Drone, localizing from its own camera

RTLS From the Cameras Already on Site

Locate people, vehicles, robots and drones against one map built from the scans you already hold, with a fresh fix every few seconds. Every camera becomes the tag, and nothing goes on the ceiling.

<5cm

median position on a multi-frame query, with heading

~2s

per cloud fix, and about 52 ms when it runs on the device

0

anchors, tags or cable runs installed on site

1map

shared by phones, vehicles, robots and drones

Built from your scansCloud or on-deviceSame map as VPS

MultiSet docs and SDKs, November 2026.

01

What is camera-based RTLS?

Camera-based RTLS works out where a device is from what its camera sees, matched against a 3D map of the site. A real-time location system answers one question for a site: where is everything right now. Tag-and-anchor systems answer it with radio timing between a tag on the asset and anchors on the ceiling, and most of their budget goes into installing that grid.

MultiSet answers it with vision. A device sends a frame, and the visual positioning system returns its position and heading in the site's coordinates. For the cost comparison against UWB and BLE, read vision-based RTLS, the third architecture. For every way to position indoors, see indoor positioning technologies.

02

One fix, two jobs

Locate puts every camera-carrying person, vehicle, robot or drone on the site's own scan, a few seconds behind the moment, for a management view and traffic analysis. Guide uses the same fix to take a worker to the asset and show the work order on it. One map pays for both jobs.

Four devices on one terminal map
Rail terminal concourse with the localization map's mesh traced in purple over it PLATFORM 2 18 M PHONE SMART GLASSES ROBOT DRONE
PhoneLocated
Smart glassesLocated
Ground robotLocated
DroneLocated
Located4 devices, one map

Every position and heading arrives in the same coordinate frame.

Locate puts every device on one view, with its recent path, refreshed every few seconds. Guide uses the same fix to route the phone to Platform 2 and open what belongs there.

Illustration over a partner's map, shown without its name. Positions are examples.

Located with
Any RGB camera at 640 x 480 or above, on phones, vehicles, body-worn units, smart glasses, robots and drones
Returns
A 6-DoF pose with a confidence value, in the map's coordinates or georeferenced to yours
Accuracy
Sub-5 cm median on a multi-frame query, measured on your site in the pilot
Fix time
1.5 to 2.5 s from the cloud, about 52 ms when localization runs on the device
Between fixes
Phones, headsets and robots carry the pose with their own motion tracking
Built from
E57 with panoramas, 360 video or a metric Gaussian splat, usually scans the site already holds
Site changes
Localized photos written back at their pose, changed areas added as new map versions
Scale
Large sites mapped in sections that join into one MapSet
Platforms
Unity, native iOS and Android, WebXR, ROS 2 and the REST API
Deploy
MultiSet Cloud, Dedicated Cloud, Private Cloud / VPC or on-device

Localize a camera against your own site this week.

Upload a scan you already hold on the Free plan, then query the map from a phone or any camera through the REST API.

03

Where it runs today

Drones inside buildings, technicians in process plants and robots on the floor beside them all localize this way now.

A drone's live path traced over a scanned town as it localizes from its own camera
Defense and public safety

Drones tracked inside buildings

Mithril Defense tracks drones that fly into buildings, where GPS drops out at the door. Each drone localizes from its own camera, so nothing is installed inside.

A process plant map with a pump and a valve labelled where they stand
Process plants

The right valve in a rack of identical ones

One pharmaceutical plant cut its valve search time 2.5x, across more than 7,000 tagged valves, with phones routed by the map.

Read the case study
A brewery fermentation hall with stainless tanks and pipework over a red floor
Food and beverage

Assets found on a brewery floor

Phones and smart glasses route staff to vessels, skids and pumps, with readings shown on the equipment.

Read the case study
A quadruped robot following its route to a waypoint, seen through smart glasses
Robotics

Ground robots on the same map as people

Quadruped robots localize against the map the people beside them use, through our robot adapter.

Bring us the site where positioning has failed before.

Mines, plants, terminals and tunnels. We scope a pilot on your scans and the cameras you already run.

04

The map keeps up with the site

Sites change every week, and in a mine every shift. Images that devices localize with are written back into the map at their pose, and a changed area becomes a new map version aligned to the original, so content and routes keep their coordinates. Large sites are mapped in sections that join into one MapSet. See map versioning.

One floor, scanned four times
Bare concrete floor plate on 28 May Partition walls going in on 18 June Glazed partitions and ceiling on 20 August Finished office with desks on 10 September

The same viewpoint in four versions of one map, from bare concrete to a finished office. Content placed in May keeps its coordinates in every later version, so nobody re-places it after a re-scan.

Three scans, one frame
FLOOR 1 FLOOR 2 YARD ONE ORIGIN FOR ALL THREE

Separate scans each start from their own origin. MapSet joins them into one coordinate frame, so a device walks from one floor to the next, or out into the yard, without starting over.

05

Where camera-based RTLS is the wrong tool

Assets with no camera. A pallet or a hand tool needs a tag. Most sites run both, with vision covering people, vehicles and robots.

Dark zones. With no light, a camera has nothing to match. Cover those areas with radio or leave them out.

Areas where filming is banned. Security zones that prohibit cameras stay with tags or with presence readers at the door.

06

Tag-and-anchor RTLS against camera-based RTLS

Both have a place on most sites. The question is which one carries each job.

Camera-based RTLS Tag-and-anchor RTLS
What is installed Nothing on site. A map built from existing scans Anchors on a grid, cabling, and a tag per asset
What it locates Anything carrying a camera Anything carrying a tag
What it returns Position and heading, with a confidence value Position, sometimes zone only
When the layout changes Re-scan the changed area and add it as a new map version Re-survey, and sometimes re-cable
Coordinate frame Your site's coordinates, shared by every device The vendor's own grid
Update rate A fix every few seconds, with device motion tracking in between Continuous, often sub-second

Track the cameras you already run, on the scans you already hold.

Start free on one map, or talk to us about a site-wide pilot.