Almende Continuous Real-world Activities with ...

The data set is originally used and collected for the Master's thesis Temporal Segmentation using Support Vector Machines in the context of Human Activity Recognition by Roemer Vlasveld.

DESCRIPTION

This data set contains recording of performed human activities in a real-world setting. The subjects wore a smartphone (in the right front pocket), which was tracking the inertial sensors with an Android application. Among those sensors are

• Accelerometer
• Magnetic Field
• Gyroscope

The subjects were recorded with a video camera during the performance.

The video recordings were used to annotate activity change points in the inertial data streams. They can be consulted in case of ambiguity.

The focus of this data set is on the change points between activities. Although an activity labeling is provided, we advise to not use it directly for activity classification.

CONTENTS

This package contains the following directories:

• /data contains for each run the sensor data and plots. It also provides manually annotated change points and activity labels
• /tools contains some Ruby and Octave/Matlab scripts. These scripts can be used to process (new) recorded data (with the Sensor Logger application) in the current format.
• /video_recordings contains the video files from the performed activity.

Each directory contains a README file with further explanation of the content.

CHANGE DETECTION versus ACTIVITY CLASSIFICATION

This data set is originally used to detect changes between performing activities. This was achieved using a One-Class Support Vector Machine classifier. The implementation is publicly available here. Such algorithm requires continuous recorded data, so a transition period between activities is present. For that purpose, this data set was collected.

Because the nature of the research was to find change points, the manually labeling has focused on finding these and is listed in data/run-*/change_points. We also provide a set of activity labels (which can be found in `data/run-*/labels), but these are only used to give information about the change points.

Since the data is recorded in a real-world setting, the activity labels are not defined in a very strict manner. Furthermore, each segment contains transition activity data, so care need to be taken when this set is used for direct activity classification.

REFERENCE THIS SET

Please use the following Bibtex entry to reference this data set:

@MISC{vlasveld2014acras, author = {Vlasveld, R.Q.}, title = {Almende Continous Real-world Activities with on-body smartphone Sensors Data Set}, howpublished = {\url{http://datasets.almende.com}}, year = {2014} }

The original thesis can be referenced with

@mastersthesis {vlasveld2014temporal, title = {Temporal Segmentation using Support Vector Machines in the context of Human Activity Recognition}, year = {2014}, month = {February}, school = {Universiteit Utrecht}, type = {mastersthesis}, address = {Utrecht, the Netherlands}, author = {Vlasveld, Roemer Q} }%