You say "the video will be streamed from the cameras on site." I kind of doubt this is what's happening. If so, they'd need to identify the bandwidth requirements.
The ideal architecture here is to use object recognition on camera with a classifier that isolates the license plate (or at least indentifies candidate frames that include a vehicle). Then send images to the cloud for recognition--most likely using a machine learning system.
Seems like that type of implementation would lead to the narrow camera requirements listed.
Maybe I'm giving Genetec too much benefit of the doubt, but that's a good architecture as it allows many tags to be identified most efficiently, and the cloud-side recognition can be tuned and modified without need for on-site intervention.