Extraction of Structural and Semantic Data from 2D Floor Plans for Interactive and Immersive VR Real Estate Exploration

Abstract

Three-dimensional reconstructions of indoor environments are useful in various augmented and virtual scenarios. Creating a realistic virtual apartment in 3D manually does not only take time, but also needs skilled people for implementation. Analyzing a floor plan is a complicated task. Due to the lack of engineering standards in creating these drawings, they can have multiple different appearances for the same building. This paper proposes multiple models and heuristics which enable fully automated 3D reconstructions out of only a 2D floor plan. Our study focuses on floor plan analysis and definition of special requirements for a 3D building model used in a Virtual Reality (VR) setup. The proposed method automatically analyzes floor plans with a pattern recognition approach, thereby extracting accurate metric information about important components of the building. An algorithm for mesh generation and extracting semantic information such as apartment separation and room type estimation is presented. A novel method for VR interaction with interior design completes the framework. The result of the presented system is intended to be used for presenting a large number of apartments to customers. It can also be used as a base for purposes such as furnishing apartments, realistic occlusions for AR (Augmented Reality) applications such as indoor navigation or analyzing purposes. Finally, a technical evaluation and an interactive user study prove the advantages of the presented system.

Generalizing Floor Plans Using Graph Neural Networks

“…They then use the A* algorithm to identify the final door and window entities. Most important, Gerstweiler et al [15] create graphs directly from the lines, arcs, polylines, etc. that make up CAD files, thereby avoiding the problematic transformation to the image domain.…”

“…With recent breakthroughs in GNNs and the new possibilities they bring, it is worth combining the idea from Gerstweiler et al [15] with GNNs. GNNs follow a recursive neighborhood aggregation scheme, where feature vectors are passed between neighboring nodes, capturing structural information within nodes' k-hop neighborhood [16].…”

ABGS: A System for the Automatic Generation of Building Information Models from Two-Dimensional CAD Drawings

“…Zhang et al [41] addressed this problem by eliminating the ambiguities and improving information integrity. Similar algorithms for the reconstruction of 3D models from floor plans can be also used for immersive virtual reality applications [42]. In a recent work, Beacons coupled with BIM models were used to provide location and guidance inside a building in real time [43].…”

3d Modeling of a Complex Building: From Multi-View Image Fusion to Google Earth Publication

“…As such, this section presents a brief overview of studies done on complex architecture reconstructions. There have been studies which involve the use of floor plans and existing 2D drawings of complex structures and buildings to retrieve measurements and then upscale these for 3D model production (Gerstweiler et al, 2018;Yin et al, 2009). For more advanced treatments image-based models have generally been employed.…”