Removing Unclassified Hand Tremor Motion from Computer Mouse Input with Neural Networks

Abstract

An artificial neural network based filter to remove unwanted tremor-induced motion in computer mouse input is presented and tested. A method to efficiently capture appropriate training data is shown to be important in the operation and training of the neural network filter. The architecture of the neural network as well as the numerous design choices are presented and explained. A simulation study proves the artificial neural network is successful at removing a simulated Parkinson's tremor from computer mouse movements even with minimal training data. Resulting tremor-free motion estimated by the artificial neural network is shown to be similar to normal tremor free computer mouse movements.