If you have no denominator data for the sites (eg no of vehicles passing/unit time, population density of area) you could use a comparison of Poisson variables ,using crash counts in successive time periods as basic data.
If you have denominator data you could use comparison of proportions with exact binomial statistics or normal approximation.
Compare only two locations makes sense if they are two accident scenarios equal (as prototipal accidents) or two different scenarios. In any case, a number from count of traffic flows that can be used as denominator could give more meaning to the word «frequency». good luck