Multimedia Security System Final Project Presentation

1
Multimedia Security SystemFinal Project
Presentation

• A Robust Elliptical Head Tracker
• Chih-Kai Yang
• Chun-Wei Hsu

2
Various Methods about Face Detection

• Face template
• Skin color and histogram
• Edge and boundary
• Depth
• Texture

3
Model We Choose Ellipse model

• Use a vertical ellipse with a fixed aspect ratio
  of 1.2 to match the shape of head
  

4
Four Modules used to Detect and Track

• Detected with Gradients
• Detected with Distance Transform image
• Color histogram
• Texture histogram

5
Gradient module
6
Gradient module (contd)

• S describes the location and size of ellipse
• (x, y) is the center of ellipse
• sis the length of long axis

7
Gradients The measure of match

• The normalized magnitude on boundary
• Take direction into concern

the intensity gradient of perimeter pixel
number of perimeter pixels
the unit vector normal to ellipse
8
Distance Transform module
9
Directions of edges

• Divide edges into multi-channels

10
Color Histogram

• Step 1
• R,G,B B-G , G-R ,
  BGR
• Step 2
• B-G 512(255-255) 8 bins
• G-R 512(255-255) 8 bins
• BGR 3x256 4 bins
• 8x8x4256 histogram

11
Color Histogram

• Step 3. Model histogram to be reference in
  first frame
• M(i) counts in the i th bin

12

Color Histogram

• Step 4. collect Target histogram after 2nd frame.
  
• I(i) counts in the i th bin

13
Color Histogram

• Step 5. Scan in Search Window .

?
14
Color Historam

• Step 6. Calculate Score of each scan
• Step 7Combine with Gradient Method

15
Texture Histogram

• Why do we use Texture Method?

16
Texture Histogram

• Step 1

4 scales 8 orientations Log-Gabor filter
32 dimension Texture data
4 scales 8 oreitation
Image
Ellipse Capture
17
Texture Histogram

• Step 2 transfer texture data to histogram

18
Texture Histogram

• Step 3.

?
19
The measure of match

• The normalized distance on boundary pixels
• Divide to multi-channels

20
Head Tracking

• Constant velocity model

21
Combine 2 modules to do tracking

• Convert scores to a percentage
• Find S whose score is maximun

22
DEMO

• Rotation , Zoom in/out , Sway.
• Simple occlusion
• Various conditions in monotonic background.
• Occlusion in complex background.
• Sway and Occlusion in complex background

23
Reference

• 1 Stan Birchfield. An elliptical head tracker.
  In Proceedings of the 31st Asilomar Conference on
  Signals, Systems and Computers, volume 2, pages
  17101714, 1997.
• 2 Stan Birchfield. Elliptical head tracking
  using intensity gradients and color histograms.In
  Proceedings of the IEEE Conference on Computer
  Vision and Pattern Recognition, pages 232237,
  1998.
• 3 Stanley T. Birchfield and Sriram Rangarajan.
  Spatiograms versus histograms for region-based
  tracking. In Proceedings of the IEEE Conference
  on Computer Vision and Pattern Recognition, pages
  11581163, 2005.
• 4 L. Brown. 3D head tracking using motion
  adaptive texture-mapping. In Proceedings of the
  IEEE Conference on Computer Vision and Pattern
  Recognition, pages 9981005, 2001.
• 5 Rafael C. Gonzalez and Richard E. Woods.
  Digital Image Processing. Prentice Hall,2002.