Docs and Info,
When multiple cameras are used to capture an activity in a fixed place, how many video images should be displayed to people who want to view it ?
There is an approach which displays all the video images captured by multi-camera system to viewers. This idea is derived from the viewpoint that it is good for the viewers to receive as much information as possible. However, as the purpose for viewing the videos is not to watch the place closely but to see the activity there, some of the videos are insignificant and useless for them.
Therefore, we first consider behavior of people in visual sense when they come across an activity in real world. They usually focus on a part of the activity because they can see only one subspace at a time. To do that, they move their heads and eyes so that they pay attention to their surrounding environment.
Therefore, it is desirable to generate only one video sequence to the viewers although multiple active cameras are used behind. The video is to be generated by controlling directions of the active cameras and selecting the best camera according to a situation measured by the system.
In our approach described from the next section, we use active cameras which can pan, tilt, and zoom so that our method can realize the functions like what real cameramen do.