基于全局特征构建与注意力引导融合的车道线检测算法研究OA

To address local feature dependence and scarce visual cues as well as to fully utilize the priori information of lane lines(narrow,slender and spatially spanned),this paper proposes a lane line detection model GFSCNet based on global feature construction and attention-guided feature fusion.It incorporates a global feature construction module that employs spatial attention,channel attention,and a global context module to generate multi-scale global features,enhancing the receptive field.Meanwhile,an adaptive fusion module integrates global semantic and local detail features through attention guidance,improving feature representation.The structure-aware contextual attention module further models global dependencies using a multi-branch structure,expanding the receptive field of anchor features and optimizing line-anchor regression accuracy.Experiments on the TuSimple and CULane datasets demonstrate GFSCNet outperforms other methods(SCNN,RESA,UFLD,and LaneATT)in complex road conditions.Specifically,GFSCNet achieves an F1 score of 77.76％on the CULane dataset and performs exceptionally well in challenging scenarios(glare and curved roads).