Need advice detecting object (which method)

edit

I am trying to detect bowling pins within an image.

First I tried with TrainCascade, For the positive images I took images from the entire pin. I was able to detect single pins that are clearly separated from the other pins, however pins that are partly visible could not become detected.

Then I took only the head of the pins for as positives, but it seems that then the TrainCascade does not perform very good (actually very bad)

I am not sure which method is best to detect thse pins. I also made tests with the feature-detector and homography, but here I have the problem of finding and separating multiple objects.

then I tried to "manually" analyze the picture: 1) Removed the background color 2) Removed noise but this ends up in a pattern that is still to be analyzed ..

Any help is very very welcome