Improving calibration results

mores_p
Posts: 23
Joined: Sun Oct 28, 2007 11:23 pm
Location: Lafayette, LA, USA.

Improving calibration results

Post by mores_p » Wed Oct 08, 2008 2:21 pm

Hello,

I have used the 3-camera rigid body tracking system for a while and I am at the point where I am trying to improve the tracking accuracy.

The fact is I never get calibration result (in Point Cloud Calibration Tool) better than "Fair". Are there any tricks to getting a better result? What factor affect it most: camera placment, light condition, etc? Any recommendation given we want to stick to 3 cameras to save cost of the system.

Assuming X-axis points to your right, Y points forward from you, and Z points up, (right-handed system) my tracking space is about 12"x18"x25" in X-Y-Z. Any recommendation for this specific setup?

Thank you,
Mark

mores_p
Posts: 23
Joined: Sun Oct 28, 2007 11:23 pm
Location: Lafayette, LA, USA.

Re: Improving calibration results

Post by mores_p » Wed Oct 08, 2008 2:43 pm

I use Flex:V100 cameras. Environment includes both natural and fluorescent light.

VincentG
Posts: 7728
Joined: Mon Jul 17, 2006 5:00 am
Location: Corvallis, Oregon

Re: Improving calibration results

Post by VincentG » Wed Oct 08, 2008 2:46 pm

Focusing on the wording (fair, good, great) is not ideal. The ideal way is to cover as much of the capture area as possible, without blocking the camera views. Also try placing the camera at different heights, so that there is a greater difference in angle views, between the cameras.

mores_p
Posts: 23
Joined: Sun Oct 28, 2007 11:23 pm
Location: Lafayette, LA, USA.

Re: Improving calibration results

Post by mores_p » Thu Oct 09, 2008 2:55 pm

Hello,

Below is a visualization of my error measurements for our 3-camera OptiTrack system. I lay down a reference grid on a floor, placed marker at each point on the grid, and had OptiTrack reported the measurements. White grid below is an ideal case when tracking is perfect, green grid is what I came up with using OptiTrack. (each grid square is 1"x1"). It's easy to see from the picture that there is a discrepancy, and this is what I am trying to correct right now. (by ways of camera placement and calibration.)

From top-down view, all 3 cameras were beyond the top edge of the picture. The center camera was oriented to look straight along the center (vertical) line of the grid. The one to left, and the one to the right were angled slightly to look at about the same point as the center camera.

From this camera configuration and error measurement, do you have any recommendation as how to correct the distortion?

Mark




Image

VincentG
Posts: 7728
Joined: Mon Jul 17, 2006 5:00 am
Location: Corvallis, Oregon

Re: Improving calibration results

Post by VincentG » Thu Oct 09, 2008 3:42 pm

It this setup - how far away are the cameras from the top/sides of the grid?

Would you say the outer cameras are at the upper left and right corners of the grid?

Also, to improve the accuracy of the tracking, be sure to wand over the entire capture volume (your grid) as evenly as possible during calibration.

mores_p
Posts: 23
Joined: Sun Oct 28, 2007 11:23 pm
Location: Lafayette, LA, USA.

Re: Improving calibration results

Post by mores_p » Fri Oct 10, 2008 12:13 pm

The center camera was about 24" from the top side of the grid.

The outer cameras were a little bit closer, about 2"- 3" closer. They were angled a bit to have overlapping capture volumes.

All cameras were at about 13" from the floor.

I tried out different camera configurations and some gave better results than others. One time, I got half of the grid at very good correspondance, while the other half had noticeable distortion. I will continue to experiment and post interesting results here.

I have a question about wand waving. I chose "1-meter radius" for the capture volume. When waving, should I focus to cover only my tracking space, eg. my grid space, or I should try to cover the whole 1-meter radius capture volume?

Mark

VincentG
Posts: 7728
Joined: Mon Jul 17, 2006 5:00 am
Location: Corvallis, Oregon

Re: Improving calibration results

Post by VincentG » Fri Oct 10, 2008 12:58 pm

How far apart are the cameras, from themselves?

You want to try and capture as much of the capture volume as possible

yoshi
Posts: 174
Joined: Sun Jul 24, 2005 5:00 am
Location: silicon valley

Re: Improving calibration results

Post by yoshi » Fri Oct 10, 2008 1:27 pm

Having cameras aiming from different directions generally improve calibration. If the cameras were placed close together and aiming toward same direction, images seen by the cameras will be similar and calibration will have to rely on subtle differences in those images to calculate parameters.

mores_p
Posts: 23
Joined: Sun Oct 28, 2007 11:23 pm
Location: Lafayette, LA, USA.

Re: Improving calibration results

Post by mores_p » Fri Oct 10, 2008 2:13 pm

Cameras are about 20" apart.

When you said "Capture Volume", did you mean a specific 3D-space (for a particular application) in which we are interested in tracking a rigid body? For example, if my application is interested in tracking a rigid body within 12"x12"x12" cube, is the cube a capture volume?

I already did as Yoshi suggested. I also tried having different heights for different cameras. One thing I am not sure is where in the tracking space to point each camera to? Like, if my tracking space is 12"x12"x12" cube, where in the cube should each camera look at?

VincentG
Posts: 7728
Joined: Mon Jul 17, 2006 5:00 am
Location: Corvallis, Oregon

Re: Improving calibration results

Post by VincentG » Fri Oct 10, 2008 3:27 pm

You want to point the cameras at the center point of the viewing angles of all three cameras -

the captture volume is the view overlay for all three cameras - and will vary, as the camera setup varies

Post Reply