Hi there! Have you considered rendering in real-time with the condition that the surface can only change in elevation x-centimeters every 1/30th of a second? That would allow you to hold your hand higher than that x-centimeter mark and it would avoid detection, however if you bring it down to the surface, you can still interact with the liquid in real-time. That real-time liquid interaction really looks like something special to me. And possibly a simple range filter could quiet down that terrain noise.
A hand hovering over the sandbox gives a very steep raise in the depth data that could be filtered out. A hand very close to the sand is more difficult to detect. Have you tried with different or shorter values for the delay?
+ReBootYourMind that, if you detect something higher than the sandbox box height, then it is a hand and shouldn't be part of the height of the sand. As for the low noise pass, can't you apply it after the data has been processed but just before you output the image to the projector ? You might be able to apply the low noise transparently this way, while not having the latency, but data might be less accurate.
My team and I (ECED educator) are thinking to build one of these and put into our kindergarten program while sharing with professors at University of British Columbia and let children/teacher candidates to play with it. Any existing story that this has been brought and used in kindergarten program?
Do you use kinect V1 or V2 ? Did you think about using multible kinect to reduce the amount of noise? Sure it would make it more complicated but would probably result in more stable results.
+fffmalzbier look older videos, this guy was one of the first to use multiple kinect to 3d map things and rooms. But it seem you get even more noise that way.
Could you not just only apply the latency to areas which only change a small amount in the Z from frame to frame? Really cool stuff anyway, love watching your videos.
+wonay92 The kinect can not track temperature, it has no sensor for that. The kinect just tracks the depth input and can recognizes shapes that meet the definition of a hand.
fffmalzbier I'm positive it has an infrared emitter and scanner built in, it's in the centre of the device. (Bear in mind I'm talking about a gen 2 Kinect not a gen 1)
+Ryan McCall The IR sensor is working for night vision but it's not capable of capturing the wavelength that would be needed for thermal imaging. The Kinect V2 IR sensor does a range of 827-850nm wavelength but the wavelength for thermal imaging of humans is 8-15 µm (8000 to 15000) nm (nanometers).
+Ryan McCall Both Kinect v1 and v2 work in the near-visible IR band, I'm guessing somewhere around 900 nm wavelength. I can still see the emitted IR light as a faint red glow. Thermal infrared radiation, the kind that is generated by warm objects around body temperature, and sensed by infrared thermometers and "heat vision" cameras, is in an entirely different part of the spectrum, somewhere between 8000 and 15000 nm.
You are right, It was one of our problem also when we train our box. thanks for your explanation. You can check here ru-vid.com/group/PLp5c0th7CDR7JB50FScNNWXQskuJBdWb5. Thanks for your time. Awesome!!