LAUSR

This paper presents an effective signal processing scheme of hand gesture recognition with a superior accuracy rate of judging identical and dissimilar hand gestures. This scheme is implemented with the air sonar possessing a pair of cost-effective ultrasonic emitter and receiver along with signal processing circuitry. Through the circuitry, the Doppler signals of hand gestures are obtained and processed with the developed algorithm for recognition. Four different hand gestures of push motion, wrist motion from flexion to extension, pinch out, and hand rotation are investigated. To judge the starting time of hand gesture occurrence, the technique based on continuous short-period analysis is proposed. It could identify the starting time of the hand gesture with small-scale motion and avoid faulty judgment while no hand in front of the sonar. Fusing the short-time Fourier transform spectrogram of hand gesture to the image processing techniques of corner feature detection, feature descriptors, and Hamming-distance matching are the first-time, to our knowledge, employed to recognize hand gestures. The results show that the number of matching points is an effective parameter for classifying hand gestures. Based on the experimental data, the proposed scheme could achieve an accuracy rate of 99.8% for the hand gesture recognition.