My opinion on this: it's going to be very difficult to create an algorithm that can reliably (no false positive, no false negative) detect the steel rods in your images. There's not enough difference between the rods an other impurites or other artefacts in the image.
The first thing you should be looking at is changing your image acquisition method to make the image processing easier. Perhaps you can change the lighting method so that the rods come out more contrasted. Maybe use structured lighting but I'm not sure it would help in your case.
Another option would be to use an infrared camera instead of a standard one. I believe the emissivity of concrete is much greater than steel (depending on the steel) so the two should look vastly different in an infrared image.
