arpan pal csi2012

14
1 Personal Context Discovery using Unobtrusive Sensing 1 st December 2012 CSI 2012 Arpan Pal Principal Scientist and Research Head Cyber physical Systems Innovation Lab, Kolkata Tata Consultancy Services (TCS)

Upload: arpan-pal

Post on 15-Aug-2015

40 views

Category:

Technology


1 download

TRANSCRIPT

1

Personal Context Discovery using Unobtrusive Sensing

1st December 2012CSI 2012

Arpan PalPrincipal Scientist and Research HeadCyber physical SystemsInnovation Lab, KolkataTata Consultancy Services (TCS)

2

Human-in-Loop Cyber-physical Systems

Humans

Computing Infrastruct

ure

ICT Systems

3

Human-in-Loop Cyber-physical Systems

Humans

Physical Objects

and Infrastruct

ure

Computing Infrastruct

ure

Cyber-physical Systems

4

Human-in-Loop Cyber-physical Systems

Humans

Physical Objects

and Infrastruct

ure

Computing Infrastruct

ure

Human-in-loop Cyber-physical Systems

Perso

nal

Conte

xt

Disco

very

5

Personal Context Discovery

Context - patterns of individual, group and societal behaviours.

Broadly classified into three categories –Personal Physical Network Discovery

Who is interacting with whom? What is the level of interaction? Who all are part of similar-interest networks?

Individual Context Discovery Who is doing what? Who is thinking what?

Community Context Discovery Can we discover how a community / group behaves as a whole?

6

Example Use Cases

Customer Behavior Study in Retail Stores Customer movement pattern Customer interaction pattern with shelves / merchandize

Crowdedness measure in public places Efficient scheduling of public transport

Wellness Activity and Work-out Quantification Pulse and other physiological parameter measurement

Organizational Behavior Team Efficiency / Best Practice Study Workspace Ergonomics - Stress Analysis People Profiling – Cognitive Load Analysis

Ref. - Alex Pentland et. al., MIT media Lab

7

What do we need to Sense

Location and Proximity

Activity

Identity

Cognitive Load

Physiological Parameters

Provide Personal Context

discovery as a Service

8

How to Sense

Requirement• Needs to be Ubiquitous and Unobtrusive• There should not be any new hardware / device to carry

for an individual

Approach• Use smartphone-based sensors (GPS, accelerometer,

compass, microphone, camera)• Use 3D surveillance cameras (like Kinect)• Use wearable EEGs with mobile phone as gateway• Augment with social network data and email data

analytics• Multimodal fusion of all the above

Issue• Privacy can be an issue – needs to be handled on an use

case-by-use case basis• Privacy vs. Utility

9

Mobile Phone Based Sensing

Proximity / presence– Using Bluetooth – Using Wi-Fi

Location– Using ultrasound beacon– Using GPS (outdoors)– Using Accelerometer /

compass / gyroscope

Activity– Using Accelerometer

Interaction Level– Using Microphone Audio

Identity– From Network ID

Physiological Sensors– Pulse rate using camera - PPG

On-board sensorsAccelerometer, GPS, Compass, GyroscopeCamera, Microphone

NetworkBluetooth, WiFi, 2G/GPRS, 3G

Network2G/GPRS, Bluetooth

On-board sensorsMicrophone, Camera

10

Kinect Based Sensing

Human Identification– Skeleton Model Based– External Stimulus based

refinement

Network Discovery– Network discovery through

proximity– Level of Interaction through

Audio

• 2D Camera with IR depth sensor• Excitation by IR light pattern• Directional Mic.

Human Interaction – Activity Detection on 3D Point

Cloud– Physical object Identification– Interaction with ObjectsExample Activities• People Discussion• Give/Put/Take an object• Enter/leave a room• Handshaking

11

EEG Based Sensing

Cognitive Load Detection– Wearable EEG headset – Biggest challenge is collecting

annotated data and removing movement artifacts

Signal Acquisition

Pre-processing (Common Spatial

Pattern filter)

Feature Extraction

Classification / Score generation

Measure of the cognitive load

Feature-1

Featu

re-

2

High cognitive load

Low cognitive load

12

Soft sensing from Web

Unstructured Data• Social network posts such as

tweets, facebook• Blog posts• Email bodies

Structured Data• Social network profiles and

network information• Email headers• Tweet Attributes

Personal

Network

13

Summary

Multimodal Fusion

Location Identity Activity

Sensors• GPS, accelerometer, compass, microphone, camera on SmartPhone • 2D / 3D surveillance cameras (like Kinect)• Wearable / Mobile-phone hosted Physiological Sensors – Pulse, ECG, EEG• Soft sensors from web and social network

Cognitive Load & Phys. Sensing

Context DiscoveryPhysical, Individual & Community

Applications

Thank You

[email protected]