Gesture recognition using the convolutional neural networks and the Leap Motion controller
This code uses the MatConvNet toolbox in Matlab for the convolutional neural network implementations, which comes from Oxford's VGG: http://www.vlfeat.org/matconvnet/
The toolbox can run on the CPU or GPU with the change of a single boolean in the code.
The data that the code runs on is located here: http://spiegel.cs.rit.edu/~hpb/LeapMotion/ It is approximately 1.5 GB and consists of over 9 thousand individual gestures collected from students, faculty, and staff at RIT. The gestures collected consist of the following:
{'capE', 'CheckMark', 'e', 'F', 'Figure8', 'Swipe', 'Tap', 'Grab', 'Release', 'Tap2', 'Wipe', 'Pinch' };
Further information on the collection of the data can be found in the DataCollector repository.