A visual positioning system (VPS) works out exactly where a camera is by comparing what it sees against a prebuilt 3D map of the space. It returns 6-DoF pose: three numbers for position, three for orientation, in the map's coordinate frame. Not "near aisle 14", but the precise point the camera occupies and the direction it faces.
MultiSet captures any environment with any scanner, normalizes it through Vision Fusion into a unified compression-optimized map, then returns a centimeter-true pose in seconds when devices query the map.
| Visual positioning (VPS) | GPS / GNSS | Visual SLAM | LiDAR SLAM | UWB | BLE / Wi-Fi | QR / markers | |
|---|---|---|---|---|---|---|---|
| Typical accuracy | Sub-5 cm | 3 to 10 m, worse indoors | Drifts without a map | 2 to 5 cm | 10 to 30 cm | 1 to 5 m | Exact at the marker |
| Returns orientation | Yes, full 6-DoF | Heading only | Yes | Yes | No, position only | No | Yes, at the marker |
| Works indoors | Yes | No | Yes | Yes | Yes | Yes | Yes |
| Absolute or relative | Absolute, against a shared map | Absolute | Relative to where it started | Relative to where it started | Absolute | Absolute | Absolute at each marker |
| Hardware to install | None, uses the existing camera | None | None | LiDAR sensor per device | Anchors plus tags | Beacons, battery cycle | Printed targets |
| Survives app restart | Yes | Yes | No, session resets | No, session resets | Yes | Yes | Yes |
| Multi-user shared frame | Yes | Yes | No | No | Yes | Yes | Yes |
| Main cost driver | Capture the space once | None | None | Sensor cost | Installation and survey | Install and battery maintenance | Placing and maintaining targets |
VPS becomes useful the moment a worker, robot drone or asset needs to know exactly where it is. Three patterns drive most enterprise rollouts:
Satellite signals do not survive a roof and a steel frame, so everything called indoor positioning is a way of replacing that missing signal. There are four families, and the choice between them is mostly a question of what you are willing to install.
A camera against a map. The device compares what it sees to a prebuilt 3D reconstruction of the building and gets back a full position and orientation. Nothing to install, but the space has to be captured once.
Radio time-of-flight. Ultra-wideband anchors and tags exchange precisely timed pulses. Ten to thirty centimetres is realistic, and it is an infrastructure project: anchors need power, mounting and surveying, and every tracked thing needs a tag.
Radio signal strength. Wi-Fi and Bluetooth beacons trilaterate from measured signal strength. One to five metres, degrading as people and inventory move around, with batteries to replace on a cycle.
Printed targets. QR codes and markers at surveyed positions. Exact at the marker, drifting between them, and someone has to maintain thousands of stickers in an environment that scuffs and repaints them.
The table above sets the four against each other on accuracy, orientation, persistence and cost.
No, and the difference decides whether a deployment works.
SLAM builds a map as it goes and tracks against it. The coordinate frame starts wherever the device happened to start, so two devices never agree with each other and nothing survives closing the app. That is fine for a robot navigating a space on its own. It is useless for content that has to stay put.
A visual positioning system localizes against a map that already exists and is shared. Every device gets a pose in the same frame, and it is the same frame next week. Anchors persist, multiple users see the same thing in the same place, and a robot and a technician can be resolved into one coordinate system.
In production they run together. SLAM handles smooth frame-to-frame tracking, and the VPS supplies the absolute fix and corrects the accumulated drift.
The full comparison is on VPS vs SLAM, and how both sit against UWB, BLE and markers is on indoor positioning technologies.
It depends entirely on the job, and being honest about which job you have is the difference between a project that pays back and one that does not.
Metres are enough to route a person to a room, a gate or a store. If that is the requirement, beacons are cheaper and they work.
Centimetres are required the moment the overlay has to land on a specific object among objects that look the same. One valve among forty. One bin among nine identical aisles. One pump on a utilities floor. At that point nothing under centimetre accuracy is useful at all, because a label on the wrong unit is worse than no label.
Paying for centimetres you do not need is the most common way these projects get expensive. So is discovering at pilot stage that metres were never going to be enough.
The vocabulary used on this page is defined in the glossary. For the wayfinding layer built on top of positioning, see indoor navigation.
