The most common failure mode in a camera system is not a camera that stops working. It is a camera that sends so many pointless alerts that everyone turns the alerts off.
What basic motion detection actually measures
Older and cheaper motion detection watches for pixels changing between frames. It has no idea what caused the change. From its point of view, a person climbing a fence and a branch moving in the wind are the same event.
So the list of things that trigger it is long and mostly boring: headlights sweeping a wall, rain, snow, a spider on the housing, shadows moving across a lot through the afternoon, and every cat in the neighborhood.
Why it gets worse at night
Infrared illumination attracts insects, and an insect a few centimeters from the lens fills a large part of the frame. That single mechanism accounts for an enormous share of the three in the morning alerts people get, and it is why night alert volume is so often many times daytime.
The fix that works
Classification rather than pixel change. A system that decides whether the thing that moved was a person or a vehicle before alerting removes most nuisance triggers at source. It is the single upgrade with the largest effect on whether anyone still trusts the alerts a year later.
The fixes that help but are not enough alone
Zones, so the camera ignores the road and watches the gate. Useful and worth doing, but a zone still cannot tell a person from a shadow inside that zone.
Sensitivity, which is genuinely a trade: turn it down enough to stop the false alerts and it starts missing real ones. Sensitivity tuning is what you do when the underlying detection is not good enough, and it treats the symptom.
Schedules, so a warehouse only alerts outside working hours. Sensible, and it does nothing for the hours you care about.
The number that tells you the truth
Ask how many alerts the system sends in a normal week, and how many of those anybody looked at. If the second number is close to zero, the alerts are decoration and the system is really a recorder you consult after the fact. That may be fine, but it should be a decision rather than a disappointment.
Who receives them matters as much as what triggers them
An alert with no owner is not an alert. Plenty of systems are configured to notify a general mailbox, or every manager at once, which reliably produces the assumption that somebody else is dealing with it.
Worth settling at handover: who receives which alerts, at what hours, and what they are expected to do. A camera on a dock door that alerts overnight needs a person who is awake or a service that is, and if neither exists it should be recording rather than alerting.
It is also worth deciding what happens when the alerts stop being useful. Systems drift, and the right response to a rising false alert rate is to change the configuration rather than to mute the notifications, because muting is invisible and permanent.
The test at the end of the first month is simple. Ask the person receiving them whether they still read them. Their answer tells you whether the system is working better than any setting does.
An alert nobody trusts is worse than no alert, which is why Elite Security would rather set up fewer and better ones. There is more like this in the blog, and AI analytics and LPR covers the detection side.