ACCURACY ASSESSMENT OF MODELING ARCHITECTURAL STRUCTURES AND
DETAILS USING TERRESTRIAL LASER SCANNING
M. Kedzierski a , P.Walczykowski a, A.Orych a *, P. Czarnecka a
a
Department of Remote Sensing and Photogrammetry, Geodesy Institute, Faculty of Civil Engineering and Geodesy, Military University of Technology, Warsaw, Poland - (mkedzierski, pwalczykowski, aorych)@wat.edu.pl, [email protected]
Commission III, WG III/2
KEY WORDS: Laser Scanning, 3D model, Accuracy, Historical Structures, Materials, Test
ABSTRACT:
One of the most important aspects when performing architectural documentation of cultural heritage structures is the accuracy of both the data and the products which are generated from these data: documentation in the form of 3D models or vector drawings. The paper describes an assessment of the accuracy of modelling data acquired using a terrestrial phase scanner in relation to the density of a point cloud representing the surface of different types of construction materials typical for cultural heritage structures. This analysis includes the impact of the scanning geometry: the incidence angle of the laser beam and the scanning distance. For the purposes of this research, a test field consisting of samples of different types of construction materials (brick, wood, plastic, plaster, a ceramic tile, sheet metal) was built. The study involved conducting measurements at different angles and from a range of distances for chosen scanning densities. Data, acquired in the form of point clouds, were then filt(ered and modelled. An accuracy assessment of the 3D model was conducted by fitting it with the point cloud. The reflection intensity of each type of material was also analyzed, trying to determine which construction materials have the highest reflectance coefficients, and which have the lowest reflection coefficients, and in turn how this variable changes for different scanning parameters. Additionally measurements were taken of a fragment of a building in order to compare the results obtained in laboratory conditions, with those taken in field conditions.
1. INTRODUCTION
As with any measurement technique, TLS measurements are not perfect and subject to errors caused by different factors affecting the measurement process. Determining the sources of errors in terrestrial laser scanning measurements is complicated due to the large number of such factors, and the fact that they are usually interlinked. the complexity of this issue is enhanced by the fact that different laser scanners use different measurement methods, for example, they operate at different wavelengths and using different beam deflection units. Careful examination of all of these factors and the errors being introduced into the data, provides a solid basis for assessing the quality of these data and the performance parameters of the scanner, which is necessary when deciding on the suitability of TLS for a particular application. The most definitive indication of the quality of the data is described by the accuracy, i.e. the degree of consistency between the measurement and the true value of the measured quantity (Iavarone 2002 r.). In TLS you can use the term "precision of the 3D coordinates of points in the point cloud" as an appropriate indicator of data quality. Because TLS is a technique for measuring area/a surface, and the final product of the laser scanning is often a 3D model, we can use the term "accuracy of the model" as an appropriate indicator of the quality of the information obtained through laser scanning. In both cases, the accuracy reflects the level of compliance between the data or the model and the real world. Sources of measurement errors include instrumental errors resulting from the construction of the scanning device. The accuracy is also affected by errors originating from both the laws of physics that affect the operation of the laser and the system directing the beam, as well as errors related to the mechanics of the scanner, i.e. the laser's optical system
errors, system errors associated with sending a laser beam. Measurement errors can also result from improper data acquisition techniques, eg. an inadequate selection of measurement stations and their number, or the wrong choice of scanning density relative to the accuracy requirements. Undoubtedly, parameters such as the intensity of the reflection beam, angle of incidence and surface properties (material, surface roughness) influence the final result of the process of obtaining the point cloud and its subsequent processing (Reshetyuk, 2006). The following describes these sources of error in more detail.
2. ANALYSIS OF FACTORS AFFECTING THE ACCURACY OF TLS
2.1. Review of factors affecting the accuracy of TSL
Terrestrial laser scanning is a technique of recording data using reflectorless distance measurement. Therefore, one of the main sources of error is the reflectivity of the scanned surface - SNR (Signal-to-noise ratio). It is defined as the ratio of the emitted and reflected signals. The SNR is also affected by: the material properties associated with the electrical and magnetic permeability, the color of the surface, the laser wavelength λ, the surface roughness, which depends on the wavelength and frequency, polarization, the surface temperature and surface moisture (Thiel et al., 2004 ).
Additionally, the intensity of the reflection and the accuracy of coordinate measurements from the point cloud are directly affected by the scanning geometry. The number of elements that describe the point cloud is dependent on the density of the cloud, which in turn depends on the parameters of the scan, i.e. the angular resolution, but also on the angle of incidence and distance to the subject. The density of the point The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XL-5/W7, 2015
25th International CIPA Symposium 2015, 31 August – 04 September 2015, Taipei, Taiwan
This contribution has been peer-reviewed.
cloud decrease angle of incide The angle of between the n direction of th angle of incid diameter. The the scanner be Measurement factors such as vibrations. T atmosphere is a deformation intensity of th Rayleigh and water vapor a in unfavorable return' phenom
3. RES
Research wor external factor of 3D models. a real structur Focus 3D pha The test field c of materials u types of eleva brick, glazed size, shape a spherical and figure 1:
Figure 1.
Textures of ma different types
es with an incre ence (Lindenbe incidence of th normal, directed
he laser beam (S dence on the su e ideal scanning eam falls at a rig errors are al s temperature, p The basic phe the propagatio of the returnin he laser light. T
Mie scattering and CO2, O3 (R e weather cond mena can be obs
SEARCH
rk was perform rs and scan para
The work was re. Measuremen ase scanner.
consists of a se used in constru ation plaster, br
tile and wood and colour. So some cylindric
Test field cons
aterials were al s of plaster are s
Figure 2. Ac
easing distance ergh et al., 2005 he beam is def d toward the tes Soudarissanane urface affects th g conditions ar ght angle to the lso affected b pressure, humid enomenon that on of the laser b ng pulse occurs This is due to g, as well as t Rueger, 1990). ditions, 'dropo
served. (Granth
med to analyz ameters to the r carried out on nts were taken
lection of the m uction. These rick, sheet meta
(fig. 1). The s ome elements al. Chosen sam
sisting of differ
lso analyzed. E shown in figure
crylic plaster tes
and an increasi 5).
fined as the an st surface, and e et al, 2007). T he scanning be re those in wh scanned object by environmen dity, lighting or t occurs in beam. As a resu
weakening of the occurrence the absorption
When measuri out pixel' or 'fa ham et al, 1997)
ze the impact resultant accura a test field and using the FAR
most popular typ include: differ al, plastic, clink samples varied
were flat, so mples are shown
ent materials
Examples of es 2 and 3:
Measurements w ser’s beam an easurements o canning distanc nd 25 meters. N canning angles. eters from th nchanged throu eld was then
easurement. As fferent angles: rocess was con 53mm ; 3,07 m e affect of the s oint cloud and ngles.
s a result, 30 s rocessing consis iminating red gistered by the hich were not duce coarse err f the modelled s
fferent errors w n the type of ma n order to com
easurements w eld conditions, eritage structur aterials, includi nd plastic and m resent on the t onduct a compa oth measuremen
3.1. Analys
comparison an chitectural stru odel for all test
igure 4. Model
Figure 3. Grain
were taken at di nd from a ra of the test fie es. The chosen Next, measurem
In this case, t he test field ughout this part rotated by 1 s a result the te
0°, 15°, 30°, nducted with mm, 6,14mm; 7 scanning density d the intensity
scanned scenes sted of filtering dundant measu
scanner. It wa on the surface rors when mode surfaces showe when fitted wit aterial making u
mpare the res with their future , measurement re constructed ing: light, dark metal railings. A
test field and arative analysis nt phases.
sis of the result
nd accuracy asse ctures was carri samples. (fig. 4
of the surface o
n plaster test
ifferent inciden ange of distan eld were take n distances are ments were taken
the scanner wa and its posi t of the experi 5° relative to test field was m
45°, 60°, 75°. different scan 7,67 mm, in or ty, the errors in values for v
were generate g the point cloud
urement point as crucial to rem e of the test fi elling this surfa ed, that differen th the point clo up that surface. sults obtained e practical imp ts were taken
with a numbe and grain plast All of these ele therefore it w s of the results
ts
essment of acqu ried out on a gen
4)
of the acrylic p
nt angles of the nces. At first, en at different e 5, 10, 15, 20 n with different as set up at 10 tion remained iment. The test o the previous measured at six . The scanning ning intervals: rder to analyze fitting with the aried scanning
ed. Their initial d, which means ts and noise move all points eld in order to ace. An analysis nt samples have oud, depending
in laboratory plementation in of a cultural er of different ter, plastic, tiles ements are also was possible to obtained from
uiring data for nerated 3D
laster test
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XL-5/W7, 2015
25th International CIPA Symposium 2015, 31 August – 04 September 2015, Taipei, Taiwan
This contribution has been peer-reviewed.
The mean valu cloud was ca reflection inte scanning dista fitting errors. Analyses were angle of the b results of these
Figure 5. A pl point cloud at
The value of th the ceramic p beam, the fitt 0,04mm. For value reaches for plastic and RMS = ± 0.0 materials leas and clinker br incidence was of from 0.7 m errors are redu
Figure 6. A pl scan
The chart show the point cloud scanner and th be observed fo distance of 15 then again, at mm. Clinker b at 10m decre materials, for fitting error va
ue of the error alculated, as w ensity. These ance has a ve
e based on two beam and the d
e analyses.
lot of the fitting t different angle on the
he fitting error plate. At low a ting error was
angles equal to ± 0.2 mm. Sim d brick. For an
5 mm, while f st accurately fitt
rick. The error s perpendicular mm to 0.8 mm uced to ± 0.4 mm
lot of the fitting nned point clou
ws that the ave d gradually inc he test field. A or the metal she 5m equal to ± 1 a distance of 2 brick behaves si
eases in relati example wood alues for four co
in fitting the m well as the me studies have ery significant
o basic scannin distance. Graph
g error of differ es of incidence e test field
into a surface w angles of incid
in the range o 60 ° and 75 milar error valu n angle of 0 ° for 75 ° RMS =
ted into a plane r obtained whe to the test fiel m. When you in
m.
g error of differ ud at variable di
erage value of t reases with dist large randomn eet, with an RM 1 mm, which th
0m, increases t imilarly, becau on to the err d, are characte onsecutive scan
model to the po ean value of proven that influence on
ng parameters: s below show t
rent materials in of the laser bea
was the lowest dence of the la from 0.3 mm
° the RMS er ues are noticeab
the error value = ± 0.2 mm. T e were sheet me en the laser be
ld, is in the ran ncrease the ang
rent materials in istances
the fitting error tance between ness of results c MS error value a hen decreases, a to a value of ±
se the fitting er or at 5m. Oth rized by const nning distances.
oint
tudies have sho mpact on the va easuring points stance, the num ven material dividual points hich one is th etermines the a f the given samp
rors. Therefore e fitting errors amples of the te ature of the rela
e distance betw or all of the st auses a decrease
A
his paper has be echnology, the epartment of Re
ranthamn JW, S adar in Adverse 997
varone A., Lase
indenbergh, R., nalysis of the eformation mo Workshop, Lase
nschede, The N
eshetyuk Y., eoreferencing in
Geodesy, Ro eodesy, 2009
ueger, J. M., ntroduction. 3rd
erlin – Heidelbe ong- Kong. 266
oudarissanane S 007. Error budg e incidence a DNordOst 2007
hiel, K.-H. and ser scanners— nalysis. Interna emote Sensing,
4. CON
own that scanni alue of the err s. Together wit mber of points decreases. T s are measured hicker. This ‘th ccuracy of the ple. The greater e, for the greate indicate the lo est field. The c ationship betwe ween the scanne tudies samples e in the reflectio
ACKNOWLED
een supported b Faculty of Civi emote Sensing
REFERE
Stargardt CD, D e Weather. In:
er Scanning Fun
Pfeifer, N. and Leica HDS30 onitoring. In: er scanning 2 Netherlands, pp.
2009. S n terrestrial las oyal Institute
,. Electronic d totally revise erg – New York 6 pages. 1990
S., van Ree J., get of terrestria angle on the 7, Berlin, Germa
Wehr, A., 2004 —An overview ational Archiv
Vol. 36–8/W2
NCLUSIONS
ning distance ha rors of fitting t
th an increase i representing th The precision
also decreases hickness of th approximation er this is, the gre
test scanning d owest accuracy conducted resea een the reflectio
er and the meas s, an increase i
on intensity.
DGEMENTS
by the Military U il Engineering a
and Photogram
ENCES
Dungey C, Mei SPIE Proceedin
undamentals, 20
d Rabbani, T., 2 000 and feasib
Proceedings o 2005, Vol. X 24–29
Self-calibration ser scanning. L of Technology
distance mea ed edition, Sp rk – London – P
Bucksch A., L al laser scannin scan quality any
4, Performance and measurem ves of Photog 2, pp. 14–18
as a substantial the model with in the scanning he surface of a with which s, as a result of he point cloud, n of the surface eater the fitting distance – 25m, y for individual arch shows the on intensity and sured structure. in the distance
University of and Geodesy, mmetry.
idunas E. Laser ngs, Vol. 3065,
002
2005. Accuracy bility of tunnel of the ISPRS XXXVI(3/W3),
and direct Licentiate thesis y, Division of
asurement. An pringer Verlag, Paris – Tokyo –
Lindenbergh R., ng: influence of y. Proceedings
e capabilities of ment principle grammetry and l The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XL-5/W7, 2015
25th International CIPA Symposium 2015, 31 August – 04 September 2015, Taipei, Taiwan
This contribution has been peer-reviewed.
