Visual indoor localization for smart indoor services is a growing field of interest as cameras are now ubiquitously equipped on smartphones. In this study, a hierarchical indoor localization algorithm is designed and validated based on 3D facility scan data, which are originally collected for facility modeling purposes. The study has shown promising results in indoor localization. The study also demonstrated a scalable approach to generate high-quality images with reference poses from laser scan data, opening doors to generate labeled images to train end-to-end pose regression model (i.e., PoseNet). In this regard, this study is the first attempt to leverage facility scan data, which are commonly collected for Building Information Modeling (BIM) purpose, for indoor localization. As more facilities are documented with laser scanners, our algorithm can unlock additional values of collected data for intelligent applications.

中文翻译：

---

使用 3D 设施扫描数据进行精确的室内定位

智能室内服务的视觉室内定位是一个越来越受关注的领域，因为摄像头现在无处不在地安装在智能手机上。在这项研究中，基于最初为设施建模目的收集的 3D 设施扫描数据设计和验证了一种分层室内定位算法。该研究在室内定位方面显示出有希望的结果。该研究还展示了一种可扩展的方法，可以使用激光扫描数据的参考姿势生成高质量图像，打开大门以生成标记图像以训练端到端姿势回归模型（即 PoseNet）。在这方面，本研究首次尝试利用通常为建筑信息模型 (BIM) 目的收集的设施扫描数据进行室内定位。随着越来越多的设施被激光扫描仪记录下来，