Augmented Reality
Every approach to asset tracking except one requires you to attach something to the asset first. RFID needs a tag. Ultra-wideband needs a tag and a grid of anchors. Barcodes need a printed label and a person to scan it. Camera-based identification needs nothing on the asset at all, because the asset itself is what gets recognised.
That one difference decides most deployments. It is worth being precise about where it helps and where it does not, because the honest answer is that both approaches win somewhere.
The tag is the cheap part. The cost is everything around it. Someone has to attach a tag to every unit you want to find, register it in a system, replace the battery on the active ones, and repeat all of that for whatever arrives next quarter. The burden scales with the number of things you want to identify, which is the wrong way round. A site with two hundred assets absorbs it. A site with twenty thousand does not.
There is a second cost that only shows up later. A tag tells you where the tag is, not where the asset is. Tags fall off. They get swapped between units during maintenance. They stay on a pallet after the equipment has moved on. The system reports that the asset is in bay 14 because that is where its tag is, and nobody finds out otherwise until someone walks to bay 14.
Three cases, and they are real ones.
No line of sight. Radio passes through boxes, crates and shelving. A camera does not. If the thing you are looking for is inside a sealed case or buried three pallets deep, RFID will find it and vision will not.
High count, low value. Tens of thousands of small items that all look identical and are worth a few pounds each. Tagging is cheap per unit at that scale and the alternative is not obviously better.
Continuous position on things that never stop moving. Forklifts, tuggers and AGVs benefit from ultra-wideband, which holds ten to thirty centimetres continuously without anyone pointing a camera at anything.
The inverse cases, and they are just as real.
Large fixed and semi-fixed equipment. Pumps, panels, compressors, machine tools. Too few to justify a tagging programme, too many to remember, and they stay put long enough that a map of the space stays accurate.
Anything already modelled. If the asset exists in CAD or in a scan, the model is the recognition target. No new data collection.
Telling apart units that look identical. This is the case tags were supposed to solve and often do not. Twelve identical valve assemblies in one skid, and the work order is for the third from the left. A tag answers that only if the right tag is on the right unit, which is exactly the assumption that fails in the field.
| RFID | UWB | Barcode or QR | Vision | |
|---|---|---|---|---|
| Needs something on the asset | Yes | Yes | Yes | No |
| Needs fixed infrastructure | Readers | Anchor grid | No | No |
| Works without line of sight | Yes | Yes | No | No |
| Identifies which specific unit | If the tag is right | If the tag is right | If the label is right | From the asset itself |
| Typical accuracy | Zone level | 10 to 30 cm | At the label | Centimetre level |
| What drives the cost | Tagging every unit | Installing the grid | Labelling and scanning | Capturing the space once |
People collapse these into one and then get confused by the accuracy figures, so it is worth separating them.
Where am I, and where is that asset in this building? That is positioning. A visual positioning system answers it by matching what the camera sees against a prebuilt map of the space, returning position and orientation to sub-5 cm. It is what puts a marker on the right pump from across the room.
Where is this specific object relative to my camera right now? That is object tracking, which recognises the asset from a 3D model of it and locks to sub-millimetre. It is what lets an overlay sit on the correct port of the correct manifold while a technician works.
The two numbers belong to two different products. Positioning gets you to the asset. Object tracking works on it.
Most sites that run this properly end up with both. Tags on the sealed and the small, vision on the large and the fixed, and a single map underneath so the two systems agree on where anything is. Without that shared frame you have two inventories in two coordinate systems and a person reconciling them.
The map is the part worth getting right first. Everything else plugs into it.
How many units do you actually need to identify, and is tagging all of them a one-time job or a permanent one? Can a person get a camera in front of the asset, or is it enclosed? Do you need to know which specific unit, or is the zone enough? And when the layout changes, what has to be redone?
That last question is the one that decides whether the system is still in use in two years. If you want to see how the map side of this holds up when a site keeps changing, the warehouse walkthrough covers it on a live floor.
