iHand: Hand Recognition-Based Text Input Method for Wearable Devices

Abstract

Text input using hand gestures is an essential component of human–computer interaction technology, providing users with a more natural and enriching interaction experience. Nevertheless, the current gesture input methods have a variety of issues, including a high learning cost for users, poor input performance, and reliance on hardware. To solve these problems and better meet the interaction requirements, a hand recognition-based text input method called iHand is proposed in this paper. In iHand, a two-branch hand recognition algorithm combining a landmark model and a lightweight convolutional neural network is used. The landmark model is used as the backbone network to extract hand landmarks, and then an optimized classification head, which can preserve the space structure of landmarks, is designed to classify gestures. When the landmark model fails to extract hand landmarks, a lightweight convolutional neural network is employed for classification. Regarding the way letters are entered, to reduce the learning cost, the sequence of letters is mapped as a two-dimensional layout, and users can type with seven simple hand gestures. Experimental results on the public datasets show that the proposed hand recognition algorithm achieves high robustness compared to state-of-the-art approaches. Furthermore, we tested the performance of users’ initial use of iHand for text input. The results showed that the iHand’s average input speed was 5.6 words per minute, with the average input error rate was only 1.79%.

Keywords

• wearable device
• hand recognition
• text input method
• human–computer interaction