By: Heng Wei Jian
Motivation of work
 Main objective at conference -> Network with other
 participants

 Golden opportunity that is often wasted

 Build a new mobile platform that facilitates this
 process

 Finding and presenting essential information to the
 user using augmented reality technologies
Motivation of work
In a conference:

 Who should I talk to?

 Who is that guy over there? He seems familiar.

 How do I find out more about him?

 How do I get an opportunity to talk to him?

 How do I approach him?
Project Objectives
 To create a mobile application that enables users to
 network effectively with other participants in a
 conference

 Evaluate the usefulness of the system.
Approach
Platform: iPhone 4

Main Features:
(Who to talk to?)
 Real-time mobile Facial Search
 Conference booklet with QRCodes

(How to find out more about him?)
 Facebook integration

(How to contact him?)
 Real-time personalized message board
Research Topics
 Balance between privacy concerns and ease of use

• Find out best possible way to gather information about
 participants in a conference without intruding their
 privacy but requiring minimal user input

• Can make use of existing social networks to get
 participants information but needs to be appropriate
 in a conference context
Research Topics (cont)
 Usefulness of the various user search tools for
 identifying people.

 Textual Search


 Facial Search


 QR Search
Research Topics (cont)
 Effectiveness of mobile tool for conference
 networking purposes

 Evaluation of system
    Ease of use
    Error frequency
    Interface design
    Task suitability
    Satisfaction
    Privacy concerns
Technical Design
Tools:

 IOS4.0
 PHP remote server
 MySQL database
 OpenCV 2.0
 Facebook SDK
 ARPlus toolkit
Application Flowchart
                                  Facebook
                                    Login
                     Login

     Account                       Normal
     Settings                       Login
                   Conference
                      List


     Textual                      Personal
                   Conference     Message
     Search
                     Details       board


                   Participant   Participant
   Facial Search
                      List         Details



    QR Search
Core Feature: Facial Search
 Allows user to easily identify other participants in the
  conference using facial detection and facial
  recognition technologies

 Non-intrusive and appropriate in conference context

 Training images can be obtained from social networks
  to relieve user manual input

 Results augmented on screen
General Approach

 Grab image                             Augment
               Facial        Facial
from camera                             results on
              Detection   Recognition
   frame                                  screen
Challenges
 Running speed on Mobile Devices
    Most algorithms require fast CPU speed and high
     memory
 Accuracy
    Accuracy is heavily dependent on pose and illumination
 Obtaining Training Images
    Get sufficient quality training images without heavy user
     input
 Capturing moving images
 Distance factor
Current Progress
iConference
Current Progress
Face Detection
   OpenCV
   Implements Viola-Jones object detection framework
   Makes use of Haar Classifier to describe and find general
    facial features
   Accuracy level for frontal view : 95%
   Already tested on the phone – average of 1 to 2s
Current Progress
Face Recognition
1st Method: Face.com
->3rd party web-based recognition tool

 Advantages:
    Easy to use
    Accuracy level: 70%


 Disadvantages:
    Not open source
    Huge overhead to post image to web to get results
    Slow
Current Progress
Face Recognition
2nd Method: Eigenfaces
-> Using PCA (Principal Component Analysis)

 Advantages
    Fast
    Uses less memory


 Disadvantages
    Build from scratch
    Proclaimed accuracy level: 60%
Current Progress
Image Pre-processing Techniques

                                                    Illumination
                           Face Alignment
                                                   Normalization




     Original           Rotated,Cropped,Resized   HistogramEqualized
Timeline:
                                                               Mar   Apr   May   Jun   Jul   Aug   Sep   Oct   Nov
    1 Research and Implementation


    V    Research on QR's current implementation

    V    Implement QR algorithm in ObjectiveC

    V    Implement QR tracking in the booklet
   WJ    System Design and Modelling

      Building the framework and foundation of the
   WJ application
    2 Integration and Iteration


    V Ensure Accuracy and Tweaks to QR              tracking

   WJ    Adding extra features to Application


V & WJ   Modifications to application based on tests
    3 Usability and Thesis

V & WJ   Carry out usability tests at conferences

V & WJ   Thesis and Technical Paper
DEMO
Screenshots (Face.com)
Screenshots (EigenFaces)
Screenshots




 Original   Cropped   Greyscale and   Equalized
                      resized

iConference

  • 1.
  • 2.
    Motivation of work Main objective at conference -> Network with other participants  Golden opportunity that is often wasted  Build a new mobile platform that facilitates this process  Finding and presenting essential information to the user using augmented reality technologies
  • 3.
    Motivation of work Ina conference:  Who should I talk to?  Who is that guy over there? He seems familiar.  How do I find out more about him?  How do I get an opportunity to talk to him?  How do I approach him?
  • 4.
    Project Objectives  Tocreate a mobile application that enables users to network effectively with other participants in a conference  Evaluate the usefulness of the system.
  • 5.
    Approach Platform: iPhone 4 MainFeatures: (Who to talk to?)  Real-time mobile Facial Search  Conference booklet with QRCodes (How to find out more about him?)  Facebook integration (How to contact him?)  Real-time personalized message board
  • 6.
    Research Topics Balancebetween privacy concerns and ease of use • Find out best possible way to gather information about participants in a conference without intruding their privacy but requiring minimal user input • Can make use of existing social networks to get participants information but needs to be appropriate in a conference context
  • 7.
    Research Topics (cont) Usefulness of the various user search tools for identifying people.  Textual Search  Facial Search  QR Search
  • 8.
    Research Topics (cont) Effectiveness of mobile tool for conference networking purposes  Evaluation of system  Ease of use  Error frequency  Interface design  Task suitability  Satisfaction  Privacy concerns
  • 9.
    Technical Design Tools:  IOS4.0 PHP remote server  MySQL database  OpenCV 2.0  Facebook SDK  ARPlus toolkit
  • 10.
    Application Flowchart Facebook Login Login Account Normal Settings Login Conference List Textual Personal Conference Message Search Details board Participant Participant Facial Search List Details QR Search
  • 11.
    Core Feature: FacialSearch  Allows user to easily identify other participants in the conference using facial detection and facial recognition technologies  Non-intrusive and appropriate in conference context  Training images can be obtained from social networks to relieve user manual input  Results augmented on screen
  • 12.
    General Approach Grabimage Augment Facial Facial from camera results on Detection Recognition frame screen
  • 13.
    Challenges  Running speedon Mobile Devices  Most algorithms require fast CPU speed and high memory  Accuracy  Accuracy is heavily dependent on pose and illumination  Obtaining Training Images  Get sufficient quality training images without heavy user input  Capturing moving images  Distance factor
  • 14.
  • 15.
    Current Progress Face Detection  OpenCV  Implements Viola-Jones object detection framework  Makes use of Haar Classifier to describe and find general facial features  Accuracy level for frontal view : 95%  Already tested on the phone – average of 1 to 2s
  • 16.
    Current Progress Face Recognition 1stMethod: Face.com ->3rd party web-based recognition tool  Advantages:  Easy to use  Accuracy level: 70%  Disadvantages:  Not open source  Huge overhead to post image to web to get results  Slow
  • 17.
    Current Progress Face Recognition 2ndMethod: Eigenfaces -> Using PCA (Principal Component Analysis)  Advantages  Fast  Uses less memory  Disadvantages  Build from scratch  Proclaimed accuracy level: 60%
  • 18.
    Current Progress Image Pre-processingTechniques Illumination Face Alignment Normalization Original Rotated,Cropped,Resized HistogramEqualized
  • 19.
    Timeline: Mar Apr May Jun Jul Aug Sep Oct Nov 1 Research and Implementation V Research on QR's current implementation V Implement QR algorithm in ObjectiveC V Implement QR tracking in the booklet WJ System Design and Modelling Building the framework and foundation of the WJ application 2 Integration and Iteration V Ensure Accuracy and Tweaks to QR tracking WJ Adding extra features to Application V & WJ Modifications to application based on tests 3 Usability and Thesis V & WJ Carry out usability tests at conferences V & WJ Thesis and Technical Paper
  • 20.
  • 21.
  • 22.
  • 23.
    Screenshots Original Cropped Greyscale and Equalized resized

Editor's Notes

  • #3 The problem we are trying to solve….
  • #4 Right now, there is no easy way to get these answers. Our app hopes to be an integrated platform for users to obtain these kind of information that they need. Research shown that Form stronger memory associationsExample scenario:
  • #7 For app to be useful and popular -> easy to use and fuss-freeWhere to go get the info?People are too busy. Too tedious to fill in data manually
  • #12 Example scenarios: Familiar face, fast forward sessionsAccuracy problems ->search
  • #14 People wear name tags but might be too far to see.Reasonable distance factor is about one room distance.