Add input stream for the Kinect.

[damellis]

[nebgnahz]

We need to address:

1. Find a Kinect.
2. Kinect on Mac or Openframeworks for Windows.

And the priority is low, I assume?

[nebgnahz]

Accidentally closed it...

[damellis]

I think the Kinect in particular is not a high priority, although I think we need a third sensor source to go with the color sensor and accelerometer... A reliable audio example would be good.

[damellis]

Some notes. I was able to get skeleton from a Kinect (model no. 1414) using ofxOpenNI which is a (largely unmaintained and undocumented) openFrameworks wrapper around an old version of OpenNI.

This required:

• Placing ofxOpenNI in third-party/openFrameworks/addons.
• Renaming ofGetGLTypeFromPixelFormat to ofGetGLFormatFromPixelFormat as the function has been renamed in recent versions of openFrameworks.
• Adding the ofxOpenNI source files to the ESP Xcode project.
• Adding the sub-directories of ofxOpenNI/include to the Xcode target's header search path.
• Copying the lib directory from ofxOpenNI/mac/copy_to_data_openni_path to Xcode/ESP/bin/data/openni/ directory, adding the latter directory to the project's/target's library search path, and adding the contained .dylib files to the Xcode project.
• Copying the config directory from ofxOpenNI/examples/openNI-SimpleExamples/bin/data/openni to Xcode/ESP/bin/data/openni/.

To make the Kinect data actually useable, we'd probably want to:

• draw the Kinect camera image / depth map (e.g. so you can see whether or not you're in the frame)
• support depth thresholding
• support selection of joints and coordinates space for them (e.g. world vs. body)
• find a way of recording training samples that doesn't require the user to have their hands on the keyboard. (Note that it can take a while for OpenNI to find the user's skeleton, so even a fixed delay at the start of a gesture may not work.)
• have a way to draw joint data as a skeleton / wire-frame

Alternatively, some of this is taken care of by Synapse, and there's even a Synapse example for the GRT. This seems kind of silly, since Synapse is built on ofxOpenNI but it might be easier than using ofxOpeNI directly.

[damellis]

Also, ofxKinectFeatures looks useful.

[damellis]

If we use Synapse, we don't need to show the depth / RGB image ourselves.

We can rely on the user to trim each training sample. We'll need a way to toggle recording of training samples, rather than pressing and holding a key while recording.

[damellis]

Use Nick's OSC receiving code.