How to Get Accurate AR Measurements in a Room App: Guide

You are scoping an AR feature that lets a customer measure a wall or drop a sofa into their living room, and one number decides whether people trust it: the scale. If the app says a cabinet is 80 centimetres wide and it arrives at 95, the customer does not file a bug report. They return the cabinet and stop using the app. Accurate AR measurement is the difference between a feature that reduces returns and one that quietly creates them.

This guide is written for founders and product leads scoping a room or furniture app, from someone who has shipped AR. It explains how AR scale accuracy actually works, what breaks it, how to design for it, and how to test it. The whole search results page quotes a tidy accuracy figure and never explains it. We will explain the mechanism instead.

How does an AR app measure a room?

An AR app measures a room by tracking where the device is, building a map of the space, and finding flat surfaces to place points on. When the user marks two points, the app already knows each point's position in real world units, so the distance is arithmetic. There is no laser tape unless the device has a depth sensor.

The tracking underneath is called world tracking, or motion tracking. Apple's ARWorldTrackingConfiguration tracks the device with six degrees of freedom: the three rotation axes, roll, pitch and yaw, and the three translation axes, movement in x, y and z. Google describes the same idea in its ARCore fundamentals, where the system uses "simultaneous localization and mapping, or SLAM, to understand where the phone is relative to the world around it." It does that by combining camera images with the phone's motion sensors.

To measure, the app needs surfaces. Both platforms detect planes. ARCore locates "clusters of feature points that appear to lie on common horizontal or vertical surfaces, like tables or walls," and exposes them as planes. ARKit does the same and adds each surface to the session as a plane anchor. A point the user taps is placed on one of these planes, which is why floors and tabletops measure more reliably than mid-air. If you are choosing between the two engines, our comparison of ARKit and ARCore for AR apps goes deeper on the trade-offs.

What actually breaks AR scale accuracy?

Seven things break scale accuracy: tracking drift, texture-poor surfaces, low light, fast device motion, long measurement spans, reflective surfaces, and the absence of a depth sensor. Each one degrades the position estimate the measurement depends on. None of them is exotic. They are the normal conditions of a real living room, which is why lab numbers rarely survive contact with customers.

Take them one at a time. Drift is the slow accumulation of small pose errors as the user walks around. The tracker estimates position frame by frame, and tiny mistakes add up, so a point set early can appear to slide by the time you return to it. Texture-poor surfaces starve the tracker of the feature points it needs. Google is blunt about it: "flat surfaces without texture, such as a white wall, may not be detected properly." A plain white room is the hardest room to measure.

Fast motion and low light both attack the camera feed. Apple exposes this directly through its camera tracking state, which drops to a limited quality with a stated reason. Its documented reasons include excessiveMotion, "the device is moving too fast for accurate image-based position tracking," and insufficientFeatures, "the scene visible to the camera doesn't contain enough distinguishable features." If your app ignores that state, it will happily report a confident measurement built on bad tracking. Long spans compound error over a single line, and glass and mirrors create false planes and fool depth. The table below pairs each factor with its effect and the practical fix.

What is the 98 percent accuracy number, really?

You will see one figure repeated across almost every article on this topic: 98 percent accuracy. It is worth being clear about where it comes from. IKEA quoted it when IKEA Place launched, describing how close its on-screen furniture sat to real scale, and agency blogs have parroted it ever since without explaining it.

Treat it as a vendor's marketing number under good conditions, not a law of physics and not a promise for your app. It says nothing about a dim room, a bare wall, or a user waving the phone around. We do not claim any specific accuracy percentage as our own, because the honest answer depends on the device, the space and the design choices below. A number you cannot reproduce on your own test rig is a number you should not put in your marketing.

Does LiDAR or a depth sensor improve accuracy?

Yes, clearly, on the devices that have it. A depth sensor measures distance directly rather than inferring it from camera motion, which lifts the accuracy floor in exactly the conditions that break camera-only tracking. It is not required for a good measuring app, but it turns the hardest rooms from unreliable into workable.

Google's ARCore Depth API shows why. Without special hardware it estimates depth from motion: "the algorithm takes multiple device images from different angles and compares them to estimate the distance to every pixel as a user moves their phone." That is why the app asks the user to move around, and why the docs note the best results come "when the device is half a meter to about five meters away from the real-world scene." Add a hardware depth sensor such as a time-of-flight sensor and, in Google's words, it "provides better depth on surfaces with few or no features, such as white walls, or in dynamic scenes with moving people or objects." Depth also drives realistic occlusion, so a virtual sofa can sit behind a real table.

On Apple devices the equivalent hardware is the LiDAR scanner, and Apple's RoomPlan builds on it to capture a room's dimensions and layout. Whether you should require it is a real product decision with a cost attached, which we work through in do you need LiDAR for an AR furniture app and in the practical guide to what Apple RoomPlan is and when to use it. The short version: LiDAR raises accuracy and speed, but it narrows the range of phones you reach, so design for the camera-only case first.

How do you design an AR app for accurate scale?

Design for accuracy in eight moves: calibrate before placing anything, anchor objects to detected planes, show live dimensions on screen, let the user confirm the scale, warn when tracking is poor, reuse depth data where the device has it, author 3D models in real world units, and re-measure to check. Most of this is design work, not hardware spend.

The two that teams skip most often are the two that build trust. Letting the user confirm the scale means asking them to check one known dimension, a door width or a tile, before they rely on the app. It costs one screen and turns a silent guess into a shared measurement. Warning on poor tracking means reading the platform's tracking state and telling the user to add light or slow down, rather than printing a confident number over bad data. Silence here is what produces the returns you were trying to avoid.

Two more are pipeline choices. Models must be authored to real scale, or a perfectly measured room will still show the wrong-sized sofa. That is a 3D asset discipline, covered in how to make 3D models for an AR furniture app, and it depends on exporting in formats that carry real units, which is the point of USDZ and glTF. Get the models wrong and no amount of tracking accuracy saves you.

How do you test AR measurement accuracy?

Test accuracy against known objects, repeatedly, across devices and conditions. One reading that matches a tape measure proves nothing. What you want to know is your spread: how close the app stays to the truth, and how much it varies run to run. Here is a test you can run in an afternoon. Log every number.

1. Pick known references. A standard door, a table edge, a floor tile. Measure each with a tape measure and write down the true value.
2. Measure each one in the app, five times. Reset between runs. Record all five, not just the best. The range between your highest and lowest reading is your real repeatability.
3. Change the room. Repeat in a bright room, a dim room, against a bare wall and against a busy bookshelf. Accuracy that only holds in one room is not accuracy.
4. Change the device. Test a depth-capable phone and a camera-only phone. The gap tells you what your minimum supported device really delivers.
5. Test the long line. Measure a short span and a long span across the same wall. Watch how much error the long one adds.
6. Test bad behaviour on purpose. Move the phone fast, then in the dark. Confirm the app warns instead of reporting a clean number.

Write the results into an acceptance target before launch, for example a maximum acceptable error and a maximum spread on your minimum device. That single table settles most arguments later, and it stops anyone quoting a marketing percentage the product cannot hit.

When AR measurement is not worth it

AR measurement is not worth it when the job needs certified precision, when your users work in conditions AR cannot handle, or when a simpler input would do. Consumer AR is an estimator, not a calibrated instrument. If a wrong number carries real cost, AR should assist a professional measurement, not replace it.

Be honest about the cases. A customer checking whether a bookshelf fits a nook is a good fit, because a few centimetres of error is forgivable and easy to confirm. A contractor ordering made-to-measure blinds is not, because the tolerance is tighter than consumer AR can promise and the cost of an error lands on you. Warehouses, dark rooms, empty white spaces and glass-heavy interiors will all fight the tracker. In those cases, either require a depth-capable device or keep AR as a rough guide with a clear disclaimer, and let a tape measure or a laser take the final number.

The trust point runs the other way too. Wrong scale is the fastest way to lose a shopper, and it drives the returns an AR feature is meant to reduce. The same logic sits behind virtual try-on: the value only appears when the fit is believable. We covered that economics in how a virtual try-on app boosts revenue, and it applies to furniture just as squarely.

What drives the cost of getting this right?

The cost drivers are the depth of accuracy you promise, the range of devices you support, the calibration and confirmation flows, the testing rig, and the quality of your 3D catalogue. A basic place-and-view feature is modest. A measuring tool you are willing to stand behind, tested across devices and conditions, is a real engineering line item, because the testing and the safeguards are the work. It is the kind of scope our app development team sizes up front rather than discovering later.

appico builds AR features as part of a mobile app engagement, with fixed scope and transparent pricing: our mobile app builds start from $12,000, a well-scoped MVP from $10,000, and larger multi-platform builds run up to about $150,000, with maintenance on a separate monthly plan. You own the code, the models and the accounts from day one. To sketch a number for your own scope, the cost calculator is a fast first pass, and the broader build economics sit in our note on the cost and time to develop a try-on app.

Our take

Accuracy in AR is not a sensor you buy. It is a set of decisions: calibrate, anchor, show the number, let the user confirm it, warn when the tracking is bad, and test the whole thing against real objects until you know your true spread. Do that and a camera-only phone measures a normal room well enough to sell furniture. Skip it and even a LiDAR device will produce numbers your customers learn not to believe.

If you are scoping an AR room or furniture feature and want it built by a team that treats scale as a trust problem, not a demo, you can talk to us about your AR app. Bring the rooms your customers actually live in, not the showroom.

Frequently asked questions