Human Activity Recognition — Smartphone Sensor Analytics

Team project for ELEC 292 (Instrumentation and Data Analysis). Used smartphone accelerometer data to classify motion types and visualize telemetry signals.

PythonMachine Learning Logistic RegressionGUI Signal Processing
Human Activity Recognition graph

Overview

We collected smartphone accelerometer data using Phyphox and segmented 5 s windows for walking vs jumping activities. The data were filtered and feature-extracted per axis, normalized (z-score), and stored in HDF5 for training.

Key Contributions

  • Applied moving-window filtering and extracted time-domain statistics (mean, RMS, variance).
  • Normalized each feature using z-score scaling and managed datasets in HDF5.
  • Trained logistic-regression model in Python achieving ≈96 % test accuracy.
  • Built a Tkinter GUI to visualize 3-axis accelerometer telemetry in real time.