In this paper, we propose a micro hand gesture recognition system usingultrasonic active sensing, which uses micro dynamic hand gestures within a timeinterval for classification and recognition to achieve Human-ComputerInteraction (HCI). The implemented system called Hand-Ultrasonic-Gesture (HUG)consists of ultrasonic active sensing, pulsed radar signal processing, andtime-sequence pattern recognition by machine learning. We adoptedlower-frequency (less than 1MHz) ultrasonic active sensing to obtainrange-Doppler image features, detecting micro fingers' motion at a fineresolution of range and velocity. Making use of high resolution sequentialrange-Doppler features, we propose a state transition based Hidden Markov Modelfor classification in which high dimensional features are symbolized, achievinga competitive accuracy of nearly 90% and significantly reducing the computationcomplexity and power consumption. Furthermore, we utilized the End-to-Endneural network model for classification and reached the accuracy of 96.32%.Besides offline analysis, a real-time prototype was released to verify ourmethods potential of application in the real world.
展开▼
