Device sensors provide real-time insights into what people are doing (walking, running, driving, etc.).
Knowing users’ activity allows, for instance, to interact with them through an app.
You can apply machine learning to detect activities by reading and processing sensor data in this regard.
Contents
Objectives
Human activity recognition (HAR) aims to classify a person’s actions from a series of measurements captured by sensors.
Nowadays, collecting this type of data is not an arduous task. With the growth of the Internet of Things, almost everyone has some gadget that monitors their movements. It can be a smartwatch, a pulsometer, or even a smartphone.
Typically, this is accomplished using a fixed-length sliding window approach for feature extraction. Here, two parameters need to be fixed: the window size and the shift.
These are some of the data you could use:
- Body acceleration.
- Gravity acceleration.
- Body angular speed.
- Body angular acceleration.
- Etc.
The machine learning model used for activity recognition relies on top of the devices’ available sensors.
However, analyzing this data can be a big challenge. Indeed, human activities are complex, and individual differences exist.
Benefits
Activity recognition is the basis for the development of many potential applications in health, wellness, or sports:
MONITOR HEALTH
Analyze a person’s activity based on information collected from various devices.
DISCOVER ACTIVITY PATTERNS
Discover which variables determine which activity a person is doing.
DETECT ACTIVITY
Develop a predictive model that can identify a person’s activity based on the signals received by the sensors.
IMPROVE WELLBEING
Design individualized exercise tables to enhance a person’s health.
Approach
Neural networks are the ideal algorithms for determining a person’s physical activity. This is due to their ability to recognize the patterns behind the data.
The following graph illustrates a neural network that classifies different activities using smartphone data.
Conclusions
Human activity recognition has a wide range of applications due to its impact on overall well-being.
It is becoming a fundamental tool in healthcare solutions such as preventing obesity or caring for elderly persons.
Data sets
Further reading
- Deep learning for sensor-based activity recognition: A survey.
- Chernbumroong, S., Cang, S., Atkins, A., & Yu, H. (2013). Elderly activities recognition and classification for applications in assisted living. Expert Systems with Applications, 40(5), 1662-1674.
- Anguita, D., Ghio, A., Oneto, L., Parra, X., & Reyes-Ortiz, J. L. (2013, April). A public domain dataset for human activity recognition using smartphones. In Esann.
- Mannini, A., & Sabatini, A. M. (2010). Machine learning methods for classifying human physical activity from on-body accelerometers. Sensors, 10(2), 1154-1175.
