Yes. I checked the Wikipedia article on the Skellam distribution. It answers your question on two conditions:
(1) You can accept the Poisson model for traffic on both roads over the course of a year. Of course if you are worri8edd about the assumption you can look for over-or-underdispersion as Simon says.
(2) You assume independence between accidents on the roads. But perhaps there are factors creating correlation between the two roads. What come to mind is the weather conditions which would be almost the same on both roads given they are in the same local area.
If you have data on these factors then a Poisson regression model might help again as Simon says. But if there is no correlation then I think the inference could be based on a test that the mean of the Skellam distribution is zero.
-------------------------------------------
Michael Chernick
Director of Biostatistical Services
Lankenau Institute for Medical Research
-------------------------------------------