Lifelong and Fast Transfer Learning for Gesture Interaction
Hanchao Yu · Journal of Information and Computational Science · 2014
Aimed at the problem that gesture interaction model is not robust to everyone and can not quickly transfer to new users, this paper presents an extreme learning machine with lifelong learning mechanism and fast transfer learning ability. Combining the proposed extreme learning machine and gesture interaction makes gesture interactive system be able to update gesture recognition model continually by the increase of using number, and then makes the model become more and more robust to everyone. At the same time, the model can quickly and accurately recognize the gesture of current user by adjusting the transfer strategy of model. Finally, based on the proposed gesture interaction method, we develop a fingers guessing game facing cerebral stroke early warning. Experiment results show that the proposed extreme learning machine can update the model continually and rapidly transfer it to new users.