Ping

[Aaron Brown]

Howdy,

Is this group alive? I'm just getting started on OpenCV and I've
got a few questions, particularly around Haar Classifier best
practices.

To whit:

Can Haar classifiers be "chained" very well? (That is, can I
check an image for instances of classifier A then, if there are none,
instances of classifier B, etc.) I'm trying to locate instances of
>100 different kinds of objects, fairly specifically, and I'm not sure
if Haar classifiers are the way to go.

Any advice on building good databases of positive images?

How many images? How many "uniques" vs. how many created vs.
CreateSamples utility?

How much variation of the images? (e.g. Should there be images
in
various rotations, tons of different angles, etc.?)

Any general advice or gotchas on taking / selecting / collecting
positive images?

As for negatives, I'm using the negatives collection at:
http://tutorial-haartraining.googlecode.com/svn/trunk/data/negatives/
Has anyone else used this collection? And if so, have you had good
luck with it?

-AMB