A. Amer, 
"Object-based video retrieval based on motion analysis and description",
Tech. Rep. 99-12, INRS-Télécommunications, June 1999, 

Finding video of interest in a large database  is a problem of
increasing importance given the ever-increasing amount of available
video information. In this presentation, an approach for retrieval of
video content in a video database will be introduced.
Here, the user can give a qualitative description of a query video by
specifying video-global features, such as global motion, and  object
features, such as  basic feature (e.g., color), spatial relationship
features (e.g., object i is closed to object j), location features
(e.g., object i is in the bottom of the image), and semantic features
(e.g., object action: object i moves left and then disappears).


An advantage of such a retrieval strategy is that  it allows the
construction of intuitive queries oriented to the observations that most
people's interpretation of real-world domains is vague and ill-defined
and  that users usually   memorize objects (``who'' is in the scene),
their action (``what'' he/she is doing), and their location (``where''
the action is done) while viewing a video. In the absence of a specific
application, such a generic model allow extensibility (e.g., by
introducing new definitions of object actions).


The emphasis of the proposed retrieval approach is, therefore, on video
analysis and interpretation methods allowing linking of basic visual
features to high-level semantics (e.g., actions and events). Since in
retrieval application, fast responses are expected, main component of
the object-based retrieval are mainly devised for on-line retrieval
applications. For that purpose, fast approaches for image segmentation, 
motion estimation, object tracking, and object feature interpretation are 
proposed.


Keywords:
Content-based video retrieval, Spatio-temporal object detection, Image
segmentation, Morphological operators, Feature extraction, Motion estimation, 
Object tracking, Similarity measures.