[Board] Google Geo Challenge

[Frank Warmerdam]

Folks,

I received the following email from an associate.  The general concept
seems good, and my question to the board is whether it is felt that it
would make sense for OSGeo to try and assemble a project to pursue
funding from Google to accomplish something like this.  The timeline is
short.  I would likely be interested in involvement in such a project
but I'm not sure I would want to lead it.

"""
   I've sent this around to a couple of prominent people I know in the 
geospatial community, to see if anyone is interested, or knows of anyone 
interested in working on a project like this. Feel free to forward it to anyone 
you care to, or ignore it entirely. I'm happy to answer questions from anyone 
about it. As I said below, I have no personal interest in it, other than I 
think the geospatial world would be a better place if this project were to come 
to fruition.

   http://www.google.org/geochallenge.html

I'm sure you've heard of it. There's an idea I've been wishing someone would 
pick up and run with, but I don't have the time and resources to pursue. 
Perhaps with the Google Geo Challenge Grants, someone else could make it 
happen. I'm wondering if it might be a cause you'd be interested in championing 
and finding if someone else is interested in?

Let me be clear, I don't want any part of the project. I don't want to work on 
it, and I don't even (personally) have any need for it. but, I realize that it 
could benefit many geo-practitioners, and a variety of other groups (especially 
search and rescue operations). And, I would like to see them have this 
capability. I think Google too, and all web mapping products and services could 
benefit.


   Here' the gist of it:

Orthophotos are difficult to produce rapidly. Aerial platforms capable of 
acquiring digital airphotos quickly are becoming more common and inexpensive 
(even hobby UAVs with digital cameras). During a time-sensitive event like a 
wildland fire, other disaster, or search and rescue operation, it could be very 
valuable to acquire up-to-the-moment ortho imagery. Many times, aircraft will 
already be flying, but they are up there for other purposes and not dedicated 
orthophotography platforms. The challenge is to take less-than-perfect 
(probably digital) imagery and turn it into the best orthoimagery possible 
given the constraints. And do it quickly and economically. Even not-great 
imagery is better than no imagery if you're looking for a lost hiker.

I believe this can be done through a combination of existing technologies and 
web 2.0 crowdsourcing.

First, the air vehicle needs to record the best images it can, and record any 
metadata at the same time. This means a GPS at a minimum. Some news helicopters 
have powerful stabilized zoom lenses -- it would be great if something like Red 
Hen's video metadata storage could be used to record camera aim and zoom too. 
But this is outside the scope of the project -- we just want to develop the 
software and process.

Imagery needs to be transmitted to the ground at some point. Worst-case, this 
would be done when the aircraft landed. Best case, HD video from a news helo, 
or digital images from a UAV or the like could be downloaded in-flight via 
radio. Once the images and metadata were on the ground, the rectification 
process would be done.

First, the basic GPS position (and aim/zoom) would be used to roughly place the 
image geospatially. Red Hen already does this. At this point, either or both of 
the methods below could be used.

   #1. Simple crowdsourcing. A simple AJAX-y web based system would allow users 
(local or online) to specify tie points and compute the image rectification. 
Reference map layers from all major online maps (Google, MS, WMS sources, etc) 
could be drawn into the UI to allow users to find matching locations in the 
already-rectified map and in the new imagery. Also, the user should be able to 
define regions of the new image with varying "quality". So, if part of the 
image is obscured, poorly focused, etc, it can be marked, and in the final 
mosaic, if that area is available from multiple images, preference can be
given to sources marked with higher quality. Some groupware features are
needed to make notes about problems found or rationales for what a given 
operator did. The original image and tiepoints will be stored in case a problem 
is found later and a different operator wants to go back and revise the 
rectification.

   #2. Automated tiepoint matching. This would be a core program, usable either 
as a batch process, or as a user-driven tool for #1, (above) to try to 
automatically match features between an existing (old) georeferenced image and 
a current non-georeferenced image. I believe David Lowe's SIFT algorithm:

http://en.wikipedia.org/wiki/Scale-invariant_feature_transform
http://www.cs.ubc.ca/~lowe/keypoints/

would perform this task well. It is patented, but I hope permission could be 
obtained to use it in this non-profit fashion.

SIFT would allow an operator to batch-add a large number of images to the 
system and get initial quick registration of them. images that initially failed 
to register could be queued for manual registration (#1 above). Even after 
initial registration, operators could revisit each batch-registered image and 
tune the registration if needed.


The final output would be a mosaic of up-to-date images. These could be viewed 
in a local command environment on a large display. They could also be available 
on a web-mapping system so that a crowdsourced group of people could examine 
each one. Similar groupware features would allow the crowd to make notes: "Red 
splotch in lower right corner has been checked and found to be rusty car, not 
red parka of missing hiker". Or, "unknown blue object spotted in upper right 
corner yesterday, has moved 500ft in today's photos, may be a mobile human." 
The crowd could have the option to queue anything they found for further 
examination by the Search and Rescue team, or revisit by the aircraft for 
better imaging.

A realtime map could be created showing all areas that have been imaged, which 
areas have been imaged with good quality (versus areas imaged, but with poor 
lighting, angle, etc), which areas have been thoroughly examined by photo 
searchers and which areas have been examined by ground searchers. this would 
greatly aid in ensuring that all areas were covered adequately and no important 
areas were overlooked.


This is sort of an outgrowth of the search processes used to look for Steve 
Fossett:

  http://en.wikipedia.org/wiki/Steve_Fossett#Disappearance_and_search
and Jim Gray:
  http://en.wikipedia.org/wiki/James_N._Gray#Disappearance_at_sea_and_search

I am not myself in a position to lead a project like this. I'm not a 
non-profit, and I don't think I can be anytime soon. I'm wondering if you know 
of any open source geospatial people who would be interested in looking at a 
project like this, and picking it up and running with it? I don't really want 
to be involved, in practice or name. i think someone else would do a much 
better job of leading and implementing this technology. I simply want
to see it happen so that everyone can benefit from it in the future, and I 
think the opportunity to have Google help fund it would be a very good one. 
Google's deadline for submissions is December 22nd, 2008, so a proposal would 
have to come together fairly quickly. I think it's exactly the sort of project 
Google would fund through their "while development of tools that enable many 
organizations to create maps might receive US$100,000." criteria.

"""

Best regards,
-- 
---------------------------------------+--------------------------------------
I set the clouds in motion - turn up   | Frank Warmerdam, warmerdam at pobox.com
light and sound - activate the windows | http://pobox.com/~warmerdam
and watch the world go round - Rush    | Geospatial Programmer for Rent