1) Easy access to good vision libraries (in particular OpenCV and ITK)
2) Fairly easy to parallize algorithms w/ OpenMP or Intel's TBB.
3) Compilers (well, Intel C++ anyway) are good at vectorizing code.
4) No garbage collection. I can work with fairly large images and I don't worry about g.c. (This may be a conservative hangup of mine. As of a couple years ago anyway g.c. would run out of memory when crunching though a set of large images).
5) Decent GUI libraries for cross platform interactive GUIs (wxwidgets and QT in particular).
(And, to be honest, it's also the language I know best and I'm most comfortable in).
Interesting. What is the advantage to C++ in the context of CV that outweights the long compile time for you?