MULTIPLE MOVING OBJECTS TRACKING FOR VIDEO SURVEILLANCE SYSTEMS

1
MULTIPLE MOVING OBJECTS TRACKING FOR VIDEO
SURVEILLANCE SYSTEMS
2
Outline

• Foreground Regions Detection
• Features Extraction
• Tracking1. Isolated object2.Occlusion object.

3
Foreground Region Detection (Flow chart)
4
Foreground Regions Detection

• Background subtraction
• Binarization
• 2-stage median filter
• Update Background

5
Foreground Regions Detection (Result)
6
Features Extraction

• Centroid
• Area
• Mean intensity

7
Isolated object tracking

• Mean intensity difference
• Area difference
• Weighted sum similarity function
• Distance similarity function
• Final

8
Dynamic Template Matching

• Template is created using previous frame (before
  occlusion)
• The correlation is then performed between the
  template and the current frame.
• The high score of the normalized
  cross-correlation matrix is occurred where the
  template is best correlated.
• The spatial position at which closes copy of the
  searched object is located.
• This spatial position represents the centroid of
  the tracked object in the current frame.
• The minimum distance from the objects position
  in the previous frame. (have many high score)

9
Parameters

• Gray-level video imagery with frame rate 24 fps
  and frame size 320240 pixels
• c10.6, c20.4
• Range of the weighted sum function is 0.65,
  0.81, and for the distance function is 3.1,
  8.8
• Th10.85, Th210

10
Experimental Results
11
Experimental Results
12
Experimental Results