How does the decision of a feature is a feature of our object?

Question

Can anybody explain how does OpenCV make a decision about a feature of an object, when doing train_cascade???

Answer 1

This is in the train_cascade documentation

opencv_traincascade supports both Haar [Viola2001] and LBP [Liao2007] (Local Binary Patterns) features.

If you don't know what those features are, you can read the papers listed at the bottom of the documentation page or, to get a basic idea, Wikipedia: Haar-like features, Local binary patterns

You select which feature is used using the -featureType<{HAAR(default), LBP}> option to the opencv_traincascade command.

Or perhaps you are asking about how the actual feature selection occurs. OpenCV uses variants of AdaBoost. Here is a useful presentation on AdaBoost.