Computational Sports Broadcasting: Automated Director Assistance for Live Sports

Since its introduction in 1927, live TV sports broadcast has grown into a major form of entertainment. One crucial part of a sports broadcast is the director, whose job it is to select the best possible camera from all available cameras at all times: the story should be consistent with the announcer’s description and complete, important events should be emphasized, the camera transitions should be perceived as natural by the audience, and the overall video composition should be aesthetically pleasing. However, with a large number of cameras in stadiums, merely scrutinizing all camera signals at the same time can already be quite overwhelming. As more cameras are introduced, this problem will only become more challenging. In this work, we present an interactive, intelligent and intuitive system for sports broadcasting designed to assist directors in finding and selecting desired camera views from a large number of possible cameras. We propose two different approaches to determine a camera ranking using machine learning techniques. Our first method is based on cinematographic rules supported by user data from a general audience, whereas our second method is able to respect a director’s artistic style based on the director’s previous footage. Both methods show promising results, demonstrate the potential of data-driven approaches for camera selection. A topic that becomes increasingly relevant as the number of cameras increases. Our method aims not to replace the director, but to simplify their task by removing irrelevant cameras, and ranking the top choices, allowing him or her to make the final broadcast choice from these possibilities.