Post on 08-Jun-2020
Definiens Enterprise Image Intelligence Technology
Dr. Waldemar Krebs – Account Manager Earth Sciences
Prag - Jan. 2008
Remote Sensing Imagery: A Great Source for Geointelligence
• From overview to detail
• Recent information on any areas around the world
• More than eyes can see
Extract Relevant Information in Time: The Great Challenge
• Select within a tsunami of imagerythe right data set and area within the imagery
• Provide objective analysis based on challenging data under huge time pressure
• Automation of the human expertise has proved challenging or impossible before now
Multi SourceDate Provision Geo - Intelligence
Information Extraction: The Bottleneck
Change Detection
Target Detection
Mapping
Target Recognition
GIS
Thermal
RADAR / LIDAR
OpticImagery
overwhelming volume of data
highly experienced image analystrequired to screen huge amounts of data
time consuming & error prone
available systems for automation areproprietary solutions for single applicationswith huge maintenance costs
critical information is detectedtoo late or not at all
expensive quality assurance process
no accurate information extractionwith reasonable automation degree available
Information Extraction
Definiens CognitionNetwork Technology ®
Multi SourceDataProvision
Geo - Intelligence
The Solution: Geo-Intelligence Supply Chain using eCognition
Change Detection
Target Detection
Mapping
Target Recognition
GIS
Thermal
RADAR /LIDAR
OpticImagery
Information Extraction
Implemented in Definiens Enterprise Image Intelligence® Suite
1995: Research ‚Think-Tank‘Created by Physics Nobel Prize Laureate Prof. Dr. Gerd Binnig
1998: Cognition Network Technology (CNT)20+ international patents
2000: CommercializationFunded by TVM and CIPIO Partners
2003: Focus on Image Intelligence
2004: Focus on Life Sciences and Geospatial Intelligence
2007: Staff 80+, 2500+ licenses worldwide,
HQ Munich, US operations
Company Background
About Definiens Enterprise Image Intelligence
� Definiens Enterprise Image Intelligence is the only technology that
understands images similar to the human mind.
� Definiens provides breakthrough Enterprise Image Intelligence to customers
like NGA, Lockheed Martin, DigitalGlobe, European Commission etc.
� Other technologies are still focused on single PC performance and pixel
based image analysis, whereas Definiens uses object and context based
technology.
� Definiens Image Intelligence provides scalable solutions from laptop to
production centre and thus guarantees maximum performance.
Applications Overview
Urban MonitoringUrban Land Cover Mapping
GMES-GUS
Semi-Automatic Land Use Classification
17%
13%
4%
4%
0%
9%2%
16%
6%
6%
13%
1%
9%
1.1.1.1 Residential continuous dense urban fabric
1.1.1.2 Residential continuous medium dense urban fabric
1.1.2.1 Residential discontinuous urban fabric
1.1.2.2 Residential discontinuous sparse urban fabric
1.2 Industrial, commercial and transport units / 1.3 Mine,dump and construction sites1.2.2.1 Road networks and associated land
1.2.2.2 Rail networks and associated land
1.4 Artificial non agricultural vegetated areas
2.1 Arable land
2.3 Pastures
3.1 Forests / 2.2 Permanent crops / 3.2 shrub and/orherbaceous vegetation associations4. Wetlands
5. Water bodies
17%
13%
4%
4%
0%
9%2%
16%
6%
6%
13%
1%
9%
1.1.1.1 Residential continuous dense urban fabric
1.1.1.2 Residential continuous medium dense urban fabric
1.1.2.1 Residential discontinuous urban fabric
1.1.2.2 Residential discontinuous sparse urban fabric
1.2 Industrial, commercial and transport units / 1.3 Mine,dump and construction sites1.2.2.1 Road networks and associated land
1.2.2.2 Rail networks and associated land
1.4 Artificial non agricultural vegetated areas
2.1 Arable land
2.3 Pastures
3.1 Forests / 2.2 Permanent crops / 3.2 shrub and/orherbaceous vegetation associations4. Wetlands
5. Water bodies
GUS - Consortia - funded by European Space Agency (ESA)
Urban Land Use – Present Land Use
Legend
1.1.1 Continuous urban fabric
1.1.1.2 Residential continuous medium dense urban fabric
1.1.1.1 Residential continuous dense urban fabric
1.1.2 Discontinuous urban fabric
1.1.2.1 Residential discontinuous urban fabric
1.1.2.2 Residential discontinuous sparse urban fabric
1.2 Industrial, commercial and transport units /1.3 Mine, dump and construction sites
1.2.2 Road and rail networks and associated land
1.2.2.1 Road networks and associated land
1.2.2.2 Rail networks and associated land
1.4 Artificial non agricultural vegetated areas
2.1 Arable land
2.3 Pastures
3.1 Forests / 2.2 Permanent crops /3.2 Shrub and/or herbaceous vegetation associations
4. Wetlands
5. Water bodies
GMES Urban Services
Urban Land Use – Present Land Use
Statistics
8.87 19.84 5. Water bodies
0.86 1.92 4. Wetlands
12.91 28.86 3.1 Forests / 2.2 Permanent crops / 3.2 Shrub and/or herbaceous vegetation associations
6.31 14.11 2.3 Pastures
6.14 13.73 2.1 Arable land
15.62 34.93 1.4 Artificial non agricultural vegetated areas
1.83 4.08 1.2.2.2 Rail networks and associated land
9.45 21.13 1.2.2.1 Road networks and associated land
0.07 0.16 1.2 Industrial, commercial and transport units / 1.3 Mine, dump and construction sites
3.73 8.34 1.1.2.2 Residential discontinuous sparse urban fabric
4.20 9.40 1.1.2.1 Residential discontinuous urban fabric
13.22 29.57 1.1.1.2 Residential continuous medium dense urban fabric
16.75 37.46 1.1.1.1 Residential continuous dense urban fabric
Portion of overall area [%]
Overall area [km²]
17%
13%
4%
4%
0%
9%2%
16%
6%
6%
13%
1%
9%
GMES Urban Services
1. Spot 5
pan: 2,5m resolution
ms: 10m resolution
(nir, red, green, swir)
2. Landsat ETM
pan: 15m resolution
ms: 30m resolution
(b, g, r, nir, swir, mir, thermal)
Input Data - Raster
Munich
Munich
Input Data – DEM
SRTM (SAR Radar Topography Mission)Ground Resolution: 90
referenced in: UTM, WGS84
tiled in: 1201 x 1201 pixel
Covering: world wide
Dresden
Input Data – Vector Data
1. TeleAtlas (used in navigation systems)Format: *.shp file / standard GIS format
Available for Western Europe and United States
Used features: Road network in 8 classes (from motorway to local road)
2. NavTeq (used in navigation systems)Format: *.shp file /standard GIS format
(possible using in future)
3. Digitized Road systemdigitized from Quickbird-Data (Barcelona) / Ikonos-Data aerial photos / using ArcGIS/ArcView
4. Customers delivery
Classification Data - Classes
Residential denseurban fabric
Residential mediumdense urban fabric
Residential sparseurban fabric
Industrial / Commercial / Public
Urban Trees
Classification – Results
Example Munich
Classification – Roads and Houses
Classification – Urban Structures
Classification – Urban Landuse
Classification – Impervious Areas
Useful for:
� Land consumption
� Flood protection
Forestry
eCognition Forester
eCeC--ForesterForester
TreeTree CountingCounting
inYoungForestStandsinYoungForestStandstest site: Scotland
2006 trees/ha
A study for Foresty Commission / Stewart Snape by Definiens AG
eCognition Forester 1.1
A study for Forestry Commission of Great Britain
tree topsHigh resolution ortho-photo (RGB) Exportable classification results:
subset
0 250 500 [m]
2006 trees/ha2006 trees/ha
1844 trees/ha1844 trees/ha
Processing time: 30 sec/ha Inaccuracy: + - 5%
eCognition Forester
Automatic classification of tree tops in forested areas
Automatic export of results to GIS
Visualisation of tree tops with graduated symbols, e.g. based on area of tree tops
Workflow
eCognition Forester
Transfer of rule base to ortho-photos of different areas with appropriate
additional thematic layers.
Transferability
2006 trees/ha
Subsets of ortho-photo Classification result Exported tree position
eCognition Forester
Transfer of rule base to ortho-photos of different areas with appropriate
additional thematic layers.
Transferability
3244 trees/ha
3244 trees/ha
2875 trees/ha
Subsets of ortho-photo Classification result Exported tree position
eCognition Forester
Transfer of rule base to ortho-photos of different areas with
appropriate additional thematic layers.
Transferability
2460 trees/ha
Subsets of ortho-photo Classification result Exported tree position
eCognition Forester
Have a look at details of a subset
eCognition Forester
Details - Input dataDetails - Classification of single trees
Have a look at further details ����
eCognition Forester
Further details –Classification of single trees
eCognition Forester
Further details –Classification of single trees
Definiens & Silvatech developed a robust strategy for new cutblock identification usingmultitemporal Landsat5 and Landsat7 data.
eCognition technology provides� Robust change detection based on satellite imagery
� Efficient fusion of different satellite imagery and ancillaryinformation like DEM.
� Automated extraction of polygons leads to signficant time savings
� Definiens Professional Services allowed� Silvatech to explore the full potential of eCognition and
achieve impressive results in a short time frame -fully satisfying their customer (MSRM).
10 km
Consultancy project carried out for: 1) MSRM: Ministry of Sustainable Resource Management, British Columbia - Terrestrial Information Branch
DEMLandsat 5 dataset
Input data
Cutblock Water
Landsat 7 dataset
Canada - Cutblock Inventory
Cutblock Inventory in British Columbia for MSRM1 Consultancy project together with SilvatechConsulting Ltd.
Clear cut extraction withcurrent technology
clear cut extraction with eCognition‘s polygonsLandsat data set (5 and 7)
GIS procedures for extracting inventory polygons from satellite remote sensing data
Canada - Cutblock Inventory
Different input datasets are used to optimize results
Landsat scene Exported results in *.shp format
Development of eCognition rule bases on subsets
Utilization of existing satellite information along with auxiliary data
Automatic generation of geo-information for entire image mosaic
Robust Extraction of Clearcuts from Multi-temporal Landsat Data
Canada - Cutblock Inventory
Developed Prototype:
• automatically creates polygon shape files representing cutblockpolygons, coastlines and water
• delivers according statistic
• saves developed strategy in protocol file for automatic reuse
Output shp files of initial data set(Rivers Inlet, BC – June 2000)
Clear Cut statistics over whole area
CutblockWater
Class cutblock water_L3 Objects 137 96Sum Area 3.20E+07 1.68E+08Mean Area 233489 1.75E+06StdDev Area 203644 1.07E+07Min Area 10350 11250Max Area 1.03E+06 1.02E+08Sum Length 132280 128810Mean Length 965.548 1341.77
Second Data set (Kamloops BC,- Aug 2001)
Robust technology allows successful strategy transfer
Canada - Cutblock Inventory
Pipeline Planning
Export Suitability Map
shp. with attributes
Schematic Workflow
data fusion
vectorlayer(GIS)
image layer
Input Processing Final Result
data input result
landcoveranalysisEO data
GIS analysis+ distances
slopecalcualation
EO data
Suitability Map
•suitability classes•GIS attributes•context attributes•experts comments
action
Analysis of Input Data
Intersection and Categorization
Overall Concept
� using cost effective, publicavailable data
� coarse evaluation of suitable/risky areas
� the in Level 1 identified areasare detailied analysed
� additional more detailied GIS and remote sensing data areused
Level 2:Detailed Suitability/Risk Calculation
Level 1:Coarse Suitability/Risk Calculation
Level 1: Coarse Suitability
Coarse Suitability Map
the used data are free of charge and public avaliable
Input 1. Analysis
e.g. streets, existing tracks, soil map,….
slope calculation
up to date land cover
suitability criteria in
eCognition PIPEMON
high suitability
good suitability
limited suitability
reduced suitability
excluded
GIS
Elevation
Landsat
Level 2: Detailed Suitability
land use, risks, legal restrictions, infrastructure…
land cover
information derived in Level 1 analysis
GIS
Level1
Aerial Image
Detailed Suitability MapInput 1. Analysis
Suitability criteria in
eCognition PIPEMON
high suitability
good suitability
limited suitability
reduced suitability
excluded
GUI eCognition PIPEMON – Test site Dorsten
GUI eCognition PIPEMON – Test site Dorsten
GUI eCognition PIPEMON – Test site Dorsten
eCognition PIPEMON - Steps to Suitability Map
GUI eCognition PIPEMON – Test site Dorsten
Conclusion
�Flexible: use any kind of GIS and remotesensing data available
�Scalable:different levels of detail possible
�GIS Data are updated and areas evaluated
�Expert knowledge can easilybe implemented and retracably administrated
�Exported results contain all relevant information used forsuitability/risk assessment
SuitabilityHigh SuitabliltyGood SuitabliltyReduced SuitabliltyLimited SuitabliltyExcluded
010001
11.39909
04_limitedsuitability6
010001
11.39909
04_limitedsuitability5
000101
10.7790902_goodsuitable4
000101
10.7790902_goodsuitable3
000101
10.7790902_goodsuitable2
010001
11.62909
04_limitedsuitability1
010001
11.62909
04_limitedsuitability0
exclude
limited
reduced
Good suited
High suited
Slope
srtm
Class_ID
Layer
Handle Classid
Route Planning in Arid Areas
Feasibility Study for EADS - Dornier
Data: Cheap and nearly everywhere available
• Aster Level 1B VNIR• Radiometric resolution: 3 bands• Spatial resolution: 15m
Automatic extraction of rivers and wadis to enable
• manual route planning for different routes on actual data sets
• identify, count and report cross-overs
Route Planning in Arid Areas
river bed / Wadis 0 5 10 km
River and wadi extractionin 3.6 s/ km2 with high accuracyon Windows XP; Pentium 4; 1,4 GHz; 2 GB of RAM:
Use extracted featuresin mobile environment
• Export of river bed polygons(ESRI Shape)
• Easy post-processing• Transfer to PDA possible for
mobile use
Next Steps
64Route 3 Crossing River/Wadi
20Route 2 Crossing River/Wadi
13Route 1 Crossing River/Wadi
Route Planning in Arid Areas
Marine Applications
Off Shore Oil Detection
Prestige Tanker Disaster (Coast of Galicia, Spain)
Data: ASAR Wide Swath
Oil and Coast Line Detection
Security Applications and Projects
Road centerline:
Rapid Mapping – Mobility Database
Basic landcover classification
Successful test at Fort Knox: Fully automated map production using eCognition
automatedroad detection
¢erline
export
Route planning
Analysis for Route Planning
+ Landuse based on IKONOS+ Slope Analysis based on SRTM+ Soil type using USGS tpye+ actual weather using METARS
criteria forGO/No GO
Information layer combininginformation from multiple sources
Rapid Mapping – Mobility Database
Workflow: Fully automated map production using eCognition based using EO imagery
Ship Detection
Norfolk SubsetsAssisted Ship Detection / Recognition
Data courtesy: Digital Globe, work partly funded by RESTEC and NGA
Automated AttributionAssisted Ship Detection / Recognition
Data courtesy: Digital Globe, work partly funded by RESTEC
On screen review
Execution and quality control with eCognition Analyst
Information on processed images
Information on created tiles
Number of detected Ships /Submarines
Used processor
Tiled image
Complete image
Assisted AirplaneDetection / Recognition
Image Resolution: 11.2 cm
Automated Airplane Detection in Airborne Imagery
Data courtesy and work funded by: DRDC - Defense Research and Development Canada Valcartier Optronic Surveillance, Data Exploitation Group
Assisted Airplane Detection / Recognition
Data courtesy and work funded by: DRDC - Defense Research and Development Canada Valcartier Optronic Surveillance, Data Exploitation Group
Automated Airplane Detection – aerial photography
Automated Airplane Detection in Airborne Imagery
Data courtesy and work funded by: DRDC - Defense Research and Development Canada Valcartier Optronic Surveillance, Data Exploitation Group
Wizard Guided Target Recognitionand GIS Integration
Evaluation of Wizard Guided Target Recognition in Aerial Imagery
Data courtesy and work funded by: DRDC - Defense Research and Development Canada Valcartier Optronic Surveillance, Data Exploitation Group
Automated Vehicle Detection
Image Data Results
High Resolution Airborne Off-Nadir 30º Image
Automated Vehicle Detection
nadir 30 degrees
parking lotvehicles
High Resolution Airborne Nadir and Off-Nadir Images
Vehicle Detection with ATRc eCognition Wizard
�eCognition Rulebase forVehicle Detectionusing spectral + shape features
�Classification
�InputHigh Resoluted Airborne Data different Angles
�Export as Statisticsand for GIS
Data courtesy and work funded by: DRDC - Defense Research and Development Canada Valcartier Optronic Surveillance, Data Exploitation Group
Video
Detection Results on Video Stream of Infrared Imagery
detected vehiclehit missed false alarm
Conclusion
Unique and Proven TechnologyModels Human Cognition, context and understanding
Secure and ReliableObject based reliability, consistent quality, IP protection
Fully ScalableFrom Single Workstation to Enterprise Level
Multi Source IntelligenceMulti data and information fusion
Highly FlexibleModular architecture, easy adaptable
Thanks for your attention !
Dr. Waldemar KrebsAccount Manager Earth SciencesDefiniens AG
Phone: +49(0)89 231180-27Fax: +49(0)89 231180-90E-mail wkrebs@definiens.com
Mission
Global leader in Enterprise Image Intelligence
We Understand ImagesTM