基于多损失融合和混洗注意力的车载LiDAR点云道路标线提取方法

Road marking extraction method from mobile LiDAR point clouds based on multi-loss fusion and shuffle attention

道路标线的准确提取在高级辅助驾驶系统和高精度地图的开发中具有重要意义。针对现有的基于阈值的车载激光点云道路标线提取方法在反射强度与点密度分布不均、道路标线与路面对比度低时提取效果较差的问题,本文提出了基于多损失融合和混洗注意力的车载LiDAR点云道路标线提取方法。选取典型高速公路试验样区进行道路标线提取试验,并与常规方法进行了精度对比分析。试验表明,本文方法在道路标线提取精度方面优于其他方法,有望更好地服务于自动驾驶的高精度地图开发应用。

The accurate extraction of road markings is of great significance in the development of advanced driving assistance system and high precision map.Since the point clouds have uneven distribution on reflection intensity and density or low contrast between road line and its surrounding road surface,the existing thresholding method is difficult to extract road line accurately,So this paper proposes the vehicle-mounted LiDAR point cloud road marking extraction method based on multi-loss fusion and mixing and shuffling attention,and selects a typical highway test sample area to conduct the road marking extraction test and compare and analyze the accuracy of the method with that of the conventional method.The accuracy comparison analysis is carried out with the conventional method.The test shows that the method in this paper is better than other methods in improving road marking extraction accuracy,which is expected to better serve the development and application of high-precision maps for autonomous driving.