Step Visualizer

A Flutter project to visualize raw sensor data and predict step counts for the left and right foot. Step predictions are generated using a combination of signal processing (Butterworth low-pass filtering and SciPy peak detection) and machine learning models.

Overview

• Prediction Models:

  • Sequential Models (Time-Series):
    • UnifiedCountLSTM: 2-layer LSTM with dropout and ReLU.
    • SimpleLSTM: Single-layer LSTM for simplicity.
    • CNN-LSTM: 1D CNN (with max pooling) + LSTM to capture local and temporal features.
    • UnifiedCountGRU: GRU-based model with faster convergence.
  • Aggregated-Feature Models:
    • Random Forest Regressor
    • XGBoost Regressor

• Comparison:
  Learning curves and MSE for the sequential models (displayed in a 2×2 grid) and error metrics (MAE) for the aggregated models were used to select the best approach.

• Data Processing:
  Raw data is preprocessed by standardizing timestamps, filtering the acceleration magnitude using a Butterworth low-pass filter, and extracting key features. Peak detection on the filtered signal yields surrogate step labels for training.

How to Run

1. Clean and fetch Flutter dependencies:

   flutter clean
   flutter pub get

2. Run the project on Chrome:

   flutter run -d chrome

For detailed code, see stepcount.ipynb and for full documentation, refer to StepCountEversion.pdf.