Real-Time Lane Line Detection via Millimeter-Wave Radar

Abstract This paper proposes a high-precision lane detection and correction method based on roadside millimeter-wave radar. After deploying millimeter-wave radars on traffic light poles to collect lane point clouds, spatiotemporal adaptive DBSCAN clustering is employed to extract lane clusters, with adaptive neighborhood and temporal weighting improving clustering accuracy. To compensate for sensor offsets, an ICP algorithm with curvature keypoint selection is introduced for rigid alignment, enhancing correction stability. Lane shapes are fitted using a segmental sliding-window PCA combined with convex hull calculation and B-spline interpolation, achieving high-precision modeling for complex curves and nonlinear lanes. Experimental results demonstrate that the proposed method effectively handles environmental variations and noise, significantly improving the accuracy and robustness of lane detection and correction.