An improved computer interface comprising a recurrent neural network and a natural user interface

We developed an interface system by which a user can operate a computer with hand and finger movements. To implement the interface, we used a gesture sensor to acquire the movement-based data. A recurrent neural network (RNN) was included to discriminate types of gestures. Using the proposed interface, high recognition rates were obtained for simple gestures, while the recognition rates of complicated gestures were low. To improve the rate of accuracy in recognizing complicated gestures, we investigated the dependency of factors on the rate of recognition in the RNN learning process and identified settings to refine these factors.