Lane Mark Detection Algorithm Based on Deep Learning

  • Zhisong Guo, Zhaohong Liu, Jun Li


Lane mark detection is a critical part of automatic driving or Advanced Driving Assistant system (ADAS). Due to the diversity and complexity of environment, it is still a challenge to detect lane mark accurately and efficiently. This paper studies lane mark detection technology based on computer vision and presents a lane mark detection algorithm using deep learning convolutional neural networks. The proposed approach mainly consists of the following steps: first, data augmentation and data clean are applied to the datasets; and then a new neural architecture similar to U-Net is designed. In this structure, EfficientNet is adopted as its backbone. Finally, the empirical result shows that the proposed method can achieve high detection accuracy and strong robustness in many different road scenes.