Industries / Retail, hospitality and personal services / Dry cleaning and laundromats
Tracking garments
Tracking garments is knowing whose item is whose, and where it is, from the moment it is handed in or collected until it goes back: the tags, codes and chips on each piece and the records behind them.
4
Milestones achieved, from 1945 to 2024.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1945
Police read laundry marks
Laundries and dry cleaners marked each customer's items with their own codes, some written by hand and some invisible to the eye, so the items came back to the right owner. Sergeant Anthony F. Nelligan of the Baltimore police began collecting these marks as a patrolman in 1940, and in April 1945 his collection became the department's official laundry and cleaning mark file. By 1950 it held over 3,000 marks from almost every laundry and dry-cleaning business in Maryland, and Nelligan handled about 150 requests a year to trace clothing found in investigations, making a positive identification, he estimated, in 95% of them, by the FBI Law Enforcement Bulletin's account of November 1951.
LessonEach laundry had its own system of marks, so reading them took a file of all of them; the Bulletin judged a series of local files more practical than one national file.
Credit
Anthony F. Nelligan, Baltimore City Police Department
2008
Trousers that went to court
In May 2005 Roy Pearson, a lawyer, left a pair of suit trousers at Custom Cleaners in Washington, DC, for alteration, then said the owners, the Chung family, had lost them and offered him another pair. He sued over their "Satisfaction Guaranteed" sign for damages that Judge Judith Bartnoff calculated could have reached $67 million. Soo Chung testified that the number on the red ticket on the trousers she offered, 182, matched the red tag on his claim ticket. Judge Bartnoff found for the Chungs on every claim after a trial without a jury, and the DC Court of Appeals affirmed on 18 December 2008.
LessonThe shop's proof of which garment was which came down to two matching paper tags and the owner's memory of an unusual waistband.
Credit
DC Superior Court (Judge Judith Bartnoff); DC Court of Appeals (Judge Kramer)
2010
Tags that survive the wash
By 2013 commercial laundries had been tracking uniforms and linen with radio-frequency identification (RFID) tags for more than 15 years, but the early low- and high-frequency tags had to be read close up. UHF (ultra-high-frequency) tags built to survive commercial laundering became available in 2010, letting a reader record a whole cart of uniforms at once, Jeff Welles of InvoTech Systems told RFID Journal. Disney built its own system on Fujitsu tags and ThingMagic readers, and Fallsview Casino Resort in Niagara Falls adopted InvoTech's in 2011. The Scottish linen-rental firm Fishers Services said earlier tags had not been robust enough to survive repeated laundering; by 2016 it read passive UHF tags on linen, bags and cages at its new Glasgow laundry, and said it lost about £1 million of product a year across its sites, on revenue of about £36 million.
LessonTracking single pieces only became practical once a tag could outlast the item it was sewn into.
Credit
InvoTech Systems; Fujitsu Frontech; Disney; Fishers Services and UBI Solutions
2024
Linen that disappears
A survey by the Textile Services Association (TSA) of its UK commercial laundry members found that, on average, nearly half the linen they buy each year is lost, damaged or stolen. By the guide it published with the trade body UKHospitality, 40% of linen items disappear from the process and only 10% are rejected because they are worn out, and half of products last less than three months instead of the two years they would if looked after.
Credit
the Textile Services Association and UKHospitality
Watch for
Sorting untagged items by camera
The JENSEN-GROUP says Inwatec's HEIMDAL station combines RFID with cameras to recognise both tagged and untagged articles by colour, pattern, size and fabric type, so a laundry can sort items that carry no chip.
Achieved whenan outside party publishes HEIMDAL's identification error rate on untagged items at a working laundry.
Claimed by
the JENSEN-GROUP, 2026, in an article that reports its August 2026 purchase of the rest of Inwatec
Open
01
Match garments without tags
An open set of photographs of the same garments taken at the counter and again after cleaning and pressing, and a baseline model that matches each cleaned garment back to its order from the photo alone, scored against the paper ticket.
The briefCredit · unclaimed02
Laundry tag survival table
An independent table of how many industrial wash cycles each model of RFID laundry tag survives while still reading at a set distance. Sew tags from several makers into test towels, have a cooperating laundry run them through its normal process, and read each at a fixed station after every ten cycles.
The briefCredit · unclaimed
Open 01 · the brief
Match garments without tags
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open set of photographs of the same garments taken at the counter and again after cleaning and pressing, and a baseline model that matches each cleaned garment back to its order from the photo alone, scored against the paper ticket.
Why it is still open
Cleaners still prove which garment is whose with paper or heat-sealed tags, as the Pearson case showed. Inwatec's HEIMDAL claims to recognise untagged articles, but by category (colour, pattern, size, fabric type) and with no published accuracy. Searches of the web for camera matching of garments to dry-cleaning orders, and of GitHub ("garment reidentification"), found no public set of the same garments before and after cleaning.
Built with what exists
a phone on a fixed stand with one lamp; an open image model that turns each photo into a list of numbers for comparison, such as Meta's DINOv2 or OpenAI's CLIP, fine-tuned on one graphics card (see the Note "One GPU to fine-tune"); the open labelling tool Label Studio; a Hugging Face repository for the photos.
First step
Photograph 100 donated or staff garments at a cooperating cleaner, with their owners' consent, at drop-off and after cleaning, and measure how often an off-the-shelf DINOv2 model picks the right one first.
Done looks like
1,000 garments from at least two cleaners photographed before and after cleaning, and the model's top-1 and top-5 match rates on 200 held-out garments, with look-alike cases such as plain white shirts counted separately.
Open 02 · the brief
Laundry tag survival table
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An independent table of how many industrial wash cycles each model of RFID laundry tag survives while still reading at a set distance. Sew tags from several makers into test towels, have a cooperating laundry run them through its normal process, and read each at a fixed station after every ten cycles.
Why it is still open
Makers rate their own tags; HID, for one, says its tags keep performing "through hundreds of wash cycles" without naming a number. The laboratory washing tests that Crossref lists (IEEE conference papers from 2014, 2016 and 2017) are, by their titles, of research tags made of conductive textile, paint or embroidery, not of the commercial tags laundries buy. Fishers Services gave tags that failed in the wash as the reason its first RFID plans stalled.
Built with what exists
commercial UHF laundry tags bought from several makers, such as HID, Datamars and Fujitsu Frontech; a fixed UHF reader on a jig that holds each towel at the same distance; a spreadsheet; a laundry that washes the towels in its normal work, so the team runs no industrial machines itself.
First step
Buy 30 tags each of two models, sew them into 60 towels, record each tag's read distance on the jig, and agree with one laundry to run the towels through its wash.
Done looks like
30 tags of each of five models through 200 cycles, with read success and maximum read distance logged every ten cycles, and survival curves and failure counts published by model.