[Japanese | Thesis | Researches in Minoh Lab | Minoh Lab]
It is possible to participate in a lecture from a remote place thanks to the power-up of computers and speed-up of the transmission lines. Video media plays an important role in viewing a lecture at a remote place. As each person wants to see different objects in a lecture, it is desirable to provide different videos appropriate for each person.
In this paper, we propose a real-time method of making multiple lecture videos appropriate for multiple users using multiple cameras.
A director and cameramen usually imagine an average taste of favorite imaging ways among many users on producing video of an event like a lecture. In this research, we realize the virtual director and the virtual cameramen and propose a method of producing video which each user hopes.
Since favorite imaging ways of the users are not unique in each situation of a lecture, each user is required to describe his/her favorite imaging way in each situation. We call this description ``camera-work''. A camera-work consists of 3 elements: object, camera angle, and camera range. Our proposed method of producing video for multi-user is realized by the following procedures. The virtual director snatches a situation of lecture. According to users' camera-works requested in a snatched situation, the director assigns camera-works to virtual cameramen and switches the videos of users. The cameramen shoot objects under the camera-work assigned by the director.
In our realization, there are two problems that should overcome to achieve our purpose. We focus on these two problems and explain how to resolve them.
The first problem is that not all the camera-works requested by the users can be satisfied in some situations. One camera can realize only one camera-work and the number of the cameras is usually less than the number of camera-works which all the users request. So we propose a method of searching for a set of the camera-works which satisfies as many users as possible. We call this method ``camera-work mediation''.
The second problem is steep image change that would occur when the situation changes. In response to the change of the situation, a video to a user should sometimes be switched. In this case, the user may fail to recognize camera angle and object in his/her video because the location of the shooting camera jumps and steep image change occurs. In order to prevent the steep image change, we define distance between videos and propose a method of selecting the least distance path in video switching. We call this method ``smooth video switching''.
We evaluate validity of ``camera-work mediation'' and ``smooth video switching'' by experiment and questionnaire.
In the experiment for ``camera-work mediation'', we produced a lecture video in 8 minutes for 16 users using 8 cameras. We applied our prototype system and got the number of users who satisfied with his/her video in the lecture. As a result, our method satisfied 10.3 users against 11.3 in theoretical maximum. The result showed that our method had almost the same ability to search for the most appropriate camera-work set.
In the questionnaire of ``smooth video switching'', we prepared 6 pairs of videos and asked 9 users to select a video which he/she feels mild video out of two videos. 72.2% of users' selection is matched with the selection by our method.