[Purposes] Optimizing the noise area and rough edges in the lane line detection results based on deep learning. [Methods] A new method combining deep learning algorithm and post-processing is proposed.The fully connected conditional random fields (Fully Connected CRFs) is introduced to modify the lane line probability map output by the ENet-SAD algorithm, fit the probability map with the original image to get the lane line detection result. The algorithm in this paper is trained and tested on the self-built data set and the CULane data set. [Findings] The results show that on the self-built data set, the F1-score of the algorithm in this paper in four scenarios of normal, strong light, shadow and occlusion were 90.0%, 73.1%, 81.5%, and 76.6%, respectively. On the CULane dataset, the F1-score of the algorithm in this paper in conventional scenarios reached 91.0%. [Conclusions] The lane detection algorithm proposed herein demonstrates adaptability to various environmental scenarios, and it is an effective lane line detection algorithm.