Camera distortion during calibration

[bencrossman]

When I perform a 3 point wand calibration with 4xV120:SLIM cameras and 3.5mm wide lenses I get weird shapes.
Using the default settings I get:


If I change the lense from auto to 3.5mm I get better results:



but still looks wrong. Should all cameras distortion look the same?

[Seth Steiling]

When the lens distortion grid gets super distorted, like in the first picture, it's likely because the calculation got off track early and should just be reset. This happens on occasion, when the calibration is set to auto-detect the lens type. Seeding the lens type in advance, like you did, will usually prevent this from happening at all.

The second image could be a matter of the calibration just needing more time to crunch. If you're only a minute or two in, and the distortion grids look a bit different still, that's probably OK. All calibrations start off with lens distortion grids looking a bit different, before it's converged on a solution. Just let it go longer and see if it improves over time. If you're 10 minutes in, and it still looks like this, it's probably an indication that something needs to be changed.

Possibilities:

- Insufficient or uneven wanding. Make sure that you have enough samples to cover as much of the 2D view as possible.

- Insufficient overlap. With only 4 cameras, it's possible that you don't have enough overlap between cameras to grab enough wand samples (as the wand sample will be ignored if it's not seen by several cameras at once). That could have the same effect as above.

- Calibration quality setting too low. It's a good idea to step up your quality setting as the calibration calculates. Starting at Medium will usually jump start the calibration, giving it a more loose set of solve parameters to get its bearings with. Then, as the results move toward "Excellent" and "Exceptional," you can bump it up to High, and then Very High. Keeping it at Low or Medium will potentially allow the calibration to retain relatively higher distortion error--in essence being satisfied with a result like what you have above, in the second picture.

[Seth Steiling]

I just noticed that you have V120:SLIM cameras in your setup. Are you using a custom LED wand, or are you illuminating active markers through some other method? If the latter, what are you using for your strobe?

[bencrossman]

Thanks for the info.

We are using a custom wand with reflective markers and lighting from each camera using these:
http://www.amazon.com/Compact-Infrared- ... B002OU3IC6

[Seth Steiling]

Hi Ben,

While any of the potential calibration issues I mentioned above might also be an issue, I'm much more concerned about the camera setup. There are several possible weak links...

1) The V120:SLIM really isn't designed to track passive markers. For that application, we'd strongly recommend one of our motion capture cameras, which include illumination that is optimized for the camera and lens. You can use a SLIM and supplement it with an external illumination source, but there can be problems with this:

- Our markers are retro-reflective, so they focus the reflected light in a relatively tight pattern (+/- 1.5%) back to the camera. This means that the illumination source needs to be as close to the lens as possible to stay within that tolerance angle. Ideally, you want to position the LEDs in a ring around the lens, like we do with our mocap cameras.

- In looking at your IR emitter, I'm guessing you have it positioned right next to each camera. At longer ranges, this can be OK (though definitely not ideal), because it will probably still be within the tolerance angle. But the closer the marker gets to the camera, the more likely the angle of reflection becomes too large, giving your a poorly illuminated marker.

- Also--the FOV, positioning, and aiming of the LEDs on our mocap cameras are tuned to match the FOV of the lens, ensuring that the camera view is sufficiently illuminted all the way to the edge of the image. With a setup like yours, this is nearly impossible to achieve, resulting in further degradation of marker illumination. When a marker moves anywhere outside of the lens center, the illumination will be even worse.

- In reading the specs and reviews for your particular IR emitter, it sounds like the effective range and the beam width are both too small, likely further compounding both the above issues.

2) The 3.5mm lens is a tad more prone to distortion, blurring, and lower marker illumination than the default 4.5mm. It's still a great, functional lens when used to track passive markers with one of our mocap cameras, but it does potentially compound all of the above issues when used with an external illumination setup.

3) While Tracking Tools does support custom calibration wands, it's not ideal and not really recommended unless critical to a unique application. Custom wands can be more prone to calibration error, if the dimensions are off even by a little. This, too, could compound all of the issues above. Is there a reason you're using your own wand instead of an official OT wand?

To summarize all of that--it's likely that you're having calibration and tracking issues because your markers aren't sufficiently and symmetrically illuminated, and potentially because your custom wand is introducing error into the calibration.

My recommendation would be to use V100:R2 cameras, at a minimum, with one of our official calibration wands.

- Seth

[bencrossman]

Hi Seth

Thanks for the info. Ideally we would have gone the V100:R2 but at the time it seemed to be better and cheaper to go the SLIM's. This was based off the table that said:
SLIM 120FPS 8.3ms
V100:R2 100FPS 10ms
and a cost which meant you were paying $200 for a nice case with some IR leds. We also really wanted the best latency we could get.
We went the wide lens because at the time I wasnt quite sure how the tracking worked (thought it was similar to a PS3 Move tracking instead of trigonometry) and thought if we went two wide angled lens we could cover the area. This was a mistake and we should have gone the standard lens.

We have our custom LEDs just above the SLIM cameras. It seems to produce ok results and have the range but yes it would have been easier to get the V100:R2's

We went with a custom calibration wand because I just couldnt bring myself to spend $250 on a black stick with three markers on. Dont want to sound like a cheap skate but this does seem a little over priced. Same with the calibration square.

The setup we are trying to create needs to have a low cost point.
I will discuss with my management and see if we can sell the 4 SLIMs we have and purchase some R2's and a wand.

Cheers
Ben

[DonBurroni]

I get this too, with V100:R2 even when the camera lens is set to the correct value, in the calibration set-up.
Ive stoppod using Tracking Tools for calib now, and use it just to aim the cameras the interface is better, as you can zoom in on cameras.

[Seth Steiling]

[quote=DonBurroni]I get this too, with V100:R2 even when the camera lens is set to the correct value, in the calibration set-up.
Ive stoppod using Tracking Tools for calib now, and use it just to aim the cameras the interface is better, as you can zoom in on cameras.
[/quote]
It will occasionally happen regardless of software, camera, or methodology as there is some randomness involved in what samples are taken and how they set the trajectory for the calibration estimation process. When it does happen, changing the quality setting up can usually reset the calibration enough to fix it. If not, a simple rewanding should suffice.

I wouldn't recommend abandoning TT for calibration, as it has a fuller implementation of what is essentially an identical calibration module (i.e. the guts are the same, but TTs UI is more verbose). So perceived discrepancies in the calibration output over time are coincidental--it's the same math whether you're in Arena or Tracking Tools.

[Seth Steiling]

[quote=Ben_Crossman]Hi Seth

Thanks for the info. Ideally we would have gone the V100:R2 but at the time it seemed to be better and cheaper to go the SLIM's. This was based off the table that said:
SLIM 120FPS 8.3ms
V100:R2 100FPS 10ms
and a cost which meant you were paying $200 for a nice case with some IR leds. We also really wanted the best latency we could get.
We went the wide lens because at the time I wasnt quite sure how the tracking worked (thought it was similar to a PS3 Move tracking instead of trigonometry) and thought if we went two wide angled lens we could cover the area. This was a mistake and we should have gone the standard lens.

We have our custom LEDs just above the SLIM cameras. It seems to produce ok results and have the range but yes it would have been easier to get the V100:R2's

We went with a custom calibration wand because I just couldnt bring myself to spend $250 on a black stick with three markers on. Dont want to sound like a cheap skate but this does seem a little over priced. Same with the calibration square.

The setup we are trying to create needs to have a low cost point.
I will discuss with my management and see if we can sell the 4 SLIMs we have and purchase some R2's and a wand.

Cheers
Ben
[/quote]
I understand, Ben.

As long as you are:

1) Not planning on scaling your head tracking data (so, 1:1 only),
2) Only tracking those glasses, and
3) (Nearly) always facing the same wall

You might consider our V120:Trio. I think it could perform well for you (not on par with 4-6 R2 cameras, but far better than what you're currently working with). It comes pre-calibrated from our factory, so you don't even need to worry about the calibration component. Plus, it's been used in a host of prominent CAVE and other VR/AR installations.

http://www.naturalpoint.com/optitrack/p ... depth.html

It even comes with a free, perpetual license for Tracking Tools (usually $799).

[Seth Steiling]

[quote=Ben_Crossman]This was a mistake and we should have gone the standard lens.[/quote]
Not necessarily.

If you do swap to R2s, I think that the 3.5mm lens might still be a good option for you. You'll probably need to change some of the default 2D filtering settings (like loosening up the circle filter, and maybe changing the imaging mode) to prevent the data from being filtered out (as the wider FOV doesn't deliver quite as precise data). We can help you with this if need be. When used in a more... standard... camera setup environment, it should perform well.