Lane line extraction based on multi-frame fusion of millimeter wave radar

Aiming at the failure of traditional lane line detection methods due to wear and occlusion of lane lines in complex road scenes, an adaptive lane line extraction method based on multi-frame point cloud of millimeter wave radar is proposed. By integrating dynamic meshing and multi-frame spatio-temporal feature enhancement techniques, a three-stage processing framework including point cloud density distribution analysis-improved kernel density estimation-segmented cubic polynomial modeling is constructed, and an evaluation mechanism is introduced to quantify the model reliability. Experiments show that the average lateral error of the lane centerline of this method is 0.14m in straight line and curved road scenarios, which is better than the traditional DBSCAN clustering method fused with the quadratic fitting method, and the maximum error of the curved road region is reduced to 0.25m. This method significantly improves the topological modeling integrity of lane lines in road scenarios through spatio-temporal fusion and geometric constraint optimization, and provides a highly robust solution for all-weather road sensing. robust solution for all-weather road sensing.