Support Vector Machine Learning for Gesture Signal Estimation with a Piezo-Resistive Fabric Touch Surface

The design of an unusually simple fabric-based touchlocation and pressure sensor is introduced. An analysisof the raw sensor data is shown to have significant nonlinearities and non-uniform noise. Using support vectormachine learning and a state-dependent adaptive filter itis demonstrated that these problems can be overcome.The method is evaluated quantitatively using a statisticalestimate of the instantaneous rate of information transfer.The SVM regression alone is shown to improve the gesturesignal information rate by up to 20% with zero addedlatency, and in combination with filtering by 40% subjectto a constant latency bound of 10 milliseconds.