Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (15): 214-222.DOI: 10.3778/j.issn.1002-8331.2204-0330

• Graphics and Image Processing • Previous Articles     Next Articles

Lane Detection Algorithm Based on Introduction of Improved GAN Network in Night Vision Environment

LIU Yan, QIU Tiantian, XIAO Yanqiu, ZHU Fubao, WANG Jingwen   

  1. 1.College of Computer and Communication Engineering, Zhengzhou University of Light Industry, Zhengzhou 450001, China
    2.College of Mechanical and Electrical Engineering, Zhengzhou University of Light Industry, Zhengzhou 450001, China
  • Online:2023-08-01 Published:2023-08-01

融合改进GAN网络的夜视环境车道线检测

刘岩,仇甜甜,肖艳秋,朱付保,王靖雯   

  1. 1.郑州轻工业大学 计算机与通信工程学院,郑州 450001
    2.郑州轻工业大学 机电工程学院,郑州 450001

Abstract: The lane detection methods based on deep learning have effectively promoted the development of automatic driving technology. However, these methods still have some shortcomings in dealing with the problem of lane detection in night vision environment. Aiming at the problems of low illumination and weak detection accuracy in lane detection in night vision environment, this paper combines the attentive mechanism-based generative adversarial network(Attentive GAN) and spatial convolutional neural network(SCNN) algorithm to propose a lane detection method for night vision environment. Firstly, the Attentive GAN network is used to improve the quality of night road images, highlight the lane features in the road images, and then ResNet-18 is used to extract features, and then the SCNN network is used to transfer the image information row by row and column by column, and the cubic spline curve is used to carry out the information transfer. The probability map is fitted to obtain the final lane detection result. This paper uses the simulated TuSimple dataset to verify the detection performance of the proposed method. The experimental results show that the laned etection method in this paper has great performance of lane detection in night vision environment.

Key words: lane detection at night, attentive generative adversarial networks(GAN), ResNet-18 network, spatial convolutional neural network(SCNN)

摘要: 基于深度学习的车道线检测方法有效地促进了自动驾驶技术的发展,然而这些方法在处理夜视场景下车道线检测问题仍然存在一定的不足。针对夜视场景下车道线检测存在的检测精度弱问题,将基于注意力机制的生成对抗网络(attentive GAN)和空间卷积神经网络(spatial convolutional neural network,SCNN)算法相结合,提出一种针对夜视场景的车道线检测方法。该方法利用Attentive GAN网络提高夜间道路图像质量,突出道路图像中的车道线特征,再利用ResNet-18网络提取车道线特征,随后利用SCNN网络进行图像信息的逐行逐列传递,并利用三次样条曲线进行概率图拟合,得到最终的车道线检测结果。在利用模拟后的TuSimple数据集验证了方法的检测性能,实验结果表明,提出的车道线检测方法在夜视场景下具有良好的车道线检测性能。

关键词: 夜间车道线检测, 注意力机制的生成对抗网络(Attentive GAN), ResNet-18网络, 空间卷积神经网络(SCNN)