AYDES-CS (Crowdsourcing) (KİTLE KAYNAK TABANLI ANALİZ) ENTERPRISE CROWDSOURCING PLATFORM for GEOSPATIAL IMAGE ANALYSIS Muhammed KALKAN Muhammed.kalkan@tubitak.gov.tr *: Crowdsourcing
Famous Example: Wikipedia. What is Crowdsourcing? Crowdsourcing is the term used to describe the process of getting work-force from a crowd of people in an online setting. Famous Example: Wikipedia. In 14 years, Wikipedia has become the first reference source with its many volunteer writers and a few specialist editors . Wisdom of Crowd: The quality of crowdsourcing is described as large groups of people are smarter than a few number of elite, no matter how brilliant they are or better at solving problems.
Crowdsourcing for Geospatial Data Using only image processing algorithms is not sufficient to produce generalised damage assessment results. Even if disaster type is same, is it possible to find an image processing algorithm that works for all image types? Rural -building types Urban – building types Industrial -building types As a result, Image Exploration and Damage Assesment Analysis needs «Human in the Loop» systems.
Crowdsourcing for Geospatial Data In todays global world, a disaster is never related to a nation itself. It becomes an international problem and it must be handled in a collaborative manner. Haiti Earthquake, 2010 Japan Tōhoku Earthquake and Tsunami, 2011 Philippines Haiyan Typhoon, 2013 Malaysia Airlines Disappearing Flight 370, 2014 These kind of disasters increased importance of maps geospatial data of the impacted areas geographically-distributed information sharing communities.
HAITI EARTHQUAKE and CROWDSOURCING BİLGİ Satellite: QuickBird Resolution: 0.64m City: Port-Au Prince File Size: 779MB Region: 360km2 SORU We need Rapid Damage Assessment Report After the Earthquake Constraints Time Efficiency Kitle Kaynak Tabanlı Görüntü Analizi Dr. Fatih KAHRAMAN
Kitle Kaynak Tabanlı Görüntü Analizi Dr. Fatih KAHRAMAN CLOSER LOOK Kitle Kaynak Tabanlı Görüntü Analizi Dr. Fatih KAHRAMAN
CROWDSOURCING for HAITI EARTHQUAKE
Task Division Strategies on Crowdsourcing Platform Several different division strategies presented to divide problem into small pieces and distribute these to the crowd combine resolutions from the crowd These strategies are Parallel Iterative Division Cascade Iterative Division Serial (Non-Iterative) Division
Parallel Iterative Division Each small piece in image analysed by at least two people Plain image piece (wout any tags) presented to each user Analysis results combined using statistical evaluation methods and reliability of the user Volunteer-1 Statistical Reliability Analysis Volunteer-2 ... Reliability values assigned to each results Volunteer-n Dr. Fatih KAHRAMAN
Cascade Iterative Division Similar to «Parallel Iterative Division» Analysis done on small image piece before shown to each user User taking labelling done by other user into account during analysis Sequential analysis of image pieces Analysis result determined by final user Volunteer-1 Volunteer-2 Volunteer-3 Volunteer-n
Non-Iterative Division Each small piece of image assigned to a different user Analysis of each piece by only one user Enables us to have analysis result of whole image in a fast manner
AFAD CROWDSOURCING PLATFORM (AYDES-CS PORTAL)
AYDES-CS END USER PERSONEL DASHBOARD (CLIENT)
AYDES CROWDSOURCING PLATFORM (CLIENT ANALYSIS INTERFACE) Dr. Fatih KAHRAMAN
AYDES CROWDSOURCING PLATFORM (ADMIN INTERFACE)
AYDES CROWDSOURCING PLATFORM (ADMIN INTERFACE)
AYDES CROWDSOURCING PLATFORM (SOCIAL MEDIA INTERFACE)