2024: 5382‒5388. https://doi.org/10.1109/ICRA57147.2024.10610087., articleTitle=null, refAbstract=null), Reference(id=1319955010444706109, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=20, rfOrder=19, authorNames=null, journalName=null, refType=null, unstructuredReference=Luo Y R, Cui S G, Li Z. Dv-3DLane: end-to-end multi-modal 3D lane detection with dual-view representation[C]//Proceedings of the 12th International Conference on Learning Representations, Vienna, Austria, 2024., articleTitle=null, refAbstract=null), Reference(id=1319955010507620670, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=21, rfOrder=20, authorNames=null, journalName=null, refType=null, unstructuredReference=罗鑫, 黄影平, 梁振明. 轴注意力引导的锚点分类车道线检测[J]. 光电工程, 2023, 50(7): 230079., articleTitle=null, refAbstract=null), Reference(id=1319955010591506751, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=21, rfOrder=21, authorNames=null, journalName=null, refType=null, unstructuredReference=Luo X, Huang Y P, Liang Z M. Axial attention-guided anchor classification lane detection[J]. Opto-Electron Eng, 2023, 50(7): 230079., articleTitle=null, refAbstract=null), Reference(id=1319955010671198528, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=22, rfOrder=22, authorNames=null, journalName=null, refType=null, unstructuredReference=Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[C]//Proceedings of the 3rd International Conference on Learning Representations, San Diego, USA, 2015., articleTitle=null, refAbstract=null), Reference(id=1319955010746696001, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=23, rfOrder=23, authorNames=null, journalName=null, refType=null, unstructuredReference=He K M, Zhang X Y, Ren S Q, et al. Deep residual learning for image recognition[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 770‒778. https://doi.org/10.1109/CVPR.2016.90., articleTitle=null, refAbstract=null), Reference(id=1319955010822193474, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=24, rfOrder=24, authorNames=null, journalName=null, refType=null, unstructuredReference=Lin M, Chen Q, Yan S C. Network in network[Z]. arXiv: 1312.4400, 2014. https://doi.org/10.48550/arXiv.1312.4400., articleTitle=null, refAbstract=null), Reference(id=1319955010918662467, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=25, rfOrder=25, authorNames=null, journalName=null, refType=null, unstructuredReference=Hu J, Shen L, Sun G. Squeeze-and-excitation networks[C]//Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 7132‒7141. https://doi.org/10.1109/CVPR.2018.00745., articleTitle=null, refAbstract=null), Reference(id=1319955012629938500, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=26, rfOrder=26, authorNames=null, journalName=null, refType=null, unstructuredReference=Li C G, Shi J, Wang Y, et al. Reconstruct from top view: a 3D lane detection approach based on geometry structure prior[C]//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition, New Orleans, USA, 2022: 4369‒4378. https://doi.org/10.1109/CVPRW56347.2022.00483., articleTitle=null, refAbstract=null)], funds=null, companyList=[AuthorCompany(id=1319954997215871206, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, xref=null, ext=[AuthorCompanyExt(id=1319954997224259815, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, companyId=1319954997215871206, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Opto-Electronic and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China), AuthorCompanyExt(id=1319954997232648424, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, companyId=1319954997215871206, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=上海理工大学光电信息与计算机工程学院,上海 200093)])], figs=[ArticleFig(id=1319955000663589121, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, label=Fig.1, caption=Schematic diagram of camera setup and vehicle coordinate system, figureFileSmall=xqlyZ+Da1NFaIwxtH+S6nw==, figureFileBig=dxzC38NNCeraR7o9I9bepg==, tableContent=null), ArticleFig(id=1319955000734892290, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, label=图1, caption=相机设置与车辆坐标系示意图, figureFileSmall=xqlyZ+Da1NFaIwxtH+S6nw==, figureFileBig=dxzC38NNCeraR7o9I9bepg==, tableContent=null), ArticleFig(id=1319955000944607491, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, label=Fig.2, caption=Geometric transformation and anchor point representation diagram, figureFileSmall=4YzFrjRRfXO22PtQJaQxxQ==, figureFileBig=yBMmFdD/c0RyXGIbdGZkUA==, tableContent=null), ArticleFig(id=1319955001011716356, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, label=图2, caption=几何变换与锚点表示图, figureFileSmall=4YzFrjRRfXO22PtQJaQxxQ==, figureFileBig=yBMmFdD/c0RyXGIbdGZkUA==, tableContent=null), ArticleFig(id=1319955001099796741, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, label=Fig.3, caption=Schematic diagram of the overall network structure, figureFileSmall=TAfgSApsX4TN8y6Rf4NbZQ==, figureFileBig=9djJXS2S8WKpE3HYHuxs4g==, tableContent=null), ArticleFig(id=1319955001175294214, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, label=图3, caption=网络整体结构示意图, figureFileSmall=TAfgSApsX4TN8y6Rf4NbZQ==, figureFileBig=9djJXS2S8WKpE3HYHuxs4g==, tableContent=null), ArticleFig(id=1319955001259180295, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, label=Fig.4, caption=Schematic diagram of the hybrid encoder structure, figureFileSmall=wAGANtSYTogwy6GE8Py1yw==, figureFileBig=Ej3k+U8LR8OjBRAoh6JMrw==, tableContent=null), ArticleFig(id=1319955001464701192, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, label=图4, caption=混合编码器结构示意图, figureFileSmall=wAGANtSYTogwy6GE8Py1yw==, figureFileBig=Ej3k+U8LR8OjBRAoh6JMrw==, tableContent=null), ArticleFig(id=1319955001565364489, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, label=Fig.5, caption=Schematic diagram of the location and structure of the attention module, figureFileSmall=dHjlIukYR4Pe5M26JMgeJw==, figureFileBig=dl6BGjnu6dP5mhqaHvr7FQ==, tableContent=null), ArticleFig(id=1319955001695387914, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, label=图5, caption=注意力模块添加的位置与结构示意图, figureFileSmall=dHjlIukYR4Pe5M26JMgeJw==, figureFileBig=dl6BGjnu6dP5mhqaHvr7FQ==, tableContent=null), ArticleFig(id=1319955001829605643, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, label=Fig.6, caption=Comparative visualization of attention-based feature representations across different scenarios, figureFileSmall=uFb1lu0njp31IKH+kHvAOw==, figureFileBig=Tl2v9WSTi5DXhaehRo+5oQ==, tableContent=null), ArticleFig(id=1319955001926074636, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, label=图6, caption=不同场景下注意力机制特征可视化对比图, figureFileSmall=uFb1lu0njp31IKH+kHvAOw==, figureFileBig=Tl2v9WSTi5DXhaehRo+5oQ==, tableContent=null), ArticleFig(id=1319955003599601933, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, label=Fig.7, caption=Comparison of lane detection performance on straightaways and curves at night in rare scenarios, figureFileSmall=h/JTMeKe3beMOTAgH5R1Og==, figureFileBig=qg8ouQIZlojW0yXPcRnzcA==, tableContent=null), ArticleFig(id=1319955003712848142, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, label=图7, caption=罕见场景下夜晚车道线检测直道与弯道效果对比图, figureFileSmall=h/JTMeKe3beMOTAgH5R1Og==, figureFileBig=qg8ouQIZlojW0yXPcRnzcA==, tableContent=null), ArticleFig(id=1319955003821900047, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, label=Fig.8, caption=Comparison of lane line detection curve performance in rare scenarios (morning and afternoon), figureFileSmall=oPIR74nEWqi/0sXv3pX/cQ==, figureFileBig=tQCVSDEOdzp9H1oZkC/jSA==, tableContent=null), ArticleFig(id=1319955003918369040, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, label=图8, caption=罕见场景下上午与下午车道线检测弯道效果对比图, figureFileSmall=oPIR74nEWqi/0sXv3pX/cQ==, figureFileBig=tQCVSDEOdzp9H1oZkC/jSA==, tableContent=null), ArticleFig(id=1319955003981283601, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, label=Fig.9, caption=Comparison of lane line detection ramp performance in rare scenarios during the morning, figureFileSmall=ega7GuDVqOE39WQ3Dy5k0g==, figureFileBig=sNEYwGGx9U6QeI5sllfgWA==, tableContent=null), ArticleFig(id=1319955004069363986, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, label=图9, caption=罕见场景下早晨车道线检测坡道效果对比图, figureFileSmall=ega7GuDVqOE39WQ3Dy5k0g==, figureFileBig=sNEYwGGx9U6QeI5sllfgWA==, tableContent=null), ArticleFig(id=1319955004161638675, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, label=Tab.1, caption=

Distribution of synthetic datasets

, figureFileSmall=null, figureFileBig=null, tableContent=
SceneTraining set sizeTest set size
Balanced scene59921496
Rare subset5992320
Visual variations3968472
), ArticleFig(id=1319955004232941844, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, label=表1, caption=

合成数据集分布

, figureFileSmall=null, figureFileBig=null, tableContent=
SceneTraining set sizeTest set size
Balanced scene59921496
Rare subset5992320
Visual variations3968472
), ArticleFig(id=1319955004358770965, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, label=Tab.2, caption=

Ablation experiment results of visual change scenes

, figureFileSmall=null, figureFileBig=null, tableContent=
Geometric anchorHybrid encoderAttention mechanismLearning strategyAP/%$ \uparrow $F1/%$ \uparrow $
√84.081.9
√√86.584.4
√√86.884.7
√√√86.984.9
√√√89.887.9
√√√90.488.5
√√√√91.589.5
), ArticleFig(id=1319955004430074134, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, label=表2, caption=

视觉变化场景消融实验结果

, figureFileSmall=null, figureFileBig=null, tableContent=
Geometric anchorHybrid encoderAttention mechanismLearning strategyAP/%$ \uparrow $F1/%$ \uparrow $
√84.081.9
√√86.584.4
√√86.884.7
√√√86.984.9
√√√89.887.9
√√√90.488.5
√√√√91.589.5
), ArticleFig(id=1319955004530737432, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, label=Tab.3, caption=

Sensitivity analysis of the dimensionality reduction ratio of the attention mechanism under visual variation scenarios

, figureFileSmall=null, figureFileBig=null, tableContent=
Reduction ratio/rMethodsAP/%$ \uparrow $F1/%$ \uparrow $Ex/F (m)$ \downarrow $Ex/N (m)$ \downarrow $Ez/F (m)$ \downarrow $Ez/N (m)$ \downarrow $
4VGG+SE86.984.90.09130.6010.01920.293
8VGG+SE87.785.80.11100.5810.02260.306
16VGG+SE86.884.70.11000.5610.02110.305
32VGG+SE81.179.30.12000.5850.02440.271
8Hybrid+SE88.886.70.06880.4700.01340.273
16Hybrid+SE91.589.40.06100.4510.01400.276
16-8Hybrid+SE89.587.40.07480.4840.01540.296
), ArticleFig(id=1319955004778201369, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, label=表3, caption=

视觉变化场景注意力机制降维率的敏感性分析表

, figureFileSmall=null, figureFileBig=null, tableContent=
Reduction ratio/rMethodsAP/%$ \uparrow $F1/%$ \uparrow $Ex/F (m)$ \downarrow $Ex/N (m)$ \downarrow $Ez/F (m)$ \downarrow $Ez/N (m)$ \downarrow $
4VGG+SE86.984.90.09130.6010.01920.293
8VGG+SE87.785.80.11100.5810.02260.306
16VGG+SE86.884.70.11000.5610.02110.305
32VGG+SE81.179.30.12000.5850.02440.271
8Hybrid+SE88.886.70.06880.4700.01340.273
16Hybrid+SE91.589.40.06100.4510.01400.276
16-8Hybrid+SE89.587.40.07480.4840.01540.296
), ArticleFig(id=1319955004874670362, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, label=Tab.4, caption=

Comparison with other methods on standard and rare scenarios of the Apollo-Sim3D dataset

, figureFileSmall=null, figureFileBig=null, tableContent=
SceneMethodsAP/%$ \uparrow $F1/%$ \uparrow $Ex/F (m)$ \downarrow $Ex/N (m)$ \downarrow $Ez/F (m)$ \downarrow $Ez/N (m)$ \downarrow $
Balanced scene3DLaneNet[3]89.386.40.06800.4770.01500.202
3DLaneNet_ext (baseline)91.789.70.06690.4530.01330.241
Gen-LaneNet[4]89.287.30.06360.5340.01270.247
LS-3DLane[1]89.689.40.08340.4800.02170.271
3DLaneNet (DA)[6]93.291.00.08200.4390.01100.242
GenLaneNet (DA)[6]92.490.30.08000.4730.01100.247
Reconstruct from Top[26]93.891.90.04900.3870.00800.213
CLGo[10]94.291.90.06100.3610.02900.250
Ours (hybrid+se_L)93.891.80.04400.3830.00800.230
Rare subset3DLaneNet[3]74.672.00.16600.8550.03900.521
3DLaneNet_ext (baseline)82.481.20.15500.8820.03160.622
Gen-LaneNet[4]77.976.90.14700.9770.31700.633
LS-3DLane[1]79.978.10.17300.9520.04690.705
3DLaneNet (DA)[6]85.884.10.28900.9250.02500.625
GenLaneNet (DA)[6]83.281.70.28300.9150.02800.653
Reconstruct from Top[26]85.283.70.12600.9030.02300.625
CLGo[10]88.386.10.14700.7350.07100.609
Ours (hybrid+se_L)86.684.70.10900.8280.02190.616
), ArticleFig(id=1319955005193437467, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, label=表4, caption=

在Apollo-Sim3D数据集的标准与罕见场景上与其他方法的比较

, figureFileSmall=null, figureFileBig=null, tableContent=
SceneMethodsAP/%$ \uparrow $F1/%$ \uparrow $Ex/F (m)$ \downarrow $Ex/N (m)$ \downarrow $Ez/F (m)$ \downarrow $Ez/N (m)$ \downarrow $
Balanced scene3DLaneNet[3]89.386.40.06800.4770.01500.202
3DLaneNet_ext (baseline)91.789.70.06690.4530.01330.241
Gen-LaneNet[4]89.287.30.06360.5340.01270.247
LS-3DLane[1]89.689.40.08340.4800.02170.271
3DLaneNet (DA)[6]93.291.00.08200.4390.01100.242
GenLaneNet (DA)[6]92.490.30.08000.4730.01100.247
Reconstruct from Top[26]93.891.90.04900.3870.00800.213
CLGo[10]94.291.90.06100.3610.02900.250
Ours (hybrid+se_L)93.891.80.04400.3830.00800.230
Rare subset3DLaneNet[3]74.672.00.16600.8550.03900.521
3DLaneNet_ext (baseline)82.481.20.15500.8820.03160.622
Gen-LaneNet[4]77.976.90.14700.9770.31700.633
LS-3DLane[1]79.978.10.17300.9520.04690.705
3DLaneNet (DA)[6]85.884.10.28900.9250.02500.625
GenLaneNet (DA)[6]83.281.70.28300.9150.02800.653
Reconstruct from Top[26]85.283.70.12600.9030.02300.625
CLGo[10]88.386.10.14700.7350.07100.609
Ours (hybrid+se_L)86.684.70.10900.8280.02190.616
), ArticleFig(id=1319955005315072284, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, label=Tab.5, caption=

Comparison with other methods on visual variation scenarios of the Apollo-Sim3D dataset

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodsAP/%$ \uparrow $F1/%$ \uparrow $Ex/F (m)$ \downarrow $Ex/N (m)$ \downarrow $Ez/F (m)$ \downarrow $Ez/N (m)$ \downarrow $
3DLaneNet[3]74.972.50.11500.6010.03200.230
3DLaneNet_ext (baseline)84.081.90.10700.5830.02490.328
Gen-LaneNet[4]87.285.30.07400.5380.01500.232
LS-3DLane[1]89.087.30.08480.5290.03210.330
3DLaneNet (DA)[6]87.485.40.11800.5590.01800.290
GenLaneNet (DA)[6]88.586.80.10400.5440.01600.294
Reconstruct from Top[26]92.189.90.06000.4460.01100.235
CLGo[10]89.287.30.08400.4640.04500.312
Ours (ResNet34)84.582.50.10700.5730.02500.331
Ours (hybrid)86.584.40.11600.6040.01780.311
Ours (vgg+se)86.684.70.10900.5610.02100.305
Ours (hybrid+se)86.984.90.08960.5640.01870.288
Ours (vgg+se_L)89.887.90.05750.4620.01350.271
Ours (hybrid_L)90.488.50.06340.4570.01330.258
Ours (hybrid+se_L_T)91.289.20.07000.4730.01400.263
Ours (hybrid+se_L)91.589.40.06100.4510.01400.276
), ArticleFig(id=1319955006057464093, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, label=表5, caption=

在Apollo-Sim3D数据集的视觉变化场景上与其他方法的比较

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodsAP/%$ \uparrow $F1/%$ \uparrow $Ex/F (m)$ \downarrow $Ex/N (m)$ \downarrow $Ez/F (m)$ \downarrow $Ez/N (m)$ \downarrow $
3DLaneNet[3]74.972.50.11500.6010.03200.230
3DLaneNet_ext (baseline)84.081.90.10700.5830.02490.328
Gen-LaneNet[4]87.285.30.07400.5380.01500.232
LS-3DLane[1]89.087.30.08480.5290.03210.330
3DLaneNet (DA)[6]87.485.40.11800.5590.01800.290
GenLaneNet (DA)[6]88.586.80.10400.5440.01600.294
Reconstruct from Top[26]92.189.90.06000.4460.01100.235
CLGo[10]89.287.30.08400.4640.04500.312
Ours (ResNet34)84.582.50.10700.5730.02500.331
Ours (hybrid)86.584.40.11600.6040.01780.311
Ours (vgg+se)86.684.70.10900.5610.02100.305
Ours (hybrid+se)86.984.90.08960.5640.01870.288
Ours (vgg+se_L)89.887.90.05750.4620.01350.271
Ours (hybrid_L)90.488.50.06340.4570.01330.258
Ours (hybrid+se_L_T)91.289.20.07000.4730.01400.263
Ours (hybrid+se_L)91.589.40.06100.4510.01400.276
), ArticleFig(id=1319955006141350174, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, label=Tab.6, caption=

Speed comparison of different methods

, figureFileSmall=null, figureFileBig=null, tableContent=
SceneLATR[15]BaselineHA-Lane Net
Balanced scene34.8550156.670176.580
Rare subset35.296759.62064.530
Visual variations36.423389.26593.735
), ArticleFig(id=1319955006451728671, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, label=表6, caption=

推理速度对比表

, figureFileSmall=null, figureFileBig=null, tableContent=
SceneLATR[15]BaselineHA-Lane Net
Balanced scene34.8550156.670176.580
Rare subset35.296759.62064.530
Visual variations36.423389.26593.735
), ArticleFig(id=1319955008196559142, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=EN, label=Tab.7, caption=

Computational complexity comparison

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodsParams/MGFLOPsGMACs
Baseline20.60 121.046 60.523
HA-Lane Net28.88 104.616 52.308
), ArticleFig(id=1319955008557269288, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, language=CN, label=表7, caption=

复杂度对比表

, figureFileSmall=null, figureFileBig=null, tableContent=
MethodsParams/MGFLOPsGMACs
Baseline20.60 121.046 60.523
HA-Lane Net28.88 104.616 52.308
)], attaches=null, journal=Journal(id=1278641367198941188, delFlag=0, nameCn=光电工程, nameEn=Opto-Electronic Engineering, nameHistory1=null, nameHistory2=null, issn=1003-501X, eissn=2097-4019, cn=51-1346/O4, coden=null, periodic=0, language=CN, oaType=null, ccby=null, superviseOffice=null, ownerOffice=null, pubOffice=null, editorOffice=null, officeType=null, aims=null, clcCode=null, officeProv=null, officeCity=null, officeAddr=null, officeZip=null, officeEmail=null, officePhone=null, editDirector=null, officeDirector=null, officeDirectorPhone=null, officeStaffNum=null, officeEmpNum=null, coverPicUrl=4Vimkd+qXLWNxtdpr9mFNw==, journalPrice=null, startedYear=null, abbrevIsoEn=Opto-Electronic Engineering, journalRemark=null, publicationField=null, createdTime=1782781457950, updatedTime=1784021784694, createdBy=18614031015, updatedBy=13041195026, firstLetterCn=G, firstLetterEn=G, subjectCode=Engineering, subjectName=null, subjectCodeEn=Engineering, subjectNameEn=null, picCn=4Vimkd+qXLWNxtdpr9mFNw==, picEn=vAq9s20WLs1ODDfbWq+Gjg==, jcr=null, cjcr=null, exts=[JournalExt(id=1283843675721536154, language=CN, name=光电工程, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1784021784954, updatedTime=1784021784954, createdBy=13041195026, updatedBy=13041195026, submissionGuidelinesUrl=, submissionAuthorUrl=http://www.manuscripts.com.cn/gdgc, submissionEditorUrl=http://www.manuscripts.com.cn/gdgc, submissionReviewUrl=http://www.manuscripts.com.cn/gdgc, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""}), JournalExt(id=1283843675771867803, language=EN, name=Opto-Electronic Engineering, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1784021784966, updatedTime=1784021784966, createdBy=13041195026, updatedBy=13041195026, submissionGuidelinesUrl=, submissionAuthorUrl=http://www.manuscripts.com.cn/gdgc, submissionEditorUrl=http://www.manuscripts.com.cn/gdgc, submissionReviewUrl=http://www.manuscripts.com.cn/gdgc, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""})], databaseList=null, tenantJournalId=1278651732997652489, websiteList=[Website(id=1278723867418018151, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1278651732997652489, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/oee/CN, language=CN, createTime=1782801127533, createBy=18614031015, updateTime=1782804494442, updateBy=18614031015, name=光电工程-中文, tplId=1146099689490845704, title=光电工程, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1278738091150128034, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=articleTextType, value=kx, createTime=1782804518735, updateTime=1782804518735, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091120767903, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=banner, value=null, createTime=1782804518728, updateTime=1782804518728, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091171099557, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=grayFlag, value=0, createTime=1782804518740, updateTime=1782804518740, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091108184990, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=logo, value=https://castjournals.cast.org.cn/joweb/oee/CN/file/pic?fileId=A1C6uwqtMazluiWkEpR0Mg==, createTime=1782804518725, updateTime=1782804518725, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091179488167, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=minRunFlag, value=0, createTime=1782804518742, updateTime=1782804518742, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091141739425, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/oee/CN/file/pic, createTime=1782804518733, updateTime=1782804518733, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091175293862, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=silenceFlag, value=0, createTime=1782804518741, updateTime=1782804518741, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091129156512, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1782804518730, updateTime=1782804518730, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091154322339, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=themeColor, value=null, createTime=1782804518736, updateTime=1782804518736, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738091162710948, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867418018151, code=themeStyle, value=null, createTime=1782804518738, updateTime=1782804518738, creator=18614031015, updator=18614031015)]), Website(id=1278723867522875769, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1278651732997652489, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/oee/EN, language=EN, createTime=1782801127558, createBy=18614031015, updateTime=1782804490442, updateBy=18614031015, name=光电工程-英文, tplId=1146101810881728533, title=Opto-Electronic Engineering, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1278738063660659607, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=articleTextType, value=kx, createTime=1782804512181, updateTime=1782804512181, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063635493780, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=banner, value=null, createTime=1782804512175, updateTime=1782804512175, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063924900762, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=grayFlag, value=0, createTime=1782804512244, updateTime=1782804512244, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063606133651, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=logo, value=https://castjournals.cast.org.cn/joweb/oee/EN/file/pic?fileId=A1C6uwqtMazluiWkEpR0Mg==, createTime=1782804512168, updateTime=1782804512168, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063937483676, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=minRunFlag, value=0, createTime=1782804512247, updateTime=1782804512247, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063652270998, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/oee/EN/file/pic, createTime=1782804512179, updateTime=1782804512179, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063933289371, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=silenceFlag, value=0, createTime=1782804512246, updateTime=1782804512246, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063639688085, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_en_623/, createTime=1782804512176, updateTime=1782804512176, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063664853912, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=themeColor, value=null, createTime=1782804512182, updateTime=1782804512182, creator=18614031015, updator=18614031015), WebsiteProps(id=1278738063916512153, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1278723867522875769, code=themeStyle, value=null, createTime=1782804512242, updateTime=1782804512242, creator=18614031015, updator=18614031015)])], journalTitle=光电工程, weixinUrl=null, journalUrl=https://www.oejournal.org/oee, iacademicId=null, status=1, seqNo=null, journalTitleEn=Opto-Electronic Engineering, journalPhotoCn=4Vimkd+qXLWNxtdpr9mFNw==, journalPhotoEn=vAq9s20WLs1ODDfbWq+Gjg==, journalFirstLetter=G, journalRecommend=null, journalNew=null, journalCollection=null, jcrJf=null, cjcrJf=null, jcrJfStr=null, cjcrJfStr=null, submissionFirstDecision=null, sciSubjectClassification=null, casSubjectClassification=null, citeScore=null, totalCitationFrequency=null, icpCode=null, psCode=null, advertisingLicenseCode=null, copyrightInformation=null, country=null, option=, provinceCode=null, provinceName=null, collectFlag=false, interPubPlatform=, interPubPlatformUrl=null), detailUrlCn=https://castjournals.cast.org.cn/joweb/oee/CN/10.12086/oee.2026.250359, detailUrlEn=https://castjournals.cast.org.cn/joweb/oee/EN/10.12086/oee.2026.250359, pdfUrlCn=https://castjournals.cast.org.cn/joweb/oee/CN/PDF/10.12086/oee.2026.250359, pdfUrlEn=https://castjournals.cast.org.cn/joweb/oee/EN/PDF/10.12086/oee.2026.250359, aliStartDate=0, aliEndDate=0, collectionFlag=false, citedCount=null, citedUrl=null, previewStatus=0, delFlag=0, hasFullText=1, orderTime=1782316800000, fullTextJson=null, articleText=null, reference=null)
收藏切换
混合编码器融合注意力机制的单目三维车道线检测
收藏切换
PDF下载
郑奎 , 黄影平 *, * , 罗鑫
光电工程 | 科研论文 2026,53(6): 250359-1
收起
收藏切换
光电工程 |科研论文 2026 , 53 (6) : 250359 -1
混合编码器融合注意力机制的单目三维车道线检测
全屏
[Author(id=1319954997291368682, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=2049482458@qq.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1319954997366866156, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, authorId=1319954997291368682, language=EN, stringName=Kui Zheng, firstName=Kui, middleName=null, lastName=Zheng, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=School of Opto-Electronic and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1319954997492695277, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, authorId=1319954997291368682, language=CN, stringName=郑奎, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=上海理工大学光电信息与计算机工程学院,上海 200093, bio={"img":"kjk1n7LA69c7R4qwswQfxw==","content":"

郑奎 (1999-),男,硕士研究生,主要从事计算机视觉的研究。E-mail:

"}, bioImg=kjk1n7LA69c7R4qwswQfxw==, bioContent=

郑奎 (1999-),男,硕士研究生,主要从事计算机视觉的研究。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1319954997215871206, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, xref=null, ext=[AuthorCompanyExt(id=1319954997224259815, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, companyId=1319954997215871206, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Opto-Electronic and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China), AuthorCompanyExt(id=1319954997232648424, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, companyId=1319954997215871206, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=上海理工大学光电信息与计算机工程学院,上海 200093)])]), Author(id=1319954999212359919, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=huangyingping@usst.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1319954999313023217, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, authorId=1319954999212359919, language=EN, stringName=Yingping Huang, firstName=Yingping, middleName=null, lastName=Huang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=School of Opto-Electronic and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1319954999396909298, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, authorId=1319954999212359919, language=CN, stringName=黄影平, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, *, address=上海理工大学光电信息与计算机工程学院,上海 200093, bio={"img":"2FIfRjA/OlSmV4dXkTeDMg==","content":"

黄影平 (1966-),男,教授,从事汽车电子、计算机视觉的研究。E-mail:

"}, bioImg=2FIfRjA/OlSmV4dXkTeDMg==, bioContent=

黄影平 (1966-),男,教授,从事汽车电子、计算机视觉的研究。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1319954997215871206, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, xref=null, ext=[AuthorCompanyExt(id=1319954997224259815, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, companyId=1319954997215871206, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Opto-Electronic and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China), AuthorCompanyExt(id=1319954997232648424, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, companyId=1319954997215871206, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=上海理工大学光电信息与计算机工程学院,上海 200093)])]), Author(id=1319954999480795380, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1319954999745036534, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, authorId=1319954999480795380, language=EN, stringName=Xin Luo, firstName=Xin, middleName=null, lastName=Luo, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=School of Opto-Electronic and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1319954999824728311, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, authorId=1319954999480795380, language=CN, stringName=罗鑫, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=上海理工大学光电信息与计算机工程学院,上海 200093, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1319954997215871206, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, xref=null, ext=[AuthorCompanyExt(id=1319954997224259815, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, companyId=1319954997215871206, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Opto-Electronic and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China), AuthorCompanyExt(id=1319954997232648424, tenantId=1146029695717560320, journalId=1278651732997652489, articleId=1319954959790097377, companyId=1319954997215871206, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=上海理工大学光电信息与计算机工程学院,上海 200093)])])]
郑奎 , 黄影平*, * , 罗鑫
作者信息
  • 上海理工大学光电信息与计算机工程学院,上海 200093
通讯作者:
作者简介:

作者贡献

郑奎:研究资料的收集、分析,实验的设计与完成,论文写作;黄影平、罗鑫:指导论文选题、论文修改。

郑奎 (1999-),男,硕士研究生,主要从事计算机视觉的研究。E-mail:

黄影平 (1966-),男,教授,从事汽车电子、计算机视觉的研究。E-mail:

HA-Lane Net: Hybrid feature encoding and attention for monocular 3D lane detection
Kui Zheng , Yingping Huang , Xin Luo
Affiliations
  • School of Opto-Electronic and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China
  • Corresponding Author:
出版时间: 2026-06-25 doi: 10.12086/oee.2026.250359
文章导航
收藏切换

三维车道线检测是自动驾驶领域的关键技术,旨在从二维图像中恢复车道线在三维空间中的几何形态。受三维信息的不确定性与复杂场景下表征歧义的限制,现有基于卷积神经网络的方法常面临特征分辨率与语义聚合之间的权衡:浅层特征具备更强的几何细节保真度,而深层特征语义判别性强但易出现空间量化损失与定位偏移。为此,本文提出一种新型三维车道线检测网络HA-Lane Net:1)设计了一种混合编码器,利用VGG浅层捕获细粒度纹理,结合ResNet深层提取高阶语义,通过通道适配与分级耦合构建均衡的特征层次;2)在输出层上引入投影前通道注意力机制,对车道相关通道进行自适应重标定,以增强跨视角一致性并抑制背景干扰。在Apollo-Sim3D数据集上的实验结果表明,该方法有效缓解了跨尺度与跨视角的不一致性。在不引入Transformer等复杂结构的前提下,HA-Lane Net的性能指标相较于3DLaneNet与Gen-Lane Net均有显著提升。

三维车道线检测  /  混合编码器  /  注意力机制  /  自动驾驶
Objective

Three-dimensional lane detection provides essential geometric information for autonomous driving systems, supporting vehicle localization, trajectory planning, and lateral control. Compared with traditional two-dimensional lane detection, three-dimensional lane detection aims to recover the spatial structure of lane markings directly from monocular images. This task remains challenging because monocular imagery lacks explicit depth cues and is affected by perspective distortion, scale variation, and complex road environments. Recent studies have increasingly adopted bird's-eye-view representations that transform front-view image features into a top-down spatial layout. Such representations simplify geometric reasoning and facilitate the estimation of lane structures in three-dimensional space. However, existing convolutional neural network architectures still face difficulties in simultaneously preserving detailed lane geometry and extracting high-level semantic information. Shallow layers capture fine textures and structural boundaries, whereas deeper layers emphasize semantic abstraction and contextual understanding. Excessive dependence on deep features may introduce localization offsets after spatial projection, while shallow features alone are insufficient for reliable detection under challenging conditions such as illumination changes, occlusions, or background interference. A network architecture capable of integrating geometric detail with semantic representation is therefore important for improving monocular three-dimensional lane detection.

Methods

To address this issue, a monocular three-dimensional lane detection framework named HA-Lane Net was developed. The framework focuses on constructing balanced feature representations through a hybrid encoding strategy together with an attention-based feature refinement mechanism.

The backbone of the proposed framework is a hybrid encoder that combines complementary properties of two classical convolutional architectures. Shallow layers derived from VGG were used to capture detailed textures and structural patterns in lane regions. These features preserve spatial cues that are important for accurate geometric reconstruction. In parallel, deeper layers based on ResNet were employed to extract semantic representations and contextual information from complex road scenes. To integrate these heterogeneous features effectively, channel adaptation and hierarchical coupling mechanisms were introduced. These operations align feature dimensions and progressively fuse shallow geometric cues with deep semantic representations across multiple network stages. As a result, the encoder constructs a multi-scale feature hierarchy that maintains geometric precision while retaining strong semantic discrimination.

Results and Discussions

Feature refinement was further introduced before spatial projection through a pre-projection channel attention mechanism. Inspired by the squeeze-and-excitation principle, this module recalibrates feature responses by strengthening channels related to lane structures and suppressing background signals. Applying attention prior to the front-view-to-bird's-eye-view projection improves feature consistency across viewpoints and reduces the influence of image blur, lighting variation, and environmental clutter. This design helps preserve structural lane information during the spatial transformation process.

Training stability was improved using a cosine annealing learning-rate schedule, which facilitates smoother optimization of the multi-task loss function. The framework was trained and evaluated on the Apollo-Sim3D dataset, which contains diverse simulated driving scenes with accurate three-dimensional lane annotations. During inference, the network directly predicts spatial coordinates and shapes of lane markings from monocular images, enabling reconstruction of lane geometry in three-dimensional space without relying on additional sensors.

Results and Discussions

Experiments on the Apollo-Sim3D dataset show that HA-Lane Net improves three-dimensional lane detection performance compared with representative monocular approaches such as 3DLaneNet and Gen-Lane-Net. The hybrid encoder effectively integrates geometric detail with semantic context, enabling more accurate estimation of lane geometry.

The pre-projection attention mechanism further improves robustness by highlighting lane-related feature responses and suppressing irrelevant background information. This targeted refinement enhances feature consistency during the transformation from front-view images to bird's-eye-view representations and reduces localization errors caused by occlusion, illumination variation, or scene complexity. Consequently, the network produces more stable detection results across different driving scenarios.

Quantitative comparisons indicate improved accuracy in three-dimensional lane coordinate estimation relative to the reference methods used for comparison. Detection results also show increased stability across diverse road environments. These improvements arise from the combined effect of hybrid feature encoding and attention-based feature refinement, which strengthen lane structure representation while mitigating background noise. In addition, the proposed framework maintains a relatively efficient architecture. Performance improvements are achieved without introducing Transformer-based components or other computationally intensive modules, allowing the network to retain manageable complexity for real-time perception tasks.

Conclusions

A monocular three-dimensional lane detection framework named HA-Lane Net has been presented for improved lane perception in autonomous driving environments. The proposed approach integrates a hybrid encoder that combines shallow VGG layers with deep ResNet layers, enabling balanced extraction of geometric detail and semantic representation. A pre-projection channel attention mechanism further refines lane-related features before spatial transformation, improving robustness against environmental interference.

Experimental evaluation on the Apollo-Sim3D dataset shows that the proposed framework improves three-dimensional lane detection performance compared with representative monocular approaches such as 3DLaneNet and Gen-Lane-Net. The hybrid encoding strategy together with attention-based refinement enables more reliable estimation of lane geometry while maintaining a relatively efficient network structure. These results suggest that the proposed framework provides a practical solution for monocular three-dimensional lane perception and offers useful insights for future research on autonomous driving scene understanding.

3D lane detection  /  hybrid encoder  /  attention mechanism  /  autonomous driving
郑奎, 黄影平, 罗鑫. 混合编码器融合注意力机制的单目三维车道线检测. 光电工程, 2026 , 53 (6) : 250359 -1 . DOI: 10.12086/oee.2026.250359
Kui Zheng, Yingping Huang, Xin Luo. HA-Lane Net: Hybrid feature encoding and attention for monocular 3D lane detection[J]. Opto-Electronic Engineering, 2026 , 53 (6) : 250359 -1 . DOI: 10.12086/oee.2026.250359
车道线检测是无人驾驶车辆实现路径规划与横向控制的核心基础。相较于仅能提供图像平面坐标的传统二维检测,三维车道线检测致力于重建车道线的三维几何属性,提供对道路环境更精确的理解,是无人驾驶技术深入发展的必然需求[1]。
当前,基于单目视觉的三维车道线检测已成为主流研究方向,其中基于鸟瞰图 (bird's-eye-view, BEV)为代表的方法表现尤为突出。此类方法多遵循“图像特征提取-空间视角变换-结构化预测”的范式:利用卷积神经网络提取前视 (front view,FV)特征,通过投影变换将其映射至BEV空间以消除透视效应,从而将三维检测简化为BEV平面上的二维检测与高度回归任务。尽管该范式在理论上具有可行性,但在实际应用中仍面临以下瓶颈:1)特征编码器表征能力受限:现有基线模型多采用通用卷积网络,在面对光照剧变、遮挡及恶劣天气时,难以提取具有高判别性的深层语义特征。2)通道信息冗余与干扰:标准卷积缺乏对特征通道间依赖关系的显式建模,导致关键的车道线语义信息易被背景噪声淹没,降低了特征的信噪比。3)视角变换的误差级联效应:空间映射对输入特征质量极度敏感。前视图像中的局部模糊或特征缺失经几何变换后会被显著放大,直接削弱了BEV视角下车道线空间定位与几何拟合的精度。
针对上述挑战,本文的主要贡献如下:
1) 设计了一种混合编码器:协同融合VGG网络的浅层细粒度纹理特征与ResNet网络的深层高阶语义,并引入余弦退火策略优化训练,构建了兼顾局部细节与全局语义的均衡特征表示。
2) 引入注意力机制优化:在编码器深层嵌入Squeeze-and-Excitation (SE)模块,通过“压缩-激励-重标定”机制显式建模通道依赖关系,有效增强车道线相关特征响应并抑制了背景噪声。
3) 在Apollo-Sim3D数据集上的实验证明,本文方法在复杂挑战场景下的车道线坐标检测回归精度及稳定性均取得显著提升。
实际道路场景具有高度动态性,受光照变化、遮挡及恶劣天气等因素影响,传统二维车道线检测方法仅能在图像平面提供像素级位置信息,缺乏真实的三维几何信息,导致其在坡度路面等复杂场景中性能不佳。为满足自动驾驶对高精度三维感知的需求,主要借助单目视觉、3D激光雷达及多传感器融合获取所需的三维信息[1]。
单目三维车道线检测旨在从单帧二维图像中重构缺失的深度信息。根据空间表征方式的不同,现有算法主要可分为基于BEV和不依赖BEV两类[1]。
此类方法通过相机内外参将前视特征重投影至俯视图平面,使非线性的三维几何结构在BEV空间内呈现为规则的二维布局[2]。3DLaneNet[3]作为开创性工作,首次构建图像视角与BEV视角的双分支网络并引入逆透视映射实现特征域转换,以锚点形式输出连续三维车道线。随后,Gen-Lane Net[4]通过解耦图像语义分割与几何推理,并以几何引导锚点替代稠密标签,降低了对三维标注的依赖。为进一步提升复杂形状的表达能力,3D-LaneNet+[5]引入半局部无锚点表示,而PersFormer[6]将Transformer融入特征变换,利用相机参数与前视图局部区域生成表达力更强的BEV特征。此外,BEV-LaneDet[7]则提出虚拟相机机制以有效规避几何畸变,CLGO[8]则通过几何先验约束估计相机姿态,确保投影的一致性。Chen等[9]提出了一种分解式交叉注意力机制,通过统一建模BEV表示与车道线表示,实现二者同步学习,从而兼顾精度与计算效率。针对动态场景,Bi2Lane[10]利用双向时序优化进一步缓解了遮挡导致的信息缺失。
相比于依赖显式BEV表征的方法,BEV-free方案侧重前视特征与三维空间之间的直接交互。SALAD[11]证明了在无需相机参数的前提下,借助于伪深度图可实现从二维平面到三维空间的坐标恢复。PVALane[12]则利用先验引导生成稀疏锚点进行特征对齐。另一类趋势是直接在三维空间建模,如CurveFormer[13]将车道线表示为有序三维锚点构成的曲线查询,通过Transformer迭代优化预测结果。随后,Anchor3DLane[14]提出以三维锚点 (按俯仰/偏航定义的射线)作为结构化表示,并将其投影到前视图上以进行分类与回归,在此基础上LATR[15]进一步将查询机制细化为“点级”与“车道级”,并引入三维动态地面位置嵌入与跨注意力结构融合多尺度语义与空间定位信息。
三维激光雷达 (LiDAR)通过测量脉冲飞行时间获取高精度点云,不受光照变化影响,具有极佳的几何精度与鲁棒性[16]。Huang等[17]通过利用RANSAC算法结合点密度分类可实现结构化道路边界的提取。然而,LiDAR设备成本高昂且对路面反射特性敏感。
为综合视觉的语义优势与雷达的几何优势,研究者们提出了多模态融合策略。M2-3DLaneNet[18]在BEV空间内融合图像特征与雷达深度信息,提升了远距离定位精度。LiLaDet[19]将点云几何特征嵌入BEV表征实现高效融合,DV-3DLane[20]则通过双视图融合和三维可变形注意力机制,进一步增强了模型在动态和复杂场景下的鲁棒性。
总体来看,LiDAR可提供高精度三维信息,多模态融合方法又兼顾几何与语义,但受成本与计算复杂度限制,单目方法仍是目前研究主流。随着硬件性能提升和LiDAR成本下降,多模态融合方法有望在未来获得更广泛地应用。
基于锚点的车道线检测方法的核心思想在于:在图像感兴趣区域内预定义一组具有固定几何含义的锚点,将连续的车道线结构建模问题转化为对离散锚点属性的联合回归与分类问题,从而显著降低建模复杂度[21]。该思想已在3D-LaneNet[3]中得到验证,但其直接在图像平面进行三维回归易受视角变化与透视畸变影响,导致几何稳定性不足。为此,本文采用Gen-Lane Net[4]提出的几何引导锚点表示,通过显式几何建模在虚拟俯视视角 (BEV)中构建锚点坐标系,以削弱视角变化对空间推理的干扰。如图1所示,定义相机中心C及车辆坐标系 (x,y,z)。其中相机高度为h,俯仰角为$ \theta $,并以地面为参考平面建立相机与车辆坐标之间的几何关系。
如图2左所示,在该几何约束下,相机中心C、三维车道点P在虚拟俯视图中的投影点$ \mathit{P}' $满足共线关系。由相似三角形原理,可推导出从BEV平面到车辆坐标系的显式几何变换关系:$ x=x_{\mathrm{v}}\cdot\left(1-{\textit{z}}/h\right) $;$ y=y\mathrm{_v}\cdot\left(1-{\textit{z}}/h\right) $。该变换表明,在已知BEV平面位置与高度信息的条件下可恢复唯一对应的三维空间坐标。
基于此,本文在BEV中空间中沿纵向预定义一组等间距锚线$ {y}_{1},{y}_{2},\cdots ,{y}_{K} $。如图2右所示,在每条锚线上设置锚点,网络预测每个锚点对应的几何属性$ \{ ({x}_{k},{{\textit{z}}}_{k},{v}_{k})\}_{k=1}^{K} $。其中$ {x}_{k} $表示锚点的水平偏移量;$ {{\textit{z}}}_{k} $为该位置的真实路面高度;$ {v}_{k} $为可见性指示变量。最终,一条完整的三维车道线可表示为$ \{ ({x}_{k},{{\textit{z}}}_{k},{v}_{k})\}_{k=1}^{K} $及其存在概率p,即$ Anchor_{\mathrm{dim}}=3K+1 $。该锚点表示法通过显式引入高度与可见性约束,有效缓解了车道线的起伏和遮挡条件下的三维回归不稳定问题,并为后续端到端学习提供了结构化、物理一致的几何先验。
本文提出的 HA-LaneNet (hybrid-encoder with pre-IPM channel attention)总体架构如图3所示,主要由混合编码器、基于IPM的几何映射模块、BEV特征融合模块以及车道线回归头构成。输入图像首先经混合编码器提取兼具多尺度语义与几何细节的特征表示,并通过通道注意力机制增强特征;随后通过几何变换将图像特征映射至BEV空间,实现跨视角特征对齐;在BEV空间内,多尺度特征进一步融合以增强结构一致性,最终由回归分支输出三维车道线表示。该端到端架构在保持几何约束一致性的同时,有效提升了复杂场景下三维车道线检测的稳定性与精度。
针对三维车道线检测任务对精细几何建模与高层语义理解的双重需求,本文设计了一种融合浅层与深层特征的混合编码器 (hybrid encoder),以实现几何与语义信息的协同表达。如图4所示,浅层分支采用VGG16[22]的前7层结构,通过连续$ 3\times 3 $卷积提取高分辨率的局部纹理与边缘信息,并在1/4与1/8的空间尺度处输出特征以保留细粒度几何结构。深层分支基于ResNet34[23],仅保留layer3与layer4的输出以增强高层语义表征能力。为实现跨层特征对齐,本文在融合路径中引入$ 1\times 1 $卷积进行通道映射,并结合Network-in-Network[24] (NIN)结构完成通道维度重整与特征压缩。该混合编码策略在保证空间分辨率的同时增强了语义判别性,为后续IPM投影与BEV特征构建提供了一致且可投影的特征基础。
为进一步提升编码特征的表达质量,本文在混合编码器的高层特征输出处引入由Hu等[25]提出的基于通道注意力SE模块,对语义特征进行自适应重标定。
如图5所示,将SE模块嵌入ResNet的layer3、layer4输出之后。针对输入特征图$ X $,经卷积变换$ F_{{\mathrm{tr}}}\left(\cdot,\theta\right) $得到初始特征$ U $。随后,通过压缩操作$ F_{{\mathrm{sq}}}\left(\cdot\right) $ (即全局平均池化)对$ U $的每个通道进行空间维度压缩,获得$ 1\times1\times C $的通道级统计描述;接着进行激励操作$ F_{{\mathrm{ex}}}\left(\cdot,W\right) $,利用包含参数$ W $且降维率$ r=16 $的两层全连接网络建立通道依赖,并通过Sigmoid函数生成通道权重;最后,通过尺度重标定操作$ F_{{\mathrm{scale}}}\left(\cdot\right) $,利用该权重对原特征$ U $逐通道重加权,得到增强特征$ \tilde{X} $。该机制能够显式强化与车道线结构相关的判别性响应,同时抑制冗余背景,在不显著增加计算开销的前提下增强网络的特征选择能力。
为同时监督车道线的存在性、几何结构与可见性,本文采用多任务联合优化策略构建端到端损失函数。设模型对第i条车道的预测结果为$ {\hat{A}}_{i}=\left\{{{{{{{{\hat{p}}_{i}},\hat{v}}_{i}},\hat{x}}_{i}},\hat{{\textit{z}}}}_{i}\right\} $,对应的真值标签为$ \mathcal{A}_i^{{\mathrm{g}} {\mathrm{t}}}=\{{p}_{i},{v}_{i},{x}_{i},{{\textit{z}}}_{i}\} $。其中$ {p}_{i}\in 0{,}1 $表示车道线存在标签,$ {v}_{i} $为逐点的可见性掩膜。总损失函数定义为分类损失$ {L}_{\rm{cls}} $、几何回归损失$ {L}_{\rm{geo}} $与可见性损失$ {L}_{\rm{vis}} $的加权求和。
$\begin{split} L=&-\sum_i \underbrace{\ell_{{\mathrm{b c e}}}\left (\hat{p}_i, p_i\right)}_{L_{{\mathrm{c l s}}}}+\sum_i p_i \\&\cdot\left (\underbrace{\left|v_i \cdot\left (\hat{x}_i-x_i\right)\right|_1+\left|v_i \cdot\left (\hat{{\textit{z}}}_i-{\textit{z}}_i\right)\right|_1}_{L_{ \mathrm{g eo}}}+\underbrace{\left|\hat{v}_i-v_i\right|_1}_{L_{{\mathrm{v i s}}}}\right)\;,\end{split} $
其中:$ {L}_{\rm{cls}} $采用二元交叉熵监督车道线存在性;$ {L}_{\rm{geo}} $基于L1范数对车道线在BEV坐标系下的水平偏移x与高度z进行回归,并显式引入$ {v}_{i} $以避免不可见点对几何建模的干扰;$ {L}_{\rm{vis}} $则用于约束模型对车道线采样点可见性的预测准确性。该联合损失设计在保证结构连续性的同时提升了模型对遮挡与远距离车道线的感知能力。
为验证本文算法的有效性,实验采用与Gen-Lane Net[4]一致的Apollo-Sim3D仿真数据集。该数据集包含10500帧高分辨率 ($ 1920\;{\mathrm{pixel}}\times 1080\;{\mathrm{pixel}} $) RGB图像及对应的三维车道线标注。数据集涵盖了标准场景、罕见场景 (如急转弯、陡坡)以及视觉变化场景 (如极端光照、雨雾天)三大类别,各场景样本分布如表1所示。本文采用$ {F}_{1} $分数及平均精度 (average precision,AP)作为核心性能指标。$ {F}_{1} $定义为:
$ F_1=\dfrac{2 \times { P \times R }}{ { P }+ { R }} \;, $
$ { P }=\dfrac{T P}{T P+F P} \;, $
$ { R }=\dfrac{T P}{T P+F N}\;, $
其中:真正例 ($ TP $)的判定遵循Gen-Lane Net设定三维匹配流程:对预测值与真值 (GT)进行等间距采样,计算对应采样点间的三维欧氏距离。当采样点中满足距离阈值$ {d}_{{\mathrm{max}}}=1.5\;\text{m} $的比例不低于75%时,该预测车道线被判定为$ TP $。通过对P-R曲线进行积分计算得到$ AP $:
$ {AP}=\sum \limits_{k=1}^{N}P\left (k\right)\cdot \Delta R\left (k\right)\;, $
式中:$ P\left (k\right) $为第k个预测的精确率;$ \Delta R\left (k\right) $为召回率增量;$ N $为预测结果数量。
此外,为进一步评估几何回归的精度,本文分别统计了近距离 ($ 0\leqslant y\leqslant40\ \mathrm{m} $)与远距离 ($ 40\ \mathrm{m}\leqslant y\leqslant100\mathrm{\ m} $)范围内的横向误差$ {E}_{X} $与纵向误差$ {E}_{Z} $。
本实验模型基于PyTorch框架开发,硬件环境为NVIDIA RTX 4090D GPU (24 GB),操作系统为Ubuntu 22.04,开发环境为cpython 3.12,编辑器为Neovim。在数据预处理阶段,将输入图像统一调整为$ 360\;{\mathrm{pixel}}\times 480\;{\mathrm{pixel}} $。模型在BEV空间构建了尺寸为$ 208\;{\mathrm{pixel}}\times 108\;{\mathrm{pixel}} $的空间网格,其对应的三维物理范围设为:x轴:$ \left[\begin{matrix}-10{,}10\\ \end{matrix}\right] $ m;y轴:(3, 5, 10, 15, 20, 30, 40, 50, 65, 80, 100) m;z轴:$ \left[\begin{matrix}-1{,}1\\ \end{matrix}\right] $ m。为降低远景区域的透视畸变影响,设置前向投影参考高度$ {Y}_{{\mathrm{ref}}}=5 $ m。网络权重采用正态分布随机初始化,使用Adam优化器进行端到端训练。初始学习率为$ 2\times {10}^{-4} $,并引入余弦退火的策略进行动态调整,公式如下:
$ {\eta }_{t}={\eta }_{{\mathrm{min}}}+\dfrac{1}{2}\left ({\eta }_{0}-{\eta }_{{\mathrm{min}}}\right)\left[1+\cos \left (\dfrac{\pi t}{{\tau }_{{\mathrm{max}}}}\right)\right] \;, $
其中:$ {\eta }_{t} $为当前学习率;t为当前epoch,总训练周期$ {\tau }_{{\mathrm{max}}}=60 $。训练过程中批大小设为8,最终学习率衰减至$ 1\times {10}^{-6} $。
为了系统验证混合编码器、注意力机制以及学习策略对模型性能的贡献,本文基于视觉变化场景构建了消融实验方案。实验以几何锚点结构作为基线,通过控制变量法对各核心组件进行组合测试。评估指标涵盖的$ AP $与$ {F}_{1} $分数及坐标预测误差,$ AP $与$ {F}_{1} $用于衡量检测精度,值越大说明性能越好;而横向误差$ {E}_{X} $与纵向误差$ {E}_{Z} $则数值越低表示预测越精确。核心组件消融结果如表2所示。
基线网络的$ AP $为84%,通过对比发现,单独引入混合编码器或注意力机制时,模型性能虽有小幅提升,但均未达到显著的增益效果,这表明架构调整需配合特定的学习策略方能充分释放其表征潜力。当引入相应的学习策略后,注意力机制的$ AP $从86.8%跃升至89.8%,体现了训练策略对结构效能的关键驱动作用。进一步的组合测试显示,同时集成混合编码器与注意力机制并辅以优化策略时,模型达到了91.5%的最高的$ AP $与89.5%的$ {F}_{1} $分数。这一结果有力地验证了混合编码器提供的层次化语义特征与注意力机制对关键信息的强化表达具有显著的互补性,两者的耦合能够系统性地增强模型在复杂视觉环境下的检测精度。
降维率r是通道注意力机制中平衡计算开销与特征表达能力的核心参数,本文对此进行了详细的参数敏感性实验,如表3所示。在基于基线网络采用的VGG模型中,随着r的增大,模型性能呈现出先升后降的趋势:当$ r=8 $时取得阶段性最优表现 ($ AP $为87.7%),而当r进一步增至32时,$ AP $出现剧烈下滑,这归因于过高的压缩率导致了关键语义信息的丢失。相比之下,引入混合编码器的模型表现出更强的稳健性与更高的精度上限。实验数据表明,当$ r=16 $时,HA-Lane Net模型在保持高精度的同时,横向预测误差$ E_{\mathit{X}}/F $降至最低值0.0610,实现了检测效率与精细化表征的最佳平衡。综上所述,降维率的选择需权衡冗余信息抑制与特征完整性保留,本文最终选取$ r=16 $作为HA-Lane Net架构的最优降维参数,以最大化提升模型在复杂场景下的坐标预测准确性。
在合成数据集上完成模型训练后,对模型进行了全面测试,并与本文主要参考方法进行对比分析。实验采用合成数据集常用的多项评价指标,包括$ AP $、$ {F}_{1} $等,结果如表4与表5所示。其中,“L”表示本文采用的新学习策略,“T”则表示将混合编码器中的ResNet34替换为ResNet18。$ AP $与$ {F}_{1} $用于衡量检测精度,值越大说明性能越好;而横向误差$ {E}_{X} $与纵向误差$ {E}_{Z} $则数值越低表示预测越精确。由于合成数据集包含三类不同场景,本文主要以视觉变化最大的场景表5作为评估基准。从多组对比实验结果可见,本文方法较3DLaneNet[3]与Gen-Lane Net[4]均表现出显著提升。然而,与部分当前先进方法 (如CLGo[10])相比,整体性能相近,但在某些特定场景下略有不足。出现这一差异的主要原因在于,本文采用的逆透视映射依赖地面平坦性假设,而实际环境中地面往往存在起伏或非理想几何形态,可能导致特征投影精度下降,进而影响最终检测效果。
从表4可以看出,在标准场景与罕见场景中,本文方法相较于基线模型均取得了约2%~4%的性能提升,且在罕见场景中表现更为突出。在表5的视觉变化场景对比中,模型性能提升可达约7.5%,表明本文方法在复杂视觉条件下具有更强的鲁棒性。
为验证HA-Lane Net在车载嵌入式平台上的工程化部署潜力,本节对模型的计算效率进行了定量评估。评估指标包括推理速度 (frames per second, FPS)、参数量 (parameters, Params)以及浮点运算量 (floating point operations per second, FLOPs)与乘加运算量 (multiplication and accumulation operations, MACs)。实验在NVIDIA RTX 4090D平台上进行,对比对象包括基线模型以及目前先进的基于Transformer的模型LATR[15]。如表6所示,本方法在Apollo-Sim3D数据集的三个场景下的推理速度上表现优异,平均达到111.62 f/s,远高于LATR的35.53 f/s,完全满足自动驾驶系统实时感知的需求 ($ \gt 30 $ f/s)。
在计算开销方面,如表7所示,虽然由于混合编码器的引入使得参数量 (28.88 M)略高于基线模型 (20.60 M),但其单帧MACs从基线的60.523 G优化至52.308 G,降低了约13.6%。实验结果表明,本方法通过混合编码架构优化了特征提取流程,在不牺牲运行速度的前提下,实现了更高效的计算开销,在保持高帧率的同时优化了单位推理成本。
为直观验证SE模块对关键特征的增强作用,本节利用通道维$ {L}_{2} $能量投影,对网络特征图进行了可视化处理。记模块前后的特征图为:${F}\in {\mathbb{R}}^{C\times H\times W} $。其中,C表示通道数,$ H,W $表示空间分辨率。令$ F_c\left(x,y\right) $表示第c个通道在空间位置$ \left (x,y\right) $处的激活响应。采用如下公式对通道维做平方和,得到每个空间位置的总体响应强度。
$ H (x,y)=\sum \limits_{c=1}^{C}{F}_{c} (x,y{)}^{2}\;. $
如图6所示,图中上半部分为视觉变化场景,从左至右依次为:基线模型与添加SE后的能量图,Top-5%注意力迁移图以及叠加在原图上的示意图。在迁移图与叠加图中,红色表示引入SE后新增的关注区域,蓝色表示基线原有的关注区域,紫色则代表两者的共同关注区域。在基线模型中,由于缺乏通道间的依赖建模,高响应区域往往弥散在路面阴影、路缘等非目标区域。而引入SE注意力机制后,特征能量显著向车道线物理结构汇聚,背景噪声干扰得到了有效抑制。通过右侧的响应迁移图,可以清晰观察到,红色区域精准覆盖了远端模糊的车道走势。这表明注意力机制通过自适应重标定通道权重,在BEV空间映射前实现了特征的“提纯”,为后续的三维坐标回归奠定了可靠的特征基础。而图6下半部分展示了在罕见场景中基线模型与HA-Lane Net的特征可视化对比。由于不同网络结构的高层特征分辨率存在差异,本文通过特征分辨率匹配算法,选取面积与长宽比最接近的层与混合编码器的高层特征进行公平对齐。实验结果进一步验证了通道注意力机制在复杂场景下的稳健性与有效性。
图7-9展示了本文方法与基线模型3DLaneNet_ext,以及Gen-Lane Net[4]在多种复杂及罕见场景下的检测性能对比,其中蓝色曲线表示真实标注 (GT),红色曲线为模型预测结果。评测场景涵盖了夜间低照度、跨时段光照显著变化以及坡道地形等典型环境。在夜间低照度场景 (图7)中,由于输入图像亮度较低且纹理信息匮乏,伴随高光与噪声干扰,基线模型在弯道路段表现出明显的曲率拟合不足与预测漂移。尽管Gen-Lane Net能够恢复部分车道结构,但仍存在拓扑不连续现象。相比之下,本方法凭借混合编码器对微弱几何纹理的提取能力,有效避免了基线模型的曲率漂移,在直道与弯道场景下均能稳定拟合真实标注,预测曲线连续且平滑,展现出在低光照条件下更优异的特征保持能力与几何推理能力。在跨时段变光照弯道场景 (图8)中,上午与下午的光照差异导致阴影形态变化、局部过曝及路面反射特性显著不同。基线模型在高亮区域出现轻微定位偏移,而Gen-Lane Net在阴影边界附近易产生结构断裂。与之形成对比,本方法在波动光照条件下均保持了稳定的车道线走势估计,预测曲线与GT的匹配精度更高,验证了模型具备良好的光照鲁棒性与跨时段泛化能力。针对坡道路面场景 (图9)中,路面坡度引发了显著的透视形变,增加了车道线三维空间结构表征的难度。基线模型与Gen-Lane Net均出现了不同程度的结构扭曲或曲率偏差,而本方法能够准确恢复车道线的三维空间形态,在重建视图中呈现出更高的平滑性与几何一致性,说明本方法在面临纵向坡度与复杂三维几何变换时,仍具备稳建的空间建模能力。
综上所述,图7-9的可视化定性分析表明,本方法在各类极端场景中均表现出更高的稳定性、鲁棒性以及车道结构恢复精度,较基线方法具有显著的技术优势。
本文针对单目三维车道线检测中编码器语义表征能力受限及特征通道冗余的问题,提出了一种轻量化的网络结构改进方法。通过构建融合VGG与ResNet特征提取优势的混合编码器,实现了浅层几何纹理信息与深层语义特征的协同建模;同时,在深层特征输出阶段引入SE模块,对通道特征进行自适应重标定,从而增强车道相关特征响应并抑制背景噪声干扰。实验结果表明,在不显著增加计算开销的条件下,该方法能够提升BEV空间特征表示能力并提高三维回归的稳定性。在Apollo-Sim3D数据集上的实验结果进一步验证了所提出方法在复杂场景下的有效性和鲁棒性,为后续多模态感知融合与时序建模研究提供了参考。

参考文献 引证文献
排序方式:
1
Luo X, Huang Y P, Cui J Y, et al. Deep learning-based lane detection for intelligent driving: a comprehensive survey of methods, datasets, challenges and outlooks[J]. Neurocomputing, 2025, 650: 130795.
2
He X, Guo H Y, Zhu K, et al. Monocular lane detection based on deep learning: a survey[EB/OL]. arXiv: 2411.16316, 2024. https://doi.org/10.48550/arXiv.2411.16316.
3
Garnett N, Cohen R, Pe'er T, et al. 3D-LaneNet: end-to-end 3D multiple lane detection[C]//Proceedings of 2019 IEEE/CVF International Conference on Computer Vision, Seoul, Korea (South), 2019: 2921‒2930. https://doi.org/10.1109/ICCV.2019.00301.
4
Guo Y L, Chen G, Zhao P T, et al. Gen-LaneNet: a generalized and scalable approach for 3D lane detection[C]//Proceedings of the 16th European Conference on Computer Vision, Glasgow, UK, 2020: 666‒681. https://doi.org/10.1007/978-3-030-58589-1_40.
5
Efrat N, Bluvstein M, Oron S, et al. 3D-LaneNet+: anchor free lane detection using a semi-local representation[EB/OL]. arXiv: 2011.01535, 2020. https://doi.org/10.48550/arXiv.2011.01535.
6
Chen L, Sima C H, Li Y, et al. PersFormer: 3D lane detection via perspective transformer and the openlane benchmark[C]//Proceedings of the 17th European Conference on Computer Vision, Tel Aviv, Israel, 2022: 550‒567. https://doi.org/10.1007/978-3-031-19839-7_32.
7
Wang R H, Qin J, Li K Y, et al. BEV-LaneDet: an efficient 3D lane detection based on virtual camera via key-points[C]//Proceedings of 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Vancouver, Canada, 2023: 21611‒21620. https://doi.org/10.1109/CVPR52729.2023.00103.
8
Liu R J, Chen D P, Liu T, et al. Learning to predict 3D lane shape and camera pose from a single image via geometry constraints[C]//Proceedings of the 36th AAAI Conference on Artificial Intelligence, 2022: 1765‒1772. https://doi.org/10.1609/aaai.v36i2.20069.
9
Chen Z Y, Smith-Miles K, Du B, et al. An efficient transformer for simultaneous learning of BEV and lane representations in 3D lane detection[EB/OL]. arXiv: 2306.04927, 2023. https://doi.org/10.48550/arXiv.2306.04927.
10
Li C X, Hu Y H, Zheng Z W, et al. Bi2Lane: Bi-Directional temporal refinement with Bi-level feature aggregation for 3D lane detection[C]//Proceedings of 2024 IEEE International Conference on Robotics and Automation, Yokohama, Japan, 2024: 2894‒2900. https://doi.org/10.1109/ICRA57147.2024.10610794.
11
Yan F, Nie M, Cai X Y, et al. ONCE-3DLanes: building monocular 3D lane detection[C]//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition, New Orleans, USA, 2022: 17122‒17131. https://doi.org/10.1109/CVPR52688.2022.01663.
12
Zheng Z W, Zhang X M, Mou Y Q, et al. PVALane: prior-guided 3D lane detection with view-agnostic feature alignment[C]//Proceedings of the 38th AAAI Conference on Artificial Intelligence, Vancouver, Canada, 2024: 7597‒7604. https://doi.org/10.1609/aaai.v38i7.28592.
13
Bai Y F, Chen Z R, Fu Z J, et al. CurveFormer: 3D lane detection by curve propagation with curve queries and attention[C]//Proceedings of 2023 IEEE International Conference on Robotics and Automation, London, United Kingdom, 2023: 7062‒7068. https://doi.org/10.1109/ICRA48891.2023.10161160.
14
Huang S F, Shen Z W, Huang Z J, et al. Anchor3DLane: learning to regress 3D anchors for monocular 3D lane detection[C]//Proceedings of 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Vancouver, Canada, 2023: 17451‒17460. https://doi.org/10.1109/CVPR52729.2023.01674.
15
Luo Y R, Zheng C D, Yan X, et al. LATR: 3D lane detection from monocular images with transformer[C]//Proceedings of 2023 IEEE/CVF International Conference on Computer Vision, Paris, France, 2023: 7907‒7918. https://doi.org/10.1109/ICCV51070.2023.00730.
16
Bi J P, Song Y C, Jiang Y H, et al. Lane detection for autonomous driving: comprehensive reviews, current challenges, and future predictions[J]. IEEE Trans Intell Transp Syst, 2025, 26(5): 5710−5746.
17
Huang J, Choudhury P K, Yin S, et al. Real-time road curb and lane detection for autonomous driving using LiDAR point clouds[J]. IEEE Access, 2021, 9: 144940−144951.
18
Luo Y R, Yan X, Zheng C D, et al. M2-3DLaneNet: exploring multi-modal 3D lane detection[EB/OL]. arXiv: 2209.05996, 2022. https://doi.org/10.48550/arXiv.2209.05996.
19
Zhao R K, Heng Y W, Wang H, et al. Advancements in 3D lane detection using LiDAR point clouds: from data collection to model development[C]//Proceedings of 2024 IEEE International Conference on Robotics and Automation, Yokohama, Japan, IEEE, 2024: 5382‒5388. https://doi.org/10.1109/ICRA57147.2024.10610087.
20
Luo Y R, Cui S G, Li Z. Dv-3DLane: end-to-end multi-modal 3D lane detection with dual-view representation[C]//Proceedings of the 12th International Conference on Learning Representations, Vienna, Austria, 2024.
21
罗鑫, 黄影平, 梁振明. 轴注意力引导的锚点分类车道线检测[J]. 光电工程, 2023, 50(7): 230079.
Luo X, Huang Y P, Liang Z M. Axial attention-guided anchor classification lane detection[J]. Opto-Electron Eng, 2023, 50(7): 230079.
22
Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[C]//Proceedings of the 3rd International Conference on Learning Representations, San Diego, USA, 2015.
23
He K M, Zhang X Y, Ren S Q, et al. Deep residual learning for image recognition[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 770‒778. https://doi.org/10.1109/CVPR.2016.90.
24
Lin M, Chen Q, Yan S C. Network in network[Z]. arXiv: 1312.4400, 2014. https://doi.org/10.48550/arXiv.1312.4400.
25
Hu J, Shen L, Sun G. Squeeze-and-excitation networks[C]//Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 7132‒7141. https://doi.org/10.1109/CVPR.2018.00745.
26
Li C G, Shi J, Wang Y, et al. Reconstruct from top view: a 3D lane detection approach based on geometry structure prior[C]//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition, New Orleans, USA, 2022: 4369‒4378. https://doi.org/10.1109/CVPRW56347.2022.00483.
2026年第53卷第6期
PDF下载
93
41
引用本文
BibTeX
文章信息
doi: 10.12086/oee.2026.250359
  • 接收时间:2025-11-21
  • 首发时间:2026-10-22
  • 出版时间:2026-06-25
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2025-11-21
  • 修回日期:2026-03-13
  • 录用日期:2026-03-20
基金
作者信息
    上海理工大学光电信息与计算机工程学院,上海 200093

通讯作者:

参考文献
分享链接
https://castjournals.cast.org.cn/joweb/oee/CN/10.12086/oee.2026.250359
分享至
全文二维码

扫描看全文

引用本文
BibTeX
本文的引用情况
2种不同金属材料的力学参数
科
Family
属数
Number of
genus
种数
Number of
species
占总种数比例
Percentage of
total species (%)
属
Genus
种数
Number of
species
占总种数比例
Percentage of total
species (%)
鹅膏菌科Amanitaceae 2 11 5.26 鹅膏菌属 Amanita 10 4.78
小菇科 Mycenaceae 2 12 5.74 丝盖伞属 Inocybe 5 2.39
多孔菌科 Polyporaceae 8 14 6.70 蜡蘑属 Laccaria 5 2.39
红菇科 Russulaceae 3 23 11.00 小皮伞属 Marasmius 6 2.87
小菇属 Mycena 11 5.26
光柄菇属 Pluteus 5 2.39
红菇属 Russula 17 8.13
栓菌属 Trametes 5 2.39
关闭全屏