Advanced Engineering Informatics 30 (2016) 312–326

Contents lists available at ScienceDirect

Advanced Engineering Informatics

journal homepage: www.elsevier.com/locate/aei

Towards generation of as-damaged BIM models using laser-scanning

and as-built BIM: First estimate of as-damaged locations of reinforced
concrete frame members in masonry infill structures
Reem Zeibak-Shini ⇑, Rafael Sacks, Ling Ma, Sagi Filin
Faculty of Civil and Environmental Engineering, Technion – IIT, Israel

a r t i c l e i n f o a b s t r a c t

Article history: After an earthquake, Terrestrial Laser Scanning (TLS) can capture point clouds of the damaged state of
Received 16 September 2015 building facades rapidly, remotely and accurately. A long-term research effort aims to develop applica-
Received in revised form 16 February 2016 tions that can reconstruct ‘as-damaged’ BIM models of reinforced concrete (RC) framed buildings based
Accepted 1 April 2016
on their ‘as-built’ BIM models and scans of their ‘as-damaged’ states. This paper focuses on a crucial step:
generating an initial ‘best-guess’ for the new locations of the façade structural members. The output
serves as the seed for a recursive process in which the location and damage to each object is refined in
Keywords:
turn. Locating the ‘as-built’ structural members in the ‘as-damaged’ scan is challenging because each
BIM
Earthquake damage
member may have different displacement and damage. An algorithm was developed and tested for the
Laser scanning case of reinforced concrete frames with masonry infill walls. It exploits the topology of the frames to
Reinforced concrete map the original structural grid onto the damaged façade. The tests used synthetic datasets prepared
Search and rescue from records of two earthquake-damaged buildings. In both cases, the results were sufficiently accurate
Structural frame to allow progress to the following step, assessment of the individual structural members.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction For these reasons, researchers have proposed the use of remote
sensing techniques and sophisticated computing methods. One
Structural engineering expertise is essential in search and res- approach has been to provide surveyors on the scene with mobile
cue (S&R) and recovery operations in the aftermath of a major applications for recording data, such as ROVER Ready [4] and
earthquake in an urban area. S&R teams need information about Urban RAT [5]. These tools help surveyors record and transmit
voids that may have formed in buildings that have sustained dam- information, but they do not acquire information directly. Remote
age in order to plan efforts to reach any survivors safely. In the sensing technologies, such as laser scanning and photogrammetry,
recovery phase, structural engineers gather information to assess can be used to automate data acquisition and to rapidly provide
the state of the buildings and the degree of damage they have point cloud and segmented geometry information describing the
suffered. exterior forms of building components after an earthquake [7–9].
To be effective, the information must be gathered rapidly. Yet in However, the point cloud and segmented face geometry cannot
post-disaster situations professional structural engineers are a be analyzed in relation to the pre-existing building components
scarce resource and gathering the information is, in and of itself, without extensive interpretation and information about the build-
a difficult and potentially hazardous activity. Under the standard ing’s original condition. Exterior scanning has two additional draw-
protocols of most countries, in the immediate post-earthquake backs: it lack the semantics needed for damage assessment of the
phase structural engineers inspect each building and classify it as exterior components and it provides no information at all concern-
safe, unsafe or dangerous [1–3]. This is laborious, slow and pro- ing the interior components and spaces of the building.
vides little of the detailed information needed. However, much more detailed and useful information can be
inferred if an ‘as-built’ model compiled using Building Information
Modeling (BIM) [6] technology is used as the starting point for
inference from the point cloud data. BIM models can provide the
⇑ Corresponding author at: Faculty of Civil and Environmental Engineering,
detailed information about buildings’ ‘as-built’ states in the form
Technion Campus, Haifa 32000, Israel.
E-mail addresses: [email protected], [email protected]
of parametric object-oriented models with component, functional
(R. Zeibak-Shini). and other non-geometric information, and they can be used to

http://dx.doi.org/10.1016/j.aei.2016.04.001
1474-0346/Ó 2016 Elsevier Ltd. All rights reserved.
 R. Zeibak-Shini et al. / Advanced Engineering Informatics 30 (2016) 312–326 313

store new information as it is acquired. This has been explored for Some other applications in condition assessment of civil infras-
the purposes of construction management, where systems attempt tructures are reviewed in [23]. However, none of the above
to estimate construction progress based on comparison of scan and attempted to reconstruct parametric building models at the level
image data with ‘as-planned’ BIM models [10,11]. of detail of individual building components.
The authors propose a framework of steps and a suite of algo-
rithms to process information from these two sources to generate
2.2. Damage assessment and modeling
useful structural engineering information for post-earthquake S&R
and recovery efforts. The overall method, described in Section 3
A number of studies have focused on identifying and evaluating
below, aims to generate an ‘as-damaged’ BIM model that contains
structural damage to exterior members by analyzing observed
both the external and internal components of a building, with both
cracks and spalling data. To assess the performance of damaged
structural and building system components. The aim is to provide
reinforced concrete structures, Farhidzadeh et al. [24] proposed a
first responders with information that can guide their S&R efforts,
crack related damage index that is capable of estimating speci-
akin to ‘x-raying’ the damaged building. The information can also
mens’ relative stiffness loss. Paal et al. [25] proposed identification
be used directly to assess the damage, first for structural assess-
of cracking and spalling of reinforced concrete columns. Their work
ment and later for estimating the costs of repair and/or for plan-
was later extended to compute the residual drift capacity of such
ning the demolition.
columns [26,27]. Lattanzi et al. [28] applied photogrammetry tech-
The framework as a whole requires six main modules. Details
niques to identify cracks on bridge columns in structural tests. The
for modules 1 and 6 can be found in [12]. This paper focuses on
feature data of the identified defects were further used to build a
module 3 of the framework, dealing with the challenge of recon-
prediction model for maximum column drift.
structing the BIM model of the exterior facade of a damaged build-
Damage to structures may have causes other than earthquakes,
ing from the point cloud data. The authors developed and tested a
of course, and TLS has been proposed for tasks such as bridge
procedure that prepares an initial estimate of the locations of the
health/damage evaluation [29]. Examples include measuring the
‘as-damaged’ structural frame members on each façade using
thickness of corroded gusset plates of collapsed I-35W Highway
information from the ‘as-damaged’ scan and the ‘as-built’ BIM.
Bridge in Minneapolis, Minnesota [30] and deformation measure-
The algorithm presented is specific to the case of reinforced con-
ment of a Hungarian bridge on the Danube during its load testing
crete framed buildings with unreinforced masonry (URM) walls.
procedure [31]. Tang and Akinci [32] formalized the workflow of
manual procedures of processing PCD for bridge inspection to
enable automation.
2. Literature review
These studies demonstrate that the TLS methods can detect
damage with high accuracy, but that the process depends on scans
2.1. Remote sensing data acquisition for earthquake damage detection
of the pre-event state. In case of a disaster, it is unlikely that a pre-
event scan of the studied area will be available so that the model
Airborne laser scanning technology is useful in post-earthquake
might not be applicable in such circumstances. Furthermore, none
phase for general damage identification at the detail level of the
of these efforts has attempted to reconstruct an ‘as-damaged’ BIM
structure as a whole [13,14]. Terrestrial Laser Scanning (TLS) can
model.
provide more detailed damage information at a much higher reso-
lution than aerial scanning, making it applicable for damage
assessment on building elements. Lindenbergh and Pietrzyk [15] 2.3. Scan to BIM – challenges in semantic interpretation
discussed applications in change and deformation detection using
static and mobile TLS. Both static and mobile TLS were able to doc- The Scan-to-BIM process has been the subject of numerous
ument the spatial geometry of a bridge in high accuracy [16]. studies. Tang et al. [33] surveyed techniques developed in civil
A novel application of TLS for assessing damage to buildings engineering and computer science that can be utilized to automate
dealt with tornado damage in the US [17]. The authors used point the process of creating as-built BIMs. The authors sub-divided the
cloud data (PCD) acquired before and after an earthquake to eval- overall process into three core operations: geometric modeling,
uate how much damage was caused to each building. They were object recognition, and object relationship modeling. They sur-
also able to estimate the path direction and the wind speed of veyed the state-of-the-art methods for each operation and dis-
the tornado by combining data for multiple buildings. Despite cussed their potential application to automated as-built BIM
the progress made in recognition of building objects within PCD creation. They also outlined the main methods used by these algo-
for generic construction management applications [11,18], neither rithms for representing knowledge about shape, identity, and rela-
Kashani et al. nor others who have applied TLS to earthquake dam- tionships. Bosché et al. [34] explored the opportunity for frequent,
age assessment have attempted to compile parametric building detailed and semantically rich assessment of as-built status of MEP
models. works in construction projects by joining three dimensional laser
In addition to laser scanning, terrestrial video and photogram- scanning and 3D/4D BIM models, and Sacks et al. [35] proposed
metry technologies can also generate dense 3D point clouds of a an approach for semantic enrichment that supports object recogni-
scene using approaches such as ‘Structure-from-Motion’ systems tion and object relationship modeling.
[19], Multi-view Stereo methods [20], and others. Some examples: Xiong et al. [36] presented a method to automatically identify
and model the main visible building components of a scanned
 German et al. [7] successfully designed and implemented an indoor environment (walls, floors, ceilings, windows, and door-
algorithm to identify the cracks in concrete columns using video ways). The authors suggested extracting planar patches from a
recordings. voxelized version of the raw point cloud. Their algorithm learns
 Torok et al. [21] used images obtained from an unmanned the unique features of different types of surfaces (vertical, horizon-
robotic platform to similarly identify cracks in the main struc- tal, etc.) and the contextual relationships between them and uses
tural members. this knowledge to automatically label patches as walls, ceilings,
 Yamazaki et al. [22] used photos collected by Unmanned Aerial or floors. The authors also presented an opening detection algo-
Vehicles (UAV) to generate a 3D model of a district damaged in rithm to detect openings in building facades. They overcame the
the 2015 Gorkha earthquake in Nepal. challenge of modeling partially occluded or occupied openings by
314 R. Zeibak-Shini et al. / Advanced Engineering Informatics 30 (2016) 312–326

learning a model of the typical size, shape, and location of openings 3. Overall system framework
from training examples. Their algorithm detects and models
rectangular-shaped openings, which is the predominant shape in Considering the limitations of the state-of-the-art and the avail-
most buildings. It uses features computed from the occlusion labels ability of BIM and laser-scanning technologies, we propose an
and from depth edges to learn a model of openings using a support approach that makes use of the pre-event BIM model in two ways:
vector machine (SVM) classifier. first, to support automated interpretation of the post-event point
Nonetheless, the Scan-to-BIM process is still not considered a clouds and produce a BIM model of the damaged building’s exte-
solved problem, because the scan only contains the geometry rior facades; and second, to generate a large solution space of can-
information, and automated compilation of a semantically rich didate collapsed building models from which the most-likely
BIM solely from the graphical model is very challenging. Semantic collapse model can be selected to predict the interior damage.
information can be inherited if an ‘as-built’ BIM is available, but the The end result of the overall system is a BIM model that includes
Scan-to-BIM task is even more challenging in case of damage rich semantic information describing the exterior and the interior
occurrence, as the ‘as-damaged’ scan represents a different geom- building components in their ‘as-damaged’ state.
etry of the same structure (e.g. openings can no longer be assumed Fig. 1 provides a high-level overview of the approach. The figure
to be rectangular-shaped features) and matching it with the ‘as- breaks the process into six main processing modules, each of which
built’ is an additional hurdle. is described briefly below. Modules 1 and 6 are detailed in Bloch
et al. [12]; Module 3 is the subject of this paper and is described
in detail in the following sections.
2.4. Skeletonization and model matching In the first module, performed before a possible earthquake, a
structural analysis ‘collapse engine’ software is used to generate
Research towards Scan-to-BIM can also draw on knowledge a large database of possible damaged or collapsed building models.
developed in computer science research concerned with matching The ‘as-built’ BIM provides the input of the building’s structural
models of different states of the same structure. The majority of components and of any building system components that might
this work is in the field of body motion and non-rigid structures. influence the structural behavior. A range of ground motions and
Skeletonization is a popular method to explore the topological directions are used to attempt to provide broad coverage of the
structure of shapes. It is widely used in research of hand motion range of possible cases. Given that time is not specifically limited,
capture [37], in which a hand model is built as a set of rigid seg- in practice extremely large solution sets can be generated and
ments connected by a hierarchical configuration of joints. The stored.
joints have determinate rotational degrees of freedom (DoF) The second module begins in the immediate aftermath of the
according to the physical constraints of a human hand; different earthquake with laser-scanning or photogrammetry of the dam-
rotational constraints are applied to different joints. The move- aged or collapsed building. Scanning is performed within a short
ment of a joint is represented as the rotation relative to its parent time, and provides point clouds registered in the relevant local
joint, up to the root joint, which is close to the wrist. Based on cadastral coordinate system (the same system in which the ‘as-
these kinematic features, given the parental joints’ position, the built’ BIM model is registered). The point clouds are then seg-
position of the end effectors (ends of the fingers) can be identified. mented using commonly available algorithms to identify planar
Alternatively, given the position of the end effectors, the move- surfaces.
ment of parental joints can also be estimated using an inverse Modules 3, 4 and 5 are run recursively, adding new BIM objects
kinematic method [37]. to the ‘as-damaged’ model until all of the external ‘as-built’ BIM
Research on matching different representations of the same objects have been matched to point cloud segments or determined
object from images generally falls into one of two categories: to be missing, in whole or in part (i.e. they have been destroyed,
Laplacian-based mesh-to-mesh matching for objects that are sub- damaged or they are occluded from the scans). Module 3 identifies
jected to non-rigid body deformation [38,39] and piecewise the structural grid, defined in the as-built BIM, in the as-damaged
shape-to-shape matching for objects that are subjected to rigid point cloud data, resulting in an initial estimate of the new loca-
body motion [37,40]. While rigid body motion may be suitable tions and orientations of the structural frame members.
for initial representation of typical damage modes of reinforced In Module 4 the goodness-of-fit between pairs of scan segments
concrete building components [41], building components can also and the ‘expected faces’ of BIM objects is evaluated. The ‘expected
break into pieces, and this obscures the shape-to-shape corre- faces’ are aggregations of the contiguous surfaces of ‘as-built’ BIM
sponding relationship. In addition, the damage propagation pattern objects that may be expected to appear as segments in the point
of building components is not pre-determined, which makes piece- clouds, such as occurs when reinforced-concrete components are
wise matching unpredictable. Thus, none of the above methods is cast monolithically with one another and they have continuous
directly applicable to this problem. and adjacent (coplanar) exterior surfaces. In each iteration cycle,
Three relevant conclusions pertaining to earthquake S&R and the pair with the best fit is used: the location of cracks or breaks
recovery operations are drawn from the literature review: on the BIM objects is determined, and a new ‘as-damaged’ repre-
sentation of the original BIM object is generated within Module 5.
 The primary remote-sensing technologies available for acquir- The new object is added to the ‘as-damaged’ BIM model and the
ing information about earthquake-damaged buildings are ter- process flow returns to the location estimation step (Module 3),
restrial laser scanning and image processing through video- or which is now based on a more accurate set of starting data.
photogrammetry. In the sixth module, a matching algorithm based on minimum
 The data collected is insufficient for compilation of building distances is applied to evaluate the goodness of fit of each of the
information models that provide semantic information about candidate collapse models (generated in Module 1) to the
the building’s components. This requires BIM model data of ‘as-damaged’ BIM model. The model with the best fit is selected
the pre-event state, but this has not been attempted. to support search and rescue operations. Finally, for the recovery
 None of the research efforts reviewed made any attempt to esti- operations, the ‘as-damaged’ model can be compared directly with
mate the ‘as-damaged’ condition of the building’s internal the ‘as-built’ model to support quantification of the damage.
components.
 R. Zeibak-Shini et al. / Advanced Engineering Informatics 30 (2016) 312–326 315

Pre-event Post-event Earthquake event, 2 Acquire and Pre-

Start Start damaged building process Scans

1 Pre-process
BIM model As-built BIM
(structural collapse)
3 Locate BIM
objects in the Scan
No

All external
6 Select full BIM objects
damaged BIM model matched? 4 Identify best-fit
BIM objects
Yes

Search and Rescue

operations As-damaged 5 Reconstruct
exterior BIM BIM objects

End

Main Process Condition Event or operation

BIM Model
Step

Fig. 1. System overview.

4. Scope – building type and damage mode joints. As the next sub-section explains, this characterization of
the joints is useful for damage representation as structural mem-
4.1. RC framed buildings with URM infill walls bers tend to shear on or near the planes of the joints.
The structural frame of the exterior facades can be considered
In terms of building structure, this research focuses on rein- as a grid composed of cells. The structural cells consist of beams,
forced concrete (RC) framed buildings with unreinforced masonry columns, joints and any URM infill between them. A structural cell
(URM) infill walls. RC framed structures are the most common may or may not contain voids in the infill. The cells can be aggre-
structure type for multi-story buildings in many countries [42]. A gated as bays (vertically) or as floors (horizontally).
typical frame unit of this kind of structure consists of two beams
(top and bottom), two columns (left and right) and a panel of
infilled masonry, as shown in Fig. 2. The structural grid can be 4.2. Damage modes considered
characterized as four linear segments (representing the four struc-
tural members) connected by four box-shaped beam–column Liel and Lynch [43] categorized earthquake damage types on RC
framed buildings with URM infills into five states: ‘Negligible’,

Joints

Top Beam

Left Right
Masonry Infill
Column Column

Bottom Beam

(a) (b)
Fig. 2. (a) RC framed buildings with URM infills. (b) A structural cell within the building frame and its members’ topological configuration.
316 R. Zeibak-Shini et al. / Advanced Engineering Informatics 30 (2016) 312–326

‘Insignificant’, ‘Moderate’, ‘Heavy’, ‘Collapse’. This research studies (3) reconstruct the beams and columns along the axes by using
RC framed buildings with URM infills that have undergone the dimensions of those elements from the original
moderate to heavy damage. The ‘collapse’ damage state is not structure.
considered.
Damage such as structural failure of beams and columns or out- In step 1, since the topology of the structural grids remains lar-
of-plane failure of URM infills is considered. Total collapse of the gely unchanged before and after the earthquake at the levels of
building or its parts in case of a soft story where RC members damage considered, an initial estimate of the damaged structural
are totally demolished is outside the research scope. However, par- grid is compiled by scaling and deforming each of the as-built
tial collapse of stories as a result of short-column failures is consid- façade grids to match the extents of each façade as defined by
ered within scope. Structural failures in beams and columns are the point cloud. Because the as-damaged point cloud compiled in
more likely than failure within the volumes of the joints. Joints Module 2 of the overall system (Fig. 1) is registered to the global
might move or rotate, whereas beams or columns tend to break location and orientation of the building, the outline of the struc-
at their interface with the joints or within their lengths. tural frame for each façade can be mapped directly onto the corre-
After an earthquake, the masonry may collapse out-of-plane sponding outline of the façade in the point cloud. The as-built grid
partially or completely, creating new voids or enlarging existing is overlaid onto the as-damaged façade point cloud by linearly scal-
voids. In the majority of cases, the top edges of new and/or ing its overall dimensions.
pre-existing voids align with the bottom face of the top beam, At this stage the grid is proportionally scaled within the extents
because masonry walls require beams or lintels above openings of the damaged façade but it does not yet account for damage
and because masonry infills always collapse downwards. The within the façade. It is therefore only useful for slight-to-
beams and columns may be bent or sheared, but in most moderate damage where the RC frame of the building remains
cases the topological relationships between the columns approximately the same. In order to treat cases of heavy damage,
and the URM/void space remains unchanged. Fig. 3 shows such as soft story collapse or pancaking, this initial estimated grid
some typical examples of the kinds of damage that are must be deformed further. This is done in step 2, using information
considered. from the voids in the façades.
As will be seen, these features can be leveraged to support In step 2, the boundaries of all of the void regions in each façade
matching the as-built and as-damaged models. The approach are automatically derived from the range image of the façade scan.
described in the next section exploits these geometrical, topologi- The void regions represent holes in the façade (which may be win-
cal and behavioral features of RC frames with URM infills to iden- dows, other openings, or regions where masonry has collapsed).
tify an initial estimate of the damaged structural grid by mapping Any background points in the scan that are more than a threshold
the as-built grid onto the as-damaged scan. depth from the façade surface (set to approximately twice the
known wall thickness) are removed from the range image. The
5. Proposed procedure top edges of each void are used as anchors for the grid deformation
because, as previously mentioned, wherever masonry infills have
The goal of the procedure described here and in the test cases in collapsed and created voids, the top edges of the voids are usually
Section 6 (module 3 of the overall system detailed in Fig. 1) is to coincident with the bottom edges of the beams above them in the
generate an initial estimate of the exterior structural frame of cell. The horizontal grid lines are therefore re-aligned to be parallel
the damaged building, as a ‘‘best guess” of the location and orien- to the void top edge lines and offset at a distance from them equal
tation of concrete elements in the scan. The flowchart presented in to the depth of each associated beams. This results in a more accu-
Fig. 4 outlines the procedure. The identities and dimensions of all rate estimate of the as-damaged structural grid. This grid may itself
RC elements forming the structural frame are read from the as- be imperfect, but as will be shown in the test cases which follow, it
built BIM. This information includes the distances between col- is likely to be sufficient to serve as the seed for the following iter-
umns and between beams and the number of bays and floors in ative steps of the procedure (as described in Section 3 above).
the façade. The matching and mapping procedure is done in three
automated steps: 6. Test cases

(1) map the original structural grid of each façade onto the The procedure for Module 3 was tested on two case studies of
range image of that façade using its boundary extents and buildings damaged in different earthquakes in Turkey. The struc-
linear interpolation for all interior axes, tural systems of both buildings consisted of RC frames with URM
(2) improve accuracy by adjusting the horizontal grid axes to infill walls made of hollow clay-tile masonry blocks. The first
align their segments with the top edges of the voids identi- building sustained moderate damaged and the second sustained
fied in the range image in each cell, heavy damage including a soft story collapse of its first floor. The

Fig. 3. Typical out-of-plane failure of URM infills in RC framed building, showing top-down collapse creating new voids that reveal the underside of the RC beams.
 R. Zeibak-Shini et al. / Advanced Engineering Informatics 30 (2016) 312–326 317

Extract the Extract object

As-built BIM structural grid dimensions

Void-driven
Map as-built grid to deformation of
as-damaged the mapped
extents structural grid

As-damaged Extract the

façade’s extents Reconstruct BIM
Scan objects in as-
damaged frame

Extract voids

Initial estimate of the

damaged façade
BIM

Main Process Condition Event or operation

Step BIM Model

Fig. 4. Flowchart of Module 3, for locating BIM objects in the scan.

inputs for the procedure, including the full ‘as-built’ BIM models
and synthetic point clouds of the damaged buildings, were pre-
pared using the methods and software detailed in Ma et al. [44].
The outputs are BIM models of the as-damaged state of each
building in IFC format. To validate the procedure, the outputs were
compared to the original pre-prepared ‘as-damaged’ models.
As will be seen in the case studies that follow, mapping the
structural grid of the original structure onto the as-damaged scan
is fairly straightforward for buildings with moderate damage, but
significantly more challenging for the case of heavy damage.
Although for the case of moderate damage it would seem that
the latter step – mapping using voids to identify the structural grid
– might be considered superfluous, the proposed procedure does
not differ with respect to damage severity.

6.1. Case 1: Kocaeli residential building – moderate damage

The first case study is a residential building damaged in a mag-

nitude 7.6 earthquake in Kocaeli, Turkey, on August 17th 1999. The
information regarding this building before and after the earth- Fig. 5. A photograph of a façade of the Kocaeli residential building [45].
quake was obtained from the EERI database [45]. To begin with,
a complete ‘as-built’ BIM model was prepared in Tekla Structures
20.0 using the 2D drawings and other information regarding the software with an angular resolution of 0.02°, yielding a point cloud
building components available in the EERI website, and exported of some 700,000 points (Fig. 6b). Thus, this model served two
as an IFC file. purposes: it represents the ground truth for the experiment and
Next, a model of the damaged building façade was prepared by it is the source data for scanning, since no laser scans of the real
a modeler by clipping and displacing the structural elements damaged building were available. Details of the procedure for
according to the evidence provided by the available photographs preparing the synthetic data is provided in Ma et al. [44].
(such as Fig. 5). Fig. 6(a) shows the façade modeled in the BIM soft- The virtual scanner was positioned directly in front of the
ware. The model was then scanned using laser scanning emulation selected façade of the building, viewing only that façade. Using
318 R. Zeibak-Shini et al. / Advanced Engineering Informatics 30 (2016) 312–326

Fig. 6. The as-damaged façade: (a) user prepared BIM model; (b) simulated laser-scan point cloud.

the same scanner viewport, all components visible to the scanner range image. The grid’s four corners are identified in the range
were then extracted from the as-built BIM IFC file (Fig. 7a). Next, image as control points for the mapping procedure. All other inter-
the beams and columns that form the structural frame were fil- mediate cell corners along the outer boundary of the grid are
tered from that file (Fig. 7b) and the floors, bays and structural grid approximated in the scan using linear interpolation to keep the
cells were identified and labeled. The dimensions of the beams and same relative division into the same number of bays and floors.
columns and of the grid itself were also extracted from the as-built Intersections between gridlines connecting those interpolated
BIM. points define the interior cell corners. Connecting all interpolated
The structural cells are readily identifiable in the as-built BIM corners results in a scaled mapping of the as-built grid onto the
model, because the beams and columns are modeled explicitly. range image (Fig. 8b).
On the other hand, the structural cells in the ‘as-damaged’ scan Where damage is heavy, however, this mapping is unlikely to be
cannot be identified easily because the texture-less building façade adequate. Beams and columns might bend or break in some of the
(with plaster on the surface) makes it difficult to distinguish the structural cells, reducing their height and distorting their rectangu-
beams, columns and infills within the façade because their exterior lar shapes. The proportionally scaled as-built grid is an imperfect
surfaces are all coplanar. The following paragraph details the skeleton of the damaged building because it is insensitive to the
reconstruction of the ‘as-damaged’ structural frame for this build- internal geometric changes the façade may have undergone.
ing using the procedure outlined in Section 5 above. Instead, its gridlines must be deformed to follow key features that
First, for the sake of efficient data processing, the point cloud is represent the new geometry.
converted into a range image (a significantly more compact data Voids within URM infills provide distinct features within dam-
representation) (Fig. 8a). The range image represents the point aged cells and are the key to a better mapping. The masonry infill
cloud in a spherical coordinate system rather than the Euclidean walls are prone to collapse as a result of natural disasters, thus
system. The first pass attempt to determine the new locations of forming new voids or merging existing ones. Three main possible
the beams and columns in the as-damaged state is achieved by cases might occur: voids in the ‘as-damaged’ model represent
proportional scaling and mapping of the as-built grid into the ‘as-built’ openings (with or without change in geometry and size);

Fig. 7. The as-built façade: (a) RC structural frame model with URM infills; (b) RC structural frame without infills.
 R. Zeibak-Shini et al. / Advanced Engineering Informatics 30 (2016) 312–326 319

Fig. 8. (a) Range image of the ‘as-damaged’ scan; (b) first pass mapping of the as-built structural grid (blue) onto the as-damaged range image. (For interpretation of the
references to colour in this figure legend, the reader is referred to the web version of this article.)

neighboring ‘as-built’ voids within the same cell are merged; or, cells from the scaled grid. Very small voids (with an area smaller
new voids are created as a result of collapse of masonry. In all three than a threshold value) are neglected.
cases the voids reveal the lower edge of the upper beam in each Table 1 shows the association results and percentage of maxi-
cell. Identification of voids’ top edges can help locate the beams mum overlap between the voids and the associated cells in the
and consequently other structural members of the cell connected façade of the Kocaeli residential building. Each void was associated
to it, thus deriving a form topologically similar in configuration with the cell with which it had the highest proportion of overlap-
to the as-built grid. ping area. For this building, 29 voids were associated with 36 cells.
The voids can be extracted by tracing the boundaries of Manual inspection of the results showed 100% success of the asso-
‘‘no-return” regions in the range image of the scan (Fig. 9a). The ciation. The numbering schemes are shown in Fig. 10.
top-left, top-right, bottom-left and bottom-right corners of each The new base levels of damaged beams are derived by extend-
void are determined using its bounding box, which in turn defines ing and connecting the top edges of the identified voids belonging
the top, bottom, right and left sides of the void (Fig. 9b). to the same cell, as is shown in Fig. 11. Each of the resulting poly-
Each damaged structural cell may contain more than one void lines is then shifted upwards by the beam’s height (which is known
and the correspondence between voids and structural cells is not explicitly from the ‘as-built’ BIM) to determine the top levels of the
obvious because the sources of the voids are unknown. To over- beams. Where no voids exist in a cell, the polylines of neighboring
come this apparent problem, the scaled grid serves as a first guess cells are extended to locate the joints and the beams are assumed
to associate voids to their corresponding cells by overlapping the to connect those joints. This makes the default assumption (for lack
scaled gridlines with the void regions. The correspondence is of any information at this stage other than the observation that the
derived based on criteria of overlapping area between voids and masonry infill in the cell is apparently intact) that the earthquake

Fig. 9. (a) Bounding box and four corners of the ‘as-damaged’ voids in the range image. (b) Bottom, top, right and left sides of the voids in yellow, green, blue and red
respectively. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
320 R. Zeibak-Shini et al. / Advanced Engineering Informatics 30 (2016) 312–326

Table 1 plane is parallel to the global façade plane (i.e. zero axial rotation
Void to cell association using maximum proportional overlapping area. is assumed). Each beam and column is thus placed with its center
Void ID Cell ID Max. overlap% (threshold 50%) of gravity placed at its associated LRF, its major axis lies along the
5 [1] 100.0 LRF X axis, and it is not rotated about the major axis.
15 [2] 100.0 Finally, given that each ‘as-damaged’ member is already associ-
17 [2] 100.0 ated with it corresponding ‘as-built’ member, all of the alphanu-
14 [3] 100.0 meric property data are inherited accordingly.
4 [7] 98.2
13 [8] 93.6
12 [9] 100.0
18 [10] 98.4 6.2. Case 2: Bingol school building – heavy damage
22 [11] 97.4
27 [12] 98.8 The second case study is a school building damaged in a magni-
3 [13] 98.5
tude 6.4 earthquake in 2003 in Bingol, Turkey, which sustained
11 [14] 93.8
10 [15] 99.0 heavy damage, as can be seen in the photograph shown in
21 [16] 98.5 Fig. 13a. In addition to the collapse of most of the URM infill walls,
24 [17] 97.7 a soft story collapse occurred and the first floor of the building pan-
26 [18] 98.2
caked. Many beams and columns were heavily damaged, undergo-
2 [19] 99.9
9 [20] 99.0 ing translation and rotation and some columns had sheared. The
8 [21] 100.0 information regarding this building before and after the earth-
20 [22] 99.6 quake was obtained from the EERI database [46]. Fig. 13b shows
23 [23] 99.4 the BIM model prepared from the drawings.
28 [24] 99.6
As explained in Sections 5 and 6, the same procedure is fol-
1 [25] 100.0
7 [26] 100.0 lowed. The text that follows therefore presents the steps for the
6 [27] 100.0 Bingol case study without repeating the details (see Fig. 14)
19 [28] 100.0 The simulated scan was performed with approximately the
25 [29] 100.0
same point of view as the as-damaged photograph (Fig. 13a), view-
29 [30] 100.0
16 [32] 100.0
ing two of the building’s façades in angular resolution of 0.02°. The
point cloud contained some 740,000 points. Fig. 15 presents the
point cloud in spherical coordinate system as a range image.
The BIM model was filtered first to remove elements that were
induced motion has not deformed the beam above that cell nor dis- outside the scan point of view or completely occluded (Fig. 16a),
connected it from the joints. and secondly to leave only RC structural frame members
The positions of beam–column joints are estimated based on (Fig. 16b). The structural grid, with three bays and three floors
the beam and column widths extracted from the ‘as-built’ BIM on each of two façades, was identified from the as-built model.
and on the positions of the damaged beams in all the structural On the other hand, the overall boundary of the damaged build-
cells. The center axes of the columns in each cell are estimated ing’s facades can be derived based on the extent of the
by connecting these joints. The resulting first estimate of the struc- ‘as-damaged’ scan. Due to the partial collapse of the ground floor
tural grid of the damaged building is shown in Fig. 12a. in the case study building, the overall height is smaller than it
The refined beam gridlines represent the estimated top edges of was before the damage. Thus, the ‘as-built’ structural grid is used
the as-damaged beams, and the column gridlines locate the center- proportionally to divide the ‘as-damaged’ scan into initial cells,
lines of the as-damaged columns. Using the as-built dimensions as shown in Fig. 17.
and the resulting as-damaged grid, the beams, columns and joints Note that the initial gridding of the scan did not match the new
were finally placed in their new estimated locations and orienta- damaged structural frame of the building due to the soft story
tions. This is done by defining a Local Reference Frame (LRF) on collapse, where the original proportions of floors’ height was not
each gridline, such that the origin is located at the center point conserved. Therefore, in this case, identifying the voids in the as-
of the grid line, the X direction is along the grid line, and the XZ damaged scan was essential for improving the mapped grid. The

[33]
[36] [35] [34] [31] 16
[32]
[27] 29 25 19 6 1
[30] [29] [28] [26] [25] 7
[21] 8
28 23 20 2
[24] [23] [22] [20] [19] 9
[15] 26 24 21 10
[18] [17] [16] [13] 3
[14] 11
[9] 12
[12] [11] [10] [7] 27 22 18 4
[8] 13
[6] [5] [3] 14 5
[4] [1]
[2] 15 17
(a) (b)
Fig. 10. (a) Cell numbering; (b) void numbering.
 R. Zeibak-Shini et al. / Advanced Engineering Informatics 30 (2016) 312–326 321

way as described in Section 6.1 above. The new end points of the
vertical gridlines were established by assuming that the endpoints
of the beam axes of adjacent cells were equidistant from the new
column centerlines, as they were in the as-built state (i.e. joint
zones are not deformed). Fig. 20 shows the deformed and refined
versions of the initial grid that represents the as-damaged struc-
tural frame.
As for case 1, the refined beam gridlines represent the estimated
top edges of the as-damaged beams, and the column gridlines
locate the centerlines of the as-damaged columns. The result is
shown in Fig. 21. Finally, as before, the LRF is defined for each grid-
line, the corresponding members are placed accordingly, and the
alphanumeric data is copied.

6.3. Performance

The procedure described above was programmed using Matlab

7.12.0 and run on a computer with IntelÒ coreTM i5 CPU processor.
The total runtime for each case study (including scan processing,
void extraction, grid mapping, void-to-cell association, gridline
deformation and reconstruction of the beams and columns) was
on the order of magnitude of 2 min. This runtime is very short in
relation to the scanning itself and will therefore not detract from
the feasibility of the system in the aftermath of an earthquake.
Fig. 11. Identification of voids’ top edges in each structural cell.
7. Verification

voids and their top edges were identified in the range image, as can The output of the procedure is an IFC file containing the struc-
be seen in Fig. 18. tural members of the façades in their best-estimated locations and
Table 1 shows the association results and percentage of maxi- orientations, yet with their original cross-section shapes and
mum overlap between the voids and the associated cells. In sum- lengths. Thus the mapping derived at this point is only an initial
mary, 75% (15 out of 20) of the voids are wholly contained estimate of the damaged state that will be refined through recur-
within their corresponding cells. The remaining voids each extend sive iteration of the solution process as a whole (as defined in Sec-
over more than one cell. In this case study building, using a thresh- tion 3 and illustrated as steps 4 and 5 in Fig. 1, and beyond the
old of 50% overlap for association, all 20 voids were associated with scope of this paper). The outputs for both buildings were verified
18 cells (see Fig. 16 and Table 2). As for the Kocaeli case study, by visual inspection of the number of members correctly placed
manual inspection of the results showed 100% success for the asso- and by the use of Boolean geometry operations to measure the
ciation (see Fig. 19). overlapping volumes of the output and the benchmark as-
Where more than one void occurred in a single cell, all of the damaged models that were prepared by the researchers. The over-
top edges of the voids were linked with straight line segments to lapping volumes were used to calculate the degrees of precision
form a polyline. This polyline represents the best estimate of the and of recall.
new location of the bottom edge of the upper beam in the cell. Fig. 22 shows the output BIM model, the original as-damaged
The grid was then relocated according to the polylines in the same BIM model prepared by the researchers, and a superposition of

Fig. 12. (a) Refinement of the initial damaged skeleton (initial skeleton in blue, refined skeleton in yellow); (b) reconstructing the as-damaged RC frame (beams in green,
columns in red and joints in blue). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
322 R. Zeibak-Shini et al. / Advanced Engineering Informatics 30 (2016) 312–326

Fig. 13. (a) A photograph of a corner of the damaged Bingol school building [46]; (b) as-built BIM model of the building.

Fig. 14. As-damaged state: (a) user pre-prepared BIM model; (b) scanned point cloud.

Fig. 15. Range image of the as-damaged scan. Gray scale represents range values – Fig. 17. Initial cells in the ‘as-damaged’ scan derived by scaling and mapping the
distant points are lighter, darker points are closer to the viewer. Black pixels ‘as-built’ grid onto the range image (in blue). (For interpretation of the references to
represent no-data. colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 16. The ‘as-built’ model: (a) RC frame with URM infills. (b) RC structural frame with beams and columns only.
 R. Zeibak-Shini et al. / Advanced Engineering Informatics 30 (2016) 312–326 323

Fig. 18. (a) Bounding boxes and corners of the voids in the range image. (b) Bottom, top, right and left sides of the voids in yellow, green, blue and red respectively. (For
interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

the event will be reconstructed using the same proportional rela-

Table 2 tionships as the original.
Void to cell association based on maximum overlapping percentage.
Similar inspection of the results for the Bingol school building
Void ID Cell ID Max. overlap% (threshold 50%) shows that 42 out of 45 structural components were identified cor-
1 [1] 100.0 rectly. As can be seen in Fig. 22c, some of the columns in the first
2 [2] 100.0 floor have sheared and their resulting parts overlap one another
3 [14] 100.0 along part of their length. However, the procedure models them
4 [8] 92.4
with their original lengths. Reconstruction of the columns includ-
5 [15] 100.0
6 [9] 66.5 ing identification of any breaks and other distortions of their
7 [3] 100.0 shapes is the subject of the next step in the overall system (Module
8 [4] 100.0 4 in Fig. 1) and beyond the scope of this paper.
9 [16] 100.0 The overlapping volume for each building, representing the
10 [10] 99.6
11 [4] 100.0
‘true positive’ (TP) result, was computed by first performing a Boo-
12 [16] 100.0 lean union of all of the beams and columns in each model (output
13 [11] 94 and pre-prepared benchmark) and then a Boolean intersection of
14 [17] 100.0 the two resulting solids. The false negative (FN) is the volume of
15 [5] 100.0
the benchmark model that remains when the output model is sub-
16 [18] 100.0
17 [12] 100.0 tracted from it (i.e. the portion of the true result that was not iden-
18 [6] 100.0 tified correctly). Similarly, the false positive (FP) is the volume of
19 [12] 90.9 the output model that remains when it is subtracted from the
20 [18] 100.0 benchmark model. Precision measures the proportion of the
true result that was predicted correctly and is computed as
TP/(TP + FP), whereas Recall measures how much of the output
the two models. Visual inspection shows that for the Kocaeli resi- was accurate and is computed as TP/(TP + FN).
dential building 70 out of 72 structural components were identi- The results of the Boolean operations and of the Precision and
fied correctly. Two mid-story beams that support the staircase Recall calculations are provided in Table 3. As expected, predicted
landings were placed incorrectly. This error results from the fact positive and the actual positive values for the Koaceli residential
that the beams are not located directly above the small openings building (case 1) are similar because both the output and the
below them, as the procedure assumes. This type of error might benchmark models contain the same structural RC members and
be overcome in the future by identifying such special situations damage is slight. This is not the case for the Bingol school building,
and applying a change detection procedure to infills and voids. A because some of the parts of the heavily damaged columns have
cell that consists of the same geometry of infills and openings after been grossly displaced. Nevertheless, the results for precision

Fig. 19. Void numbering (no brackets) and cell numbering (in brackets).
324 R. Zeibak-Shini et al. / Advanced Engineering Informatics 30 (2016) 312–326

Fig. 20. (a) Identification and linkage of voids’ top edges in each structural cell; (b) resulting structural grid (initial skeleton in blue, refined skeleton in yellow). (For
interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

(81.98% and 76.57% respectively for the two buildings) confirm

that the procedure is relatively successful at preparing the first
estimate for the overall process. Success is determined by the abil-
ity of the procedure to identify at least one member’s new location
well enough to allow the next steps, in which the shape of each
member will be refined to better fit the real as-damaged shape
as reflected in the PCD. In both cases, more than one member
was very well matched, thus surpassing this requirement.

8. Conclusion

A new, methodical procedure for performing the first step of a

Fig. 21. Reconstructed as-damaged RC frame (beams in green, columns in red). (For method for reconstructing a semantically rich BIM model of the
interpretation of the references to colour in this figure legend, the reader is referred as-damaged state of an RC framed structure that has been dam-
to the web version of this article.)
aged by an earthquake has been developed and tested. The proce-

Fig. 22. Case 1 (top row) and Case 2 (bottom row): (a) Pre-prepared as-damaged model (without infills). (b) Process output – reconstructed as-damaged model. (c)
Superposition of the two models (pre-prepared and reconstructed).

Table 3
Precision and recall results for the two case studies.

Case True Positive False Positive False Negative Predicted positive Precision TP/ Actual positive Recall TP/
(TP) (m3) (FP) (m3) (FN) (m3) (TP + FP) (m3) (TP + FP) (%) (TP + FN) (m3) (TP + FN) (%)
Kocaeli 34.13 7.49 8.12 41.62 81.98 42.25 80.77
Bingol 18.75 5.74 5.19 24.49 76.57 23.94 78.35
 R. Zeibak-Shini et al. / Advanced Engineering Informatics 30 (2016) 312–326 325

dure’s success in locating most of the members correctly for both [11] F. Bosche, C.T. Haas, Automated retrieval of 3D CAD model objects in
construction range images, Automat. Constr. 17 (2008) 499–512.
the moderate and the heavy damage cases strongly suggests that
[12] T. Bloch, R. Sacks, O. Rabinovitch, Interior models of earthquake damaged
development of a methodical procedure for the first phase of buildings for search and rescue, Adv. Eng. Inform. 30 (1) (2016) 65–76.
reconstruction of semantically rich BIM models of damaged build- [13] P.L. Dong, H.D. Guo, A framework for automated assessment of post-
ing façades using TLS is a feasible and promising approach to sup- earthquake building damage using geospatial data, Int. J. Remote Sens. 33
(1) (2012) 81–100.
port planning of S&R and recovery operations after a natural [14] W. Liu, P.L. Dong, J.B. Liu, H.D. Guo, Evaluation of three-dimensional shape
disaster. Two major problems have been tackled in this solution: signatures for automated assessment of post-earthquake building damage,
the lack of any semantic information in the scan and the need to Earthquake Spectra 29 (3) (2013) 897–910.
[15] R. Lindenbergh, P. Pietrzyk, Change detection and deformation analysis using
match the original structural members with the displaced and/or static and mobile laser scanning, Appl. Geomat. 7 (2) (2015) 65–74.
damaged objects that are present in the point cloud data. [16] NDT documentation and evaluation of the roman bridge of lugo using GPR and
The procedure serves as the first module in a broader S&R sup- mobile and static LiDAR, J. Perform. Constr. Facil. 29 (1) (2015) 06014004.
[17] A. Kashani, P. Crawford, S. Biswas, A. Graettinger, D. Grau, Automated tornado
port system these problems. In this procedure the ‘as-built’ BIM damage assessment and wind speed estimation based on terrestrial laser
model is essential for providing information defining the building’s scanning, J. Comput. Civ. Eng. (2014) 04014051.
components. A set of rational assumptions concerning the specific [18] I. Brilakis, M. Lourakis, R. Sacks, S. Savarese, S. Christodoulou, J. Teizer, A.
Makhmalbaf, Toward automated generation of parametric BIMs based on
geometry, topology and behavior of URM infilled RC frame were hybrid video and laser scanning data, Adv. Eng. Inform. 24 (2010)
necessary to facilitate the methods’ steps, which include extracting 456–465.
voids from the scan, mapping the voids to the structural cells, and [19] M.J. Westoby, J. Brasington, N.F. Glasser, M.J. Hambrey, J.M. Reynolds,
‘Structure-from-Motion’ photogrammetry: a low-cost, effective tool for
reconstruction of the structural grid of the damaged building.
geoscience applications, Geomorphology 179 (2012) 300–314.
Application of the proposed approach to two case studies of RC [20] Y. Furukawa, J. Ponce, Accurate, dense, and robust multiview stereopsis, IEEE
framed buildings damaged in the 1999 Kocaeli Earthquake (Case 1) Trans. Pattern Anal. Mach. Intell. 32 (8) (2010) 1362–1376.
and in the 2003 Bingol Earthquake (Case 2) in Turkey, showed good [21] M. Torok, M. Golparvar-Fard, K. Kochersberger, Image-based automated 3D
crack detection for post-disaster building assessment, J. Comput. Civ. Eng. 28
and satisfactory results. However, situations where columns and (5) (2014) A4014004.
beams might shear and bend cannot be identified by the method [22] F. Yamazaki, T. Matsuda, S. Denda, W. Liu, Construction of 3D Models of
because of the explicit simplifying assumption, within the current Buildings Damaged by Earthquakes using UAV Aerial Images.
[23] C. Koch, Z. Zhu, S.G. Paal, I. Brilakis, Machine vision techniques for condition
scope, of rigid body motion. assessment of civil infrastructure, in: Z. Liu, H. Ukida, P. Ramuhalli, K. Niel
Damage modes like members overlapping, out of plane damage (Eds.), Integrated Imaging and Vision Techniques for Industrial Inspection,
of structural building members, etc., must therefore be identified Springer, London, 2015, pp. 351–375.
[24] A. Farhidzadeh, E. Dehghan-Niri, A. Moustafa, S. Salamone, A. Whittaker,
in module 4 of the overall system, which uses local segmentation Damage assessment of reinforced concrete structures using fractal analysis of
of the ‘as-damaged’ scan to identify cracks and breakages in indi- residual crack patterns, Exp. Mech. 53 (9) (2013) 1607–1619.
vidual members. In this way the modules will be deployed recur- [25] S.G. Paal, J. Jeon, I. Brilakis, R. DesRoches, Automated damage index estimation
of reinforced concrete columns for post-earthquake evaluations, J. Struct. Eng.
sively to refine the model of the damaged building façade. In (2014) 04014228.
addition, more cases of damaged buildings will need to be tested [26] S. German, I. Brilakis, R. DesRoches, Rapid entropy-based detection and
to seek evidence for robustness of the procedure for Module 3 properties measurement of concrete spalling with machine vision for post-
earthquake safety assessments, Adv. Eng. Inform. 26 (4) (2012) 846–858.
and of the system as a whole.
[27] Z. Zhu, S. German, I. Brilakis, Visual retrieval of concrete crack properties for
automated post-earthquake structural safety evaluation, Automat. Constr. 20
Acknowledgments (7) (2011) 874–883.
[28] D. Lattanzi, G. Miller, M. Eberhard, O. Haraldsson, Bridge column maximum
drift estimation via computer vision, J. Comput. Civ. Eng. (2015), http://dx.doi.
This research was funded in part by the Insurance Companies org/10.1061/(ASCE)CP.1943-5487.0000527. 04015051.
Association of Israel. The assistance of undergraduate research [29] W. Liu, S. Chen, E. Hauser, LiDAR-based bridge structure defect detection, Exp.
Tech. 35 (6) (2011) 27–34.
assistants Ashrant Aryal and Jiashu Liang is greatly appreciated.
[30] NTSB, Highway Accident Report: Collapse of I-35W Highway Bridge,
Minneapolis, Minnesota, August 1, 2007, Highway Accident Report NTSB/
References HAR-08/03, National Transportation Safety Board Records Management
Division, Washington, DC, 2007, p. 2008.
[31] T. Lovas, Á. Barsi, A. Detrekoi, L. Dunai, Z. Csak, A. Polgar, A. Berenyia, Z. Kibedy,
[1] S. Anagnostopoulos, M. Moretti, Post-earthquake emergency assessment of
K. Szocs, Terrestrial laserscanning in deformation measurements of structures,
building damage, safety and usability—Part 1: Technical issues, Soil Dynam.
Int. Arch. Photogramm. Remote Sens. 37 (2008) B5.
Earthquake Eng. 28 (2008) 223–232.
[32] P.B. Tang, B. Akinci, Formalization of workflows for extracting bridge surveying
[2] FEMA, Evaluation of Earthquake Damaged Concrete and Masonry Wall
goals from laser-scanned data, Automat. Constr. 22 (2012) 306–319.
Buildings: Basic Procedures Manual-FEMA 306, Federal Emergency
[33] P. Tang, D. Huber, B. Akinci, R. Lipman, A. Lytle, Automatic reconstruction of as-
Management Agency, 1988.
built building information models from laser-scanned point clouds: a review
[3] S. Okada, N. Takai, Classifications of structural types and damage patterns of
of related techniques, Automat. Constr. 19 (7) (2010) 829–843.
buildings for earthquake field investigation, Presented at the Proceedings of
[34] F. Bosché, A. Guillemet, Y. Turkan, C. Haas, R. Haas, Tracking the built status of
the 12th World Conference on Earthquake Engineering (paper 0705),
MEP Works: assessing the value of a scan-vs-BIM system, J. Comput. Civ. Eng.
Auckland, 2000.
28 (4) (2014) 05014004.
[4] S. Hunter, L. Kohandoust, H. Schechner, R. Bakerian, Rapid Observation of
[35] M. Belsky, R. Sacks, I. Brilakis, Semantic enrichment for building information
Vulnerability and Estimation Risk (ROVER), Applied Technology Council, 2013.
modeling, Comput. Aided Civ. Infrastr. Eng. (2015).
[5] S. Ploeger, M. Sawada, A. Elsabbagh, M. Saatcioglu, M. Nastev, E. Rosetti, Urban
[36] X. Xiong, A. Adan, B. Akinci, D. Huber, Automatic creation of semantically rich
RAT: new tool for virtual and site-specific mobile rapid data collection for
3D building models from laser scanner data, Autom. Constr. 31 (2013) 325–
seismic risk assessment, J. Comput. Civ. Eng. (2015) 04015006.
337.
[6] C.M. Eastman, P. Teicholz, R. Sacks, K. Liston, B.I.M. Handbook, A Guide to
[37] A. Aristidou, J. Lasenby, Motion capture with constrained inverse kinematics
Building Information Modeling of Owners, Managers, Designers, Engineers and
for real-time hand tracking, in: 4th International Symposium on
Contractors, 2nd ed., John Wiley and Sons, Hoboken, NJ, 2011.
Communications, Control and Signal Processing (ISCCSP), 2010, 2010, pp. 1–5.
[7] S. German, J. Jeon, Z. Zhu, C. Bearman, I. Brilakis, R. DesRoches, L. Lowes,
[38] C. Stoll, Z. Karni, C. Rössl, H. Yamauchi, H.-P. Seidel, Template deformation for
Machine vision-enhanced post-earthquake inspection, J. Comput. Civ. Eng. 27
point cloud fitting, Presented at the Proceedings of the 3rd Eurographics/IEEE
(6) (2013) 622–634.
VGTC Conference on Point-Based Graphics, 2006, pp. 27–35.
[8] C. Schweier, M. Markus, Classification of collapsed buildings for fast damage
[39] M. Ye, X. Wang, R. Yang, L. Ren, M. Pollefeys, Accurate 3D pose estimation from
and loss assessment, Bull. Earthq. Eng. 4 (2) (2006) 177–192.
a single depth image, in: 2011 IEEE International Conference on Computer
[9] L.G. Dong, J. Shan, A comprehensive review of earthquake-induced building
Vision (ICCV), 2011, pp. 731–738.
damage detection with remote sensing techniques, ISPRS J. Photogramm. 84
[40] D. Smeets, J. Keustermans, J. Hermans, D. Vandermeulen, P. Suetens, Feature-
(2013) 85–99.
based piecewise rigid registration in 2-D medical images, in: 2012 9th IEEE
[10] F. Bosche, Plane-based registration of construction laser scans with 3D/4D
International Symposium on Biomedical Imaging (ISBI), 2012, pp. 696–699.
building models, Adv. Eng. Inform. 26 (1) (2012) 90–102.
326 R. Zeibak-Shini et al. / Advanced Engineering Informatics 30 (2016) 312–326

[41] L. Ma, R. Sacks, R. Zeibak-Shini, Information modeling of earthquake-damaged [44] L. Ma, R. Sacks, R. Zeibak-Shini, A. Aryal, S. Filin, Preparation of synthetic as-
reinforced concrete structures, Adv. Eng. Inform. (2015). damaged models for post-earthquake BIM reconstruction research, J. Comput.
[42] P. Haldar, Y. Singh, D.K. Paul, Identification of seismic failure modes of URM Civ. Eng. (2015) 04015032.
infilled RC frame buildings, Eng. Fail. Anal. 33 (2013) 97–118. [45] EERI, Ilgi Moble Building. [Online] Available: <http://db.concretecoalition.org/
[43] Abbie B. Liel, Kathryn P. Lynch, Vulnerability of reinforced-concrete-frame building/135>2014 (accessed 07.10.14).
buildings and their occupants in the 2009 L’Aquila, Italy, earthquake, Nat. [46] EERI, Turkey School Bldgs – Moment Frames. [Online] Available: <http://
Hazards Rev. 13 (1) (2012) 11–23. db.concretecoalition.org/building/132>2014 (accessed 07.10.14).