FMCW Radar-Based In-Air Alphanumeric Gesture Recognition With Machine Learning

The rapid advancement in computing devices and their integration into daily lives is constantly increasing the importance of natural human-computer interfaces (HCIs). In recent years, in-air writing gesture recognition using radars has gained substantial attention. Given that several alphabet and digit patterns are highly similar, existing studies perform alphabet and number recognition separately, often by using multiple radars. Unlike existing studies, this study develops a new framework to recognize 43 gestures, including 36 alphanumerics and seven special characters, using a single noncontact frequency-modulated continuous-wave (FMCW) radar. Hand movement is tracked using the range, Doppler, and angle information extracted using the FMCW radar to form a drawing pattern that serves as an input to a ShuffleNet-based deep learning model. Data from 14 participants are collected from three locations for performance evaluation. The system achieves a promising accuracy of 93.1%, validating its reliability and efficiency in real-world settings.