Abstract: With the development of mobile communication technology, the traditional Internet is migrating to the mobile Internet, and the development of intelligent wearable devices is very rapid. In particular, the smartwatch has become one of the most popular wearable devices because of its small size and ease of carrying. However, the smartwatch has an unsatisfied experience with text input. For example, keyboard input often leads to input errors because the screen is small and the number of keyboards is large. Speech recognition is vulnerable to environmental noise and privacy disclosure risks. In order to solve the above problems, a new text input method is studied, which is based on vibration perception to recognize user keystroke behavior. Design and implement a method to recognize which finger joint is tapped. Different finger joints are mapped to different keys. And then, classifying their tapping vibration signals can make up for the inconvenience of keyboard input on the small screen. Firstly, the accelerometer of LPMS-B2 module with a low sampling rate is used to collect the vibration signal generated by tapping finger joints. Then, design a algorithm to process the signal and extract features from the time-frequency map. Finally, use the Naive Bayes classification to recognize the tapped finger joint. The experimental results show that the accuracy of three classifications of finger joints is more than 90%, and the accuracy of five classifications is about 75%.

Key words: smartwatch, vibration perception, key recognition, time-frequency map, the Naive Bayes classification