Gaussian cone modeling for background.

[pv33]

Have full cone modeling or truncated cone modeling.
Most likely requires two different variance models and mean estimation methods.
Shouldn't it be some kind of exponential L1 constraint for z and exponential L2 constraint for (x,y) in the cone coordinates?
Sounds about right. will have to code this up pretty quickly if want it to work.
[Do I need to sketch it out? work out math in a Notebook?]

Call it truncated cones. More permissive of illumination changes.

Or maybe just better to do in another color space? HSV is already conical, no? What's faster though? Especially with ROI filtering with precomputed extraction and insertion using vectorized computation. (That needs to be worked out too)