So you record from the kinect using ipi recorder and use the avi in ipi desktop. When you turn to show your back ipi desktop makes the arm that has currently no exposition(fi turning left, your left arm is hidden from about 60° rotation) jitter anywhere in the space not being seen.

Is there a prediction algorithm that is better/assumes from the rest of the bones/physics how the arm is likely to move....

This is a huge issue I have with ipiDesktop

Occlusion is a big problem with single Kinect; this configuration cannot track what it cannot see. Sometimes you can get Studio to interpolate the path of a limb by tracking backwards and forwards but this is not always possible. With single Kinect it's best to avoid motions that involve occlusion. Alternatively, you can try performing at an angle to the camera so that it can see more of what's going on 'behind' the performer.

If you absolutely need to turn away from the main view then you'll probably want to upgrade to a twin Kinect or multiple PS3 Eye configuration.

Here's one more option: If you don't need to see the character's legs or feet, you can edit around this problem. For example, I originally shot Happy Box using a single Kinect, and for the 'chainsaw dance' my intention was to shoot the dance facing camera and then rotate the entire character rig 360 degrees in my 3D program, faking the turnaround motion.

In the end, I shot the turnaround properly using twin Kinects but I think the single Kinect 'cheat' would have worked out fine because you never see the character's legs in the shot. (To be honest, the final shot almost looks like I used the 'cheat' anyway.) :p

The point I'm making is that shooting with a single Kinect is limiting only if you allow it to be. There are many tricks you can do through clever composition and editing that can make the audience think they're seeing more than what's actually happening in front of the camera. (Ask Alfred Hitchcock about this.)

G.

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

Watch a one minute excerpt on Vimeo now!

I have other 2D cameras. How would you mix a stream showing my back with kinect's source?

If it was all 2D it would be 2 or more videos next to each other but what about 1 kinect and 1 other video? I could record the kinect with ipi recorder and another one with my camcorder but how would I combine them for ipi desktop to understand?

Push.

You cant.

Its 1 kinect, or 2 kinect, or 3-6 webcams... thats it.

_________________
My Site

RH is correct; you can't mix Kinect with 2D video. Just curious but why all the bother? It sounds like you're attempting to turn a fairly simple process into an unnecessarily complicated one.

Ideally, you should use the same camera type for all cameras when recording. That said, it is possible to mix different 2D video types, for example, PS3 Eye cameras and regular HD cameras. To do so, you will need to composite your video clips in a format that iPi can understand, synchronize and color match the videos, and output it as a new movie file. You can see an example of how this works by looking at 1k0's behind-the-scenes video for Danse Kabyle.

This is just my opinion but it's a lot of extra work for no real gain in quality, cost savings, or efficiency. To be clear, higher resolution doesn't necessarily improve tracking quality but it will probably increase calibration and tracking time. Higher resolution is really only useful if you're recording a larger capture space, which is possible with PS3 Eye but not so with Kinect, so in this situation it makes no sense to mix them.

FYI, framerate is more important than resolution--higher FPS means little to no motion blur which can affect tracking accuracy. The PS3 Eye can record iPi data at 60fps, which is very good for a entry level system like this. The Kinect system can only do 30fps but it's the convenience of the system makes it worth considering. You really need to choose one system--by keeping the devices in a configuration the same, you can be sure that the data will be synchronized with consistent image quality. Also, by keeping the cameras the same, iPi Recorder will do all the formatting for you automatically--no need for compositing, syncing, or color matching, just plug-and-play.

Hope his helps.

G.

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

Watch a one minute excerpt on Vimeo now!

You are not with ipisoft so you are unable to determine whether this is impossible. Coming from the nature of software it is well possible, you just don't know how to do it.

No, seriously, this is not possible with the current software. Really. Positively. Almost absolutely. Maybe? No. :)

G.

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

Watch a one minute excerpt on Vimeo now!

Hey there happy people. (Greenlaw, RH!)

I have to say I don't know why anyone would try it with a single Kinect or single camera at all. I picked up 2 kinects, and I'm running the trial at the moment but after I sorted out exactly what was going on in the calibration set up, I was able to record and produce 5 minutes of running mocap using the system and it was very very usable/clean immediately.

Personally I think JimmyRig with the Kinect is better than ipisoft with a single Kinect because JimmyRig is real-time, but you would still have the same issues with a turning character.

You know it very well.