The biggest challenge here would be self-occlusion, especially the arm hidden under the subject. I would expect to kill the keyframes for that arm and animate it manually after you've tracked the rest of the body.
In general, you'll need to position the cameras strategically to capture as much of the visible body as possible. If you were using PS Eye cameras, you would position a couple of cameras very high and some very low. Seeing you have dual Kinect, you might try one higher up and the other just a little lower than normal, favoring positions with the most visibility. Don't go too extreme though--as you many know, you don't have as much flexibility when positioning Kinect devices as you do with PS Eye cameras.
If the actor needs to do more than maintain that position, like rolling in bed while talking, that gets more complicated. If this is for a film project with camera angles that cut from shot to shot, try to edit around the motions because it will be easier to capture these motions in separate takes.
Another thing to consider: avoid shooting in an actual bed as this may cause even more occlusion problems. If the actor isn't moving around too much, consider shooting with the performer on a long bench with a firm surface or padding. If the surface is too soft (like a mattress,) the performer may sink into it, causing even more occlusion problems. If a bench or appropriate furniture isn't available, consider raised platform you can move around. The point is to keep off the floor because the system will clip any data an inch or two from the floor. (It's an issue with Kinect only.)
If you use a bench or platform, calibrate your setup before putting it into the scene. Then, when you're ready for motion capture, add the items but also be sure you run a Scene Evaluation to subtract the furniture.
Starting from a T-Pose helps as it gives you a starting point that can lead into the pose you want, even it it doesn't go 100% into the intended pose. More than likely, you'll need to adjust the body once it's near the intended pose. Since this is a complicated pose for Kinect capture, I would definitely experiment with this pose before going into an actual shoot--you may need to adapt the pose to favor what's actually trackable with Mocap Studio.
I haven't done exactly the example you've shown in the photo but based on what I have shot in the past, I think that should do it. Hopefully somebody with experience in shooting this pose will pop in with proven tips.
Good luck.
G.
|