This report presents the development of a prototype system for automatic workout tracking and augmented reality (AR) feedback. The system is designed to recognise exercise type, count repetitions, and display workout statistics in real time through an AR interface. The aim is to offer a distraction and phone-free gym experience that supports improved mental health and workout performance by eliminating the need for user interaction with smartphones or tracking methods during training.

The system integrates several components: a smartphone app (Sensor Logger) to collect accelerometer and gyroscope data from the user, support vector machine (SVM) models for exercise classification, peak detection algorithms for repetition counting, and a Unity-based AR application for displaying workout statistics via a Meta Quest 2 headset. The system was evaluated using 42 test sets involving squats and pushups. Exercise classification achieved an accuracy of 97.6%, while repetition counting achieved an accuracy within ±1 repetition for 71% of squats and 29% of pushups, and within ±2 repetitions for 96% and 57% respectively. A heuristic evaluation was conducted to assess the usability of the AR user interface and concluded the UI was well designed, aesthetic and familiar, but sometimes lacked intuitive error messages.

The results demonstrate the feasibility and potential of real-time, automatic, AR-based workout tracking systems. While classification performance and the UI design was strong, repetition counting remains an area for improvement. The project provides a functional proof of concept and lays a solid foundation for future development into a novel and commercially viable product.