3D Mapping and Change Detection in Indoor Environments Using Multisource LiDAR Point Clouds

室内移动的地图绘制对于从室内导航、危险地图绘制、自主停车、设施管理、室内设计到虚拟旅游的广泛应用是重要的。如今，具有多线、固态和区域阵列LiDAR传感器的室内移动的激光扫描（iMLS）系统可以获取点云（数据点集），这些点云不仅表示室内或地下空间中的静态对象，还表示移动对象的3D运动数据。因为在建筑物内往往无法利用全球导航卫星系统进行精确定位。通过将LiDAR传感器与低成本惯性测量单元（IMU）和相机集成，iMLS系统可以从建筑物内部捕获3D点云数据。然而，除了iMLS点云具有体积大、空间离散、点密度随距离变化等特点外，复杂室内空间中三维物体的不完整性、遮挡性和相似性也使得三维物体的定位、识别和变化检测具有相当大的挑战性。到目前为止，将点云转换为属性空间对象的过程是由操作员驱动的，尚未实现自动化。为了应对这些挑战，该研究计划旨在开发新的策略，方法和软件工具，可以快速，准确地更新3D室内地图从3D点云获得的各种类型的低成本的MLS系统在GNSS-否认室内环境。这项研究包括四个主要主题：（A）研究用于处理多源iMLS点云的鲁棒融合框架，（B）开发新的质量评估度量和多尺度iMLS数据增强方法，（C）探索用于3D对象检测和变化检测的弱监督学习方法，以及（D）开发可扩展的iMLS软件原型，以实现大规模点云的交互式分割操作。预计这项研究将对室内测绘做出三个潜在的贡献：（1）用于处理多源iMLS点云的广义3D特征描述和3D对象识别的理论框架，（2）需要较少标签数据的弱监督学习算法，以及（3）用于室内测绘和变化检测的开源软件原型。所开发的算法和软件原型将克服当前用于室内测绘的点云处理工具的缺点，并提供更多的室内环境变化的定量报告。该研究成果将提升加拿大在移动的LiDAR测绘技术和地理空间测绘行业的世界领先地位。

Indoor mobile mapping is important for a wide range of applications ranging from indoor navigation, hazard mapping, autonomous parking, facility management, interior design, to virtual tourism. Today, indoor mobile laser scanning (iMLS) systems with multi--line, solid-state, and area-array LiDAR sensors can acquire point clouds (sets of data points) that represent not only static objects but also 3D motion data of moving objects in the indoor or underground space. As precise positioning with a Global Navigation Satellite System (GNSS) is often not attainable inside buildings. By integrating the LiDAR sensors with a low-cost inertial measurement unit (IMU) and a camera, an iMLS system allows the capture of 3D point cloud data from building interiors. However, besides the characteristics of large volume, spatial discreteness, and point density variation over distances of iMLS point clouds, the incompleteness, occlusion, and similarity of 3D objects in complex indoor spaces also make 3D object localization, recognition and change detection considerably challenging. To date, the process of converting point clouds into attributed spatial objects is operator-driven and has not yet reached automation. To tackle these challenges, this research program aims to develop novel strategies, methods, and software tools that can rapidly and accurately update 3D indoor maps from 3D point clouds acquired by various types of low-cost in MLS systems in GNSS--denied indoor environments. This research includes four main themes: (A) to investigate robust fusion frameworks for handling multi-source iMLS point clouds, (B) to develop new quality assessment metrics and multi-scale iMLS data enhancement approaches, (C) to explore weakly supervised learning approaches to 3D object detection and change detection, and (D) to develop extensible iMLS software prototype towards an interactive segmentation operation of large-scale point clouds. It is anticipated that this research will make three potential contributions to indoor mapping: (1) a theoretical framework for the generalized 3D feature description and 3D object recognition to handle multi-source iMLS point clouds, (2) the weakly supervised learning algorithms that require fewer label data, and (3) an open-source software prototype for indoor mapping and change detection. The developed algorithms and software prototype will overcome the shortcomings of current point cloud processing tools for indoor mapping and provide more quantitative reports of the identified changes in indoor environments. The research results will enhance Canada's world leadership in mobile LiDAR mapping technology and geospatial mapping industry.