Introduction

As WWW technology becomes popular today, one of the main issues on serving WWW is how to create WWW content and what kind of contents we should put on WWW. Our research and development division of the Center for Information and Multimedia Studies (CIMS) in Kyoto university works on this issue for several years and have been developing new methods that can make multimedia contents from real world automatically.

We mainly take up the scenes that are related with human activities and can be commonly seen in universities. For example, we have daily classes and student activities anywhere in campus in our university.

To let people view such activities on internet, various kinds of presentation techniques have to be used. Video and audio media are typical and essential media among them because we usually understand the human activities by our eyes and ears.

Our research group studies visual medium first, and then plans to utilize audio medium for generating new type of WWW contents.

Because video medium is usually obtained by video cameras, we consider two types of camera environment for taking videos of human activities.

Multiple active cameras (dense arrangement)
Multiple static cameras (sparse arrangement)

The first type is suitable to capture the events which arises at a certain fixed place. We can apply this type of camera environment to lectures, plays, and some kinds of sports, which are commonly seen in our campus. As these activities consist of more than one person, we have to understand where the people are in the scene and who is shot and which camera is used for the shot. Until today, these issues are handled by arranging many cameramen and directors and so it is difficult to film such activities many times because it is usually hard to prepare those staffs. Therefore, we are asked to design a new method through which multiple active cameras are controlled and an appropriate camera is selected to shoot the activity. This method cuts out subspace within which an interesting behavior is observed and determine the best camera to shoot it according to spatial analysis of the scene.

Concerning the second type, a good point is wide application because it is easy to set each camera in a certain scene. On applying this environment, it is important how to extract attractive visual information and how to show it in video style because cameras are set in the scenes where events will happen but no one knows when they arise. In other words, a method that is applied in this type of camera environment should cut out interesting subsequences in a long sequence of video.

Hence we describe two issues of our research in this paper which are related with the two types of camera environment shown above. The two issues are :

Video camera control and selection of multiple active video cameras
Video skimming for people in rooms for long period

These two issues are especially focused on visual information taken by video cameras.

Currently, the first issue aims to support distance learning in universities. As classrooms are generally too large to shoot by only one video camera and there many persons to shoot, we have to prepare several cameras and control them according to a situation in the classroom. We will mention our approach in Section 2.

The second issue is to support browsing video which is taken by one video camera for long period at a place people go in and out. This is a new approach of using cameras that are connected to LAN. As it is one of the hardest works to review the video for long time, we have developed video skimming method that makes video length shorter. Details are described in Section 3.

Next: Control and Selection Up: Studies of Automatic Video Previous: Studies of Automatic Video

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