A Real-Time Wearable Virtual Keyboard Based on Inertial Finger Gesture Recognition

Kurzfassung

With the increasing demand for hands-free and portable text input methods in AR/VR and assistive technologies, traditional keyboards are becoming less practical. This paper presents a methodology for the implementation of a virtual keyboard using MEMS sensors. A set of MPU6050 sensors mounted on finger rings is used to capture motion data, which is processed using a complementary filter to obtain stable orientation features. A lightweight Support Vector Machine (SVM) classifier is deployed on an ESP32 microcontroller using a lightweight C implementation generated with the micromlgen library for real-time keystroke recognition. The system achieves a classification accuracy of 73.5% on unseen users with an average prediction latency of 0.000015 seconds. Additionally, a host-side LSTM-based word prediction module enhances typing efficiency. The proposed system provides a portable, low-cost, and hands free solution for text input, with potential applications in assistive technologies and AR/VR environments.