As mentioned in my last blog, our lab got two Intel Realsense stereo cameras. I got a chance to try one of them for testing.
Ken told me he got an idea about multitouch detection by utilizing depth, and we wanted to see if it can work out on the Realsense camera. By pointing down the camera at a table from above, the camera knows how far it is placed from the table surface, and all variances happened in between the surface and a plane with a small threshold height above the surface will be recognized as hand touch. The idea is similar to the Kinect painting software that Ken talked about in one of his blogs a few years ago (http://blog.kenperlin.com/?p=6485).
The depth information from the Realsense camera is a bit noisy, so it will get a lot of false positive if we set a higher sensitivity, and false negative if lower sensitivity. But we got a pretty good result when we set higher sensitivity with proper filters and signal amplification procedures. The result is shown below: