Lane detection is a key technology in intelligent driving. Previous DETR-based works utilized instance queries to detect lanes. However, the number of these queries is artificially assigned and learned entirely by the model itself, which has certain limitations. To solve this problem, A novel DERT architecture object detector called QED-Former is developed for end-to-end use, including Query Enhanced Decoder Layer (QEDL) for Lane Detection and Video Query Decompose Transformer Decoder (VQDTD), and get a competitive performance on VIL100. Our method QED-Former can enable the query to learn the symmetry inherent in the lane itself, thereby enhancing its expressive power. In addition, our query contains information about lanes throughout the entire video. This improves the accuracy of subsequent detection and tracking effectively. Our proposed method has archived 88.5 F1 score of 0.5 IoU and 95.1 accuracy score on VIL 100 dataset. In addition, our newly added module also slightly improves the convergence speed of the mode. Our model currently shows the competitive performance on the VIL 100 dataset.