ETH - IfE-Wearable Computing - Hierarchical Context Recognition Dataset

A key part of the dataset are the smart objects, objects that sense their motions and recognize manipulations on them. The recognized events are transmitted to the network for fusion.

Scissors with integrated Tmote Mini and acceleration sensor.

Two test subjects collaboratively assembling a shelf.

This site presents a dataset consisting of sensor data collected from various sensors worn on the body and integrated in tools used by people in an ambient intelligence environment. The dataset is meant to be used as a benchmark for activity classification algorithms working at different layers:

Classification of atomic activities on sensor nodes. This generates a local view of what is happening.
Fusion of the classification results from multiple sensor nodes available in the network.
Inference larger scale activities from the fused atomic activity recognition taking time into account

The dataset consists of 8 different scenarios measured by 12 wireless sensors with 5 sensing modalities. The activities are made up of 64 atomic activities which make up 17 composed activities.

More details on the dataset are given in the publications listed below. Please reference one of them in case you publish results based on this dataset.

Plot of which atomic activities occur during which scenarios

Dataset Structure

All data is stored in CSV files for easy opening. Every recording session has been stored into an individual directory featuring the following files:

readme.txt	Contains a short description and comments about the recording session.
sensordev.csv	Signal names of the data columns presented in data.csv. These names are given in a separate file for ease of loading. For each signal, the sensor name (location) is used as prefix, and the signal as suffix. The suffixes for different sensor types are: Acceleration sensors: x,y,z for the axes, :m for the magnitude of the three axes. Light sensors: value for the light value Glove sensor: x,y,z for the acceleration axes, thumb,index,middle,ring,small for the bend sensors. Computer: mouse for the number of mouse events within a second, keys for the number of key presses within a second. Pyroelectric infrared motion sensor (PIR): value for the reading value
data.csv	The values of the sensor signals. All values have been synchronized and resampled to 50Hz. The columns are organized as described by sensordev.csv, each row corresponds to one sample point.
groundtruth.csv	Ground truth labels indicating start and end sample (or row within data.csv), activity identifier, and confidence (always 1).
labels_*.csv	Recognized labels for each individual sensor. Each row stats the start and end sample (or row within data.csv), activity identifier, and confidence (between 0 to 1, greater meaning more confident).
class_definitions.txt	This file is provided once for the whole dataset. It contains the mapping between activity identifier and human readable description. Three types of activities are available: atomic activities: basic gestures performed by hand or the tools composite activities: sequences of atomic activities, describing a more abstract or long-term activity. Macro activities: describes the settings recorded and contains a sequence of composite activities.

Download

Download the resampled sensor data, groundtruth labels, and sensor recognition results from here.

The download includes all data from subject 1. In case you would like to obtain more data, please contact Clemens Lombriser.

Details on how the recognition results are described in a paper which is currently under review. Please by patient to wait for those results.

Publications

Experiences with Experiments in Ambient Intelligence Environments
Piero Zappi, Clemens Lombriser, Elisabetta Farella, Luca Benini and Gerhard Tröster
in: IADIS International Conference Wireless Applications and Computing, 2009
[ BiBTeX RIS PDF ]

Links

UCI Machine Learning Repository
PlaceLab datasets
Dataset of Fine-Grained Activity Recognition by Aggregating Abstract Object Usage ISWC 2005"
Datasets for Activity Recognition
BoxLab - List of Home Datasets
Education Sensor Kit for Wearable Computing. Includes wireless sensors as well as software to collect datasets for context recognition.

Wichtiger Hinweis:
Diese Website wird in älteren Versionen von Netscape ohne graphische Elemente dargestellt. Die Funktionalität der Website ist aber trotzdem gewährleistet. Wenn Sie diese Website regelmässig benutzen, empfehlen wir Ihnen, auf Ihrem Computer einen aktuellen Browser zu installieren. Weitere Informationen finden Sie auf
folgender Seite.

Important Note:
The content in this site is accessible to any browser or Internet device, however, some graphics will display correctly only in the newer versions of Netscape. To get the most out of our site we suggest you upgrade to a newer browser.
More information