Hi Snapz,
Thanks for the comments. That's very interesting, although some of what you wrote contradicts my own experience over the years:
Snapz wrote:
...I am wondering if you ever tried using a black mat or rug on the floor? I believe has to be solid black or very very dark color, all I tried was black.
We happen to have a large brownish patterned rug on our living room floor which has been very useful. I learned a few years ago that when you have a shiny floor, a rug is very effective for knocking out reflection, whether you use PS Eye or Kinect, but for different reasons.
Two things:
1. In my experience, a matte black carpet might be a poor choice because that specific surface tends to absorb the IR light and you want the surface to bounce rays back to the sensor for them to register the depth. At least, that's what I've noticed about matte black for clothing and props. I haven't thought about what it means for the floor--maybe interfere with proper floor detection?
2. The 'reflection' I'm talking about is actually more visible on the rug directly in front of the opposite sensor, and the polished floor outside of the rug. So technically, it's doing exactly what it's supposed to do--the issue is that the near sensor is picking up rays from the opposing sensor--I think this is why it reads as 'yellow'.
Quote:
I have tried this with dual Kv2s, and it makes no data return from the covered floor area, (solid yellow), which is fine, it doesn't affect the floor grid setting, or the recorded performer point cloud and allows more of the feet to be detected when on the floor.
I found yellow areas not to be a bad thing with Kinects, it eliminates some of the background point cloud noise in the processing stage and has no effect on the point cloud actor tracking.
That's very interesting. It sounds like instead of having the software accurately create the ground, you are in effect canceling out the ground scanning altogether. I guess since the orientation of the ground plane can be assumed with just a little bit of data sampled over a large area, maybe that's all you need?
Don't know. Personally, I still think more data would lead to greater accuracy though. For example, even though the 'near' sensor was canceling out IR rays from the far sensor, I still got some of the best calibration and motion recordings last night. I think this is because I still got very clean floor data for most of floor in the the actual performance area. (Although, I should also attribute improved results to the recent Kinect-specific improvements in Mocap Studio.)
Quote:
I will try a demo with the Kv1 and see for myself using this theory, if it does then should do the same with Kv2, I already know it worked for the floor with Kv2.
Yes, multiple Kinect One (v2) sensors do work significantly better than multiple (v1) sensors. The main thing I've noticed in comparison is that the captured data is less noisy with Kinect One (v2) so there is far less 'jitter' in the data to start with. Also, I'm finding Kinect One (v2) works much better in daylight than v1 did--v1 is apparently more susceptible to IR interference from sunlight. (Although the recording last night was obviously done at, well, night.) :)
