Thanks for your feedback. I'll try to explain what is going on.

Actor scaling from center (rather then from base) is not a bug but a feature. The logic behind it is the following:
It is possible to re-scale the actor after you did some tracking (for instance, if you found out that initial guess of scene scale was a bit off). Scaling from base would result in feet (in new animation) being close to floor but head and hands being way too far from desired locations. Scaling from center results in much more reasonable estimate of new locations for all body parts. Of course, it is recommended that you re-track entire animation clip if you changed actor scale. But scaling from center at least gives a reasonable initial guess for the new animation. Scaling from center is a consequence of our skeleton design (root bone is hips - typical for many animation rigs).

Now to why iPi Studio doesn't want to translate the rig down to align with the ground after scaling it smaller. I think this can be an effect of trajectory filtering (you change one frame and trajectory filtering tries to "fix" actor translation based on adjacent frames to avoid discontinuity). My guess is that setting trajectory filtering to zero should allow you to change actor translation as you want. (You can reset trajectory filtering to default "1" later, after tracking).

About temporal offset in data from two Kinects: I believe this is a result of Kinects not running in sync. By design Kinect does not have any hardware capabilities for synchronization of multiple sensors. iPiRecorder captures data from two Kinects not running in perfect sync. The synchronization offset is random and is anywhere between 0 and 1 frames. Currently there is no way to fix this in software. (In the future, we may invent a way to mitigate this problem in software by using various tricks, but it is unknown how much time it may take).

You are right that PS Eye cameras can help in capture of fast motions, because PS Eye can run at 60 frames per second (twice the framerate of Kinect). Even without any means of hardware synchronization in PS Eye cameras, just the fact that PS Eye can run at higher framerate helps to deal with synchronization problems.

Curiously, motion blur with PS Eye cameras also helps (to a certain degree) to mitigate the effect of cameras not running in perfect sync. Spaciotemporal data from multiple PSEye cameras looks much "smoother" then from multiple Kinects. By design Kinect produces very little motion blur and this results in "sharp" discontinuities in multiple Kinect data streams. Motion blur is very small with Kinect because Kinect works with short exposure times. This is possible because Kinect has its own light source and also because Kinect uses a monochrome camera working in near Infrared light (modern CMOS light sensors have best sensitivity in monochrome IR light).

About optimized body shapes for children tracking: yes, we will be working in this direction. New body models in upcoming v.2.0 should allow us to make such optimizations. It may take some time (probably till v.2.1 or v.2.2) because currently we still need to do a lot of tuning of new models for more common cases of adult man and adult woman.



Yes, we are always looking for more test data for various interesting cases. We would appreciate it you could send us the corresponding video. ( http://wiki.ipisoft.com/Troubleshooting ... ch_support )