2015
MAS 2015 Proceeding
An improved conversive hidden non-markovian model-based touch gesture recognition system with automatic model creation
Authors: Tim Dittmar, Claudia Krull, Graham Horton
Abstract
Mobile devices like smartphones and tablets that are controlled via a multi-touch interface have become ubiquitous. In previous work a touch gesture recognition system based on Conversive Hidden non- Markovian models has been proposed that is able to recognize similar gestures with different execution speeds based on recorded examples. With this work, we improved the system by eliminating the major drawback of manually and tediously creating models for every gesture from recorded training data. To achieve this, the gesture model design has been adapted to include an additional structure that represents a map of all known gesture examples. Experiments conducted on two different datasets show that the new system can distinguish gestures with different speeds with good accuracy and fast detection times. Ideas to further improve the system are discussed and we believe that such a system could be the basis for a new gesture authentication system in the future.
I3M Scientific Sponsors
I3M Industrial Sponsors
I3M Media Sponsors
