Autonomous camera systems make small events, such as lectures and amateur sporting competitions, available to much larger audiences.

Traditionally, autonomous camera systems follow an object-tracking paradigm, such as “follow the lecturer,” and implement camera planning smoothing the data from the object tracking. The output videos produced by such systems tend to look robotic, particularly for dynamic activities such as sporting events.

To overcome these problems, Disney has developed techniques for an autonomous camera system to more effectively mimic a human camera operator. Basically, a supervised regression problem is applied to formulate camera planning.

For example, the techniques developed by Disney may be used to record a basketball game. The automatic broadcasting application receives video inputs from two cameras capturing the same event. A first video input is captured by a human-operated camera. A second video input is captured by a stationary camera with a wider field of view, such as a camera with a higher vantage point of the game or with different lens settings. Then, the automatic broadcasting application extracts feature vectors from the second video input and exemplary camera pan-tilt-zoom states from the first video input. In videos of basketball games, it may be assumed that tilt and zoom are constant, as a wide shot is typically maintained with a mostly constant tilt angle. Accordingly, the automatic broadcasting system determines pan angles of the first human-operated video input.

Once trained, the automatic broadcasting application controls motors in the autonomous robotic camera to achieve the planned pan-tilt-zoom settings and capture a video.

These autonomous cameras will be able to record dynamic situations like sporting games without any human supervision.

us20160277673