|
That's very strange about your Kinect results. A bit of yellow can be expected in the background and it usually has little effect on tracking so long as the performer data and floor is good. What can mess things up is when the performer gets too close to a wall or furniture because the volumes start to merge, creating a confusing shape for the tracker. But that doesn't sound like your problem.
To me, this issue sounds like poor calibration data. If you're using two sensors and the flat plane method for calibration, make sure your plane is adequately large enough and has sharp corners. The board must be rigid--I like to use a large piece of 1/4" foam core purchased from an office or art supply store. It should not be shiny as that may reflect the IR or black as that may absorb the IR. Move it mostly in depth (z), less side-to-side (x,y), but also rotate from side-to-side (along x and y). Don't move the board too quickly--the software needs to figure out the shape in 3D as clearly as possible. I usually record about 600 to 700 frames. You really don't need more than that--IMO, too much data just slows down the process and won't appreciably improve accuracy. If you do record more, you can trim down the ROI.
Alternatively, you might also try Maglite calibration, which works with two or three Kinect sensors. (The board method does not work for three sensors.) Maglite can produce more accurate results but it can be trickier to record in a brightly lit room. Also, it's important that the light is visible to all sensors as much as possible.
What was the arrangement of your Kinect sensors, how many devices? when using two, I usually have both at one end of the room angled at about 70 to 90 degrees. This is good for many situation, even 360 capture if the action isn't too complicated. If 360 capture it important, I move the second Kinect to the opposite side of the room. I don't usually do this because I feel the fidelity is better when both Kinects are on the same side and closer together, but some actions obviously do need wider coverage and you just have to do 360. Placing a Kinect on the far side also decreases the size of my capture space because our living room isn't terribly huge.
When I'm using three sensors, I arrange them in a triangle with each devices a evenly spaced as possible. I find this harder to calibrate but the results can be very good. I use this for motions that are more complicated, but I still use the two sensor config for most situations because it's so much easier and quicker to set up and use.
To tell the difference between an XBox and a Windows Kinect, the XBox version says 'XBox 360' on the front and the Windows version says 'Kinect'. Functionally, there are a few differences, so which relate to iPi and some that do not. For example, the Windows version has much better video adjustment controls, which I find very useful for getting clear 'well lit' RGB video at night. The Windows version also features Near Mode but that's not used by iPi Mocap Studio; the Xbox version dose not have this feature.
Hope this helps.
G.
|
