License plate correction method, device, system and readable storage medium
Through the license plate correction method, the vehicle picture features and parking behavior model are used to fuzzy matching and correction of license plate numbers, solving the accuracy of license plate recognition under occlusion, tilt and extreme weather conditions, and improving the accuracy and stability of license plate recognition .
Patent Information
- Application Number
- CN202111292352.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-11-03
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2041-11-03
AI Technical Summary
The existing license plate recognition technology reduces the recognition accuracy when the license plate is blocked or tilted, and has high requirements for image quality, so it cannot effectively deal with the blurred problem of license plates under extreme weather conditions.
By obtaining the correction data of the camera terminal equipment, the vehicle picture characteristics and parking behavior are extracted, and the comprehensive similarity calculation is performed using the pre-established vehicle picture characteristic model and parking behavior model to perform fuzzy matching and correction of license plate numbers, combined with DBSCAN algorithm and deep learning algorithm establish and update models, handle the replacement and filtering of license plate numbers, and improve identification accuracy.
When the license plate characters are slightly obstructed or tilted, the license plate can be correctly identified, which significantly improves the accuracy of license plate recognition and detects identification errors early, which is suitable for assisting existing license plate recognition technology.
Smart Images

Figure CN114092927B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of smart parking, and in particular to a license plate correction method, device, system and readable storage medium. Background Art
[0002] Smart parking is an important part of smart city construction. Making full use of on-street parking on urban roads is an important part of realizing smart parking. High-position video is an important solution to solve on-street parking charges and monitoring. Therefore, improving the accuracy of license plate recognition of high-position video is a very urgent and pressing issue. In the actual environment, high-position video faces many challenges, which are specifically manifested in: license plates will be blocked by various factors, such as pedestrians, bicycles, electric vehicles and other electric vehicles; compared with barriers, license plate features of high-position video are more diverse, and license plates will tilt at various angles, increasing the difficulty of character recognition. For example, 0 and D, 6 and G are highly similar due to tilt; extreme weather causes license plates to be blurred, such as snow, heavy rain, fog, etc.
[0003] There are two main solutions for license plate recognition: one is based on traditional image processing technology, which segments and classifies license plate characters to finally identify the license plate; the other is a recognition solution based on deep learning, where the license plate image is directly output after passing through the machine's "neural network". However, these solutions have many defects mentioned above, such as the inability to correctly recognize the results when the license plate characters are partially blocked, the decline in recognition accuracy due to severely tilted license plates, the high image quality requirements, and the inability to perceive license plate recognition errors. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a license plate correction method, device, system and readable storage medium in view of the above-mentioned defects of the prior art.
[0005] The technical solution adopted by the present invention to solve the technical problem is:
[0006] On the one hand, a license plate correction method is constructed, the method comprising:
[0007] Acquire calibration data of the camera terminal device, and extract vehicle image features, parking behavior, and the license plate number to be calibrated from the calibration data, wherein the parking behavior includes the time of entry or exit event and the parking space;
[0008] Input the license plate number to be calibrated into the pre-established license plate number model to obtain the license plate number on fuzzy matching;
[0009] Traversing the license plate numbers on the fuzzy match to perform a first comprehensive similarity calculation, including: inputting each license plate number on the fuzzy match into a pre-established vehicle image feature model and a parking behavior model, and performing a first comprehensive similarity calculation based on the fuzzy match information, the vehicle image features and parking behaviors output by the vehicle image feature model and the parking behavior model, and the vehicle image features and parking behaviors extracted from the calibration request;
[0010] If the highest value of the calculated first comprehensive similarity is greater than the preset similarity value, the license plate number corresponding to the highest value is selected for output.
[0011] Preferably, the method further comprises:
[0012] If the license plate number to be corrected fails to be extracted from the correction data, determining whether the parking behavior extracted from the correction data includes an entry event or an exit event;
[0013] If it is an entry event, saving the correction data;
[0014] If it is an exit event, then find the license plate number of the most recent entry event on the parking space in the correction data that is not matched with an exit event, input the found license plate number into the vehicle image feature model and the parking behavior model, and perform a second comprehensive similarity calculation based on the vehicle image features and parking behaviors output by the vehicle image feature model and the parking behavior model, as well as the vehicle image features and parking behaviors extracted from the calibration request. If the calculated second comprehensive similarity value is greater than the preset similarity value, output the found license plate number.
[0015] Preferably, the method further comprises: using completed normal parking orders to pre-establish or update the parking behavior model, vehicle image feature model, and license plate number model;
[0016] The parking behavior model includes the license plate number and the time of entry and exit events corresponding to each license plate number and the parking space;
[0017] The vehicle image feature model includes a license plate number and a vehicle image feature corresponding to each license plate number, wherein the vehicle image feature includes a vehicle color feature, a vehicle type classification feature of at least one dimension, and a vehicle brand feature;
[0018] The license plate number model is used to obtain a license plate number that can be fuzzily matched with the input license plate number.
[0019] Preferably, the establishment or updating of the parking behavior model includes:
[0020] Get the normal parking orders that have been completed in the recent period, and extract the orders of the license plate numbers whose order times are greater than the preset times;
[0021] For the extracted orders, the DBSCAN algorithm is used to extract the clustered time periods of exit and entry events for the orders with the same license plate number;
[0022] For orders in the same clustering time period, the DBSCAN algorithm is used to extract clustered berths;
[0023] The center point of the gathering time period of the exit event and the entrance event, the center point of the gathering berth, and the license plate number are associated and saved.
[0024] Preferably, the establishment of the vehicle image feature model includes: extracting vehicle images of the same license plate number in different orders, extracting the vehicle image features through a deep learning algorithm, and saving the vehicle image features and license plate information.
[0025] Preferably, the vehicle type classification features of the at least one dimension include:
[0026] Classification features based on the compartment dimension: single compartment, hatchback, hatchback and a half, and sedan.
[0027] Classification features based on the door dimension: two doors, four doors.
[0028] Classification features based on roof dimensions: panoramic sunroof, small sunroof, no sunroof.
[0029] Classification features of other classification dimensions: pickup trucks, sedans, SUVs, vans, sports cars, and small trucks.
[0030] Preferably, the license plate number model comprises:
[0031] Eliminate the province abbreviation of the license plate number to be corrected to obtain the target license plate number;
[0032] Replace the characters 0 and D, 6 and G in the target license plate number to obtain all possible replacement license plate numbers;
[0033] The target license plate number and all its replacement license plate numbers are output as license plate numbers on fuzzy matching.
[0034] Preferably, the method further comprises: after obtaining the license plate number on the fuzzy match using the license plate number model, first filtering the license plate number on the fuzzy match, and then performing the traversal on the filtered license plate number;
[0035] The fuzzy matching license plate numbers are first filtered, including: calculating the matching degree between each fuzzy matching license plate number and the license plate number to be calibrated based on the calculation formula L=1-N / M, and selecting the license plate number with a matching degree greater than the matching threshold, where L represents the matching degree, M represents the number of letters and numbers in the license plate number to be calibrated, and N represents the standard Levins distance between the fuzzy matching license plate number and the license plate number to be calibrated;
[0036] Among them, the calculation formula for the first comprehensive similarity is S1=K0*L+K1*L1+K2*L2, S1 represents the first comprehensive similarity, L1 represents the similarity between the vehicle image features output by the vehicle image feature model and the vehicle image features extracted from the calibration request, L2 represents the similarity between the parking behavior output by the parking behavior model and the parking behavior extracted from the calibration request, and K0, K1, and K2 are coefficients.
[0037] Preferably, the calculation formula of the second comprehensive similarity is S2=K3*L1+K4*L2, S2 represents the second comprehensive similarity, and K3 and K4 are coefficients.
[0038] In a second aspect, a license plate correction device is constructed, characterized in that it includes a processor and a memory, the memory stores a computer program, and when the computer program is executed by the processor, the steps of the method described above are implemented.
[0039] In a third aspect, a readable storage medium is constructed, characterized in that a computer program is stored therein, and when the computer program is executed by a processor, the steps of the method described above are implemented.
[0040] Fourthly, a license plate correction system is constructed, characterized in that the system includes a camera terminal device and a service gateway. After the camera terminal device recognizes a parking behavior and recognizes the license plate number, the license plate correction interface is called. After the license plate correction interface is called, the data is passed to the service gateway and the corrected license plate number is returned by the service gateway. The service gateway is used to execute the steps of the method as described above.
[0041] The license plate correction method, device, system and readable storage medium of the present invention have the following beneficial effects: the present invention performs license plate correction based on image and parking behavior modeling, and can correctly identify the license plate when the characters are slightly obscured or when the license plate is severely tilted. It can be used to assist existing license plate recognition technology and significantly improve its accuracy. It can be used for anomaly detection, and when license plate recognition errors occur, they can be discovered early and manual intervention can be performed in advance. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below. Obviously, the drawings in the following description are only embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on the provided drawings without creative work:
[0043] Figure 1 It is a schematic diagram of an application scenario of the present invention;
[0044] Figure 2 It is a flow chart of a license plate correction method in a specific embodiment of the present invention. DETAILED DESCRIPTION
[0045] In order to facilitate understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Typical embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.
[0046] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.
[0047] Terms including ordinal numbers such as "first" and "second" used in this specification may be used to describe various components, but these components are not limited by these terms. The purpose of using these terms is only to distinguish one component from other components. For example, without departing from the scope of the present invention, the first component may be named as the second component, and similarly, the second component may also be named as the first component.
[0048] The general idea of the present invention is that the main users of urban on-street parking are local urban users, and the life characteristics of local residents have certain regularity. For example, most people's daily life pattern is to go from residence to work in the morning and from work to residence in the evening. Therefore, it is an effective method to desensitize and model these data features for auxiliary license plate recognition. For this purpose, reference Figure 1 The present invention proposes to use order data in advance, clean, convert, desensitize, establish or update the data model and then store the data. After the camera terminal device recognizes a parking behavior and recognizes the license plate number, it can call the license plate correction interface through the Internet to correct the license plate.
[0049] In order to better understand the above-mentioned technical scheme, the above-mentioned technical scheme will be described in detail below in conjunction with the accompanying drawings and specific implementation methods. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical scheme of the present application, rather than limitations on the technical scheme of the present application. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.
[0050] Embodiment 1
[0051] This embodiment provides a license plate correction method, which includes steps S1-S10. Before executing step S1, it is necessary to use the completed normal parking orders to pre-establish or update the parking behavior model, vehicle image feature model, and license plate number model.
[0052] The following are the three models:
[0053] A) Parking behavior model:
[0054] The parking behavior model includes the license plate number and the time of entry and exit events corresponding to each license plate number and the parking space. The establishment or update of the parking behavior model includes:
[0055] a1: Get the normal parking orders completed in the recent period, and extract the orders of the license plate numbers whose order times are greater than the preset times;
[0056] For example, extract normal parking orders that have been completed in the last month, and extract parking orders for license plates with more than 20 order times.
[0057] a2: For the extracted orders, use the DBSCAN algorithm to extract the clustered time periods where exit and entry events occur for the orders with the same license plate number;
[0058] Among them, the DBSCAN algorithm is a relatively well-known algorithm. When analyzing the clustering of time, for entry events, the distance is the difference between two entry times, and for exit events, the distance is the difference between two exit times, that is:
[0059] Entry event: Dist (distance) = entry time 1 - entry time 2;
[0060] Exit event: Dist (distance) = exit time 1 - exit time 2;
[0061] It is understandable that the clustering time period may be the one that is ranked first during cluster analysis, or may be several of the ones that are ranked first.
[0062] a3: For the clustering time period analyzed in the previous step, use the DBSCAN algorithm to extract clustering berths for orders in the same clustering time period;
[0063] When analyzing the concentration of berths, the distance is defined as:
[0064] Dist (distance) = ABS (berth number 1 - berth number 2) / total number of berths;
[0065] Among them, ABS() means taking the absolute value.
[0066] a4: Associate the center point of the gathering time period of the exit event and the entrance event, the center point of the gathering parking space, and the license plate number and save them.
[0067] The center point of the aggregation time period of the exit event is saved as the exit time of the license plate number, the center point of the aggregation time period of the entry event is saved as the entry time of the license plate number, and the center point of the aggregation parking space is saved as the parking space of the license plate number.
[0068] B) Vehicle image feature model:
[0069] The vehicle image feature model includes a license plate number and a vehicle image feature corresponding to each license plate number, wherein the vehicle image feature includes a vehicle color feature, a vehicle type classification feature of at least one dimension, and a vehicle brand feature.
[0070] The establishment of the vehicle image feature model includes:
[0071] b1: Extract vehicle images of the same license plate number in different orders;
[0072] b2: Extracting the color features of the vehicle through a deep learning algorithm, such as whether the color is black, white, gray, etc.
[0073] b3: Extracting the vehicle type classification feature of the at least one dimension through a deep learning algorithm, wherein the vehicle type classification feature of the at least one dimension includes:
[0074] Classification features based on the compartment dimension: single compartment, hatchback, hatchback and a half, and sedan.
[0075] Classification features based on the door dimension: two doors, four doors.
[0076] Classification features based on roof dimensions: panoramic sunroof, small sunroof, no sunroof.
[0077] Classification features of other classification dimensions: pickup trucks, sedans, SUVs, vans, sports cars, and small trucks.
[0078] b4: extracting the vehicle brand features through a deep learning algorithm;
[0079] b5: The vehicle color feature, at least one dimension of vehicle type classification feature, vehicle brand feature and license plate information are associated and saved.
[0080] C) License plate number model:
[0081] The license plate number model is used to obtain a license plate number that can be fuzzily matched with the input license plate number, and the license plate number model includes:
[0082] c1: Eliminate the province abbreviation of the license plate number to be corrected to obtain the target license plate number;
[0083] c2: Replace the characters 0 and D, 6 and G in the target license plate number to obtain all possible replacement license plate numbers;
[0084] For example, if the target license plate number is B01058, the replacement license plate numbers are: BD1058, B01D58, BD1D58. c3: Output the target license plate number and all its replacement license plate numbers as the license plate numbers on the fuzzy match.
[0085] After establishing or updating the parking behavior model, vehicle image feature model, and license plate number model, refer to Figure 1 After the camera terminal device recognizes a parking behavior and the license plate number, it will provide correction data. We can use this correction data and the pre-established model to perform license plate correction. It should be noted that the execution subject of the method of this embodiment is not limited, and it can be any hardware that can run a program to implement the correction method, such as the camera terminal device itself, a remote service gateway, or a specific correction device.
[0086] refer to Figure 2 , the correction method of this embodiment includes:
[0087] S1: Acquire calibration data of the camera terminal device, and extract vehicle image features, parking behavior, and the license plate number to be calibrated from the calibration data, wherein the parking behavior includes the time of entry or exit events and the parking space;
[0088] For example, the license plate number to be calibrated is A01058, and the vehicle image features are: color is black, hatchback, four doors, panoramic sunroof, SUV, Audi. The parking behavior is: exit, 10:00, parking space 0008.
[0089] S2: Determine whether the license plate number to be calibrated is successfully extracted, if the license plate number to be calibrated is successfully extracted, go to step S3, if the license plate number to be calibrated is not extracted from the calibration data, go to step S6;
[0090] Because the camera terminal device may not have successfully recognized the license plate number before, the license plate number cannot be calibrated directly during calibration, and the license plate number needs to be determined first.
[0091] S3: input the license plate number to be calibrated into the pre-established license plate number model to obtain the license plate number on fuzzy matching;
[0092] For example, the license plate numbers A01058 mentioned above, obtained by fuzzy matching using the license plate number model, are AD1058, A01D58, and AD1D58.
[0093] S4: Traversing the license plate numbers on the fuzzy match to perform a first comprehensive similarity calculation, including: inputting each license plate number on the fuzzy match into a pre-established vehicle image feature model and a parking behavior model, and performing a first comprehensive similarity calculation based on the fuzzy matching information, the vehicle image features and parking behaviors output by the vehicle image feature model and the parking behavior model, and the vehicle image features and parking behaviors extracted from the calibration request.
[0094] Preferably, before traversal, the license plate numbers on the fuzzy match are filtered first, specifically: the matching degree between each license plate number on the fuzzy match and the license plate number to be calibrated is calculated based on the calculation formula L=1-N / M, and the license plate number with a matching degree greater than the matching threshold is selected, where L represents the matching degree, M represents the number of letters and numbers in the license plate number to be calibrated, and N represents the standard Levins distance between the license plate number on the fuzzy match and the license plate number to be calibrated.
[0095] Here, the standard Levins distance refers to the number of steps required to change from the license plate number to be calibrated to the license plate number on the fuzzy match. For example, in the previous example, A01058 is the license plate number to be calibrated, and the change to AD1058 is to change the second 0 to D, and the number of steps required is 1, that is, the standard Levins distance is 1; for another example, suppose that a license plate number to be calibrated is blocked, and only A0105 is left, then fill in the missing characters, and the distance is 1.
[0096] Among them, the calculation formula for the first comprehensive similarity is S1=K0*L+K1*L1+K2*L2, S1 represents the first comprehensive similarity, L1 represents the similarity between the vehicle image features output by the vehicle image feature model and the vehicle image features extracted from the calibration request, L2 represents the similarity between the parking behavior output by the parking behavior model and the parking behavior extracted from the calibration request, K0, K1, and K2 are coefficients, and the sum of K0, K1, and K2 is 1. Considering that the various factors corresponding to K0, K1, and K2 objectively contribute differently to the final similarity, it is recommended that K0 is greater than K1, and K1 is greater than K2 (that is, the image similarity is better than the parking behavior similarity), for example, K0 is 0.7, K1 is 0.2, and K2 is 0.1.
[0097] The calculation method of vehicle similarity and parking behavior similarity is not specifically limited here, and different methods can be used according to actual scenarios in specific applications.
[0098] S5. If the highest value of the calculated first comprehensive similarity is greater than the preset similarity value, the license plate number corresponding to the highest value is selected for output, and the method ends.
[0099] The preset similarity value may be an empirical value, such as 0.85. Assuming that the largest S1 calculated in S4 comes from AD1058 and is greater than 0.85, AD1058 is output.
[0100] It should also be noted that if the license plate number to be calibrated and its replacement license plate number do not exist in the model, it means that this is the first time that the car has parked in the monitoring area. At this time, it can be directly determined according to the license plate number recognized by the camera terminal device, or an abnormal work order can be generated for manual intervention and return of the abnormal order.
[0101] S6: determining whether the parking behavior extracted from the correction data includes an entry event or an exit event;
[0102] S7: If it is an entry event, then the correction data is saved and the method ends;
[0103] S8: If it is an exit event, searching for the license plate number of the most recent entry event that is not matched to an exit event on the parking space of the correction data;
[0104] S9: Input the found license plate number into the vehicle picture feature model and the parking behavior model, and perform a second comprehensive similarity calculation based on the vehicle picture features and parking behaviors output by the vehicle picture feature model and the parking behavior model, and the vehicle picture features and parking behaviors extracted from the calibration request.
[0105] Specifically, the calculation formula for the second comprehensive similarity is S2=K3*L1+K4*L2, where S2 represents the second comprehensive similarity, K3 and K4 are coefficients, and the sum of K3 and K4 is 1. Considering that the various factors corresponding to K3 and K4 objectively contribute differently to the final similarity, it is recommended that K3 be greater than K2, for example, K3 is 0.6 and K4 is 0.4.
[0106] In addition, it should be noted that if the license plate number cannot be found in the entry event, an abnormal work order can be generated for manual intervention and an abnormal order can be returned.
[0107] S10. If the calculated second comprehensive similarity value is greater than the preset similarity value, the found license plate number is outputted, and the method ends.
[0108] Embodiment 2
[0109] This embodiment discloses a license plate correction device, including a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the steps of the method described in the first embodiment are implemented. The specific implementation process can be found in the description of the above method embodiment, which will not be repeated here.
[0110] Embodiment 3
[0111] This embodiment discloses a readable storage medium storing a computer program. When the computer program is executed by a processor, the steps of the method described in the first embodiment are implemented. The specific implementation process can be found in the description of the above method embodiment, which will not be repeated here.
[0112] Embodiment 4
[0113] The present embodiment discloses a license plate correction system, which includes a camera terminal device and a service gateway. After the camera terminal device identifies a parking behavior and the license plate number, the license plate correction interface is called. After the license plate correction interface is called, the data is passed to the service gateway and the corrected license plate number is returned by the service gateway. The service gateway is used to execute the steps of the method described in Example 1. The specific implementation process can be found in the description of the above method embodiment, which will not be repeated here.
[0114] In summary, the license plate correction method, device, system and readable storage medium of the present invention have the following beneficial effects: the present invention performs license plate correction based on image and parking behavior modeling, and can correctly identify the license plate when the characters are slightly obscured or the license plate is severely tilted. It can be used to assist existing license plate recognition technology and significantly improve its accuracy. It can be used for anomaly detection, and when license plate recognition errors occur, they can be discovered early and manual intervention can be performed in advance.
[0115] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.
Claims
1. A license plate correction method, characterized in that: The method comprises: Acquire calibration data of the camera terminal device, and extract vehicle image features, parking behavior, and the license plate number to be calibrated from the calibration data, wherein the parking behavior includes the time of entry or exit event and the parking space; Input the license plate number to be calibrated into the pre-established license plate number model to obtain the license plate number on fuzzy matching; Traversing the license plate numbers on the fuzzy match to perform a first comprehensive similarity calculation, including: inputting each license plate number on the fuzzy match into a pre-established vehicle image feature model and a parking behavior model, and performing a first comprehensive similarity calculation based on the fuzzy match information, the vehicle image features and parking behaviors output by the vehicle image feature model and the parking behavior model, and the vehicle image features and parking behaviors extracted from the calibration request; If the highest value of the calculated first comprehensive similarity is greater than the preset similarity value, the license plate number corresponding to the highest value is selected for output.
2. The license plate correction method according to claim 1, characterized in that: The method further comprises: If the license plate number to be corrected fails to be extracted from the correction data, determining whether the parking behavior extracted from the correction data includes an entry event or an exit event; If it is an entry event, saving the correction data; If it is an exit event, then find the license plate number of the most recent entry event on the parking space in the correction data that is not matched with an exit event, input the found license plate number into the vehicle image feature model and the parking behavior model, and perform a second comprehensive similarity calculation based on the vehicle image features and parking behaviors output by the vehicle image feature model and the parking behavior model, as well as the vehicle image features and parking behaviors extracted from the calibration request; if the calculated second comprehensive similarity value is greater than the preset similarity value, output the found license plate number.
3. The license plate correction method according to claim 1, characterized in that: The method further includes: using the completed normal parking order to pre-establish or update the parking behavior model, the vehicle image feature model, and the license plate number model; The parking behavior model includes the license plate number and the time of entry and exit events corresponding to each license plate number and the parking space; The vehicle image feature model includes a license plate number and a vehicle image feature corresponding to each license plate number, wherein the vehicle image feature includes a vehicle color feature, a vehicle type classification feature of at least one dimension, and a vehicle brand feature; The license plate number model is used to obtain a license plate number that can be fuzzily matched with the input license plate number.
4. The license plate correction method according to claim 3, characterized in that: The establishment or update of the parking behavior model includes: Get the normal parking orders that have been completed in the recent period, and extract the orders of the license plate numbers whose order times are greater than the preset times; For the extracted orders, the DBSCAN algorithm is used to extract the clustered time periods of exit and entry events for the orders with the same license plate number; For orders in the same clustering time period, the DBSCAN algorithm is used to extract clustered berths; The center point of the gathering time period of the exit event and the entrance event, the center point of the gathering berth, and the license plate number are associated and saved.
5. The license plate correction method according to claim 3, characterized in that: The establishment of the vehicle image feature model includes: extracting vehicle images of the same license plate number in different orders, extracting the vehicle image features through a deep learning algorithm, and saving the vehicle image features and license plate information; The vehicle type classification features of the at least one dimension include: Classification features based on the compartment dimension: single compartment, hatchback, half hatchback, and sedan; Classification features based on the door dimension: two doors, four doors; Classification features based on roof dimensions: panoramic sunroof, small sunroof, no sunroof; Classification features of other classification dimensions: pickup trucks, sedans, SUVs, vans, sports cars, and small trucks.
6. The license plate correction method according to claim 3, characterized in that: The license plate number model includes: Eliminate the province abbreviation of the license plate number to be corrected to obtain the target license plate number; Replace the characters 0 and D, 6 and G in the target license plate number to obtain all possible replacement license plate numbers; The target license plate number and all its replacement license plate numbers are output as license plate numbers on fuzzy matching.
7. The license plate correction method according to claim 2, characterized in that: The method further comprises: after obtaining the fuzzy matched license plate number by using the license plate number model, filtering the fuzzy matched license plate number first, and then performing the traversal on the filtered license plate number; The fuzzy matching license plate numbers are first filtered, including: calculating the matching degree between each fuzzy matching license plate number and the license plate number to be calibrated based on the calculation formula L=1-N / M, and selecting the license plate number with a matching degree greater than the matching threshold, where L represents the matching degree, M represents the number of letters and numbers in the license plate number to be calibrated, and N represents the standard Levins distance between the fuzzy matching license plate number and the license plate number to be calibrated; The calculation formula of the first comprehensive similarity is S1=K0*L+K1*L1+K2*L2, S1 represents the first comprehensive similarity, L1 represents the similarity between the vehicle image feature output by the vehicle image feature model and the vehicle image feature extracted from the calibration request, L2 represents the similarity between the parking behavior output by the parking behavior model and the parking behavior extracted from the calibration request, and K0, K1, and K2 are coefficients; Among them, the calculation formula for the second comprehensive similarity is S2=K3*L1+K4*L2, S2 represents the second comprehensive similarity, L1 represents the similarity between the vehicle image features output by the vehicle image feature model and the vehicle image features extracted from the calibration request, L2 represents the similarity between the parking behavior output by the parking behavior model and the parking behavior extracted from the calibration request, and K3 and K4 are coefficients.
8. A license plate correction device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 7 are implemented.
9. A readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.
10. A license plate correction system, characterized in that: The system includes a camera terminal device and a service gateway. After the camera terminal device identifies a parking behavior and recognizes the license plate number, it calls a license plate correction interface. After the license plate correction interface is called, it passes data to the service gateway and returns the corrected license plate number fed back by the service gateway. The service gateway is used to execute the license plate correction method as described in any one of claims 1-7.
Citation Information
Patent Citations
Intelligent parking lot license plate identification method
CN106251637A
Training modeling method, vehicle charging method, management system and storage medium
CN110598758A
