After i solved my secondary usb controller problem i started to try it.

Initially i set my 2 tripots for the kinects. i used piece of rubber. i set them to the same height. for 2 kinects calibration web page there is no info for the distance that kinect will be far from the floor. mine is 1.30 m.

there should be pivot points on the screen like center or thirds. so we can place camera for a reference object. that helps calibration.

calibration takes long time and it uses cpu. tracking uses gpu. idle state ipi recorder uses %15 my cpu. so after recording it is better to close it.

i calibrated and saved scene. then i closed ipi studio. i reopened it. i opened video. i loaded scene. i zoomed out. it shows my cameras at front of actor. actually they are placed 90 degree.

to be continued...

Try reloading your scene data. I've seen this happen here and reloading the file fixes this glitch for me.

I'm surprised that calibration is taking a long time for you. For me, it's been relatively quick--about 5 to 10 minutes depending on the duration of the footage. 500 to 600 frames should be plenty. Are you doing both recording and calibrating/tracking on a laptop? If so, that may be the issue--laptop graphic cards are not nearly as fast as their desktop counterparts.

G.

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

Watch a one minute excerpt on Vimeo now!

there should be an indicator for calibration quality.

when hands near the body tracking loose pose.

i stand front of 2 cameras. depth data also says that. but ipi studio wants me to set t pose like i am front of the first camera. i have to move and rotate the pose. with one kinect it was easier. rotation is not easy. there should be shortcuts for top, side and front views like blender has Num 1, 3, 7. it is very hard to rotate view especially when you look from top.

edit : solution for T pose it to stand to first camera :-) no to both.

I believe the three green dots on the surface of the calibration board are the indication of a good track. In general, you should see this on the board throughout the entire scene. So long as the depth of the calibration board can be fully determined throughout the process (about 500 to 600 frames at least,) calibration should be automatic and 'hands off'. FWIW, I've had perfect calibration each time with Dual Kinect and have never needed to make manual post adjustments in Studio. If your calibration result is so far off that you need to manually adjust something, it's actually much easier and probably better to simply reshoot the calibration video. Getting good calibration (as far as Studio is concerned,) is absolutely necessary for good tracking, and it's actually very easy to do if you follow a few guidelines.

To get good calibration here are a few things to be aware of:

When calibrating, watch that the range of depth is visible to both cameras. Over rotating the board will cause Studio to lose its track for one or both cameras. Understanding the degrees of rotation is easy to learn by looking at the of range of colors on the computer screen. In general, I do not push this rotation to the extremes, keeping it just under the edges of visibility. Once you understand the range, you can easily figure this out without a monitor by just paying attention to the positions of each camera.

Also, make sure there is plenty of space all around you when calibrating and capturing. If walls or furniture are near you, this may compromise the quality of the volume generation. Also, try to stand in the center of the capture space...this is area where calibration and tracking will be the most accurate.

If you look at this video here, it may give you a good idea of where you should stand for calibration and performance, and where the two Kinect Sensors should be. This is what has worked almost flawlessly for me with the beta software. (Even with the occasional interference of two cats, my tracking results have been good.)

As for where to face a T-Pose, my T-poses for all the motion capture for Happy Box were done facing between the two Kinect Sensors and not favoring either one, and the results have been excellent.

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!

One more tip: You should make sure the FOV of each sensor is aligned fairly symmetrically. This means angle and pitch should be similar (doesn't need to be exact, I just 'eyeball it',) and that the space where they cross is fairly balanced. You can determine this by placing an object like a box or a chair in the center of the room and looking at the monitor to make sure it really in the center for each Kinect camera. If not, adjust the camera by physically rotating it and using the motorized pitch control.

I think this is a key bit of info that has guaranteed good calibration for me.

G.

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

Watch a one minute excerpt on Vimeo now!

thanks for info.

can you try this pose?

t pose, hug yourself turn several times half t pose, turn several times, full t pose. i have problem with this workflow. tracking looses

it is based on this :-)
http://www.youtube.com/watch?v=29SMYaEwGyM

i worn gloves and shoes to make hands and feet more traceable. i tried calibration with a bigger board.

i tried simple moves with the pose close to t pose. my turns has no problem.

i have problems when my arms close to body.

Hi Atmuc,

You may have to track a few portions of your backwards,

(about once or twice) You may have to use the rotate tool on the shoulders and or forearm.

Luckily these tools are now available.

Are you using Kinect? If so, gloves and shoes make no difference as iPi Studio does not use color video data from Kinect, only the depth data. If fact, the color video stream is purely optional. For example, if bandwidth is an issue, you can disable this feature completely without affecting tracking quality and results. (The color video is useful for visual reference only.)

The 'looseness' of your clothing will make a difference though. If your clothing is baggy, then obviously the shape of your body becomes more difficult to track.

If you're using PS3 Eyes, then that's a different story. Unlike Kinect, the PS3 Eye setup uses color video data for tracking. However I don't think gloves will help as I think the software expects your hands to be same color as your face. The color of your shoes may make a difference (I wear white socks when using PS3 Eyes.)

As for hugging yourself or crossing your arms, Kinect can do this but I found that PS3 does this much better. The issue is that when your arms are that close to the body, the volume can 'blob up' and get confusing. You can see this happen if you study your volume shapes. You can avoid this problem by not completely hugging, making sure there is a gap between your body and arms. Study the volume and practice for best results.

I don't have a posted Kinect video of me crossing my arms but I did something like that in this PS3 Eye video about a year and a half ago (it occurs near the end of the video):

iPi Studio Markerless Motion Capture: Deform Test Mocap

This was done using four PS3 Eye cameras and much earlier version of iPi DMC.

If you watch some of the other videos where I used Kinect, you can see the arms get close to the body. I was not successful with my first takes but after studying the issues I was able to adjust my performance in a later takes which were very successful. The key to good capturing is to understand how the software works, understand what the limits are so you can adjust for them, and rehearse your performances.

Like anything worth doing, using this software successfully takes a little skill and practice.

G.

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

Watch a one minute excerpt on Vimeo now!

Yeah. I'm wondering if there might be some solution in the future that utilizes the RGB data to help with this issue. In some of the Kinect hacks, you can see the 3D data with color info mapped in real time. I imagine this gets processed within the Kinect itself. Not sure though.