Cameras that tell you what matters

The camera labels a moving object as human, vehicle or other and ignores leaves, rain, insects and small animals before it sends an alert.

What the data says

Hikvision says AcuSense can remove up to 90 percent of false alarm events. Dahua lists a false-alarm rate under 1 percent for SMD Plus (under 2 percent in some documents) and under 0.5 percent for small animals with SMD 4.0. All of these are vendor test results.

Honest limit

Results depend on the scene. Set zones and sensitivity on site and review the first week of alerts.

Best for

Homes, compounds, shops, warehouses, anywhere an alert must mean something.

Draw a virtual fence or zone. A person who crosses the line or stays inside the zone triggers an alert, and only people and vehicles count.

What the data says

This is the filtered version of motion alarms, so the 94 to 98 percent false-alarm history of ordinary alarms is exactly the problem it targets.

Honest limit

A camera sees a flat picture, so the line must follow the real ground. We mark it with a person walking the route.

Best for

Boundary walls, rooftops, factory yards, schools and mosques after hours.

When a person is detected, the camera flashes a strobe and plays a recorded warning, so a visitor learns they are being watched before anything is taken.

What the data says

Hikvision’s audio-and-strobe AcuSense cameras have a built-in speaker up to about 60 dB covering roughly 15 metres, an adjustable strobe pattern and two-way audio from the mobile app, in 4 MP to 8 MP models.

Honest limit

The warning must be set to the right hours and message so it does not disturb neighbours or staff.

Best for

Shops after closing, warehouses, construction sites, building entrances.

Compare faces against a list for attendance, door access, VIP recognition or a watch list. Usually runs on a recorder or an AI server.

What the data says

NIST publishes continuous tests of face algorithms at a fixed false-match rate of one in a million, and its researchers warn that a single “accuracy percentage” is misleading. The Stanford AI Index reported error rates on visa and mugshot style photos falling from near 50 percent to a fraction of a percent by 2020.

Honest limit

Lab photos are not corridors. Backlight, angle, masks and distance lower accuracy, and consent and notices are your responsibility.

Best for

Offices, banks, schools, factories with controlled entrances.

A dedicated camera reads plates at a gate and can open a barrier for approved vehicles or log every visitor.

What the data says

One accuracy analysis puts modern ANPR above 99 percent in controlled lanes and at 90 to 98 percent in real-world conditions. In a vendor test with the camera about 6 feet from the car, plates were read correctly up to a 70 degree angle and failed at 75 degrees.

Honest limit

Headlights, speed, dirty plates and steep angles decide the result. Use an infrared ANPR camera, not a general-purpose one.

Best for

Factory and campus gates, parking, residential towers.

Count entries and exits, measure queue length and see on a floor plan where people spend time. Staff rosters and shelf layouts can follow the data.

What the data says

This is a business tool more than a security one: the camera produces numbers by hour and zone that you can export, instead of footage you have to watch.

Honest limit

Counting works best from an overhead or steep-angle camera over a single doorway. Wide scenes lose accuracy.

Best for

Retail stores, bank halls, showrooms, malls.

Ask the recorder for “people at the back gate between 2 and 4 am” and jump straight to those clips instead of scrubbing hours of video.

What the data says

Vendors and their customers describe investigations dropping from hours to minutes. These are customer stories rather than controlled studies, so we plan for real savings, not the headline claim.

Honest limit

Search is only as good as the metadata, so cameras must be calibrated and the clock must be correct.

Best for

Every site that has ever had to find “what happened”.

AI-assisted compression such as H.265+ lowers the bitrate when the scene is quiet, so the same disks hold more days of video.

What the data says

Hikvision’s own 24-hour tests showed H.265+ files 79 to 82 percent smaller than H.264. Independent IPVM tests found smart codecs from several makers cut bandwidth 50 to 90 percent, most in still scenes, while early plain H.265 cameras saved only 20 to 30 percent.

Honest limit

Busy scenes save much less. We size disks on the cautious end of the range, not the brochure.

Best for

Any system with long retention or many cameras.