Design Methodology for ContextAware Wearable Sensor Systems

1
Design Methodology for Context-Aware Wearable
Sensor Systems

• Urs Anliker, Holger Junker,
• Paul Lukowicz, Gerhard Troster
• Zurich,Schweiz

2
Introduction

• Optimization problem where to put the sensors,
  how to organize the system
• Traditional way trial and error approaches
• Formalize the optimization problem and pave the
  way towards an automated system design process.

3
? previous work A systematic approach to the
design of distributed wearable systems. ? 3
components - architecture model -
problem specification - exploration
environment
4
Architecture Model

• Generic model
• A set of computing modules distributed over the
  users body
• Module
• Devices processors, sensors
• Communication channel interface IO interface
• Inter-module communication physical channel
• Problem specific model
• a set of constraints on following from problem
  specification
• system topology
• module resource set
• task resource set

5
Problem Specification(cont.)

• Usage profile
• Workload characteristics of the wearable system
• Task
• Self-contained unit
• Input data
• Computational load
• Use instruction mixture
• Output data
• Application
• Scenario
• Assigning 2 timing parameters
• Repetition frequency R
• Maximal latency Dmax

6
Problem Specification(cont.)

• Information flow
• A set of body locations to each I/O related task
  of the usage profile specification.
• Physical constraints
• The devices size, weight
• Power consumption
• Power weight vector wp

7
Problem Specification

• Hardware resources
• A set of computation and communication channel
  devices available for the design.
• Computing devices
• 2 characteristics of
• Execution speed
• Energy consumption per cycle
• Operation modes sleep, idle, execution
• Communication channel
• 4 operation states transmitting, receiving,
  idle, standby
• 2 modes continuous mode, burst mode

8
Exploration Environment

• 3 optimization criteria
• Functionality
• By timing constraints R and Dmax
• Battery Lifetime
• Wearability
• The sum of the abstract wearability factors of
  all allocated resources for a specific module m
  at a given location p.

9
Extensions to the Model(cont.)

• Context
• 2 parameters
• Update rate
• Minimum classification performance
• Context lt-gt application
• Feature
• Mathematical function, which preprocesses raw
  sensor data.
• Feature lt-gt task
• For data reduction, dataout ? datain

10
Extensions to the Model(cont.)

• Sensor
• Generates data at a sampling frequency (fsample)
  with a specific resolution (res).
• 2 power states sampling (Psample), idle (Pidle)
• Be specified in the hardware resource
  specifications

11
Extensions to the Model

• Mapping and Binding
• Introduce context resource set and feature
  resource set
• Sensors are used as data source

12
Objectives(cont.)

• Power consumption
• Sensor load(SLoad)

frep update rate
13
Objectives(cont.)

• Classification performance
• Recognition accuracy
• Mutual information
• If MI(XY) 0, then X, Y is independent

feature F
class variable C
14
Objectives

• Error probability (Fanos inequality)
• The error probability is decreased if MI(CF)
  gets larger.

of classes
15
Case Study
Force sensitive resistors
???
16
(No Transcript)
17
(No Transcript)
18
Conclusion and Future Work

• Describing a methodology designed to formalize
  and automate the choice and placement of sensors
  and the selection of features for wearable
  context sensitive systems.
• Future work Apply it to more complex case
  studies and recognition tasks

19
Contribution

• Systematic framework for sensor selection based
  on power consumption and classification
  performance for wearable computing applications.
• Taking classification performance via mutual
  information for computation and communication.

20
Q A

• Thank you !!