Intelligent 3D world building from mobile terrestrial LiDAR point clouds

移动激光扫描(MLS)提供快速、高效和经济实惠的数据收集。正如这项技术的早期采用者进行的研究所强调的那样，使用移动激光扫描可以减少80%-90%的现场时间，但增加200%-300%的后台处理。处理的挑战来自于从大量的激光点数中区分不同特征类型的困难。即使可以应用一些高效的数值算法来加速三维点云的处理，但它们往往依赖于用户的大量交互。MLS工作流程中重要的“增值”在于将3D地理参考点云转换为城市基础设施和环境的智能3D模型。这种类型的语义描述不仅比原始数据要紧凑得多，而且还支持分析空间使用情况或规划翻新等高级任务。这是目前该领域面临的真正挑战。拟议研究计划的长期目标旨在设计新的方法和处理解决方案，以高效地从移动地面激光雷达点云构建智能3D世界。这一目标取决于加快从海量注册数据中检测和提取重要结构的过程，提供包括几何和功能知识的模型，并将MLS点云与互补数据集融合，产生多尺度和多模式表示。提出的研究工作的基础是兴趣点的概念和适用于MLS点云的相关3D描述符。与计算机视觉相比，过去对这一主题的研究很少，而计算机视觉是一个研究得很好的领域。MLS兴趣点的概念可能会对创建多尺度3D模型并将语义附加到这些模型的能力和便利程度产生强烈影响。这将有助于满足目前对地理信息的巨大需求，这些信息必须更新、更3D、更容易通过互联网访问。

Mobile laser scanning (MLS) provides fast, efficient and cost-effective data collection. As studies carried out by various early adopters of this technology have highlighted, using mobile laser scanning allows cutting down field times by 80-90% but increases back office processing by 200-300%. The processing challenge comes from the difficulty of distinguishing different feature types from the huge amount of laser points. Even if some efficient numerical algorithms can be applied to speed up the 3D point cloud processing, they often rely on significant interaction of the user. The significant "value add" in the MLS workflow comes in transforming 3D georeferenced point clouds into intelligent 3D models of our urban infrastructure and environment. Not only is this type of semantic description significantly more compact than the raw data, but it supports high-level tasks like analysis of space usage or planning for renovations. This is the real challenge currently facing this field.The long-term objective of the proposed research program aims at designing new methodology and processing solutions to efficiently build intelligent 3D worldsfrom mobile terrestrial LiDAR point clouds. This objective depends upon speeding up the process of detecting and extracting significant structures from the huge amount of registered data, providing models including geometric and functional knowledge, and fusing MLS point clouds with complementary data sets, yielding multi-scale and multi-mode representations. The cornerstone of the proposed research work is the concept of interest point and related 3d descriptors adapted to MLS point clouds. There has been little research on such a topic in the past, compared to computer vision where this is a well researched area. The concept of MLS interest point could have strong impact on the capability and facilityto create multi-scale 3D models and to attach semantics to these models. This would help meeting the current tremendous need for geo-information which has to be more up to date, more 3D, more accessible over the Internet.