$40,000 prize for algorithm to do change detection on aerial imagery

This is a really important problem to solve; hope someone does! From Popular Mechanics:

An engineering firm wants to make sense of the abundance of aerial photos we now shoot from planes, satellites, and drones. Draper, out of Massachusetts, has put a $40,000 bounty on an algorithm that can accurately sort and sequence images taken from miles in the sky, with implications, perhaps, to track wildlife, observe erosion, direct traffic, or notice climate change in real time.

The contest, which runs until June 27, requires the algorithms to take a series of images taken of California and correctly sequence them in the order they occured. The hope is that subtle changes can then be noticed (simple enough) and also understood (much trickier).

It might be years before we could understand all the data that correctly sequenced images can convey, but global security operations have been doing similar things for the past 80 years. As the world swims in ever more photos, we could stand to gain a lot by being able to automatically interpret them. The right algorithm could have wide-ranging effects for scientists, corporations, governments, and hobbyists alike.

Views: 2238

Comment by Bjorn Nellerson on May 5, 2016 at 7:42pm

Give me a cluster of 1000 GPUs and 10000 datasets and it'll be done ;P

Comment by Curt Olson on May 5, 2016 at 9:08pm

I'm probably missing something obvious, but all my cameras know the time and write this into the exif meta data.  Then all I have to do is sort by timestamp.  Who do I send my mailing address to, and that's Curt with a "C", and Ols "o" n.

Comment by Bjorn Nellerson on May 5, 2016 at 9:44pm

@Curt Olson Not sure if you're serious, but you won't get access to that data I presume.

Comment by Chet on May 5, 2016 at 9:56pm

See ERDAS Imagine . . . 

Comment by benbojangles on May 5, 2016 at 10:16pm

Time & GPS, simples.

Comment by Ben on May 5, 2016 at 11:09pm

Notice the third image has shadows at a different angle, that's probably part of the complexity.

Comment by Gary Mortimer on May 5, 2016 at 11:26pm

You could almost make some sort of App store for this sort of thing

This is called pattern of life BTW and is what the military does.
Comment by Jonas Bohlin on May 6, 2016 at 3:43am

Ok, so now I have read the all the instructions for the contest on Kraggle, and the contest is about time-sorting the images. As I have been working with satellite and aerial images from many different sensors and also with time series and change detection for more than 10 years, I must say that I think that it is highly unlikely that the images from these small satellites should come without a time stamp. Why would the images have a geo tag (so they know what where/what they are covering) but no time tag? Or don't they even have other meta data like trajectory or orientation, which is even more unlikely, making them completely useless. 

Also the images in this competition is much higher resolution than the resolution of the satellites, which have about 10m resolution, which require different algorithms (e.g. cars would not be possible to detect).

So this challenge is about something else. Where would one find large datasets of non-tagged, location or time, high resolution aerial (or not) images...? 

Comment by John Arne Birkeland on May 6, 2016 at 8:57am

I agree, this makes little sense. Normally you look at the time-stamp and see what has changed. But they want you to look at visual changes and find the time stamp. In short they are turning an easy task backwards to make it hard.


You need to be a member of DIY Drones to add comments!

Join DIY Drones


Season Two of the Trust Time Trial (T3) Contest 
A list of all T3 contests is here. The current round, the Vertical Horizontal one, is here

© 2019   Created by Chris Anderson.   Powered by

Badges  |  Report an Issue  |  Terms of Service