Facial Recognition for Persistent Surveillance Architecture - PowerPoint PPT Presentation

1 / 11
About This Presentation
Title:

Facial Recognition for Persistent Surveillance Architecture

Description:

Corinne Russell (crussel2) The Problem ... Can use realtime lookup on a face that enters frame. Chose techniques in particular areas to utilize their speed or ... – PowerPoint PPT presentation

Number of Views:44
Avg rating:3.0/5.0
Slides: 12
Provided by: slappy
Category:

less

Transcript and Presenter's Notes

Title: Facial Recognition for Persistent Surveillance Architecture


1
Facial Recognition for Persistent Surveillance
Architecture
  • Rockwell Collins Team
  • Corinne Russell (crussel2)

2
The Problem
  • Warfare and crime becoming more prevalent in
    urban environments
  • Real-time feedback
  • matching people of interest
  • Historical record
  • look back for more information
  • Balance speed and accuracy

3
Our Solution
  • Keeps database of faces, locations, times
  • allows for historical access
  • Can use realtime lookup on a face that enters
    frame
  • Chose techniques in particular areas to utilize
    their speed or accuracy when appropriate
  • Haar -- very fast
  • Clustering -- slow, but compresses well

4
The Process Detect a Face
  • Find a face in the frame
  • Haar, other heuristics
  • Track already detected face
  • Lucas-Kanade
  • Store enter and exit times
  • Store location

5
The Process Insert Image
  • Make request to server with face
  • multi-threaded so as to continue detecting faces
  • Feature extraction Landmarks (Face print)
  • distance between eyes
  • depth of eye sockets
  • width of nose, chin
  • skin texture (luminosity, hue)

6
The Process Insert Image
  • Check for a good match within database
  • Must have efficient search structure
  • hierarchical -- reduce max amount of searching
  • trees, probabilistic method
  • return/insert based on match results

7
The Process Cluster
  • Many images exist in database, some faces might
    be the same
  • Cluster images to combine identical faces
  • Keep entity for each unique face
  • store all locations, times for this face

8
Usage
  • Query database based on image, location, times
  • Still have realtime recognition
  • Add relevant information into database manually

9
Implications
  • Could flag particular people
  • when face detected, sets off alarm, etc.
  • Mine for correlations between people, people and
    activities

10
Possible Use Case Scenario
  • New knowledge of criminal
  • Look back through database for locations, times
    of person
  • Find correlations between criminal and other
    people, criminal and activities
  • Still have realtime lookup if criminal appears
    again

11
The End
  • Questions?
Write a Comment
User Comments (0)
About PowerShow.com