hi. few points.
1. for skeleton tracking all CPU is taken by kinect runtime. driver does not perform intensive calculations.
2. if you have led tracking low cpu will be only when all ball colours properly calibrated and tracked!!! if even one colour is not tracked then driver needs to find it so instead of searching in previous region it must scan again whole image. what I have seen it happens very occasionally during gameplay.
i don’t think that using ROI should affect quality of trackng but not 100% sure.
3. if you use only colour tracking (no skeleton tracking) then kinect runtime does not start processing skeleton therefore only CPU used is by driver to detect and process led balls.
if you use skeleton tracking for headset and led for hands compared to led only scenarion then kinect will start processing skeleton but on the other hand we will not have driver processing of one ball that is used to track head.
i have not tested which one is better.
i still have few ideas, not so classic as using roi so we will see.
anyway I think this one is good for release as I can see no extra problems were found.