GetFit: A Teachable, Open-Source Fitness Tracker Using On-Device Machine Learning

Accurate and effective exercise tracking is essential for tracking one's own health.By combining embedded hard ware with on-device machine learning, this study presents an intelligent, affordable solution for real-time fitness tracking.The suggested system makes use of an Arduino Nano 33 BLE Sense microcontroller, which uses its built-in 9-axis IMU to continuously monitor user movements.The high variability in user-specific motions, which conventional rule-based programming cannot handle, is a major problem in fitness tracking.Bluetooth Low Energy (BLE) is used for control and data collection, and it is controlled by a custom web application.The Edge Impulse platform is used to train a Long Short-Term Memory (LSTM) or similar neural network in order to achieve high accuracy and provide a "teachable" system.To identify and count an infinite number of exercises, this model is directly applied to the microcontroller (on-device inference).For daily and weekly calorie burn estimates, the system sends exercise counts and personal information to a Firebase real-time database.According to test results, an open-source, functional system that can be customized for new exercises is a big improvement over static, pre-programmed fitness trackers.