|
...
If you want to check the tracking possibilities with more than 3 actors, you can try to just use single actor process and track each video by the number of actors in scene. Would work better for more isolated actors in the scene, I can not test it with Kinects, but it could be a possibility for some situations.
You can then try to judge how body occlusions will affect the tracking, although you won't get the transparent (grayed out) actors as you do with 3 actor process selected, you can test the tracking with over 3 actors this way anyway.
Use the 3 actor selection for first 3 actors if you want, then just switch to single actor selection for any other actors, but it would always be best to try to keep some actor separation when possible, this is true with all camera based mocap systems for best results.
The way you had the Kinects set up was the best way for maximum depth of field, adding 1 more to split the front area will give 1 more view, but to try to use a circular set up will cut down on the capture volume with Kinect sensors, use the semi circle set up but don't make the outter 2 sensors face each other straight on, let them point at the angle they were shown in the previous post.
This way your actors can perform further away from the sensors, as well as closer, you just have to be able to get a good calibration from the set up, which it seems you had pretty good.
I believe also you would get better tracking and larger capture volume with multiple PS Eyes, but there are more requirements to be followed for best performance.
...
On another note, I don't know if this would work, but maybe to be able to use more Kinects to record the same scene, it may be possible to use 2 separate Modems (different IP's) along with using a little different semi circle sensor config. where the middle sensors would overlap views a bit and using 2 Recorders with a Master/Slave for 4 sensors on each set up.
Would still take 8 computers to achieve this, with a bit different sensor config., but it would in theory give a wider capture volume ability, while maintaining the better depth of field that could be walked between the 2 set ups, then in post, splice the animations back together at the appropriate points, removing the unneeded frames from the overlapped sensor tracking exported animations.
You would just have to manually position the actor in studio to refit the transition frame for tracking, which isn't hard, as a T-pose is never "really" required to refit the actor pose, and you can export in a T-pose first frame on export anyway if one is needed.
This wouldn't be optimal for most users, but in your case it could possibly be an alternative for additional sensor use and a bit wider volume, it's just a thought, I have no way to test it.
...
|
