UltraDigit: An ultrasound signal‐based in‐air digit input system via transfer learning

Abstract

Abstract With the popularity of smart devices, people pay more attention to the experience of text input. The authors propose an in‐air digit input system, namely UltraDigit, which enables users to input digits in the air using acoustic signal via transfer learning. Different from other technologies, UltraDigit can obtain the finger movement through a speaker and a microphone built in the smartphone, which does not need the assistance of other hardware. In addition, the authors use the idea of transfer learning with a small amount of data to train classification model, which achieves good performance with low training cost (such as time cost and data cost). Experimental results show the average accuracy of digit recognition is 92.9%, which verifies the validity of the ultrasound signal‐based contactless digit input.

Keywords

• acoustic signal detection
• doppler shift
• hand gesture recognition
• image classification
• time‐frequency analysis