In Internal Testing
What it does:
This model flags video of rain, snow, or condensation that will make computer vision highly unlikely to be successful. It can be set to send an alert when it occurs, so that a manual check, like a security guard walking through an area, can be initiated or so that the maintenance team could determine an appropriate strategy to deal with the root cause, such as an overflowing gutter or lack of an overhang.
This model also will flag sections of time in 24/7 footage view where the camera view was obscured and where accurate detections are unlikely to have taken place.
Why it is needed:
Not all conditions are conducive to computer vision. If the camera cannot see an area, survail cannot accurately assess whether people, vehicles, or other objects are present. This is most obviously the case when a camera is covered in rain, snow or condensation while also in night vision, infrared mode recording (as seen in the image above). The possibility of getting good data in these conditions is very poor, as the camera really cannot see anything very well. In the image above, it is incredibly difficult, even for a human to tell if the lights are a vehicle's headlights or a building's flood lights, because there is so much water on the camera lens.
Interaction with on-camera vandalism alerts
Our cameras already have a series of Vandalism Detection Analytics that send you an alert when the camera is experience either a loss in function or a loss (if the cable is cut or the camera smashed) or change in viewport (if someone covers or paints the camera or just points it towards a wall). These alerts usually result in no footage being recorded or no events being detected. A camera with water on the lens, will continue to create record, but is very likely to not know if people or vehicles are in them.