Semantic 3D Reconstruction of Furnished Interiors Using Laser Scanning and RFID Technology

Abstract

Terrestrial Laser Scanning (TLS) technology is increasingly used for the generation of accurate three-dimensional (3D) models of objects and scenes. But converting the acquired 3D point cloud data into a representative, semantic 3D model of the scene requires advanced processing and skills. This research field is challenging, particularly when considering inhabited, furnished environments that are characterised by clutter and occlusions. This paper presents a TLS data-processing pipeline aimed at producing semantic 3D models of furnished office and home interiors. The structure of rooms (floor, ceiling, and walls with window and door openings) is created using Boundary Representation (B-Rep) models that not only encode the geometry of those elements, but also their connectivity. Windows and doors are recognized and modeled using a novel method based on molding detection. For the furniture, the approach uniquely integrates smart technology [radio frequency identification (RFID)] that is increasingly used for Facilities Management (FM). RFID tags attached to furniture are sensed at the same time as laser scanning is conducted. The collected IDs are used to retrieve discriminatory geometric information about those objects from the building’s FM database; this information is used to support their recognition and modeling in the point cloud data. The manuscript particularly reports results for the recognition and modeling of chairs, tables, and wardrobes (and other similar objects like chests of drawers). Extended experimentation of the method has been carried out in real scenarios yielding encouraging results.