There is also the issue of recognition software that doesn't have too many false positives, and properly tuning it for each crossing, in all light situations. The amount of time and work needed can really be very large. I'm not against it though. Some pilot projects would be a good idea.