Lane detection is a crucial technology in au-tonomous driving. Current lane detection methods focus on individual images(frame) and ignore associations along the video. In this work, we reevaluate the limitation of video detection methods based on DETR, which have predominantly focused on the attention mechanism, disregarding the inherent characteristics of the query instances themselves. We observe a consistent inherent where lane lines appear in pairs and exhibit horizontal symmetry when observed by a forward-facing camera. Building upon this insight, we introduce a novel approach named Symmetry Feature Enhancement for Decoder Layer (SFED). SFED enables the query to capture the intrinsic symmetry present in lane markings, thereby enhancing its expressive capacity. Additionally, the SFED module contributes to a modest improvement in convergence speed. Our method enriches the query representation with information about lanes throughout the entire video, significantly enhancing the accuracy of subsequent detection and tracking processes. Our proposed approach achieves a new state-of-the-art performance on the VIL100 dataset.