Optimised scan planning for LiDAR data acquisition towards 3D indoor understanding

Resumen

Changing habits are leading people to spend more time indoors. This fact is aggravated in older people who are more prone to suffer from some kind of impairment that affects their mobility and orientation. Advances in positioning systems, smart phones and tablets, together with access to semantic building models have stimulated interest in the development of applications for assisted indoor navigation. Since BIM models provide rich geometric and semantic information, they are an interesting option to provide contextual support for navigation. However, these models are not always available, may be outdated and their flexibility to changes in the environment is limited. In contrast, indoor mapping systems can generate a highly accurate 3D model in the form of a point cloud. Despite data acquisition is considered a fast process, scanning large and complex sites can became a time-consuming task if the scanning is not properly planned beforehand. Raw point clouds provided by scanning systems do not contain semantic information, which makes their use in indoor navigation challenging. Fortunately, with the development of Artificial Intelligence techniques, promising results are being obtained in extracting semantic information from point clouds in an automated way. The semantic information of buildings is essential for the interpretation of the indoor space in order to provide support for a safe and accurate indoor navigation oriented to the user's context. In This Doctoral Thesis, different methods with the objective of implementing path planning for contextual indoor navigation from point clouds are presented. The developed methods address the three global procedures that have been identify as necessary to achieve the objective of this Doctoral Thesis: Data acquisition, indoor understanding and indoor navigation. The methods developed in the first part of the thesis propose solutions for scan planning problem with the aim to optimise data acquisition in terms of time ensuring the quality and completeness of the data. Room segmentation and classification methods were implemented to extract semantic information from raw point clouds addressing indoor understanding problem. Semantically enriched point clouds are exploited to partition the indoor on basis different contexts deriving in a hierarchical path planning that supports the efficient computation of precise routes oriented to the agent's context. All the proposed methods in this Doctoral Thesis were tested in real case studies obtaining encouraging results. The methods and results were presented as a compendium of six scientific articles, five of them have been published in international journals with high impact factor and another one is in peer-review process. Three articles were published in international journals indexed on the Journal Citation Report (JCR), and two papers were presented in international conferences.