Real-time workout classification using Arduino Nano 33 BLE Sense, Edge Impulse, and Web Bluetooth.
The Smart Dumbbell can classify four types of movements namely Idle, Biceps Curl, Lateral Raise, and Overhead Press. A confidence score is also displayed showing the model's confidence in the prediction. The only hardware required is an Arduino Nano 33 BLE Sense and a battery attached to a dumbbell. For more details, see the documentation.
The data collection, model training and testing was done using Edge Impulse Studio. The Arduino code relies on the smartbell_inferencing library which contains the TinyML model. The library and other project data can be accessed here.
The idea for the Smart Dumbbell came from this project.