​Real-time Fitness Assistance via WiFi

Project Description

Workers cannot dedicate appropriate time during the day to travel to dedicated exercise places. Instead, they perform their exercise in an office/home environment. Unfortunately, it is difficult to analyze their form while doing their exercises without incurring the significant cost of a personal trainer or the discomfort of smart sensors on their person. Our solution is a device-free personalized fitness assistant that analyzes the channel state information of existing WiFi infrastructure.

When completed, our system will differentiate individuals when they are performing an exercise and assess the workout in real time. To detect individuals, we plan on using a deep neural network (DNN) with two layers: one to differentiate between exercises, and one deeper layer to differentiate individuals. To assess the workout, we will analyze the workout quality and provide a workout review for individuals to improve their exercises.

Tools

We used a TP-LINK router with 2.4GHz and 5GHz frequencies. We used a Dell Laptop with an Ubuntu 14.02 kernel and an Intel WiFi Wireless Link 5300 MIMO radio, also known as the IWL5300. We used the Linux 802.11n CSI tool (created by Daniel Halperin, Wenjun Hu, Anmol Sheth, and David Wetherall), implemented in Matlab and C, to extract the CSI from the channel measurements. For the two-layer deep neural network, we used Python and Tensorflow.

People

Justin Esposito Electrical and Computer Engineering Class of 2022 Rutgers University

Sachin Mathew Electrical and Computer Engineering Class of 2022 Rutgers University

Amit Patel Electrical and Computer Engineering Class of 2022 Rutgers University