Project

1
Project 2Leveraging the snBench

• Tim Walsh
• Samir Keco

2
Motivation

• A user wishes to track the occupants of an
  enclosed area such as an office, club, airport,
  casino, etc.
• The user may be interested in HOW MANY occupants
  there are, as well as WHO the occupants are

3
Previous Project

• Lisa Premerlani, Luke Skelly, and Wajeeha Akram
  approached the question of HOW MANY?

4
We ask the question WHO?

• Camera watches a doorway, motion triggers camera
  to take snapshot
• Face detection algorithm is run on snapshot,
  cropped images of faces are stored in database
• Face images in database are displayed on
  automatically updated webpage for user

5
Face Detection

• Based on searching for skin-colored pixels
• Fine to coarse grain filtering
• Assumptions
• - Fixed lighting
• - Fixed distance from camera to door

6
(No Transcript)
7
?
8
(No Transcript)
9
Face Detection

• Obviously not perfect
• - Some false positives
• - Output images not quite normalized
• Makes many assumptions
• Improved algorithm would use statistical
  analysis, neural net, etc.

10
Process of Flow
Detect motion (outputs snBoolean)
grabframe (outputs snImage)
Triggers
11
Process of Flow
Take snImage
FaceDetection
Returns snImages
12
Database
snImage
Store on Harddrive
Figure out the Link
Convert To Bytes
Store Link
Database
13
Database
snImage
Store on Harddrive
Figure out the Link
Convert To Bytes
Store Link
Database
14
SXE.CORE.SQL

• ConvertToByte (snImage)
• returns the image as an byte
• WriteToDatabase(bytes)
• Writes to Database
• ReadFromDatabase()
• Returns the data from Database
• ConvertToImage (byte)
• Recreates an image from bytes

15
Possible Extension 1

• Take snapshots on the other side of door to
  identify people leaving
• Match up these faces to those in the database and
  remove them
• Maintain an accurate database of current
  occupants along with counter
• Solves weakness of previous system which did not
  account for people entering or leaving
  simultaneously

16
Possible Extension 2

• Compare faces as they are found to another
  database (uploaded by the user) containing
  known faces
• Alert user is match is found
• Good for airport, casino security

17
Looking Back

• Took a long time to define narrow goals for the
  project and get started
• Good division of labor
• Tim Vision algorithms and opcodes, maintaining
  website
• Samir STEP logic for integrating camera,
  opcodes, database

18
Going Forward

• Turn face detection module into working snBench
  opcode
• Finish STEP logic for interacting with camera to
  watch for motion, take snapshots
• Writing to and building webpage from database
• Face recognition algorithm