CN114782938A - License plate character recognition method and device - Google Patents
License plate character recognition method and device Download PDFInfo
- Publication number
- CN114782938A CN114782938A CN202210291417.2A CN202210291417A CN114782938A CN 114782938 A CN114782938 A CN 114782938A CN 202210291417 A CN202210291417 A CN 202210291417A CN 114782938 A CN114782938 A CN 114782938A
- Authority
- CN
- China
- Prior art keywords
- character
- license plate
- characters
- data
- images
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Character Discrimination (AREA)
Abstract
The application discloses a license plate character recognition method and device. The method comprises the steps of obtaining a plurality of character license plate images, wherein each character is marked with a real character type and a real character position, obtaining the number of each real character type, screening the plurality of characters, and forming a plurality of data enhanced copy characters and a plurality of non-data enhanced copy characters, wherein the number of the character types of the data enhanced copy characters is smaller than that of the character types of the non-data enhanced copy characters; replacing at least part of the non-data enhanced copy characters with data enhanced copy characters to obtain a plurality of extended character license plate images, and performing data enhancement processing on the plurality of extended character license plate images and the plurality of character license plate images to obtain a plurality of converted character license plate images; performing model training on the character target detection model according to the plurality of converted character license plate images to obtain a trained character target detection model; and inputting the license plate image of the character to be detected into the trained character target detection model for detection.
Description
Technical Field
The present application relates to the field of image processing technologies, and in particular, to a license plate character recognition method and apparatus.
Background
The license plate recognition technology has an important role in a plurality of tasks such as urban traffic management, vehicle recognition, parking lot charge management, violation processing and the like. The license plate recognition technology comprises the steps of detecting the position of a license plate and recognizing characters of the license plate. And detecting the position of the license plate, and positioning by shooting an image to obtain the position of a license plate area. The license plate characters can be recognized from the detected license plate characters in the license plate area.
However, the license plate recognition technology is still a challenging task due to the influence of a plurality of factors, such as illumination conditions, weather conditions, image definition, shooting angles of license plates, and colors of license plates. The traditional license plate recognition method is used for recognizing through a fully-connected network or a recurrent neural network, so that the dependency on the license plate data volume is large, and further, massive manually marked license plate data is needed for model training and optimization. Therefore, the traditional license plate recognition method is based on a large amount of artificially marked license plate data for recognition, huge labor and resource costs are required to be consumed, the data types in the data samples are easily unbalanced, and the license plate recognition effect is poor.
Content of application
The method aims to solve the technical problem that the license plate recognition effect of the traditional license plate recognition method is poor. In order to achieve the above purpose, the present application provides a license plate character recognition method and device.
The application provides a license plate character recognition method, which comprises the following steps:
acquiring a plurality of character license plate images, wherein each character in each character license plate image is marked with a real character type and a real character position, and acquiring the number of each real character type;
screening the characters according to the number of each real character category to form a plurality of data enhanced copy characters and a plurality of non-data enhanced copy characters, wherein the number of the character categories corresponding to the data enhanced copy characters is smaller than the number of the character categories corresponding to the non-data enhanced copy characters;
replacing at least part of the non-data enhanced copy characters in each character license plate image with the data enhanced copy characters to obtain a plurality of extended character license plate images, and performing data enhancement processing on the plurality of extended character license plate images and the plurality of character license plate images to obtain a plurality of converted character license plate images;
performing model training on a character target detection model according to the plurality of converted character license plate images to obtain a trained character target detection model;
and inputting the license plate image of the character to be detected into the trained character target detection model, and obtaining the character category and the character position of each character in the license plate image of the character to be detected.
In one embodiment, the replacing at least a part of the non-data enhanced copied characters in each character license plate image with the data enhanced copied characters to obtain a plurality of extended character license plate images, and performing data enhancement processing on the plurality of extended character license plate images and the plurality of character license plate images to obtain a plurality of converted character license plate images includes:
acquiring the image size of at least part of the non-data enhanced copy characters in the character license plate image, and adjusting the image size of the data enhanced copy characters according to the image size of the non-data enhanced copy characters to obtain the data enhanced copy characters with the adjusted size;
replacing at least part of the non-data enhanced copy characters with the data enhanced copy characters after size adjustment to obtain a plurality of expansion character license plate images;
and marking the data enhanced copy character with a corresponding real character type and a real character position after the size is adjusted.
In one embodiment, the replacing at least a part of the non-data enhanced copied characters in each character license plate image with the data enhanced copied characters to obtain a plurality of extended character license plate images, and performing data enhancement processing on the plurality of extended character license plate images and the plurality of character license plate images to obtain a plurality of converted character license plate images further includes:
rotating the plurality of expanded character license plate images and the plurality of character license plate images to obtain a plurality of rotated character license plate images;
performing brightness conversion on the plurality of rotary character license plate images to obtain a plurality of brightness conversion character license plate images;
cutting the license plate images with the brightness conversion characters to obtain license plate images with the conversion characters;
each character in each converted character license plate image is marked with a real character type and a real character position.
In one embodiment, the performing model training on the character target detection model according to the plurality of converted character license plate images to obtain a trained character target detection model includes:
inputting each converted character license plate image into the character target detection model, and outputting the predicted character category and the predicted character position of each character in each converted character license plate image;
constructing a classification loss function according to the predicted character category and the real character category;
constructing a position regression loss function according to the predicted character position and the real character position;
and training and optimizing the character target detection model according to the classification loss function and the position regression loss function to obtain the trained character target detection model.
In one embodiment, after the inputting each transformed character license plate image into the character target detection model and outputting the predicted character category and the predicted character position of each character in each transformed character license plate image, before the constructing the classification loss function according to the predicted character category and the real character category, the method further includes:
obtaining the confidence of a detection frame corresponding to each predicted character position in each converted character license plate image;
and sequencing the detection frames corresponding to the predicted character positions in each converted character license plate image according to the confidence, and outputting the predicted character positions and the predicted character types corresponding to high confidence according to the preset license plate character number.
In one embodiment, at least a part of the non-data enhanced copied characters in each character license plate image are replaced by the data enhanced copied characters, and a plurality of extended character license plate images are obtained, wherein the number of the data enhanced copied characters in each extended character license plate image is 1/8-4/7 of the total number of characters.
In one embodiment, the present application provides a license plate character recognition apparatus, comprising:
the data acquisition module is used for acquiring a plurality of character license plate images, wherein each character in each character license plate image is marked with a real character type and a real character position, and the number of each real character type is acquired;
the data enhanced copy character acquisition module is used for screening the characters according to the number of each real character type to form a plurality of data enhanced copy characters and a plurality of non-data enhanced copy characters, wherein the number of the character types corresponding to the data enhanced copy characters is smaller than the number of the character types corresponding to the non-data enhanced copy characters;
the converted character license plate image acquisition module is used for replacing at least part of the non-data enhanced copy characters in each character license plate image with the data enhanced copy characters to obtain a plurality of extended character license plate images, and performing data enhancement processing on the plurality of extended character license plate images and the plurality of character license plate images to obtain a plurality of converted character license plate images;
the character target detection model acquisition module is used for carrying out model training on the character target detection model according to the plurality of converted character license plate images to obtain a trained character target detection model;
and the detection module is used for inputting the license plate image of the character to be detected into the trained character target detection model to obtain the character category and the character position of each character in the license plate image of the character to be detected.
In one embodiment, the transformed character license plate image obtaining module comprises:
the size adjusting module is used for obtaining the image size of at least part of the non-data enhanced copy characters in the character license plate image, adjusting the image size of the data enhanced copy characters according to the image size of the non-data enhanced copy characters, and obtaining the data enhanced copy characters after size adjustment;
the license plate image acquisition module is used for replacing at least part of the non-data enhanced copy characters with the data enhanced copy characters after size adjustment to obtain a plurality of license plate images with the extended characters;
and marking the data enhanced copy character with a corresponding real character type and a real character position after the size is adjusted.
In one embodiment, the transformed character license plate image obtaining module further comprises:
the rotation module is used for rotating the plurality of expanded character license plate images and the plurality of character license plate images to obtain a plurality of rotation character license plate images;
the brightness conversion module is used for performing brightness conversion on the plurality of rotary character license plate images to obtain a plurality of brightness conversion character license plate images;
the cutting module is used for cutting the license plate images with the brightness conversion characters to obtain the license plate images with the conversion characters;
each character in the converted character license plate image is marked with a real character type and a real character position.
In one embodiment, the character target detection model obtaining module includes:
the predictive data acquisition module is used for inputting each converted character license plate image into the character target detection model and outputting the predictive character category and the predictive character position of each character in each converted character license plate image;
the classification loss function building module is used for building a classification loss function according to the predicted character category and the real character category;
the position regression loss function building module is used for building a position regression loss function according to the predicted character position and the real character position;
and the training optimization module is used for training and optimizing the character target detection model according to the classification loss function and the position regression loss function to obtain the trained character target detection model.
In one embodiment, the apparatus further comprises:
the confidence coefficient acquisition module is used for acquiring the confidence coefficient of a detection frame corresponding to each predicted character position in each converted character license plate image;
and the output module is used for sequencing the detection frames corresponding to the predicted character positions in each converted character license plate image according to the confidence degrees and outputting the predicted character positions corresponding to high confidence degrees and predicted character categories according to the preset number plate character digit.
In one embodiment, in the transformed character license plate image obtaining module, the number of the data-enhanced copied characters in each of the expanded character license plate images is 1/8 to 4/7 times the total number of characters.
In the license plate character recognition method and the license plate character recognition device, the data enhanced copy characters and the non-data enhanced copy characters are formed by screening a plurality of characters. The number of character categories corresponding to the data-enhanced copy characters is smaller than the number of character categories corresponding to the non-data-enhanced copy characters. Screening and comparing according to the number of different characters, replacing the non-data enhanced copy characters with the data enhanced copy characters, adjusting the number proportion of different characters in the newly generated extended character license plate image, and carrying out balance adjustment on the number proportion of different characters. The data-enhanced copied characters and the non-data-enhanced copied characters are from the real character license plate image, so that the expanded character license plate image formed after replacement still comes from the conversion of the real character license plate image, and the problem that excessive noise is introduced due to the fact that an anti-network is generated for recognition in the traditional method is solved.
Furthermore, the plurality of extended character license plate images and the plurality of character license plate images are fused together, and the fused extended character license plate images and the plurality of character license plate images are subjected to data enhancement processing and then serve as a training data set of the character target detection model, so that the training data are more real and the number of different characters is balanced. Therefore, in the training data set of the character target detection model, the size difference among a plurality of single characters is small, and the proportion of different license plate character types is balanced after data enhancement, so that the target detection task difficulty is low, the detection precision is high, a complex network structure is not required to be designed, the model training of the character target detection model is facilitated, and the character target detection model which is more stable and optimized is obtained. According to the trained character target detection model, different types of license plates such as two-layer freight cars, motorcycle license plates, military license plates, harbor and Australian area license plates or embassy license plates can be processed, and the character category and the character position of each character in the license plate image are obtained. And sequencing the characters according to the character types and the character positions, and outputting a complete license plate.
Drawings
Fig. 1 is a schematic flow chart illustrating steps of a license plate character recognition method provided by the present application.
Fig. 2 is a schematic structural diagram of a license plate character recognition device provided in the present application.
Detailed Description
The technical solution of the present application is further described in detail by the accompanying drawings and embodiments.
Referring to fig. 1, the present application provides a license plate character recognition method, including:
s10, acquiring a plurality of character license plate images, wherein each character in each character license plate image is marked with a real character type and a real character position, and acquiring the number of each real character type;
s20, screening a plurality of characters according to the number of each real character category to form a plurality of data enhanced copy characters and a plurality of non-data enhanced copy characters, wherein the number of the character categories corresponding to the data enhanced copy characters is less than that of the character categories corresponding to the non-data enhanced copy characters;
s30, replacing at least part of non-data enhanced copy characters in each character license plate image with data enhanced copy characters to obtain a plurality of extended character license plate images, and performing data enhancement processing on the plurality of extended character license plate images and the plurality of character license plate images to obtain a plurality of converted character license plate images;
s40, performing model training on the character target detection model according to the plurality of converted character license plate images to obtain a trained character target detection model;
s50, inputting the license plate image of the character to be detected into the trained character target detection model, and obtaining the character type and the character position of each character in the license plate image of the character to be detected.
In S10, a plurality of character license plate images are obtained by extracting image frames containing vehicle license plates from the surveillance video and further extracting license plate images from the image frames containing vehicle license plates. The plurality of character license plate images are vehicle license plate images captured by real road traffic scene monitoring, and the license plate images cover license plate images under different visual angles, different backgrounds and different illumination intensities. The license plate images with the characters comprise license plate information of different types of vehicles such as cars, new energy vehicles, buses, motorcycles, electric vehicles and trucks.
A character license plate image can also be understood as an image containing the entire license plate information, for example an image containing license plate jing a 12345. Each character in each character license plate image is labeled and contains information of real character types and real character positions. And displaying the character position in the character license plate image in the form of a detection frame. When each character in the license plate image of each character is labeled, an algorithm method or a manual labeling method can be adopted for labeling. The categories of real character categories may include 10 Arabic numerals (0-9), 24 capital letters (A-Z, except I and O), and 50 Chinese characters. And calculating the number of each real character type in the license plate images with multiple characters to obtain the number corresponding to each real character type.
In S20, a screening, division, and comparison are performed according to the number corresponding to each real character category, and a smaller number of character categories are screened out as data enhancement objects, thereby forming a plurality of data enhancement copy characters. The number of character categories corresponding to the non-data enhanced copy characters is greater than the number of character categories corresponding to the data enhanced copy characters relative to the data enhanced copy characters.
In S30, each character license plate image includes non-data enhanced copy characters or/and data enhanced copy characters. The non-data enhanced copy characters in the license plate image of each character are replaced by the data enhanced copy characters, so that the number corresponding to the data enhanced copy characters is increased, and the number corresponding to the non-data enhanced copy characters is reduced. Therefore, at least part of the non-data enhanced copy characters in each character license plate image are replaced by the data enhanced copy characters, and the problem of unbalanced character type data caused by large difference of different character type quantities in a model training data set is solved.
In S40, the character target detection model includes a plurality of convolution layers, a normalization layer, a non-linear active layer, and a residual convolution module. The feature size of each transformed character license plate image is adjusted to 64 × 256 × 3(H × W × C), and the feature size is adjusted to 4 × 16 × 256 by sequentially passing through a plurality of convolution layers, a normalization layer, and a nonlinear activation layer. And after the size of the feature map of each converted character license plate image is adjusted to 4 × 16 × 256, further feature extraction is carried out through a residual convolution module, and the size of the feature map is unchanged. The residual convolution module comprises a plurality of residual convolution neural networks connected. Normalization layers include, but are not limited to, instance normalization layers, adaptive instance normalization layers, and the like. The nonlinear activation layer includes, but is not limited to, ReLu, Leaky ReLu, and other nonlinear activation functions. And performing model training on the character target detection model by using the plurality of converted character license plate images as a model training set of the character target detection model, and finally obtaining the trained character target detection model.
In S50, the license plate image of the character to be detected is predicted according to the trained character target detection model, and the character type and the character position of each character in the license plate image of the character to be detected can be obtained. The specific reason of each character can be obtained according to the character type, the specific coordinate of each character can be obtained according to the character position, and then the complete license plate in the license plate image of the character to be detected can be obtained.
In this embodiment, a data enhanced copy character and a non-data enhanced copy character are formed by screening a plurality of characters. The number of character categories corresponding to the data-enhanced copy characters is less than the number of character categories corresponding to the non-data-enhanced copy characters. Screening and comparing according to the number of different characters, replacing the non-data enhanced copy characters with the data enhanced copy characters, adjusting the number proportion of different characters in the newly generated extended character license plate image, and carrying out balance adjustment on the number proportion of different characters. The data-enhanced copied characters and the non-data-enhanced copied characters are from the real character license plate image, so that the expanded character license plate image formed after replacement still comes from the conversion of the real character license plate image, and excessive noise caused by recognition by using a generated countermeasure network in the traditional method is avoided.
Furthermore, the plurality of expanded character license plate images and the plurality of character license plate images are fused together, and the fused expanded character license plate images and the plurality of character license plate images are subjected to data enhancement processing and then serve as a training data set of the character target detection model, so that training data are more real and the number of different characters is balanced. Therefore, in the training data set of the character target detection model, the size difference among a plurality of single characters is small, and the proportion of different license plate character types is balanced after data enhancement, so that the target detection task difficulty is low, the detection precision is high, a complex network structure is not required to be designed, the model training of the character target detection model is facilitated, and the character target detection model which is more stable and optimized is obtained. According to the trained character target detection model, different types of license plates such as two-layer trucks, motorcycle license plates, military license plates, harbor and australian region license plates or embassy license plates can be processed, and the character category and the character position of each character in the license plate image are obtained. And sequencing the characters according to the character types and the character positions, and outputting a complete license plate.
In one embodiment, S10 obtains a plurality of character license plate images, each character in each character license plate image is labeled with a real character type and a real character position, obtains the number of each real character type, and performs license plate detection on an image frame in a surveillance video by using a license plate target detection algorithm, so as to obtain position information corresponding to a license plate. According to the position information corresponding to the license plate, the image frame can be intercepted, and a character license plate image is obtained. Or the image frame is marked and intercepted by adopting a manual marking mode to obtain a character license plate image. And labeling each character in the character license plate image by using a rectangular frame by adopting a 2D target detection labeling tool to obtain a labeling label of each character license plate image. The labeling label of each character license plate image comprises the real character type and the real character position of each character.
In one embodiment, the license plate target detection algorithm includes, but is not limited to, two-stage or single-stage target detection methods such as fast-RCNN, YOLO, SSD, etc. The labeling label data of each character license plate image comprises but is not limited to storage by using txt, json, xml and other format files.
In one embodiment, S20, the characters are filtered according to the number of each character category to form a plurality of data-enhanced copy characters and a plurality of non-data-enhanced copy characters, the number of the character categories corresponding to the data-enhanced copy characters is smaller than the number of the character categories corresponding to the non-data-enhanced copy characters, and the ratio of the number of the data-enhanced copy characters to the number of the non-data-enhanced copy characters is 1: 2.
In this embodiment, the quantity ordering among the characters can be obtained by counting the quantity corresponding to each character type. According to the quantity arrangement condition, a plurality of characters are divided into two types of data enhanced copy characters and non-data enhanced copy characters. The number and the proportion of different characters in the newly generated license plate image with the expanded characters can be balanced and adjusted by setting the number proportion of the data enhanced copy characters to the non-data enhanced copy characters to be 1:2 and replacing at least part of the non-data enhanced copy characters with the data enhanced copy characters. In addition, the data enhancement copy character and the non-data enhancement copy character are both from a real image, so that the problem that excessive noise is introduced by adopting a generation countermeasure network in the traditional method can be avoided.
In one embodiment, S30, replacing at least a portion of the non-data enhanced copied characters in each character license plate image with data enhanced copied characters to obtain a plurality of extended character license plate images, and performing data enhancement processing on the plurality of extended character license plate images and the plurality of character license plate images to obtain a plurality of converted character license plate images, includes:
s310, obtaining the image size of at least part of non-data enhanced copy characters in the character license plate image, and adjusting the image size of the data enhanced copy characters according to the image size of the non-data enhanced copy characters to obtain the data enhanced copy characters after size adjustment;
s320, replacing at least part of the non-data enhanced copied characters with the data enhanced copied characters after size adjustment to obtain a plurality of extended character license plate images;
and the data enhanced copy character after the size adjustment is marked with a corresponding real character type and a real character position.
In this embodiment, at least a part of the non-data enhanced copy characters in the character license plate image are the non-data enhanced copy characters that need to be replaced. Character positions, i.e., coordinate information, of the non-data enhanced copy characters to be replaced are obtained. And calculating and obtaining the image size of the non-data enhanced copy character to be replaced according to the coordinate information. Further, the image size of the data-enhanced copy character is adjusted to the image size of the non-data-enhanced copy character to be replaced, and the replacement is performed. In the replacement process, the replacement of the character category and the character position is realized at the same time. Therefore, each expanded character license plate image in the plurality of expanded character license plate images formed after replacement is marked with a real character type and a real character position.
At least part of the non-data enhanced copy characters in the license plate image of the character can be one or more non-data enhanced copy characters at any position. For example: the license plate characters of the character license plate image are 'Jing A12345', the 'Jing', the 'A', the '3' and the '4' are replaced by 'Tibetan', and the license plate characters of the license plate image with the expansion characters are 'Tibetan 12 Tibetan 5'. Wherein, the 'Jing', 'A', '3' and '4' are non-data enhanced copy characters, and the 'Tibetan' is a data enhanced copy character. By replacing at least part of the non-data enhanced copy characters with the data enhanced copy characters after size adjustment, a plurality of different extended character license plate images can be obtained. The expanded character license plate image is completely different from the character license plate image, and the model training data set is fully expanded, so that the training data is more diversified, the problem of unbalanced character type data caused by large difference of different character type numbers in the model training data set is solved, and a more optimized model can be obtained more favorably.
In one embodiment, in S30, at least a portion of the non-data enhanced copied characters in each character license plate image are replaced with data enhanced copied characters, and a number of data enhanced copied characters in each extended character license plate image is 1/8 to 4/7 of the total number of characters in the extended character license plate image is obtained.
In the embodiment, the proportion of the total number of characters in the license plate image of the expanded characters is 1/8-4/7 of the total number of characters by adjusting the data-enhanced copied characters, and the non-data-enhanced copied characters are replaced by the data-enhanced copied characters, so that the number and proportion of different characters in the newly generated license plate image of the expanded characters can be well balanced and adjusted, and the overall balance of a training data set is realized. In addition, the data enhancement copy character and the non-data enhancement copy character are both from a real image, so that the problem that excessive noise is introduced by adopting a generation countermeasure network in the traditional method can be avoided.
In one embodiment, S30, replacing at least a portion of the non-data enhanced copied characters in each character license plate image with data enhanced copied characters to obtain a plurality of extended character license plate images, and performing data enhancement processing on the plurality of extended character license plate images and the plurality of character license plate images to obtain a plurality of converted character license plate images, further includes:
s330, rotating the plurality of expanded character license plate images and the plurality of character license plate images to obtain a plurality of rotated character license plate images;
s340, performing brightness conversion on the license plate images with the plurality of rotating characters to obtain license plate images with the plurality of brightness conversion characters;
s350, cutting the license plate images with the brightness conversion characters to obtain a plurality of converted character license plate images;
each character in each converted character license plate image is marked with a real character type and a real character position.
In S330, the license plate images with different rotation angles can be obtained by rotating the multiple extended character license plate images and the multiple character license plate images, so as to simulate oblique license plate image data appearing in a real shooting scene. When the plurality of extended character license plate images and the plurality of character license plate images are respectively rotated, the images can be rotated by different angles of 10 degrees, 15 degrees, 30 degrees, 50 degrees, 120 degrees and the like. Therefore, after each expansion character license plate image or each character license plate image is rotated, the number of the obtained rotation character license plate images is more than multiple times of the number of the expansion character license plate images or the character license plate images, and the number and the diversity of the model training data sets are increased.
In S340, license plate image data in the real shooting scene under different brightness conditions in the morning, the middle and the evening can be simulated by performing brightness conversion on the license plate images with the plurality of rotating characters. When the brightness of the license plate images with the plurality of rotating characters is converted, the adjustment of the morning brightness, the noon brightness and the night brightness can be realized, and the license plate image data under different environmental conditions can be obtained. The number of the obtained license plate images with the brightness conversion characters is more than three times that of the license plate images with the rotation characters, and the number and the diversity of the model training data sets are further increased.
In S350, by cutting the license plate image with the plurality of luminance conversion characters, license plate image data under a shielding condition encountered in a real shooting scene can be simulated. When the license plate image of a plurality of luminance conversion characters is cut out, the license plate image can be cut out to have different areas such as 1/4, 1/3 or 1/2 of the original image. And obtaining corresponding license plate image data under different shielding conditions. The number of the obtained converted character license plate images is more than multiple times of the number of the brightness conversion character license plate images, and the number and diversity of the model training data sets are further increased. Compared with the sum of the number of the expanded character license plate images and the number of the character license plate images, the number of the converted character license plate images is increased by more than multiple times, and richer and more diverse data are provided for model training.
In the embodiment, a plurality of expansion character license plate images and a plurality of character license plate images are rotated to form a plurality of rotation character license plate images, and the rotation character license plate images are subjected to brightness conversion and cutting processing in sequence to form a plurality of conversion character license plate images serving as a model training data set, so that license plate images acquired under a plurality of factors such as illumination conditions, weather conditions, shooting angles of license plates, colors of license plates and the like can be simulated really, and richer and more diverse data are provided for model training. Therefore, the data enhancement processing is carried out on the plurality of extended character license plate images and the plurality of character license plate images, the applicability of the character target detection model is improved, the license plate images collected under various conditions can be detected, and the method and the device can be applied to various application environments such as urban traffic management, parking lot charging management, violation processing and the like.
In one embodiment, the S40, performing model training on the character target detection model according to the plurality of transformed character license plate images to obtain a trained character target detection model, including:
s410, inputting each converted character license plate image into a character target detection model, and outputting a predicted character type and a predicted character position of each character in each converted character license plate image;
s420, constructing a classification loss function according to the predicted character category and the real character category;
s430, constructing a position regression loss function according to the predicted character position and the real character position;
and S440, training and optimizing the character target detection model according to the classification loss function and the position regression loss function to obtain a trained character target detection model.
In this embodiment, the classification loss function and the position regression loss function form a total loss function of the character target detection model. The classification Loss function can be a cross entropy Loss function or a Focal Loss function, and the like. The position regression Loss function can be an L1 Loss mean absolute error Loss function, an L2 Loss mean square error Loss function, an IoU Loss cross-over ratio Loss function, and other position regression Loss functions. The character target detection model may be a deep learning based target detection algorithm. And optimizing and adjusting parameters of the character target detection model by adjusting the classification loss function and the position regression loss function, finally obtaining a stable character target detection model, and detecting the license plate image of the character to be detected.
In one embodiment, the classification loss function is model trained using a cross-entropy classification loss function.
Wherein the classification loss function is as follows:
where P indicates the probability that the predicted character type is the true character type, and y-0 and y-1 indicate the case where the true character type corresponds to 0 or 1, respectively.
In one embodiment, the position regression Loss function adopts an IoU Loss cross-over ratio Loss function to carry out license plate character position regression. The IoU Loss cross ratio Loss function is based on IoU cross ratios between predicted and true character positions. The predicted character position and the real character position both exist in the form of a detection box. The IoU Loss of pass ratio Loss function can also be understood as being based on the cross ratio between the prediction detection frame and the real detection frame, and is expressed as follows:
LIoU＝1-IoU。
in one embodiment, after S410 and before S420, S40 further includes:
s411, obtaining the confidence of a detection frame corresponding to each predicted character position in each converted character license plate image;
s412, sequencing the detection frames corresponding to the predicted character positions in the license plate image of each converted character according to the confidence, and outputting the predicted character positions and the predicted character types corresponding to high confidence according to the preset number of the license plate character digits.
In this embodiment, a Non-Maximum Suppression algorithm (NMS) is used to remove a redundant box with a low confidence level. The predicted character position may be indicated by a detection box. And sequencing the detection frames corresponding to the positions of the predicted characters in the license plate image of each converted character according to the numerical value of the confidence coefficient, so as to obtain a plurality of detection frames with the confidence coefficients sequenced from top to bottom. The preset number of license plate characters can be understood as the number of license plate characters, for example, if the number of characters in "jing a 12345" is 7, the number of characters in the preset number of license plate characters can be set to 7. The preset number of characters of the license plate can be set according to actual conditions, and specific numerical values can be positive integers such as 6, 7 and 8. And when the number of the detection frames is larger than the preset number of the characters of the license plate, the license plate is regarded as a redundant frame. And outputting the detection frames corresponding to high confidence degrees, namely outputting corresponding predicted character positions according to the sequence of the confidence degrees of the detection frames and the preset number of the characters of the license plate. And then, acquiring the corresponding predicted character type according to the output predicted character position. By removing the redundant detection frames with low confidence coefficient, the recognition accuracy of the license plate character recognition method can be further improved.
In one embodiment, S50 inputs the license plate image with the characters to be detected into the trained character target detection model, obtains the character type and the character position of each character in the license plate image with the characters to be detected, and outputs complete license plate characters according to the character type and the character position of each character and the reading habit from left to right, which is suitable for a line of license plate characters.
In one embodiment, S50 inputs the license plate image with the characters to be detected into the trained character target detection model, obtains the character type and the character position of each character in the license plate image with the characters to be detected, and outputs complete license plate characters according to the character type and the character position of each character and the reading habit from top to bottom and from left to right according to the character type and the character position of each character, which is suitable for the license plate characters in two rows.
According to the license plate character recognition method, real license plate characters are randomly combined to generate a new character license plate image, the number and the proportion of different characters in the new character license plate image are adjusted according to the number and the proportion of the different characters, the number proportion of the different license plate characters is balanced to form a plurality of expansion character license plate images, and the expansion character license plate images are used for model training. The character target detection model is based on a deep learning target detection algorithm and is trained according to the model training data set to form a trained character target detection model which is used for detecting a license plate image of a character to be detected and obtaining a complete license plate character.
Referring to fig. 2, in one embodiment, the present application provides a license plate character recognition apparatus 100 including a data acquisition module 10, a data enhancement copy character acquisition module 20, a converted character license plate image acquisition module 30, a character target detection model acquisition module 40, and a detection module 50.
The data acquisition module 10 is configured to acquire a plurality of character license plate images, where each character in each character license plate image is labeled with a real character type and a real character position, and acquire the number of each real character type. The data enhanced copy character acquisition module 20 is configured to filter a plurality of characters according to the number of each real character category, to form a plurality of data enhanced copy characters and a plurality of non-data enhanced copy characters, where the number of the character categories corresponding to the data enhanced copy characters is smaller than the number of the character categories corresponding to the non-data enhanced copy characters.
The converted character license plate image acquisition module 30 is configured to replace at least a portion of the non-data enhanced copy characters in each character license plate image with data enhanced copy characters to obtain a plurality of extended character license plate images, and perform data enhancement processing on the plurality of extended character license plate images and the plurality of character license plate images to obtain a plurality of converted character license plate images. The character target detection model acquisition module 40 is configured to perform model training on the character target detection model according to the plurality of transformed character license plate images, and acquire a trained character target detection model. The detection module 50 is configured to input the license plate image of the character to be detected into the trained character target detection model, and obtain the character type and the character position of each character in the license plate image of the character to be detected.
In this embodiment, reference may be made to the description of S10 in the above embodiment for the related description of the data obtaining module 10. The related description of the data enhanced copy character acquisition module 20 may refer to the related description of S20 in the above embodiment. The related description of the transformed character license plate image obtaining module 30 can refer to the related description of S30 in the above embodiment. The relevant description of the character target detection model obtaining module 40 can refer to the relevant description of S40 in the above embodiment. The relevant description of the detection module 50 can refer to the relevant description of S50 in the above embodiment.
In one embodiment, the transformed character license plate image acquisition module 30 includes a resizing module (not shown) and an extended character license plate image acquisition module (not shown). The size adjusting module is used for obtaining the image size of at least part of the non-data enhanced copy characters in the character license plate image, adjusting the image size of the data enhanced copy characters according to the image size of the non-data enhanced copy characters, and obtaining the data enhanced copy characters after size adjustment. The extended character license plate image acquisition module is used for replacing at least part of the non-data enhanced copy characters with the data enhanced copy characters after size adjustment to obtain a plurality of extended character license plate images. And the data enhanced copy character after the size adjustment is marked with a corresponding real character type and a real character position.
In this embodiment, the relevant description of the resizing module can refer to the relevant description of S310 in the above embodiment. The related description of the extended character license plate image acquisition module can refer to the related description of S320 in the above embodiment.
In one embodiment, the transformed character license plate image acquisition module 30 further comprises a rotation module (not shown), a brightness conversion module (not shown), and a cropping module (not shown). The rotation module is used for rotating the plurality of expanded character license plate images and the plurality of character license plate images to obtain a plurality of rotation character license plate images. The brightness conversion module is used for performing brightness conversion on the license plate images with the plurality of rotating characters to obtain the license plate images with the plurality of brightness conversion characters. The cutting module is used for cutting the license plate images with the brightness conversion characters to obtain a plurality of converted character license plate images. Each character in each converted character license plate image is marked with a real character type and a real character position.
In this embodiment, reference may be made to the description of S330 in the above embodiment for the related description of the rotation module. The relevant description of the luminance conversion module can refer to the relevant description of S340 in the above embodiment. The related description of the clipping module can refer to the related description of S350 in the above embodiment.
In one embodiment, the character object detection model obtaining module 40 includes a prediction data obtaining module (not shown), a classification loss function building module (not shown), a position regression loss function building module (not shown), and a training optimization module (not shown).
The prediction data acquisition module is used for inputting each converted character license plate image into the character target detection model and outputting the predicted character category and the predicted character position of each character in each converted character license plate image. And the classification loss function building module is used for building a classification loss function according to the predicted character category and the real character category. And the position regression loss function building module is used for building a position regression loss function according to the predicted character position and the real character position. And the training optimization module is used for training and optimizing the character target detection model according to the classification loss function and the position regression loss function to obtain the trained character target detection model.
In this embodiment, reference may be made to the description of S410 in the above embodiment for the description of the prediction data obtaining module. The relevant description of the classification loss function building block may refer to the relevant description of S420 in the above embodiment. The relevant description of the positional regression loss function building block can refer to the relevant description of S430 in the above embodiment. The related description of the training optimization module can refer to the related description of S440 in the above embodiment.
In one embodiment, the license plate character recognition apparatus 100 further includes a confidence level obtaining module (not shown) and an output module (not shown).
The confidence coefficient acquisition module is used for acquiring the confidence coefficient of the detection frame corresponding to each predicted character position in each converted character license plate image. The output module is used for sequencing the detection frames corresponding to the predicted character positions in the license plate image of each converted character according to the confidence coefficient, and outputting the predicted character positions and the predicted character types corresponding to high confidence coefficient according to the preset number of character digits of the license plate.
In this embodiment, the relevant description of the confidence level obtaining module may refer to the relevant description of S411 in the above embodiment. The relevant description of the output module can refer to the relevant description of S412 in the above embodiment.
In one embodiment, in the transformed character license plate image acquisition module, the number of the data enhanced copy characters in each expanded character license plate image is 1/8-4/7 of the total number of the characters.
In this embodiment, the ratio of the number of the data enhanced copy characters in each extended character license plate image to the total number of the characters may refer to the corresponding description in the above embodiments.
In the various embodiments described above, the particular order or hierarchy of steps in the processes disclosed is an example of exemplary approaches. Based upon design preferences, it is understood that the specific order or hierarchy of steps in the processes may be rearranged without departing from the scope of the present disclosure. The accompanying method claims present elements of the various steps in a sample order, and are not intended to be limited to the specific order or hierarchy presented.
Those of skill in the art will further appreciate that the various illustrative logical blocks, units, and steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate the interchangeability of hardware and software, various illustrative components, elements, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design requirements of the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the embodiments of the present application.
The various illustrative logical blocks, or elements described in this application may be implemented or operated by a general purpose processor, a digital signal processor, an Application Specific Integrated Circuit (ASIC), a field programmable gate array or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described. A general-purpose processor may be a microprocessor, but, in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices, e.g., a digital signal processor and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a digital signal processor core, or any other similar configuration.
The steps of a method or algorithm described in the embodiments herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may be stored in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. For example, a storage medium may be coupled to the processor such that the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium may be integral to the processor. The processor and the storage medium may reside in an ASIC, which may be located in a user terminal. In the alternative, the processor and the storage medium may reside as discrete components in a user terminal.
The above-mentioned embodiments, objects, technical solutions and advantages of the present application are described in further detail, it should be understood that the above-mentioned embodiments are merely exemplary embodiments of the present application, and are not intended to limit the scope of the present application, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present application should be included in the scope of the present application.
Claims (12)
1. A license plate character recognition method is characterized by comprising the following steps:
acquiring a plurality of character license plate images, wherein each character in each character license plate image is marked with a real character type and a real character position, and acquiring the number of each real character type;
screening the characters according to the number of each real character category to form a plurality of data enhanced copy characters and a plurality of non-data enhanced copy characters, wherein the number of the character categories corresponding to the data enhanced copy characters is smaller than the number of the character categories corresponding to the non-data enhanced copy characters;
replacing at least part of the non-data enhanced copy characters in each character license plate image with the data enhanced copy characters to obtain a plurality of extended character license plate images, and performing data enhancement processing on the plurality of extended character license plate images and the plurality of character license plate images to obtain a plurality of converted character license plate images;
performing model training on a character target detection model according to the plurality of converted character license plate images to obtain a trained character target detection model;
and inputting the license plate image of the character to be detected into the trained character target detection model, and obtaining the character category and the character position of each character in the license plate image of the character to be detected.
2. The license plate character recognition method of claim 1, wherein the replacing at least a portion of the non-data enhanced copy characters in each of the character license plate images with the data enhanced copy characters to obtain a plurality of extended character license plate images, and performing data enhancement processing on the plurality of extended character license plate images and the plurality of character license plate images to obtain a plurality of converted character license plate images comprises:
acquiring the image size of at least part of the non-data enhanced copy characters in the character license plate image, and adjusting the image size of the data enhanced copy characters according to the image size of the non-data enhanced copy characters to obtain the data enhanced copy characters after size adjustment;
replacing at least part of the non-data enhanced copy characters with the data enhanced copy characters after size adjustment to obtain a plurality of extended character license plate images;
and marking the data enhanced copy character with a corresponding real character type and a real character position after the size is adjusted.
3. The license plate character recognition method of claim 2, wherein the replacing at least a portion of the non-data enhanced copy characters in each of the character license plate images with the data enhanced copy characters to obtain a plurality of extended character license plate images, and performing data enhancement processing on the plurality of extended character license plate images and the plurality of character license plate images to obtain a plurality of converted character license plate images, further comprises:
rotating the plurality of expanded character license plate images and the plurality of character license plate images to obtain a plurality of rotated character license plate images;
performing brightness conversion on the plurality of rotary character license plate images to obtain a plurality of brightness conversion character license plate images;
cutting the license plate images with the brightness conversion characters to obtain license plate images with the conversion characters;
each character in each converted character license plate image is marked with a real character type and a real character position.
4. The license plate character recognition method of claim 3, wherein the performing model training on the character target detection model according to the plurality of transformed character license plate images to obtain the trained character target detection model comprises:
inputting each converted character license plate image into the character target detection model, and outputting the predicted character category and the predicted character position of each character in each converted character license plate image;
constructing a classification loss function according to the predicted character category and the real character category;
constructing a position regression loss function according to the predicted character position and the real character position;
and training and optimizing the character target detection model according to the classification loss function and the position regression loss function to obtain the trained character target detection model.
5. The license plate character recognition method of claim 4, wherein after the inputting each of the transformed character license plate images into the character target detection model and the outputting of the predicted character class and the predicted character position of each character in each of the transformed character license plate images, before the constructing of the classification loss function according to the predicted character class and the real character class, the method further comprises:
acquiring the confidence of a detection frame corresponding to each predicted character position in each converted character license plate image;
and sequencing the detection frames corresponding to the predicted character positions in each converted character license plate image according to the confidence, and outputting the predicted character positions and the predicted character types corresponding to high confidence according to the preset license plate character number.
6. The license plate character recognition method of claim 1, wherein at least a portion of the non-data-enhanced copied characters in each of the character license plate images are replaced with the data-enhanced copied characters to obtain a plurality of extended character license plate images, and a ratio of the number of the data-enhanced copied characters in each of the extended character license plate images to the total number of characters is 1/8 to 4/7.
7. A license plate character recognition device, comprising:
the license plate display system comprises a data acquisition module, a display module and a display module, wherein the data acquisition module is used for acquiring a plurality of character license plate images, each character in each character license plate image is marked with a real character type and a real character position, and the number of each real character type is acquired;
the data enhanced copy character acquisition module is used for screening the characters according to the number of each real character type to form a plurality of data enhanced copy characters and a plurality of non-data enhanced copy characters, wherein the number of the character types corresponding to the data enhanced copy characters is smaller than the number of the character types corresponding to the non-data enhanced copy characters;
the converted character license plate image acquisition module is used for replacing at least part of the non-data enhanced copy characters in each character license plate image with the data enhanced copy characters to obtain a plurality of extended character license plate images, and performing data enhancement processing on the plurality of extended character license plate images and the plurality of character license plate images to obtain a plurality of converted character license plate images;
the character target detection model acquisition module is used for carrying out model training on the character target detection model according to the plurality of converted character license plate images to obtain a trained character target detection model;
and the detection module is used for inputting the license plate image of the character to be detected into the trained character target detection model to obtain the character category and the character position of each character in the license plate image of the character to be detected.
8. The license plate character recognition device of claim 7, wherein the transformed character license plate image obtaining module comprises:
the size adjusting module is used for obtaining the image size of at least part of the non-data enhanced copy characters in the character license plate image, adjusting the image size of the data enhanced copy characters according to the image size of the non-data enhanced copy characters, and obtaining the data enhanced copy characters after size adjustment;
the extended character license plate image acquisition module is used for replacing at least part of the non-data enhanced copy characters with the data enhanced copy characters after size adjustment to obtain a plurality of extended character license plate images;
and marking the data enhanced copy character with a corresponding real character type and a real character position after the size adjustment.
9. The license plate character recognition device of claim 8, wherein the transformed character license plate image acquisition module further comprises:
the rotation module is used for rotating the plurality of expanded character license plate images and the plurality of character license plate images to obtain a plurality of rotation character license plate images;
the brightness conversion module is used for performing brightness conversion on the license plate images with the plurality of rotating characters to obtain license plate images with a plurality of brightness conversion characters;
the cutting module is used for cutting the license plate images with the brightness conversion characters to obtain the license plate images with the conversion characters;
each character in each converted character license plate image is marked with a real character type and a real character position.
10. The license plate character recognition device of claim 9, wherein the character target detection model obtaining module comprises:
the prediction data acquisition module is used for inputting each converted character license plate image into the character target detection model and outputting the predicted character category and the predicted character position of each character in each converted character license plate image;
the classification loss function building module is used for building a classification loss function according to the predicted character category and the real character category;
the position regression loss function building module is used for building a position regression loss function according to the predicted character position and the real character position;
and the training optimization module is used for training and optimizing the character target detection model according to the classification loss function and the position regression loss function to obtain the trained character target detection model.
11. The license plate character recognition device of claim 10, further comprising:
the confidence coefficient acquisition module is used for acquiring the confidence coefficient of a detection frame corresponding to each predicted character position in each converted character license plate image;
and the output module is used for sequencing the detection frames corresponding to the predicted character positions in each converted character license plate image according to the confidence degrees and outputting the predicted character positions corresponding to high confidence degrees and the predicted character types according to the preset license plate character digit number.
12. The license plate character recognition device of claim 7, wherein the transformed character license plate image acquisition module is configured to obtain the augmented character license plate image with the data enhanced copy characters from 1/8 to 4/7 of the total number of characters.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210291417.2A CN114782938A (en) | 2022-03-23 | 2022-03-23 | License plate character recognition method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210291417.2A CN114782938A (en) | 2022-03-23 | 2022-03-23 | License plate character recognition method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114782938A true CN114782938A (en) | 2022-07-22 |
Family
ID=82424887
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210291417.2A Pending CN114782938A (en) | 2022-03-23 | 2022-03-23 | License plate character recognition method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114782938A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115345321A (en) * | 2022-10-19 | 2022-11-15 | 小米汽车科技有限公司 | Data augmentation method, data augmentation device, electronic device, and storage medium |
-
2022
- 2022-03-23 CN CN202210291417.2A patent/CN114782938A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115345321A (en) * | 2022-10-19 | 2022-11-15 | 小米汽车科技有限公司 | Data augmentation method, data augmentation device, electronic device, and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111259758B (en) | Two-stage remote sensing image target detection method for dense area | |
| CN110826429A (en) | Scenic spot video-based method and system for automatically monitoring travel emergency | |
| Gong et al. | Unified Chinese license plate detection and recognition with high efficiency | |
| CN111523439B (en) | Method, system, device and medium for target detection based on deep learning | |
| CN107704797B (en) | Real-time detection method, system and equipment based on pedestrians and vehicles in security video | |
| CN111563516B (en) | Method, terminal and storage medium for fusion display of pedestrian mask and three-dimensional scene | |
| CN105095835A (en) | Pedestrian detection method and system | |
| CN112990136B (en) | Target detection method and device | |
| CN113723377A (en) | Traffic sign detection method based on LD-SSD network | |
| CN114038004A (en) | Certificate information extraction method, device, equipment and storage medium | |
| CN113901972A (en) | Method, device and equipment for detecting remote sensing image building and storage medium | |
| CN115909059A (en) | Natural resource sample library establishing method and device | |
| CN114782938A (en) | License plate character recognition method and device | |
| CN114842285A (en) | Roadside berth number identification method and device | |
| CN114898243A (en) | Traffic scene analysis method and device based on video stream | |
| Hu et al. | Vehicle color recognition based on smooth modulation neural network with multi-scale feature fusion | |
| Lin et al. | A lightweight, high-performance multi-angle license plate recognition model | |
| CN114155363A (en) | Converter station vehicle identification method and device, computer equipment and storage medium | |
| CN116386018A (en) | License plate character classification method and system | |
| CN111126187A (en) | Fire detection method, system, electronic device and storage medium | |
| CN116091964A (en) | High-order video scene analysis method and system | |
| CN116682101A (en) | License plate number recognition method and system | |
| CN114519791A (en) | License plate character recognition method and device | |
| CN115909140A (en) | Video target segmentation method and system based on high-order video monitoring | |
| Peng et al. | A new end-to-end secondary network for high-efficient vehicles and license plates detection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |