Real-time Human Activity Classification with IMU Sensor

This project explored using an IMU sensor to classify human activities (walking, running, jumping) in real-time and was completed for ENGR 845: Neural Machine Interface at San Francisco State University.

• We collected sensor data during these activities and processed it in Matlab.

• Feature extraction involved calculating statistics from the acceleration signal.

• Ensemble classifiers (bagged trees, boosted trees, random forests) were trained with the Matlab classification learner app to categorize the data.

• A bagged tree ensemble achieved the highest accuracy (95-99%) and was implemented in a Simulink model for real-time classification.

• The model showed promising results but limitations were identified:

  • Occasional misclassification during testing suggests room for improvement.

  • A trade-off exists between classification accuracy (favored by larger window lengths) and processing speed in Simulink.

This project demonstrates the feasibility of using IMU sensors and machine learning for real-time human activity classification, highlighting areas for further development.