Abstract Most mainstream methods mainly regard lane detection as a pixel-by-pixel segmentation task, resulting in high computational cost and time-consuming, and the accuracy is influenced by severe occlusion and extreme lighting conditions. To tackle these issues, we propose a novel Attention-based Row Selecting Networks(ARS-Net), which utilizes the row selecting method based on global features to detect lanes, greatly improves the detection speed. At the same time, channel and spatial attention mechanisms are integrated into ResNet as the backbone to focus on important features and suppress unimportant ones, so as to adjust feature weights and reduce information loss. Besides, group normalization is employed to replace batch normaliza-tion, which enhances the stability of accuracy. We carry out immense amounts of experiments on two international public datasets TuSimple and CULane, the experimental results show that our method achieves the state-of-the-art performance in both accuracy and speed and significantly outperforms other methods for real-time and efficient lane detection in real-world applications.