Hello forum,
First of all some background...
Having spent a good 2 years playing about with an Skywalker, Ursus and a Canon camera running CHDK, and designing my own uav, which actually flew recently ,I have come to the conclusion that flying wings may just be the best type of platform for short range aerial mapping tasks to create DEM, DSM models for GIS purposes.
The Skywalker was okay (but quite fragile, and now in pieces!), but I think it is too small to fit a roll gimbal on it. And I've found that if pictures are taken on a slight angle, it is very difficult to use software to tile them up nicely. this is my best effort, but far from perfect. The roads on the right do not line up as the camera must have been on a 45 degree angle when the picture was taken. (click for larger resolution) - It was taken using an ixus80 attached to an URSUS Airframe
Realizing that I need to use an airframe that can accommodate a stabilized roll gimbal, i started to design and build my own twin boom design. It took 6 months from initial design, to first test flight. It has a wingspan of 2.4m - AUW of 5kg, and a flight duration of around 30 minutes with 2 x 5000mah packs. However it needs a heck of a lot of room to take off and land, something i don't have in the areas i need to take photos.
Picture of twin boom plane...
What are peoples experiences with using flying wings for this reason? I have the following types of flying wings, and wondering which one is the best to begin with?
- jp-si 48" span Zagi
- Scaled up 60" span jp-si Zagi (hot wire cut by myself)
- x8 2m wing
I have been using the 60" wing to test the APM2 and 3dr radios. And was using the URSUS (now in pieces after a nasty crash!) as a test platform to practice with the Canon CHDK Kit.
Here's a few pictures of the wing and camera....(click images for larger version.)
My question is how straight forward do you think it is to fit a roll stabilized gimbal to any of the above airframes?
Replies
Dense reconstruction complete - obviously the quality is at least 4 times lower than it should be!
VisualSFM is working much better now, it has managed to use almost all the photos from the set, about 296 out of 330. There were quite a few opaque pictures in there, which i didnt remove prior to processing the pictures, so that's about right.... Here is the sparse reconstruction looking from over the sea up the river
I think this is as far as I can go with the limitations of the processing power I have....but i would expect this data is going to produce a very nice textured surface model....
r
Right then, I've been processing the dataset from today's flights at the estuary, I think this set of photos is going to work out quite well, after reducing the photo sizes from 4000x3000 to 1000x750, I am able to work with it given my limitations with internet connection for Photosynth, and computer limitations with the laptop.
Some interesting stats from today...
3 flights covering 0.398 km² (98.338 Acres). 336 photographs.
Here is the raw point cloud generated with Photosynth..
r
Thanks Grant for that info - very interesting
http://wiki.openstreetmap.org/wiki/RTKLIB
Supported GPS Hardware is the uBlox LEA-6T Precision Timing
APM 2.5 uses the uBlox LEA-6H
Hi Rich, I think your journey have answered many questions many of us are looking for. Well done, mate. $3.5k....that's steep. There are some other packages that might come in under that...not free though. Is that a X5 on your catapult?
Another question - have you referenced the model with ground control points? In other words, can the point cloud be used for design and quantity purposes.
Good work
Antonie
I've been processing the images from my last flight, and have had some good results from my trial of Photoscan, but not so good from Visual SFM,
for visual SFM, I am using VLFeats SIFT program with CPU detection, and also I had to half the image sizes from 4000x3000, to 2000x1500. If I was able to use the default SIFT in Visual SFM, and full size images, I may get better results...Anyway, from 69 photographs, Visual SFM made 4 models, the best one utilising about 26 images.
Here is the dense CMVS reconstruction....quite good, but the reconstruction covers less than half the area photographed. I would also like to use the photo data to fill in the gaps to produce a nice textured model, but i'm not sure how to do that.
Photoscan on the other hand, produced a model utilisiting all 69 images. skipping none. I gave photoscan the same pictures to use. The 2000px downsizes ones.
I was also able to produce a photomosaic from the pictures... downsized to 1500px for web...
So, looks like Photoscan is the software i'll need to buy...@$3500, i may need to put this project on hold for a while!!
I'd like ot be able to achive the same results with VisualSFM, but i believe that cannot be used for commercial use anyway....
good good...
Okay, Just got back from the field and did two flights, also made the tracks a bit closer together to get more overlap, also used 1/1500 shutter speed, as opposed to 1/1000. The photos look crisper.
I am not able to put the better Canon camera (sx230) into the plane yet as i don't think the 9 x 4 prop will take it, so ordered a 9 x 6 for a bit more power on climbout, It struggles gain height nicely already, that could be because i don't have enough up movement on the elevons though.
I did use a slightly better camera today, an ixus120, It takes slightly higher resolution photos, 4000x3000. The pictures look good, and fingers crossed i will have more overlap, here is the waypoints....
I'll run a batch downsize of the files, so they are 1000px wide, and have a play over the weekend with VisualSFM to see if i can get a decent point cloud of the area. I think i covered a good 500m2.
The plane did go into FBW-A mode for some reason when it was quite a distance away, even though the throttle fail-safe is enabled and is supposed to go into circle, then rth. weird.
If I don't have any luck with the pics, maybe i'll take you up on your offer of processing them for me Brenden! I am looking at speccing up a PC, but that's probably going to be a little while!
Cheers, R
Ah yes, now I remember why I didn't use manual focus/infinity...I knew I would forget to set it sometimes on start up.
And you removed all the other features that Luca mentioned? That should help.
Another thing that should be noted is the lens IXUS 38mm lens vs the sx230 28mm. The wider lens will help with coverage.
One of the things I really like about the sx260 is its 25mm lens, which offers great coverage at lower altitudes.
I'll have to time it but with the intervalometer I'm using I'm getting sub 1 second intervals with the sx260 when set to 0 delay.
After reading that processing scripts also slows down the process, I stripped the script I use down to bare minimum.
You can find the very simple set of scripts I use here: http://flightriot.com/remote-sensing/chdk/
It does require a USB trigger though. I use this USB trigger method: https://www.youtube.com/watch?v=i4YSind9dXw
Once the class 10 is in there, should I be expecting a photo every 1 second? I've set the cera to tv priority mode, but what's stopping it from taking blurred photos? Surely I should be setting the shutter speed to 1/1600 minimum?
-
5
-
6
-
7
-
8
-
9
of 14 Next