specs
specs copied to clipboard
onvif
Add Likelihood attribute to color type
When using a classifier to determine the color of an object the Likelihood attribute describes how confident the algorithm is that the color is correct.
If we make changes such that the color example will look like this:
<tt:Color Likelihood="0.73">
<tt:ColorCluster>
<tt:Color X=”58” Y=”105” Z=”212”/>
<tt:Covariance XX=”7.2” YY=”6” ZZ=”3”/>
<tt:Weight>0.5</tt:Weight>
</tt:ColorCluster>
</tt:Color>
Is this in line with your suggestion?
When the color is determined by a classifier there is normally no weight and covariance:
<tt:Color Likelihood="0.73">
<tt:ColorCluster>
<tt:Color X=”58” Y=”105” Z=”212”/>
</tt:ColorCluster>
</tt:Color>
Conclusion at ONVIF meetings that schema change of original suggestion is best approach, since there is no backward compatible way to have an array of color descriptors.
Nevertheless there are missing extension points. PR #216 has been filed for adding these.
Note that the text still needs substantial rework to clarify both weight and likelihood usage. As indicated the above changes blur the weight approach.
There has been no further change updates either to this PR or the Issue #220 since 1st July.
I am afraid that the likelihood concept is incompatible with the weight concept and multiple colors: the current definition allows to define that 10% of the area is blue + 15% is red each of them optionally with a color variance. With an additional likelihood it is unclear whether the result means that both are alternatives or occur both at the same time.
To me it looks like the weight concept cannot be mixed with the likelihood concept. If likelihood is important it would need to be on the level of ColorDescriptor to allow multiple sets of color clusters with different likelihoods.
There is now a separate example using Likelihood as it does not make sense to combine it with Weight. So I think the requested change has been done.
