Improving calibration results

[mores_p]

I get a lot better result this time. I think the main reason is how I wave the wand.

Previously, I tried to cover the whole volume created by the intersection of each camera's viewing volume. This time, I focus only on my 3D-grid space, the 3D-space (cube) in which tracking will occur in my application. We don't care much about tracking outside this space.

Is this the way I should wave my wand? Focusing only on the space where tracking will occur in the application? or Should I try to cover the whole space created by the intersection of each camera's viewing volume. I am asking to make sure that I do not just get lucky this time.

Also, do you have any suggestion if I want to improve the error further? What is the best accuracy (in terms of error) I should expect to achieve from this 3-camera setup?

[dashesy]

I have some questions that come in mind when reading this thread.

I am doing an assessment on the available options to track a 4"x6" object in a volume as small as 20"x20"x20" or 40"x40"x40" in realtime.

1- Is this an unusual setup to track (1 or more) rigid bodies?
I mean, I guess most of your users are interested in larger volumes to capture lets say human body motion. So your system should be doing fine for larger volumes.
So far I think Tracking Tools 2.0 with 3 V100 cameras would do our job. But since 3 markers is the minimum number (while for most of our customers 2 would be enough), I have to find small and lightweight markers, or I have to use patches or paints which is not ideal.
2- What is an estimation on the resolution of tracking if I use only 3 cameras?
3- Can I perform better with more cameras, considering that occlusion is very unlikely in my case?
4- In addition, is there any "best camera setup" that I can stick to? (I would rather a mathematical solution rather than pure heuristic explanation)
5- When do you think, you can offer external synchronization of cameras?
Synching your cameras with active markers (pulsed IR LEDs) would decrease the need for eliminating the reflecting objects in the scene and increase the accuracy, I believe. Plus, I can sequence the LEDs to get unique positions in each frame without rigid body calculations (current 100fps is really more than our requirement so why not divide it for better tracking).
Therefore, I would like to know the possibility that we can also offer active LED tracking to our customers, some time in the future.

We want to integrate your system to ours and offer it to our customers as an add-on.

dashesy

[VincentG]

1 - We have many customers, that use the cameras to track in a more confined tracking volume, and the cameras are designed to track anywhere from 18", out to 20', depending on their setup

2 - This will depend on camera placement, and calibration, but up to sub-millimter

3 - With 3 cameras, in a confined area, you should be just fine.

4 - For setup, I would recommend setting up in a triangular fashion, with the camera heights varied as well, to get the most angular difference between cameras, and thereby getting a more accurate calibration and capture.

5 - We are looking to add it, as a feature, but I can't verify a timeline for release.