A Method for Generating Network-based Indoor Topology Data using an Omnidirectional Image Dataset

People spend most of their time indoors. Therefore, indoor spatial information is vital for leading a safe and high-quality life. Nowadays, as digital twins and smart cities become popular, various fields are trying to build indoor spatial information to simulate disaster or terrorist situations in advance and find countermeasures or to provide efficient routes and query services to users. There are various methods for visually modeling indoor spaces, but recently, modeling using image data that are efficient in economic feasibility and usability has been increasing interest. However, since image data does not contain geometric or semantic information, there are limitations in providing image-based indoor spatial information services. In order to provide services such as routing, spatial or non-spatial queries, and object identification, these services must use topology data in conjunction with image data. Multiple studies have proposed methodologies to link image and topology data, but existing studies have limitations in requiring external reference data or too much manual work. This study aims to overcome the limitations of previous studies and effectively construct indoor spatial information by establishing relationships between image data and generating network-based topology data from the image dataset. Accordingly, we present a series of methodologies to automatically create an NRS (Node-Relation Structure) dataset using an omnidirectional image dataset. Specifically, we present methods to acquire an image dataset, detect object space, and generate nodes and edges using image and object space data. The final output of this study is NRS data consisting of nodes and edges.