高分辨率遥感影像的车道级高精地图要素提取

高精地图是城市道路场景下自动驾驶必不可少的基础设施之一。针对目前高精地图静态层在生产和更新中存在的采集成本高、周期长、数据处理复杂等问题，本文提出了一整套基于高分辨率遥感影像的高精地图静态层的车道级要素提取方案。首先，通过SURF算法对多时相影像进行配准，同时结合影像的光谱信息对像元进行判断，实现了对车道中动态车辆的检测和去除；然后，基于面向对象的方法提出了一种要素对象的特征选择方法，并结合统计学理论对各要素进行阈值分析，实现了对虚线车道线、人行横道等车道级要素的检测和提取；最后，结合试验数据，验证了本文所提出的基于遥感影像的车道级要素大范围快速提取方案的有效性。

High-precision lane-level map elements extracting based on high-resolution remote sensing image

High-precision map is automatically under the city road scene driving one of the necessary infrastructures. Aiming at the problems of the high acquisition cost,long period,and complicated data processing in the production and update of the static layer of high-precision map,the lane-level element extraction scheme of high-precision map static layer based on high-resolution remote sensing image is proposed.Firstly,multi-temporal images are registered by the SURF algorithm,and pixels are judged by combining the spectral information of the image,thus realizing the detection and removal of dynamic vehicles in the lane.Secondly,based on the object-oriented method,a feature selection method of element objects is proposed,and the threshold analysis of each element is carried out in combination with the statistical theory to realize the detection and extraction of lane-level elements such as the dashed lane line crosswalk. Finally,combined with experimental data,the effectiveness of the proposed lane-level elements extraction scheme based on remote sensing image is verified.