Near-Field Radio-Frequency for Real-Time Facial Expression Recognition

The advent of smart wearable computing platforms in the form of VR/AR enhanced devices has unlocked new strides in mental wellness (through tele-health counseling), personalized spatial computing, and interactive content consumption. However, most current devices rely upon vision and audio based user inputs that pose privacy risks and often consume high amounts of power. In this work, we introduce a novel near-field radio-frequency (NFRF) based facial expression recognition (FER) system that is personal, privacy-respecting, and may be extended to commercially available wearable form-factors. The proposed system demonstrated high discriminability across common facial expressions, including reciprocal expressions occurring across individual facial hemispheres, achieving a test accuracy of 83.01% with sub-millisecond inference time, making it a suitable candidate for wearable computing as well as innovative healthcare devices, such as for objective measurement of post-stroke palsy rehabilitation.