Fast and Robuse Algorithm of Tracking Multiple Moving Objects for Intelligent Video Surveillance Systems by linzhengnd

VIEWS: 4 PAGES: 19

									  Fast and Robust Algorithm of
Tracking Multiple Moving Objects
for Intelligent Video Surveillance
              Systems

   Jong Sun Kim, Dong Hae Yeom, and
         Young Hoon Joo,2011
                    Goal
• detecting and tracking multiple moving
  objects
• real-time detecting
• robustness against the environmental
  influences and the speed
                    Outline
• Introduction
• Previous Methods
• Detecting Moving Objects
  – Extraction of Moving Objects
  – Grouping Moving Objects
• Tracing Moving Objects
• Implementation and Experiment
• Conclusions
               Introduction
• In the traditional systems that a person should
  always monitor video.
• intelligent video surveillance systems are
  high-cost and low-efficiency
• Environment affects a lot.
• This paper propose a method detecting and
  tracking multiple moving objects in real-time.
           Previous Methods
• particle filter ,extended Kalman filter
• Background modeling (BM) or the Gaussian
  mixture model (GMM)

• gray-scale BM shows the image information is
  excessively attenuated.
    Extraction of Moving Objects
• Using RGB color BM instead of gray-scale BM
• Each pixels will compare with previous pixels
  in little group.
• If it is stationary, the pixels will be black.
• The parameter δ is proposed to overcome the
  sensitivity problem .
• δ would be different on different camera.
Extraction of Moving Objects
Extraction of Moving Objects
Extraction of Moving Objects
      Grouping Moving Objects
• The individual tracking of neighboring or
  overlapping objects requires a lot of
  computational capacity .
• The 4-directional blob-labeling is employed to
  group moving objects.
      Grouping Moving Objects
• Contour Tracing
       Grouping Moving Objects
• its initial search position is set to be
       d+2 (mod 8)
Tracing Moving Objects
Tracing Moving Objects
 Implementation and Experiment
• The 33Mbit IP camera provides the input
  image with 704x480 pixels.
• The surveillance image is transmitted through
  Internet.
• 2.66GHz CPU and 4GB RAM PC for the image
  signal processing and the proposed algorithm.
Implementation and Experiment
Implementation and Experiment
Implementation and Experiment
               Conclusions
• Real-time detecting and tracing
• Only for fixed camera.
• Future works can be on predicted position.

								
To top