Face reorientation in video conferencing by approximating normal distributed depth - PowerPoint PPT Presentation

Loading...

PPT – Face reorientation in video conferencing by approximating normal distributed depth PowerPoint presentation | free to view - id: 219e34-ZDc1Z



Loading


The Adobe Flash plugin is needed to view this content

Get the plugin now

View by Category
About This Presentation
Title:

Face reorientation in video conferencing by approximating normal distributed depth

Description:

Face re-orientation in video conferencing by approximating normal distributed depth ... Foreground objects and background objects are preserved ... – PowerPoint PPT presentation

Number of Views:23
Avg rating:3.0/5.0
Slides: 15
Provided by: ben122
Category:

less

Write a Comment
User Comments (0)
Transcript and Presenter's Notes

Title: Face reorientation in video conferencing by approximating normal distributed depth


1
Face re-orientation in video conferencing by
approximating normal distributed depth
Hey Ben! Speak up and look at me!
eh I am!
Ben Yip Jesse S Jin The University of Sydney
2
Background
  • Known since 1969
  • Hardware solutions not easily accessible
  • Monocular approaches
  • Involves predefined face mask
  • Stereo binocular approaches
  • Personalized face model, feature points mapping
  • Multiple viewpoints approaches
  • Occluded feature points, eye gaze detection
  • Our approach
  • Monocular, with no predefined face model
  • Initially, use 4 images to obtain face depth data
  • Rotate the normal curve

3
Our approach - overview
  • Obtain front view and camera facing view
  • Obtain additional images
  • Calculate a rough depth map
  • Calculate the mean and standard deviation of the
    depth values, and construct the 3D normal curve
  • Rotate the 3D normal curve

4
Obtain front view and camera facing view
  • Use viewpoint determination algorithm to decide
    the angle of the camera.
  • In this example, it is 9.521 above and 5.878 to
    the left of the user.

5
Additional images
  • Two additional images are used in our experience.
  • Head pose is calculated for each image.
  • Then find the depth map of the face.

6
Calculate a rough depth map
  • Feature mapping for all images to the front view
  • Depth calculating using trigonometry
  • Could be better if more feature points, but too
    time consuming

7
Construct the 3D normal curve
Where W is width of the users face and
8
Why 3D normal curve?
  • It is a smooth function
  • Guarantees the peak of the curve is at the middle
    of the head
  • Values far away from the mean (usually is
    background) are practically zero
  • Simple to calculate

9
Rotate 3D normal curve
  • Model a downward looking scene using the 3D
    normal curve, and then rotate the head upward

10
Performance
  • 19 feature points to compare
  • Data is scaled to ES Units (eye separation units
    in MPEG-4)
  • Middle point of the eyes are aligned in the
    centre for comparison

11
(No Transcript)
12
Result
Before
After
13
Summary
  • Advantages
  • Foreground objects and background objects are
    preserved
  • No feature registration involved during video
    conferencing
  • Result is not perfect, but acceptable
  • Disadvantages
  • Initial depth and 3D normal curve takes awhile
  • Not real time yet

14
Thank You!
questions?
About PowerShow.com