#A Tutorial on Human Activity Recognition Using Body-worn Inertial Sensors
MATLAB toolbox for the publication
A Tutorial on Human Activity Recognition Using Body-worn Inertial Sensors
Andreas Bulling, Ulf Blanke and Bernt Schiele
ACM Computing Surveys 46, 3, Article 33 (January 2014), 33 pages
DOI: http://dx.doi.org/10.1145/2499621
@article{bulling14_csur,
title = {A Tutorial on Human Activity Recognition Using Body-worn Inertial Sensors},
author = {Andreas Bulling and Ulf Blanke and Bernt Schiele},
url = {https://github.com/andyknownasabu/ActRecTut},
issn = {0360-0300},
doi = {10.1145/2499621},
year = {2014},
journal = {ACM Computing Surveys},
volume = {46},
number = {3},
pages = {33:1-33:33},
abstract = {The last 20 years have seen an ever increasing research activity in the field of human activity recognition. With activity recognition having considerably matured so did the number of challenges in designing, implementing and evaluating activity recognition systems. This tutorial aims to provide a comprehensive hands-on introduction for newcomers to the field of human activity recognition. It specifically focuses on activity recognition using on-body inertial sensors. We first discuss the key research challenges that human activity recognition shares with general pattern recognition and identify those challenges that are specific to human activity recognition. We then describe the concept of an activity recognition chain (ARC) as a general-purpose framework for designing and evaluating activity recognition systems. We detail each component of the framework, provide references to related research and introduce the best practise methods developed by the activity recognition research community. We conclude with the educational example problem of recognising different hand gestures from inertial sensors attached to the upper and lower arm. We illustrate how each component of this framework can be implemented for this specific activity recognition problem and demonstrate how different implementations compare and how they impact overall recognition performance.},
keywords = {}
}
If you find the toolbox useful for your research please cite the above paper, thanks!
Version 1.4, 19 August 2014
-
The data should be arranged in a MATLAB matrix with rows denoting the frames (samples) and columns denoting the different sensors or axes -> matrix NxM (N: frames, M: sensors/axes) IMPORTANT: make sure the matrix does not contain any timestamp columns as often added by data recording toolboxes, such as the Context Recognition Network Toolbox
-
The ground truth labels should be integers, arranged in a MATLAB vector with rows denoting the frames -> vector Nx1 (N: frames)
-
The data matrix should be loaded into the variable
data
, the ground truth label vector into the variablelabels
-
The NULL class needs to have label 1, the remaining classes labels 2:n
-
If you want to modify the default parameters of the different classifiers have a look at
setClassifier.m
-
This toolbox requires the following MATLAB toolboxes:
-
To compile the different third-party libraries have a look at the documentation
Execute run_experiments_paper.m
in MATLAB
-
Have a look at
settings.m
This file contains all settings available in the toolbox and their defaults. All settings are stored in a MATLAB structSETTINGS
. Set the different fields in this struct according to the requirements of your planned experiment. -
Have a look at
Experiments/expTutorial.m
and run the script This file contains a (simple) example structure of an experiment. Note howsettings.m
is executed first, followed by modifications to theSETTINGS
fields.optional: Install all third-party libraries you plan to use (see list below). Archives of all supported libraries are provided in the subdirectory "Libraries". The libraries should be installed in the same directory. If you prefer to install the libraries in a different path, adapt the library paths in
settings.m
accordingly (line 33 and following) -
To create your own experiment
-
Copy
Experiments/expTutorial.m
toExperiments/expOwn.m
-
Write code in
expOwn.m to
modifySETTINGS
according to your experiment's requirements, in particular:
SETTINGS.CLASSIFIER (default: 'knnVoting')
SETTINGS.FEATURE_SELECTION (default: 'none')
SETTINGS.FEATURE_TYPE (default: 'VerySimple')
SETTINGS.EVALUATION (default: 'pd')
SETTINGS.SAMPLINGRATE (in Hz, default: 32)
SETTINGS.SUBJECT (default: 1)
SETTINGS.SUBJECT_TOTAL (default: 2)
SETTINGS.DATASET (default: 'gesture')
SETTINGS.CLASSLABELS (default: {'NULL', 'Open window', 'Drink', 'Water plant',
'Close window', 'Cut', 'Chop', 'Stir', 'Book', 'Forehand', 'Backhand', 'Smash'})
SETTINGS.SENSOR_PLACEMENT (default: {'Right hand', 'Right lower arm', 'Right upper arm'})
SETTINGS.FOLDS (default: 26)
SETTINGS.SENSORS_AVAILABLE = {'acc_1_x', 'acc_1_y', 'acc_1_z', ...
'gyr_1_x', 'gyr_1_y', ...
'acc_2_x', 'acc_2_y', 'acc_2_z', ...
'gyr_2_x', 'gyr_2_y', ...
'acc_3_x', 'acc_3_y', 'acc_3_z', ...
'gyr_3_x', 'gyr_3_y'};
SETTINGS.SENSORS_USED (default: {'acc_1', 'acc_2', 'acc_3', 'gyr_1', 'gyr_2', 'gyr_3'})
-
Change the
EXPERIMENT_NAME
andIDENTIFIER_NAME
variables inexpOwn.m
For example,EXPERIMENT_NAME
could be set to 'kNN' andIDENTIFIER_NAME
to 'k_5' if your experiment involves using a kNN classifier with k fixed to 5. -
Put your data files in subdirectories of "Data" named according to the scheme: subjectX_Y
- X denotes the index of the subject (
1:SETTINGS.SUBJECT_TOTAL
) - Y denotes the type of dataset (
SETTINGS.DATASET
plus additional ones) For example, the toolbox datasets are stored in the following subdirectories: subject1_walk, subject1_gesture, subject2_walk, subject2_gesture The data files should be called "data.mat" and should contain both variablesdata
andlabels
- X denotes the index of the subject (
-
Run
expOwn.m
and wait for the script to finish. Extracted features will be saved in "Output/features" whereas the experiment output will be saved in "Output/experiments/EXPERIMENT_NAME/IDENTIFIER_NAME"
-
libSVM
URL: http://www.csie.ntu.edu.tw/~cjlin/libsvm/ -
liblinear
URL: http://www.csie.ntu.edu.tw/~cjlin/liblinear/ -
mRMR
URL: http://penglab.janelia.org/proj/mRMR/ -
SVMlight
URL: http://svmlight.joachims.org/ -
jointboosting by Christian Wojek
URL: none -
HMM Toolbox for MATLAB by Kevin Murphy
URL: http://www.cs.ubc.ca/~murphyk/Software/HMM/hmm.html -
Performance evaluation scripts by Jamie Ward
URL: http://www.jamieward.net/research/performance/