Application of bidding automata for measuring the informativeness of hand motion characteristic features

Abstract

This research evaluates how informative are the feature (coordinates, velocities and accelerations) captured by 3D motion tracking system for segmentation of human hand movements. Six gestures and their transitions have been studied. To measure the informativeness of the features the bidding automaton was constructed that compares the human-defined segmentation points (“techer” reference data set) to the changes of the real feature values in neighborhood of those gesture transition/switching points. Based on the comparison certain weights, describing correlation between changes of certain feature and teacher defined switching, were assigned to each feature at each switching point.