With the introduction of full convolutional neural product networks, semantic segmentation networks have also been widely used in the field of deep learning. Most lane detection tasks are currently done on the basis of semantic segmentation networks, so the development of semantic segmentation also directly determines the progress of lane detection. Methods: The development of semantic segmentation networks and the performance comparison between different model frames are used to summarize the improvement points as well as the advantages and disadvantages of each approach. Current lane detection network models with good performance based on semantic segmentation networks are described and the performance between the models is compared. Result: The current development of deep learning-based lane detection methods has been very fruitful, with significant improvements in network performance, but they cannot yet be applied in practice. For example, lightweight networks are not stable enough in extracting features, while deep neural networks are too ineffective in real time. Conclusion: Lane detection is of high research value as a key technology for unmanned driving. However, most of the current neural network methods have not been studied from a practical point of view, and there are few methods that use multiple frames as a basis for research. Therefore, in the future how to efficiently use continuous images for lane detection is a key direction to be researched in the future.