Real-time recognizing human hand gestures

Abstract

The development of a system for classifying and interpreting human hands motion is considered in this paper. This is obtained by locally approximating motion data with rank-1 structures. The approximation is obtained in two steps: first the time series is decomposed into simpler sub-series (segmentation), then each subseries labelled by a unique vector. The effectiveness of the proposed strategy is shown on sensory data from a data-glove when a human picks a tin can and a pencil. The strategy proves to be simple and reliable, even in the presence of unknown data corrupted by noise, and can be used as a basis for real-time automated recognition and interpretation of human gesture. © Springer-Verlag Berlin Heidelberg 2012.