Test methodology for Hand gesture recognition application in Kinect

Question

I am developing an Immersive image navigation project in Kinect, which uses hands-free gestures. I have decided upon a set of gestures, which I shall use in the project.

I am working on the algorithms. What I want to know what is the general methodology for testing the various components of such Kinect related projects.

How must I design the test suite, what will be its fields? How must the different gesture recognition algorithms be tested? What is an optimum number of tests which establish data worth presenting? How many participants must be involved? And what set of data must be collected for the required coverage of test information?

Answer 1

With Kinect SDK you can record the gestures and then replay them to verify that your code can properly detect them.

Doing gestion recognition is hard, though - try to record different people doing different gestures and then see if your code can recognize them all.

Most of the time machine learning is involved in creating such recognizers - for machine learning to be working you need to record many many people for training the recognizer - I would advise against this and reuse the gestures offered by the Kinect SDK (grip, release for example) unless you know what you are doing.