Gesture Classification with Hierarchically Structured Recurrent Self-Organizing Maps

New input devices need clever algorithms to process input information. We constructed a hierarchically structured neural network assembly based on recurrent self-organizing maps which is able to process and to classify motion data. We derived motion data using a so called Gesture Cube (M. Kranz et al., 2006), a cubic tangible user interface developed for one-handed control of media appliances in a home environment. This previously recorded data was automatically pre-processed by our biologically inspired neural network and classified by a improved k-nearest neighborhood classifier. In this paper we shortly describe the platform used for data acquisition but focus on the novel algorithms used for classification.