From Lens to Prism: Device-Free Modeling and Recognition of Multi-Part Activities

Abstract

Exploiting radio frequency signals with different physical characteristics of refraction and reflection of different parts and movements of the human body, most of the existing work focuses on the identification of single activity from a special part, and the elimination of mutual interference in different activities recognition of various parts of the body is deliberately diluted. A reliable, real-time, and non-invasive human body combination activities recognition system, namely, WiCoach, is proposed in this paper. WiCoach extends the possibility of utilizing the concept of sensorless sensing and the effective use of wireless signal side channels. Leveraging well-designed signal features and robust denoising methods, WiCoach models the speed and duration between channel state information and the body's multiple concurrent activities, and then recognizes activity of different parts and guides the individual fitness. The experimental results show that WiCoach can effectively discriminate the activities of different parts of the human body, and no longer relies on modeling the signal characteristics of single part, compared with the current work.