Error message

Deprecated function: Array and string offset access syntax with curly braces is deprecated in include_once() (line 20 of /netapp/www/home/devkin/domains/

Kinect Hand Pose Recognition

Kinect Hand Pose Recognition

One of the original founders of the OpenKinect community, Joshua Blake, has been hard at work, toiling in his secret gesture recognition laboratory hidden somewhere deep beneath the Atlantic Ocean. Now I'm not sure about the validity of that previous statement, but that's how I envision where really smart people go to do their best work. Anyway, I popped by the OpenKinect Google Groups page and noticed that Mr. Blake posted something very interesting the other week.

He's been developing a Hand Pose Recognition system for use with the Kinect. This is incredible work and has a ton of potential, mainly because our hands are so multifaceted. We can perform a wide variety of unique gestures with our hands which could then be translate into computer commands. These gestures essentially would be captured and interpreted by the Kinect, then sent as a unique command to a computer or any other digital deveice.

Right away I can see how this would providing a huge advantage for deaf or hard of hearing people. If this application is able to recognize all sign language gestures including letters of the alphabet, the Kinect could essentially act as an automated stenographer for a number of people signing in a room. You could also -- and I assume this is the point of the entire process -- create a custom set of gestures to control a number of computer related actions such as keyboard shortcuts or launching applications and even mouse functions.

The decision forest used to recognize the hands in this image was trained with 625 computer generated and labeled images representing the combination of five images for each of four degrees of freedom for the hand/wrist range of motion. Training three trees in the forest took just over 11 hours on a Core i7 2600. Additional degrees of freedom are planned, but take longer to train.

Exciting stuff to say the least! The program was written in C# and hopefully will be available for us all to check out once he gets the "go-ahead" to release an excecutable. Furthermore, the program works with both OpenNI the Kinect SDK. If you'd like to follow the development of the project, here's the LINK to the google group post.

Check out the video below for a sneak peek of Josh Blake's Hand Pose Recognition application in an early development stage.