【摘要】 高精度地图主要利用已采集图像的地面信息生成,但是在真实环境中图像的地面信息容易受动态障碍物遮挡,同时GPS难免会有抖动误差,导致地图拼接效果并不理想。针对上述问题提出了一种基于路面分割的动态障碍物去除与图像配准方法。使用深度学习对全景图像进行语义分割并提取路面信息,在去除动态障碍物干扰后,利用路面特征进行图像配准。融合GPS与里程计信息进行定位优化,利用多帧图像叠加填补空缺形成地图。最终,实验验证了该方法在去除动态障碍物的同时也提高了地图的精度。更多还原

【Abstract】 High-precision maps are mainly generated by using ground information of acquired images. But in the real environment, the ground information of image is easily occluded by dynamic obstacles, and GPS will inevitably have jitter errors, resulting in the map mosaic problems. Therefore, this paper proposes a dynamic obstacle removal method and an image registration method based on the road segmentation for the aforementioned problems. In-depth learning is used to semantically segment panoramic images and extract pavement information. Pavement features are used for image registration and removal of dynamic obstacle occlusion can also improve the accuracy of registration. GPS and odometer information are fused for positioning optimization, and multi-frame image overlay is used to fill the gaps to form a map. Finally, the experimental results show that this method improves the accuracy of map while removing dynamic obstacles.更多还原