In: Paparoditis N., Pierrot-Deseilligny M„ Mallet C., Tournaire O. (Eds), 1APRS. Vol. XXXVIII. Part ЗА - Saint-Mandé, France. September 1-3. 2010 
gions, each of them referred to its own plane. This task allows 
to interpolate the raw' data to obtain a grid DEM for each region. 
Considerations about scan georeferencing and preliminary filter 
ing from vegetation are reported in (Alba et al., 2010). Hereafter, 
the processing is carried on by computing point-wise differences 
between both DEMs of the same region corresponding to differ 
ent epochs. The ADEM achieved this way is then analysed along 
a three-step procedure. First, possible systematic effects or low- 
frequency deformations are extracted by looking for a linear com 
ponent in the ADEM. If this component is statistically signifi 
cant, it is removed from the dataset. Thus the ADEM is checked 
against major changes, i.e. loss of material or vegetation growth. 
Finally, deformations are looked for by analyzing the mean de 
formations computed on square windows of a few decimeter side, 
process that should theoretically improve the original precision of 
each point. Deformations are detected on the basis of statistical 
testing, which requires an estimate of the ADEM precision (see 
Par. 2.6.1). 
2009). An approach based on the relative scan coregistration by 
means of ICP algorithm might allow to improve the relative ac 
curacy, even though it requires that stable parts are in the scene. 
The technique that is currently implemented for the analysis of 
surfaces requires that both point clouds to be compared are re 
sampled on a regular grid (DEM) established with respect to a 
reference plane. If the entire rock face under investigation has a 
complex morphology and cannot be referred to one plane only, 
a segmentation step is required to split up the whole dataset in 
smaller 2.5D regions. The method proposed by Roncella and 
Forlani (2005) is well suitable to this aim. This is based on 
a RANSAC segmentation technique requiring the definition of 
two input parameters: the maximum allowed distanced of a point 
from its reference plane, and the minimum number of points per 
region. In order to avoid different segmentations at two epochs, 
this procedure is applied to point cloud 1. and then the boundary 
of each region Regu is utilized to subdivide point cloud 2. Here 
after, each region Regk will be separately analysed; accordingly, 
the subscript index k will be omitted. A plane 7r is estimated by 
Least Squares on the basis of points belonging to a specific re 
gion only. Finally, data can be resampled to a grid lattice, whose 
resolution Sdem (from now on called ‘DEM Unit’-DU) is very' 
close to the one of the original data to avoid loss of information. 
Each Reg at epoch t gives rise to a surface that is described by a 
rectangular matrix DEM 1 to be used as input for next processing 
steps. 
2.2 Computation of the ADEM 
The deformation of a cliff consists in the change of the shape 
of its surface, due to sliding of rock masses along discontinu 
ities. Usually deformations are preliminary to rockfalls, whose 
magnitude can depend on several factors. Thus the deformation 
analysis of a cliff’s surface surveyed at two or more epochs must 
comprehends two main stages. The first one is the change de 
tection (CliDet), which is focused to find rocks that fell down 
between two observation epochs and to filter out the grown veg 
etation. Regions that are interested by these two processes must 
be excluded from the next stage of the analysis. This is repre 
sented by the deformation analysis (DefAn) aimed to locate the 
areas that were affected by shape deformations. 
The first step before proceeding with both items is to compute 
a new matrix ADEM resulting from the difference of the two 
DEMs concerning the same region (Reg): 
ADEM 12 = DEM 2 - DEM 1 . (1) 
Figure 1: Workflow of the procedure for rock face deformation 
analysis. 
2.1 Preliminary data processing 
Deformation analysis is carried out on the basis of two datasets 
taken under the same operational conditions at epochs t\ and 12. 
Details about data acquisition planning and instruments can be 
found in Alba et al. (2005). At least one epoch data must be geo- 
referenced into a GRS in order to align the 2 axis of the project 
RS (PRS) to the local plumb-line, and to refer data to the national 
mapping system. Indeed, in the geomorphological analysis these 
tasks are generally needed to define the spatial orientation of rock 
discontinuities. On the other hand, the use of GCPs to register 
point cloud at epoch t-2 into the same RS cannot be enough to 
obtain an accuracy suitable to detect deformations (Alba et al., 
The area-based methods for deformation analysis which have 
been quoted in Section 1 make use of interpolations of the orig 
inal point clouds. Unfortunately, this approach can’t be directly 
applied to the problem discussed here because of irregular sur 
faces. In absence of deformations and changes, the ADEM eval 
uated from Eq. 1 should be flat and regular, so that an area-based 
technique could be applied to it. Theoretically, elevations z(i,j) 
of the ADEM should be normally distributed as N(0, a 2 ), where 
a z is the std.dev. of each ADEM point along 2. Both algorithms 
ChDet and DefAn will look for discrepancies with respect to this 
stochastic model. 
On the other hand, the use of ADEM introduces the following 
drawbacks: (i) the computation of ADEM requires interpolation 
of the original point clouds, task that increases the correlation 
between points at the same epoch; (ii) the use of DEMs results 
in a low-pass filtering, with consequent loss of information; (iii) 
misalignment errors on both DEMs at epochs t\ and t2 might be 
emphasized when computing the ADEM.