Camera Placement for a New York Storefront

Almost every storefront camera system we are asked to look at has the same fault, and it is never the camera count. The cameras are pointed at areas rather than at the two or three places where a person’s face is guaranteed to pass within a known distance of a lens.

The result is footage that shows an incident happened and cannot show who did it: the worst possible outcome, because the store paid for cameras, believed it was covered, and finds out otherwise on the day it matters.

The only spec that matters: pixels across the target

Megapixels do not tell you what a camera can see. Pixel density does: how many pixels the sensor spends on each foot of scene at the distance you care about. It is standardized in IEC/EN 62676-4, whose 2015 edition defined the DORI scale most of the industry still quotes:

Task	Pixel density	Approx. pixels per foot	What you actually get
Detection	25 px/m	~8 PPF	A person is there. Not who, not what they are wearing.
Observation	62.5 px/m	~19 PPF	Clothing color, general build, direction of travel.
Recognition	125 px/m	~38 PPF	You can tell it is someone you already know.
Identification	250 px/m	~76 PPF	A stranger’s face, in good light, standing still.

The current edition, in force since October 2025, replaces DORI with a two-group scheme and raises the bar: reliable identification of a person moved from 250 px/m to 500 px/m, about 152 pixels per foot, because 250 did not survive low light, motion blur and compression. Almost nobody selling storefront systems has updated their design rules for it.

So: if you want a usable face of a stranger, design for 76 PPF as a floor and 150 PPF where you can get it. Under about 40 PPF you can only recognize someone you could already name.

Doing the math on the camera you already own

Pixel density is a division problem. Take the camera’s horizontal resolution and divide it by the width of the scene at the distance you care about:

PPF = horizontal pixels ÷ scene width in feet

A 4 MP camera is 2560 pixels wide. So:

• Covering a 34-foot scene width: 2560 ÷ 34 = 75 PPF. Right at the old identification threshold, in daylight, if the subject faces the lens.
• Covering a 17-foot scene width: 2560 ÷ 17 = 150 PPF. Comfortably identification-grade, and it still meets the new standard.
• Covering the 60-foot width of a typical storefront and the sidewalk in front of it: 2560 ÷ 60 = 43 PPF. That is recognition, not identification. It will not give a detective a face.

An 8 MP camera is 3840 pixels wide, so it buys 50% more scene width at the same density, but if it achieves that on a smaller sensor with smaller pixels, it loses to the 4 MP camera the moment the sun goes down. Choose resolution to hit the density you need at the distance you need, then take the largest sensor that still delivers it.

Focal length and distance

Scene width, distance and lens are locked together: W = (D × sensor width) ÷ focal length, and rearranged, focal length = (D × sensor width) ÷ W. In practice this means the two levers are the same lever: doubling focal length halves the scene width and doubles pixel density.

On a real storefront: a 2.8 mm lens is a very wide angle, roughly 100 to 110° horizontal on a common sensor. It ships in almost every box, it makes a small store look fully covered on the monitor, and at 25 feet it spreads 2560 pixels across roughly 60 feet of scene. A 6 mm lens on the same camera covers about half that width and doubles your pixel density. A 12 mm lens aimed at a doorway gives you a face. One caution: nominal sensor designations (1/3″, 1/2.8″, 1/1.8″) are not real dimensions and vary by manufacturer, so use a varifocal camera and set it on site against a person standing where the subject will be.

The camera over the door pointed at the street gives you nothing

This is the most common position on New York storefronts, and it fails for four reasons at once. It is too wide: capturing the sidewalk means covering 50 to 70 feet of scene, which puts you in the 35 to 50 PPF band. It is too high and looking down: mounted above the header at ten or eleven feet and tilted at the sidewalk, it photographs the tops of hats, and analytics want a near-horizontal view of a subject just as much as people do. It is backlit for most of the day, metering a sunlit sidewalk while the person in the doorway falls into shadow. And it is watching the wrong moment: people on the sidewalk have not done anything, while the threshold gives you a guaranteed distance, a guaranteed direction of travel and a doorway that funnels everyone through the same three feet.

Keep a wide street-facing camera for context. Just do not let it be the camera your identification depends on.

Where the pixels belong: the choke point

Every storefront has at least one place where everybody who enters passes through a known, narrow gap, facing the same direction, at a distance you control. That is where the identification camera goes.

The best version is a vestibule: a camera inside the second door at roughly eye level to slightly above (six and a half to eight feet), looking back toward the outer door. Everyone entering walks straight into it, and a door opening is a known width, typically 36 inches for a single leaf, which gives you a built-in scale reference. At an 8-foot working distance and a 6-foot scene width, a 4 MP camera is at 426 PPF. You will get eye color.

With no vestibule, the equivalent is a camera behind and above the customer line aimed at the inside face of the entry door, positioned so the subject is 8 to 12 feet away as they turn in. Mount at 8 to 9 feet: high enough not to be casually reachable, low enough that tilt stays under about 30° and you are looking at a face rather than a scalp.

The register and the back of house

Overhead-only coverage of a register is the classic mistake. A camera directly above the point of sale shows the drawer, the hands and the top of two heads: excellent for a till dispute, worthless for identifying the person on the customer side. The professional layout is two cameras working together:

• An overhead on the drawer and counter surface, tight enough to read what changes hands. Pair it with point-of-sale text overlay if your system supports it, so a search on a void or a no-sale jumps straight to the video.
• A camera behind the register at counter height plus two feet, facing the customer. A four-to-six-foot working distance puts even a 2 MP camera far past identification density. This is the camera that produces the picture the police can use.

Behind the counter, the positions that repay themselves are the back door, the stockroom door and the safe or cash office, where internal loss actually happens. The back door camera should see the door from the inside and cover the floor space in front of it, where goods get staged, not just the doorway.

Night, glass, and the two things that never work

Daytime backlight. A camera indoors looking toward a glass storefront sees a silhouette. The fix is true wide dynamic range: multi-exposure WDR that captures separate frames biased for highlights and shadows and fuses them. The accepted threshold to call it true WDR is 120 dB; digital WDR is a single exposure with tone mapping applied afterward, typically claims 90 to 110 dB, and cannot recover information the sensor never captured. Specify true WDR at 120 dB or better on every camera aimed at a doorway or window. It is roughly a $100 difference on the camera and it is the difference between a face and a black outline.

Infrared through glass never works. Point an IR camera through your own glazing at night and the illuminator reflects straight back into the lens, returning a white sheet. No setting fixes it. The answers are to put the camera outside the glass, or disable IR and light the scene with white light. A camera inside looking out at night is also fighting its own room lights reflected in the window.

Color is the single most useful suspect descriptor there is, so where a storefront already has decent sidewalk lighting, a true low-light color camera beats an IR camera every time. And cap the maximum shutter time and gain: in low light the camera stretches its exposure to gather light and people smear into ghosts. A dark sharp face is evidence; a bright smear is not.

Sidewalk sheds, scaffolding and the year your sightline disappears

This is the New York constraint no national camera guide mentions, and it has ruined more storefront systems than any equipment failure.

A sidewalk shed puts a solid deck roughly eight feet above the sidewalk across your whole frontage. Every camera above that deck (which is where most exterior cameras go, precisely to keep them out of reach) is now looking at plywood. Everything under it sits in permanent deep shade by day and under the shed’s own strip lighting at night, which is a completely different exposure problem than the one it was set up for. And sheds do not go up for a weekend: facade work in this city routinely keeps one in place for a year or more.

What we do about it when a shed is up or expected:

• Mount exterior identification cameras below the shed line, tight to the building face, on the door surround rather than out at the awning edge.
• Specify vandal-rated housings and tamper-resistant hardware. A shed and a pipe scaffold are a ladder; a camera that was fifteen feet up is now four feet from a work platform. IK10 is the top of the impact scale at 20 joules, paired with IP66 or IP67 outdoors.
• Re-aim and re-expose after the shed goes up, and again after it comes down. It is a half-hour visit almost no store books, and it is why a system that worked in March is useless in September.
• Do not run exposed cable along the shed structure. It leaves with the shed.

Roll-down gates

A roll gate creates three specific problems. The gate housing occupies the header above the doorway, which is exactly the mounting position an installer instinctively reaches for. Drilling into or beside a gate box risks the mechanism, and a camera fixed to the housing vibrates every time the gate runs. A closed solid gate blinds every interior camera that was looking outward, so if your only night coverage looks out the front, you have no night coverage. And open-mesh gates, the see-through type the city has pushed storefronts toward for years, are better than solid but not transparent to a camera: the grid interacts with the sensor to produce moiré patterning, and at night IR bounces off every slat and blooms.

The layout that works: one camera inside covering the storefront interior and the gate line, one outside beside or above the gate covering the approach, IR disabled on anything shooting through mesh, and acceptance testing done at night with the gate down, not at 2 p.m. with it open.

Where a camera legally cannot go

New York is specific about this, and the penalties attach per device.

General Business Law §395-b makes it an offense for an owner or manager to knowingly permit a camera to surreptitiously observe the interior of a fitting room, restroom, bathroom, washroom, shower, or a guest room in a hotel, motel or inn. Each device is a separate violation. One narrow exception applies to fitting rooms only, where conspicuous written notice is posted at the entrance. In a city of a million or more, meaning New York City, that notice must be in both English and Spanish.

Labor Law §203-c is separate and covers employees: no employer may cause a video recording of an employee in a restroom, locker room, or a room designated for changing clothes, absent a court order. Recordings made in violation cannot be used for any purpose, and the statute carries a private right of action with damages and fees.

The rule we work to: no camera sees into a space where a person would reasonably undress, and none sees into an employee restroom or changing area, including through a doorway, a mirror or a reflective surface. Every field of view gets checked against those doors, and if a view clips one, the camera moves or gets a privacy mask documented in the handover pack.

One more, if you are considering face-matching analytics rather than plain video: New York City requires retail stores, food and drink establishments and places of entertainment that collect biometric identifier information to post a clear and conspicuous sign at every customer entrance, and separately bans profiting from that information. There is a private right of action. Talk to counsel first. Plain video without identification software is a different thing.

The audio trap

This one catches good operators, because the usual summary is technically true and practically misleading.

New York is a one-party consent state. Under Penal Law §250.05, eavesdropping is a class E felony, and the definitions turn on consent: recording a conversation is lawful if at least one party to it consents. If you are in the conversation, you can record it.

Here is the trap. The definition of mechanical overhearing covers recording a conversation without the consent of at least one party, by a person not present. An unattended camera microphone in your ceiling, recording two customers talking to each other, has no consenting party and nobody present. That is a felony, not a civil exposure, and the microphone ships enabled in a large share of the cameras sold today.

Our default, and our recommendation to every retail client: audio disabled at the camera, verified at commissioning, documented. If there is a genuine reason to capture audio at one position, that is a conversation to have with an attorney first, with posted notice, and scoped to that position rather than the whole store.

A storefront layout that actually works

Position	Job	Spec that matters
Vestibule or inside the entry door, 6.5 to 8 ft, facing incoming traffic	Identify every person who enters	Scene width under 8 ft; true WDR 120 dB; tilt under 30°
Behind the register, counter height +2 ft, facing the customer	Identify the person at the point of sale	4 to 6 ft working distance; no backlight from a window behind the customer
Overhead on the drawer and counter	See the transaction and what changes hands	Tight field of view; POS text overlay if available
Interior wide, high corner	Context, movement, how many people, which direction	Wide lens is fine here; this is the one camera that may be 2.8 mm
Back door, interior side	Deliveries, staff exits, staged goods	Covers the door and the floor space in front of it; IP-rated if it sees weather
Exterior over the entrance, below the shed line	Approach, vehicles, time reference	IP66/IK10; white light rather than IR if there is glass in frame

Six positions. Most storefronts we survey have more cameras than that and none of these angles. The fastest second opinion on an existing system costs nothing: stand where a customer stands, have someone pull up the live view, and see whether they can read your face. If they cannot, no amount of resolution on the quote will fix it.