Abstract
Portable text entry systems on mobile computing devices have been increasingly demanded to facilitate daily use. Unfortunately, existing virtual keyboards struggle to trade off text input accuracy and user-friendliness well. Thus, inspired by the well-accepted and oft-used QWERTY layout, a novel IMU-based virtual keyboard system is proposed to achieve efficient and user-friendly text input. In this system, an exquisite data glove integrated with six IMU sensors is designed to capture hand motion signals during keystrokes. With the aim of predicting the corresponding typing outputs, a hand gesture decoding algorithm called VK-GR is proposed, which can effectively handle noise interference and extract spatial details as well as sequential features. To support the training and validation of the algorithm, a keystroke gesture dataset from 13 individuals is collected. Experiments show that the proposed system enables users to type on mobile devices with limited learning costs, achieving an offline accuracy of 95.61% and an online accuracy of 78.3%, along with a real-time input speed of 29.99 WPM.