he solid is
hat besides
rapping’ and
stigated and
acy of object
1 the angle
ce. Mapping
ingly; control
ly applied in
sophisticated
manding is a
! users to re-
ogrammetry.
g introduced
(reported by
Karras et al.,
; rather than
à further step
1oscopic and
architectural
. thesis, De -
f surfaces of
ric measure-
International
2. Analysis of
urnal of Geo-
9. 91-109.
1g of Cylindri-
ing, N.T.U.A.
| Differenzial-
elen aus der
bhe Mitteilun-
U. Wien.
IMAGE SEQUENCE ANALYSIS
Dana Klimesova, Tomas Suk
Institute of Information Theory and Automation,
Academy of Sciences of the Czech Republic
Pod vodarenskou vézi 4, 182 00 Praha 8, Czech Republic
Telephon: 42 2 6605 2247, 42 2 6605 2586
Telefax: 42 2 6641 4903
e-mail: klimes@utia.cas.cz
Commission V, IWG VIIII - Image Sequence Analysis
KEYWORDS: Forestry, Analysis, Transformation, Algorithm, Multitemporal
ABSTRACT:
The contribution deals with temporal context information in the field of temporal image data sets processing. We
propose the special comparative transform which make possible to evaluate the objects dynamics along the temporal
axis. We present examples of forest dynamics analysis using aerial data sets from the last about fifty years.
1. INTRODUCTION
Multisource and temporal analysis represents a very
effective method how to gain information about the state,
dynamics and future trends of observed landscape
objects and phenomena. Together with the technology of
geographical information systems it is a powerful tool to
provide a new quality of information. We used time series
of aerial photos to monitor the changes during fifty years
aimed at forest devastation, prevention process
evaluation and at the mapping of the range of mining
activity in the selected territory
2. TEMPORAL ANALYSIS
2.1. Temporal layers creating and preprocessing
Having in disposal temporal data sets from the observed
territory we are able even visually recognize some
significant changes but we are not able to quantify them
and distinguish enough all changes and their dynamics.
Usually we are dealing with data sets from different
sources which differs in resolution and technology of
scanning. In the case we have to use more than five
temporal layers the problems usually occur because of
control points selection. To cover the selected area it was
necessary to consider 48 photos from 12 years. The first
step we need to do was the geometric transformation of
all images (digitized aerial data sets). The registration
had to be done both inside one time layer and for all
overlaying layers. Great changes of landscape in case of
long period make impossible to use the same set of
control points for corresponding temporal layers. This fact
is more complicated when highly corrugated terrain is
processed. The difficulties are simplified when suitable
map layer is used as a reference image. We have to
estimate the type and parameters of the mapping
function. For the purposes of error estimation the
translation, similarity, affine, projective, quadratic and
surface spline transform have been applied with at least
20 control and 12 test points for temporal layer.
2.2 Error estimation
Error estimation in the number of points with respect to
control and test points was as follows:
Type of transform control points test points
translation 39,362 87,034
similarity 4,544 4,669
affine 3,835 3,486
projective 3,909 3,643
quadratic 2,476 2,646
spline 0,000 1,925
It means that the error less then 2 pixels was only yielded
by the surface spline transform.
n
iue 9 l 22
u=a, +a,x+a,y+ ) ff ln r,
iz]
v=b+bx+b,y+) 8 r nr
I
izl
where
2 2 2 2
r, z(x-x,) +(y-y,;) 1212." H,
n is the number of control points,
(x, Yi) and (u, ,v,) are coordinates of control
points in transformed and reference images respectively
(x, y) and (u,v) are coordinates of transformed
and reference image.
The coefficients
0,0450, D, 5, . 5, f, and g,
iz12 ..
we can obtain as solution of the systems of H+3
equations
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
