Hello

I have had no problems with calibrating 4 eye cameras, but with 6 eye cameras I cannot get an accurate setup it seems.

I've tried multiple times, and although all cameras are roughly the same height from ground , i get results like this picture:
http://s14.postimg.org/68daqwl81/problem.png

How can camera#6 just be located several meters above the others and still the dots are green? It must be a mathematically incorrect solution to the camera system matrix?

Compared with 4 cameras calibration its not possible to keep the light source visible in all cameras at the same time, is that really required?
I'm sure i can get a accurate camera setup by using 6 pictures from each camera using a program like pfTrack with the user track module (which uses simple photo grammetery as well).

The only problem i can think of is that the light source doesnt start from top and go down , but is that really required?

Thank you very much for helping me out with this!

That's really weird. Have you checked that the lens switch on the camera is the same as the others? You can dial the camera lens for wide or narrow FOV, and maybe the one flying high is set for narrow?

G.

_________________
Greenlaw
Artist/Partner - Little Green Dog | My Demo Reels (2013,) (2015,) (2017,) and (2019)

Watch a one minute excerpt on Vimeo now!

The marker may not visible to all cameras at the same time. But it should be good visible to each individual camera most of the time (80% minimum). That will lead to a sufficient intersection of marker detections between various combinations of cameras.
The pattern of moving the marker is not important. Just ensure the visibility, good capture space coverage, and making several ground touches.

Check the FOV setting on cameras, as Greenlaw suggested.
What is the "marker misdetection" values for cameras? (They are shown at the "Scene" tab after the marker detection pass)
Do you position the cameras in proper order before running the calibration with auto detection of positions off?

Hi,

Thanks very much for the quick reply.

Within the next calibration video after i posted i was actually able to get a successful calibration. It looks very accurate finally, great!
My simple tip is that a longer video seem to work less good for 6 cameras, try to focus on having a short video with the light source visible in all cameras instead.
Probably like you said the movement shouldnt matter as long as you have sparsity enough and atleast 3 ground points (i had exactly 3 and its enough).

Unfortunately i get some bad results with the actor arms when tracking, they get double folded and penetrate the actor torso, probably the arm color is too close to the ground color so will try to put on a long sleeved shirt. Mistracking seem to happen when the arm is too close to the body. I'm also using a small virtual prop for the actor.

Btw, wouldnt it be great to add support for being able to change how wide the actor pants are (or simply thickness of legs), i was thinking of using tights when i record to make leg movements more pronounced?

Greenlaw, thanks.

Yes, they are all wide PS eye's, believe thats 75 degrees FoV.
The problem was probably just too much data, i made a much shorter video that also had the marker visible in close to all cameras all the time and that worked much better.
Not much difference in tracking quality from 4 to 6 cameras though :(

Still problems like arms double folded and penetrating the chest. Why doesnt Ipi studio have more constraints? Why does it need to support broken looking arms and body parts penetrating chest etc? Wouldnt tracking be much more solid if those type of solutions could be ruled out? For example why not have a settable angle span that each joint would be restricted to?

We plan to add more customization to the actor's model in v3.


We already have internal constraints on individual joint rotations, closely related to anatomical limitations. Maybe we also make these configurable in v3.
There are many ways to improve tracking consistency that we plan to work on.

Many of the tracking errors can be easily cleaned up by tracking backwards, manual editing + refit pose.