Pirouette Recognition LSTM
IMU time-series classification with a small recurrent network.
| Where | UC San Diego, under Professor Phuong Truong |
|---|---|
| When | 2025 |
| Tools | MATLAB, Deep Learning Toolbox |
| Data | Public IMU dataset, about 85 labelled turns across six dancers |
| Result | Over 95% classification accuracy |
I worked from a public IMU dataset of about 85 labelled turns across six dancers, backing out joint torques from inertia estimates to run alongside the raw time series. I trained an LSTM to classify turn type from the combined signal.
It reached over 95% accuracy, though with a dataset that small the number says more about fitting six dancers than about generalising to new ones. Written up under Professor Phuong Truong’s guidance.