In: Paparoditis N., Pierrot-Deseilligny M.. Mallet C.. Tournaire O. (Eds). 1APRS. Vol. XXXVIII. Part ЗА - Saint-Mandé, France. September 1-3. 2010 
UNDERSTANDING CHANGES AND DEFORMATIONS ON MULTI-TEMPORAL ROCK 
FACE POINT CLOUDS 
M. Scaioni. M. Alba 
Politecnico di Milano. Dept. B.E.S.T., via M. d Oggiono 18/a, 23900 Lecco, Italy - e-mail: {mario.alba, marco.scaioni}@polimi.it 
Commission III/2 
KEY WORDS: Point Cloud. Change Detection, Deformation Monitoring. Terrestrial Laser Scanning. Digital Surface Analysis 
ABSTRACT: 
The paper outlines an approach to compare two digital surfaces of a rock tace in order to extract geometric changes and deformations. 
The method requires the preliminary registration of each point cloud by using techniques typical of TLS. Then, point clouds are 
segmented in several regions, each of them referred to a plane; this task allows to interpolate all data to obtain a set of grid DEMs 
per epoch. Then each pair of corresponding DEMs are subtracted point-wise to obtain the ADEM of the differences along elevation. 
This is then processed along a three-step procedure. First possible systematic errors or low-frequency deformations are extracted by 
looking for a linear component. Secondly, the remaining ADEM is check against major changes, i.e. loss of material or vegetation 
growth. Finally, deformations of the cliff are enhanced by analyzing the mean displacement through the convolution with a square 
window applied to the ADEM filtered out from holes and bushes. This technique improves the original precision of each measured 
point, because deformation is evaluated as mean of a sample on the ADEM. The method has been tested so far on both synthetically 
generated and simple real data sets. 
1 INTRODUCTION 
Today the high potential of 3D surface reconstruction provided 
by Terrestrial Laser Scanning (TLS) has opened up many new 
prospects of application. Among these, deformation monitoring 
based on the comparison of multi-temporal point clouds is one 
of the most challenging. The advantage of this approach is rel 
evant: geodetic monitoring techniques can achieve very precise 
measurements but limited to few control points, while point cloud 
analysis extends the observation to whole surfaces, including ar 
eas which usually are not investigated. In literature different ex 
periences are reported, involving for the most applications to Ge 
ology, Civil and Building Engineering (Vosselmann and Maas, 
2010). One of the common issues afforded by several authors 
is how to cope with uncertainty in point clouds, being this the 
bottle-neck of deformation measurement by TLS. Three main as 
pects contribute to the error budget, which is larger with respect 
to the standard monitoring techniques: precision of intrinsic mea 
surements, point cloud registration, and data modelling. An in 
teresting strategy that was adopted to overcome the problem of 
the measurement uncertainty is given by the so called area-based 
techniques. These make use of surfaces (planes or other regu 
lar shapes) interpolating the point clouds to be compared. This 
task might be performed on the whole object, when it features 
a known shape (Lindenbergh et ah, 2005. Schneider. 2006, Gor 
don and Lichti, 2007), or on some parts of it (Lindenbergh and 
Pfeifer. 2005). In both cases, the object surface should be regular, 
like frequently occurs in the analysis of man-made structures. A 
higher degree of complexity is involved in the geological field, 
especially w-hen dealing with deformation analysis of rock faces. 
Indeed, different applications were successfully carried out on 
terrain slopes and landslides, due to the fact the displacements 
to detect are very often larger than the accuracy of the adopted 
sensors (Abelian et ah. 2006, Teza et ah. 2007). When dealing 
with cliffs where the rockfall risk is relevant the problem becomes 
more complex, because the accuracy needed for failure forecast 
ing is very often lower than the uncertainty of the adopted ob 
servations. A small number of papers were published so far on 
this subject, and no one presents an exhaustive and general ap 
proach. Interesting inputs can be found in (Abelian et ah, 2009). 
Besides the problem of the required accuracy in data acquisition 
and modelling, ranging in the order of few cm up to 0.5 mm ac 
cording to the size and the topography of the site, some further 
problems have to be tackled. Deformations might generally occur 
on an entire portion of a slope, or they might affect a local region 
only. The former requires to establish a stable ground reference 
system (GRS), calling for the use of high precision geodetic tech 
niques. Alternatively, a comparison with external stable rock ar 
eas is needed, but this solution usually does not guarantee enough 
accuracy. The latter might be overcome by considering relative 
displacements between close regions. In addition, on rock faces 
vegetation can grow, and blocks can fall down between observa 
tion epochs, resulting in significant major changes on the surfaces 
where deformations occurred. 
In this paper a method to perform a deformation analysis of a rock 
face is presented, accounting for both the detection of local major 
changes and widespread deformations. Examples of application 
to a synthetically generated dataset and to a simple real dataset 
are reported in Section 3. 
2 A TECHNIQUE FOR THE COMPARISONS OF ROCK 
FACE SURFACE ALONG TIME 
The basic concept that was followed here is to exploit the data re 
dundancy of a point cloud to improve the precision of detectable 
deformations and changes. However, the application of this prin 
ciple to rock faces is more complex than it is in case of man 
made structures, due to the presence of irregular surfaces which 
prevent from interpolation with analytical functions, to possible 
major changes on point clouds, and to the millimetric precision 
required. 
The full data processing workflow proposed and discussed in this 
paper is shown in Fig. 1. The method requires the preliminary 
acquisition and registration of each point cloud by using stan 
dard methods of the adopted sensor technology. Here the use 
of TLS is assumed, but 3D modeling through Photogrammetry 
could be also used, if enough accuracy and resolution are pro 
vided. Then, point-clouds are segmented into several 2.5D re 
67