Lead by a team from Disney Research and the California Institute of Technology, computer engineers are teaching sports cameras to better broadcast games autonomously by watching the actions of sports videographers and how they anticipate and follow the action, and recover from errors.
While cameras can already follow the players to some extent, they can’t yet follow the ball, or at least not smoothly. Sports like soccer that lack easily identifiable player positions are even harder for these automatic cameras to record. Researchers believe that by teaching the software using actual human videographers, they’ll be able to develop an algorithm that can record sports broadcasts without the jerkiness that current automatic tech produces.
“This research demonstrates a significant advance in the use of imitation learning to improve camera planning and control during game conditions,” Jessica Hodgins, vice president at Disney Research, said. “This is the sort of progress we need to realize the huge potential for automated broadcasts of sports and other live events.”
During field tests, senior research engineer Peter Carr said that the computer was less aggressive than human operators, but that in a few cases, the computer actually performed better. For example, he explained, on a fast break in a basketball game, the videographer incorrectly anticipated a dunk while the computer, reading the player’s positions, correctly prepared for a pass play.
“Having smooth camera work is critical for creating an enjoyable sports broadcast,” he said, “The framing doesn’t have to be perfect, but the motion has to be smooth and purposeful.”
While the method of teaching computers through human observation is an intriguing one, the potential of automated sports broadcasts raises a few concerns. First, the technology could mean that some of the human operators the cameras learn to imitate could be out of a job. And even with human operators watching, how well can a computerized camera really make split-second decisions involving an infinite number of possibilities?
While those questions have yet to be answered, the researchers are expected to present their project at the IEEE Conference on Computer Vision Pattern Recognition on Sunday.
