OSense

Abstract

Observing that, how we grasp objects is highly correlated with geometric shapes and interactions, we propose the use of hand postures and motions as an indirect source of inputs for object-activity recognition. This paradigm treats the human hand as an always-available sensor, and transforms all sensing problems to the data analysis for the "sensor hand". We envision this paradigm to be generalizable for all objects regardless of whether they are acoustically or electromagnetically active, and that it detects different motions while holding the same object. Our proof-of-concept setup consists of six IMU sensors mounted on the fingers and back of the hand. Our experiments show that when the posture is combined with the motion, the personalized object-activity detection accuracy increases from 80% to 87%.