Analyzing Human Gestural Motions using Acceleration Sensors

A novel approach to human gestural motion analysis and recognition on the basis of acceleration sensors is proposed. Using accelerometers for motion analysis has the advantage of very high temporal resolution (measurement rates ≥ 1 kHz are easily possible) and low instrumentation overhead. In this paper, a theoretical model for the description of gestural motion trajectories and the derivation of expected accelerations as well as the estimation of trajectories from measured acceleration data is presented. The theory is applied to measurements performed with a system of two combined triaxial accelerometers. Applying results from the theory of curves and differential geometry, features for pattern classification are extracted from the data. As an example, the nearest-neighbour classificator is used to match the features of gestural trajectories against stored templates. Preliminary trials indicate good recognition rates for a set of 9 different elementary gesture motions of varying size and speed. Also, a short overview of the interdisciplinary research project “Gesture Recognition with SensorGloves” at the Technical University of Berlin is given.