Speaking of magic, universal uses Kinect for Harry Potter in Knockturn Alley. Was surprised to see a simple Kinect running it (hard to see but it’s in a box to the left of the skeleton magic wand experience).

The Kinect has done wonders for amusement park and museum efforts. At this point I'm more surprised when those sorts of museum and amusement projects aren't using a Kinect. (...and more often than that it's simply because an old expensive system hasn't broken yet, but soon as it breaks you expect it to be replaced with a Kinect.)

For that reason, the commodification of the Kinect really has been a boon for science and entertainment. There are some commodity Kinect knock-offs out there (Intel's cameras and sensors come to mind), but selling millions of Kinects means that they will probably stay important to museum projects.

The secondary market will probably remain flooded with Kinects for a while, but maybe (hopefully) by the time it becomes hard to get a decent priced Kinect for such cool little projects there will be a Kinect 3 or similar ready. (Or Microsoft will sell the Kinect brand to a hardware manufacturer.)

I went to a few museums last week on vacation. I was surprised to see how ubiquitous the Kinect was - almost every new-ish exhibit had something interactive that made use of the Kinect. Just when I was thinking about how much fun it would be to get one to play with, they killed it :(

It will be interesting if Microsoft quietly makes a "Kinect" platform that targets businesses / museums / etc. for this exact reason. (Basically, going the Google Glass route of discontinuing a consumer "flop" that still ends up being utilized, and produced, on the B2B end).

this? https://youtu.be/uUE_DaM8ymk?t=19s

Yep, that's the one.

Kinectimals was actually well beyond the standard skeletonization system - I think they rewrote a chunk of the stack to allow for younger kids to use it and have it perform well.

I think it was more similar to how Double Fine's Happy Action Theater / Kinect Party handled the system (I recall they had some great write ups of the compromises they found), which came later than my hazy memories of my young cousins but also fit that young of an audience well. The Kinect provided a spectrum of accuracy. The most accurate skeletonizations required the most trade-offs in number of users and unobstructed views, etc, but the Kinect also provided lots of less accurate data, such as the raw depth sensing, and you didn't have to rely on just one data stream. So you might use the depth sensing alone as your primary tool, and there were enough basic gestures you can easily watch that way (like reaching in to "pet a cat" can be a very clear depth change), and use more accurate gestures and skeleton tracking for cases where you actually need more accurate gestures and/or skeleton tracking. You would use one data source to backup the other rather than only programming to the easiest/most accurate/most in-depth model.

I find Kinect programming retrospectives fascinating. Particularly that fascinating feel that with the Kinect you have so much data at your fingerprints that good Kinect programming is as much figuring out (quickly) which data to ignore as which data to use, but also realizing that you don't have to trust just one source of the data you want a gestalt of it. (...and getting a good gestalt to be performant in real time is certainly a challenge.)