To address the detection challenges of on-board monocular vision in complex scenarios such as perspective distortion, illumination variation and lane marking wear, as well as the limited computing power of embedded platforms, this paper proposes a lightweight detection method based on geometric structure parsing of lane markings. Free from reliance on complex models and large-scale annotated data, this proposed method explicitly models the geometric priors including parallelism and continuity of lane markings, and establishes a complete pipeline consisting of adaptive region of interest extraction, directional edge enhancement, improved inverse perspective transformation, reliable region selection and dynamic polynomial fitting to achieve the enhancement and stable extraction of lane features. The verification experiments are carried out on a self-built multi-scenario dataset covering straight roads, curved roads, urban roads, night, rainy days and fisheye distortion. The precision, recall and F1-score of the proposed method reach 0.953, 0.948 and 0.950 respectively, which are significantly better than those of traditional methods. Meanwhile, this method exhibits stronger robustness and higher accuracy than two advanced segmentation-based methods. The average processing time is 24.2 ms per frame without GPU acceleration, meeting the real-time requirement of vehicle-mounted applications. The research conclusions provide reliable environmental perception support for advanced driver assistance systems and autonomous driving.
针对上述挑战,现有车道检测方法主要分为传统图像处理与深度学习两大类。传统方法通常依赖车道线的低层视觉特征和先验几何模型。例如,霍夫变换及其改进方法被广泛用于直线车道检测,李荣彬[5]通过结合车道线平行与长度特性优化霍夫变换,提升结构化道路中的检测精度,但其对弯道及复杂曲率车道的适应性有限。为克服透视畸变影响,逆透视变换被引入以获取车道线的鸟瞰视图。刘景锋等[6]提出改进的逆透视映射(inverse perspective mapping,IPM)算法,结合直方图粗定位与滑动窗口拟合实现车道检测,但在剧烈光照变化下稳定性不足。此外,部分研究侧重于自适应感兴趣区域与特征融合以提升鲁棒性。刘丹萍[7]提出的感兴趣区域(region of interest,ROI)自适应定位方法,结合梯度引导与浅层网络学习,增强复杂场景下的检测能力,但仍需较多人工干预与参数调整。近年来,一些研究尝试将图像分割技术引入车道检测。王畅等[8]提出一种融合图像分割与变尺度窗口的车道线距离检测方法,利用分割结果引导滑动窗口,提高距离测量的准确性。黄艳国等[9]优化阈值分割结合滑动窗口,这些方法在特定场景下效果良好,但分割过程仍易受光照和阴影干扰。Zeng等[10]将IPM与卡尔曼滤波结合用于车道跟踪,但模型灵活性不足。总体而言,传统方法计算量小、可解释性强,但在复杂场景下的泛化能力与精度难以兼顾。
本文提出一种基于车道线几何结构解析的车载单目车道检测方法,通过自适应ROI提取-几何结构约束-改进逆透视变换-可靠区域选择-动态多项式拟合流程,显式建模车道线平行性与连续性先验,无需标注数据与GPU加速。方法将方向性边缘增强、改进IPM与动态拟合结合,有效消除透视畸变,提升破损车道线检出能力。多场景实验表明:本文方法精确率、召回率、F1值分别达0.953、0.948、0.950,显著优于传统方法;夜间与雨天场景 F1 值达0.941,优于两类分割对比方法;无GPU加速下单帧耗时24.5 ms,满足车载实时检测需求。低纹理路面与严重遮挡场景下鲁棒性不足,后续将融合多传感器时空跟踪,并拓展车道线类型识别功能。
YeXinchen, WangXuesong, TianWei. Readiness evaluation on freeways in plain regions for automated vehicles based on lane detection performance[J]. China Journal of Highway and Transport, 2024, 37(11): 220-234.
[4]
PittnerM, JanaiJ, ConduracheA P. LaneCPP: continuous 3D lane detection using physical priors[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. June 16-22, 2024, Seattle, WA, USA. IEEE, 2024: 10639-10648.
MaQinglu, ZhangLi, MaLian, et al. AI adaptive recognition method for tunnel lane lines in autonomous driving[J]. Journal of Southwest Jiaotong University, 2026,61(2):452-465.
LiuJingfeng, LiYanliang, JiangKui, et al. Lane detection algorithm based on inverse perspective transform[J]. Intelligent Computer and Applications, 2021, 11(3): 109-112.
WangChang, LiYonghang, ZhangKaichao, et al. Lane line distance detection based on fusion segmentation and a variable-scale window[J]. China Journal of Highway and Transport, 2023, 36(7): 212-222.
HuangYanguo, ZhongYong, RaoZehao. Lane departure warning based on optimized threshold segmentation of maximum inter-class variance and sliding window method[J]. Automobile Technology, 2023(6): 9-16.
[17]
ZengL J, AoB Q, LiQ. Lane-line detection based on inverse perspective mapping and Kalman filter[J]. Advances in Engineering Technology Research, 2024, 11(1): 1.
[18]
LeeS, KimJ, YoonJ S, et al. VPGNet: vanishing point guided network for lane and road marking detection and recognition[C]//2017 IEEE International Conference on Computer Vision. October 22-29, 2017, Venice. IEEE, 2017: 1965-1973.
[19]
TabeliniL, BerrielR, PaixoT M, et al. PolyLaneNet: lane estimation via deep polynomial regression[C]//2020 25th International Conference on Pattern Recognition. January 10-15, 2021, Milan, Italy. IEEE, 2021: 6150-6156.
[20]
GuoA W, LiM Y, LiuD K, et al. Lane detection in complex scenes based on ultra-fast structure-aware[J]. Highlights in Science, Engineering and Technology, 2024, 119: 783-791.
LouLu, HuZhenkun, WeiWenjie, et al. Monocular 3D lane detection for intelligent driving in complex road scenarios[J]. Application Research of Computers, 2025, 42(12): 3807-3814.
[25]
JainM. Lane line detection[J]. International Journal for Research in Applied Science and Engineering Technology, 2024, 12(4): 4932-4938.
LiuNa, DaiJiguang. A line segment sequence detection method for expressing image object edge[J]. Journal of Liaoning Technical University (Natural Science), 2024, 43(2): 208-216.
[28]
PengY, ChenZ, WuQ M J, et al. Traffic flow detection and statistics via improved optical flow and connected region analysis[J]. Signal, Image and Video Processing, 2018, 12(1): 99-105.
[29]
SadeliA F, LawandaI I. Recall, precision, and F-measure for evaluating information retrieval system in electronic document management systems (EDMS)[J]. Khizanah al-Hikmah: Jurnal Ilmu Perpustakaan, Informasi, dan Kearsipan, 2023, 11(2): 231-241.