Hi, guys! I'm having some trouble in making dual kinect to work...
I've followed the step by step and watched the recommended videos on the wiki page, but nothing seems to work.
I'm starting to feel reall bad about it, since everyone say that this is very simple to configure.
I was considering buying the ps eye set up due to its higher frame rate, but now I'm not so sure. If the simplest one is getting me in trouble, if I may never get the ps eye to work.

Hence I don't know what I'm doing wrong I got a few screen captures of the process to see if you guys could help me out.

A) The calibration stage. My region of interest had 486 frames. I've got 419 good frames according to IPI. Is it a nice proportion?


B)Saved the scene

C) Opened a new video and loaded the previous saved scene into this one. ( I don't know if I should be worried that the camera is so messed up right now. It's way to tilted.. I real life, the kinect is on a flat surface)


D)now things go pretty bad...
I've ajusted the height


E) After I hit refit pose I got this


F) Then by the time I track it forward, I goes insane!





Any suggestion regarding what I'm doing wrong?
Thanks guys.

Hi Kyle,

Yes you need at least 500 frames to calibrate I suggest you use 800 frame region of interest.

Make sure only one depth sensor is showing in the display.

dont touch the camera during the 800 frame calibration.

LEt me know if that helps, it should.
Cheers

Also,

FOr 180 degree calibration you need to hang the card board in the area like a crane.

Not stand behind the cardboard, that is only for 30 degree calibrations.

Cheers

Also for your TPOSe use a Palms down.

I notice your hands are facing the camera.

Hands should face the floor in your tpose.

Cheers

When I do dual Kinect '90-degree' calibration, I tilt the board up slightly--I think this helps Kinect/iPi Studio read the surface better since there's more variation in angle this way. Also, when you rotate from side-to-side, be careful not to over rotate the board--the surface needs to be seen by both cameras at all times.

What is the surface of your plane like? I use a flat white foam core board, but a mid value color is probably best. Do not use black or very shiny surfaces--Kinect cannot read anything from these surfaces, and you will just get a big 'hole' in your scene where the board should be.

As for palm direction, I always use face palms toward the front (or camera.) I feel this gives the system a bigger 'target' to look at when it begins tracking, which may improve accuracy.

Hope this helps.

G.

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

Watch a one minute excerpt on Vimeo now!

I assume you're doing a 90-degree setup. FWIW, I prefer this setup because it seems to be a bit more accurate over 180--I'm guessing this is because of the geometry overlap that may increase resoultion. However, if I run into occlusion issues, then I switch to the 180 setup, which of course can nearly see completely around the performer. Without any overlap, however, it can be a little more 'jittery'. Be advised that this is just my opinion based on personal observations; other users have reported differently so be sure to experiment and see what works best for you, as your environment and requirements may be different.

G.

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

Watch a one minute excerpt on Vimeo now!

mrBones, greenlaw thank you very much for your feedback. I'll take all of them into consideration next time I'm doing my captures.
In my latest test, I finally got it to work. However, I still don't now what I was doing wrong...
I changed the Pc and kinects position. Maybe it was the limited capture space, don't really now.
Which frame count do you consider ideal? For instance, in a 500 frames shot, I should get 500 good frames on calibration?

Thank you very much for your help!

For calibration, I just rotate the plane a few times, maybe three or four times and I set a range that seems 'good enough'. Typically, 500 to 600 frames should be sufficient. It appears that iPi DMC averages the data so, while adding more frames may improve calibration quality, at some point adding more frames may simply become redundant. Variables in environment, accessories, and configuration can affect quality, so it's probably a good idea to just figure out what works best for you.

Here's a suggestion: record an extra long calibration video (rotate the panel several times) and the set your calibration range in expanding stages, and take note of how much the calibration quality improves. At some point, the difference will become negligible and you'll know what you're optimum frame range is. Use this as your 'rule of thumb' for future calibration.

There is no relationship between how long a motion clip should be and how long a calibration video should be.

BTW, I hope you're not recording a calibration video for every motion in a session. you only need one calibration video for an entire session--you can use the same scene data for all motion files, assuming the cameras are not moved. I usually record one calibration video before recording all motions, and then I record a second calibration video after the last motion just in case a camera got 'bumped'.

Hope this helps.

G.

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

Watch a one minute excerpt on Vimeo now!