Crowdsourcing approach for large scale mapping
of built-up land
Kavinda Gunasekara
Geoinformatics Center
Asian Institute of Technology, Thailand.
Regional expert workshop on land accounting for SDG monitoring and reporting – 26 Sep 2017
Content • Introduction to AIT and Geoinformatics Center
• Project details
• Why we need build-up land
• Image classification approach
• Crowdsourcing approach
• Monitoring and Quality Control
• Project on data sharing platform in Pacific Island countries, support by UNESCAP and implementation by AIT
• Drones for mapping
Establish in 1959 as a Post Graduate School
Catering for higher education in Asia
• Over 1,600 Graduate students from 40+ countries • 14,000 alumni from 74 countries • 22,000 short-term trainees from 71 countries • Over 100 faculty members from 26 countries
Geoinformatics Center Established in 1999
Activities of the Geoinformatics Center - AIT
• Projects and Consulting Works
• Capacity building Programs, primarily in Asia and the Pacific
• QZSS GPS Monitoring Station and GNSS Research
• Emergency Disaster Response Mapping
• Rapid Mapping Support for Sentinel Asia & IDC
• Applied Research (DRR, Poverty, Environment, etc.)
• Exchange Programs: Students, Researchers, Experts
• Information Sharing and Publications: Journal, Conference, Reports, Manuals etc.
(ongoing)
http://arcg.is/2r9Lw5m
Project details
http://www.uttarakhand-dra.in/ Facebook: https://www.facebook.com/UttarakhandDRA/
Why we need built-up land
• Multi hazards (Flood, Earthquake, Landslide and Industrial) risk assessment project in Uttarakhand state • Size of Uttarakhand state: 53,483 km2
• Population: 10.08 million
• Location and extent of built-up land is essential for the risk assessment task
• Extent of built-up land is needed below the village level
• All the censes data at village level
Why? Interpretation on Google Earth
Red lines: village boundary Yellow polygons: built-up land
Image Classification Approach
• Tested methods a. Normalized Built-up Area Index (NBAI)
b. Pixel-based classification
c. Object oriented classification This methods gave best results and further extended
Tested on Google Earth images
Tested on Digital Globe Map API
• Limitations • High resolution image coverage over the state
• Commercial satellite images/budget
Google Earth Pro vs Digital Globe Map API
Object Oriented Classification – Segmentation
Object Oriented Classification – Classification
• Classification for a small area
• Results are better than a pixel based classification
Object Oriented Classification – Classification
• We can use object attributes to refine the classification
• E.g. Differentiate between roads and buildings using their length/width ratio.
Object Oriented Classification – Classification
Preliminary results of object oriented classification (built-up land shows in red)
Urban Area
(a) (b)
Use of object attributes to refine the classification (Yellow represents the built-up land) (a) Supervised classification (b) Refined classification using object attributes (e.g. width:length ratio)
Urban Area
Rural Area
Yellow polygons: built-up land
Summary of Object Oriented image classification
• Small area was able to classify successfully
• Batch processing was not able to perform for larger area
• Rural area classification was not successful
• High cost of high resolution images
Crowdsourcing approach
Architecture of crowdsourcing approach
Internet
1. Send Remaining Random Grid
3. Digitize
2. Get background Bing Map image
4. Send Digitize Data
5. Write Digitized data into database and make grid as
completed
USER 1
USER 2
USER 3
USER n
Number of Grids: 75563 Grid size: 1.2km * 0.8 km
Crowdsourcing approach
• Source and details • High Resolution Satellite Product – Bing Map (free) • Scale – 1 : 20,000 • Image Acquisition Dates – 2010 to 2015
• Training and quality control • Before Operational Stage - Each GIS Digitizers were trained before start the work to
enhance their ability to interpret high resolution Satellite Data • Operational Stage – Quality of digitizing work was randomly assessed daily basis and
guide GIS Digitizers improve their quality of work • After Operational Stage – After finishing all the grids, each grid was assessed by GIS
expert and identified around 12 % of the grids which haven’t done properly. And those grids were removed from the products and re-digitized by the best GIS Digitizers.
Notes: These buildings are very important to be digitized as those highly vulnerable to flood
Notes: Near the river and need to digitize all the building, could digitize as building cluster
Notes: Level of this details would be enough
Notes: All the buildings near the river and mountainous area need to be digitized
Notes: Hope you can understand level of details we expect to be digitized
Progress of the work
Live demonstration of crowdsourcing tool
• http://www.geoinfo.ait.ac.th/ukd/
Monitoring and Quality Control
Continuously monitoring the quality of the work and communicating with crowdsourcing people
Final product
Summary of crowdsourcing approach
• 8 non-GIS undergraduates utilized for the interpretation
• 4 GIS experts worked on the initial interpretation
• Digitization of whole state was completed within 3 weeks
• 4 GIS experts used for accuracy assessment and refined the errors within 2 weeks
Pilot project on implementation of data sharing platform in
Pacific Island Countries
Geo-portal (GEONODE)
Geo-portal is a centralized platform to share all the spatial data
• Open Data with Public
and
• Restricted data only among the Agencies
GEONODE is one such open source solution for a Geoportal
Image Source: https://e3geoportal.ecdc.europa.eu/E3%20Images/E3Geoportal_home.png
End User (Public)
Agency 1 Agency 2 Agency 3
Geoportal (GEONODE)
• Mainly Intended to Facilitate one way data flow from Data Providers to Public (End User) • Additionally, restricted data can be shared among only intended Agencies
Example from Tonga Pilot Project
3 Types of Data
• Layers – GIS or Satallite Image Data • Maps – Combination
of Layers • Documents
Data can be easily accessed
by,
• Searching with Key-
words • By Category
Current Data in Tonga Geoportal
Other Operational Geo-portals in Pacific
• Micronesia http://www.geoportal.oeem.gov.fm/
• Fiji http://www.fijigeoportal.gov.fj/
• Tonga http://202.134.25.30
Drones for mapping • Building low cost fixed-win drones
• Low cost drones for mapping
• Land cover mapping
• Vegetation health mapping
• 3D mapping/ 3D buildings
Thank you
Top Related