I think it all depends on where the test statistic (KSI/fatalities/etc) originally lay in comparison to the mean value.
If the speed camera is removed when the test statistic was low in comparison to the mean, then regression to the mean will occur and will make it look like the removal of the camera has caused more accidents.
If the test statistic was high in comparison to the mean, then regression to the mean will happen in the opposite direction and it will appear that removal reduced accidents.
If the test statistic was near to the mean when the camera was removed, then taking an average for the next few years should hopefully provide a reasonable conclusion on whether or not that camera reduced, increased or had no bearing on accidents.
I hope that's all correct
.