The document describes Citron, a framework for context acquisition using personal sensory devices. It was developed to utilize multiple sensors on Muffin, a prototype sensory device, to enable reliable context recognition through analysis of information from different sensor aspects. Citron includes context analysis modules that work independently and share context information. A sample application demonstrates detecting a user's walking state and speed in real time using various sensor modules. Experiments showed coordination of analysis modules improves context acquisition. Future work includes coordinating personal and remote sensors for context recognition.
Why device, WIFI, and ISP insights are crucial to supporting remote Microsoft...
Citron : Context Information Acquisition Framework on Personal Devices
1. Citron : Context Information Acquisition Framework on Personal
Devices
Distributed Computing Laboratory
Waseda University
Tetsuo Yamabe, Ayako Takagi, Tatsuo Nakajima
2. Outline
1. Introduction
2. Muffin
3. Sensors on Muffin
4. Context acquisition on Muffin
5. Citron
6. Sample application
7. Experiments result
8. Conclusion and future direction
3. Introduction
• It is expected that personal devices acquire a perceptual
ability and recognize a user’s context information.
– Why personal devices?
• Tight partnership with a user
• Connectivity to a user and context-aware services
– How they recognize?
• Incorporate sensors and analyze acquired values
" What type of sensors are useful to acquire a user’s context ?
" What is required in the process of context acquisition ?
4. • We have developed Muffin, which is a prototype of a
sensory personal device, to investigate sensors’
characteristics and data processing process.
• Also, we have developed a framework named Citron…
– to utilize the advantage of multiple sensory personal device.
– to implement context analysis modules on it.
• By running context-aware application on top of Citron,
we present…
– how Citron bring out Muffin’s capability
– possibilities of personal devices fabricated with multiple sensors
5. What is Muffin??
• Muffin is a prototype of the future sensor device for
research on ubiquitous computing area.
– Developed by a collaboration work with Nokia Research Center
– Sensing capability for context-awareness
• 15 kinds of sensors in a PDA size box
– Linux OS
– Wired / Wireless interface
• Bluetooth, IrDA, WLAN
• USB, Serial port, PCMCIA slot
6. Sensors on Muffin
• Sensors on Muffin are roughly divided into 4 categories.
• Environmental sensors
Alcohol gas sensor • Physiological sensors
Relative humidity sensor • Motion/Location sensors
Air temperature sensor • Other sensors
Skin resistance sensor Rear camera
Grip sensor
Front camera
RFID reader GPS
Microphone
Pulse sensor
Barometer
Compass / Tilt sensor
3D Linear accelerometer
Skin temperature sensor Ultrasonic range finder
7. Context acquisition on Muffin
• We performed some experiments about context
acquisition on Muffin; and found that…
" Validity of sensor value and analysis algorithm changes
frequently according to a user’s taking style.
" Some sensors’ characteristics require long term data logging.
Muffin Waist-mounted Held Held
User Not watching Not watching Watching
Pulse invalid valid valid
Standing or not ? invalid invalid valid
8. " Multiple sensors enable reliable context acquisition by
analyzing information from multiple aspects of view.
Walking or running or not
Under watch or not Moving or stop
Activity(1- 5)
Held or not
Top side
Skin resistance Activity(1 - 5)
Accel Ultra range finder
" We should reflect the already recognized context…
– to select an appropriate set of sensors and analysis algorithms
– by modeling relationships among other context
→ Middleware support should be offered to application
programmers.
9. Citron: architecture overview
Application
• A framework for context acquisition
Citron API
on sensory personal devices
– Citron Worker
• Context analysis module
put
• Work independently Citron Worker Citron Space
• Enable parallel context processing
– Citron Space
• Shared space for storing context
read
• Core module of a blackboard model
Output
• Citron supports … - Analyzed context
– Hierarchical context abstraction Context
– Context analysis from multiple
aspects of view
Sensor
– Switching analysis module Input
- Sensor data
according to context - Context
10. Sample application : StateTracer
• StateTracer displays the track of walking route with
user’s state in real time.
– Not only walking or not, but also walking speed and resting time
– No location systems or infrastructure
Walking At rest
11. Working modules on StateTracer
Orientation Walking (state, speed)
Can detect speed,
but time consuming Walking (FFT) Walking (Threshold)
Citron
worker Watching Activity
Can not detect speed,
Sensor Holding Top_side but responsive
compass skin resistance accel_x, accel_y, accel_z
Orientation : “N”, “NW”, “W”, “SW”, “S”, “SE”, “E”, “EW”
Walking_State : “walking”, “resting”
Walking_Speed : “0”, “1”, “2”, “3”, “4”
12. Experimental result
• Walk around a lot (50m x 100m)
Start and Goal
– Change walking speed
– Two stop point Stop point
• Change working analysis modules Walk fast
– Case1 : Walking (threshold) worker only
Walk slowly
– Case2 : Walking (FFT) worker only
– Case3 : Both workers
Case 1 Case 2 Case 3
13. Conclusion and future direction
• Coordination among analysis module with sharing
context information is flexible and effective way to
acquire context on Muffin.
– Bring out capability of Muffin and its perceptual ability
– Enable reliable context acquisition in practical usage
• We continue to research on context acquisition on
personal devices based on Muffin and Citron.
– Rearrange placement of sensors and reshape its form
– Distribute sensors as a wearable sensor device
– Coordination with remote resource over network
14. Cookie : Coin size wearable sensor
• Size • Sensor
– 24mm x 22mm x 8~10mm – Compass
– Almost same size as 10 Yen coin. – Ambient Light Sensor
• Three stacked board structure – Pulse sensor
– Main board – Skin temperature sensor
– Sensor board – GSR sensor
– Extension board – UV sensor
• Running time – RGB color sensor
– About 1 hr (with 2032 size battery) – 3-Axis Linear Accelerometer
– Vibration motor