With this in mind I find myself back at my computer in my
home office ready to login to Tomnod to volunteer to search
imagery for another campaign. I have an interest to know how
the Tomnod Platform at DigitalGlobe develops the interface
and gathers the imagery. Luke Barrington explains one way a
campaign gets started is by DigitalGlobe being approached by a
client needing assistance to analyze and solve a problem. At this
point the Tomnod Team will determine if existing DigitalGlobe
imagery can be used or whether one of their five satellites will
need to be tasked to acquire imagery for the area of interest.
This current campaign I have volunteered for is in the
Hawaiian Islands. The Nature Conservancy has approached
DigitalGlobe with a current need to find invasive weeds that
are aggressively spreading through Hawaii’s native forests.
The outcome of this campaign will lead to helping to save
Hawaii’s high-elevation rainforest that remain untouched by
civilization. This is a unique project as the Nature Conservancy
is providing aerial imagery to be used by Tomnod. For this like
other campaigns, the Tomnod Team develops project specific
web interfaces.
When first joining the Hawaii Campaign Challenge you will
see this screen.
The Hawaii Campaign was deployed on June 17, 2014 and
within two weeks there was more than 5000 volunteers on
the Tomnod Platform. The imagery I am provided to review
will also be reviewed by other participants. Barrington states
that the basic idea is to have three people review each pixel
and with this “wisdom from the crowd starts to emerge”. As
you place tags in the imagery for the invasive weeds, the
Australian Tree Fern and the African Tulip Tree (see figures 1
and 2), the tags are recorded by the web interface.
DigitalGlobe’s Tomnod team has developed a suite of
advanced reliability algorithms (CrowdRank) to analyze the
many tags created by volunteers. The algorithms ensure that
tags from the crowd are properly filtered and weighted to
maximize accuracy. The CrowdRank analysis is not just for
where the consensus of tags are located but also how accurate
are the people who are selecting the points. CrowdRank is
able to rank the individual members of the crowd by tracking
the areas of agreement and disagreement. This enables the
system to identify inaccurate and sometime malicious tags.
This in essence will truly leverage the wisdom of the crowd.
For example in thinking about the analysis of the Hawaii
tags, CrowdRank will determine that one person consistently
selects the weed type and never selects a location where a
weed does not exist in the imagery. If this person has created
a login account with Tomnod, then the Tomnod Platform
will remember this person as a reliable volunteer for future
campaigns.
After a campaign is completed, the high confidence locations
designated by CrowdRank will be further reviewed by image
analysts at DigitalGlobe. Or the final CrowdRank analysis is
the delivery requirement by the client who then may conduct
ground truth research.
The results for the Hawaii Campaign are proving how well
an approach first developed to search for Genghis Khan Ruins
is helping environmental conservation efforts. To date there
have been about 390,000 tags of the Australian Tree Fern,
430,000 tags of the Partial Australian Tree Fern and 26,000
tags of the African Tulip Tree. This includes overlap in the
tags from multiple users, which raises the probability that
what multiple people are tagging is accurate.
I hope you get the opportunity to volunteer for a DigitalGlobe
Tomnod Campaign. It truly does provide you with the chance
to look at imagery from around the world and gives you a sense
of helping society solve geospatial related issues.
Jim Peters works in the Geospatial Industry. For ASPRS, he
currently serves as the Chair for the Electronic Communications
Committee.
Figure 1. Australian Tree
Fern.
Figure 2. African Tulip Tree.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
August 2014
719
