Environmental Engineering and Management Journal [616784]

Environmental Engineering and Management Journal
June 2016, Vol.15, No. 6, 1293-1303
http://omicron.ch.tuiasi.ro/EEMJ/

“Gheorghe Asachi” Technical University of Iasi, Romania

ANALOGOUS VS. DIGITAL CAMERAS
FOR BUILDINGS 3D MODELS CREATION

Ersilia Oniga, Constantin Chiril ă

„Gheorghe Asachi” Technical University of Iasi, Faculty of Hydrotechnical Engineering, Geodesy and Environmental
Engineering, Department of Terrestrial Measurements and Cadastre, 65 Prof. Dr. Docent D. Mangeron Str., 700050 Iasi,
Romania

Abstract

In this paper, the steps used in the pro cess of building 3D model creation, using images acquired with the UMK 10/1318 metric
terrestrial photogrammetric camera and digital images acquired with the Canon EOS Rebel XSi/450D digital non-metric camera,
are presented in detail. For the calibration process of both cameras, the Heikkila and Silven algorithm and a 3D calibration ob ject
were used. In order to save time, both on the field and back to the office, instead of taking and processing UMK images by the
classic method of stereophotogrammetry, a new methodology was suggested. So, the UMK 10/1318 photogrammetric camera is
used in a modern process of acq uisition and processing of terre strial images through the multi- image photogrammetric process.
The buildings 3D models accuracy was evaluated based on the coordinates of the characteristic points and by a proposed method which implies the transformation of the 3D model into a 3D mesh surface. While applying this method, the 3D model is
compared with a reference model created based on TLS (Terrestrial Laser Scanner) data, using as comparison metric the
Hausdorff distance. The purpose of this work is to analyse the degree of confidence when using a non-metric digital camera to
create a building 3D model and to determine the differences between two 3D models, especially CAD models, with high
precision, in a completely automated way.

Key words: 3D model, analogous cameras, calibration, digital cameras, Hausdorff distance

Received: August, 2015; Revised final: March, 2016; Accepted: April, 2016

 Author to whom all correspondence should be addressed: e-mail: [anonimizat]; Phone: +[anonimizat] 1. Introduction

1.1. General assertions

Non-metric digital cameras have known a
great development in the last few years, being used in a wide range of applications, such as: preservation of
cultural heritage (Chen and Romise, 2009; De Reu et
al., 2013; Grussenmeyer et al., 2008), architectural heritage (Pukanska et al., 2014; Sužiedelyt ė-
Visockien ė et al., 2015; Pérez Ramos and Robleda
Prieto, 2015), the wave movement for coastal protection, soil erosion, buildings 3D modelling etc.
Metric cameras, are still of great use in areas
such as architecture and cultural heritage (Caprioli
and Scognamiglio, 2003; Mata et al., 2004; Piech, 2013; Sadjadi, 2008; Trieb and Kilpatrick, 2004) and
also environmental imaging.

1.2. A short review of the related work

In the last decade, many studies have been
conducted in order to evaluate 3D models created
based on digital images, in terms of accuracy,
recommendations, advantages and disadvantages, but only a few of them refer to the accuracy of an entire
building 3D model. The images are acquired with
digital cameras (Cardenal et al., 2004; Chandler et al., 2005; Remondino et al., 2008; Skarlatos and
Kiparissi, 2012; Sužiedelyt ė-Visockien ė
and Brucas,
2009) and metric cameras respectively (Böhler and
Marbs, 2004; Grussenmeyer et al., 2008; Piech,

Oniga and Chiril ă/Environmental Engineering and Management Journal 15 (2016), 6, 1293-1303

12942013). As far as accuracy is concerned, the
comparison was made with the TLS data or total stations measurements (Koutsoudis et al., 2013).
Regarding the precision of an entire building
3D model, created based on digital images taken with a digital non-metric camera, Braybon (2011)
demonstrated that the overall positional accuracy of
the building characteristic points is within 0.65 m, over a distance of 140 m, for a complete loop, by
using the Canon EOS 450D digital camera.
Metric and digital non-metric cameras, have
been subject of comparison studies in the last years, but only small buildings parts were modeled and
compared (i.e. one wall), not the entire building, as
mentioned above. On the other hand, most comparative studies focused on the advantages and
disadvantages of both methods, i.e cost, time,
required equipment, problems encountered, etc.
(Caprioli and Scognamiglio, 2003; Cardenal et al.,
2005; Sadjadi, 2008) and only a few on accuracy
(Cardenal et al., 2004). Also, the images acquired with a metric camera are taken and proccesed by
standard stereophotogrammetry.
So, this paper brings to attention a new
method of acquisition and processing of terrestrial
metric images through the multi-image
photogrammetric process. This way, the metric camera, whose lenses have minimum aberrations and
ensure a high optical and geometrical quality, is used
in a modern process of acquisition and processing,
almost half reducing the working time as compared to stereophotogrammetry. In this paper, the accuracy
evaluation process have taken into acount the 3D
models created based on digital images, acquired with digital cameras and me tric cameras respectively,
for the entire building, not only for some parts of the
building.
Taking into account the numerous data
sources used to create the buildings 3D models, a
new method for comparing them, both from the point of view of accuracy and the representation
completeness, it is necessary.
Over the years, there were made comparative
studies on 3D models created based on different data
sources, most of them being focused on comparing
characteristic points coordinates, 3D distances, areas or buildings volumes. The problem of assessing the
accuracy of the 3D model creation has already
appeared in literature. Akca et al. (2010) proposed internal accuracy assessment of the 3D model based on bringing the three-dimensional surfaces in
correspondence, by the least squares method and
calculate the Euclidean distances between the two surfaces. Elberink and Vosselman (2011) made a
complex analysis of errors that influence the 3D
modelling precision of building different elements (especially roofs), is presented. The proposed method
for assessing the accuracy is based on a comparison
of the model with data resulted from airborne laser scanning.

2. Study area and materials

For this case study the dean's office building
from the “Faculty of Hydrotechnics, Geodesy and
Environmental Engineering”- „Gheorghe Asachi” University of Iasi, Romania, has been chosen, which
is a complex shape building, the roof structure having
a shape of a hyperbolic paraboloid.
In this paper the UMK 10/1318FP was used,
equipped with a Lamegon 8/100 lens (f/8, f=100mm),
that has a distortion less than 12 micrometres (µm)
for object distances between infinity and 3.6 m (Mikhailand Fryer, 1989). As a photosensitive
material this camera uses glass plates 13×18 cm,
which have the usable image area of 120 mm×166 mm. Since these glass plates are no longer produced,
an adjustment was required: on the glass plate, a plan
film with the same format was placed.
As a digital camera a Canon EOS Rebel XSi
/450D digital SLR camera (12.2 Mega pixel),
equipped with a 22.2 mm by 14.8 mm image sensor and a Canon EF-S 18-55mm IS lens, was used. The
digital image has a resolution of 4272 x 2848 pixels
with a 5.2 µm pixel size. The topographic measurements of the points situated on the building
edge were made using the Leica TC(R) 405 total
station, which has the accuracy in the angle measurement of 5” and in the case of distances of 5
mm ± 2 ppm. The TLS data were aquired with a fast,
very accurate long-range, tripod-mounted terrestrial
laser scanner, namely the ScanStation2. It is
produced by Leica Geosystems, has an integrated
digital camera of 5 megapixels and a measurement
precision of 6 mm at a distance of 50 m.

3. Metric and non-metric camera calibration

A non-metric camera is a camera that has an
interior orientation completely or partially unknown
and frequently unstable, the images being
characterized by the lack of fiducial marks (Faig, 1976). A metric camera, on the other hand, is
characterized by a stable, known, and repeatable
interior orientation, defined by fiducial marks and determined through calibration, either in the
laboratory, on the job or by self-calibration.
The experiments conducted by Samper et al.
(2011) for synthetic and real point calibration
models, demonstrate that the model using a planar
target produces a larger reconstruction error of the
world system coordinates than the model using 3D calibration objects. In the mentioned context, in this
paper the Heikkilä and Silven’s algorithm was used
to determine the parameters of the UMK 10/1318 terrestrial camera and the Canon EOS Rebel
XSi/450D digital camera, using a 3D calibration
object. Calibration using 3D calibration objects yields very efficient results, although the calibration
elements must be accurate and require an elaborate
configuration (Zhang, 2000).

Analogous vs. digital cameras for buildings 3D models creation

1295We built a target consisting in a number of 42
points, 36 of them being placed in the corners of 9 wooden cubes with different heights and 6 of them
on a board, to be further used as a calibration object
for the calibration model. The 42 control points have a 18 mm diameter and consist of metal parts
manufactured by means of a lathe (Oniga and Diac,
2013).

3.1. Measuring the world coordinates of the control
points
In order to place the target in the world
coordinate system (X,Y,Z), a coordinate measuring
machine (CMM) produced by Aberlink was used. The coordinates of the 42 control points were
measured with a precision of 2 m, in the world
coordinate system, with the XOY plane, the board plane and the OZ axis perpendicular to the board
plane (Oniga and Diac, 2013).
For each control point were measured four
points, the software automatically determines the best circle that approximates the four points, then
calculates all automatically, the centre coordinates
and the circle diameter. Then, the coordinates of the control points were automatically registered to a
computer.

3.2. Image observations of the 3D calibration object

The UMK 10/1318 terrestrial
photogrammetric camera was placed in three
different positions: at 1.0 m, 1.2 m and 1.4 m from
the 3D object, the images being taken in normal position. The exposure time was 1s for all images
and the aperture was 16.
In the case of the Canon EOS Rebel
XSi/450D digital camera, seven images were taken at
different distances from the calibration object and the
focal length was fixed at minimum zoom (f = 18
mm). Both cameras were placed on a tripod for stability (Oniga and Diac, 2013).

3.3. Measuring the image coordinates of the control
points and calculating the cameras parameters

For the measuring process of the image
coordinates, a specific Matlab code was written. For
the metric and non-metric camera parameters
calculation, the Heikkila and Silven’s calibration algorithm was used (Heikkila and Silven, 1997;
Heikkila, 2000), implemented in a Matlab toolbox
available at www.ee.oulu.fi/~jth/calibr.
3.4. The Canon EOS Rebel XSi /450D digital camera
and the UMK 10/1318 terrestrial photogrammetric
camera calibration

Having a 3D target, one image is enough to
estimate the camera parameters through the
calibration process, but for this experiment seven images were used, taken from different positions, as we mentioned before and the camera parameters
were calculated as an average. The control points are often circular, because they are easy to make and
accurate to measure in subpixel precision from digital
images. The images (negative film) taken with the UMK 10/1318 terrestrial photogrammetric camera
were scanned with a desktop scanner Epson V750 at
3200dpi. The size of the image is 12614 by 9070 pixels.
The intrinsic parameters such as focal distance
(f), optical centre point ( u
0,v0), correction of radial
distortion ( k1, k2), correction of decentering distortion
(p1, p2) and the image scale factor su, were calculated,
separately, for each image taken with the Canon EOS
Rebel XSi /450D digital camera, then the average was computed.

4. Creating the building 3D model

4.1. Modelling the building’s roof
In order to approximate the roof's structure
with the most probable geometric form of a
hyperbolic paraboloid, the following computation steps were made (Oniga and Chirila, 2012):
– approximating the main axis for the hyperbolic
paraboloid using the least squares method;
– approximating the parabola from the XOZ plane
of the paraboloid using the least squares method;
– computing the necessary parameters to transform
the detail points coordinates from the national system stereographic-1970 in the paraboloid
system;
– approximating the geometric shape of the
hyperbolic paraboloid using the least squares
method.
Although from the ALS (Airborne Laser
Scanner) point cloud, planes are determined very
well, the roof edges are determined with difficulty
using only these data. Therefore, to obtain a very good accuracy in 3D reconstruction of the objective
roof, additional measurements made with the total
station were carry out. Uniting the points measured on the building edge, the roof limit was obtained,
which was superimposed on ALS data to accurately
identify points belonging to the roof surface (Fig. 1).
To calculate the parameters of the first
hyperbolic paraboloid, a total of 566 ALS points
were used. After going through three iterations, the maximum correction calculated for the normal altitudes along the Z axis of the hyperbolic
paraboloid was -0.193 m. The parameters a and b of
the hyperbolic paraboloid, resulting from the adjustment process by means of indirect
observations, can be affected by gross errors, in this
case, they may be caused by misclassified ALS points or the points belonging to the upper edge of
the roof. The corrections resulting from the
adjustment process should not be used directly to identify large errors. The re siduals result not only due
to errors in the observations (points ALS) but also
because the functional model error compensation.

Oniga and Chiril ă/Environmental Engineering and Management Journal 15 (2016), 6, 1293-1303

1296
4.2. Statistical blunder detection applied to the ALS data

The concept of statistical blunder detection in
surveying was introduced in the mid-1960s and
utilizes the cofactor matrix Qxx for the residuals
(Ghilani and Wolf, 2006). To develop this matrix, the adjustment of a linear problem can be expressed in
matrix form as given by Eq. (1), where B is the
coefficient matrix, X is the unknowns vector, L is the
free terms vector, C is a constants vector, and V is the
corrections vector. Eq. (1) can be rewritten for V in
the form of Eq. (2).

11 11 r, n n, r, r, r,BX C L V (1)

vv VQ W ( 2 )
The Qvv matrix represents the covariance
matrix for the vector of corrections, v
i, W is the
weigh matrix and ε is the vector of true errors for the
observations. The standardized residualsiv are
calculated based on the principal diagonal elements
of the covariance matrix Qvv as Eq. (3), where vi are
the residuals and qii the diagonal element of the Qvv
matrix. Using the Qvv matrix, the standard deviation
in the residual is ii 0qs. Thus, if the denominator
of Eq. (3) is multiplied by so, a t statistic is defined. If
the residual is significantly different from zero, the
observation used to derive the statistic is considered to be a blunder. The test statistic for this hypothesis
test is (Eq. 4). The equation for blunders rejection is
given in the form of Eq. (5)

ii
iivv
q ( 3 )
ii ii
vo oi ivv vtss sq
(4)

i
io
iivvs r e jection level
q 
(5)

For this case study the rejection level from the
Eq. (5) was considered 2.8, the value of the
confidence level of 95%, the points whose ivvalues
exceed 2.1356, being eliminated. After applying this
threshold, a single ALS point was eliminated. The
adjusted parameters of the hyperbolic paraboloid that
best fit the 565 ALS points, remaining after the statistical test application, are: a= 5.226m and b=
5.554m , the maximum correction being -0.188 m. A
code in Matlab programming language was written for the calculation. A file with the *.txt extension was
created, which contains the ALS points coordinates
inventory in three-dimensional system. In the same
manner, the calculations were carried out for the second hyperbolic paraboloid, resulting in the
following parameters : a= 5.209m and b= 5.489m .
After removing two ALS points identified as gross errors using the Eq. (6) and the 2.8 rejection level,
the maximum correction was -0.107 m.
Using the parameters calculated for the two
hyperbolic paraboloids, the roof limit and the 0.5 m
intervals for the X and Y coordinates, was
automatically generated, using a code writen in Matlab programming language, a coordinates
inventory (X,Y,Z) which verifies the equations of the
two hyperbolic paraboloids and belong to the roof surface of the Dean's Office building from the
Faculty of “Hydrotechnical Engineering, Geodesy
and Environmental Engineering” from Ia și. Using the
Delaunay triangulation, two surfaces were created:
the best fitting hyperbolic paraboloid shape and the
roof surface modelled in 3D using the ALS points.
The differences between them are clearly noticeable in (Fig. 2) through the use of different colours.

a) b)

Fig. 1. The roof limit superimposed on the ALS points which belong to (a) „ground” class and (b) „building” class

Analogous vs. digital cameras for buildings 3D models creation

1297

a) b)

Fig. 2. The roof surface modelled in 3D using the ALS points (green colour) and the best fitting hyperbolic paraboloid shape
(red colour) (a) from the west side, (b) from the east side

4.3. The 3D model generation based on images
acquired with the UMK 10/1318 photogrammetric
camera and processed by multi-image photogrammetry (proposed method)

Photogrammetric data acquisition and
processing: To create the 3D model of a building
based on UMK images, but a lot quicker, we suggest
a new method for image processing, namely the multi-image process. Thus the classical
photogrammetric camera, with minimum aberrations
lenses, is used in a modern process of collecting and
processing the images. Based on the study made on the shape of the building and its surrounding areas,
ten camera positions have been established on the
ground plan of the building, situated at approximately 18 m from the building, from which
10 images were taken (Wolf et al., 2014).
The images acquired using the UMK 10/1318
photogrammetric camera, were imported to the
„PhotoModeler Scanner 2012” software (Skarlatos et
al., 2010) and for this case study the „Marking & Referencing” method was used, every detail point of
the building being manually marked and referenced.
Following the bundle adjustment process (Barazzetti et al., 2010), the „PhotoModeler Scanner2” software,
calculated the three-dimensional coordinates of 397
characteristic points of the „Dean’s office” building,
in a local coordinate system, based on a number of ten UMK 10/1318 images, as well as the exterior
orientation parameters for each camera position.
To convert the coordinates of the building
characteristic points from the local ccordinate system
in the world coordinate system defined by the control
points, for this case study the National Projection System „Stereographical on unique secant plan-
1970” and the „Black Sea 1975” reference system for
heights, the coordinates of three artificial control points were introduced (two located on the main
facade and one in the middle of the main side
facade). Thus, in Fig. 3, can be seen the results of the bundle adjustment process, the camera positions and
orientations towards the building to be photographed
and the building characteristic points in the world coordinate system. In Fig. 4, it can be seen the ten
UMK images, oriented in the world coordinate
system. The 3D building model (Fig. 5) was created
based on the building characteristic points, using the
„PhotoModeler Scanner” software specific functions.
Since the roof was not photographed, in order
to create the final 3D model of the „Dean’s office”
building, a *.txt file which contains the coordinates of the points that define the most probable hyperbolic
paraboloid, determined automatically using a Matlab
code, was imported into the software „PhotoModeler Scanner”. To create the surface, the imported points
were interpolated by Delaunay triangulation method,
using the „Automatic Surfacing / Triangulation”
function.
Quality assessment: For this case study, all
the image coordinate errors were less than 5 pixels
(Table 1), tolerance suggested by „PhotoModeler Scanner”. The overall residual of the project was
2.83 pixels, less than the recommended 5 pixels. The
errors for determining the world coordinates have values between 1.5 cm and 7.8 cm. The angles
between the projection rays range between 20
0.6130
÷ 890.9489 with an average angle of 650.2396 (Table
1). Due to the lack of visibility in some areas,
because of trees situated around the building,
especially in the left side of the building, around 1% of the points were calculated using small angles
between 20
0÷300.

4.4. The 3D model generation of the „Dean’s office”
building, based on images acquired with the digital
camera

Photogrammetric data acquisition and
processing: All images were taken with the 18 mm
minimum focal length of the Canon EOS Rebel XSi/450D camera lens, mounting the camera on a
tripod at each station point for stability. Following
the bundle adjustment process, the „PhotoModeler Scanner” software calculated the three-dimensional coordinates of 458 characteristic points of the
„Dean’s office” building, in a local coordinate
system, based on a number of 19 digital images, as well as the exterior orientation parameters for each
camera position. To convert the coordinates of the
building characteristic points from the local coordinate system in the world coordinate system
defined by the control points, the coordinates of three

Oniga and Chiril ă/Environmental Engineering and Management Journal 15 (2016), 6, 1293-1303

1298artificial control points were introduced, the same
used in the case of the proposed method.
The 3D building model (Fig. 6) was created
based on the characteristic building points (Hanan et
al., 2015), using the „PhotoModeler Scanner” software specific functions.
Quality assessment: All the image coordinate
errors were less than 5 pixels, except one point, for which the error was 9.88 pixels, being determined
only from two images and also being placed on the
edges of both images, where the distortions are maximum.

Fig. 3. The resulted elements of the bundle adjustment,
scaling and rotation processes
Fig. 4. The UMK 10/1318 terrestrial images orientated in the
world coordinate system

a) b)

Fig. 5. The „Dean’s office” building model, created in the “PhotoModeler Scanner” software, based on the UMK images,
processed by multi-image photogrammetry, (a ) north-east view, (b ) south-west view

Table 1. Parameters representing the precision determination of the building characteristic points coordinates

Measurement errors of image coordinates
[pixels] The average angles between optical axes
[0]
0-1 40% 20.6130-30 1%
1-2 29% 30-50 17%
2-3 18% 50-70 35%
3-4 7% 70-90 47%
4-5 6% – –

a) b)

Fig. 6. The „Dean’s office” building model, created in the “PhotoModeler Scanner” software, based on the digital images
(a) perspective view of the main facade, (b) perspective view of the main side facade

Analogous vs. digital cameras for buildings 3D models creation

1299
If this point is being eliminated and the bundle
adjustment process repeated, the interior accuracy of
the 3D model is improved with only 2.2 cm. Being
an important characteristic point of the building, we couldn’t eliminate it. The overall residual of the
project was 1.541 pixels. The errors for determining
the world coordinates have values between 1 cm and 10.9 cm. The angles between the projection rays
range between 9
0.79810÷890.7531 with an average
angle of 580.8552.

5. Comparing the building 3D models

5.1. Comparing the buildings 3D models based on
the characteristic points coordinates
The accuracy of the 3D modelling process
was evaluated first by pointing out and verifying the
differences between the values of two sets of rectangular coordinates (X, Y, H) (Table 2). These points represent 40 characteristic points of the
building (Fig. 7) (window edges, door etc.), located
on the building façades. The first set of coordinates was obtained by the 3D model interrogation in
AutoCAD software, created based on TLS
(Terrestrial Laser Scanner) data which represent the
reference coordinates (Lerma García, 2008). The second set of coordinates was the result of the 3D
model interrogation in AutoCAD software, created
based on digital metric and UMK images processing, respectively.
The reference 3D model was created based on
TLS data (point cloud) by using lines and surfaces with the existing functions in the Cyclone v. 6.0
software. The TLS point cl oud resulted after the
direct georeferencing process of five scans, acquired with the ScanStation2 laser scanner.

5.2. Comparing the buildings 3D models by the proposed method

In order to compare 3D models derived from
two different data sources, we suggested the comparison method used on two mesh surfaces
(network of triangles), as triangles are the most used
primitives to create 3D models. In most cases, the buildings 3D models are of CAD type, therefore the
surfaces must be transforme d in mesh surfaces to be
compared through the suggested method. The method for two mesh surfaces comparison, has been
introduced since 1987 (Cignoni et al., 1998a; Cignoni
et al., 1998b; Roy et al., 2002; Roy et al., 2004) to measure and highlight the errors caused by a mesh
simplification, as a representation of a complex 3D
surface, consume the computer resources and it is not always necessary. Hausdorff Distance – named after
Felix Hausdorff, is the most famous metric for
comparing two mesh surfaces, providing a global comparison (Aspert et al., 2002).
To determine the accuracy of the buildings 3D
models by the proposed method, these must be compared with a reference model considered for this
case study, the 3D model created based on the TLS
data, as mentioned above. In order to evaluate the
interior accuracy of the 3D model created based on the TLS point cloud, it will be compared using the
„CloudCompare” software with the point cloud
resulted from the scanning and manual filtering processes (Fig. 8a) (Girardeau-Montaut, 2015).
Quality assessment of the “Dean’s office”
building 3D model, created based on the TLS data: By analysing the differences distribution
histogram, between the TLS point cloud and the
„Dean’s office” building 3D model, created based on the TLS data, it can be seen that most of the
Hausdorff distances values range between -2.6 cm ÷
+2.0 cm and the standard deviation is 2.4 cm (Fig.
8b). Certainly, TLS data processing involves manual
steps, the building components being created by
approximation with their mathematical shape (finding the best fitting geometric elements like
patch/plane, cylinder, sphere, etc.), using the „Leica
Cyclone” software (Oniga et al., 2012), therefore the
accuracy of the 3D model is not the same with the terrestrial laser scanning system measurement
accuracy (Grussenmeyer et al., 2008).
An important observation is that the TLS
points belonging to closely scan lines, located on the arc-shaped structural element from the southeast of
the „Dean’s office” building, are colored differently,
the calculated distances being 1 mm and -6.3 cm.
Quality assessment of the “Dean’s office”
building 3D models, created based on the digital
metric and non-metric images: After comparing the
“Dean’s office” building 3D models, created based
on two data sources, with the reference 3D model of
the building, the following differences were obtained (Table 3).
Analysing the colours diagram, obtained after
comparing the „Dean’s office” 3D model, created based on digital images, with the reference model, the following can be observed:
– for the main facade from the building left side,
the colours in shades of green show that the differences are of the order of 1 mm ÷ 10 cm and for
the 1÷4 points which belong to this facade, the
differences between the two sets of coordinates are in ranges of 3.7 cm ÷ 8.3 cm;
– the maximum differences were record ed for the
points situated on the exterior part of the main entrance structural element. Also, for this points the
image coordinates measurement errors were
maximum;
– large errors were also re corded for the front part
of the cylinder located above the main entrance and
the bottom right corner of the main facade from the building right side, differences being within the range
of 15 cm to 30 cm;
– for the main façade (right side) colours in
shades of yellow to orange show that the differences
are of the order of 14 cm ÷ 25 cm, and for 8÷20

Oniga and Chiril ă/Environmental Engineering and Management Journal 15 (2016), 6, 1293-1303

1300points belonging to this facade, differences ranges
between 17.8 cm ÷ 28.7 cm;
– for the secondary facade colours in shades of
green show that the differences are of the order of 1
mm ÷ 10 cm, and for 31÷35 points that belong to this facade, differences yielded between 4.9÷8.9 cm;
– for the left side facade, the colours in shades of
green to blue show that the differences are of the order of 1 mm ÷ 10 cm and for the 38÷40 points
which belong to the facade, and differences were in
ranges of 1.9 cm ÷ 10.8 cm;
– for the roof, differences are null, because its
surface, previously determined by the method of least
squares, was imported into “PhotoModeler” software.
By doing the same analysis for the 3D
building model created based on UMK images, it can
be observed that there is a direct correspondence
between the Hausdorff distances and the euclidian distances calculated between the two sets of
coordinates. If, however, the 3D model created based
on the TLS data, will be compared with the one created based on digital metric or non-metric images,
the differences will be different because they will highlight the building missing components, reflecting
the representation completeness (Oniga, 2014).
Another way to highlight the missing
elements is to subdivide each triangle, component of
the mesh surface representing the 3D model of the
building to be compared, in this case the one created based on the digital images, the Hausdorff distances
being calculated also inside the triangle not only in
its vertices.
The only inconvenience in using this method
for 3D models comparison, is when the components
of the reference 3D model were created by using the
“extrude” function, because when there is a translation between the two 3D models facades, it is
possible that this may not be highlighted correctly.
The Hausdorff distances are calculated from the nearest surface, which in this case may be the one
from the interior side of the element.

Fig. 7. The main facade characteristic points, us ed to evaluate the 3D model precision

Table 2. The residual errors of the 3D models characteristic points coordinates

Data sources RMSE RMS
[m] Minimum value
[m] Maximum value
[m]
Digital images 0.019 0.287 0.151
UMK 10/1318 images, processed by
multi-image photogrammetry 0.025 0.149 0.101

a) b)

Fig. 8. The differences between the TLS point cloud and the “Dean’s office” building 3D model, created based on the TLS data,
using the “Leica Cyclone” software, (b) the differences distribution histogram

Analogous vs. digital cameras for buildings 3D models creation

1301Table 3. The residuals of the „Dean’s office” building 3D models, obtained by the proposed method

Digital images

Differences:
Standard deviation: 4.8 cm
Maximum positive: 31.9 cm
Maximum negative: 17 cm
Mean: 8 mm UMK 10/1318 images, processed by multi-image
photogrammetry

Differences:
Standard deviation: 3.1 cm
Maximum positive: 15.8 cm
Maximum negative: 14.2 cm
Mean: 6 mm

6. Conclusions

6.1. Conclusions regarding the building 3D model
creation based on digital images, taken with the
CanonEOSRebelXSi/450D digital camera

After comparing the spatial coordinates (X, Y,
H) of 40 characteristic points of the building 3D model created based on images taken with the
CanonEOSRebelXSi/450D digital camera, with the
ones belonging to the 3D model obtained based on the TLS data, the cumulative root mean square error
was 15.1 cm , so the accuracy is about 1:643 of the
object size.
The accuracy of the 3D models created based
on digital images depends on the type of the camera
used and its technical characteristics, the geometry of the point station network for taking the images, the
accuracy of the image identification process of the
object space points and their number (at least 10 referenced points in every image). It is very
important not to mark on the image points that can’t
be clearly identified due to blur or obstacles, fewer points providing better accuracy than points that were wrong marked. Other factors that directly influence
the accuracy of the 3D models creation process, are:
the determination accuracy of the control points coordinates in the world coordinate system, used to
georeference the 3D model from the object reference
system (local) to the world reference system, but also for the scaling process; the number of the used
images; the accuracy of the camera calibration
parameters calculation.
The required time to acquire the digital
images with the CanonEOSRebelXSi/450D digital
camera, was approximately 30 minutes and the processing time was 18 hours.
Some disadvantages of using this method for
buildings 3D models creation are:
– the final accuracy of the building 3D model
created through the close-range photogrammetry
method depends upon respecting some conditions, an important one being the optimal geometry of the image taking stations. This condition can't be always
fullfilled, due to different obstacles surrounding the
buildings (bushes, statues, other nearby constructions
etc.), which do not allow images to be taken from the
desired position;
– the accuracy of the building 3D model is in the
range of dozens of centimeters and cannot be
compared to the one obtained by means of using TLS data, which is of few centimeters;
– in order to obtain the form and the geometry of
the 3D model, one has to determine the coordinates of the specific points of the building, whose number
increases according to the size and the complexity of
the object, thus leading to an increase in the amount of time necessary for the creation of the final 3D
model;
– the model obtained through this method must be
scaled and then georeferenced.

6.2. Conclusions regarding the building 3D model
creation based on images taken with the UMK
10/1318 terrestrial photogrammetric camera and
processed by multi-image photogrammetry (proposed method)

After comparing the spatial coordinates (X, Y,
H) of 40 characteristic points of the “Dean’s office” building 3D model created based on UMK images,
with the ones belonging to the 3D model obtained
based on the TLS data, the cumulative root mean square error was 10.1 cm. so the accuracy is about
1:961 of the object size.
The proposed method for taking and
processing the UMK images by multi-image
photogrammetry, reduce th e time required for the
images acquisition and processing by stereoscopy (at about half) and was used to compare the results
obtained with a classical photogrammetric camera
whose lens has minimal aberrations with those obtained with a digital camera, whose lens has large
aberrations.
The difference between the cumulative root
mean square errors of the 3D models created based

Oniga and Chiril ă/Environmental Engineering and Management Journal 15 (2016), 6, 1293-1303

1302on UMK images (10.1 cm) and digital images (15.1
cm), represented 33% of the cumulative root mean
square error of the 3D model obtained based on digital images.
The required time to acquire the images with
the UMK photogrammetric camera, was
approximately 2 hours and the processing time was
24 hours.
In order to be processed through digital
technologies, UMK images undergo supplementary
processing procedures, such as developing and
scanning, thus inducing new sources of distortions that turn this process into an uneconomical one. This
is the only disadvantage of using UMK images in the
process of building 3D model creation in comparison of using digital images acquired with a digital
camera.

6.3. Conclusions regarding the proposed method for
buildings 3D models comparison

When using this method for buildings 3D
models comparison, an inconvenience may occur if
the components of the reference 3D model are
created by using the “extrude” function, as presented above. It offers a comprehensive assessment of the
differences between two 3D models by using a
colour palette, the user can quickly and accurate identify the errors of the 3D model that was
compared with a reference model considered with no
errors.
It illustrates the missing elements of the 3D
model that was evaluated, by changing the reference
model with the compared model, or by subdividing the triangles surfaces which are components of the
mesh surface corresponding to the transformed 3D
model.
References

Akca D., Freeman M., Sargent I., Gruen A., (2010),
Quality assessment of 3D building data, The
Photogrammetric Record , 25, 339–355.
Aspert N., Santa-Cruz D. , Ebrahimi T., (2002), Mesh:
Measuring Errors between Surfaces Using the
Hausdorff Distance , In Proc. of the IEEE International
Conference in Multimedia and Expo (ICME),
Lausanne, 1, 705-708.
Barazzetti L., Scaioni M., Remondino F., (2010),
Orientation and 3D modelling from markerless
terrestrial images: Combining accuracy with
automation, Photogrammetric Record , 25, 356-381.
Böhler W., Marbs A., (2004), 3D Scanning and
Photogrammetry for Heritage Recording: a
Comparison, Proc. of 12th Int. Conf. on
Geoinformatics, Gävle, Sweden, 291-298.
Braybon J., (2011), Traversing the UNSW campus using
Terrestrial Photogrammetry , MSc Thesis, UNSW
School of Surveying and Spat ial Information Systems,
Sydney, Australia.
Caprioli M., Scognamiglio A., (2003), Photogrammetry
and Laser Scanning in Surveying and 3D Modelling of
Architectural Heritage, FIG Working Week 2003,
Paris, France, 1-7. Cardenal J., Mata E., Castroa P., Delgado J., Hernandez
M.A., Perez J.L., Ramos M., Torres M., (2004), Evaluation of digital non metric camera (Canon D30), for the photogrammetric recording of historical
buildings,
The International Archives of the
Photogrammetry, Remote Sensing and Spatial
Information Sciences , 34, 1-6.
Cardenal J., Mata E., Ramos M., Delgado J., Hernandez
M.A., Perez J.L., Castroa P., Torres M., (2005), Low
cost digital photogrammetric techniques for 3D
modelization in restoration works. A case study: St.
Domingo de Silos’ church (XIVth century, Alcala la Real, Spain)
, CIPA 2005 XX International Symposium,
Torino, Italy.
Chandler J.H., Fryer J.G., Jack A., (2005), Metric
capabilities of low-cost dig ital cameras for close range
surface measurement, Photogrammetric Record , 20,
12–26.
Chen F., Romice O., (2009), Preserving the cultural
identity of Chinese cities in urban design through a
typomorphological approach, Urban Design
International , 14, 36-54.
Cignoni P., Rocchini C., Scopigno R., (1998a), Metro:
Measuring error on simplified surfaces, Computer
Graphics Forum , 17, 167–174.
Cignoni P., Montani C., Scopigno R., (1998b), A
comparison of mesh simplification algorithms,
Computers & Graphics , 22, 37-54.
De Reu J., Plets G., Verhoeven G., De Smedt P., Bats M.,
Cherretté B., De Maeyer W., Deconynck J.,
Herremans D., Laloo P., Van Meirvenne M., De Clercq W., (2013), Towards a three-dimensional costeffective registration of the archaeological
heritage,
Journal of Archaeological Science , 40, 1108-
1121.
Elberink S.O., Vosse lman G., (2011), Quality analysis on
3D building models reconstructed from airborne laser
scanning data, ISPRS Journal of Photogrammetry and
Remote Sensing, 66, 157–165.
Faig W., (1976), Photogrammetric potentials of non-metric
cameras, Photogrammetric Engineering and Remote
Sensing , 42, 47-49.
Ghilani C.D., Wolf P.R., (2006), Adjustment
Computations. Spatial Data Analysis , 4th Edition, John
Wiley&Sons Inc., New York.
Girardeau-Montaut D., (2015), Cloud Compare version
2.6.1 – user manual, On line at:
http://www.danielgm.net/cc/doc/qCC/CloudCompare
%20v2.6.1%20%20User%20manual.pdf.
Grussenmeyer P., Landes T., Voegtle T. and Ringle K.,
(2008), Comparison methods of terrestrial laser scanning, photogrammetry and tacheometry data for recording of cultural heritage buildings,
International
Archives of Photogrammetry, Remote Sensing and
Spatial Inform ation Sciences , 37, 213–218.
Hanan H., Suwardhi D., Nurhasanah T., Bukit E. S.,
(2015), Batak Toba Cultural Heritage and Close-range Photogrammetry,
Procedia – Social and Behavioral
Sciences , 184, 187-195.
Heikkila J., Silv en O., (1997), A Four-Step Camera
Calibration Procedure with Implicit Image Correction
, Proc. of the 1997 Conference on
Computer Vision and Pattern, San Juan, IEEE,
DOI:10.1109/CVPR.1997.609468.
Heikkila J., (2000), Geomet ric camera calibration using
circular control points, IEEE Transactions on Pattern
Analysis and Machine Intelligence , 22, 1066-1077.
Koutsoudis A., Vidmar B., Ioannakis G., Arnaoutoglou F.,
Pavlidis G., Chamzas C., (2013), Multi-image 3D

Analogous vs. digital cameras for buildings 3D models creation

1303reconstruction data evaluation, Journal of Cultural
Heritage , 15, 73-79.
Lerma García J.L., (2008), Theory and Practice on
Terrestrial Laser Scanning. Training Material based
on Practical Applications, Prepared by the Learning
Tools for Advanced Three-Dimensional Surveying in
Risk Awareness Project (3DRiskMapping) , University
Politécnica de Valencia, Valencia, Spain.
Mata E., Cardenal J., Castro P., Delgado J., Hernandez
M.A., Perez J.L., Ramos M., Torres M., (2004),
Digital and analytical photogrammetric recording applied to cultural heritage. A case study: “St. Domingo de Silos’ Church (XIVth century, Alcala la Real, Spain)”,
The International Archives of the
Photogrammetry, Remote Sensing and Spatial
Information Sciences, 34, 1-6.
Mikhail E.M., Fryer J., (1989), Non-Topographic
Photogrammetry , Second Edition, Karara H.M. (Ed.),
American Society for Photogrammetry and Remote
Sensing, 210 Little Falls Street Falls Church, Virginia.
Oniga V.E., Chirila C., (2012), Complex roof structures
generalized representa tion established by
approximating the mathematical shape using the least-
squares method, Scientific Journal Mathematical
Modelling in Civil Engineering , 8, 178-187.
Oniga V.E., Chirila C., Șutu M., (2012), Terrestrial laser
scanner surveying versus to tal station surveying for
3D building model generation, Scientific Journal
Mathematical Modelling in Civil Engineering , 8, 168-
177.
Oniga V.E., Diac M., (2013), Metric and non-metric
cameras calibration for the improvement of real-time monitoring process results,
Environmental
Engineering and Management Journal , 12, 719-726.
Oniga V.E., (2014), A new method for buildings 3D
models comparison, Journal of Geodesy and Cadastre
RevCAD , 17, 60-67.
Pérez Ramos A., Robleda Prieto G., (2015), 3D
virtualization by close range photogrammetry indoor
gothic church apses. The case study of church of San
Francisco in Betanzos (La Coruna, Spain), The
International Archives of the Photogrammetry,
Remote Sensing and Spatial InformationSciences, XL-
5/W4 , 201-206.
Piech I., (2013), Geodetic and Photogrammetric
measurements in the area of historic grange in
Msciwojow, Geomatics, Landmanagement and
Landscape, 1, 73–81.
Pukanska K., Bartos K., We iss G., Rakay S., (2014), The
Application of Close-Range Photogrammetry for a
Documentation of Metallurgical Art-Historical Objects,
SGEM2014 Conference Proceedings, Albena,
vol. 3, 327-334. Remondino F., Baltsavias E., El-Hakim S., (2008). Image-
based 3D modelling of the Erechtheion, Acropolis of
Athens, The International Archives of the
Photogrammetry, Remote Sensing and Spatial
Information Sciences , XXXVII , 1083-1092.
Roy M., Foufou S., Truchetet F., (2002), Generic attribute
deviation metric for assessing mesh simplication
algorithm quality , In Proceedings of IEEE
International Conference on Image Processing, 817-
820.
Roy M., Foufou S., Truchetet F., (2004), Mesh comparison
using attribute deviation metric, International Journal
of Image and Graphics , 4, 127-140.
Sadjadi S.Y., (2008), An Integration of Close-Range
Photogrammetry and CAD System for Cultural
Monuments: Preliminary Findings, WSEAS
International Conference on Engineering mechanics,
structures, engineering geology, Heraklion, Crete
Island, Greece, 142-150.
Samper D., Santolaria J., Brosed F.J., Majaren A.C.,
Aguilar J.J., (2011), Analysis of Tsai calibration
method using two and three-dimensional calibration
objects, Machine Vision and Application s, 24, 117-
131.
Skarlatos D., Agapiou A., Rova M., (2010),
Photogrammetric Support on an Underwater
Archaeological Excavation Site: The Mazotos
shipwreck case , Euromed 2010, Digital Heritage,
Lemesos.
Skarlatos D., Kiparissi S., (2012), Comparison of laser
scanning, photogrammetry and SFM-MVS pipeline aplied in structures and artificial surfaces,
ISPRS
Annals of the Photogrammetry, Remote Sensing and
Spatial Information Sciences , I-3, 299-304.
Sužiedelyt ė-Visockien ė J., Brucas D., (2009), Digital
photogrammetry for building measurements and
reverse-engineering, Geodesy and Cartography , 35,
61–65.
Sužiedelyt ė-Visockien ė J., Bagdži ūnaitė R., Malys N.,
Maliene V. (2015), Close-range photogrammetry
enables documentation of environment-induced deformation of architectural heritage,
Environmental
Engineering and Management Journal , 14, 1371-
1381.
Wolf P. R., Dewitt B. A., Wilkinson B. E., (2014),
Elements of Photogrammetry with Applications in
GIS, 4th Edition, Terrestri al and Close-Range
Photogrammetry, Chapter (McGraw-Hill Professional,
2014), AccessEngineering.
Zhang Z., (2000), A flexible new technique for camera
calibration, IEEE Transactions on Pattern Analysis
and Machine Intelligence , 22, 1330-1334.