no operators`s intervention

Authors: Krištof Oštir, Aleš Marsetič, Peter Pehani, Matej Perše,
Tatjana Veljanovski, Klemen Zakšek, Janez Zaletelj, Tomaž Rodič
Prototype of an Automatic Near Real-Time
Satellite Image Processing Chain
Space-SI is developing a complete automatic (no operators’s intervention)
near real-time processing chain from raw optical satellite image to webdelivered map-ready products. Prototype presented here is an functional
subset of the planned full chain. Currently prototype is operational in the
area of Slovenia and for the RapidEye imagery.
Full implementation is planned for the end of 2013, upgrades are possible
afterwards.
Processing newly
obtained images
Firewall and
switch
Main
Control
Module
User
requests
Image
Processing
Server
Pre-processing:
- Geometric corrections
(orthorectification)
- Radiometric corrections
(topographic normalisat.)
Application
(web)
servers
JAVA/IDL
bridge
Processing:
- NDVI
Database and
Image server
Database
JAVA
IDL , C++
System architecture.
Motivation
Automatic and fast processing of satellite images is an everlasting challenge in remote sensing.
Scheme of processing chain
ground station /
raw optical images
Prototype basics
More emphasis given to the pre-processing steps (geometric and radiometric corrections) than towards the processing steps (data interpretation), since the first are prerequisites for successful completion of all subsequent operations.
All processing is fully automatic, mutually connected and managed by the
main control module.
The chain is triggered by the appearance of a raw satellite image on the
FTP server.
Individual processing (sub)modules are developed in IDL and C++, main
control module and web viewer however in Java.
P 1.1
automatic
geometric
correction
D
s c rip
Deesc
riptio
tionn
R
s uults
Rees
lts
S U R F p o in ts
S U R F p o in ts a re a re a -b a s e d
S U R F po in ts a re a re a-ba sed
fe a tu re s w ith lo c a l a re a d e s c rip to r.
ferie
a tun re
s wnith
rea
scrip
O
ta tio
a nlo
d ca
s cl aale
in vde
a ria
n t. to r.
N
e to
e lo
a ll
Noott re
relia
liabble
le.. DDuue
to lalarg
rge
locca
d iffe re n c e s b e tw e e n s a t. a n d a e r. im a g e
d iffe
sa nt.sand
r. aim a ge
in
te n ren
s itiece
s ,s pbe
o intw
t ee
lo cna tio
a n dae
a re
sitie
loaca
tion
din
e te
s cn
rip
to rss,d po
o nin
o tt m
tc h
. s and a rea
O rie n ta tion and scale in va rian t.
de scrip to rs do n o t m a tch .
P
se
P hh aa se
c
rrela
latio
tionn
c oorre
M
h in
h a sse
e sspectru
p e c tru m
ak
Maatctch
ingg ooff ppha
m pp eea
kss
to fin d th e d is p la c e m e n t.
N
e in
te nnssitie
itie ss aare
o
Noott re
relia
liabble
le.. Im
Imaagge
inte
re to
too
ddiffe
n t,t, im
o sssib
s ib le
iffere
ren
imppo
le to
to fin
findd
c o rre s p o n d in g p e a k s .
Im
Im aa ggee
c
rrela
latio
tionn
c oorre
AAbbsso
o lu
lid e aan
n d
ta n
e
lute
te oorr EEuucclide
dis
ista
n cce
b e tw e e n a lig n e d im a g e s
N
m ee re
g io n ss hha
a vve
e
Noott re
relia
liabble
le.. SSaam
region
d iffe re n t im a g e in te n s ity .
O
jecctt-b
s edd
O bb je
-b aa se
c o rre la tio n
Im
e ddiffe
n cce
e is
a lc
u la
d fo
Imaagge
iffere
ren
is cca
lcu
la te
ted
fo rr
d is tin c t o b je c ts o f s a te llite im a g e
B
r. SSm
n cce
e w
n im
e ss
Beette
tter.
maallll ddiffe
iffere
ren
whheen
im aa gge
aare
o rre
tly aalig
n eed
d ..
re cco
rrecctly
lign
c o rre la tio n
R
R oo aa dd
m a tc h in g
m a tc h in g
to find th e d isp la ce m en t.
co rre spo nd in g pea ks.
be tw e en a ligned im a ges
R
a d ss lin
e sse
e ggm
m een
n ts
x tra ccte
te d
Rooad
line
ts eextra
d
fro
m im
e ss.
. M
h in
ed o
from
imaagge
Maatc
tch
ingg is
is bb aa ssed
onn
g e o m e try a n d lo c a l im a g e
ge
om
e
try
a
nd
lo
ca
l
im
a
ge
in fo rm a tio n .
Table left: For the implementation of the sub-module P1.2
Automatic GCP extraction we tested several methods, the
most promising seems road matchig.
Figure below: workflow of image matching.
R ap id E ye Im a ge
d iffe ren t im a ge in ten sity.
d istin ct ob je cts o f sa te llite im a ge
info rm a tio n .
Im a ge P re p ro ce ssin g
PPro
a d lin
e ggem
e m een
n ts
rom
mis
isin
ingg.. RRooad
linee sse
ts aare
re
w e ll d e fin e d o n b o th im a g e s . D iffic u lt to
w e ll defined o n bo th im a ge s. D ifficu lt to
c la s s ify ro a d s o n a e ria l im a g e .
F e a tu re
D e te ction
F e a tu re
D e scrip to rs
F e a tu re
M a tch in g
A e ria l Im a ge
cla ssify ro ad s o n ae ria l im a ge .
Robust estimation
of global image
transformation parameters
Im a ge P re p ro ce ssin g
F e a tu re
D e te ction
Im a ge
T ran sfo rm a tion
F e a tu re
D e scrip to rs
P 1.2
P 1.3
P1.3 Sensor modelling and orthorectification sub-module
P1.3 Sensor model and ortorectification:
A generic physical sensor model was defined to accommodate various (full-frame
and pushbroom) optical sensor types.
Tests are being performed on RapidEye
images. Ground control and check points
are obtained from module P1.2.
Orthorectification is done with single-ray
backprojection and a nationwide DTM.
form ation of
collinearity e qu ations
exterior orienta tion
m odellin g
P 2.1
self-calibratio n
automatic
radiometric
correction
m etadata
integration in
collinearity e qu ations
groun d control and
check points
form ation of w eight
m atrix
iterative le ast squares
adjustm e nt
form ation of w eight
m atrix
accuracy check
P 2.2
D E M (D S M )
orthorectificatio n
Figure above: Geometric model workflow.
Figure right: orthorectification of a pushbroom image.
Final implementation (end of 2013)
Support to various full-frame and pushbroom sensors will be added.
M
Meeth
thoodd
S U R F p o in ts
Supported sensor: RapidEye
Region: Slovenia (national coordinate reference system)
Refinement and optimisation of prototype modules.
The following modules will be added:
* P2.1 Sub-module for atmospheric correction
* R1 Change detection sub-module
* R2.2 NDVI changes sub-module
* S1 Data catalog
* S3 End user/administrator triggered re-processing
ground station /
raw optical images
P1.2 Automatic ground control points extraction sub-module
Prototype implementation (June 2012)
P1 Automatic geometric correction
* P1.1 Data preparation sub-module
* P1.2 Automatic ground control points extraction sub-module
* P1.3 Sensor modelling and orthorectification sub-module
P2 Automatic radiometric correction
* P2.2 Sub-module for topographic correction
R Automatic generation of results
* R2.1 NDVI generation module
S End user services
* S2 Web mapping of products
Processing chain module details
2 datasets
P2.2 Sub-module for topographic correction
automatic
interpretation
R 2.1
R1
P2.2 Topographic correction was solved
with an extended combination of Image
Processing Workbench (IPW) method and
Minnaert method.
LN =
2 datasets
L
illm K
illm = p
R 2.2
cos i
+ (1 − p ) svf + a (1 − svf )
cos z cos e
direct
irradiance
diffuse
irradiance
from the
sky
diffuse
irradiance
from the
terrain
p = fraction of direct irradiance within total irradiance, i = incidence angle, z = sun zenith angle
e = slope, K = Minnaert coefficient, a = albedo, svf = sky-view factor (portion of visible sky)
Figure left: Comparison between false
colour composites: orthorectified (above)
and topographically normalized images
(below) for the area around Maribor (left
part of the figure urban area, middle part
ski-centre covered by snow). As seen on
the hills around the city the corrected
results reveal almost no influence of the
topography.
Well-known processing workflow
with no operator's intervention!
R2.1 Module for NDVI generation
database
Main control module, database
Main control module is the “brain” of the processing chain, with the following tasks:
* control of available processing resources (e.g. CPU, memory, etc.),
* control of correct execution of individual image processing modules,
* communication with all other parts of the system (i.e. the web application and database) which
were developed in Java environment.
It is developed in Java. The Java/IDL bridge and input/output XML forms are used to pass/receive
the processing commands and parameters to/from the individual modules of image processing
chain which are developed in IDL and C++.
Database is implemented on the PostGIS infrastructure.
* raw optical images
* orthorectified images
* radiometrically corrected images
* interpretation results
* metadata about processing
> 0.6
0.4
0.2
0.0
< -0.2
S 1, S 2, S 3
Figure left: RGB rendering of NDVI for the
area around Maribor for situation in
March 2011.
end user /
products delivery
Figure below: On the operational level the execution of the prototype processing chain is controled by the application window.
end user /
products delivery
R2.1 NDVI (Normalized Difference Vegetation Index) is a standard product in
remote sensing of vegetation. To improve
visual interpretation of NDVI and to
enable easier comparison between two or
more NDVI products from two or more
different satellite images, we use RGBrendering based on a fixed color scheme.
S2 Service: Web mapping of products
Figure above: Planned full processing chain. The modules are denoted as:
P - Pre-processing modules (geometric and radiometric corrections)
R - modules for automatic interpratation yielding Results
S - end user Services
Modules implemented in the prototype are marked in darker colour.
S2 Web mapping of products is one of the
possibilities of final product dissemination. All the key products are passed to
web mapping application which enables
GIS functionalities such as geo-location of
processed images and some basic feature
querying (e.g. sensor type, available products, date, etc.).
Additionally a WMS server was developed
to facilitate the use and integration of
products into other third-party applications.
Figure left: End user viewer.
W: www.space.si
E: offi[email protected]
Corresponding authors: [email protected] (overall), [email protected] (overall), [email protected] (P1.2),
[email protected] (P1.3), [email protected] (P2.2), [email protected] (S2 and MCM)