return to the home page Media Processing Framework

Time Of Interest Component (mpf-toi)

Published June 2, 2009, 11:21 am | by mbiggs

Published June 1, 2009, 5:44 pm | by shawn_mcmurdo

I have checked in a first version of the mpf-toi (Time Of Interest) component.  This will allow for segmenting or analyzing media streams based on the stream time.  The source code is in MPF/trunk/components/mpf-toi in the svn repository.

The mpf-toi component currently specifies the time of interest by setting the following parameters.  In future versions in may be interesting to add other methods of specifying interest such as mathematical functions or user provided functions.

Input
The input parameters accepted are:

start: The start time in seconds. Default is 0.0.

end: The end time in sedonds. Default is -1 which indicates the end of stream.

interval: The repeating interval in seconds. Default is 0. O means it does not repeat.

duration: The duration in seconds. Default is 0. O means a single frame only.

in: The interest level when inside the specified times expressed as an integer from 0-100. Default is 100.

out: The interest level when outside of the specified times expressed as an integer from 0-100. Default is 0.

Output

The mpf-toi component outputs a simple RDF graph containing the current level of interest as follows:

this rdf:type appscio:interest

this appscio:currentInterestLevel <value>

where <value> is the current interest level.

Any thoughts on whether the "in" and "out" parameters should be floats beetween 0 and 1 rather than ints between 0 and 100?

Categories:

Two comments: * 'in' and

Submitted by wstidolph on Sat, 06/06/2009 - 12:20.

Two comments:

* 'in' and 'out' sounds like flows or directions, rather than  levels - you should rename the params ('interest' and 'outerest'? :)

* I'm comfortable with ints at this point; I expect we'll have to quickly revisit the concept of interest so keeping it simple now is good.

FWIW, I'm thinking we'll need to generalize the concept of 'interest' to a vector showing elements of interest as contributed by multiple recognizers, and we'll need to overlay some duration-concept so we can easily express complex conditions like "(interest.face>20).for 5.seconds AND interest.keyword>50"