RIS-Assisted mmWave Radar for Robust Hand Gesture Recognition Using Lightweight CNNs

Identificador:  imarina:9470731

Handle: https://hdl.handle.net/20.500.11797/imarina9470731

Autores:  Morabet F; Lazaro A; Lazaro M; Villarino R; Girbau D

Resumen:
Accurate and robust hand gesture recognition is a key enabler of intuitive, contactless human–machine interaction in applications such as healthcare, smart environments, automotive systems, and wearable electronics. While radar-based methods offer significant advantages over vision- and contact-based systems, conventional approaches that rely on Doppler or phase information are often limited by multipath interference, environmental clutter, and user variability. This paper presents a novel amplitude-domain sensing framework that integrates a 24 GHz continuous-wave (CW) radar with modulated frequency-selective surfaces (FSS) to achieve reliable hand gesture recognition without the need for Doppler or phase processing. Each FSS panel introduces a distinct modulation pattern, allowing spatial encoding of gesture-induced occlusions as temporally varying changes in the radar return signal. Rather than directly tracking hand motion, the system infers gestures by identifying attenuation patterns in the modulated reflections, which are transformed into time–frequency spectrograms and classified using a lightweight convolutional neural network optimized for real-time embedded inference. The system was evaluated using eight dynamic hand gestures, including swiping, zigzag, and circular motions. The findings demonstrate a peak classification accuracy of 97% over an interaction range of 0.5 m to 1.5 m.