DURAARK – Deriving architectural information from point clouds

DURAARK – Deriving architectural information from point clouds

SHARE

The European research project, Durable Architectural Knowledge (DURAARK) investigates the automated detection of meaningful architectural information within 3D laser scan point clouds and how to generate semantically rich BIM models for building industry.

Land surveyors stood typically in the beginning of any planning process, whether related to new build or renovation. Their discipline is in its core related to the abstraction of the world’s complexity to understandable information for users that want to process this information. 3D laser scanning makes surveying today extremely fast, precise and as well relatively cheap, when considering the relation of cost and precision gained. Land surveyors are however facing the problem that clients are rarely willing to pay for these products. This is as the product 3D point cloud is not only data-heavy, but is hard to integrate with the internal workflows of their clients in the building practices.

Figure 1: Pointcloud separated with a process developed in the DURAARK research project. Architectural information valuable for the use within Facility Management processes is automatically detected: Spaces, spatial connectivity, doors, area and volume. 3D scan of Bispegaard courtesy of Statsbyg Norway.

These workflows are in the Scandinavian region predominantly based on Building Information Modelling (BIM). Architectural representations that follow a BIM logic consist of discreet elements, structured in a hierarchical logic, as from the overall building to the door handle. 3D geometrical representations are however only one of many other types of information that BIM can hold, further are i.e. element properties, building schedules and links to external libraries and classifications. BIM models are hence semantically rich and linked to live online databases. They are as well customised to specific types of usage, which means that their structure will vary between individual projects and clients.

To derive this data from 3D scans requires architectural knowledge, as the unstructured point clouds have to be interpreted through architectural terms in order to obtain a semantically enriched 3D model, representing both physical and functional characteristics of a physical structure.

Current approaches to generate BIM data from point clouds use 3D modelling functionality in a BIM software with the segmented and classified point cloud as an underlay or by plan- and section views. This ‘process of “as-built” BIM is mainly a manual process that can be tedious, intensive and subjective’ (Hichri, et al., 2013). Commercial software products, which focus on a Scan to BIM process, as Kubit VirtuSurv, approach the modelling of BIM models through the choice of predefined element types and selection of points in the planar image produced by the 3D scanner. In this and similar approaches the user is augmented by software – the main modelling task rests however still on him.

As these processes are not akin to the expertise of land-surveyors, Scan to BIM tasks are usually executed by specialized subcontractors from architecture or oversea areas. These deliver BIM files, which need to be validated by land-surveyors in order to assess whether an agreed structure, level of meta-information and detailing is met.

Research into Scan to BIM

Two of the main challenges tasks in a Scan to BIM process are hence:

  1. To find the right fitting simplified representation in a client specific BIM structure
  2. To evaluate the quality of the product and detect eventual deviations

Research into Scan to BIM investigates approaches that provide eventually fully automated approaches for the detection of architectural features as spaces and discrete building components from point clouds. The aim is to create semantically rich products, where detected elements are rich in information and link for instance back to the original point cloud or to other BIM models and online data.

With the scope to develop more sustainable practices in information handling, the European research project DURAARK (Durable Architectural Knowledge www.duraark.eu) investigates how architectural meaningful information can be found within point clouds and how this can be linked to existing BIM models and semantic information sources in the web.

In a collaboration of the Centre for IT and Architecture (CITA) in Copenhagen and the Rheinische Friedrich-Wilhelms-Universitaet Bonn (UBO) and in exchange with stakeholders, such as the Danish land surveying company LE34 (http://www.le34.dk/) the project developed approaches to tackle the named challenges.

Indoor point cloud structuring

The structuring of indoor point clouds has been a focus and is topic of a recent scientific publication (Ochmann et al. 2014). Apart from a high-level semantic segmentation of the formerly unstructured point clouds into stories and rooms, these methods additionally allow the extraction of attributed graphs in which nodes represent rooms (including room properties like area or height), and edges represent connections between rooms (doors or staircases) or indicate neighborhood relationships (separation by walls). The current research prototype creates BIM models in the IFC format and uses the IFC_Space definitions to store information about geometry, areas and volumes (Fig. 2-7).

Figure 2-7: An example for space segmentation and determination of room relationships: The point cloud before (2) segmentation, initial rough point-to-space assignments (3), final segmentation (4). The affiliations of points to rooms are color-coded. Automatically recognized room neighborhood graphs (5) and room connectivity graphs (6, 7). 3D scan of Bispegaard courtesy by Statsbyg Norway.

This approach is suitable for stakeholders such as institutional building owners. These lack often up to date data and three dimensional data of their building stock. An automated extraction of spatial information and relations from point clouds into BIM formats allows a data creation of sematic 3D data in contemporary facility management system such as DaluxFM (Fig. 8).

Figure 8: Example for an IFC extracted from the point cloud shown in Figure 2-7. One of the IFC_Spaces is highlighted. 3D scan of Bispegaard courtesy by Statsbyg Norway.

Verification of data

The creation of 3D representations puts higher demands on the quality data, which needs to be assessed and documented. In terms of Scan to BIM processes this is in first line an assessment, whether all relevant architectural objects have been modelled. To assess, whether a complex BIM model deviates from the underlying gigabyte big point clouds is technically challenging and part of ongoing research in the DURAARK project.

The prototypical assessment tools operate in two steps: A first registration step of the BIM model and the point clouds, which is followed by an analysis cycle. The first step is necessary, as model and point clouds often deviate in their position in space.

Registration of Point Clouds and BIM

The DURAARK approach uses a variant of the Iterative Closest Point (ICP) as proposed by Besl and McKay (Besl and McKay 1992) for the registration. This approach is semi-automatic in nature and requires a manual, coarse pre-alignment of the building representations. This first manual step is followed by an automatic fine alignment between the different building representations. The used ICP approach requires information about corresponding point pairs in two given (subsampled) representations A and B. In case of two point clouds, this correspondence is obtained by searching for nearest points in representation B from all points in representation A (or vice-versa). In case of registration against a BIM model, the surfaces of the model are first sampled point-wise in order to generate a point cloud representation internally. Subsequently, the method for aligning two point clouds is applied (Fig. 9-14).

The used method is robust and as it “converts” any given BIM Model internally into this 3D representation, it is suitable for the registration of BIM to BIM, Point Cloud to Point Cloud and Point Cloud to BIM models.

Figure 9-14: Registration of a BIM model and point clouds of the same building. First two images: a mesh model generated from an IFC file of the building (overview of one storey and a detail view). The ceiling structure has been removed for visualization. Middle: Corresponding scans taken in the real building. Bottom: The two representations after registration using the developed tools. 3D scan of Risklokka courtesy by Statsbyg Norway.

Analysis of differences between Point Cloud and building elements in BIM

The found corresponding Point Pairs are used in the following analysis step, which detects and highlights differences between multiple (concurrent) representations of the same building or part of building (Fig. 15-16).

Figure 15: Display of difference of modelled object in BIM (blue) and missing elements (red).

References

P. J. Besl and N. D. McKay. Method for registration of 3-d shapes. In Robotics-DL tentative, pages 586{606. International Society for Optics and Photonics, 1992.

Hichri, N., Stefani, C., De Luca, L., Veron, P. and Hamon, G., 2013. From Point Cloud to BIM: A survey of existing approaches. Ecole Nationale Superoeure d’Architecture de Marseille.

Ochmann, S., Vock, R., Wessel, R., Tamke, M. and Klein, R. Automatic Generation of Structural Building Descriptions from 3D Point Cloud Scans. In proceedings of GRAPP 2014 – International Conference on Computer Graphics Theory and Applications, Lisbon, Portugal, SCITEPRESS, Jan. 2014