Light-based wearable sensing methods for human body motion often rely on a few single light emitters and receivers, which leads to limited sensing capabilities. While increasing the number of light sources and sensors can help detect more complex motions, this increase in hardware often degrades wearability and mobility. In this paper, we employ a flexible organic photosensor matrix surrounded by an LED array as the light source to detect subepidermal images on the back of the hand. We then use computer vision and deep learning techniques to detect patterns based on blood-related changes under the skin. Our sensor system can accurately distinguish 32 hand postures and 17 gestures in user-dependent training, showing promise for ultra-light wearable systems in natural user interface applications.
D. A. Chacon, K. Shinoda, T. Yokota and K. Yatani, Demonstrating the Feasibility of Subepidermal Image Sensing for Hand Posture and Gesture Recognition. IEEE Sensors Letters, 2022. (paper)