Method
Transformer
Uses a Transformer encoder to learn walking patterns from raw sensor data. Best overall accuracy.
Detection methods
Choose a detection method
Three different approaches are available. They use the same sensor data but analyze it differently — try another if your results seem off.
Method
LSTM
Uses an LSTM network to model the sequence of motion over time. Good for varied walking speeds.
Method
FFT Centroids
Extracts frequency features from the motion signal and matches against stored centroids. Fastest inference.