Plate formats

The detector finds plates from many countries, but a reading is only accepted when it fits a plate layout you have turned on. ANPR_PLATE_FORMATS lists them, in order of preference. The default is tr, Türkiye only. Since 0.3.0.

export ANPR_PLATE_FORMATS=de,fr,tr

Countries

Code	Country	Layout	Canonical form	Region
tr	Türkiye	2-digit province, 1–3 letters, 2–4 digits	34 ABC 1234	province code and name
de	Germany	2–5 letters, 1–4 digits without a leading zero, at most 8 together, optional E or H	MAB 1234	none
fr	France (SIV, 2009 on)	2 letters, 3 digits, 2 letters; no I, O, U; no SS, no 000	AB-123-CD	none
it	Italy (1994 on)	2 letters, 3 digits, 2 letters; no I, O, Q, U	AB 123 CD	none
es	Spain (2000 on)	4 digits, 3 consonants; no Ñ, no Q	1234 BCD	none
be	Belgium (2010 on)	1 digit, 3 letters, 3 digits	1-ABC-123	none
pt	Portugal	AA-00-AA, 00-AA-00, 00-00-AA or AA-00-00	AA-00-AA	none

German plates come out as letters and digits, MAB 1234, not M-AB 1234. The district code is one to three letters, so the split between the district and the rest can’t be told from the characters alone; that needs the list of district codes, which isn’t built in yet. Matching isn’t affected: hashes ignore spaces and hyphens, so M-AB 1234 added to the watchlist still matches.

Only Turkish plates have a region. For the others, province_code and province_name are null in every response.

How a reading is checked

1. The OCR text is stripped of spaces and put in upper case.
2. Each format in ANPR_PLATE_FORMATS tries it as it stands, in order. The first that accepts it wins.
3. If none does, each format tries again with look-alike characters swapped for what its layout expects in each position: 0 and O, 1 and I, 8 and B, 5 and S, 2 and Z, 6 and G. A Spanish I2B4BCD becomes 1284BCD.
4. If that fails too, parsed is null and the reading isn’t confirmed on the stream.

An exact reading in any format beats a corrected one. When two formats accept the same text as it stands, the one listed first wins: AB123CD is valid in both France and Italy, so with ANPR_PLATE_FORMATS=it,fr it is read as Italian.

Turn on only the countries you expect. Every extra format is another way for a misread to look valid, and the short Portuguese layouts in particular accept many six-character strings.

In the responses

Every parsed object has a country field with the code from the table:

"parsed": {
  "country": "de",
  "province_code": null,
  "province_name": null,
  "letters": "MAB",
  "digits": "1234",
  "canonical": "MAB 1234"
}

anpr infer and anpr watch show the province for Turkish plates and the country code for the others. Stored detections have a country column, which GET /api/v1/detections returns; the service adds it to a database made by an older version on start-up, and rows from before then keep it null. The anpr_confirmed_plates_total metric has a country label.