ETH - IfE-Wearable Computing

News | People | Showcase

Contact | Sitemap | Help

Research | Education | Resources

Open Positions

ETH Zurich - D-ITET - IfE - Wearable Computing - Resources - Educational and research kit for activity recognition from on-body sensors

Resources

Educational and research kit for activity recognition from on-body sensors

Activity recognition datasets

Context awareness and information processing in opportunistic ubiquitous systems

Educational and research kit for activity recognition from on-body sensors

Media

Electronics Lab

Job Links

Open Position »»

What is it?

The initial motivation for this project is to support hands-on teaching and experience of real-time activity or gesture recognition from on-body sensors within a 2-hours lab session with undergraduates. In particular, it focuses on the trade-offs related to the signal processing and machine learning aspect in the design of activity recognition systems.

Besides, the kit supports data acquisition from various on-body sensors for research purposes.

The kit contains sensors, educational and support software. Despite forming an integrated system the kit components remains generic so that the kit can be used for other purposes.

Open source hardware and software

This kit is available with all schematics and source code for use by third parties. The only requirement is that work using this kit and work deriving new components from this kit acknowledge this project by citing the following publication:

D. Roggen, M. Bächlin, J. Schumm, T. Holleczek, C. Lombriser, G. Tröster, L. Widmer, D. Majoe, J. Gutknecht. "An educational and research kit for activity and context recognition from on-body sensors", Proc. IEEE Int. Conf. on Body Sensor Networks (BSN). 277--282, 2010

Hardware

Basics

The hardware components include a set of Bluetooth-based wireless sensor nodes:

NTMotion:AccGyro: An 3-axis acceleration and 2-axis rate of turn sensor
NTECG: A 2 electrode ECG sensor (for heart rate detection)
NTKBD: Annotation keypad
NTProto: prototyping node.

In addition, a charging/programming station is available.

All nodes share these characteristics:

ATMEL ATmega32a micro-controller
AmberWireless BlueNiceCom bluetooth module
Slide-in connector to program/charge the node via
6+h battery life with a 300mAh battery, continuous data acquisition
User configurable sample rate, data channels, etc.
Identical firmware code-base

Sensor nodes

NTMotion:AccGyro

NTECG

Motion sensor for activity recognition from on-body sensors (e.g. for HCI)

Characteristics:

ADXL330 3-axis accelerometer
IDG650 2-axis gyroscope

ECG sensor for heart rate monitor (e.g. for sports application).

Characteristics:

2 electrodes
Gain ~400 (fc=35Hz)
Optional 8-th order hardware low-pass elliptic filter (fc=35Hz)

Only suitable for heart rate monitoring (due to the analog / low-pass filtering characteristics)

NTKBD

NTPROTO

Annotation keypad, to mark events of relevance during a recording.
Characteristics:

12 keys mapping to 12 different labels

Customizable sensor node derived from the NTKBD node (without the keypad attached).
Characteristics:

Space for custom user circuit
4 analog inputs or digital I/O
3 digital I/O ports

Charging station

Characteristics:

5 charging-only slots
1 charging/programming slot
USB powered (only for charging/programing slot) or external power supply
Interface to Atmel STK500 (or compatible) programming kit

Download

Firmware

The source for all the sensor node firmware is available here.

All the nodes share the same firmware code base. Essentially, only the sensor initialization and data acquisition are specifically defined for each node. This makes it straightforward to adapt the firmware for new sensor nodes derived from this kit.

Educational software for activity recognition

Two educational software are provided to allow exploring the trade-offs in the design of an activity recognition system.

Hidden Markov models (HMM) are frequently used for hierarchical activity recognition. The first software allows to teach the design choices and trade-offs in using HMMs in
an activity recognition chain with explicit segmentation.

The second software is designed to be more visually appealing, by combining activity recognition with a game engine. It allows to teach choices and trade-offs to design a low-latency activity recognition chain (important for HCI or gaming scenarios),
with a simple NCC classifier operating on a sliding window.

Typical activity recognition chain used in wearable computing for activity/gesture recognition - the characteristics of the education software along the recognition chain is indicated below.

Matlab-based online activity recognition with hidden Markov models

A Matlab-based software allows to use separate discrete hidden Markov models to classify activities/gestures using the acceleration values of the motion sensor, typically placed on the forearm at the wrist. The software acquires directly the sensor data, visualizes data in real-time, and does real-time classification (recognition) of gestures and visualization of recognition likelihoods.

The main educational aspects are: i) the trade-offs between the complexity of the HMM (number of states, observations,
architecture), the computational time and the need for sufficient training data, ii) the trade-offs between sensitivity and
specificity of the segmentation method and the resulting types of classification errors, iii) the sensitivity to sensor orientation and placement variability, and gesture execution variability.

SensorTux - A gesture-based SuperTux game using wearable acceleration sensors

A C++-based software allows to recognize simple HCI gestures to control the linux open source computer game SuperTux, using one motion sensor typically placed on the forearm.

The main educational aspects are: i) introduction to the topic of human-computer interaction; ii) obtaining low-latency gesture classification; iii) understanding what are meaningful features to classify the desired gestures.

Additional software

The software below a set of softw

	SensHub - Sensor data acquisition, server and labeler SensHub has three key functions: Data aquisition from one or more sensors Data server over TCP to allow remote access to the sensors Data labeling, to indicate relevant events in the sensor data streams It provides various online data merge capabilities that are designed to handle incoming bursts of data, typical on many wireless (and even wired) connections. The software is generic and can be used with sensors others than those presented here.
	Java bluetooth gateway The Java BluetoothGateway is a tool to discover the sensors, record their data into a file, or act as a TCP server for a remote client. It receives data from multiple sensors and synchronizes and resamples their data streams.
	DScope - Fast and flexible oscilloscope DScope is a digital oscilloscope optimized for fast rendering and flexibility of configuration. It does real-time display of data coming from sensors interfaced over a serial, RFCOMM, USB or TCP connection. Incoming data format include plain text format and fully configurable binary streaming format. Acquired signal data can be exported to a file. The software is generic and can be used with sensors others than those presented here.
	SynScopeV - Signal/Video Visualization and Alignment Tool SynScopeV allows the interactive alignment of signals, and the alignment of signals with video footages. SynScopeV supports multiple signal sources that may differ in offset and/or sample rate, as well as multiple videos sources.
	CRNT Toolbox Driver The CRN Toolbox (context-recognition toolbox) is an open-source data-flow-oriented open-source signal acquisition and processing framework developed by the University of Passau. A generic driver (a reader task in the toolbox terminology) is available to acquire the data of sensors following the frame-based format, such as those presented here.

Documentation

Acknowledgements

The project would like to acknowledge the following persons:

March Bächlin: Initial concept of the motion sensor, initial concept and redesign of the charging station
Johannes Schumm: Initial design of the ECG sensor
Urban Suppiger: Redesign and additional features of the motion sensor
Clemens Lombriser: Java Bluetooth gateway development
Lars Widmer: Gestural game control developer
Thomas Holleczek: Matlab GUI for hidden Markov model based activity recognition
Mirco Rossi, Bernd Tessendorf, Sebastian Feese: definition of exercises for HMM-based activity recognition

Part of this work was financed by the Swiss-funded project Educational Kit for Wearable Computing under the NanoTera programme

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