This is a preliminary interface to a search engine for retreiving
video clips based on contents and actions. Most clips are of humans
doing various actions. This research is sponsored by the
National Science Foundation
and the
Army Research Office.
Some details of the first year program are included
here.
You will need viewers for mpeg, quicktime and avi to actually view
clips if you choose to download them after the search.
-
For Unix: xanim is one
of the most versatile viewers. It can be used to view all the file formats
used here.
-
For MS Windows:
Xing MpegPlay for mpeg videos.
Standard quicktime and avi players generally come installed in MS Windows and
files can be played using the Windows Mediaplayer.
Search the video database with content and action keywords
The Query Interface and Menu
The following interface provides the capability for specifying objects
and associated actions in a temporal sequence. The temporal depth
provided by the menu is 3. Additional simultaneous specifications of
objects and actions can also. Currently the simultaneous number of
actions that can be specified is also three.
A Java based GUI is under development, which will permit
unrestricted models of hierarchical and temporal specifications of video
content.
THEN sets the temporal sequence in the the query specification.
AND specify action object sets occuring simultaneosly.
Example queries:
-
Subject: man Action: ride Object: bike THEN
Subject: man Action: on Object: bike AND
Subject: bike Action: jump Object: ramp THEN
Subject: bike Action: land
-
Subject: aircraft Action: fly AND
Subject: man Action: jump THEN
Subject: man Action: freefall
Optional backgrounds can also be specified as in:
room or sky
Webmaster: Arvind Mohan
Feedback: benarie@ece.uic.edu
© 1998-2003 Machine Vision Laboratory.