The best approach to detect slightly different objects

edit

Hey guys,

I just wondering... After saw a thousands of videos about "object detection" I could have an idea of how efficient haar cascada can be be in order to detect objects/face/eye.. .etc

But came to my mind the question: "How can I detect two or more objects that are slightly different of each other?" I can give you an example: WHEELS. Wheels are circular format and basically only the style is different.

Wheel 1: http://www.tirerack.com/images/winter...

Wheel 2: https://www.carid.com/images/dorman/f...

How haar cascade could be efficient to detect those images?

Thanks