BioMotion : A Multimodal Training Monitoring System Using IMU, EMG, and heart rate sensors

Sammanfattning

This project investigates how a multimodal embedded system can be used to monitor exercise performance in real time. Instead of relying on just one data source, the system combines data from three different sensors. The sensors were a inertial measurement unit (IMU), an electromyography (EMG) sensor and a heart rate sensor to capture different aspects of movement, muscle activity and physical effort. The aim was to determine whether a practical and low-cost prototype could provide functional real-time feedback during strength training exercises. The system was developed using an engineering-oriented and iterative approach. Each sensor was first tested separately and was then integrated into a complete system consisting of a Raspberry Pi Pico W, embedded firmware, a Flutter mobile application, and a FastAPI backend with a Random Forest classification model. The prototype detects repetitions, distinguishes between full and partial movements, and classifies repetitions as GOOD, INCOMPLETE, or UNSTABLE. Sensor data and feedback are transmitted to the mobile application using Bluetooth Low Energy. The prototype was evaluated during practical exercise sessions using four exercises. In the repetition detection evaluation, all 160 tested full and half repetitions were detected correctly. In the separate AI classification evaluation, 231 of 240 repetitions were classified correctly, corresponding to an overall classification rate of 96.25%. Connectivity and calibration checks achieved a success rate of 92.62%. The results show that the BioMotion system functioned as an integrated prototype under the tested conditions. However, the evaluation was limited to the project members, a small number of exercises, and a relatively small dataset. The results show that the prototype worked under the tested conditions but they cannot be applied to all users or training situations.