Security Cameras & AI Video

How today’s AI cameras work

Newer cameras do a lot more than record. They can sort what they see into people, vehicles and animals, read license plates and, if you choose to turn it on, recognize faces you’ve enrolled. On many systems, that analysis happens on the camera and recorder at your property, so video doesn’t have to go to the cloud to be analyzed (one example: Ubiquiti: AI detections).

1. The camera seesA 4K camera picks out people, vehicles and animals as they move through its view.
2. The recorder sorts itFaces, plates, vehicle types and line crossings are tagged on site and saved with the video.
3. Rules decideYou choose what’s worth an alert: which areas, which hours, which direction.
4. Your phoneAn alert with the clip. Search later for “person” or a plate, and export footage for police or insurance.

Detection features from Ubiquiti’s AI detections chart. Other brands offer similar features under different names.

The virtual fence

A virtual fence is a line drawn across a camera’s view. When a person or vehicle crosses it in the direction you chose, the system sends an alert (line crossing). The trick is where you draw it. We place one-way “tripwires” just inside the property edge, so people walking down the public sidewalk never set them off. Only someone coming in does.

The fence lines can be armed on a schedule, such as dusk to dawn, while lines at the entrances simply log people and vehicles all day so they’re easy to search later. Recording keeps going on site even if the internet drops.

Facial recognition: useful, with limits

Face recognition lets the system tag people you enroll, like family or staff, so search and alerts can tell a familiar face from someone who isn’t enrolled. It’s handy at front doors, gates and entrances.

• It needs a good look at the face. Accuracy depends heavily on image quality: lighting, angle and distance. NIST’s testing also found that accuracy can vary across demographic groups (NIST). So we place recognition cameras where people face them up close, at doors and gates, instead of expecting a wide yard camera to identify faces.
• It’s off until you turn it on. On UniFi, face recognition has to be switched on in the system’s settings (Ubiquiti).
• For businesses, it’s personal information. Under California’s privacy law, facial imagery used to identify someone is biometric information, a kind of sensitive personal information, and businesses covered by the law have to give notice when they collect it (Civil Code 1798.140, California Attorney General). We’re electricians, not lawyers. For a business, HOA or apartment property, please check with your attorney before turning it on.

License plate recognition

Regular cameras can catch plates in good light, but a dedicated plate camera is built for the job. The UniFi AI LPR camera, for example, has a 3× optical zoom and infrared out to about 50 feet, and reads plates on vehicles moving up to about 55 mph (Ubiquiti specs).

Placement matters more than the camera. Ubiquiti’s guidance is to mount it no higher than about 16 feet, read plates within about 39 feet, and keep the angle to the lane under 25°, or under 15° for faster traffic (setup guide). A plate on your list can also open a gate; see our access control and gates page.

Coverage we model, not guess

Before we suggest a camera count, we model the property. On a recent multi-building townhome property, we placed 583 test points every 10 feet across the site and checked each one against every camera’s real field of view (about 110°, with roughly 80 feet of useful range), with the buildings blocking views the way they really do. Then we chose the smallest set of cameras that sees every point. In that design, 73% of the points are seen by exactly one camera, so the owner isn’t paying for two cameras watching the same spot.

We offered three options built on the same design, so stepping up later doesn’t mean starting over. And we confirm coverage on a site walk, because patio fences, trees and parked vehicles can block views that a drawing can’t show.

How much recording, for how long

Thirty days of recording is a common ask. Whether you actually get it depends on how many cameras you have, their bitrate, and whether they record all the time or only on motion. That’s why we put the settings in writing. On the recent townhome design, the 30-day retention was tied to continuous 24/7 recording at an average of 8 Mbps per 4K camera, on a recorder with surveillance-rated drives in RAID 5, which keeps recording if a single drive fails.

The math is simple. 8 megabits per second × 86,400 seconds in a day ÷ 8 bits per byte is about 86.4 GB per camera per day, or about 2.6 TB per camera over 30 days. Try your own numbers:

• 8 baysdrive bays in the UniFi UNVR G2 Pro recorder
• Up to 504K cameras on that recorder
• 10Gfiber (SFP+) port on that recorder

Recorder specs from Ubiquiti. The calculator is arithmetic only.

Fiber, PoE and battery backup

• PoE: one cable per camera. Power over Ethernet carries power and video on one cable. A UniFi G6 Bullet draws under 10 watts that way (Ubiquiti specs).
• Fiber between buildings. All-dielectric fiber has no metal in it, so it can’t carry lightning-induced current or create ground loops between buildings (Fiber Optic Association). Each building still gets proper grounding of its own. We plan the shortest route and use bores under walkways and hand trenching to keep sawcutting to a minimum.
• Battery backup. On the recent design, a pure sine-wave battery backup on the rack (about 30 minutes, with power-fail alerts) plus a small one at each building switch. For longer outages, see generators and battery backup.
• Backup internet. A second line, such as cellular or satellite, can plug into the gateway’s backup port. Many gateways switch over automatically when the main line fails, and back when it returns (one example: Ubiquiti). Recording continues on site either way.
• Night vision. The G6 Bullet’s infrared reaches about 98 feet (specs); good lighting still helps color and faces.
• Wi-Fi kept separate. Guest or resident Wi-Fi runs on its own network, apart from the cameras.

Privacy and California law

• Audio. California generally requires the consent of everyone involved to record a confidential conversation (Penal Code 632). We leave camera audio off unless there’s a clear reason and you’ve had legal advice.
• Restrooms and locker rooms. California employers may not record audio or video in restrooms, locker rooms or changing rooms (Labor Code 435). For vaping problems in those spaces, a camera-free vape sensor is the right tool.
• Faces and plates. See the notes on face recognition above. Businesses covered by the CCPA have notice duties (California Attorney General).
• Neighbors. We aim cameras at your property, not into a neighbor’s windows or yard.

This is general information, not legal advice.

How much detail do you need?

“Can it see a face?” depends on pixels, not megapixels. What matters is how many pixels land on each foot of the scene at the distance you care about. The IEC 62676-4 standard sorts this into four levels, which Axis lays out clearly (Axis):

Rounded from Axis’s per-meter figures. A 2025 edition of the standard renames and extends these levels (Axis).

That’s why we often pair a wide camera that sees the whole area with a narrower one aimed at the door or gate, where you need a face or a plate.

Night vision: infrared or color?

Since we’re electricians, we can add or re-aim the exterior lighting at the same time, which often helps more than a pricier camera.

The right camera for each spot

Find it in seconds, not hours

Keeping the cameras themselves secure

Cameras are computers on your network. CISA’s advice for any connected device is simple: change default passwords, keep firmware updated, turn off remote features you don’t use, and keep these devices on their own network segment (CISA) (CISA). Use the checklist to see where you stand:

One more thing to know: the FCC keeps a list of equipment it considers a national security risk. Video surveillance gear from some manufacturers is on it, for uses such as public safety, government facilities and critical infrastructure (FCC). If your property falls in those areas, ask us before you buy.

Sensors, sirens and “if this, then that”

Camera systems now work with door, motion, glass-break, leak and smoke sensors and sirens on the same app (Ubiquiti). Rules tie them together: when a camera sees a person in the pool area after hours, it can turn on lights, sound a siren, move a PTZ camera, lock a door or send a webhook to another system (Ubiquiti).

1. Something happensA sensor trips or a camera sees a person in a set area.
2. A rule runsOnly during the hours and in the areas you chose.
3. The site respondsLights on, a siren, a camera turns, a door locks.
4. You decideYou get the alert and the clip on your phone.

Where it fits

• HomesDoorbell and driveway views, a virtual fence after dark, and alerts that skip the deer.
• HOAs and apartmentsEntrances, parking, mail and trash areas, with plate logging at the gates and fiber between buildings.
• BusinessesLots, loading doors and entrances, with searchable footage for incidents and insurance.
• SchoolsEntrances and outdoor areas, paired with vape sensors (not cameras) in restrooms.
• Ranches and ag sitesLong driveways and gates, with plate cameras and backup internet where service is spotty.

Choosing a camera system

There are many good camera systems. These are the questions we help you answer, whatever the brand:

• Recording: a recorder on site, the cloud, or both, and how many days of video you want to keep.
• Subscriptions: whether features need a monthly plan, and what still works without one.
• One app or several: some platforms run cameras, door access and the network together; others work best on their own.
• Resolution and lenses matched to the distance and what you need to identify, like a face at a door or a plate at a gate.
• Wired PoE vs Wi-Fi: wired is steadier and powers the camera on the same cable.
• Privacy and access: who can view video, from where, and with what login.

We’re glad to work with the system you already have, too.