License plate attribute recognition and locating method and apparatus, and terminal device and storage medium

Abstract

The present application is applicable to the technical field of vehicles. Provided are a license plate attribute recognition and locating method and apparatus, and a terminal device and a storage medium. The license plate attribute recognition and locating method comprises: acquiring a first image feature of a license plate image and a first text feature of text information, wherein the text information comprises at least one task description, and one task description corresponds to one license plate attribute; fusing the first text feature on the basis of the first image feature to obtain a second image feature; fusing the first image feature on the basis of the first text feature to obtain a second text feature; and on the basis of the second image feature and the second text feature, determining a recognition result and a locating result of at least one license plate attribute. By means of the present application, one or more license plate attributes can be recognized and located, and the accuracy of recognition and locating of the license plate attributes are improved.

Description

License plate attribute recognition and positioning method and device, terminal equipment and storage medium TECHNICAL FIELD

[0001] The application belongs to the technical field of vehicles, and particularly relates to a license plate attribute recognition and positioning method and device, terminal equipment and storage medium. BACKGROUND

[0002] In identifying and positioning license plate numbers, license plate countries, license plate cities, license plate regions, license plate colors, vehicle types and other license plate attributes, the prior art usually uses multiple models for identification and positioning. For example, a license plate number recognition model is used to identify the license plate number, a classifier corresponding to each of other license plate attributes such as license plate country, license plate city, license plate region, license plate color and vehicle type is used for identification, a character detection model is used to position the license plate number, and an attribute detection model is used to position the above-mentioned other license plate attributes. That is, a single model in the prior art does not have multiple license plate attribute perception capabilities, cannot identify and position multiple (i.e., at least two) license plate attributes, and has low identification accuracy and positioning accuracy of license plate attributes. TECHNICAL PROBLEM

[0003] Embodiments of the application provide a license plate attribute recognition and positioning method, device, terminal equipment and storage medium to realize identification and positioning of one or more license plate attributes and improve the identification accuracy and positioning accuracy of license plate attributes. TECHNICAL SOLUTION

[0004] In a first aspect, embodiments of the application provide a license plate attribute recognition and positioning method, which includes:

[0005] obtaining a first image feature of a license plate image and a first text feature of text information, the text information including at least one task description, one task description corresponding to one license plate attribute;

[0006] fusing the first text feature on the basis of the first image feature to obtain a second image feature;

[0007] fusing the first image feature on the basis of the first text feature to obtain a second text feature;

[0008] determining an identification result and a positioning result of at least one license plate attribute based on the second image feature and the second text feature.

[0009] In a second aspect, embodiments of the application provide a license plate attribute recognition and positioning device, which includes:

[0010] The feature acquisition module is configured to acquire a first image feature of the license plate image and a first text feature of text information, the text information including at least one task description, and one task description corresponding to one license plate attribute;

[0011] The first fusion module is configured to fuse the first text feature on the basis of the first image feature to obtain a second image feature.

[0012] The second fusion module is configured to fuse the first image feature on the basis of the first text feature to obtain a second text feature.

[0013] The result determination module is configured to determine an identification result and a positioning result of at least one license plate attribute on the basis of the second image feature and the second text feature.

[0014] In a third aspect, an embodiment of the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the license plate attribute recognition and positioning method according to the first aspect.

[0015] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executable on a processor to implement the steps of the license plate attribute recognition and positioning method according to the first aspect.

[0016] In a fifth aspect, an embodiment of the present application provides a computer program product, which, when running on a terminal device, enables the terminal device to perform the steps of the license plate attribute recognition and positioning method according to the first aspect. Advantages

[0017] As can be seen, the scheme can obtain the first image feature of the license plate image and the first text feature of the text information, the text information includes at least one task description, one task description corresponds to one license plate attribute, the first text feature is fused on the basis of the first image feature, the second image feature fused with the text feature can be obtained, the first image feature is fused on the basis of the first text feature, the second text feature fused with the image feature can be obtained, and the recognition result and the positioning result of at least one license plate attribute can be determined based on the second image feature and the second text feature. The task description in the text information can limit the license plate attribute to be recognized and positioned, and the second image feature fused with the text feature and the second text feature fused with the image feature provide richer feature information than the first image feature and the first text feature, so that the recognition and positioning of one license plate attribute or multiple license plate attributes can be realized based on the second image feature and the second text feature, and the recognition accuracy and the positioning accuracy of the license plate attribute can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0019] FIG. 1 is an implementation flowchart of a license plate attribute recognition and positioning method according to an embodiment of the present application;

[0020] FIG. 2 is an implementation flowchart of a license plate attribute recognition and positioning method according to another embodiment of the present application;

[0021] FIG. 3 is a structural example diagram of a bidirectional attention module;

[0022] FIG. 4 is a structural example diagram of a character decoder;

[0023] FIG. 5 is a structural example diagram of a detector;

[0024] FIG. 6 is a structural example diagram of a multi-task model;

[0025] FIG. 7 is a structural schematic diagram of a license plate attribute recognition and positioning device according to an embodiment of the present application;

[0026] FIG. 8 is a structural schematic diagram of a terminal device according to an embodiment of the present application. Embodiments of the present application

[0027] In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular architectures, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known methods, devices, circuits, and

[0028] It is to be understood that the terminology "includes", "has", "holds", "contains" or "comprises", "comprising", or "including" when used in this specification and in the following claims, specifies the presence of stated features, integers, steps, operations, elements, or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or groups thereof.

[0029] It is also to be understood that the terminology "and / or" when used in this specification and in the following claims, refers to at least one of the items, or any combination of the items, or any combination of one or more of the items with one or more of the other items.

[0030] As used in this specification and in the claims, the terms "if" and "when" can be interpreted to mean "upon" or "in response to determining," or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [a described condition or event] is detected" can be interpreted to mean "upon determining" or "in response to determining," or "upon detecting [the described condition or event]" or "in response to detecting [the described condition or event]," depending on the context.

[0031] In addition, the terms "first", "second", "third", etc. in the description of the application and in the claims are used only to distinguish descriptions, and cannot be understood as indicating or implying relative importance.

[0032] Reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the application. Thus, the appearances of the phrases "in one embodiment", "in some embodiments", "in other embodiments", "in additional embodiments", and so on, in various places throughout this specification are not necessarily all referring to the same embodiment, unless otherwise specified. The terms "comprise", "comprising", "has", "having", "includes" and "including" as well as variations thereof, mean "including but not limited to", unless expressly specified otherwise.

[0033] The training process of the model in the prior art is as follows: a license plate data set is constructed based on manually collected license plate data, and license plate numbers are labeled, the license plate numbers are embodied in character form, license plate countries, license plate cities, license plate regions, license plate colors, vehicle types and other license plate attributes are labeled, the other license plate attributes are embodied in classification labels, and the positions of the characters in the license plate numbers and the positions of the other license plate attributes are labeled. Based on the license plate data set and the labeled license plate numbers, a license plate number recognition model is trained; based on the license plate data set and the labeled each other license plate attribute, a classifier is trained for each other license plate attribute, such as a classifier for license plate countries and a classifier for license plate colors. Based on the license plate data set and the labeled positions of the characters and the other license plate attributes, a character detection model and an other license plate attribute detection model are trained.

[0034] The prior art has the following disadvantages:

[0035] 1. License plate number recognition, other license plate attribute classification, license plate character detection, other license plate attribute detection, etc. require training multiple models, and a single model does not have all license plate attribute perception capabilities.

[0036] 2. Multiple tasks are independent of each other, and the model cannot be trained uniformly, and training and optimizing the model is time-consuming and labor-intensive.

[0037] 3. License plate attributes in different countries and different regions differ, and license plate data sets differ, and different regions may have multiple different attribute recognition and positioning models.

[0038] 4. The model is limited by the number of license plate data in different regions, and the performance of the model trained based on the license plate data in a region with a small amount of license plate data is poor.

[0039] 5. Multiple license plate attribute recognition and positioning require multiple models, and multiple models in series require complex logic design to adapt to different regional license plates, and cannot directly input data to obtain recognition results and positioning results.

[0040] The embodiments of the present application can use a multi-modal algorithm of text and image to uniformly implement license plate number recognition, other license plate attribute classification, license plate character detection, other license plate attribute detection, etc. using a multi-task model, that is, to uniformly implement recognition and positioning of a license plate attribute or multiple license plate attributes, and a multi-task model has all license plate attribute perception capabilities.

[0041] The embodiments of the present application can train a multi-task model using a large amount of license plate data (i.e. license plate samples, license plate images for training the model), and uniformly train the model, which is efficient.

[0042] The embodiment of the present application can improve the performance of the recognition algorithm and the positioning algorithm by using the multi-task model structure, and solve the problem of dependence of license plate data in each region. All license plate data in different regions can be used to train the multi-task model, and the multi-task model can correctly recognize and position the license plate data even if the number of license plate data in a certain region is small.

[0043] The embodiment of the present application uses a multi-task model, no longer needs a long pipeline, directly outputs end to end, does not need complex logic to adapt to license plates in different regions, and a single model can adapt to all license plate data.

[0044] It should be noted that the multi-task model described above can refer to program code or software for implementing license plate attribute recognition and positioning in the embodiment of the present application.

[0045] The license plate attribute recognition and positioning method provided by the embodiment of the present application can be applied to electronic devices such as mobile phones, tablet computers, wearable devices, vehicle-mounted devices, augmented reality (AR) / virtual reality (VR) devices, notebook computers, ultra-mobile personal computers (UMPCs), netbooks, personal digital assistants (PDAs), and the like. The embodiment of the present application does not make any limitation on the specific type of electronic device.

[0046] It should be understood that the size of the serial number of each step in the embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0047] In order to illustrate the technical solutions described in the present application, the following will be described by specific embodiments.

[0048] Referring to FIG. 1, it is an implementation flowchart of a license plate attribute recognition and positioning method provided by an embodiment of the present application, and the license plate attribute recognition and positioning method is applied to a terminal device. As shown in FIG. 1, the license plate attribute recognition and positioning method can include the following steps:

[0049] Step 101, obtaining a first image feature of a license plate image and a first text feature of text information.

[0050] The text information includes at least one task description, and one task description corresponds to one license plate attribute. If the text information includes one task description, the license plate attribute corresponding to the task description can be recognized and positioned by the application. If the text information includes multiple task descriptions, the license plate attributes corresponding to the multiple task descriptions can be recognized and positioned by the application, that is, multiple license plate attributes can be recognized and positioned.

[0051] For example, the text information includes five task descriptions: output license plate number, output license plate country, output license plate city, output license plate color, and output license plate region. The license plate attribute corresponding to the task description "output license plate number" is the license plate number, and the corresponding task is to recognize and position the license plate number. The license plate attribute corresponding to the task description "output license plate country" is the license plate country, and the corresponding task is to recognize and position the license plate country. The license plate attribute corresponding to the task description "output license plate city" is the license plate city, and the corresponding task is to recognize and position the license plate city. The license plate attribute corresponding to the task description "output license plate color" is the license plate color, and the corresponding task is to recognize and position the license plate color. The license plate attribute corresponding to the task description "output license plate region" is the license plate region, and the corresponding task is to recognize and position the license plate region. The five license plate attributes: license plate number, license plate country, license plate city, license plate color, and license plate region can be recognized and positioned by the application.

[0052] The license plate image can be any image that needs to be recognized and positioned for license plate attributes. The license plate image can be collected by a camera on a terminal device, stored on the terminal device, or sent to the terminal device by another device (such as a vehicle-mounted camera), and the application does not make any limitation.

[0053] In an optional embodiment, the license plate image can be input into an image feature extractor, and the image feature extractor outputs a first image feature, so as to obtain the first image feature of the license plate image. The license plate image can be input into a text feature extractor, and the text feature extractor outputs a first text feature, so as to obtain the first text feature of the license plate image.

[0054] The network structure of the image feature extractor can be any of Vit, Swintransformer, Resnet, etc. Of course, the network structure of the image feature extractor can also be other structures, which are not limited in the present application. The network structure of the text feature extractor can be BERT (Bidirectional Encoder Representations from Transformers), which is a pre-trained natural language processing model. Of course, the network structure of the text feature extractor can also be other structures, which are not limited in the present application.

[0055] Step 102, fusing the first text feature on the basis of the first image feature to obtain a second image feature.

[0056] The terminal device fuses the first text feature on the basis of the first image feature to obtain a second image feature that has fused the text feature, thereby realizing multi-modal feature fusion. Multi-modal feature fusion can reduce the heterogeneity difference between text and image, while maintaining the integrity of specific semantics of each modality. Therefore, when recognizing and positioning the license plate attribute based on the second image feature, the recognition accuracy and positioning accuracy of the license plate attribute can be improved.

[0057] Step 103, fusing the first image feature on the basis of the first text feature to obtain a second text feature.

[0058] The terminal device fuses the first image feature on the basis of the first text feature to obtain a second text feature that has fused the image feature, thereby realizing multi-modal feature fusion. Multi-modal feature fusion can reduce the heterogeneity difference between text and image, while maintaining the integrity of specific semantics of each modality. Therefore, when recognizing and positioning the license plate attribute based on the second text feature, the recognition accuracy and positioning accuracy of the license plate attribute can be improved.

[0059] Step 104, determining the recognition result and positioning result of at least one license plate attribute based on the second image feature and the second text feature.

[0060] Optionally, the recognition result of the license plate attribute can be represented in text form. For example, when the license plate attribute is license plate number and license plate country, the recognition result of the license plate number can be "License plate number: SSS-xxxx", and the recognition result of the license plate country can be "License plate country: North America". Of course, the recognition result of the license plate attribute can also be represented in other forms, which are not limited in the present application.

[0061] The positioning result of the license plate attribute can refer to position information of the license plate attribute in the license plate image, and specifically can refer to position information of a bounding box of the license plate attribute (for example, coordinates of the upper left corner and the lower right corner of the bounding box), where the bounding box is a bounding box of the license plate attribute. When the license plate attribute is a license plate number, the position information of the license plate number can refer to position information of each character in the license plate number.

[0062] Since the second image feature and the second text feature are both subjected to multi-modal feature fusion, richer feature information is provided compared to the first image feature and the first text feature, so that based on the second image feature and the second text feature, not only the recognition and positioning of one license plate attribute or multiple license plate attributes can be realized, but also the recognition accuracy and positioning accuracy of the license plate attribute can be improved.

[0063] The task description in the text information can limit the license plate attribute to be recognized and positioned, and the second image feature fused with the text feature and the second text feature fused with the image feature provide richer feature information compared to the first image feature and the first text feature, so that based on the second image feature and the second text feature, not only the recognition and positioning of one license plate attribute or multiple license plate attributes can be realized, but also the recognition accuracy and positioning accuracy of the license plate attribute can be improved.

[0064] Referring to FIG. 2, it is an implementation flowchart of a license plate attribute recognition and positioning method provided by another embodiment of the present application, and the license plate attribute recognition and positioning method is applied to a terminal device. As shown in FIG. 2, the license plate attribute recognition and positioning method can include the following steps:

[0065] Step 201: obtaining a first image feature of a license plate image and a first text feature of text information.

[0066] This step is the same as step 101, and for details, please refer to the related description of step 101, which will not be repeated here.

[0067] Step 202: inputting the first image feature into a third self-attention module to obtain a third image feature.

[0068] The third self-attention module realizes the function of attention. The input dimension and the output dimension of the third self-attention module are the same, and the input and the output are both image features. The output (i.e., the third image feature) fuses image attention features compared to the input (i.e., the first image feature), so that the terminal device can pay attention to the correlation between different parts of the first image feature, and the performance of the multi-task model is improved.

[0069] Step 203: inputting the first text feature into a fourth self-attention module to obtain a third text feature.

[0070] The fourth self-attention module realizes the function of attention. The input dimension and the output dimension of the fourth self-attention module are the same, and the input and the output are both text features. The output (i.e., the third text feature) fuses the text attention feature compared with the input (i.e., the first text feature), so that the terminal device pays attention to the correlation between different parts in the first text feature, and the performance of the multi-task model is improved.

[0071] In step 204, the third image feature and the third text feature are input into a text-to-image attention module to obtain a fourth image feature.

[0072] The text-to-image attention module is used to fuse the text feature into the image feature. The fourth image feature can be abstractly understood as a high-dimensional feature that fuses the text feature.

[0073] After the third image feature and the third text feature are input into the text-to-image attention module, the feature Q1 can be obtained based on the third text feature by using a convolution operation through the text-to-image attention module, and the feature K1 and the feature V1 can be obtained based on the third image feature by using a convolution operation. The feature Q1 and the feature K1 are multiplied to obtain a first similarity matrix. The first similarity matrix is multiplied by the feature V1 to obtain an output (i.e., the fourth image feature). The input dimension and the output dimension of the text-to-image attention module are the same, and the output fuses the text feature compared with the input. The third image feature input does not contain the text feature, and the fourth image feature output fuses the text feature through the text-to-image attention module.

[0074] In step 205, the fourth image feature is input into a sixth neural network to obtain a second image feature.

[0075] The sixth neural network can be a feedforward neural network (FNN), which essentially is a two-layer fully connected layer. The fully connected layer can further extract features and enhance the expression ability of the features. Of course, the sixth neural network can also be other neural networks, which are not limited in the present application.

[0076] In step 206, the third image feature and the third text feature are input into an image-to-text attention module to obtain a fourth text feature.

[0077] The image-to-text attention module is used to fuse the image feature into the text feature. The fourth text feature can be abstractly understood as a high-dimensional feature that fuses the image feature.

[0078] After the third image feature and the third text feature are input into the image-to-text attention module, the feature Q2 can be obtained based on the third image feature by using a convolution operation, the feature K2 and the feature V2 can be obtained based on the third text feature by using a convolution operation, the feature Q2 is multiplied by the feature K2 to obtain a second similarity matrix, and the output (i.e., a fourth text feature) is obtained by multiplying the second similarity matrix by the feature V2. The input dimension and the output dimension of the image-to-text attention module are the same, and the output fuses the image feature compared with the input. The third text feature input does not contain the image feature, and the fourth text feature output fuses the image feature through the image-to-text attention module.

[0079] In step 207, the fourth text feature is input into a seventh neural network to obtain a second text feature.

[0080] The seventh neural network can be a FNN, which is essentially a two-layer fully connected layer. The fully connected layer can further extract features and enhance the expression ability of the features. Of course, the seventh neural network can also be other neural networks, which are not limited in the present application.

[0081] In an embodiment, the third self-attention module, the fourth self-attention module, the text-to-image attention module, the image-to-text attention module, the sixth neural network and the seventh neural network can constitute a bidirectional attention module. The first image feature and the first text feature are input into the bidirectional attention module, and the multi-modal feature fusion can be performed to obtain the second image feature and the second text feature. As shown in FIG. 3, it is a structural example diagram of the bidirectional attention module.

[0082] In step 208, the second image feature, the second text feature and the first start-stop symbol feature are input into a character decoder to obtain a recognition result of at least one license plate attribute.

[0083] The character decoder can be a language model (LM) for generating a recognition result in an autoregressive manner. The first start-stop symbol feature in step 208 is an initial value of the character decoder, which can be set in advance. The first start-stop symbol feature and the output are cyclically generated or updated.

[0084] For example, in the initial stage: the first start-stop symbol feature is a start-stop symbol [start], and the output is the first character: car; in the next stage: the first start-stop symbol feature is a start-stop symbol [start car], and the output is the first character: plate; in the next stage: the first start-stop symbol feature is a start-stop symbol [start car plate], and the output is the first character: number; in the next stage: the first start-stop symbol feature is a start-stop symbol [start car plate number], and the output is the first character: code; in the next stage: the first start-stop symbol feature is a start-stop symbol [start car plate number code], and the output is the first character: “:”; in the next stage: the first start-stop symbol feature is a start-stop symbol [start car plate number code:], and the output is the first character: S; the cycle continues; until the next stage: the first start-stop symbol feature is a start-stop symbol [start car plate number code: SSS-xxxx], and the output is the first character: [end]; indicating the end of generation (i.e., the end of the cycle).

[0085] The second image feature can provide image feature information for the character decoder, and the second text feature can provide text feature information for the character decoder, so that the character decoder can implement license plate attribute recognition based on multi-modal features.

[0086] In an optional embodiment, as shown in FIG. 4, the character decoder includes a first self-attention module, a first image cross-attention module, a first text cross-attention module, a first neural network, and a second neural network; and the identification process based on the character decoder is as follows:

[0087] The first start-stop symbol feature is input into the first self-attention module to obtain a second start-stop symbol feature;

[0088] The second start-stop symbol feature and the second image feature are input into the first image cross-attention module to obtain a third start-stop symbol feature;

[0089] The third start-stop symbol feature and the second text feature are input into the first text cross-attention module to obtain a fourth start-stop symbol feature;

[0090] The fourth start-stop symbol feature is input into the first neural network to obtain a fifth start-stop symbol feature, and the fifth start-stop symbol feature is taken as the first start-stop symbol feature to return to execute the step of inputting the first start-stop symbol feature into the first self-attention module and subsequent steps until the number of returned executions reaches N times, N being an integer greater than zero;

[0091] The fifth start-stop symbol feature is input into the second neural network to obtain an identification result.

[0092] The first self-attention module realizes the function of attention. The input dimension and the output dimension of the first self-attention module are the same, and the input and the output are both start and end symbol features. The output (i.e., the second start and end symbol feature) fuses the attention feature of the start and end symbol compared with the input (i.e., the first start and end symbol feature).

[0093] The first image cross-attention module can be a Transformer structure. The feature Q3 can be obtained based on the second start and end symbol feature by using the convolution feature. The features K3 and V3 can be obtained based on the second image feature by using the convolution operation. The third similarity matrix is obtained by multiplying the feature Q3 and the feature K3. The output (i.e., the third start and end symbol feature) is obtained by multiplying the third similarity matrix and the feature V3. The third start and end symbol feature of the output fuses the image feature.

[0094] The first text cross-attention module can be a Transformer structure. The feature Q4 can be obtained based on the third start and end symbol feature by using the convolution feature. The features K4 and V4 can be obtained based on the second text feature by using the convolution operation. The fourth similarity matrix is obtained by multiplying the feature Q4 and the feature K4. The output (i.e., the fourth start and end symbol feature) is obtained by multiplying the fourth similarity matrix and the feature V4. The fourth start and end symbol feature of the output fuses the text feature on the basis of the third start and end symbol feature, that is, the fourth start and end symbol feature fuses the image feature and the text feature.

[0095] The character decoder fuses the image feature and the text feature in the process of autoregressive decoding by using the first image cross-attention module and the first text cross-attention module, so as to realize the multi-modal feature fusion.

[0096] The first neural network can be a FNN, which is essentially a two-layer fully connected layer. The fully connected layer can further extract features and enhance the expression ability of the features. Of course, the first neural network can also be other neural networks, which are not limited in the application.

[0097] The second neural network can be a FNN, which is essentially a two-layer fully connected layer. The fully connected layer can output the recognition result of the license plate attribute.

[0098] The embodiment can make the start and end symbol feature better fuse the image feature and the text feature and enhance the expression ability of the features by performing N+1 times of multi-modal feature fusion and N+1 times of feature enhancement on the start and end symbol feature, so as to improve the recognition accuracy of the license plate attribute. Optionally, only one time of multi-modal feature fusion and one time of feature enhancement can be performed on the start and end symbol feature, that is, the step of returning to perform the multi-modal feature fusion and the feature enhancement is not performed.

[0099] Step 209, input the second image feature, the second text feature and the first target query feature into the detector to obtain a positioning result of at least one license plate attribute.

[0100] The detector is a detection decoding module, which can fuse image features and text features in a decoding process to realize multi-modal feature fusion. Inputting the text feature into the detector can make the target query feature understand the input text information, which is used to guide the detection content (i.e., positioning which license plate attribute or several license plate attributes). Inputting the image feature into the detector can make the target query feature understand the input image information, which is used to guide the generation of the positioning result.

[0101] The second image feature can provide image feature information for the detector, and the second text feature can provide text feature information for the detector, so that the detector can realize license plate attribute positioning based on multi-modal features.

[0102] The target query feature is used to query and position the license plate attribute in the target detection task. The first target query feature in step 209 is composed of a group of pre-defined vectors, each of which represents a potential license plate attribute.

[0103] In an optional embodiment, as shown in FIG. 5, the detector includes a second self-attention module, a second image cross-attention module, a second text cross-attention module, a third neural network and a fourth neural network; and the positioning process of the detector is as follows:

[0104] Input the first target query feature into the second self-attention module to obtain a second target query feature;

[0105] Input the second target query feature and the second image feature into the second image cross-attention module to obtain a third target query feature;

[0106] Input the third target query feature and the second text feature into the second text cross-attention module to obtain a fourth target query feature;

[0107] Input the fourth target query feature into the third neural network to obtain a fifth target query feature, and take the fifth target query feature as the first target query feature to return to execute the step of inputting the first target query feature into the second self-attention module and the subsequent steps until the number of returned executions reaches M times, M being an integer greater than zero;

[0108] Input the fifth target query feature into the fourth neural network to obtain the positioning result.

[0109] The second self-attention module realizes the function of attention. The input dimension and the output dimension of the first self-attention module are the same, and the input and the output are both target query features. The output (i.e., the second target query feature) fuses the attention features of the target query compared with the input (i.e., the first target query feature).

[0110] The second image cross-attention module can be a Transformer structure. The feature Q5 can be obtained based on the second target query feature by using convolution features, and the features K5 and V5 can be obtained based on the second image feature by using convolution operations. The fifth similarity matrix is obtained by multiplying the feature Q5 and the feature K5. The output (i.e., the third target query feature) is obtained by multiplying the fifth similarity matrix and the feature V5. The third target query feature of the output fuses the image features.

[0111] The second text cross-attention module can be a Transformer structure. The feature Q6 can be obtained based on the third target query feature by using convolution features, and the features K6 and V6 can be obtained based on the second text feature by using convolution operations. The sixth similarity matrix is obtained by multiplying the feature Q6 and the feature K6. The output (i.e., the fourth target query feature) is obtained by multiplying the sixth similarity matrix and the feature V6. The fourth target query feature of the output fuses the text features based on the third target query feature, that is, the fourth target query feature fuses the image features and the text features.

[0112] The third neural network can be an FNN, which is essentially two fully connected layers. The fully connected layers can further extract features and enhance the expression ability of the features. Of course, the third neural network can also be other neural networks, which are not limited in the present application.

[0113] The fourth neural network can be an FNN, which is essentially two fully connected layers. The fully connected layers can output the positioning result of the license plate attribute.

[0114] The target query feature can better fuse the image features and the text features, and the expression ability of the features can be enhanced by performing M+1 times of multi-modal feature fusion and M+1 times of feature enhancement on the target query feature, so as to improve the positioning accuracy of the license plate attribute. Optionally, only one time of multi-modal feature fusion and one time of feature enhancement can be performed on the target query feature, that is, the steps of performing multi-modal feature fusion and feature enhancement are not returned.

[0115] It should be noted that M and N can be the same or different, which are not limited in the present application. For example, M and N are both 6.

[0116] In an optional embodiment, the detector further includes a fifth neural network, and after obtaining the fifth target query feature, the detector further includes:

[0117] The fifth target query feature is input into the fifth neural network to obtain a category of the license plate attribute.

[0118] The fifth neural network can be a FNN, which essentially is a two-layer fully connected layer, and through the fully connected layer, the category of the license plate attribute can be output. By identifying the category of the license plate attribute, the prediction boxes of the multiple license plate attributes can be distinguished when the multiple license plate attributes are positioned, so that the user can know which license plate attribute each prediction box corresponds to.

[0119] As shown in FIG. 5, the detector includes a second self-attention module, a second image cross-attention module, a third text cross-attention module, a third neural network, a fourth neural network and a fifth neural network, and through the detector, the positioning result and the category of the license plate attribute can be output.

[0120] As shown in FIG. 6 is a structural example diagram of a multi-task model, which includes an image feature extractor, a text feature extractor, a bidirectional attention module, a character decoder and a detector, and by inputting the license plate image and the text information into the multi-task model, the recognition and positioning of at least one license plate attribute in the license plate image can be realized. It should be noted that for the license plate attribute whose position cannot be positioned, the positioning result is empty or the positioning result is not output. For example, in the license plate image in FIG. 6, there is no region where the license plate attribute such as license plate color, license plate country, license plate color, etc. is located, i.e., the position of these license plate attributes cannot be positioned in the license plate image, so in the positioning result shown in FIG. 6, the prediction boxes of these license plate attributes are not displayed.

[0121] Before the multi-task model is used to recognize and position the license plate attribute, the character decoder and the detector in the multi-task model need to be trained, and the training process is as follows:

[0122] (1) At least one license plate sample and a text sample "license plate number" are input into the multi-task model to obtain the first predicted recognition result of each license plate sample output by the character decoder and the first predicted positioning result and the first predicted category of each license plate sample output by the detector; based on the first predicted recognition result and the first true recognition result of the at least one license plate sample, a first loss function of the character decoder is calculated; based on the first predicted positioning result and the first true positioning result of the at least one license plate sample, a first positioning loss function of the detector is calculated; based on the first predicted category and the first true category of the at least one license plate sample, a first classification loss function of the detector is calculated; the weighted sum of the first loss function of the character decoder, the first positioning loss function and the first classification loss function of the detector is calculated to obtain a first total loss function; based on the first total loss function, the parameters of the multi-task model are adjusted to complete the training of the multi-task model, so that the trained multi-task model can realize the recognition, positioning and classification of the license plate number.

[0123] (2) input at least one license plate sample and text sample "license plate country" into the multi-task model to obtain a second predicted recognition result of each license plate sample output by the character decoder and a second predicted positioning result and a second predicted category of each license plate sample output by the detector; based on the second predicted recognition result and the second real recognition result of at least one license plate sample, calculate the second loss function of the character decoder; based on the second predicted positioning result and the second real positioning result of at least one license plate sample, calculate the second classification loss function of the detector; based on the second predicted category and the second real category of at least one license plate sample, calculate the second classification loss function of the detector; calculate the weighted sum of the second loss function of the character decoder, the second positioning loss function and the second classification loss function of the detector, and the second total loss function can be obtained; based on the second total loss function, adjust the parameters of the multi-task model, complete the training of the multi-task model, so that the trained multi-task model can realize the recognition, positioning and classification of the license plate country.

[0124] (3) input at least one license plate sample and text sample "license plate city" into the multi-task model to obtain a third predicted recognition result of each license plate sample output by the character decoder and a third predicted positioning result and a third predicted category of each license plate sample output by the detector; based on the third predicted recognition result and the third real recognition result of at least one license plate sample, calculate the third loss function of the character decoder; based on the third predicted positioning result and the third real positioning result of at least one license plate sample, calculate the third positioning loss function of the detector; based on the third predicted category and the third real category of at least one license plate sample, calculate the third classification loss function of the detector; calculate the weighted sum of the third loss function of the character decoder, the third positioning loss function and the third classification loss function of the detector, and the third total loss function can be obtained; based on the third total loss function, adjust the parameters of the multi-task model, complete the training of the multi-task model, so that the trained multi-task model can realize the recognition, positioning and classification of the license plate city.

[0125] (4) input at least one license plate sample and text sample "license plate color" into the multi-task model to obtain a fourth predicted recognition result of each license plate sample output by the character decoder and a fourth predicted positioning result and a fourth predicted category of each license plate sample output by the detector; calculate a fourth loss function of the character decoder based on the fourth predicted recognition result and the fourth real recognition result of at least one license plate sample; calculate a fourth positioning loss function of the detector based on the fourth predicted positioning result and the fourth real positioning result of at least one license plate sample; calculate a fourth classification loss function of the detector based on the fourth predicted category and the fourth real category of at least one license plate sample; calculate a weighted sum of the fourth loss function of the character decoder, the fourth positioning loss function and the fourth classification loss function of the detector to obtain a fourth total loss function; adjust the parameters of the multi-task model based on the fourth total loss function to complete the training of the multi-task model, so that the trained multi-task model can realize the recognition, positioning and classification of the license plate color.

[0126] (5) input at least one license plate sample and text sample "license plate area" into the multi-task model to obtain a fifth predicted recognition result of each license plate sample output by the character decoder and a fifth predicted positioning result and a fifth predicted category of each license plate sample output by the detector; calculate a fifth loss function of the character decoder based on the fifth predicted recognition result and the fifth real recognition result of at least one license plate sample; calculate a fifth positioning loss function of the detector based on the fifth predicted positioning result and the fifth real positioning result of at least one license plate sample; calculate a fifth classification loss function of the detector based on the fifth predicted category and the fifth real category of at least one license plate sample; calculate a weighted sum of the fifth loss function of the character decoder, the fifth loss function and the fifth classification loss function of the detector to obtain a fifth total loss function; adjust the parameters of the multi-task model based on the fifth total loss function to complete the training of the multi-task model, so that the trained multi-task model can realize the recognition, positioning and classification of the license plate area.

[0127] Optionally, the types of the first loss function, the second loss function, the third loss function, the fourth loss function and the fifth loss function of the character decoder can be cross-entropy loss functions. Of course, they can also be other types of loss functions, which are not limited in the present application. The types of the first positioning loss function, the second positioning loss function, the third positioning loss function, the fourth positioning loss function and the fifth positioning loss function of the detector can be Focal loss functions, which are used to train the positioning function of the detector, so as to realize the positioning of the license plate attribute. Of course, the positioning loss function can also be other types of loss functions, which are not limited in the present application. The types of the first classification loss function, the second classification loss function, the third classification loss function, the fourth classification loss function and the fifth classification loss function of the detector can be L1 loss functions and Giou loss functions, which are used to train the classification function of the detector, so as to identify the category of the license plate attribute. Of course, the classification loss function can also be other types of loss functions, which are not limited in the present application.

[0128] The present embodiment proposes a novel license plate attribute recognition and positioning scheme based on multi-modal and multi-task. The multi-task model is a brand-new network structure, and a unified network framework solves the license plate perception problem. By using the character decoder and the detector, combined with the text information, a multi-task loss function can be constructed, so that the multi-task model can learn different tasks by itself, and realize multi-task license plate attribute recognition and positioning. Since the text information and the perception task are combined, all labeled data can be used to train the multi-task model, thereby solving the problem that license plate data cannot be shared across regions, and being suitable for license plate attribute recognition and positioning in different regions and different countries.

[0129] By inputting the first image feature and the first text feature into the bidirectional attention module, the second image feature and the second text feature can be obtained. By inputting the second image feature, the second text feature and the first start-stop symbol feature into the character decoder, the recognition result of at least one license plate attribute can be obtained. By inputting the second image feature, the second text feature and the first target query feature into the detector, the positioning result of at least one license plate attribute can be obtained, so as to realize the recognition and positioning of one license plate attribute or multiple license plate attributes, and improve the recognition accuracy and positioning accuracy of the license plate attribute.

[0130] Referring to FIG. 7, it is a structural schematic diagram of a license plate attribute recognition and positioning device provided by an embodiment of the present application. For the convenience of description, only the parts related to the embodiments of the present application are shown.

[0131] The license plate attribute recognition and positioning device includes:

[0132] The feature acquisition module 71 is configured to acquire a first image feature of the license plate image and a first text feature of text information, the text information including at least one task description, and one task description corresponding to one license plate attribute.

[0133] The first fusion module 72 is configured to fuse the first text feature on the basis of the first image feature to obtain a second image feature.

[0134] The second fusion module 73 is configured to fuse the first image feature on the basis of the first text feature to obtain a second text feature.

[0135] The result determination module 74 is configured to determine a recognition result and a positioning result of the at least one license plate attribute on the basis of the second image feature and the second text feature.

[0136] Optionally, the result determination module 74 includes:

[0137] The attribute recognition unit is configured to input the second image feature, the second text feature and the first start-stop symbol feature into a character decoder to obtain the recognition result.

[0138] Optionally, the character decoder includes a first self-attention module, a first image cross-attention module, a first text cross-attention module, a first neural network and a second neural network; and the attribute recognition unit is specifically configured to:

[0139] input the first start-stop symbol feature into the first self-attention module to obtain a second start-stop symbol feature;

[0140] input the second start-stop symbol feature and the second image feature into the first image cross-attention module to obtain a third start-stop symbol feature;

[0141] input the third start-stop symbol feature and the second text feature into the first text cross-attention module to obtain a fourth start-stop symbol feature;

[0142] input the fourth start-stop symbol feature into the first neural network to obtain a fifth start-stop symbol feature, and take the fifth start-stop symbol feature as the first start-stop symbol feature to return to execute the step of inputting the first start-stop symbol feature into the first self-attention module and subsequent steps until the number of times of returning to execute reaches N times, N being an integer greater than zero;

[0143] input the fifth start-stop symbol feature into the second neural network to obtain the recognition result.

[0144] Optionally, the result determination module 74 includes:

[0145] The attribute positioning unit is configured to input the second image feature, the second text feature and a first target query feature into a detector to obtain the positioning result.

[0146] Optionally, the detector comprises a second self-attention module, a second image cross-attention module, a third text cross-attention module, a third neural network and a fourth neural network; and the attribute positioning unit is specifically configured to:

[0147] input the first target query feature into the second self-attention module to obtain a second target query feature;

[0148] input the second target query feature and the second image feature into the second image cross-attention module to obtain a third target query feature;

[0149] input the third target query feature and the second text feature into the second text cross-attention module to obtain a fourth target query feature;

[0150] input the fourth target query feature into the third neural network to obtain a fifth target query feature, and take the fifth target query feature as the first target query feature to return to execute the step of inputting the first target query feature into the second self-attention module and subsequent steps until the number of times of returning to execute reaches M times, where M is an integer greater than zero;

[0151] input the fifth target query feature into the fourth neural network to obtain the positioning result.

[0152] Optionally, the detector further comprises a fifth neural network; and the attribute positioning unit is further configured to:

[0153] input the fifth target query feature into the fifth neural network to obtain the category of the license plate attribute.

[0154] Optionally, the license plate attribute recognition and positioning apparatus further comprises:

[0155] an image extraction module configured to input the first image feature into the third self-attention module to obtain a third image feature;

[0156] a text extraction module configured to input the first text feature into the fourth self-attention module to obtain a third text feature;

[0157] the first fusion module 72 is specifically configured to:

[0158] input the third image feature and the third text feature into a text-to-image attention module to obtain a fourth image feature;

[0159] input the fourth image feature into the sixth neural network to obtain the second image feature;

[0160] the second fusion module 72 is specifically configured to:

[0161] input the third image feature and the third text feature into an image-to-text attention module to obtain a fourth text feature;

[0162] The fourth text feature is input into the seventh neural network to obtain a second text feature.

[0163] The license plate attribute recognition and positioning device provided by the embodiments of the present application can be applied in the method embodiments described above, and details are described in the description of the method embodiments, which will not be described here.

[0164] Referring to FIG. 8, it is a structural schematic diagram of a terminal device according to an embodiment of the present application. As shown in FIG. 8, the terminal device 8 of the embodiment includes one or more processors 80 (only one processor is shown in the figure), a memory 81, and a computer program 82 stored in the memory 81 and executable on the processor 80. The processor 80 implements the steps in the above-mentioned various license plate attribute recognition and positioning method embodiments when executing the computer program 82.

[0165] The terminal device can include, but is not limited to, the processor 80 and the memory 81. Those skilled in the art can understand that FIG. 8 is only an example of the terminal device 8 and does not constitute a limitation on the terminal device 8, which can include more or fewer components than those shown, or combine certain components, or different components, for example, the terminal device can also include an input / output device, a network access device, a bus, etc.

[0166] The processor 80 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0167] The memory 81 can be an internal storage unit of the terminal device 8, for example, a hard disk or a memory of the terminal device 8. The memory 81 can also be an external storage device of the terminal device 8, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device 8. Further, the memory 81 can also include both the internal storage unit and the external storage device of the terminal device 8. The memory 81 is used to store the computer program and other programs and data required by the terminal device. The memory 81 can also be used to temporarily store data that has been output or will be output.

[0168] It can be clearly understood by those skilled in the art that, for the convenience and brevity of description, only the division of the above functional units and modules is exemplified, and in actual application, the above functions can be completed by different functional units and modules according to needs, that is, the internal structure of the apparatus is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit, and the integrated unit can be realized in the form of hardware or in the form of software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction, and do not limit the protection scope of the present application. The specific working process of the units and modules in the apparatus can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0169] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps in each of the above method embodiments can be implemented.

[0170] The embodiment of the present application further provides a computer program product, which, when running on a terminal device, enables the terminal device to execute the steps in each of the above method embodiments.

[0171] In the above embodiments, the description of each embodiment has its own emphasis, and the parts not described or recorded in detail in a certain embodiment can be referred to the related description of other embodiments.

[0172] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0173] In the embodiments provided by the present application, it should be understood that the disclosed apparatus / terminal device and method can be implemented in other ways. For example, the apparatus / terminal device embodiments described above are merely schematic, and the division of the modules or units is merely a logical function division, and there can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection between interfaces, devices or units can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0174] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e. can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0175] The integrated module / unit, if implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned various method embodiments can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0176] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for license plate attribute recognition and localization, the method comprising: The license plate attribute recognition and positioning method comprises: obtaining first image features of a license plate image and first text features of text information, the text information comprising at least one task description, one task description corresponding to one license plate attribute; fusing the first text features on the basis of the first image features to obtain second image features; fusing the first image features on the basis of the first text features to obtain second text features; determining at least one recognition result and positioning result of the license plate attribute based on the second image features and the second text features.

2. The method of claim 1, wherein, Determining at least one recognition result of the license plate attribute based on the second image features and the second text features comprises: inputting the second image features, the second text features and first start-stop symbol features into a character decoder to obtain the recognition result.

3. The method of claim 2, wherein, The character decoder comprises a first self-attention module, a first image cross-attention module, a first text cross-attention module, a first neural network and a second neural network; inputting the second image features, the second text features and first start-stop symbol features into the character decoder to obtain the recognition result comprises: inputting the first start-stop symbol features into the first self-attention module to obtain second start-stop symbol features; inputting the second start-stop symbol features and the second image features into the first image cross-attention module to obtain third start-stop symbol features; inputting the third start-stop symbol features and the second text features into the first text cross-attention module to obtain fourth start-stop symbol features; inputting the fourth start-stop symbol features into the first neural network to obtain fifth start-stop symbol features, and taking the fifth start-stop symbol features as the first start-stop symbol features to return to execute the step of inputting the first start-stop symbol features into the first self-attention module and subsequent steps until the number of returned executions reaches N times, N being an integer greater than zero; inputting the fifth start-stop symbol features into the second neural network to obtain the recognition result.

4. The method of claim 1 to 3, wherein, Determining at least one positioning result of the license plate attribute based on the second image features and the second text features comprises: inputting the second image features, the second text features and first target query features into a detector to obtain the positioning result.

5. The method of claim 4, wherein, The detector comprises a second self-attention module, a second image cross-attention module, a third text cross-attention module, a third neural network and a fourth neural network; inputting the second image features, the second text features and first target query features into the detector to obtain the positioning result comprises: inputting the first target query features into the second self-attention module to obtain second target query features; inputting the second target query features and the second image features into the second image cross-attention module to obtain third target query features; inputting the third target query features and the second text features into the second text cross-attention module to obtain fourth target query features; inputting the fourth target query feature into the third neural network to obtain a fifth target query feature, taking the fifth target query feature as the first target query feature, and returning to execute the step of inputting the first target query feature into the second self-attention module and subsequent steps until the number of times of returning to execute reaches M times, where M is an integer greater than zero; inputting the fifth target query feature into the fourth neural network to obtain the positioning result.

6. The method of claim 5, wherein, The detector further includes a fifth neural network; After obtaining the fifth target query feature, the method further includes: inputting the fifth target query feature into the fifth neural network to obtain the category of the license plate attribute.

7. The method of claim 1 to 3, wherein, After obtaining the first image feature of the license plate image and the first text feature of the text information, the method further includes: inputting the first image feature into a third self-attention module to obtain a third image feature; inputting the first text feature into a fourth self-attention module to obtain a third text feature; The step of fusing the first text feature on the basis of the first image feature to obtain a second image feature includes: inputting the third image feature and the third text feature into a text-to-image attention module to obtain a fourth image feature; inputting the fourth image feature into a sixth neural network to obtain the second image feature; The step of fusing the first image feature on the basis of the first text feature to obtain a second text feature includes: inputting the third image feature and the third text feature into an image-to-text attention module to obtain the fourth text feature; inputting the fourth text feature into a seventh neural network to obtain the second text feature.

8. A license plate attribute recognition and positioning apparatus, characterized by, The license plate attribute recognition and positioning apparatus includes: a feature acquisition module configured to acquire a first image feature of a license plate image and a first text feature of text information, the text information including at least one task description, one task description corresponding to one license plate attribute; a first fusion module configured to fuse the first text feature on the basis of the first image feature to obtain a second image feature; a second fusion module configured to fuse the first image feature on the basis of the first text feature to obtain a second text feature; a result determination module configured to determine, based on the second image feature and the second text feature, at least one recognition result and positioning result of the license plate attribute.

9. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the license plate attribute recognition and positioning method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the steps of the license plate attribute recognition and positioning method according to any one of claims 1 to 7.

Images