![](bg3.png)
2.3 Camera Control Strategy
What does ACE capture? One solution for this problem is to take the scene that students
want to watch, but many scenes are probably requested by many students at the same
time. Although this solution needs the consensus of all students, it is very dif cult to
make it. We have decided, therefore, that ACE captures the most important thing from
a teacher’s point of view. When we designed the previous version of ACE, we assumed
that the most important thing is the latest object written on the blackboard. The previous
version of ACE took a shot zoomed in on the object after the teacher had written it on
the blackboard. After a-few-second zooming, the previous version of ACE zoomed out
and take a shot containing the latest object and a region near it. However, this strategy
has some problems. The previous version of ACE cannot take a suitable shot when
the teacher explains object written before. The lecture scene captured by the previous
version of ACE changes at short intervals because the latest object is often found when
he/she goes on writing objects on the blackboard for a long periods of time. Such a
video is not appropriate for students who take part in the distant lecture.
We have adopted, therefore, the strategy that ACE captures an object explained by
a teacher. When the teacher is writing object on a blackboard, ACE captures the teacher
and the object. When he/she is explaining to his/her students, ACE captures the latest
object. If he/she is explaining the objects written on the blackboard before, ACE also
captures them. On the other hand, the image processing component on the PC checks
whether the next target which ACE should capture is included in the current capturing
area or not. If the next target is included, ACE need not move the active camera. This
solves the problem that the lecture scene changes at short intervals. When he/she is
explaining the objects on the screen, ACE captures whole of the screen even if the
objects are anywhere on the screen.
3 Teacher’s Behavior Model
3.1 Teacher’s Behavior in a Lecture
We observed some lecture videos (Table 1) to nd out what behavior a teacher is doing
in his/her lecture. We found out that the behavior of the teacher could be categorized
into three kinds: “Writing”, “Explaining” and “Moving”. When the teacher was writing
some objects on the blackboard, we categorized his/her behavior as “Writing”. When
the teacher was explaining objects on the blackboard or on the screen, we categorized
his/her behavior as “Explaining”. We categorized the other kind of behavior as “Mov-
ing”. Table 1 shows the ratio of each behavior in the lectures.
3.2 Creating Teacher’s Behavior Model
We got the position of the teacher’s centroid g(t) = (g
x
(t), g
y
(t)), face f (t) = (f
x
(t), f
y
(t))
and hand h(t) = (h
x
(t), h
y
(t)) from the Video A in Table 1 by the hand work in 2 fps.
The positions may be represented as a time series I(0), I(1), · · · , I(T ), where I(t)
denotes the position of the teacher’s centroid, face, and hand at time t.
I(t) = (g
x
(t), g
y
(t), f
x
(t), f
y
(t), h
x
(t), h
y
(t)) (1)
17