摘 要：鉴于车道线检测在自动驾驶中占据重要地位，文章提出一种基于迁移学习的车道线检测方法，用 Darknet53 替代原 DeepLabv3+ 的特征提取网络。针对图片中车道线类与背景类分布极不平衡的问题，使用 Focal Loss 函数替代 CE Loss 函数。实验结果表明，该方法的检测效果比原网络好，平均交并比达到 76.95%，具有良好的准确性。
关键词：车道线检测；迁移学习；DeepLabv3+
DOI:10.19850/j.cnki.2096-4706.2021.20.021
中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2021）20-0082-05
A Lane Line Detection Method Based on Transfer Learning
WANG Zihao, LI Xiangjun
(Dalian Ocean University, Dalian 116023, China)
Abstract: Since lane line detection plays an important role in automatic driving, this paper proposes a lane line detection method based on transfer learning, replacing the original DeepLabv3+’s feature extraction network with Darknet53. To solve the problem that the distribution of lane line class and background class in the picture is extremely unbalanced, the Focal Loss function is used to replace the CE Loss function. The experimental results show that the detection effect of the method is better than that of the original network, and the average cross-merge ratio reaches 76.95%, which has good accuracy.
Keywords: lane line detection; transfer learning; DeepLabv3+
参考文献：
[1] 隋靓，党建武，王阳萍 . 基于分段切换模型的快速车道线检测 [J]. 计算机应用与软件，2017，34（8）：201-205.
[2] MA C，XIE M.A method for lane detection based on color clustering [C]//2010 Third International Conference on Knowledge Discovery and Data Mining. Phuket：IEEE，2010：200-203.
[3] 杨金鑫，范英，樊祺超，等 . 基于动态区域搜索框及K-means 聚类的三车道检测算法 [J]. 科学技术与工程，2019，19（27）：253-257.
[4] PIZZATI F，GARCÍA F. Enhanced free space detection in multiple lanes based on single CNN with scene identification [C]//2019 IEEE Intelligent Vehicles Symposium (IV). Paris：IEEE，2019：2536-2541.
[5] 丁冰，杨祖莨，丁洁，等 . 基于改进 YOLOv3 的高速公路隧道内停车检测方法 [J]. 计算机工程与应用，2021，57（23）：234-239.
[6] LEE S，KIM J，YOON J S，et al. Vpgnet:Vanishing point guided network for lane and road marking detection and recognition [C]//2017 IEEE International Conference on Computer Vision (ICCV).Venice：IEEE，2017：1965-1973.
[7] Deng J，DONG W，SOCHER R，et al. ImageNet：A largescale hierarchical image database [C]//IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway：IEEE，2009：248–255.
作者简介：王子豪（1996—），男，汉族，辽宁葫芦岛人，硕士研究生在读，研究方向：控制科学与控制理论、计算机视觉；李向军（1963—）女，汉族，辽宁大连人，教授，硕士生导师，博士，研究方向：控制科学与控制理论。