Databases Reference
In-Depth Information
8
Spatio-temporal Browsing
for Video Databases
Masatoshi Arikawa
Center for Spatial Information Science
The University of Tokyo
ABSTRACT
Time-series spatial description data concerning a camera's movement are expected to
be automatically generated by various sensors in the near future. Such description
data as a camera's time-series positions, directions and zoom ratios will provide a
rich environment for retrieving and browsing video data spatially. Real-time 3D CG
(three dimensional computer graphics) is used for user interfaces to browse videos in
a virtual space corresponding to an existent space in the real world. Cameras'
movements or video sequences are represented as 3D icons in the virtual space. If we
click one of the 3D icons, the corresponding video sequence will be replayed in the
virtual space. This paper presents a basic principle of the 3D spatial hypermedia for
video data browsing, and shows some demonstrations of our prototype system. The
basic procedure for producing virtual spaces dynamically is realized by dynamically
retrieving and visualizing spatial data stored in spatial databases using the rule of
LoD (Levels of Detail)
. In addition to the spatial data,
time
data automatically
generated as description data of the video data are useful for video retrieval and
structuring. This paper introduces a new concept
time walk-through
for retrieving
video data using
time
dimension. The concept enables users to travel time in a virtual
space. The time walk-through is based on “time” extension to LoD. Our prototype
system “TimeWalk” based on our proposed framework is explained with real
examples.
1
Introduction
There is a large amount of video data produced by electronic consumer products
such as video cameras and personal computers. Finding parts of our interests
from a large amount of video data becomes a big problem. Some researchers have
tackled the problem using various technical methods, such as keyword searches,
natural language processing
5)
and pattern recognition
2)
. As an example of keyword
searches and natural language processing, all news programs on TV in the United
States are required to have a channel that provides superimposed dialogue for
114
