融合语义特征与GPS位置的地面激光点云拼接方法

拼接是地面激光点云数据处理的必要步骤，但基于同名点的点云拼接方式已成为阻碍点云处理效率提升的长期瓶颈，而直接匹配点云识别同名特征的方法亦对点云重叠区域具有较高的要求。本文提出一种融合语义特征与GPS位置的地面激光点云拼接方法，通过语义知识自动识别出原始三维点云中所包含的地面特征与建筑物立面特征，并使用这两种面状特征结合点云测站中心的GPS位置作为同名标靶进行点云初始拼接，随后使用点到面最小距离约束下的ICP进行点云精确拼接。实验表明，本方法可以有效提高地面激光点云拼接的整体效率，尤其对于包含平面结构（如马路、建筑物）的场景具有良好的拼接效果。

Registration of Terrestrial Laser Point Clouds by Fusing Semantic Features and GPS positions

Registration is an inevitable process during the procedure of point cloud processing, while point based registration has been the bottleneck of the whole point cloud processing procedure, and pure data driven points matching methods demand strict conditions for the overlapping regions between base and registration stations. This article proposes a new registration method for terrestrial laser point clouds by fusing semantic features and GPS positions. First, the semantic features such as ground and wall fa?ade are automatically recognized using knowledge, then together with GPS positions of the scan station, two point clouds can be roughly registered, and followed by a fine tuning using ICP which minimizes point to plane distance. Two test cases indicate that this method achieves higher registration efficiency for terrestrial laser point clouds, and is especially applicable to scenes containing planar structures such as roads and buildings.