Human Activity Recognition with Wearable Sensors
Architecture
§ Data Signals [1]
Minh Nguyen, Liyue Fan, Cyrus Shahabi Integrated Media Systems Center University of Southern California
Classifier Method
Related Research § [1] O. D. Lara and M. A. Labrador. A survey on human activity recognition using wearable sensors. IEEE Communications Surveys & Tutorials 15(3), pp.1192-1209. 2013; 2012.DOI:10.1109/SURV.2012.110112. 00192. § [2] J. Parkka, J. Parkka, M. Ermes, M. Ermes, P. Korpipaa, J. Mantyjarvi, J. Peltola and I. Korhonen. Activity classification using realistic data from wearable sensors. IEEE Transactions on Information Technology in Biomedicine 10(1), pp. 119-128. 2006. . DOI: 10.1109/TITB.2005.856863.
Introduction § Motivation:
+ The exceptional development of wearable sensors/devices
+ Human interaction with the devices as part of daily living
+ Human activity data analysis + Useful healthcare services
§ Devices’ Accelerometers & Machine Learning algorithms to recognize locomotion type § Providing users with human performance status
Conclusion & Future Work
n Evaluation for each classification method: Activity Confusion/Overall Accuracy
n Application for Healthcare Informatics
IMSC Retreat 2015
Communication
Sen
sor Acceleration 3-axis accelerometer
Environmental Attribute
Light, temperature, noise, location
Physiological Signal
Heart rate, respiration rate, galvanic skin response
Data Collection
Data Preprocessing
Feature Extraction Classifier Recognition
Result
Integration Storage
Human Activity Data Signals
Data Segmentation
Mean, Standard Deviation,
Energy, etc.
Decision Tree, NB, kNN, etc.
Walk, sit, stand, lie,
etc.
Decision Tree [2] K-nearest Neighbors
Naïve Bayesian Support Vector Machine
+ Activity A - Features Xi + Independence assumption between features
+ Calculating distances between points + Finding k nearest feature points of feature points
+ Decision Tree (C4.5, ID3) + Root 1: Input (extracted features) + Node 1, 2, 3, 4, 5, 6: Based on the value of the features, the nodes decide which activity is labelled
+ SVM finds the hyperplane that maximizes the margin between the data points
Other Methods
+ Neural Networks, HMM, Fuzzy Basis Function