Cameras that can still identify someone at 40 feet

Nearly every camera system we get called to look at is recording perfectly well. That is not the problem. The problem shows up four months later, when a package theft, a lobby assault or a slip-and-fall claim means somebody has to sit down and export a clip, and the person in the frame is a gray shape thirty feet away with no face, no clothing color anyone would swear to in a deposition, and a plate that reads as four smeared rectangles.

That outcome was decided at design time by one number, and it is not megapixels and it is not camera count. It is pixel density on the target: how many pixels land across each meter of scene width at the distance the subject will actually be.

The standard that governs this is IEC/EN 62676-4. The 2015 edition gave the trade the DORI bands (Detect, Observe, Recognize, Identify) and most installers still quote them. The 2025 edition, in force since 9 October 2025, replaced DORI with a two-group scheme and moved the bar for reliable identification of a person from 250 px/m to 500 px/m, on the finding that 250 fell apart under low light, motion blur and compression. Which is to say, it fell apart under precisely the conditions that produce the incidents you wanted footage of.

If a quote in front of you does not state a distance and a pixel figure for at least the cameras covering your entrances and cash points, it is a parts list, not a design.

The scene width a lens covers at a given distance is W = (D × sensor width) / focal length, and the pixel density you land on is simply horizontal resolution / W. Solve the first for focal length and you get the lens you should have specified: f = (D × sensor width) / target width.

Here is what that means on a real wall. A 4 MP camera puts 2,560 pixels across the frame. Point it down a 12 m (40 ft) lobby on the 2.8 mm lens it shipped with, so it covers the full width, and you get 2,560 ÷ 12 = 213 px/m. That is short of even the old identification bar and less than half the current one. Everyone in that lobby is a silhouette with a coat on.

Put the same camera on a 12 mm lens so it covers a 3 m doorway instead and you get 2,560 ÷ 3 = 853 px/m. Now you have a face. The camera did not change. What changed is that we stopped asking one camera to cover a whole room and gave it one job.

That is the whole argument for a system built as a wide camera for context plus a tight camera on each chokepoint (door, register, stair, elevator lobby, dock) rather than four wide cameras in four corners. It also explains why 4K is not automatically the answer. Doubling resolution on the same lens does double pixel density, but it does it by putting smaller photosites on the same piece of silicon, and smaller photosites collect less light. A 4 MP camera on a 1/1.8-inch sensor will beat an 8 MP camera on a 1/2.8-inch sensor in a dark corridor every time. Pick the resolution that hits your pixel target at your distance, then buy the largest sensor that still delivers it.

Backlight is why your entrance camera produces silhouettes

A lobby camera looking out toward a glass storefront in daylight has a bright exterior behind the subject and very little light on their face. Wide dynamic range is the fix, and there is a real line between the two kinds. True WDR captures multiple exposures and fuses them; the accepted threshold to call a camera true WDR is 120 dB. Digital WDR is one exposure with tone mapping applied afterward, typically claiming 90 to 110 dB, and it cannot recover detail the sensor never captured. On anything aimed at a doorway we specify true WDR and put the dB figure on the quote.

The exception is license plate work. Multi-exposure fusion smears a moving plate, so on plate-capture cameras we turn WDR off and control exposure with shutter speed instead (typically 1/500 s as a floor and faster on a roadway), keep the total camera-to-vehicle angle under about 30°, and aim for 100 to 150 pixels across the plate itself rather than across the scene. Note also that plate capture (a readable image a human can use) and plate recognition (an automated text string in a database) are different products at different prices, and vendors are not always careful about which one they are selling you.

Housing, form factor and the ratings that get conflated

Turrets are the default for most fixed positions, because with no bubble there is no infrared bouncing off the inside of a dome, the single most common reason a camera looks fine by day and milky by night. Domes stay where tamper resistance and a low ceiling profile matter. Bullets aim easily and throw infrared far, but the housing is a grab handle and a perch for spiders whose webs sit directly in the IR beam. Multi-sensor units earn their price on building corners and corridor intersections, where one cable and one mount give uniform pixel density across the whole width instead of three cameras arguing about overlap.

Outdoors, two ratings get muddled constantly. IP is ingress: IP66 is dust-tight plus powerful water jets; IP67 adds thirty minutes under a meter of water, which is what an areaway, a sidewalk vault or anything low enough to be hosed actually needs. IK is impact in joules, a separate scale entirely: IK08 is 5 J, IK10 is 20 J and is the top of the standard, so anything sold as IK11 is marketing. An exposed street-level camera in New York wants IP67 and IK10, plus security-pin or Torx hardware.