Automatic license plate recognition

from Wikipedia, the free encyclopedia

Automatic number plate recognition (including automatic number plate recognition or automated license plate control ) is a video surveillance method , the handwriting recognition uses (OCR) to license plate can be seen on vehicles. Automatic license plate reading systems (AKLS) can currently (as of 2005) evaluate about one vehicle per second at a driving speed of up to 160 km / h. For this purpose either permanently installed video surveillance cameras, photo and video cameras in speed measurement systems or specially set up mobile devices are used. Such systems are used by authorities for automatic evidence when collecting tolls and for traffic monitoring (such as speed and distance measurements or compliance with the red light at traffic light-controlled intersections).

French license plates: The system must be able to recognize different types of license plates based on color, shape and graphic structure.

A technically suitable system can both save the recorded images and read out the recognized text, and in some cases also save a photo of the driver. Usually infrared light is used for illumination in order to be able to take pictures regardless of the time of day. The systems also use flashlight , on the one hand to increase the image quality and on the other hand to signal the driver of his incorrect behavior. The systems used differ in detail, in particular due to country-specific differences in the license plates used.

The software used runs on PC hardware and can communicate with other programs or databases. After the license plate is located in the photo, this area is visually normalized and qualitatively improved. A character reading is then performed to obtain the alphanumeric text.

The systems evaluate either on the spot or photos are collected and sent to an outsourced computer system, where recognition takes place with a time delay. If the recognition is carried out on the spot, the entire recognition process takes around 250 milliseconds, whereby the text on the number plate, the date of the recording, the lane and other relevant data (such as the offense charged) are extracted. This relatively compact information is transmitted or stored for later retrieval. If the data is transmitted immediately without reducing the data volume, then it is processed by a powerful server, such as with the London Congestion Charge . Systems without data reduction, however, require a higher bandwidth of the data connection for the transmission of the image data.


There are a number of terms that refer to automatic license plate recognition. The following terms are used in English:

  • Automatic Number Plate Recognition (ANPR)
  • Automatic vehicle identification (AVI)
  • Car plate recognition (CPR)
  • License plate recognition (LPR)


Letters on a Dutch license plate

The most important part in the software is the character recognition on the photos ("Optical Character Recognition"; OCR ). In addition to many technical aspects, attention must also be paid to the design of the marks and the characters used on them. In Germany, the " forgery-increasing script " has been used for new license plates since 2000 . In this font, similar letter forms (e.g. number zero / letter O, 3/8, E / F or P / R) are designed to be deliberately different. Although primarily introduced to make license plate modifications more difficult by criminals, this also simplifies character recognition.

Following a similar pattern, some changes were made to the font when the Dutch license plates were converted in 2002. For example, small spaces were added to the characters P and R in order to be able to distinguish them more easily and to simplify the use of automatic license plate recognition. Some license plate systems use different font sizes and positions, which makes automatic recognition significantly more difficult. Complex recognition systems can recognize different types of license plates; however, most of them are geared towards a specific system.

Existing radar cameras or other surveillance cameras can be used as the camera system , but also mobile systems that are mostly installed in vehicles. Some systems use infrared cameras to get a better picture of the license plate.


Steps 2, 3 and 4: The license plate is first brought to a uniform contrast and brightness (normalization) and then segmented for character recognition

The following algorithms are applied to the image to recognize the text on the license plate:

  1. License plate localization: recognizes the position of the license plate in the picture
  2. License plate orientation and size: compensates for differences in the spatial position and size of the license plate
  3. Normalization: adjusts the brightness and contrast of the image
  4. Character segmentation: identifies and separates the characters in the text
  5. Character recognition: recognizes the individual alphanumeric characters.

The quality of each individual level influences the accuracy of the overall system. During the third phase, some systems try to recognize the edges of the characters - the difference in color between the characters and the background. Similarly filter used to compensate for optical distortions.


There are a variety of potential problems that the software must consider and compensate for. These are for example:

  • Bad image quality due to the great distance between the camera and the vehicle or insufficient image resolution
  • blurry images, especially at high speeds and with mobile cameras
  • poor lighting, especially poor contrast or excessive lighting due to reflections
  • hidden license plates, e.g. due to dirt or trailer hitches
  • different labels on the front (e.g. purely decorative, if allowed)
  • deliberate thwarting of admission

While some of these can be corrected by the software, the majority can only be achieved with better hardware or changes to the overall system. By placing the camera higher, the problem that other objects - such as vehicles - can cover the number plate can be eliminated. On the other hand, this creates additional problems due to the stronger distortion of the license plate.

Many countries use retroreflective labels. These reflect the light in the direction of the source, resulting in better contrast. Non-reflective signs are also often used, which increases the contrast even under poor lighting conditions. Infrared cameras are also well suited for use in such systems - in conjunction with an infrared emitter and a normal light filter in front of the camera. However, this can only be used with systems that are specially adapted for this; conventional radar cameras are usually not suitable for this. If true color images are also required for evidence purposes (e.g. of the driver), the infrared camera can be coupled with a normal light camera in order to take two images.

Blurred images make it difficult to recognize characters

Blurry images make character recognition difficult. Cameras with a very short exposure time are therefore used in order to minimize motion blur . Ideally, the exposure time is 11000  seconds. If the camera is mounted very low or if the traffic moves slowly, this time can be longer. At a speed of about 64 km / h, an exposure of 1500  second is sufficient, at 8 km / h 1250  second is sufficient .

Other vehicles, towbars, tow bars or similar objects can obscure one or two signs, as can bicycles on bike racks. Most of these obstacles can be corrected by placing the camera higher up. With regard to bicycle racks, there are some laws (e.g. in Austria , New South Wales , Australia ) that stipulate the installation of an additional, clearly visible label when bicycle racks or similar visual obstacles are used.

Depending on the application, minor errors can also be accepted. If a recognition system is used to provide access to an otherwise closed area, the incorrect or non-recognition of an individual character can usually be tolerated. It would be very unlikely that a vehicle with almost the same license plate would want to enter the site. In most areas of application, however, correct recognition of the entire license plate is a prerequisite for the functioning of the overall system.

Technical restrictions

A high error rate and possible misidentifications are criticized in general and in individual cases. Due to constant further development, however, the reliability and selectivity of the systems are constantly increasing. It can be assumed that test series will be carried out repeatedly with each new generation of devices.

The failure rate of older systems was alarming: a critic of the London system states that four out of ten number plates were incorrectly recognized. This inevitably leads to a high number of cases in which fees are charged to uninvolved citizens. For a payment of £ 10, they can have  proof that the costs are justified, for example by handing over the evidence photo. Improvements in the technology used have drastically reduced the error rates, although unjustified payment claims still occur frequently.

Further areas of application

Systems for automatic license plate recognition can also be used or are already being used for the following purposes:

  • When crossing borders
  • at petrol stations for automatic payment settlement
  • at parking lots and multi-storey car parks for billing and access control
  • for traffic management, to evaluate the speed of vehicles in the traffic flow
  • for traffic jam forecast by evaluating the traffic flow, taking into account the city and region code
  • to analyze traffic behavior (route selection, source and destinations) for traffic planning

Situation by country

Situation in Germany

In Germany, automatic license plate recognition is currently in permanent use in Saxony , Bavaria and Hesse . Brandenburg , Mecklenburg-Western Pomerania , Lower Saxony and North Rhine-Westphalia use the system for specific searches. The introduction is planned in Saarland and Baden-Württemberg .

Rhineland-Palatinate, Saxony-Anhalt, Schleswig-Holstein and Thuringia do not use automatic license plate recognition.

According to research by BuzzFeed News , the systems used in Germany are extremely flawed. The error rate in Hesse is 93%, in Saxony 97% and in Bavaria even 99%. The error rate describes the ratio of hits reported by the system to so-called "real hits", which remain after a manual check by a police officer. This obviously does not achieve the hoped-for goal of relieving the police, as more than 90 out of 100 hits have to be checked manually, which means that the alleged hit has to be discarded.

According to the research, the originally envisaged goal of using license plate recognition to investigate "correspondingly serious crimes or dangers" will not be achieved. In half of all identified crimes, only vehicles were found for which liability insurance had not been paid. In addition, offenses such as fuel fraud or license plate theft dominated. "The proportion of so-called 'Schengen hits' - i.e. cross-border vehicle theft or illegal entry within the Schengen area - is hardly measurable: In Saxony it accounts for 0.07 percent of all hits reported to the police," says BuzzFeed News.

In a decision of December 18, 2018, the first Senate of the Federal Constitutional Court declared the automatic number plate recognition according to the Bavarian Police Task Act to be largely unconstitutional as a violation of the right to informational self-determination .

Police test series in Bavaria

In Bavaria , from October 2002 to March 2003, a large-scale test was carried out under police supervision, in which the detection technology was tested at several locations under operational conditions, including at two border crossings to the Czech Republic and on the A8 Munich-Salzburg. Since 2006, the number plates of vehicles with stationary and mobile devices have been automatically recorded in Bavaria. The a utomatisierte K ennzeichen e rkennung AKE scans this month about eight million mark. Stationary devices are currently distributed across twelve locations on Bavarian motorways.

Police test series in Thuringia

A similar test was carried out from June 2003 with an unapproved device in Thuringia (Rennsteig tunnel on the A 71 ).

Police test series in Brandenburg

In October 2003 there was a test run in Brandenburg with a mobile device for number plate recognition.

Since early 2012, construction and agricultural equipment thefts by police automatic is in Brandenburg to educate passenger car, K ennzeichen- E rfassungs sy stem (KESY) set up temporarily for five years. With KESY, vehicles advertised for search could be located at eleven locations in the country.

In March 2019, it became known that in Brandenburg in 2018, through 95 temporally overlapping court decisions, more than 55,000 license plates were recorded daily using KESY and stored in advance.

After criticism from the state data protection officer Hartge , the police changed the procedure in February 2020 so that the data is now stored for a maximum of 3 months instead of indefinitely.

Police test series in Hessen

In Hesse , the Ministry of the Interior carried out a temporary technical function test on Elzer Berg in autumn 2003 without any data being saved.

Use in the United States

Mobile systems for automatic license plate recognition are widespread in the USA ; around 71% of all police offices use them.

The storage duration of the data differs greatly between the states.

A test on the border with Mexico found that of 780,000 license plates recorded, more than 1,300 were linked to a crime, including four murders.

Use in the UK

General police use

Widespread surveillance in central London , here cameras in a residential area near Trafalgar Square

After the license plate has been identified, it can be searched for and entered in a police database. This can be used to detect stolen vehicles or vehicles that have been used for a crime. Some systems also check whether the vehicle's insurance or registration is still valid.

On November 18, 2005, the British policewoman Sharon Beshenivsky was shot dead in Bradford during a robbery . The escape vehicle was registered a little later via an automatic license plate recognition system, whereupon six suspects could be arrested. The system has been described by a representative of the British police as a "revolutionary tool for fighting crime".

Project Laser (United Kingdom)

In March 2005 it was announced that over 2000 automatic license plate recognition systems would be installed across the UK. So in 2006 the UK should be the first country where virtually every vehicle movement could be monitored and recorded.

This was a logical consequence of Project Spectrum , in which all 43 police units in England and Wales were equipped with such, but mobile systems. The original project ran from September 2002 to March 2003, with the devices being tested by nine police units. From summer 2003 to summer 2004, the second phase was a field test in 23 offices. The Driver and Vehicle Licensing Agency (DVLA) has already collected data from unregistered and uninsured vehicles.

The project was seen as a great success, although reports suggest the error rate was up to 40% of misrecognized vehicles. In return, the project led to over 100 arrests per police officer per year, which was about ten times the national average. Further tests and changes to the system (introduction of infrared systems and software improvements) resulted in the error rate being reduced to 5%.

A total of around 28 million license plates were recognized during the one-year test phase; 1.1 million of them (3.9%) could be found in the database. 180,543 vehicles were stopped (101,775 of them directly due to automatic detection), which led to 13,499 arrests (7.5% of the arrests) and 50,910 penalties (28.2%). 1,152 stolen vehicles were discovered, drugs worth £ 380,000 and stolen goods valued at £ 640,000 were confiscated.

The main objective of the second phase was to determine whether the system would amortize itself. The result was that only 10% of the costs incurred could be recovered through penalties. It was stated, however, that many of the punished did not pay on time and that this was the only reason that this low figure was reached. It was recommended by the operators to continue the project and install it nationwide.

Funding for the nationwide system is now secured. The basis for this is a data center that can recognize 50 million license plates every day. It should be operational by March 2006.

Legal basis

Systems that collect data and save it without analysis are questionable with regard to data protection and the fear of being monitored by government or other bodies (“personal movement profile creation”).

In addition, the “intimidating effect” of the measure is often criticized - for example by Clemens Arzt , police lawyer and professor for constitutional and administrative law at the Berlin School of Economics and Law. The fact that the data is recorded covertly and without notice and that citizens therefore have no opportunity to judge an administrative act directed against them is repeatedly criticized. Dieter Müller, Professor of Road Traffic Law at the University of the Saxon Police and Chairman of the Legal Advisory Board in the German Road Safety Council, also points out that, in his opinion, the system can only be used lawfully “if you pull this person out immediately and you get the hit Suspicion of a criminal offense is held up. ”Müller sees the current variant as an unlawful measure,“ if one then has the hit, but not immediately pursued it. For me that even goes in the direction of thwarting punishment in office. "

Patrick Breyer , among others, had lodged a constitutional complaint against the authorization for the Federal Police . In December 2018, the Federal Constitutional Court ruled that the state laws in Bavaria, Hesse and Baden-Württemberg were partly unconstitutional, among other things because the states lacked the legislative competence for criminal prosecution . For the purpose of police security , however, the court allowed the registration of license plates.


Other concerns include data retention of the recorded data. These could be used to collect information about people, such as driving behavior or even daily scheduling (“personal movement profiles”). Such personal data are normally strictly protected by the legal systems within the framework of the fundamental right to informational self-determination .

Legal situation in Germany

Decision of the German Federal Constitutional Court on March 11, 2008

On March 11, 2008, the German Federal Constitutional Court declared the Hessian and Schleswig-Holstein law on automated license plate recognition null and void. The court stated that the automated registration of license plates encroaches on the protection of the basic right to informational self-determination if the license plate is not immediately compared with the wanted list and deleted immediately without further evaluation. The court denied that the interventions in this fundamental right legitimized by the two state laws are based on a constitutional legal basis, among other things because the challenged provisions do not meet the constitutional requirement of proportionality .

Operation of collection systems after the court ruling

The federal states of Bavaria , Baden-Württemberg and Lower Saxony in particular continue to use license plate recognition. The former Baden-Württemberg Interior Minister Heribert Rech (CDU) said that the use of automatic license plate reading systems is an important tool in the search work.

The license plate data are meanwhile (status: 2010) compared in real time with the police search computer. If this results in a hit, the responsible motorway police inspection is automatically informed without delay in order to pull the vehicle in question out of the traffic.

The Munich Administrative Court and the Bavarian Administrative Court declared the operation of the systems in Bavaria to be permissible. According to a study by the data protection expert Alexander Roßnagel commissioned by the ADAC in April 2009, however, almost all of the relevant state laws are in part unconstitutional.

Bavaria operates (as of November 2013) 30 systems at twelve permanent locations and an additional three mobile systems. The systems are used to prosecute light and serious crimes. An average of eight million vehicles are photographed per month in Bavaria , with 500 to 600 hits being reported per month.

According to a draft law of the federal government of January 23, 2017 (printed matter 18/10939), the automatic scanning of license plates at the German borders is to be enabled under certain conditions according to § 27b BPolG (draft) and the monitoring of traffic routes is to be expanded accordingly.

Decision of the German Federal Constitutional Court on December 18, 2018

On December 18, 2018, the Federal Constitutional Court once again declared the statutory regulations for automatic license plate recognition in Bavaria, Baden-Württemberg and Hesse to be unconstitutional and void. In all three Länder, the Länder already lack legislative competence for the regulations , namely insofar as the automatic license plate recognition is intended to serve criminal prosecution and this is not limited to criminal offenses of considerable weight, since criminal prosecution falls within the competing legislative competence of the federal government. This also applies to the Bavarian regulation insofar as the automatic number plate recognition is intended to prevent illegal immigration, since immigration law falls under the exclusive legislative competence of the federal government. The Hessian and Baden-Wuerttemberg regulations further violate the quotation requirement , as they restrict the basic right to freedom of assembly without the basic right concerned being cited in the law.

Opinion regarding the prosecution of serious crimes / use of the truck toll systems

The restriction on the use of the systems used for billing and monitoring truck tolls , which are based on the same technology, only for the purpose of toll collection, has been heavily criticized by the police, the judiciary and broad circles of politics since its introduction in 2005 ( for example from the former Federal Public Prosecutor Nehm ). A nationwide uniformly regulated expansion of the use of this technology for the prosecution of serious crime cases is therefore very likely in the foreseeable future.


Web links



Individual evidence

  1. Stockholm Traffic Cameras ( Memento from March 20, 2009 in the Internet Archive )
  2. ^ Markus Friedrich, Prokop Jehlicka, Johannes Schlaich, 2008. " Automatic number plate recognition for the observance of travel behavior ". Retrieved January 11, 2011.
  3. a b c d e f g The police have recorded hundreds of millions of license plates, although the system does not work for them . In: BuzzFeed . ( [accessed October 16, 2018]).
  4. ^ German Bundestag - answer of the federal government: answer of the federal government to the small question of the deputies Jan Korte, Herbert Behrens, Jens Petermann, other deputies and the faction DIE LINKE. Printed matter 17/14431. Retrieved October 15, 2018 .
  5. Federal Constitutional Court - Press - Automated vehicle license plate checks according to the Bavarian Police Task Act in parts unconstitutional. Retrieved April 12, 2020 .
  6. Automatic License Plate Recognition - Where your license plate is captured . In: . October 23, 2014. Retrieved June 3, 2015.
  7. ^ Volkmar Krause: Police action ready for film . In: . June 17, 2013. Retrieved June 3, 2015.
  8. Brandenburg and Kesy take action against car theft . In: . May 5, 2012. Retrieved June 3, 2015.
  9. Ministry of the Interior: License plates recorded every day in 2018 . In: . May 9, 2019. Retrieved May 11, 2019.
  10. Ismahan Alboga: License plate registration system - dispute over Kesy . In: . May 9, 2019. Retrieved May 11, 2019.
  11. Data protection officer criticizes the planned search for license plates. Retrieved April 12, 2020 .
  12. Brandenburg police change license plate recognition. Retrieved April 12, 2020 .
  13. German Police 4/2004, p. 16
  14. Archive link ( Memento from January 29, 2013 in the Internet Archive )
  16. License plate data hits 42M mark 2013
  17. Constitutional complaint: No vehicle mass comparison at border crossings! [added on May 27, 2018] . In: Patrick Breyer . May 19, 2018 ( [accessed October 16, 2018]).
  18. Decision of the First Senate of December 18, 2018. Federal Constitutional Court, May 31, 2019, accessed on December 18, 2018 .
  19. Federal Constitutional Court : Hessian and Schleswig-Holstein regulations for the automated recording of license plates are void , press release of March 11, 2008
  20. Mass controls: Bavaria, Lower Saxony and Baden-Württemberg are blocking the Autoscan stop. In: Spiegel Online . March 11, 2008, accessed June 9, 2018 .
  21. ( Memento from May 3, 2013 in the web archive ) from September 23, 2009 (Ref. M 7 K 08.3052)
  22. Judgment (PDF; 274 kB) of December 17, 2012 (Az. 10 BV 09.2641)
  23. ADAC - traffic advocacy group for number plate recognition ( memento of August 25, 2014 in the Internet Archive ) (PDF; 115 kB)
  24. Stefan Krempl: Bavaria defends license plate scanning. In: Heise . November 29, 2013, accessed August 22, 2015 .
  25. ^ Monika Ermert: Administrative Court of Munich: License plate scanning is legal. In: Heise . September 23, 2009, accessed August 22, 2015 .
  26. Andreas Salch: The police, your big brother. In: Süddeutsche Zeitung . October 19, 2011, accessed August 22, 2015 .
  27. BVerfG, decision of December 18, 2018, AZ 1 BvR 142/15
  28. BVerfG, decision of December 18, 2018, AZ 1 BvR 2795/09, 1 BvR 3187/10
  29. Report on about Nehm's request to use the truck toll data for criminal prosecution (January 2006) ( Memento from April 27, 2009 in the Internet Archive )