Automatic reconstruction of fully volumetric 3D building models from oriented point clouds

https://doi.org/10.1016/j.isprsjprs.2019.03.017Get rights and content

Abstract

We present a novel method for reconstructing parametric, volumetric, multi-story building models from unstructured, unfiltered indoor point clouds with oriented normals by means of solving an integer linear optimization problem. Our approach overcomes limitations of previous methods in several ways: First, we drop assumptions about the input data such as the availability of separate scans as an initial room segmentation. Instead, a fully automatic room segmentation and outlier removal is performed on the unstructured point clouds. Second, restricting the solution space of our optimization approach to arrangements of volumetric wall entities representing the structure of a building enforces a consistent model of volumetric, interconnected walls fitted to the observed data instead of unconnected, paper-thin surfaces. Third, we formulate the optimization as an integer linear programming problem which allows for an exact solution instead of the approximations achieved with most previous techniques. Lastly, our optimization approach is designed to incorporate hard constraints which were difficult or even impossible to integrate before. We evaluate and demonstrate the capabilities of our proposed approach on a variety of complex real-world point clouds.

Introduction

The challenging problem of generating high-quality, three-dimensional building models from point clouds has been approached in a variety of ways in recent years by the computer graphics, remote sensing, and architecture communities. The concept of Building Information Modeling (BIM) is increasingly used as a modern and versatile means for integrating various aspects of construction planning and facility management into a common, digital data basis. A widely adopted implementation of BIM is the Industry Foundation Classes (IFC) standard which defines a broad range of geometric and abstract entities as well as their relationships. In contrast to the geometric representation of a building in the form of e.g. unordered point clouds, sets of unconnected surfaces, or boundary meshes, a BIM/IFC model closely resembles the physical building structure by defining semantically annotated, volumetric building entities such as walls and floor slabs, usually including information how these elements are interconnected. While such rich, high-level descriptions are useful for a variety of applications such as verification of construction processes by means of e.g. detecting differences between as-planned BIM models and as-built point cloud measurements (Brodie et al., 2017, Hyland et al., 2017, O’Keeffe et al., 2017), or energy simulations (Garwood et al., 2018), they may not be readily available especially in case of the existing, legacy building stock. Since manual generation of accurate models is often a time-consuming process even when using point cloud data as a template, automated approaches for generating BIM models from measurements have become an interesting topic of current research.

Most previous reconstruction approaches do not represent buildings using parametric, volumetric entities but instead either result in completely separate, planar surfaces without information about their relations (Sanchez and Zakhor, 2012), watertight boundaries of the whole building (Oesau et al., 2014), or boundaries of separate rooms (Mura et al., 2016, Turner et al., 2015). Also, some methods which support multi-story buildings assume that these can be globally separated by horizontal cuts through the building (Macher et al., 2017, Turner et al., 2015) in order to process each story separately. This assumption can be limiting in practice since varying floor or ceiling heights within a story (see e.g. Fig. 9) cannot be represented. None of these approaches yields a representation which enables unhindered usage in the aforementioned scenario. While one recent approach (Ochmann et al., 2016) models buildings using volumetric walls, the method is restricted to single-story buildings which limits its usability without laborious manual separation of the point cloud data into separate stories. Additionally, the final wall elements are generated in a post-processing step without being integrated into the used optimization framework which may lead to locally implausible results. Other methods (Liu et al., 2018, Murali et al., 2017) aiming at reconstructing BIM models make the severe assumption that walls are positioned in a Manhattan world constellation which is often violated in real-world buildings.

Our proposed method overcomes limitations of previous approaches by alleviating the requirements on the input data and by providing a flexible optimization framework for indoor building reconstruction. Some prior methods (Mura et al., 2016, Ochmann et al., 2016) require separate scans and scan positions to derive an initial segmentation into rooms. In contrast, our fully automatic room segmentation approach does not depend on the availability of such information and does not impose particular rules for scanning (e.g. one scan per room). We base our optimization approach on integer linear programming (ILP) which has recently been proposed for polyhedral modeling (Nan and Wonka, 2017) and modeling of building exteriors (Kelly et al., 2017). In addition to solving an energy minimization problem, as often done using Graph-Cut based methods, this formulation provides flexible means to steer the reconstruction using a ruleset of hard constraints that the model is guaranteed to fulfill. Additional constraints can be added interactively to further guide the reconstruction process and edit the resulting model. While some previous approaches regularize the resulting model based on room boundary complexity (Mura et al., 2016, Oesau et al., 2014, Turner et al., 2015), they fail to account for dependencies between related surfaces belonging to opposite sides of the same wall. Our formulation based on volumetric entities enables better regularization of the model with respect to the actual volumetric walls and slabs used to represent the building. In contrast to any previous approach, our method automatically reconstructs the complete geometry of walls and slabs from complex, multi-story point clouds, including volumetric intersections which allow for direct generation of plausible BIM/IFC models. We evaluate our method on various real-world datasets, compare our results with a closely related approach (Ochmann et al., 2016) and with a manually created IFC model, and demonstrate interactive editing capabilities. In summary, our contributions are:

  • 1.

    A combination of automatic multi-story, multi-room reconstruction with a completely volumetric representation of complex rooms, walls, and wall connections.

  • 2.

    Automatic room segmentation of oriented, but otherwise unstructured, multi-story point clouds by means of Markov clustering without prior knowledge about the number or layout of rooms.

  • 3.

    A versatile ILP approach incorporating hard constraints that the reconstruction is guaranteed to fulfill, enabling modeling of complex interactions between reconstructed elements and interactive editing.

Section snippets

Related work

Research on scan-to-BIM and related approaches led to a wide range of developments in recent years and still is a current topic of ongoing work. We now give an overview and compare key features of the most closely related methods in Table 1.

Some methods aim at the generation of 2D floor plans. Okorn et al. (2010) model 2D floor plans by projecting detected structures into the horizontal plane and performing wall segment detection using the Hough transform. Ambruş et al. (2017) reconstruct floor

Overview

The input of our approach is a 3D indoor point cloud (Fig. 1 a) with known “up” direction and oriented normals (i.e. oriented towards the room interior). This information is often available from the data when using terrestrial laser scanners.

We first detect planes using an efficient RANSAC implementation (Schnabel et al., 2007) (Fig. 1 b) and compute occupancy bitmaps (i.e. a coarse grid on each detected plane in which each cell or pixel has value 1 iff at least one point lies within it, see also

Method

In this Section, we detail each of the steps of our approach with a focus on the formulation as an optimization problem.

Evaluation

We evaluate the reconstruction quality and performance on a variety of datasets, and show comparisons with groundtruth IFC and related work. We also exemplify the ability to modify the reconstruction in an intuitive, interactive manner.

Datasets. We used a variety of real-world datasets and one synthetic dataset for evaluation. Table 2 shows six multi-story point clouds measured using terrestrial laser scanners. These datasets were provided by The Royal Danish Academy of Fine Arts Schools of

Conclusion and future work

We have presented a novel approach to tackle the indoor building reconstruction problem from point clouds using integer linear programming. In contrast to previous methods, our approach reconstructs fully volumetric, interconnected walls and room topology on multi-story buildings with weak assumptions on the input data. The resulting models are very close to the requirements needed for Building Information Modeling tasks including volumetric representations of room spaces, walls, and their

Acknowledgments

We acknowledge the Visualization and MultiMedia Lab at University of Zurich (UZH) and Claudio Mura for the acquisition of the 3D point clouds, and UZH as well as ETH Zürich for their support to scan the rooms represented in these datasets. Their datasets were used in our evaluation (Fig. 6). We also used datasets provided by The Royal Danish Academy of Fine Arts Schools of Architecture, Design and Conservation (CITA) (Table 2), the Luleå tekniska universitet (LTU) (Fig. 9), from The ISPRS

References (54)

  • D. Bobkov et al.

    Room segmentation in 3D point clouds using anisotropic potential fields

  • Y. Boykov et al.

    Fast approximate energy minimization via graph cuts

    IEEE Trans. Pattern Anal. Mach. Intell.

    (2001)
  • Brodie, S., Hyland, N., Dore, C., O’Keeffe, S., 2017. The bim & scan® platform: a cloud-based cyber-physical system for...
  • A. Budroni et al.

    Automated 3D reconstruction of interiors from point clouds

    Int J. Architect. Comput.

    (2010)
  • L. Díaz-Vilariño et al.

    Door recognition in cluttered building interiors using imagery and LiDAR data

    Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci.

    (2014)
  • Dongen, S., 2000. A cluster algorithm for graphs. Tech. Rep.; Amsterdam, The Netherlands, The...
  • Durable Architectural Knowledge (DURAARK) data repository. 2013....
  • Gurobi Optimization I., 2016. Gurobi optimizer reference manual....
  • Hyland, N., O’Keeffe, S., Dore, C., Brodie, S., 2017. Automatic open standard reporting for dimensional control...
  • J. Jung et al.

    Automatic room segmentation of 3D laser data using morphological processing

    ISPRS Int. J. Geo-Inf.

    (2017)
  • T. Kelly et al.

    BigSUR: large-scale structured urban reconstruction

    ACM Trans. Graph. (TOG)

    (2017)
  • K. Khoshelham et al.

    The isprs benchmark on indoor modelling

    Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci.

    (2017)
  • C. Liu et al.

    Raster-to-vector: revisiting floorplan transformation

  • Liu, C., Wu, J., Furukawa, Y., 2018. Floornet: A unified framework for floorplan reconstruction from 3D scans. arXiv...
  • H. Macher et al.

    From point clouds to building information models: 3D semi-automatic reconstruction of indoors of existing buildings

    Appl. Sci.

    (2017)
  • C. Mura et al.

    Exploiting the room structure of buildings for scalable architectural modeling of interiors

  • C. Mura et al.

    Robust reconstruction of interior building structures with multiple rooms under clutter and occlusions

  • Cited by (148)

    • Indoor functional subspace division from point clouds based on graph neural network

      2024, International Journal of Applied Earth Observation and Geoinformation
    View all citing articles on Scopus
    View full text