计算机与现代化 ›› 2024, Vol. 0 ›› Issue (07): 87-92.doi: 10.3969/j.issn.1006-2475.2024.07.013

• 算法设计与分析 • 上一篇    下一篇

基于改进SCNN网络的车道线检测算法


  

  1. (1.无锡学院电子信息工程学院,江苏 无锡 214105; 2.南京信息工程大学电子信息工程学院,江苏 南京 210044)
  • 出版日期:2024-07-25 发布日期:2024-08-08
  • 基金资助:
    国家自然科学基金青年资助项目(62106111); 2021年无锡学院第二批产学合作协同育人项目(202102563020)

Lane Line Detection Algorithm Based on Improved SCNN Network

  1. (1. School of Electronic and Information Engineering, Wuxi University, Wuxi 214105, China; 2. School of Electronic and
    Information Engineering, Nanjing University of Information Science and Technology, Nanjing 210044, China)
  • Online:2024-07-25 Published:2024-08-08

摘要: 车道线检测是车道保持系统、车道偏离告警系统的实现前提。为了进一步提升检测的精度,本文将深度学习与车道线检测结合提出一种融合改进的SCNN车道线识别算法。该方法以改进的SCNN网络为基础,引入PSA注意力模块,并将其与VGG(Visual Geometry Group) 网络相结合提出融合上下文信息的车道线识别网络VGG-K,帮助每层中行和列像素之间进行消息传递,增强其对连续变换目标的识别能力,然后利用二次曲线模型拟合得到最终的车道线检测结果。将改进的模型在CULane数据集上进行测试,结果表明:该方法在正常场景下的综合评价指标F1数值达到92.1,恶劣场景下的数值达到75.3,与其他模型对比可知,本文算法检测能力显著提升,对于多种复杂状况下的车道线具有更好的识别效果。

关键词: 车道线检测, 深度学习, 目标识别, 特征提取

Abstract:  Lane line detection is the prerequisite for the realization of lane keeping system and lane departure warning system. In order to further improve the accuracy of detection, combining deep learning and lane line detection, this paper proposes an improved SCNN lane line recognition algorithm. Based on the improved SCNN network, this method introduces PSA attention module, and combines it with VGG (Visual Geometry Group) network to propose a lane line recognition network called VGG-K. This network fuses context information, helps the message transmission between row and column pixels in each layer, enhances its recognition ability for continuous transformation targets, and then uses quadric model fitting to obtain the final lane line detection result. The improved model is tested on the dataset CULane. Training results show that the comprehensive evaluation index F1 value of the method reaches 92.1 in normal scenarios and 75.3 in harsh scenarios. Compared with other models, the detection ability of the proposed algorithm is significantly improved, and the proposed algorithm has a better recognition for lane lines under various complex conditions.

Key words: lane line detection, deep learning, target recognition, feature extraction

中图分类号: