Calibration

Knowledge of the position, orientation and field-of-view of the camera that captured each frame of footage (referred to as camera calibration data) is important to many aspects of image processing. Without this information, tracking of people or athletes within the real 3-dimensional world would not be possible via image processing techniques alone. For one of the segmentation methods referred to this document, camera information is essential.

This camera calibration data cannot in practice be recorded directly for the vast majority of events. In this project, the camera calibration information is therefore deduced by using image processing techniques combined with knowledge of the real word positions of some features that appear in these images. Multiple, images are used to enable enough data to be extracted to estimate the location of the camera, and to build a ‘map’ of features from which the camera orientation and field-of-view for each image can be estimated.

By using the following information the camera calibration data can be estimated for the majority of frames:

• 3D location of known track markings
• Orientation of features (for example a lamp post is vertical even if its absolute location is unknown)
• Additional measured points
• Constraining the camera to a fixed position
• Optionally: Fixing the camera roll to zero

The camera calibration methods are based on those initially developed by BBC in the MATRIS project and then further developed in the VSAR project.

Previously, the camera calibration software worked from the command line. To increase the efficiency of this possess to make it suitable for use in the My eDirector 2012 project, a GUI has been developed in year 2 of the project. This GUI aids the manual process of matching points and lines in the 3D world onto the image, which is needed to initialize the tracking process and build the map of features that can then be used for automated tracking. Thus camera calibration is now faster.

This year the camera calibration algorithm has been enhanced by the addition of an “ignore region”. This ignore region is the mask from the segmentation method described in the next section. This mask area within the image will not be classified as background by the camera calibration algorithms. This has improved the reliability particularly where athletes occupy a large proportion of the image.

Camera parameters have been produced for every frame for some sequences of the athletic footage captured by the BBC in year 1. These include sequences from the following events:

• Track
  • Men's 100 meters sprint
  • Men's 5km walk
• Field
  • Long Jump

These camera parameters have been made available for use by multiple parts of the My eDirector 2012 system including: segmentation algorithm, enabling 2D person tracking to enter the 3D world, and scene understanding.