Posts Tagged: kinect


26
Feb 12

Hack, slash & play with kinect at FITC Amsterdam 2012

Today the workshops of FITC 2012 took place. I visited Koen De Weggheleire en Wouter Verweirder their Kinect workshop.

Koen and Wouter are developers of AIRKinect of which version two is almost to be released. Unfortunately my lovely new MacBook refused to install some of the drivers despite Wouter’s persistence. Fortunately they had a spare laptop to use (thanks!).

I’d seen demos and fooled around with the depth image before, but the next version of AIRKinect, used as native extension offers lots more then I’d ever seen possible in AIR (may of course be my ignorance).

Some of the features of the 2.0 preview we got to play with were:

  • depth image
  • point cloud
  • point cloud regions
  • skeletons
  • poses

Depth image

The depth image shows the image the IR camera registers in greyscale. Nearby white, far black. My first ever experiment was that I used this to, pixel by pixel, reconstruct a 2.5D image with Flash’s z-property of sprites. But…

Point cloud

The point cloud is a ByteArray with for each “pixel” an x-, y-, and z-position. This allows for precise tracking of, for instance, the closest thing the IR camera registers. Extend your arm and move stuff around, the famous Minority Report interface in the making.

Point cloud regions

Point cloud regions allow use of hotspots. If a certain region (a box) contains more than a set number of points from the cloud (the count is done by the extension), you could start whatever action you see fit. Du du, du du, du du, du du, ta ta! I can feel it coming in the air tonight … Drum like Phil Collins for instance.

Skeletons

The skeletons you may have seen before. AIRKinect supports 4 skeletons and then some additional center of mass registrations. With the skeletons you have access to the main joints, their position in 3D space, position projected onto the webcam image, orientation and more.

Poses

With the skeletons you can create poses, this is actually not done in the extension, but with a few classes in which you can define poses like assertions: right elbow, above, right shoulder. Combine a few of these rules and you can create complex poses which are easy to check.

There is no doubt more, but the distraction of and annoyance over the failures of installing macports probably made me miss out on some…

All together it requires a somewhat elaborate install sequence (which they will no doubt publish at due time) on the Mac, mine being the odd one out resisting all attempts to get it running. On windows it’s a bit lot less elaborate. Definitely will try at home to get it running on my desktop. When I succeed, hours of fun will be waiting :)

Pointcloud