Related Experiment Video
Updated: Dec 29, 2025
06:49
Evaluation of a Smartphone-based Human Activity Recognition System in a Daily Living Environment
Published on: December 11, 2015
9.3K
REAL-Time Smartphone Activity Classification Using Inertial Sensors-Recognition of Scrolling, Typing, and Watching
Sijie Zhuo1, Lucas Sherlock1, Gillian Dobbie2
1Department of Electrical, Computer and Software Engineering, University of Auckland, Auckland 1010, New Zealand.
Sensors (Basel, Switzerland)
|January 30, 2020
Summary
This study shows that smartphone motion sensors can recognize user activities like scrolling and typing with 78.6% accuracy. This opens possibilities for personalized mobile applications using inertial measurement unit (IMU) data.
Area of Science:
- Human-Computer Interaction
- Mobile Computing
- Sensor Technology
Background:
- Current smartphone applications lack real-time awareness of user activities due to privacy restrictions.
- Personalization features are limited because direct access to user actions like typing or scrolling is not feasible.
- Internal movement sensors on smartphones offer a potential, albeit policy-sensitive, avenue for inferring user behavior.
Purpose of the Study:
- To investigate the feasibility of using smartphone inertial measurement unit (IMU) sensors to classify common user activities.
- To determine if sensor data can enable real-time recognition of smartphone interactions.
- To explore the potential for developing context-aware mobile applications based on activity recognition.
Main Methods:
- A study was conducted with human participants using an Android application to collect motion data.
- Data collection included activities such as scrolling, typing, and video watching, in both seated and walking conditions, alongside non-use baselines.
- Machine learning models were trained using data from triaxial accelerometers, gyroscopes, and magnetometers to classify eight distinct states.
Main Results:
- An optimal machine learning model achieved 78.6% accuracy in recognizing smartphone activities.
- The best performance was obtained using the Extremely Randomized Trees algorithm with data sampled at 50 Hz and analyzed in 5-second windows.
- The study demonstrated the potential of IMU sensor data for classifying distinct user interactions with smartphones.
Conclusions:
- Inertial measurement unit (IMU) sensors show viability for recognizing common smartphone activities in real-time.
- The findings suggest a pathway for developing personalized mobile experiences by inferring user actions from sensor data.
- Further research can explore refining algorithms and sensor fusion for improved accuracy and broader application.
