In: Paparoditis N., Pierrot-Deseilligny M.. Mallet C. Tournaire O. (Eds). 1APRS. Vol. XXXVIII. Part ЗА - Saint-Mandé, France. September 1-3. 2010 
an area of 20x20 m with DU=2 cm. the original std.dev. of 
points is assumed as a z = 0.25 DU. The first surface (epoch t\) 
is flat, while the other one (i 2 ) presents every kind of changes and 
deformation that are dealt with in this paper: 
• a global low-frequency deformation consisting on a shift 
along z whose mean value is half the size of the std.dev. 
of points; 
• 4 cavities resulting from rock detachments featuring two dif 
ferent shapes (parallelepiped and semi-sphere), a footprint 
on the surface ranging from 7-21 DU. a depth from 2-8 DU; 
• 3 areas with grown bushes with different shape and size; 
a parallelepiped (41x11x10 DU), a cylinder (21x51x11 
DU), and a semi-sphere (R=20 DU); and 
• a deformation consisting in a shift of 0.1 DU (2 mm) over a 
window of size 62.500 DU 2 (i.e. 12.50 m 2 ). 
Deformations with reliability 
: Ш ШШШШШШШШНШШШШ 
1 щшшшшншшшшншшш 
: ШШШШШШШпШШШШШ 
. шш шшшшшшшшшшшшш 
Щ Ш8ршш8г$ш1шшшшгш 
: ШШ ШШШНШШШШШНШШШ 
. шшшшьшшшшкышшшшш 
ттатттттттттттттт 
; Щ ШШШШШНШШИШШШШ 
шшшшшшшшшшшшшшшш 
Ixxxxxxxiiixxxxtuutxiixxiuxtxxxxxx-xxxtxxxxtxxt. 
Figure 5: Results of the significance analysis on the deformation 
map shown in Fig. 4; red circles indicates the center of windows 
where displacement was tested as significant. 
In both DEMs a white noise was added up to z coordinates, and 
then the ADEM generated. Input thresholds and internal param 
eters are shown in Table 1. The low-frequency deformation was 
detected with no significant coefficients a and b of the estimated 
plane, according to the fact it consisted only on a shift along z 
direction; this was estimated with an error of 1.7%, correspond 
ing to 0.1 mm. All the cavities and the vegetated areas have been 
correctly detected. The use of median filter in general improves 
border detection and then the area computation, even tough the 
results strongly depends on the size of the adopted search win 
dow. 
The area of local deformation has been correctly detected. As 
can be seen in Table 1, two different sizes for the averaging win 
dow have been tried: 20 and 30 DU. The number of the averaged 
points in each sample window was 400 and 900, respectively, 
meaning that the precision of single points is theoretically 20-30 
times better that of the sample mean. In reality, the contribute of 
correlations in Eq. 8 and the variance propagation (Eq. 7) when 
subtracting DEMs at different epochs reduce this increment of 
precision of 0.5 times. However, both configurations did not re 
sulted in any major different outcomes. In Fig. 4 the evaluated 
deformations are reported when using a window size of 20 DU; 
in Fig. 5. red circles draw the position of the centers of windows 
where the displacement was evaluated as statistically significant. 
3-0 plM of computed deformations 
3.2 Application to a real dataset 
After validating the algorithmic efficiency and the influence of 
individual input parameters on the dataset 1, the same algorithms 
have been tested on a real dataset 2. Data from a rock face col 
lected at different times have been used. In Alba et al. (2009) 
problems concerning data acquisition, georeferencing, and veg 
etation filtering are discussed . The point clouds were acquired 
by a RIEGL LMSZ420i TLS with a linear resolution of 1.3 cm, 
and covering a rock face sizing 25 x 15 m. First of all. both point 
clouds acquired and georeferenced at different epochs w'ere seg 
mented (see Fig. 6) by the algorithm explained in Subsection 2.1, 
and after each region REGu w'as analysed. For the dataset 2 only 
a region of 4.2 x 8.3 m with a strong presence of detachments was 
employed. Input thresholds and internal parameters are shown in 
Table 1. 
Here the application of the ChDet algorithm successfully identi 
fied 5 detachments, for a total volume of 0.035 m'\ and 3 growths 
of vegetation for 0.002 m 3 (see Fig. 7). The results of the search 
for detachments were partially validated by recovering of some 
pieces of rocks corresponding to the detected holes in the ADEM 
(Fig. 8). The deformation analysis showed displacements of a 
few tenths of a millimeter, that however were not considered as 
statistically significant. 
Example 
1 
1 
1 
2 
Parameters 
DU 
cm 
DU 
cm 
Wcav 
4 
8 
2 
10 
fiZcav 
2 
4 
1 
5 
W V eg 
7 
14 
3 
15 
OZveg 
5 
10 
3 
15 
tO de f 
20/30 
40/60 
4 
20 
A W 
20/30 
40/60 
4 
20 
<7 z 
±0.25 
±0.5 
±0.01 
±0.5 
A 
1 
2 
1 
5 
n w (%) 
90 
90 
90 
90 
Table 1: Input thresholds adopted in the processing of both sim 
ulated (1) and real (2) datasets. 
4 CONCLUSIONS AND FUTURE WORK 
Figure 4: Detected deformation in dataset 1, where a search win 
dow of size 20 DU was adopted. 
In the paper a method for change detection and deformation anal 
ysis of multi-temporal digital surfaces of rock faces has been pre 
sented. The achievable results are a map of the fallen blocks,