Welcome to the Time Trust Trial contest, Round 4 (T3-4)! This round is an aerial imaging task. Here is your assignment:
Program your UAV to take photos from an altitude of ~400 feet that you then stitch together to make a single image showing an area of 500mx500m (a quarter square kilometer). Somewhere in that image, a Santa (or replica of a Santa, poster of a Santa, just you wearing a Santa hat, etc) must be seen. NO PHOTOSHOP (ie, the sample above would be disqualified)--you actually need to bring something Santa-ish to the field so your UAV can capture it in its shots (yes, I know it will be very small. Just circle the location in your image so we can enlarge and inspect--don't make us do "Where's Wally"!) This is to prove that the shots aren't actually taken from Google Earth ;-)
You can use any path strategy you want: "lawnmower", spiral, concentric circles, etc...
For stitching software, I use the free PTGui, but you're welcome to use whatever software you prefer.
And for your camera, may I suggest you hack up some cool way to trigger the shutter with our cool ServoSwitch?
KML tracks must be provided. Video is not required, but is suggested.
We've now switched to a six-week cycle, so the deadline is 12:00 midnight PST on Sunday, January 17, 2010 now Monday, January 18th due to the Martin Luther King holiday in the US.
The overall winner will be the best quality image, as decided by the judges (based on a combination of resolution, stitch quality and overall coolness--a clever Santa will win you brownie points, and a pretty area is no bad thing [note: snow is lovely, but be warned that it can confuse stitching software]), but everyone who completes the challenge will win a prize.
Enjoy!
Thanks for the info. I was hoping you'd link to something smaller :-) I've been looking for a small enough board that would support an XR5 for use on a hexacopter, but still haven't found anything small enough.
We may in reality be using the single core version, but I'm pretty sure it's dual core. At the time we purchased it, it was the only Atom product available. We could have gone smaller form factor (Nano-ITX etc.), which we used to fly, but they were VIA boards and linux drivers were crap, they were slow pc's (still fast enough for simply taking pictures, embedding telemetry, and streaming the photos back to us) , unreliable at best and they were fairly expensive. We've killed one Atom Mini ITX board in the 1.5 years we've been flying this pc, and that was likely because someone drilled a hole in the board. Now, I would buy a Zotac board instead, Nvidia onboard chipsets are far superior to the intel ones, and the unit is fanless, i.e, more reliable.
Also, they have recently introduced nano-ITX atom boards, which look awesome but they are more expensive ($400 vs. $100) and the hardware is less available, and may or may not be as reliable. I will recommend Intel over VIA (or even AMD for that matter), their stuff just seems to work better.
There are two major orthorectification packages online: the US govt funded Open Source Software Image Map (OSSIM, pronounced "awesome") is a huge toolset for remote sensing and photogrammetry. It has an load of features from simple color correction to the ability of being run on a cluster.
The second package is called Opticks. It's a remote sensing and imagery analysis framework that is unique in many ways, for example it can process video as well as stills. It's more targeted at AP than OSSIM and its interface is easier in that way.
Both are heavily deployed by civilian and military organizations and provide precise results. Both can work with tens of GBs of data at once. Both are open source and available.
I deleted the picktures from the most elevated areas of the terrain and thereby came somehow closer to a result with PTGui. Still, the output can probably best be described as art - based on 154 picktures/ 900MB of data.
We have in-house software with that capability (orthorectification). I'm fairly certain this capability doesn't exist or is limited in it's ability in commercial software implementations.
If anyone knows of a commercial product capable of orthorectification, terrain modeling, image stitching, and GPS coordination, I would love to know!
I have now looked more closely into the picktures and datas and have found that most of the picktures missing were from over the woods, (high trees). In this area the terrain was also elevated up to 50m above takeoff point. I suspect that this, in combination with the fact that I forgot to turn off autofocus caused delay to the trigger-event. The autopilots trigger was active for only 1second, which may have been too short for lots of autofocusing on "moving" trees.
Further on, great difference in terrain elevation is causing deformation when stitching picktures in PTGui. Hence, I will probably need some more heavy-duty software capable of performing digital terrain modelling and orthorectification in order to be able to pull off this one.
brakar, resize those 200+ pics to say 320x240 (i.e. very small, due to memory constrains) and just shove 'em into Hugin or equivalent, it'll show you which ones are missing automagically.
Thanks for your comment Joe. As for the telemetry data, I have the trigger-event log from the autopilot: log.txt.trig1.csv
I did not notice before now that both the kml-trace and the log tells it should have been 230 picktures, not only 201 as I actually got. Most likely I may have used too short time period for the trigger-event, or it may have been too cold for the camera (-15 deg celcisus ground temp). However, since I shot twice as many picktures as I needed, I think I will be fine - only have to figure out which picktures are missing.
Comments
To my knowledge, we are using this board:
http://www.mini-box.com/Intel-D945GCLF2D-Mini-ITX-Motherboard
We may in reality be using the single core version, but I'm pretty sure it's dual core. At the time we purchased it, it was the only Atom product available. We could have gone smaller form factor (Nano-ITX etc.), which we used to fly, but they were VIA boards and linux drivers were crap, they were slow pc's (still fast enough for simply taking pictures, embedding telemetry, and streaming the photos back to us) , unreliable at best and they were fairly expensive. We've killed one Atom Mini ITX board in the 1.5 years we've been flying this pc, and that was likely because someone drilled a hole in the board. Now, I would buy a Zotac board instead, Nvidia onboard chipsets are far superior to the intel ones, and the unit is fanless, i.e, more reliable.
Also, they have recently introduced nano-ITX atom boards, which look awesome but they are more expensive ($400 vs. $100) and the hardware is less available, and may or may not be as reliable. I will recommend Intel over VIA (or even AMD for that matter), their stuff just seems to work better.
There are two major orthorectification packages online: the US govt funded Open Source Software Image Map (OSSIM, pronounced "awesome") is a huge toolset for remote sensing and photogrammetry. It has an load of features from simple color correction to the ability of being run on a cluster.
The second package is called Opticks. It's a remote sensing and imagery analysis framework that is unique in many ways, for example it can process video as well as stills. It's more targeted at AP than OSSIM and its interface is easier in that way.
Both are heavily deployed by civilian and military organizations and provide precise results. Both can work with tens of GBs of data at once. Both are open source and available.
If anyone knows of a commercial product capable of orthorectification, terrain modeling, image stitching, and GPS coordination, I would love to know!
Further on, great difference in terrain elevation is causing deformation when stitching picktures in PTGui. Hence, I will probably need some more heavy-duty software capable of performing digital terrain modelling and orthorectification in order to be able to pull off this one.
I did not notice before now that both the kml-trace and the log tells it should have been 230 picktures, not only 201 as I actually got. Most likely I may have used too short time period for the trigger-event, or it may have been too cold for the camera (-15 deg celcisus ground temp). However, since I shot twice as many picktures as I needed, I think I will be fine - only have to figure out which picktures are missing.
-
6
-
7
-
8
-
9
-
10
of 12 Next