CenceMe - Computing and Software

advertisement
Design, Implementation and
Evaluation of CenceMe Application
COSC7388 – Advanced Distributed Computing
Presentation By
Sushil Joshi
Outline
Introduction
Architectural Design
Limitations
Split level classification
Architectural Diagram
Classifier
Phone Classifier
Backend Classifier
Performance
Power and Memory Benchmark
Experimental Deployment and feedback
Introduction
Mobile application that infers personal
presence and updates the status to social
networks.
Sensor devices like microphone,
accelerometer, GPS, camera and
bluetooth inbuilt in Nokia N95.
An always-on application needs to use energy in as
efficient way as possible.
Introduction
Sense
Learn
Share
Information and process flow in CenseMe System
Introduction
Realizing vision of automatic updates to social
networks.
Enablers – Integration of sensors to consumer
mobile devices.
Vision about bluetooth enabled cellphone
talking to
• Other devices attached in running shoes, BlueCell
dongle
• Attached to other user
• Sensor available in town ecosystem like carbondioxide or pollen sensors.
Nokia N800, N95, Nokia 5500, Tmote Mini, BlueCell Dongle.
Architectural Design (Limitations)
Symbian OS Exception handlers
API limitations – e.g. Missing JME API to
access N95 internal accelerometer
Security Limitations
Energy Management Limitations
Architectural Design (Split level
Classification)
Architectural Design (Split Level
Classification)
Advantages
Minimizes sensor data that needs to be uploaded
Resiliency when Radio/WiFi dropout by buffering
and batching primitives
Minimizes sensor data that needs to be uploaded
thus saving energy that would be used up.
Architectural Diagram (Phone
Software)
Architectural Diagram (Backend)
Classifier (Phone Classifier)
DFT of human voice sample
registered by Nokia N95 microphone
DFT of audio sample from noisy
environment as registered by Nokia
N95 microphone
Classifier (Phone Classifier)
Discriminant analysis clustering which determines the dashed lines
(threshold between talking and non-talking)
Classifier (Phone Classifier)
Data collected by Nokia N95 on-board accelerometer for different activities like
sitting and walking.
Classifier (Backend Classifier)
Rolling window of size N=5 used by
conversation classifier
Assymetric strategy
P1
P2
P3
P4
P5
p1
p2
p3
p4
p5
Primitive indicates voice
Primitive indicates no voice
Conversation
No Conversation
Classifier (Backend Classifier)
Social Context classifier
Mobility Mode Detector
Location Classifier
Historical trend of user data to identify
behaviorial pattern. e.g. Nerdy, party
animal, health conscious.
Performance
Table 2 indicates false positives which could be attributed to either sensors
grasping human voice from background or due to assymetric strategy for
conversation classification.
Performance
Conversation classifier accuracy in different ambience
Performance
Conversation Classifier accuracy with varying duty cycle
Performance
Accuracy of activity classification vs different positioning of mobile phone
Power, Memory and CPU Usages
Power consumption during sampling/upload interval
Power, Memory and CPU Usages
Screen saver mode turned on while using Nokia Energy
Profiler so as to decouple energy used to light up the
LCD screen.
Feedback From Experimental
Deployment
More likely to be used by population who
already use social networking.
Far less deletion of random images
compared to uploads.
Location feature mostly used.
Can reveal lifestyle trends e.g less
physical activity
Questions
?
Reference
[1]Miluzzo, Emiliano, Lane, Nicholas D., Fodor, Krist\'of,
sPeterson, Ronald, Lu, Hong, Musolesi, Mirco, Eisenman,
Shane B., Zheng, Xiao, Campbell, Andrew T., Sensing meets
mobile social networks: the design, implementation and
evaluation of the CenceMe application, SenSys '08:
Proceedings of the 6th ACM conference on Embedded
network sensor systems, pp. 337--350, ACM, New York, NY,
USA, 2008.
[2] Emiliano Miluzzo, Nicholas D. Lane, Shane B. Eisenman,
and Andrew T. Campbell, CenceMe – Injecting Sensing
Presence into Social Networking Applications
Download