Contents of this page are obsolete. This page is preserved and stored at this URL just from historical viewpoint. Original URL was http://www.mm.media.kyoto-u.ac.jp/members/kameda/...
Please visit www.kameda-lab.org for recent information. (2002/12/06, [email protected])

Publication, Research, Docs and Info, KAMEDA


next up previous
Next: Automatic Camera Selection Up: Control and Selection Previous: Discussion for Video

Automatic Camera Control Method

One of the features of our approach is to introduce observation cameras. There are two kinds of cameras in a classroom in our approach. One is observation camera, and the other is shoot camera.

The observation camera is like an human cameraman's naked eye. Its role is to follow the target to shoot. The cameras for this function usually zoom down at their limit so as not to miss movement of the target in the classroom.

On the other hand, the shoot camera is just like a video camera which human cameraman uses. The shoot cameras are forced to turn right and left, up and down, and zoom in and out to obtain good video image against the target.

Although the observation camera is like cameraman's eye and the shoot camera is like the video camera, we have to consider which shoot camera should be used when an object is observed at certain observation camera. We call this description ``camera control relation.'' A representation of the camera control relation has two fields. One field is observation camera, and the other is a list of controlled shoot cameras. We draw this relation in Figure 1.

  
Figure 1: Camera Control Relation

When the observation camera detects movement of an object, the shoot cameras under control of that observation camera changes their direction so as to place the object in the center of their video images. Note that one observation camera can have multiple shoot cameras, and observation cameras can share the same shoot cameras in their camera control relation.

By sharing the shoot cameras, each observation camera can generate wide variety of video images because they are taken from different camera locations. As a drawback of camera sharing, there may be a situation that one shoot camera is forced to shoot two (and more) objects at the same time if different objects are observed simultaneously at the two observation cameras and they share the shoot camera. This kind of situation is inevitable, but we can reduce the possibility of this situation to practically low level by assigning the shoot cameras under the rule below.

For example, it is acceptable to share shoot cameras among the observation cameras if they are to observe a lecturer, and there is only one lecturer in the room.

There is also a desirable rule on configuring the cameras in the classroom. That is :

A viewing volume is a subspace inside which an observation camera can watch objects with its viewing angle. Since normal video cameras can be modeled by perspective projection, a shape of the viewing volume is a pyramid shape. (See Figure 2.)

  
Figure 2: Viewing Volume

Object detection is done by inter-frame subtraction. Therefore, the observation cameras cannot change their directions during the detection. And, pan and tilt parameters for the shoot cameras are determined according to the location of the detected movement on the image planes of the observation cameras[1].



next up previous
Next: Automatic Camera Selection Up: Control and Selection Previous: Discussion for Video



Yoshinari Kameda
Tue Oct 5 01:12:48 JST 1999