Human gestures have much abundant information in non-verbal human communications, so it's expected that gesture recognition is one of promising man-machine-interfaces in the near future. Especially, gesture communication has rich quantitative information which stands for the degree of gesture meaning, i.e. “very”, “a little”.
Recently, in fields of Virtual Reality and sign-language recognition, a lot of efforts regarding human motion capturing and gesture recognition have been made. Among many sensing methodologies are widely tried, image sensing has exhibited some problems due to occlusion etc., and magnetic sensing has shown some problems in magnetic disturbances etc.
In this paper, we propose a measurement of quantitative information of gestures using gyroscopes and accelerometers, which are not influenced by occlusion or magnetic disturbances potentially. Unlike position sensors, the both sensors are dynamic ones, which are superior in capturing the quantitative information of gesture. A developed gesture-sensing-unit is introduced. A quantitative and qualitative recognizing method using an artificial neural network is also proposed. In experiments with 12 kinds of gestures (Japanese sign-language), the effectiveness of the method has been verified in the recognition of both qualitative information and quantitative information.
View full abstract