Loading...

Calibration of PTZ Cameras from Multiview Soccer Match Videos

Hadian, Mostafa | 2015

1739 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 47636 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Kasaei, Shohreh
  7. Abstract:
  8. Camera calibration in sport videos has applications such as sport video analysis, insertion of virtual graphical elements, and 3D reconstruction. Our goal in this research is to fully calibrate cameras in soccer match videos. By calibration parameters, we mean the internal and the external camera parameters. We can have a mapping from world coordinates to image coordinates if we have all the calibration parameters. The cameras used in this research are stationary pan-tilt-zoom cameras. The main challenges in camera calibration are noise, blur, bad lighting conditions, lack of useful information image,fast movement of camera and radial distortion. In addition, as another challenge, we assume all the internal parameters to be variable. The most important priority in this research is calibration accuracy. By calibration accuracy, inverse of discrete model reprojection error is meant. As a socond priority, run time of the calibration algorithm is also important. We have proposed new algorithms by which we can accurately calibrate the camera in penalty area scenes and pitch center scenes and then track the calibration in the successive frames. At first we compute the initial homography with straight lines or by the use the center circle. Then we refine the initial homography using an iterative method based on a new homography estimation method. The new homography estimation method named the Point-Line method is an alternative to the common Direct Linear Transformation method. Finally using a technique based on vanishing points, we compute the full camera calibration with the least information possible. Furthermore, if we know the camera position, given we have the homography too, we can compute full camera calibration without the use of that technique. Thus we proposed a new method to compute the optimum camera position using the information of multiple frames. We evaluated the proposed method on two different videos and showed that the average reprojection error for frames of the first video was 0.73 pixels. While the average error of the previous method on the same video was 1.45 pixels. So we have had a 49 percent decrease in error from the previous method. In addition, the error of the proposed method on the second video was 1.68 pixels. Furtheremore, the average run-time of the homography refinement error used in calibration tracking for each frame was 364 milliseconds which is a 33 percent improvement compared to a previous method which has reported the run-time to be 550 milliseconds. full camera calibration is computed and is tracked accurately. Also our results proved to be of less error than a previous research
  9. Keywords:
  10. Tracking ; Homography ; Camera Calibration ; Projection Matrix ; Radial Distribution ; Soccer Match Videos ; Vanishing Point

 Digital Object List

 Bookmark

...see more