CSE 20011048 JaeDeug Choi, Computer Vision Lab

1
CSE 20011048 JaeDeug Choi, Computer Vision
Lab Advised by Prof. JoonHee Han Master-Slave
Camera System for Intelligent Surveillance
2. Overview of Procedure
1. Introduction

• Framework
• Master Camera is a static camera that watches a
  wide fixed field of view.
• Slave Camera is a dynamic camera that pans, tilts
  and zooms so that it captures high resolution
  images.

Master Camera
Slave Camera
Background modeling and object detection
Obtain matching points
Learning
Homogeneous matrix H
Compute a homogeneous matrix
Find an object in the sliding window

• Objectives
• Current surveillance system consisted with a
  static camera, which watches the wide view,
  easily misses the information of relatively small
  objects. Also the system needs lots of expenses
  of analyzing and managing huge recorded video
  data.
• Master-slave camera system can solve these
  problems by getting high resolution images from
  slave camera.

Coordinates of the object
Execution
Pan, tilt and zoom on the mapped object
High resolution image
Map the object onto the slave camera view
3. Learning Procedure

• Background modeling using MoG

• Object detection by a connected component
  algorithm

• Obtain matching points from each view at the
  same time

• Compute a homogeneous matrix

4. Execution Procedure

• Master Camera
• Find an object in the sliding window, which
  represents the region that slave camera pans,
  tilts and zooms on.
• Map the coordinates of the object onto the slave
• camera view using the computed homogeneous
• matrix.

• Slave Camera
• First, pan and tilt on the region which contains
  the mapped object.
• Second, pan, tilt and zoom on the mapped object.
• Third, take a photo of the object and store it
  with time and coordinates.

5. Conclusion
6. References Thanks to

• Pros and Cons
• Without camera calibration, the coordinates on
  the master camera view can be mapped onto the
  slave camera view.
• High resolution images are acquired by slave
  camera.
• This system can reduce the expenses of analyzing
  huge video data.
• Sometimes slave camera misses the object because
  this system cannot predict the objects movement
  and slow PTZ speed of slave camera makes time
  delay.

1 Senior, Hampapur, and Lu, "Acquiring
Multi-Scale Images by Pan-Tilt-Zoom Control and
Automatic Multi-Camera Calibration", IEEE
Workshops on Application of Computer Vision,
2005(1) 433 2 Bodor, Morlock, and
Papanikolopoulos, "Dual-Camera System for
Multi-Level Activity Recognition", IEEE
Internalional Conference on Intellignet Robots
and System, October, 2004 643 3 Stillman,
Tanawongsuwan, and Essa, "A System for Tracking
and Recognizing Multiple People with Multiple
Cameras", Georgia Institute of Technology,
Graphics, Visualization and Usability Center,
Technical Report, August 1998. 4 Hartley, and
Zisserman, Multiple View Geometry in Computer
Vision 5 Stauffer and Grimson, Adaptive
background mixture models for real-time
tracking, IEEE Internalional Conference on
Computer Vision and Pattern Recognition, 199(2)
252 Special thanks to TaeGyu Lim, SungHoon Kim,
WoonHyun Nam, ChungSong Song, SuHa Gwak I really
appreciate their advices, comments and great
supports.