Databases Reference
In-Depth Information
One application is to simulate the movement of the camera as real-time 3D CG
animated camera icons in a virtual space. Users could appreciate the movement of
3D CG animated camera icons in the virtual space from arbitrary viewpoints. If we
click a 3D CG animated camera icon at a moment, the video corresponding to the
current 3D CG animated camera icon is replayed on a CG rectangle plane appearing
in front of the 3D CG animated camera icon in a virtual space (Figure 3). The
zoom ratio of the camera is used for determining the distance from the CG rectangle
plane to the CG camera for replaying a video. Thus, we can appreciate spatially
both the movement of the camera and replaying videos in a 3D virtual space.
The other application is to use 3D CG still arrows, each of which represents
each segment of video sequences. In the experiment, a video sequence was divided
into meaningful segments of video sub-sequences by hands. 3D CG still arrow
icons representing the segments of the video sequences were automatically generated
by visualizing the average values of the position, direction and zoom ratio of the
camera. The length of the arrow represents the average value of the zoom ratio of
the camera (Figure 4). We can walk through the 3D CG still arrow icons which
address video sequences and indicate the average values of the position, direction
and zoom ratio of the movement of the cameras. If we click one of the arrow
icons, we can appreciate replaying videos in a virtual space (Figure 5). In this
application, the camera to view the virtual space is fixed in the same position,
direction and zoom ratio of the real-world camera. We are guaranteed to see the
video in the middle of the scene in the virtual space. Compared with the previous
application, we appreciate a video from a right angle, but the position and angle
of our view cannot be changed in this application. Furthermore, we could have a
wider view compared to only a video being played in another window on a screen.
The video can be augmented as a wider view and be imposed on a virtual space so
that users can experience the video more spatially.
4
Frame Grouping
Since spatial data, that is, the position and direction describing the movement of a
camera are available, video data can be clustered depending on the movement of the
camera. We can tell different scenes using the discontinuity of the movement of the
camera. Even if the position of the camera is continuous, the movement of the
camera is clustered by means of its acceleration. For example, a camera translates,
then it stops and stays at a fixed point, then starts rotating. In this example, the
movement of the camera is divided into three phases: translation, stay, and rotation.
In order to cut some scenes, we introduce an algorithm which compares successive
video frames for grouping them. Atomic component of the video data is a video
frame. Scene cutting is realized by grouping video frames into some scenes. Figure 6
shows the results of frame grouping using our proposed algorithm. The upper bar is
the result using the recognition of humans. The lower bar is the result by our
algorithm. The two bars show that our algorithm works well.
