Page 1 of 1
How much error should I expect?
Posted: Tue Oct 30, 2007 10:21 am
by aweinstein
Hi:
Right now I am getting errors, when measuring the distance between two points, between 0.5% and 10%, for a nominal distance of 300mm. The calibration process give me good results (I get a "Great" for the three coordinates). I am using the Point Cloud SDK 1.1.032.
How much error should I expect for the system?
Re: How much error should I expect?
Posted: Tue Oct 30, 2007 5:59 pm
by VincentG
That is about one foot from the camera, which is a little closer than we would normally recommend for tracking. The higer quality tracking results usually start about 2 to 3 feet out from the cameras. The number of cameras that can see a given point will also increase the quality of the calculation, using the "show camera" rays feature can help them determine that.
Re: How much error should I expect?
Posted: Wed Oct 31, 2007 8:45 am
by aweinstein
NaturalPoint - Vincent wrote:That is about one foot from the camera
No. The 300mm is the distance _between_ the points. The distance from the points to the camera is about 1.5 meters.
The number of cameras that can see a given point will also increase the quality of the calculation, using the "show camera" rays feature can help them determine that.
Unfortunately I am getting inconsistent results with the camera rays output (see my other post about it).
How much error should I expect?
Re: How much error should I expect?
Posted: Wed Oct 31, 2007 10:51 am
by VincentG
No more than 3%.
How do you have the cameras layed out?
Re: How much error should I expect?
Posted: Thu Nov 01, 2007 8:29 am
by aweinstein
[quote=NaturalPoint - Vincent]
How do you have the cameras layed out?
[/quote]
Here are a couple of photos of the setup.
http://www.ocam.cl/fotos/optitrack/opti ... etup_1.jpg
http://www.ocam.cl/fotos/optitrack/opti ... etup_2.jpg
The cameras are about 1 meter apart, at 1 meter high, looking downward.
Re: How much error should I expect?
Posted: Wed Nov 07, 2007 9:26 pm
by beckdo
Hi Alejandro,
Something I would suggest trying is to back the cameras out further. Sometimes when you've got them in that tight it can cause problems. Also, when you're only tracking with two rays the error will be considerably higher so watch out for that.
D
Re: How much error should I expect?
Posted: Wed Nov 07, 2007 9:40 pm
by beckdo
Alejandro,
I just realized what you're main issue here is regarding error and camera rays.
Unfortunately, you're going to have a considerable amount of error using the C120s for tracking 3D markers. Your 3%-10% sounds appropriate.
During the development of the point cloud toolkit we had to make a number of improvements to the C120 in order to achieve better results. The resulting camera was our V100 and here are the main things that improve the situation over the C120:
- V100s run at higher resolution (640x480) as compared to 355x288.
- V100s have a very high quality, low distortion lens. The lens is an expensive component in the V100.
- V100s have a hardware based image recognition system for tracking very high marker counts (256) on the camera.
- V100s have a highly improved IR illumination ring to give it a much larger distance over which it can see the camera as well as a very high shutter speed for solid recognition of high speed markers.
The items listed above make the V100s have a typical error approximately 10 times less than the C120s.
It's unfortunate that the C120 cameras are unable to achieve a high level of accuracy when tracking makers. However, they do work with the point cloud library if you are able to overcome the residual error.
See camera rays post for my comments...
Re: How much error should I expect?
Posted: Wed Nov 07, 2007 10:05 pm
by beckdo
This was a recommendation was for the V100s actually, it won't help you with the C120s.
Re: How much error should I expect?
Posted: Thu Nov 08, 2007 8:00 am
by aweinstein
Thanks for your answer.
Right now I am trying with the cameras more far away, and measuring points located about 3 meters from the cameras, and I think that there is a kind of "sweet spot" volume, where I get quite good results. Does it make any sense?
I think it may be related to the fact that the cameras have wide angle lens,(I couldn't find the focal length of the cameras in the specifications, so please confirm that), which may cause some kind of perspective distortion when the points are located too close to the cameras.
Regards,
Alejandro.
Re: How much error should I expect?
Posted: Thu Nov 08, 2007 4:47 pm
by Jim
Hello:
Yes, there can be a sweet spot in the capture volume, that is something we have seen. The lenses on the C120 are good, but do have distortion, they are stock at 2.6mm, which is not that wide for the imager size, about 45 degrees of horizontal FOV.
From the pictures I saw, I would recommend tracking small objects, with points somewhat close together, as if they are far apart, the error can be more as each marker is in a different place in the capture volume
Also, if you track 3 points to create a rigid body, your error will be much less, as you will have 3 points to average for one center point, and 1mm in offset over the total area will not matter as much. It can be misleading to view the absolute accuracy as the distance between only 2 points.