|
|
Implementation:
|
|
|
|
- Semi-semantic Video Processing for Video Indexing
We shall employ the technologies developed in UIM/118, such as motion detection
and color detection, to identify the objects of interest in this semi-semantic video
processing. In addition to that, we shall also extract additional descriptors for shape,
color histogram, edge distribution, etc. in this level of processing to reduce the false
retrieval hit as much as possible. Major efforts will be devoted in finding good
trade-off between computation complexity in this level of processing and the
accuracies in identifying candidate video of interests.
- Video Indexing Scheme for Efficient Video Retrieval
We shall investigate different description and indexing schemes such as MPEG7,
LASeR (Lightweight Application Scene Representation, MPEG4-Part 20), etc. to
see which one of them could be best utilized for retrieval of surveillance type of
video. Although a direct implementation of existing description scheme is feasible,
it may not be efficient to have a full implementation of such description scheme for
real-time processing. As such, we shall investigate how these description schemes
can be trimmed down to meet the real-time constraints.
- Conceptual Query to Semantic Description Translation
We shall implement this translation technology as a module with user-friendly
graphical interface that enables user to define conceptual query in an easy and
intuitive way. The module that takes the user inputs and put them all together to
generate semantic description for retrieval purpose. We have been quite successful
in this kind of development as demonstrated in the project UIM/118, in which we
have developed an application generator which takes user input to formulate a set
of complex motion detection rules (descriptions). In this project, we shall take one
step further to make this module to be able to translate for descriptions in a broader
domain, which encapsulates the descriptors generated from the semi-semantic
processing.
- Video Data Caching for Reliable Hard Disks Operations
To implement the video data caching, we shall analyze the hard disks read-write
access pattern in a video recording servers, which is supposed to be quite different
from traditional file read-write access pattern. Based on this analysis, we shall
determine the optimal cache size, caching structure and algorithm for filling and
flushing of the data cache.
|
|
|
|
|
|
|
