Classifying Human Activities with Accelerometers Using Markov Models
Author Information
Author(s): Andrea Mannini, Angelo Maria Sabatini
Primary Institution: Scuola Superiore Sant' Anna
Hypothesis
Can Hidden Markov Models (HMMs) improve the classification of human physical activities compared to Gaussian Mixture Models (GMMs)?
Conclusion
The study demonstrates that HMMs can effectively classify human physical activities by leveraging statistical information about movement dynamics.
Supporting Evidence
- HMMs incorporate movement dynamics into the classification process.
- The study used a dataset of 20 subjects performing 20 activities.
- Classification accuracy improved with the use of HMMs compared to GMMs.
- Spurious data rejection methods enhanced classification performance.
- Leave-one-out validation showed good generalization capabilities of classifiers.
Takeaway
This study shows how computers can learn to tell what people are doing, like walking or sitting, just by using sensors that measure movement.
Methodology
The study used accelerometer data from subjects performing various activities and applied Hidden Markov Models for classification.
Potential Biases
There may be biases in the data due to individual differences in how subjects perform activities.
Limitations
The study is limited by the complexity of real-life data and the potential for overfitting due to the high number of parameters in the models.
Participant Demographics
The study involved 20 subjects performing various daily activities.
Statistical Information
P-Value
p<0.05
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website