Abstract
Many professional soccer sports clubs analyze their games either manually or using existing analytics tools.
However, the existing software often requires much manual work and spending time looking through video tapes.
Researchers and students from Simula Research Laboratory, The University of Oslo, and The University of Tromsø have created
a real-time analysis system prototype called Bagadus. It is aimed at being automatic, easy to use and integrated with both annotations, recording software, video processing and sensor-networks. The latter is providing heartbeat, speed, position and various other statistics from the players.
Bagadus capture video footage from multiple angles, and delivers a stitched panorama video. A natural extension of a system using many cameras from different angles, is a free-view functionality.
Free-view and virtual cameras means that we reconstruct an image from an imaginary virtual camera. This image is reconstructed using data from nearby real cameras.
To reconstruct the scene as seen from the virtual camera, the depth of the pixels from the nearby real cameras must be known.
In this thesis, we show look at how we can improve the Bagadus system by camera calibration and stereo vision for depth map generation needed by the virtual camera application.
We show how cameras can easily be installed and the system configured for the free-view, and how depth maps can be generated
in real-time. Finally, we discuss see how results from this work can be applied to the Bagadus system.