Posted by Errol Tazbaz on January 18, 2011 at 9:04am
I am a senior year mechanical engineering student and am working on a school project for mapping and geo-locating. I have read through quite a few posts and it seems like most people use either the copilot or BTA as a wing leveler. I was wondering if I could use one of these systems to keep a camera pointing straight down. The platform for the camera would be an actual Cessna 172. I know you guys are UAV guys but it seems like you have amassed quite a bit of knowledge and was wondering if you could help shed some light on my little project? Weight is not as big an issue because of the size of the platform and using redundant systems as backup would be completely acceptable. The monkey wrench in this whole thing is the system must work in mountainous areas and also over forest fires. Would I be able to still correctly sense the horizon in this kind of terrain or would I need something else? Thanks in advance.Errol
You need to be a member of diydrones to add comments!
You can cheat the resolution gradient by overdoing it at the photo and downscaling - ie, take the photos at, say, 300DPI equivalent, and then reduce to 280 or 250DPI for the finished image, essentially bringing the whole image down to the resolution of the lo-res corner of the trapeziodal images. Obviously, go, the further off ortho you go, the lower the resolution of the whole image, which puts you back to needing a very good mount, but it gives you a little leeway, at least.
The problem with orthorectification software is that there is a resolution gradient because of the trapezoidal image. I was thinking whether there was a way to recognize the non-ortho images and discard them.
Thanks for all of the great comments! This is a pretty cool community made up of some really smart people.
I have found a stepper motor that can achieve 1600 steps with 1/8 micro steps from Anaheim Automation for a couple hundred bucks each. We would need 4x as many steps which I think we can accomplish using a 4:1 gear reduction. They claim the servo is good for 3000 rpm and all we need is 180rpm So the gear reduction shouldn't slow the stepper motor down to a point where we aren't meeting our slew rate objective. Also we are only doing a +/- 45 degree sweep for roll and pitch. The software takes care of rotating (pan) the images to create the overlays for the mosaic. With all of that being said it looks like we will need to use a pretty decent IMU and see how closely we can meet the specs.
The problem I have seen with all of the videos of various systems is that the tracking is too slow and not accurate enough (PID settings?). Maybe there is another way to do it in software where we take a ton of images and throw out all of the non-ortho ones. We would have to figure out how the software could tell the non-ortho from the ortho ones.
There is a much more detailed maths in this document about the Direction Cosine Matrix (DCM). The DCM is a mathematical model of the orientation of the plane relative to the earth. The DCM is a rotational matrix which makes it simple to keep a servo aligned with the horizon or aligned with straight down, while it is mounted on the aircraft.
And finally, the code that does servo stabilization (for a camera) in MatrixPilot is here.
The concept works, maths and implementation works. The issue is accuracy. Will the IMU and the accuracy of the Servos be sufficient for your application within the context of an aircraft which experiences high G forces while flying.
Hm I'm working in LIDAR and photogrammetry business and I can imagine why would you like to have a 1mrad precision of rotation. In reality we are not really looking forward to keep camera looking in nadir because it is probably more expensive than using best civil grade IMU. Using this kind of IMU (approx. 200k€) and post processed GPS trajectories allows us to get orientation parameters of platform in 0.003 - 0.007 deg with 400Hz sampling. In theory of course. Practically that means that we can achieve values that are 10-times larger. But we can still produce an orthophoto without aerotriangulation or geocode LIDAR measurments with accuracy better than 5 cm.
But if you have some photogrammetry knowledge and adequate software, you can use far less capable and cheaper IMUs.
You'll have trouble with any gravity-based gimballing systems - the problem with flight is that your resultant acceleration is often different from the local gravity direction. A notable example of this is with spiral divergence mode - for those who don't know, it's a potentially dangerous mode of flight where the aircraft diverges from a bank into a downward spiral. Its dangerous because the resultant acceleration on the aircraft tends to stay pretty close to the aircraft's up/down axis, so that even when the aircraft is banked over to almost 90 degrees relative to the horizon, if you close your eyes it feels like your wings are level and you're just pulling up hard (very hard - in a 172 you'll be pulling close to 2g by the time you get to 90 degrees). Planes have crashed because they went into cloud and started spiralling, and the pilots thought they were climbing when they were actually spiralling toward the ground.
In a mode such as this, any gravity-based gimballing system will fail - in a developed spiral you could end up with the camera looking at the horizon. Obviously, that's an extreme example, but it shows that it can't be relied on. Even small perturbations will cause deviation, and you'll have a tricky task damping the gimbal so that it moves quickly enough to keep the error down, but slowly enough that turbulence and course adjustments don't cause a wobble. The 172 is a small aircraft so it won't have a great deal of inherent damping from it's own mass and wing area.
Your best bet is probably a setup similar to the APM in terms of hardware - use the sensors to figure out which way is down as normal, but change the servo control code to keep a gimballed camera pointing downward.
Ground Engineering Ltd Home Page
Ground Engineering Ltd providing Land Surveying, Topographical Surveying services all over India.
Replies
I have found a stepper motor that can achieve 1600 steps with 1/8 micro steps from Anaheim Automation for a couple hundred bucks each. We would need 4x as many steps which I think we can accomplish using a 4:1 gear reduction. They claim the servo is good for 3000 rpm and all we need is 180rpm So the gear reduction shouldn't slow the stepper motor down to a point where we aren't meeting our slew rate objective. Also we are only doing a +/- 45 degree sweep for roll and pitch. The software takes care of rotating (pan) the images to create the overlays for the mosaic. With all of that being said it looks like we will need to use a pretty decent IMU and see how closely we can meet the specs.
The problem I have seen with all of the videos of various systems is that the tracking is too slow and not accurate enough (PID settings?). Maybe there is another way to do it in software where we take a ton of images and throw out all of the non-ortho ones. We would have to figure out how the software could tell the non-ortho from the ortho ones.
Errol,
You might want to take a look at the following:
A test video of stabilizing a camera with the UAV DevBoard.
An in flight video of camera targetting.
A description of some of what is involved is shown in this docment aboutUAV DevBoard coordinates.
There is a much more detailed maths in this document about the Direction Cosine Matrix (DCM). The DCM is a mathematical model of the orientation of the plane relative to the earth. The DCM is a rotational matrix which makes it simple to keep a servo aligned with the horizon or aligned with straight down, while it is mounted on the aircraft.
And finally, the code that does servo stabilization (for a camera) in MatrixPilot is here.
The concept works, maths and implementation works. The issue is accuracy. Will the IMU and the accuracy of the Servos be sufficient for your application within the context of an aircraft which experiences high G forces while flying.
Best wishes, Pete
Hm I'm working in LIDAR and photogrammetry business and I can imagine why would you like to have a 1mrad precision of rotation. In reality we are not really looking forward to keep camera looking in nadir because it is probably more expensive than using best civil grade IMU. Using this kind of IMU (approx. 200k€) and post processed GPS trajectories allows us to get orientation parameters of platform in 0.003 - 0.007 deg with 400Hz sampling. In theory of course. Practically that means that we can achieve values that are 10-times larger. But we can still produce an orthophoto without aerotriangulation or geocode LIDAR measurments with accuracy better than 5 cm.
But if you have some photogrammetry knowledge and adequate software, you can use far less capable and cheaper IMUs.
In a mode such as this, any gravity-based gimballing system will fail - in a developed spiral you could end up with the camera looking at the horizon. Obviously, that's an extreme example, but it shows that it can't be relied on. Even small perturbations will cause deviation, and you'll have a tricky task damping the gimbal so that it moves quickly enough to keep the error down, but slowly enough that turbulence and course adjustments don't cause a wobble. The 172 is a small aircraft so it won't have a great deal of inherent damping from it's own mass and wing area.
Your best bet is probably a setup similar to the APM in terms of hardware - use the sensors to figure out which way is down as normal, but change the servo control code to keep a gimballed camera pointing downward.
Would something like this work for your application?
Regards
Martin
www.buildyourowndrone.co.uk