Placing UniFi Protect Cameras: Mounting Height, Field of View, and What Night IR Really Gives You

Do you know how many pixels your driveway camera actually puts on a face at forty feet? If you have not measured it, you are guessing, and in almost every install the guess runs optimistic.

That gap surfaces after an incident, when someone pulls the clip and finds a person-shaped smear crossing a gravel lot at 2 a.m. The camera worked exactly as specified.

A G5 Pro on a well-chosen mount produces video you can hand to an investigator. The same camera bolted under a soffit at fourteen feet, aimed across a parking area, produces a motion alert and a coat color, and it does that for the entire life of the install.

The variables that separate those two outcomes are mounting height, downtilt, horizontal field of view, and the light available in the middle of the night. None of them show up in a model comparison, and all of them get locked in during the twenty minutes you spend on the ladder.

What follows is the placement math, the honest version of what infrared gives you, and the short list of scenes where no Protect camera will ever produce usable evidence. If you have not settled on models yet, work through how to choose a UniFi camera and the UniFi camera gear guide first, then come back and place them.

Start from the standard the industry already uses, because it converts an argument about "good enough" into arithmetic. IEC 62676-4 defines the DORI scale: roughly 25 pixels per meter to detect that something is there, 125 to recognize a person you already know, and 250 to identify a stranger.

Converted to imperial, that is about 8 pixels per foot for detection, 38 for recognition, and 76 for identification. Note that identification is the only tier that survives cross-examination, and it costs ten times the pixel density of detection.

Pixel density falls off with scene width, not with the camera's megapixel count alone. A 4K Protect model spreads 3840 horizontal pixels across whatever the lens sees, and a 2K/4MP model spreads 2688 pixels across that same view.

Height enters the math twice. It sets the distance to your subject, and it sets the angle at which you see them.

The angle is the part people underestimate. Mount at fourteen feet, aim down at a walkway ten feet away, and you record the top of a hat and a pair of shoulders at excellent pixel density, with no face in it.

As a rule, keep identification cameras between 8 and 9 feet with a shallow downtilt in the range of 10 to 20 degrees. Overview cameras can go to 12 feet or higher, because their job is counting bodies and tracking direction rather than recognizing anyone.

Of course, low mounts invite tampering, and that trade-off is real. The usual resolution is a vandal-resistant turret or dome at 9 to 10 feet rather than a bullet at 7, since a bullet on an arm is the easiest camera in the catalog to grab and rotate toward the sky.

Keep in mind that identification is a zone rather than a property of the camera. Every fixed camera has a band of distance where it clears 76 pixels per foot, and outside that band it is doing a different job.

Here is how to assign that band deliberately:

• Pick the pinch point first. Doors, gates, stair landings, and hallway ends force a subject through a narrow, predictable path at a known distance, which is where identification pixels are affordable.
• Aim across the path, not down it. A camera looking at the side of an approach sees a face for several strides, while a camera looking at the back of the approach sees the back of a head.
• Let the wide view be wide. The overview camera is allowed to be detection-grade, as long as you have not quietly assigned it the identification job as well.
• Write the distance down. Record the target range for each camera at commissioning so the next person does not re-litigate the aim from scratch.

Overall, height is the cheapest thing to get right and the most expensive thing to fix. Correcting it after the fact means a new hole, a new gland, and frequently a new cable run.

Horizontal field of view determines how wide the scene is at any given distance, and the relationship is plain trigonometry. Scene width equals twice the distance multiplied by the tangent of half the horizontal FOV.

Work an example with a roughly 100-degree lens, which is typical of the wider fixed Protect models. At 30 feet that camera covers about 71 feet of width, so a 4K sensor lands near 54 pixels per foot.

That sits comfortably above recognition and well under identification. In practice it means you will know it was a man in a grey hoodie, and you will not know which man.

Narrow the same 4K camera to roughly a 45-degree field of view, which the optical-zoom models such as the G4 Pro reach at the long end of their range, and the picture changes entirely. At 30 feet the scene is about 25 feet wide, which puts you above 150 pixels per foot and deep into identification territory.

This is why lens choice beats sensor upgrades on most sites. Doubling resolution doubles pixel density, while halving the scene width does the same thing and simultaneously improves your night exposure, because the camera is now illuminating a smaller area.

The same math explains why a 360-degree fisheye such as the AI 360 is a situational-awareness device. Its pixels are distributed across the full circle, so any single direction receives a small fraction of the total, and the useful identification radius sits close to the mount.

For instance, a fisheye over a retail floor tells you how many people were in the store, which aisles they walked, and when they left. It will not tell you whose face that was at the far wall, and no firmware update changes that.

Coverage overlap follows from the same numbers. Each camera's edge should reach into the neighboring camera's usable band, so that a subject moving through the property never crosses an unrecorded gap.

What's more, overlap gives you tamper coverage at no extra cost. When one camera can see the housing of another, an attempt to spray, cover, or rotate a lens is recorded by the camera that is still working.

Two practical constraints ride along with every coverage decision. More cameras and narrower lenses mean more streams to retain and more ports to power, so check the numbers in Protect storage sizing and PoE budget planning before you commit to a camera count.

Infrared is the feature most often oversold on the spec sheet and most often disappointing in the timeline. Understanding four of its properties will save you a redesign.

First, IR is monochrome by definition. The illuminator emits near-infrared and the sensor drops its IR-cut filter to see it, which means every color descriptor you would put in an incident report disappears.

That matters more than the loss of fine detail. A grey sedan and a red sedan are the same sedan at night, and a blue jacket becomes a mid-grey jacket that matches half the wardrobes in the county.

Second, IR obeys the inverse-square law. Double the distance to your subject and the illumination reaching them drops to roughly a quarter, which is why a published IR range describes the distance at which the camera still sees something rather than the distance at which it still identifies anyone.

Your practical identification range at night is always shorter than your daylight range on the same camera. Plan around the night number and treat the daylight number as headroom.

Third, low light forces the shutter open longer. To gather enough photons the camera slows its exposure, and a person walking at normal pace smears across the frame exactly when you most need their face.

This is the failure that surprises people, because the static scene looks fine. A parked car, an empty walkway, and a closed door all render sharply under IR, and the one moving object in the frame is the one that blurs.

Fourth, IR washout is the most preventable problem on this list. When the illuminator hits a nearby wall, soffit, railing, gutter, or a spider web strung across the lens hood, that surface returns a bright reflection, auto-exposure compensates downward, and the rest of the scene goes grey.

Accordingly, get the lens out past the surfaces around it. Use a standoff or arm mount rather than recessing a camera into an eave, keep the first several feet of the view clear of hard surfaces, and check the housing seasonally, since spiders are drawn to the insects that IR attracts.

Weather amplifies all of it. Rain, snow, and fog reflect IR straight back into the sensor the way high beams reflect back in fog, and a scene that is clean in dry conditions can white out during the storm you actually needed to record.

The reliable fix is white light. A porch fixture, a soffit light on a photocell, or a PoE floodlight keeps the camera in color mode, keeps the shutter fast enough to freeze a walking subject, and preserves the clothing and vehicle colors that make a description usable.

Remember that the goal at night is a small area lit well enough that the camera never leaves color mode in the one place you care about. Pushing IR further into the dark buys far less than that.

Some placements are unrecoverable regardless of model, and recognizing them early is worth more than any settings change. All of them come down to a light source inside the frame that is far brighter than the subject.

The classic case is a camera aimed at a west-facing glass door. Wide dynamic range holds the scene together for much of the day, and for the forty minutes around sunset every person entering becomes a silhouette against a blown-out background.

Do the sun-path check before you drill. Stand at the proposed mount, look along the intended aim, and ask whether the sun ends up in that frame in the morning, in the evening, and in December when it sits low all day.

The fix is orientation rather than configuration. Aim the camera so the sun is behind it, cross-mount two cameras facing each other along an approach, or move the identification camera indoors just past the door, where the exterior light becomes the background instead of the subject.

Headlights create the same problem on a compressed schedule. A camera aimed straight down a driveway meets high beams head-on, auto-exposure protects the highlights, and the vehicle behind them collapses into a shape.

License plates deserve their own camera for exactly this reason. Keep the camera within roughly 30 degrees of the plate both horizontally and vertically, offset the mount so you are not staring into the headlights, and let a separate wide camera handle scene context.

Note that retroreflective plates behave differently from everything else in the frame. Under IR they return far more light than the car around them, which is why a plate-focused camera is tuned for the plate and produces a dark, unusable image of the driver.

There are also scenes that no amount of placement discipline saves. Here are the ones worth refusing outright:

• Wide lots on a single camera. A 4K wide lens across 80 feet of asphalt is detection-grade everywhere and identification-grade nowhere.
• Cameras shooting through glass. Interior mounts aimed out a window bounce their own IR straight back and pick up every reflection in the room after dark.
• High mounts over close subjects. Anything above 12 feet looking almost straight down produces crown-of-the-head footage at excellent resolution.
• Unlit approaches longer than the IR range. If the subject is beyond effective illumination while they choose their route, the camera only records the arrival.

All of these fail the same test. Ask what an investigator could state from the footage, and if the answer stops at "someone was there," the placement needs to move rather than the settings.

Sign-off should require more than a level camera and a live image on a laptop. A placement is proven when someone has walked the scene under the conditions that matter and reviewed the result the way it will actually be reviewed.

Run the walk test after dark rather than at install time in the afternoon. Have a colleague walk each approach at normal pace, in ordinary dark clothing, and stop at the distance you designated as the identification band.

Then review the recording instead of the live view. Open the clip in the Protect timeline, export it at full resolution, and look at it on a monitor, because a phone screen flatters footage that falls apart when it is enlarged.

Motion and smart-detection zones deserve the same scrutiny. A person crossing at the far edge of a wide scene may never occupy enough of the frame to trigger person detection, so confirm the detections fire where you need them and exclude the public road, the tree line, and the neighbor's driveway.

Check the supporting infrastructure while you are in the console. Confirm that IR draw at night has not pushed a switch past its budget, that retention still matches the recording profiles you set, and that the cameras sit on their own segment per UniFi VLANs explained rather than sharing a network with everything else.

Finally, confirm that you can reach the footage from somewhere other than the site. Set up and verify UniFi remote access in advance, because the evening of an incident is a poor time to discover that the only path to the recordings is a laptop on the local LAN.

Placement carries no line item on the quote and determines nearly everything about the outcome. Get the height, the angle, the lens, and the light right, and the hardware you already own will do work that a larger sensor on a bad mount never will.

If you are still selecting hardware for a site you have already walked, the UniFi camera gear guide covers sensor and lens options by scene type. Decide the mount first, then buy the camera that fits it.