From what I've seen, these can be spotty in terms of reliability. You're basically looking for a variation of scene changes in most cases. If the scene is fairly static and has consistent lighting, it's easy. If the scene has dynamic lighting (car headlights, sunrise/sunset shows or glare, etc) it can be more problematic.
We (VideoIQ) tackle this by using a loitering alarm... Look for someone in front of a grafitti-prone area for more than X minutes. Yes, it won't catch them immediately if the person walks up and immediately starts painting, but neither will most other systems. You will however reduce the probability of false alarms to a point that it becomes practical to monitor and respond.
The other thing with graffitti vs. people detection is just pixels on target. I can cover 200' of wall space pretty reliably, obviously based on having some flexibility on camera placement. I can have a direct view of the wall, or an off-axis view because in either case the person is going to be in front of the wall, and fairly easy to detect relative to spotting pixel changes/paint sprayed on the wall. Many of the graffiti detectors will require less coverage area per camera and more direct views of the area for reliable operation.
That's just my experience/opinion. I will say that this hasn't been a particularly high-demand item so it's very possible there is some new/cool/highly advanced system out there I'm not aware of yet, but I doubt it ;)
Ethan mentioned this a while back but no one had any knowledge or comment on it: Merlin Graffiti Detection Systems, who claims that it is "instantly triggered by spray can signature" and that "Advanced heuristics and algorithms allow the Merlin to identify and characterize a graffiti event by intelligent recognition of an aerosol fingerprint."
IPVMU Certified | 10/21/13 05:30am
The Merlin was made by Potter a very old name in the alarm business. I talked to them about it at ASIS or ISC West (can't seem to remember which) and was told they no longer offer it (although your link still seems to work). The guy told me there were too many false alarms with the sensor.
Also, you should be able to stack rules in f the analytics i have seen. So Virtual Trip Wire + Loitering + People + movement = alarm. But obviously, thats only if the camera can see.
Bosch's in-camera IVA system for detecting graffiti.
Here's their demo video:
I wonder about the angle of incident. Do you need to place the camera like that? If so, it could get expensive, and logistically complex to put a camera away from a building, pointing at the building.