EP2320384B1 - Method of licence plate recognition and categorisation of vehicles for toll purposes - Google Patents

Description

The present invention relates to a method for automatic license plate recognition of a vehicle, in particular for tolling purposes in road toll systems according to the preamble of claim 1.
Such a method is known from US 2007/008179 A1 known.

Road toll systems based on automatic license plate recognition by image capturing the vehicle are also referred to as video toll systems. The vehicles are taken when passing a toll collection station with one or more cameras. The cameras record a front and / or rear view of the vehicle in which the license plate is as clear as possible. From the image, the vehicle license plate is read using methods of automatic character recognition (OCR). In a database of the toll center, a vehicle account with the vehicle registration number is maintained for each vehicle. The vehicle passage is assigned to the vehicle account whose license plate matches the OCR license plate reading result.

When vehicles from different countries - or from different states in the USA - pass through a toll collection station, it is sometimes necessary to identify the state or coat of arms or the type of license plate in addition to the license plate and the country that issued the license plate, to enable a clear assignment of the read license plate to a vehicle account.

A problem with the known methods is the high error rate of the OCR number plate recognition. The incorrectly read license plates must be corrected manually and assigned to a vehicle account. In the worst case, it can happen that a wrongly read license plate confused with a license plate of another vehicle and this vehicle account is charged with the passage.

Improvements in the assignment of the current license plate image to a vehicle account have therefore already been proposed in which, in addition to the license plate itself, the appearance of the license plate is also taken into account. For this purpose, essential features are extracted from a current license plate image. The essential features are intended to uniquely represent the appearance of the license plate and are referred to as a "fingerprint." In addition to the license plate, a reference fingerprint is also stored for each vehicle account of the database ,

Since the fingerprints are calculated from images taken in different light and weather conditions, and also the relative position between the license plate and camera at the passage of a toll station varies, the detected fingerprints of the same vehicle are usually slightly different, making the comparison difficult , In addition, fingerprints include many features, i. are similar to high-dimensional vectors, which makes the search for similar fingerprints very compute-intensive and even with database techniques is not fully possible, especially if the databases contain very many entries, as is the case with video-based systems.

The invention has for its object to provide methods for automatic license plate recognition and toll categorization, which have a lower error rate and higher reliability than the known methods and thereby are quick and easy to implement.

This object is achieved with a method for automatic license plate recognition with the features of claim 1.

The method of the invention is based on the use of OCR tag read results and fingerprints and is based on the recognition that the errors of OCR readings are not randomly distributed, but that the same misreads occur again and again for one and the same tag. For example, it may be that a license plate ABC123 is usually read correctly, but occasionally as ABCI23 or as A8C123. Thus, when the automatic license plate recognition reads ABC123, it makes sense to consider only those vehicle accounts whose license plate is now and then read as ABC123 for an assignment of the toll passage using the fingerprint comparison. The method of the invention proposes to store in the database a list of possible reading results of the automatic license plate recognition, which contains only those known variants of license plate reading results for the license plate of the vehicle, which meet minimum quality or quantity requirements.

• Es werden weniger Zuordnungskandidaten aus der Fahrzeugdatenbank gefunden.
• Die Wahrscheinlichkeit, daß das korrekte Fahrzeugkonto in der Liste der Zuordnungskandidaten enthalten ist, ist höher.

The method of the invention thus takes into account selected past read errors of the OCR license plate recognition for each individual license plate and thereby provides better results:

• Fewer candidate candidates from the vehicle database are found.
• The probability that the correct vehicle account is included in the list of assignment candidates is higher.

In an advantageous implementation, the n: n assignment list of possible license plate reading results and vehicle license plates is successively filled during the operation of the toll system. Newly created vehicle accounts have only entered the correct indicator in the list of possible license plate reading results. For each misreading of the indicator, the reading result is checked manually, assigned to a vehicle account and added to the list of possible license plate reading results. Each read variant of the license plate only needs to be processed once and entered into the vehicle account.

The invention according to claim 1 is accordingly characterized in that, if in step d) no identical label reading result is determined or in step e) no similar or no similar candidate fingerprint can be determined, the vehicle identifier for the current label reading result is determined manually.

In these cases, the vehicle license plate in the database is assigned the current license plate reading result as a new or additional license plate reading result and preferably also the current fingerprint as a new or additional reference fingerprint if the current license plate reading result reaches a minimum frequency and / or minimum quality.

• wenn der Vergleich des aktuellen Fingerprint mit dem/den dem Fahrzeugkennzeichen zugeordneten Referenzfingerprint(s) fehlschlägt und der aktuelle Fingerprint eine bessere Qualität oder mehr Information als der bzw. die gespeicherte(n) Referenzfingerprint(s) aufweist, der aktuelle Fingerprint dem Fahrzeugkennzeichen als neuer oder zusätzlicher Referenzfingerprint zugeordnet werden; oder
• wenn zum Fahrzeugkennzeichen kein Referenzfingerprint gespeichert ist und der aktuelle Fingerprint Mindestqualitätsanforderungen erfüllt, dieser dem Fahrzeugkennzeichen als neuer Referenzfingerprint zugeordnet werden.

Preference may also be given in cases

• if the comparison of the current fingerprint with the reference fingerprint (s) associated with the vehicle identifier fails and the current fingerprint has better quality or more information than the stored reference fingerprint (s), the current fingerprint is the newer to the vehicle identifier or additional reference fingerprint; or
• if no reference fingerprint is stored for the vehicle registration number and the current fingerprint has minimum quality requirements met, this be assigned to the vehicle registration as a new reference fingerprint.

Preferably, an association is also removed from the database if it is not used within a predetermined period of time.

In the event that the OCR flag reading detects different read variants of the tag, it is also useful to use the read variants in step d) in the search for associated vehicle accounts or fingerprints and / or when creating the assignment list as possible reading results of a license plate to save.

As such, the fingerprint may be generated in any manner known in the art. It is particularly advantageous if the sensory detection takes place by means of a camera with which an image of the license plate and / or the vehicle is made, from which features are extracted and from which the fingerprint is calculated. Alternatively or additionally, the sensory detection can take place by means of sensors such as light barriers, laser scanners, induction loops, weight sensors or radio communication means with which characteristic features of the vehicle are detected, from which the fingerprint is calculated.

Investigations by the applicant have shown that a large part of the entries in the list of license plate reading results only refers to very few vehicle license plates. It may be that there are some very difficult to distinguish characteristics. For example, the flags ABC123 and ABCI23 could be very similar. Another aspect of the invention is therefore that for each entry in the list of possible license plate reading results in addition to differentiate the fingerprints of assigned vehicle license plate is stored. If, for example, both the indicator ABC123 and ABCI23 are read as ABCI23, then it makes sense, when comparing the fingerprints, to give a higher ranking to the region in which the two identifiers ("I" and "1") differ. to weight. Such "information about the relevant differences" of two or more license plates, which could be confused due to the automatic license plate recognition, is stored in the list of possible license plate reading results as a weighting for the associated fingerprints.

This makes sense, in particular, because license plates often differ only very slightly, but on the other hand, large variations of the fingerprint for one and the same vehicle may result due to soiling, light reflections or the like. Such large-scale influences may mean that the slight local differences in the fingerprint comparison are not sufficiently taken into account.

A further preferred embodiment of the invention is accordingly distinguished by storing weightings for individual regions of the assigned reference fingerprints in the database for each license plate reading result, preferably for each individual assignment thereof to a vehicle license plate, which are taken into account in the comparison of similarity. These weights form the so-called "information about the relevant differences".

In toll systems there is often the problem that vehicles that are not subject to tolls are classified as toll. An example is the classification of a toll-free large car (eg a van) as a toll vans. This misclassification stems from measurement inaccuracies in the sensors for the classification or is simply due to the fact that some vehicle classes can not be defined on the basis of defined external characteristics. In these cases, the misclassification must be corrected manually. Often, the correction must be made on the same vehicle at each vehicle passage. The inventive method can still be used profitably in these cases, if in the vehicle list an entry for the vehicle is stored in which is entered that the vehicle is not subject to toll. The problem with this is that with it an entry for a vehicle is created, which is not subject to toll: Such an approach may violate, depending on the legal situation against data protection provisions. To solve this problem, the proposed method can also be used in an advantageous embodiment so that the label reading results before they are stored in the list are encrypted.

Accordingly, a preferred embodiment of the invention is that the tag reading results are stored in encrypted form in the database and the current tag reading result for the equality search of step d) is encrypted in the same manner.

The method of the invention can preferably be used in addition to the toll categorization of the vehicle by also assigning a toll category to each reference fingerprint in the database and, if this reference fingerprint is determined to be the most similar candidate fingerprint, its assigned toll category.

• Fig. 1 ein Flußdiagram des erfindungsgemäßen Verfahrens zur automatischen Kennzeichenerkennung und
• Fig. 2 ein Flußdiagram des erfindungsgemäßen Verfahrens zur Mautkategorisierung.

Other objects, features and advantages of the invention will become apparent from the following description of the preferred embodiments thereof taken in conjunction with the accompanying drawings, in which:

• Fig. 1 a Flußdiagram the inventive method for automatic license plate recognition and
• Fig. 2 a Flußdiagram the method according to the invention for toll categorization.

Fig. 1 shows the method for automatic license plate recognition based on an image 1 of a vehicle 2 with a vehicle registration plate LP in the context of a road toll system. As used herein, the term "vehicle" includes any type of land, air and water vehicles, both self-propelled and externally propelled.

In a first step (b), an OCR reading of the license plate LP of the vehicle 2 takes place from the image 1 in order to obtain a current license plate reading result LP _OCR , eg "ABC123".

In a preceding, parallel or subsequent step (c), one or more characteristic features of the vehicle 2 are sensed in order to obtain therefrom a so-called "current fingerprint" FP of the vehicle 2. Such characteristic features are, for example, the appearance of the License plate, the vehicle color, the appearance of the vehicle front, etc., which can serve to verify the vehicle identity, and / or the vehicle dimensions, the vehicle shape, the presence of a trailer, etc., which can be used for example for the toll.

For example, the fingerprint FP can be extracted directly from the camera image 1, e.g. by determining significant bright / dark areas, calculating image parameters such as contrast, cross sums, etc., u.zw. both of the entire vehicle 2 and parts of the vehicle 2. In particular, it is possible that the fingerprint FP is formed in this way from the image of the vehicle registration plate LP itself.

Alternatively or additionally, the fingerprint FP can be obtained from other or further images 1 of the vehicle 2, from data from sensors such as laser scanners, light barriers, induction loops, video cameras, weight sensors, radio systems, e.g. RFID transponder queries, etc. are determined. Different types of sensor data can also be combined in one and the same fingerprint FP.

In a step (d), the license plate reading result LP _{OCR is} looked up in a database DB of the road toll system. In the illustrated embodiment, the database DB contains a first list 3 with stored license plate reading _results LP _{OCR, men} and a second list 4 with all vehicle license plates LP registered in the road toll system and associated reference fingerprints FP _ref . Between the lists 3, 4 there is an n: n-relational (n ∈ N ) assignment of the license plate _reading results LP _{OCR, meme} on the one hand to the vehicle license plate LP or their reference fingerprints FP _ref , as represented by the lines 5: a license plate reading result LP _{OCR, mem} can refer to one or more vehicle identifiers LP or their reference fingerprints FR _ref ; and for a vehicle identifier LP (or a reference fingerprint FP _ref ) one or more possible label _{reading results} LP _{OCR, mem may be} stored.

Thus, in the look-up step (d), the current mark reading result LP _OCR can be looked up directly in the list 3 of the stored mark _reading results LP _{OCR, mem} , since there is only one identical record LP _{OCR, mem} in the list 3 for each possible mark _{reading result} LP _OCR This makes the implementation significantly easier and faster.

In a subsequent step (e), all reference fingerprints FP _{ref of} the relevant vehicle identifier LP _that match the current license plate reading result LP _OCR are then determined as "candidate fingerprints" FP _{ref, cand} and respectively compared with the current fingerprint FP. This comparison is a similarity comparison in which the "most similar" candidate fingerprint FP _{ref, cand is} determined; the vehicle identifier LP assigned thereto is outputted as a result of the license plate recognition process (step f).

For the determination of the "most similar" candidate fingerprint, a wide variety of similarity criteria can be specified, for example the maximum value of a numerically calculated similarity measure and / or exceeding a minimum similarity measure etc. When determining the "most similar" candidate fingerprint, it can also be checked whether the similarity criteria for no other Candidate fingerprint apply and thus the "most similar" candidate fingerprint is uniquely determined, ie only a clearly most similar candidate fingerprint is used for further processing.

In the list 4 of the database DB, several reference fingerprints FP _ref can also be stored for a vehicle identifier LP. For example, LP reference fingerprints FP _ref can be calculated from different image recordings showing the license plate of the vehicle 2 under different light conditions or viewing angles for one and the same identifier. In the fingerprint similarity comparison of Step (e) then the similarity comparisons of the current fingerprint FP can be combined with all candidate fingerprints FP _{ref, cand} one and the same identifier LP to a common similarity value, eg by adding up, by selecting the maximum value, etc.

If no similar label _{reading result} LP _{OCR, mem} in the list 3 or in step (e) no similar or no similar candidate fingerprint FP _ref, can be determined in step (d), a branch is made in each case to a step (g) the license plate LP by an operator manually, in particular visually, determined from the image 1 and the current license plate _reading result LP _OCR and from this a new entry for the database DB is generated. In this case, the manually determined vehicle identifier LP is assigned the current license plate reading result LP _OCR as a new or additional license plate _{reading result} LP _{OCR, mem} and / or the current fingerprint FP as a new or additional reference fingerprint FP _ref . It can also be checked whether the current license plate reading result LP _OCR or the current fingerprint FP reaches a minimum frequency and / or a minimum quality that qualifies it or him to be included in the database DB.

Criteria are also preferably provided to remove assignments of license plate _reading results LP _{OCR, mem} to license plate LP from the list 3 of the database DB, for example if they are no longer being accessed, for example because a license plate has been renewed and thus become more readable is, so that certain license plate reading results no longer occur.

Optionally, weightings WT for reference fingerprints FP _ref can also be determined in a step (h) and stored in the database DB for each assignment LP _{OCR, mem} to an identifier LP, which weightings WT are taken into account in the fingerprint comparison of step (e), around this to focus on particularly distinctive or relevant areas of the fingerprints FP and FP _{ref, cand} to be compared.

It is also possible, for certain vehicle identifiers LP or license plate reading results LP _OCR - should there be an increased misallocation in these - to make increased demands on the fingerprint similarity comparison of step (e) or to exclude them entirely from an automatic assignment.

Some OCR tag recognition methods provide several alternative tag reading results, one most likely to be most likely, while the other are possible but less likely alternatives. The alternative label reading results may be used beneficially in step (d) and step (g). In step (d), the alternative read results can also be looked up in the list 3 of the stored label reading results LP _{OCR, mem} and the associated reference fingerprints FP _{ref can be used} for the fingerprint comparison. In step (g) all possible read variants of the flag can be entered in the database DB.

As already explained briefly, the label _{reading results} LP _{OCR, mem} stored in the list 3 can be encrypted for privacy reasons. For the equality comparison in step (d), the current label reading result LP _OCR is then encrypted in the same manner; it is therefore not necessary to use, transfer or store in the database DB or in the context of the present method license plate reading results in plain text.

The encryption can be done by any known method which ensures that label _{read results} LP _OCR , LP _{OCR, mem} can not be accessed without authorization. In fact, in the case of the method presented here, it is not necessary at all to know or decrypt the license plate reading result in plain text, ie an irreversible encryption can be used. It is enough if the encryption is a mathematically injektive function and therefore each encryption result is only generated exactly by a license reading result: For the method is based on comparing a current license plate reading result with a list stored license plate reading result and looking for exact matches, not just similar characteristics , Such a search for an exact match can also be carried out if both the current license plate reading result LP _OCR and the stored license plate _{reading result} LP _{OCR, mem are present} in a form encrypted in the same way.

Advantageously, the encryption is a so-called "one-way function", i. a difficult invertible function, e.g. an asymmetric encryption with public and private key. The private key that would be needed to decrypt can be destroyed so that no one can infer the actual label reading result from the encrypted label reading result.

Encrypting the license plate reading results also makes it possible to apply the method to section control, parking and other closed systems, which place particularly stringent data protection requirements. It is to distinguish between closed systems in which a vehicle passage is assigned at the exit from the closed system of a vehicle passage at the entrance to the closed system, and systems with registered users, in which vehicle passages are assigned to a vehicle account. For closed systems (without registration) saving a list of possible license plate reading results does not make sense. However, also in these systems, a preselection of fingerprints based on the license plate reading result, whereby the encryption of the license plate reading results may be required for data protection reasons, especially in these applications.

A very advantageous extension of the method described is to use the location and time of a vehicle passage in the recognition of the vehicles to make plausibility of the assignment of the vehicle passages to vehicle accounts. In doing so, e.g. Checked from several vehicle detections, whether a vehicle could be at a specific location at the specified time.

Fig. 2 shows a further advantageous embodiment of the invention, in addition to the automatic license plate recognition and an automatic toll categorization of a vehicle is performed, on the basis of the discussed comparison of fingerprints, the fingerprint may include features with which the vehicle characteristics can also be verified. For this purpose, in the database DB the reference fingerprint (s) FP _{ref of} an identified vehicle having an identification ID, eg its vehicle identifier LP, is also assigned one (or more) toll category (s) MK. If the current fingerprint FP of the vehicle passage with one of the above-identified candidate fingerprints FP _{ref, cand} best similar resembles, the associated toll category MK is taken from the database DB and output in step (f) (see also Fig. 1 ).

The following example illustrates a possible variant of this method.

A van passes through a toll station during a first vehicle passage. The sensors installed at the toll station, e.g. a laser scanner, detects the vehicle and classifies it (wrongly) as a pickup truck. As the van only pays for the (lower) toll category "Van", a contradiction is found.

In the first vehicle passage, which leads to a contradiction, there is a manual processing, in the course of which the contradiction is corrected and the van is correctly classified as a toll category "Van". Due to the correction of the contradiction the fingerprint is stored in the vehicle account together with the toll category determined by the processor.

If a contradiction occurs again in another passage of the same vehicle, the fingerprint of the current passage is compared with the reference fingerprint stored in the vehicle account. If the two fingerprints match (i.e., the similarity measure is above a predetermined threshold value), the toll category determined by the processor during the first passage is adopted and the contradiction is thus automatically corrected.

The method is based on the knowledge that one and the same vehicle, which has not changed in its outer shape, can be assigned to the same toll category again. In this case, in a first step, the identification of the vehicle or the assignment to a vehicle account, e.g. with the aid of the method previously described for automatic license plate recognition. In a second step, it is ensured by means of the fingerprint comparison that changes to the vehicle are detected and the toll category stored in the vehicle account is only accepted if the vehicle has remained externally unchanged. Examples of changes to the vehicle are when the vehicle is being renewed (e.g., same license plate on new vehicle) or e.g. carries a trailer or caravan.

On the other hand, known methods for checking whether the correct toll category has been paid use only the data measured by the sensors in order to determine the toll category according to a classification method identical for all vehicles. In contrast, the categorization method presented here is able to store features of individual vehicles and to reuse them for a later categorization decision.

As in the database of Fig. 2 As can be seen, a vehicle with a specific identification ID or LP can also have different reference fingerprints FP _ref for its appearance be assigned in different configurations K with associated toll categories MK, so that one and the same vehicle in different configurations K, eg with and without trailers, can be automatically categorized and vermautet.

In a simplified embodiment of the method, for example, only the outer dimensions of the vehicle can be stored as reference fingerprint FP _ref ; The presence of a trailer can then be easily recognized eg by a large change in length.

Another very advantageous embodiment is to save trailers as their own vehicles in the database DB with its own reference fingerprint. During the passage, the towing vehicle and trailer are identified separately and optionally the characteristics of towing vehicle and trailer are verified. The determined toll category MK then results from a combination of the categories of the towing vehicle and the trailer.

The determination of the toll category MK by means of fingerprints FP can be based on the presented here method for license plate recognition based on OCR license plate reading results LP _OCR , in which case can be used as a vehicle identification ID directly its flag LP. Alternatively, any other known method for vehicle identification could be used for this, for example a vehicle identification by means of an electronic tag attached to the vehicle, such as an RFID transponder chip or an onboard unit (OBU) for electronic toll debiting.

• eine den Schritt (b) ausführende Komponente "Vehicle Identification" identifiziert das Fahrzeug und ordnet es einem Fahrzeugkonto zu;
• eine den Schritt (d) ausführende Komponente "Retrieve Reference Sensor Fingerprint" lädt aus dem Fahrzeugkonto des identifizierten Fahrzeugs und den zugeordneten Einträgen der Datenbank DB den bzw. die zugeordneten Referenzfingerprint(s) als Kandidatenfingerprint(s);
• eine den Schritt (c) ausführende Komponente "Generate Sensor Fingerprint" berechnet den aktuellen Fingerprint der Fahrzeugpassage;
• eine den Schritt (e) ausführende Komponente "Fingerprint Comparison" vergleicht den aktuellen Fingerprint mit einem oder mehreren Kandidatenfingerprints und liefert als Ergebnis, ob und mit welchem Kandidatenfingerprint eine Übereinstimmung besteht;
• eine weitere Komponente "Enforcement Decision" verwendet das Ergebnis des Fingerprintvergleichs zusammen mit dem Ergebnis einer gesonderten Klassifizierung ("Automatic Classification") und einer vom Fahrzeugbenutzer angegebenen Mautkategorie ("Claimed Class") und stellt fest, ob ein Widerspruch besteht.

Fig. 2 schematically shows the components of the method for automatic toll categorization using fingerprints and other involved system components as follows:

• a component "Vehicle Identification" executing step (b) identifies the vehicle and assigns it to a vehicle account;
• a component "Retrieve Reference Sensor Fingerprint" executing step (d) loads from the vehicle account of identified vehicle and the associated entries of the database DB or the associated reference fingerprint (s) as a candidate fingerprint (s);
• a component "Generate Sensor Fingerprint" executing step (c) calculates the current fingerprint of the vehicle passage;
• a component "fingerprint comparison" executing step (e) compares the current fingerprint with one or more candidate fingerprints and as a result delivers whether and with which candidate fingerprint a match exists;
• another component "Enforcement Decision" uses the result of the fingerprint comparison together with the result of a separate classification ("Automatic Classification") and a claim category specified by the vehicle user and determines whether a contradiction exists.

The invention is not limited to the illustrated embodiments, but includes all variants and modifications that fall within the scope of the appended claims.

Claims (9)

1. Method for automatic licence plate recognition of a vehicle, in particular for toll purposes in road toll systems, with the steps:

   a) provision of a data bank (DB) for assigning (5) possible licence plate reading results (LP_OCR,mem) to vehicle licence plates (LP) and reference fingerprints (FP_ref) assigned to these,

   b) reading of the licence plate (LP) of the vehicle (2) by OCR as current licence plate reading result (LP_OCR), with preceding or subsequent

   c) detection by sensor of features of the vehicle (2) as a current fingerprint (FP) of the vehicle,

   d) determination of a licence plate reading result (LP_{OCR, mem}) that matches the current licence plate reading result (LP_OCR) in the data bank (DB) and all reference fingerprints (FP_ref) assigned to this as candidate fingerprints (FP_{ref cand}),

   e) comparison of the similarity of the current fingerprint (FP) with the candidate fingerprints (FP_{ref, cand}) to determine the closest matching candidate fingerprint, and

   f) output of the vehicle licence number (LP) assigned to the most similar candidate fingerprint (FP_{ref, cand}) as result of the automatic licence plate recognition, wherein, if no matching licence plate reading result (LP_{OCR, mem}) can be determined in step d) or no most similar or even no similar candidate fingerprint (FP_{ref, cand}) can be determined in step e), the vehicle licence number (LP) is manually determined for the current licence plate reading result (LP_OCR) and in these cases the current licence plate reading result (LP_OCR) is assigned to the vehicle licence plate (LP) as new or additional licence plate reading result (LP_{OCR, mem}) in the data bank (DB), characterised in that the new or additional assignment is only performed when the current licence plate reading result (LP_OCR) reaches a minimum frequency and/or minimum quality.
2. Method according to claim 1, characterised in that if the comparison of the current fingerprint (FP) with the reference fingerprint(s) (FP_ref) assigned to the vehicle licence plate (LP) fails and the current fingerprint (FP) has a better quality or more information than the stored reference fingerprint(s) (FP_ref), the current fingerprint (FP) is assigned to the vehicle licence plate (LP) as new or additional reference fingerprint (FP_ref).
3. Method according to claim 1 or 2, characterised in that if no reference fingerprint to the vehicle licence plate (LP) is stored and the current fingerprint (FP) meets minimum quality requirements, this is assigned to the vehicle licence plate (LP) as a new reference fingerprint (FP_ref).
4. Method according to one of claims 1 to 3, characterised in that an assignment (5) is removed from the data bank (DB) if it is not used within a given time period.
5. Method according to one of claims 1 to 4, characterised in that the detection by sensor is performed by means of a camera, with which an image (1) of the licence plate (LP) and/or of the vehicle (2) is taken, and features are taken from said image and the fingerprint (FP) is generated therefrom.
6. Method according to one of claims 1 to 5, characterised in that the detection by sensor is performed by means of photoelectric barriers, laser scanners, induction loops, weight sensors or radio communication elements, with which characteristic features of the vehicle (2) are detected, from which the fingerprint (FP) is generated.
7. Method according to one of claims 1 to 6, characterised in that weightings (WT) for individual regions of the assigned reference fingerprints (FP_ref) are stored in the data bank (DB) for each licence plate reading result (LP_OCR,mem), preferably for each individual assignment thereof to a vehicle licence plate (LP), and said weightings are taken into consideration during similarity comparison.
8. Method according to one of claims 1 to 7, characterised in that the licence plate reading results (LP_OCR,mem) are stored in encrypted form in the data bank (DB) and the current licence plate reading result (LP_OCR) for the match search of step d) is encrypted in the same manner.
9. Method according to one of claims 1 to 8, additionally for toll categorisation of the vehicle, characterised in that in the data bank (DB) a toll category (MK) is also assigned to each reference fingerprint (FP_ref) and, if this reference fingerprint (FP_ref) is determined as the most similar candidate fingerprint (FP_{ref, cand}), the assigned toll category (MK) is output with it.

Images