WiGym: Exercise recognition and counting using Wi-Fi signals

In this paper, we propose a Wi-Fi based passive tracking system WiGym, which recognizes and counts the fitness exercises performed by the user. The body movements change the multi paths reflected against the body, and thus generate the unique pattern in CSI time series. To count the number of exercises, short time energy is calculated. To extract features from CSI time series, discrete wavelet transform is performed. We implemented WiGym using commercial Wi-Fi devices and conducted extensive experiments to verify its performance. Our evaluation shows that WiGym achieves 100% recognition accuracy and 95.6% segmentation accuracy. Compared with previous on-body sensor based fitness tracking systems, it is more ubiquitous and requires no hardware modifications.