摘要
车道线作为实现精确导航和自动驾驶的基础信息,其采集方式自动化程度低、生产周期长,严重影响了高精地图的应用。因此,文章设计了一种利用无人机正射影像,基于U-Net语义分割模型,结合栅矢数据处理的高精地图车道线自动提取方法。文章利用Lovász Loss完成20%样本的U-Net模型训练,实现了IoU在75%以上的车道线语义分割,且U-Net识别出的栅格车道线经过栅矢结合手段处理后即可得到高质量矢量车道线。文章设计的综合无人机倾斜摄影、深度学习、GIS数据处理的车道线提取方法可为高精地图车道线的获取提供一种新思路、新方法,为无人驾驶提供一种新的数据支持。
Lane lines as the basic information to realize accurate navigation and autonomous driving,the current acquisition method has low automation and long production cycle,which seriously affects the application of highdefinition maps.Therefore,this paper designs an automatic lane line extraction method for high-definition maps using UAV orthophoto,based on U-Net semantic segmentation model,combined with raster vector data processing.In the experiment,the U-Net model is trained with 20%of the samples using Lovász Loss,and the semantic segmentation of lane lines with IoU above 75%is achieved,and the raster lane lines identified by U-Net can be processed by combining raster and vector means to obtain high quality vector lane lines.The lane line extraction method designed in this paper,which integrates UAV tilt photography,deep learning and GIS data processing,can provide a new idea and method for acquiring lane lines in high-definition maps and provide a new data support for autonomous driving.
作者
高正跃
李春梅
Gao Zhengyue;Li Chunmei(School of Geography,Geomatics and Planning,Jiangsu Normal University,Xuzhou 221116,China)
出处
《无线互联科技》
2023年第12期78-83,共6页
Wireless Internet Technology
基金
江苏师范大学校级科研项目,项目名称:人工智能时代下的单目视觉导航定位技术研究,项目编号:2021XKT0098。
