State‐of‐the‐art in Automatic 3D Reconstruction of Structured Indoor Environments

Abstract

Creating high‐level structured 3D models of real‐world indoor scenes from captured data is a fundamental task which has important applications in many fields. Given the complexity and variability of interior environments and the need to cope with noisy and partial captured data, many open research problems remain, despite the substantial progress made in the past decade. In this survey, we provide an up‐to‐date integrative view of the field, bridging complementary views coming from computer graphics and computer vision. After providing a characterization of input sources, we define the structure of output models and the priors exploited to bridge the gap between imperfect sources and desired output. We then identify and discuss the main components of a structured reconstruction pipeline, and review how they are combined in scalable solutions working at the building level. We finally point out relevant research issues and analyze research trends.

Automatic Reconstruction of Multi-Level Indoor Spaces from Point Cloud and Trajectory

“…A recent survey and tutorial indicated that modeling with inter-room and inter-floor connections approach is reserved for future work. Pintore et al [ 1 ] presented the modeling of an entire building; however, the connections between rooms and levels is still a complex problem. Hence, a global solution is required to reconstruct complex environments [ 2 ].…”

Topology-Aware 3d Modelling of Indoor Spaces From Point Clouds

“…Wide range of applications of 3D building models such as Building Information Modelling (BIM), facility management, Indoor Location Based Services (ILBSs), virtual and augmented reality, and emergency response motivates several researchers in photogrammetry, computer vision, computer graphics and other relevant disciplines to increase the level of automation of the reconstruction process (Mura et al, 2016;Ochmann et al, 2015;Pintore et al, 2020;Sahebdivani et al, 2020;Sanchez & Zakhor, 2012). Meanwhile, with the increasing availability of various types of the sensors for generating point clouds such as terrestrial laser scanners (Jung et al, 2018) as well as mobile laser scanners (Karam et al, 2018;Maboudi et al, 2018), and HoloLens (Hübner et al, 2020) point clouds are ubiquitous data source for 3D modelling of indoor (and outdoor) spaces.…”

A Comparison of Low-Cost Sensor Systems in Automatic Cloud-Based Indoor 3D Modeling

“…The automation of image-based reconstruction has advanced in recent years, with automated modeling providing an efficient, low-cost alternative and results of good accuracy [22]. With personal cameras being ubiquitous, image-based modeling is a widely accessible method for indoor modeling [30], though the geometric information obtainable with TLSs or RGB-D cameras aids reconstruction, and areas with sparse features remain an issue [31].…”

“…Occlusions in the point cloud pose a challenge for automatic modeling, though the reduction of their negative impact on the model quality through detecting regularities in point clouds with significant occlusions has been explored [35]. The segmentation of floors and walls can be done under the assumption that floors are horizontal planes and walls vertical planes [30,33], though in more complex architectures, the division of a building into floors based solely on height is unfeasible due to slanted roofs and walls, and spaces that span multiple levels of the building [32]. A combination of region growing and planar segmentation has been explored as an alternative for reconstructing spaces with no a priori assumptions about their geometry [36].…”

“…RGB-D cameras offer both geometric information and RGB images, though the low geometric quality of the data can make the semantic segmentation of floors and walls challenging, and the extraction of a point cloud from RGB-D data can be computationally expensive [37]. Alternatively, a collection of range and color maps in a reference frame can be used as input for the reconstruction, with the sensor's positioning information aiding reconstruction [30]. For modeling RGB-D data, deep learning approaches are common, with methods including the integration of the 2D semantic segmentation of images with a 3D reconstruction [36], the 2D reconstruction of walls as polygonal loops [37], and 3D volumetric segmentation [38].…”