Slashdot Mirror
Help Find Steve Fossett
An anonymous reader invites us to join in the hunt for the missing Steve Fossett using Amazon's Mechanical Turk. DigitalGlobe, one of Google's imaging partners, has acquired new high-resolution satellite imagery of the area where Fossett disappeared on Monday. The public can now go through this imagery and quickly flag any images that might contain Fossett's plane. Flagged images will receive further review by search and rescue experts.
The area of Nevada where he is missing is actually rugged and mountainous ( I have some proerties in those parts myself ). Look on Google earth if you don't believe me, the name 'Nevada' means ',mountains. Also area 51 is now where nearby. There's a lot of rugged and inaccessible terrain he could've gone down, unfortunatley, and 5 days is a long time without water, its dry and hot out this time of year. I'd say the situation doesn't look good at this point, but we can always hope for a miracle, best of luck to the SAR and CAP people.
Yes, Google Earth has updated images of the region. You can tell because they are obviously satellite photos and not overflight. Notice that everything is shot from straight up instead of the normal angle and also note the lack of color.
Read the article, and you'd see that they explain how authorities helped facilitate new images.
38 29' 03.51" N
119 24' 21.64" W
I keep hearing people whine about the Mechanical Turk interface. Each image only requires one mouse click and two keyboard presses (no mouse movement at all required). Here's how:
1 - When you manually accept the first hit, make sure you check "Automatically accept the next HIT".
2 - Press the END key to scroll all the way down to see the image.
3 - Click the mouse on Yes or No.
4 - Press the ENTER key to accept the HIT.
5 - Goto 2
I've found two images that are really good candidates for a crash. One was at 38.020248,-119.368515. It looks like a line of tree damage, with a bright object at the edge of the tree line.
Next, I keep hearing people saying that laypeople aren't useful for something like this. This is simply to flag interesting images so experts can spend their time looking only at the most likely candidates. Also, this is free for them. So they could use an algorithm something like this:
Show each image to at least 5 people.
Each time someone says "Yes" to a specific image, show it to two additional people, up to a max of 20 reviews.
Sort the images by descending Yes vote count and show them to the experts in that order.
Dan East
Better known as 318230.
For a distributed human image recognition project I think classify-galaxies-at-home is more rewarding than "find-Fossett's-corpse" (A bit harsh perhaps, but let's not beat around the bush). At least classifying galaxies you get to see some beautiful galaxies that no-one may ever have seen before, and your time will help scientists look for patterns in galaxy types and test theories about galaxy formation.
// MD_Update(&m,buf,j);
When I started to help this morning there were 32,000 work units (called hits, images to be reviewed) available. They were disappearing at a rate of 5-10,000 per hour, meaning that all things being equal there were 50-100 people looking at them.
:) I'm guessing that although from the side it's mostly blue, that the top of the wings are white.
However over the past half hour the work units available have been *increasing*. Currently 12,000 and increasing. Clearly they are adding more to be done faster than we're doing them. So anyone who helped out at the beginning - don't assume the hits are "all done". There could be more at any time.
In my old version of IE I couldn't see the scale bars or the example image, looking at the same coords of a unique scene in google maps I estimated the image was 125m x 125m - which would be half meter resolution. Now I see they claim the images are actually 85x85m, which would be 1.08ft resolution.
Based on that and that I've done 400 units, that mean's I've searched one full square mile.
It also means the 32,000 units I saw when I started is only 10 miles x 10 miles, 100 square miles. I heard someone else say that they only have 500 square miles of imagery. Looking at Google Earth, assuming the new imagery is the kinda-rectangular patch that is all the same color/brightness - they have approx 1700 square miles. That means there is approximately 600,000 work units in total that need done. If everone does a square mile (shouldn't take more than an hour) then we need 1700 people helping.
But as someone else noted - they're really artificially limiting the search area, considering the range on his plane. Assuming he went certain places or crashed on his way back to the ranch. That doesn't bode well.
PS: It'd be way way more effective if they showed a "image before crash" so that people could self-discover their false positives, without forcing people to download google earth and figure it's before/after out, and/or be smart enough to copy/paste the coords into google-maps satellite view.
PPS: If they were really smart, they'd have a second private pool of the public's false positives being reviewed by amateurs or employees whom they know have much much smaller false positive rates, whom they know are comparing the two available before images (google maps and google earth) against the current images.
BTW: Here are images of the actual specific plane he was flying. http://www.airport-data.com/aircraft/N240R.html (Aviation buffs take pictures and index online everything that flies, apparently