WO2021184847A1 - 一种遮挡车牌字符识别方法、装置、存储介质和智能设备 - Google Patents

一种遮挡车牌字符识别方法、装置、存储介质和智能设备 Download PDF

Info

Publication number
WO2021184847A1
WO2021184847A1 PCT/CN2020/135510 CN2020135510W WO2021184847A1 WO 2021184847 A1 WO2021184847 A1 WO 2021184847A1 CN 2020135510 W CN2020135510 W CN 2020135510W WO 2021184847 A1 WO2021184847 A1 WO 2021184847A1
Authority
WO
WIPO (PCT)
Prior art keywords
license plate
sample
occlusion
image
plate image
Prior art date
Application number
PCT/CN2020/135510
Other languages
English (en)
French (fr)
Inventor
林春伟
王健宗
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021184847A1 publication Critical patent/WO2021184847A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
    • G06V20/63—Scene text, e.g. street names
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
    • G06V20/625—License plates

Definitions

  • This application belongs to the field of information processing technology, and in particular relates to a method, a device, a storage medium, and a smart device for recognizing characters on a covered license plate.
  • the license plate is a specific number of the vehicle, which has the characteristics of character arrangement.
  • the color of the license plate and the color, content, and location of the license plate characters contain the information of the vehicle to which it belongs.
  • traffic management by capturing video images of vehicles and identifying license plates, it has become the most commonly used traffic management method.
  • the embodiments of the present application provide a method, device, storage medium and smart device for recognizing characters of a covered license plate to solve the problem of limited sample data in the prior art, which leads to insufficient efficiency and recognition accuracy of the covered license plate. .
  • an embodiment of the present application provides a method for recognizing characters on a covered license plate, including:
  • an embodiment of the present application provides a recognition device for obscuring license plate characters, including:
  • Image and strategy acquisition unit for acquiring sample license plate images and sample expansion strategies
  • a mask image extraction unit for extracting occlusion sample mask images from a preset sample occlusion mask image set based on the sample expansion strategy
  • the occlusion image construction unit is used to occlude the sample license plate image by using the extracted sample occlusion mask image to construct a sample occlusion license plate image set;
  • a deep learning model training unit configured to train a deep learning network model for occluding license plate character recognition based on the sample occluded license plate image set and the sample license plate image set formed by the sample license plate image;
  • the occlusion recognition unit is used to perform occlusion license plate character recognition by using the trained deep learning network model.
  • an embodiment of the present application provides a computer-readable storage medium that stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the computer-readable On the one hand, the method of occluding license plate character recognition is proposed.
  • an embodiment of the present application provides a smart device, including a memory, a processor, and computer-readable instructions stored in the memory and running on the processor, and the processor executes the computer When the instructions are readable, the method for recognizing characters on a covered license plate as proposed in the first aspect of the embodiments of the present application is realized.
  • the embodiments of the present application provide a computer-readable instruction product, which when the computer-readable instruction product runs on a terminal device, causes the terminal device to execute the method for recognizing characters on a covered license plate as described in the first aspect.
  • the sample expansion strategy is used to extract occlusion sample mask images from a preset sample occlusion mask image set, and then use the extracted sample occlusion mask images to occlude the sample license plate images to construct a sample occlusion license plate image set , Realize the expansion of the existing sample occlusion images, effectively increase the number and types of sample occlusion images, and then based on the sample license plate image set composed of the sample occlusion license plate image set and the sample license plate image, deep learning for the recognition of the occlusion license plate characters
  • the network model is trained, and the rich sample license plate image set can enhance the training effect of the deep learning network model for obscuring license plate character recognition.
  • the trained deep learning network model is used for obscuring license plate character recognition, thereby improving the recognition of obscuring license plate characters Precision and efficiency.
  • FIG. 1 is an implementation flowchart of a method for recognizing characters on a covered license plate provided by an embodiment of the present application
  • FIG. 3 is a specific implementation flow chart of the method S103 for obscuring license plate character recognition provided by an embodiment of the present application;
  • FIG. 4 is a specific implementation flow chart of the verification of the sample occluded license plate image in the occluded license plate character recognition method provided by the embodiment of the present application;
  • FIG. 5 is a structural block diagram of a device for obscuring license plate character recognition provided by an embodiment of the present application
  • Fig. 6 is a schematic diagram of a smart device provided by an embodiment of the present application.
  • the term “if” can be construed as “when” or “once” or “in response to determination” or “in response to detecting “.
  • the phrase “if determined” or “if detected [described condition or event]” can be interpreted as meaning “once determined” or “in response to determination” or “once detected [described condition or event]” depending on the context ]” or “in response to detection of [condition or event described]”.
  • the method for obstructing license plate character recognition provided by the embodiments of this application can be applied to smart devices such as servers, tablets, laptops, ultra-mobile personal computers (UMPCs), netbooks, and so on. There are no restrictions on the specific types of smart devices.
  • the present application proposes a method for recognizing occluded license plate characters, which specifically relates to deep learning technology, which can train a deep learning model by using abundant training samples, thereby improving the accuracy and efficiency of the deep learning model in recognizing occluded license plate characters.
  • FIG. 1 shows an implementation process of a method for recognizing characters on a covered license plate provided by an embodiment of the present application.
  • the process of the method includes steps S101 to S105.
  • the specific implementation principles of each step are as follows:
  • the sample expansion strategy is a preset strategy for constructing and occluding the sample license plate image, and the user can select an existing sample expansion strategy or modify the existing sample expansion strategy by himself.
  • the vehicle sample expansion strategy includes the occlusion type of the sample occluding the license plate and the ratio data of the occlusion sample image of the occlusion type.
  • the occlusion type refers to the occlusion type on the license plate image, including full occlusion, half occlusion, and superimposed occlusion.
  • FIG. 2 shows a specific implementation process of the method S101 for obstructing license plate character recognition provided by an embodiment of the present application, which is described in detail as follows:
  • A1 Randomly extract vehicle images from the vehicle image collection.
  • a random extraction algorithm is used to extract a specified number of vehicle images from the vehicle image collection.
  • the vehicle image collection is an existing collection in which vehicle images are stored.
  • A2 Extract a license plate image from the extracted vehicle image as a sample license plate image.
  • the license plate image collection includes already marked license plate data
  • the license plate data includes JPEG pictures and JSON files.
  • the position information of the license plate is obtained by parsing the JSON file, and the first specified field in the JSON file is extracted, such as the finename field, to obtain the corresponding vehicle image name, and the second specified field in the JSON file is extracted, such as
  • the annotation field acquires the coordinate information of the license plate image, and the coordinate information of the license plate image is used to identify the position of the license plate image in the vehicle picture. According to the name of the vehicle picture, the vehicle picture is searched, and the license plate image is extracted from the vehicle picture according to the coordinate information of the license plate image.
  • A3 Extract vehicle images from the acquired road surveillance video. Specifically, the road monitoring system is connected in real time or periodically, the road monitoring video is obtained, and the vehicle image is extracted from the obtained road monitoring video.
  • the license plate detection model is a model for detecting license plate images in vehicle images.
  • the TextBoxes++ model is used as the basic model of the license plate detection model
  • the TextBoxes++ model is a graphic detection model based on the ssd framework.
  • the original TextBoxes++ model uses the VGG16 network as the basic network.
  • Resnet is used as the basic network of the TextBoxes++ model, and multiple convolutional layers are added to construct a license plate detection model including multiple convolutional layers and sampling layers.
  • the sampling layer uses multi-scale feature sampling to extract target candidate frames for feature maps of different sizes and granularities, that is, locate and detect the sample license plate image in the vehicle image, and use maximum suppression to extract the coordinates of the sample license plate image until the output sample license plate
  • the error between the image and its license plate coordinates and the sample license plate image and its license plate coordinates marked by the sample vehicle data is within the specified error range.
  • a trained license plate detection model is obtained, and the license plate image is processed by the pre-trained license plate detection model. Position detection and obtain license plate images.
  • A5 Use the obtained license plate image as a sample license plate image.
  • many vehicle images can be generated in the traffic system every day, and the vehicle images in the vehicle video of the road monitored by the traffic system are acquired, and the vehicle images are denoised and input to the license plate detection model. , Output the license plate image and the license plate coordinates, so that a large number of license plate images can be obtained as sample license plate images.
  • the preset occlusion sample mask image is an image used for occlusion, and there are many types, for example, an image of an optical disc, an image of a glove, an image of a rag, and an image of a tissue.
  • the preset occlusion sample mask image can be separated and extracted from the existing occlusion sample images in the database.
  • the sample expansion strategy includes the occlusion type and the ratio data of the sample corresponding to the occlusion type to occlude the license plate image. Based on the occlusion type and the sample corresponding to the occlusion type, the sample occludes the license plate image from the sample corresponding to the occlusion type. In the set of occlusion mask images, the number of occlusion sample mask images corresponding to the ratio data is extracted.
  • S103 Use the extracted sample occlusion mask image to occlude the sample license plate image to construct a sample occlusion license plate image set.
  • the constructed sample occlusion license plate image set includes a occlusion sample license plate image that uses the extracted sample occlusion mask image to occlude the sample license plate image.
  • a random algorithm is used to extract samples to occlude the mask image.
  • the vehicle image used to extract the license plate image is a vehicle image taken in a natural scene, and the taken license plate is a deformed quadrilateral after projection transformation, if the license plate image is directly occluded, the occlusion is likely to be inaccurate. The quality of the license plate data that is constructed to obscure the license plate is poor.
  • the embodiment of the application further includes correcting the sample license plate image, which specifically includes: performing image correction on the sample license plate image according to the license plate coordinates and a preset correction algorithm, and obtaining the corrected image Standard sample license plate image.
  • the following transformation formula (1) is used to obtain the license plate coordinates of the corrected rectangular sample license plate image:
  • the transformation matrix can be divided into 4 parts, Represents linear transformation, [a 31 a 32 ] is used for translation, [a 31 a 32 ] T produces perspective transformation, a 33 is a constant value 1, and [x′,y′,w′] is the license plate coordinates after correction conversion , [U,v,w] is the license plate coordinates of the sample license plate image before correction, the sample license plate image is a two-dimensional image, w′ is a constant value, and then the The license plate coordinates of the sample license plate image after correction:
  • the sample license plate image is corrected to correct the deformed quadrilateral sample license plate image into a rectangular sample license plate image, which facilitates the construction of the occluded sample image later, and improves the construction efficiency and quality of the occluded sample image.
  • the above step S103 specifically includes: using the extracted sample occlusion mask image to occlude the standard sample license plate image to construct a sample occlusion license plate image set.
  • FIG. 3 shows a specific implementation process of the method S103 for obscuring license plate character recognition provided by an embodiment of the present application, which is described in detail as follows:
  • the sample license plate image is combined with a randomly selected sample occlusion sample image, and the same combination includes a sample license plate image and at least one sample occlusion mask image, that is, in a combination, a sample license plate image and a sample occlusion mask image Combination of code images, or a combination of a sample license plate image and multiple sample occlusion mask images.
  • B2 Perform image fusion of the sample license plate image and the sample occlusion mask image in each combination to generate multiple sample occlusion license plate images. Specifically, the occlusion image coordinates are generated according to the license plate coordinates and the occlusion type of the sample license plate image, the pixels of the occlusion mask image are overlaid on the sample license plate image according to the occlusion image coordinates, and the image pixel overlay operation is performed, Realize the effect of license plate occlusion.
  • the occlusion type is half occlusion, select a corrected sample license plate image and a sample occlusion mask image to combine; if the occlusion type is superimposed occlusion, select a corrected sample license plate image and at least two A sample occlusion mask image is combined.
  • the sample occlusion mask image is used to occlude the sample license plate image through the image pixel coverage operation, and the sample occlusion license plate image is constructed. Automatically construct a large number of samples to block the license plate image.
  • S104 Based on the sample license plate image set formed by the sample occluded license plate image set and the sample license plate image, train a deep learning network model for occluded license plate character recognition.
  • the sample occlusion mask image, the sample license plate image, and the occlusion type are arbitrarily combined to construct a sample occlusion license plate image set that includes a large number of sample occlusion license plate images, and the samples are occluded
  • the license plate image set and the sample license plate image set constituted by the sample license plate images are used as input to train the deep learning network model for obscuring the license plate character recognition, which greatly enriches the training samples of the deep learning network model, thereby enhancing the Describe the training effect of the deep learning network model.
  • the method for recognizing characters of a occluded license plate further includes verifying the constructed sample occluded license plate image in the set of sample occluded license plate images.
  • FIG. 4 shows the specific implementation process of the verification of the sample occluded license plate image in the method for occluding the license plate character recognition provided by the embodiment of the present application, which is described in detail as follows:
  • the license plate occlusion recognition model refers to a model used to recognize license plates.
  • the training step of the license plate occlusion recognition model includes:
  • c1 Construct a license plate occlusion recognition model, the license plate occlusion recognition model includes a convolutional layer and a cyclic neural network layer;
  • c2 Use annotated positive and negative sample license plate images as the input of the license plate occlusion recognition model, where the positive sample license plate image is an unoccluded normal license plate image, and the negative sample license plate image is a occluded license plate image;
  • c3 Perform feature extraction in the convolutional layer to obtain a feature sequence
  • c4 input the feature sequence into the recurrent neural network layer for occlusion recognition, and output the occlusion recognition result;
  • the license plate occlusion recognition model adopts the CRNN convolutional recurrent neural network algorithm, and the training process of the license plate occlusion recognition model includes taking annotated positive and negative sample license plate images as input, where the positive The sample license plate image is an unoccluded normal license plate image, and the negative sample license plate image is an occluded license plate image.
  • the feature extraction is performed through the convolutional layer, and the last layer is stitched column by column to obtain the feature sequence, and the obtained feature sequence is put into the recurrent neural network
  • the occlusion recognition is performed in the layer, and finally the repeated characters are combined through the translation layer to output the occlusion recognition result, and the model parameters of the license plate occlusion recognition model are adjusted to the optimal according to the recognition result and the error function.
  • C2 According to the output result of the license plate occlusion recognition model, verify whether the sample occluded license plate image is an effectively occluded sample license plate image.
  • the sample occlusion license plate image in the sample occlusion license plate image set is used as the input of the license plate occlusion recognition model, and according to the recognition result of the license plate occlusion recognition model, it is verified whether the sample occlusion license plate image is effective. .
  • since there are missing characters in the occluded license plate it can be judged whether the license plate occlusion is effective according to whether the recognized character field is complete.
  • the recognized character field is complete, it is determined that the license plate occlusion in the license plate occlusion image is invalid; if the recognized character field is Incomplete, it is determined that the license plate occlusion in the license plate occlusion image is valid;
  • C3 According to the sample license plate image verified as effective occlusion, construct the target sample occlusion license plate image set. Specifically, mark the sample occluded license plate image with invalid occlusion, delete the sample occluded license plate image with invalid occlusion from the set of sample occluded license plate images, and then construct a target sample occluded license plate image that contains the sample occluded license plate image that is verified as effective. set.
  • the step of training a deep learning network model for obscuring license plate character recognition based on the sample occluded license plate image set and the license plate image set including the sample license plate image includes:
  • a deep learning network model for occlusion license plate character recognition is trained.
  • the sample occluded license plate image in the sample occluded license plate image set is verified to cover the sample occlusion.
  • the mask image can still identify the occluded license plate samples of the license plate characters, thereby improving the validity of the constructed sample data.
  • S105 Use the trained deep learning network model to perform occlusion license plate character recognition.
  • the occluded license plate image to be recognized is used as the trained deep learning network model to perform occluded license plate character recognition.
  • occlusion sample mask images are extracted from a preset sample occlusion mask image set, and then the extracted sample occlusion mask image pairs are used
  • the sample license plate image is occluded, the sample occlusion license plate image set is constructed, the existing sample occlusion images are expanded, and the number and types of the sample occlusion images are effectively increased, and then based on the sample occlusion license plate image set and the sample license plate
  • the sample license plate image set composed of images is used to train the deep learning network model for obscuring license plate character recognition.
  • the rich sample license plate image set can enhance the training effect of the deep learning network model for obscuring license plate character recognition. Finally, use the training well.
  • the deep learning network model of the occluded license plate character recognition can improve the accuracy and efficiency of the occluded license plate character recognition.
  • FIG. 5 shows a structural block diagram of the device for occluding license plate characters provided by an embodiment of the present application. part.
  • the occlusion license plate character recognition device includes: an image and strategy acquisition unit 51, a mask image extraction unit 52, an occlusion image construction unit 53, a deep learning model training unit 54, and an occlusion recognition unit 55, in which:
  • the image and strategy acquisition unit 51 is used to acquire sample license plate images and sample expansion strategies
  • the mask image extraction unit 52 is configured to extract occlusion sample mask images from a preset sample occlusion mask image set based on the sample expansion strategy;
  • the occlusion image construction unit 53 is configured to use the extracted sample occlusion mask image to occlude the sample license plate image to construct a sample occlusion license plate image set;
  • the deep learning model training unit 54 is configured to train a deep learning network model for occluding license plate character recognition based on the sample occlusion license plate image set and the sample license plate image set composed of the sample license plate image;
  • the occlusion recognition unit 55 is configured to perform occlusion license plate character recognition by using the trained deep learning network model.
  • the image and strategy acquisition unit includes:
  • the vehicle image extraction module is used to randomly extract vehicle images from the vehicle image collection
  • a first sample license plate image acquisition module configured to extract a license plate image from the extracted vehicle image as a sample license plate image
  • the vehicle image extraction module is used to extract vehicle images from the acquired road monitoring video
  • the license plate image detection module is used to detect the vehicle image by using the pre-trained license plate detection model to obtain the license plate image;
  • the second sample license plate image acquisition module is used to use the acquired license plate image as a sample license plate image.
  • the device for obstructing license plate character recognition further includes:
  • An image correction unit configured to perform image correction on the sample license plate image according to the license plate coordinates and a preset correction algorithm, and obtain a corrected standard sample license plate image
  • the occlusion image construction unit is also used for:
  • the extracted sample occlusion mask image is used to occlude the standard sample license plate image to construct a sample occlusion license plate image set.
  • the occlusion image construction unit includes:
  • An image combination module configured to randomly combine the sample license plate image and the sample occlusion mask image
  • the image fusion module is used for image fusion of the sample license plate image and the sample occlusion mask image in each combination to generate multiple sample occlusion license plate images;
  • the occlusion image building module is used to occlude the license plate image based on the constructed sample, and generate a sample occlusion license plate image set.
  • the device for obstructing license plate character recognition further includes:
  • a verification model input determination unit configured to use the sample-masked license plate image in the sample-masked license plate image set as an input of a pre-trained license plate occlusion recognition model, the license plate occlusion recognition model for identifying whether the license plate is occluded;
  • the occlusion verification unit is configured to verify whether the sample occlusion license plate image is an effectively occluded sample license plate image according to the output result of the license plate occlusion recognition model;
  • the target image set construction unit is used to construct the target sample occluded license plate image set according to the sample license plate images that are verified as effective occlusion;
  • the deep learning model training unit 54 is further configured to train a deep learning network model for obscuring license plate character recognition based on the target sample occlusion license plate image set and the license plate image set including the sample license plate image.
  • the device for obstructing license plate character recognition further includes:
  • An occlusion recognition model construction unit for building a license plate occlusion recognition model, the license plate occlusion recognition model including a convolutional layer and a cyclic neural network layer;
  • the sample input determination unit is configured to use the labeled positive and negative sample license plate images as the input of the license plate occlusion recognition model, wherein the positive sample license plate image is an unoccluded normal license plate image, and the negative sample license plate image is a occluded license plate image;
  • a feature extraction unit configured to perform feature extraction in the convolutional layer to obtain a feature sequence
  • a character recognition unit configured to input the feature sequence into the cyclic neural network layer for occlusion recognition, and output the occlusion recognition result;
  • the model optimization unit is used to adjust the model parameters of the license plate occlusion recognition model to the optimum according to the recognition result and the error function.
  • occlusion sample mask images are extracted from a preset sample occlusion mask image set, and then the extracted sample occlusion mask image pairs are used
  • the sample license plate image is occluded, the sample occlusion license plate image set is constructed, the existing sample occlusion images are expanded, and the number and types of the sample occlusion images are effectively increased, and then based on the sample occlusion license plate image set and the sample license plate
  • the sample license plate image set composed of images is used to train the deep learning network model for obscuring license plate character recognition.
  • the rich sample license plate image set can enhance the training effect of the deep learning network model for obscuring license plate character recognition. Finally, use the training well.
  • the deep learning network model of the occluded license plate character recognition can improve the accuracy and efficiency of the occluded license plate character recognition.
  • An embodiment of the present application also provides a computer-readable storage medium, the computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, any one of those shown in FIG. 1 to FIG. 4 is implemented.
  • the embodiments of the present application also provide a computer-readable instruction product.
  • the computer-readable instruction product When the computer-readable instruction product is run on a smart device, the smart device executes any of the methods for obstructing license plate character recognition as shown in Figures 1 to 4 step.
  • An embodiment of the present application also provides a smart device including a memory, a processor, and computer-readable instructions stored in the memory and capable of running on the processor.
  • the processor executes the computer-readable instructions, Realize the steps of any one of the method for obscuring the license plate character recognition as shown in Fig. 1 to Fig. 4.
  • Fig. 6 is a schematic diagram of a smart device provided by an embodiment of the present application.
  • the smart device 6 of this embodiment includes a processor 60, a memory 61, and computer-readable instructions 62 that are stored in the memory 61 and can run on the processor 60.
  • the processor 60 executes the computer-readable instructions 62
  • the steps in the above-mentioned embodiments of the method for recognizing characters on a covered license plate are implemented, for example, steps 101 to 105 shown in FIG. 1.
  • the processor 60 executes the computer-readable instructions 62
  • the functions of the modules/units in the foregoing device embodiments, such as the functions of the units 51 to 55 shown in FIG. 5, are realized.
  • the computer-readable instructions 62 may be divided into one or more modules/units, and the one or more modules/units are stored in the memory 61 and executed by the processor 60, To complete this application.
  • the one or more modules/units may be a series of computer-readable instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer-readable instructions 62 in the smart device 6.
  • the smart device 6 may be a computing device such as a server and a cloud smart device.
  • the smart device 6 may include, but is not limited to, a processor 60 and a memory 61.
  • FIG. 6 is only an example of the smart device 6 and does not constitute a limitation on the smart device 6. It may include more or less components than shown in the figure, or a combination of certain components, or different components.
  • the smart device 6 may also include input and output devices, network access devices, buses, and so on.
  • the processor 60 may be a central processing unit (Central Processing Unit, CPU), or other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (ASIC), Ready-made programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor may be a microprocessor or the processor may also be any conventional processor or the like.
  • the memory 61 may be an internal storage unit of the smart device 6, such as a hard disk or a memory of the smart device 6.
  • the memory 61 may also be an external storage device of the smart device 6, for example, a plug-in hard disk equipped on the smart device 6, a smart memory card (Smart Media Card, SMC), or Secure Digital (SD). Card, Flash Card, etc.
  • the memory 61 may also include both an internal storage unit of the smart device 6 and an external storage device.
  • the memory 61 is used to store the computer readable instructions and other programs and data required by the smart device.
  • the memory 61 can also be used to temporarily store data that has been output or will be output.
  • the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.
  • the computer-readable storage medium may be non-volatile or It is volatile. Based on this understanding, the implementation of all or part of the processes in the above-mentioned embodiments and methods in this application can be accomplished by instructing relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the computer-readable instructions are executed by the processor, they can implement the steps of the foregoing method embodiments.
  • the computer-readable instruction includes computer-readable instruction code, and the computer-readable instruction code may be in the form of source code, object code, executable file, or some intermediate form.
  • the computer-readable medium may at least include: any entity or device capable of carrying computer-readable instruction codes to the device/terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access Memory (RAM, Random Access Memory), electric carrier signal, telecommunications signal, and software distribution medium.
  • ROM read-only memory
  • RAM random access Memory
  • electric carrier signal telecommunications signal
  • software distribution medium For example, U disk, mobile hard disk, floppy disk or CD-ROM, etc.
  • computer-readable media cannot be electrical carrier signals and telecommunication signals.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Multimedia (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)
  • Character Discrimination (AREA)
  • Traffic Control Systems (AREA)

Abstract

一种遮挡车牌字符识别方法、装置、存储介质和智能设备。所述方法包括：获取样本车牌图像以及样本扩充策略（S101）；基于所述样本扩充策略，从预设的样本遮挡掩码图像集中抽取遮挡样本掩码图像（S102）；利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集（S103）；基于所述样本遮挡车牌图像集与所述样本车牌图像构成的样本车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练（S104）；利用训练好的所述深度学习网络模型进行遮挡车牌字符识别（S105）。上述方法可提高遮挡车牌字符识别的精度和效率。

Description

一种遮挡车牌字符识别方法、装置、存储介质和智能设备
本申请要求于2020年03月17日提交中国专利局、申请号为202010185281.8，发明名称为“一种遮挡车牌字符识别方法、装置、存储介质和智能设备”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请属于信息处理技术领域，尤其涉及一种遮挡车牌字符识别方法、装置、存储介质和智能设备。
背景技术
车牌是车辆特定的编号，具有字符排列特征，车牌颜色以及车牌字符颜色、内容、位置蕴含有所属车辆的信息。在交通管理中，通过抓拍车辆的视频图像，识别出车牌，已经成为目前最为常用的一种交通管理手段。
现有技术中，采用深度学习技术可识别遮挡车牌。然而，发明人发现，使用深度学习技术进行遮挡车牌字符识别时，需要大量的包含遮挡样本图像的样本数据作为支撑，然而样本数据有限，导致遮挡车牌的效率和识别的精度不够高。
技术问题
有鉴于此，本申请实施例提供了一种遮挡车牌字符识别方法、装置、存储介质和智能设备，以解决现有技术中，样本数据有限，导致遮挡车牌的效率和识别的精度不够高的问题。
技术解决方案
第一方面，本申请实施例提供了一种遮挡车牌字符识别方法，包括：
获取样本车牌图像以及样本扩充策略；
基于所述样本扩充策略，从预设的样本遮挡掩码图像集中抽取遮挡样本掩码图像；
利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集；
基于所述样本遮挡车牌图像集与所述样本车牌图像构成的样本车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练；
利用训练好的所述深度学习网络模型进行遮挡车牌字符识别。
第二方面，本申请实施例提供了一种遮挡车牌字符识别装置，包括：
图像及策略获取单元，用于获取样本车牌图像以及样本扩充策略；
掩码图像抽取单元，用于基于所述样本扩充策略，从预设的样本遮挡掩码图像集中抽取遮挡样本掩码图像；
遮挡图像构建单元，用于利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集；
深度学习模型训练单元，用于基于所述样本遮挡车牌图像集与所述样本车牌图像构成的样本车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练；
遮挡识别单元，用于利用训练好的所述深度学习网络模型进行遮挡车牌字符识别。
第三方面，本申请实施例提供了一种计算机可读存储介质，所述计算机可读存储介质存储有计算机可读指令，所述计算机可读指令被处理器执行时实现如本申请实施例第一方面提出的遮挡车牌字符识别方法。
第四方面，本申请实施例提供了一种智能设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令，所述处理器执行所述计算机可读指令时实现如本申请实施例第一方面提出的遮挡车牌字符识别方法。
第五方面，本申请实施例提供了一种计算机可读指令产品，当计算机可读指令产品在终端设备上运行时，使得终端设备执行上述第一方面所述的遮挡车牌字符识别方法。
有益效果
本申请实施例中，通过样本扩充策略，从预设的样本遮挡掩码图像集中抽取遮挡样本掩码图像，然后利用抽取的样本遮挡掩码图像对样本车牌图像进行遮挡，构建样本遮挡车牌图像集，实现对已有的样本遮挡图像进行扩充，有效增加样本遮挡图像的数量和种类，再基于样本遮挡车牌图像集与样本车牌图像构成的样本车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练，丰富的样本车牌图像集可增强用于遮挡车牌字符识别的深度学习网络模型的训练效果，最后利用训练好的深度学习网络模型进行遮挡车牌字符识别，从而可提高遮挡车牌字符识别的精度和效率。
附图说明
为了更清楚地说明本申请实施例中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1是本申请实施例提供的遮挡车牌字符识别方法的实现流程图；
图2是本申请实施例提供的遮挡车牌字符识别方法S101的具体实现流程图；
图3是本申请实施例提供的遮挡车牌字符识别方法S103的具体实现流程图；
图4是本申请实施例提供的遮挡车牌字符识别方法中对样本遮挡车牌图像的验证的具体实现流程图；
图5是本申请实施例提供的遮挡车牌字符识别装置的结构框图；
图6是本申请实施例提供的智能设备的示意图。
本发明的实施方式
以下描述中，为了说明而不是为了限定，提出了诸如特定系统结构、技术之类的具体细节，以便透彻理解本申请实施例。然而，本领域的技术人员应当清楚，在没有这些具体细节的其它实施例中也可以实现本申请。在其它情况中，省略对众所周知的系统、装置、电路以及方法的详细说明，以免不必要的细节妨碍本申请的描述。
应当理解，当在本申请说明书和所附权利要求书中使用时，术语“包括”指示所描述特征、整体、步骤、操作、元素和/或组件的存在，但并不排除一个或多个其它特征、整体、步骤、操作、元素、组件和/或其集合的存在或添加。
还应当理解，在本申请说明书和所附权利要求书中使用的术语“和/或”是指相关联列出的项中的一个或多个的任何组合以及所有可能组合，并且包括这些组合。
如在本申请说明书和所附权利要求书中所使用的那样，术语“如果”可以依据上下文被解释为“当...时”或“一旦”或“响应于确定”或“响应于检测到”。类似地，短语“如果确定”或“如果检测到[所描述条件或事件]”可以依据上下文被解释为意指“一旦确定”或“响应于确定”或“一旦检测到[所描述条件或事件]”或“响应于检测到[所描述条件或事件]”。
另外，在本申请说明书和所附权利要求书的描述中，术语“第一”、“第二”、“第三”等仅用于区分描述，而不能理解为指示或暗示相对重要性。
在本申请说明书中描述的参考“一个实施例”或“一些实施例”等意味着在本申请的一个或多个实施例中包括结合该实施例描述的特定特征、结构或特点。由此，在本说明书中的不同之处出现的语句“在一个实施例中”、“在一些实施例中”、“在其他一些实施例中”、“在另外一些实施例中”等不是必然都参考相同的实施例，而是意味着“一个或多个但不是所有的实施例”，除非是以其他方式另外特别强调。术语“包括”、“包含”、“具有”及它们的变形都意味着“包括但不限于”，除非是以其他方式另外特别强调。
本申请实施例提供的一种遮挡车牌字符识别方法可以应用于服务器、平板电脑、笔记本电脑、超级移动个人计算机(ultra-mobile personal computer，UMPC)、上网本、等智能设备上，本申请实施例对智能设备的具体类型不作任何限制。
本申请提出一种遮挡车牌字符识别方法，具体涉及深度学习技术，能够在利用丰富的训练样本对深度学习模型进行训练，从而提高该深度学习模型对遮挡车牌字符识别的精度和效率。
图1示出了本申请实施例提供的遮挡车牌字符识别方法的实现流程，该方法流程包括步骤S101至S105。各步骤的具体实现原理如下：
S101：获取样本车牌图像以及样本扩充策略。
在本申请实施例中，所述样本扩充策略为预设的用于构造遮挡样本车牌图像的策略，用户可自行选择已有的样本扩充策略或者修改已有的样本扩充策略。所述车样本扩充策略包括样本遮挡车牌的遮挡类型及该遮挡类型的遮挡样本图像的比例数据。所述遮挡类型是指车牌图像上被遮挡的类型，包括全遮挡、半遮挡以及叠加遮挡。
可选地，作为本申请的一个实施例，图2示出了本申请实施例提供的遮挡车牌字符识别方法S101的具体实现流程，详述如下：
A1：从车辆图像集合中，随机抽取车辆图像。利用随机抽取算法，从所述车辆图像集合中抽取指定数量的车辆图像。所述车辆图像集合为已有的存储有车辆图像的集合。
A2：从抽取的所述车辆图像中提取车牌图像作为样本车牌图像。具体地，所述车牌图像集合中包括已经标注的车牌数据，所述车牌数据包括JPEG图片以及JSON文件。在本申请实施例中，通过解析JSON文件获取车牌位置信息，提取JSON文件中的第一指定字段，如finename字段，即可获取对应的车辆图片名称，提取JSON文件中的第二指定字段，如annotation字段，获取车牌图像的坐标信息，所述车牌图像的坐标信息用于标识所述车牌图像在所述车辆图片中的位置。根据所述车辆图片名称，查找车辆图片，根据所述车牌图像的坐标信息从所述车辆图片中提取车牌图像。
和/或，
A3：从获取的道路监控视频中提取车辆图像。具体地，实时或者周期连接道路监控系统，获取道路监控视频，从获取的道路监控视频中提取车辆图像。
A4：利用预先训练好的车牌检测模型对所述车辆图像进行检测，获取车牌图像。所述车牌检测模型为用于检测车辆图像中车牌图像的模型。在本申请实施例中，采用TextBoxes++模型作为所述车牌检测模型的基础模型，所述TextBoxes++模型为基于ssd框架的图文检测模型。原始的TextBoxes++模型采用VGG16网络作为基础网络，在本实施例，采用Resnet作为所述TextBoxes++模型的基础网络，并增加多层卷积层，构建包括多个卷积层、采样层的车牌检测模型。具体地，获取一定数量的已标注车牌坐标的样本车辆图像，使用样本车辆图像对所述车牌检测模型的训练，利用Resnet-34基础网络卷积层对输入的所述样本车辆图像进行特征提取，采样层采用多尺度特征采样，针对不同大小粒度的特征图提取目标候选框，也即定位检测所述车辆图像中的样本车牌图像，使用极大抑制提取样本车牌图像的坐标，直到输出的样本车牌图像及其车牌坐标与所述样本车辆数据标注的样本车牌图像及其车牌坐标的误差在指定误差范围内，得到训练好的车牌检测模型，利用预先训练好的车牌检测模型对所述车牌图像进行定位检测，获取车牌图像。
A5：将获取的所述车牌图像作为样本车牌图像。在本申请实施例中，交通系统中每日都可以产生许多的车辆图像，获取交通系统监控道路的车辆视频之中的车辆图像，对所述车辆图像进行去噪后输入至所述车牌检测模型，输出得到车牌图像以及车牌坐标，从而可获取大量的车牌图像作为样本车牌图像。
S102：基于所述样本扩充策略，从预设的样本遮挡掩码图像集中抽取遮挡样本掩码图像。
具体地，在本申请实施例中，所述预设的遮挡样本掩码图像是用于遮挡的图像，种类 有多种，例如，光盘图像、手套图像、抹布图像以及纸巾图像。所述预设的遮挡样本掩码图像可从数据库中已有的遮挡样本图像进行分离提取的到。所述样本扩充策略包括遮挡类型以及所述遮挡类型对应的样本遮挡车牌图像的比例数据，基于所遮挡类型以及所述遮挡类型对应的样本遮挡车牌图像的比例数据，从所述遮挡类型对应的样本遮挡掩码图像集中，抽取与所述比例数据对应数量的遮挡样本掩码图像。
S103：利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集。
具体地，构建的样本遮挡车牌图像集中包括利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡构造的遮挡样本车牌图像。其中，采用随机算法抽取样本遮挡掩码图像。
可选地，由于用于提取车牌图像的车辆图像是在自然场景下拍摄的车辆图像，拍摄的车牌是经过投影变换后的形变四边形，如果直接对该车牌图像进行遮挡，容易造成遮挡不准确，构造的遮挡车牌的车牌数据质量较差。
为提高遮挡样本图像的质量，在本申请实施例中还包括对样本车牌图像进行校正，具体包括：根据所述车牌坐标与预设校正算法，对所述样本车牌图像进行图像校正，获取校正后的标准样本车牌图像。具体地，利用如下变换公式(1)，获取校正后的矩形样本车牌图像的车牌坐标：
Figure PCTCN2020135510-appb-000001
其中，
Figure PCTCN2020135510-appb-000002
为变换矩阵，变换矩阵可分为4个部分，
Figure PCTCN2020135510-appb-000003
表示线性变换，[a 31a 32]用于平移，[a 31a 32] T产生透视变换，a 33为恒定值1，[x′,y′,w′]为进行校正转换之后的车牌坐标，[u,v,w]为进行校正之前的样本车牌图像的车牌坐标，所述样本车牌图像为二维图像，w′为恒定值，再根据下式(2)、(3)得到所述样本车牌图像进行校正之后的车牌坐标：
Figure PCTCN2020135510-appb-000004
Figure PCTCN2020135510-appb-000005
在本申请实施例中，通过对样本车牌图像进行校正，将形变四边形的样本车牌图像校正为矩形的样本车牌图像，方便之后构造遮挡样本图像，提高遮挡样本图像的构造效率和质量。
在对所述样本车牌图像进行校正之后，上述步骤S103具体包括：利用抽取的样本遮挡掩码图像对所述标准样本车牌图像进行遮挡，构建样本遮挡车牌图像集。
作为本申请的一个实施例，图3示出了本申请实施例提供的遮挡车牌字符识别方法S103的具体实现流程，详述如下：
B1：将所述样本车牌图像与所述样本遮挡掩码图像进行随机组合。具体地，将样本车牌图像与随机选择的样本遮挡样图像进行组合，同一组合中包括一个样本车牌图像和至少一个样本遮挡掩码图像，即在一个组合中，一个样本车牌图像与一个样本遮挡掩码图像组合，或者一个样本车牌图像与多个样本遮挡掩码图像组合。
B2：将每一组合中的样本车牌图像与样本遮挡掩码图像进行图像融合，生成多个样本遮挡车牌图像。具体地，根据所述样本车牌图像的车牌坐标以及遮挡类型，生成遮挡图像坐标，根据所述遮挡图像坐标将所述遮挡掩码图像的像素覆盖至所述样本车牌图像，进行图像像素覆盖操作，实现车牌遮挡效果。进一步地，若所述遮挡类型为半遮挡，选择一个校正后的样本车牌图像与一个样本遮挡掩码图像进行组合；若所述遮挡类型为叠加遮挡， 则选择一个校正的样本车牌图像与至少两个样本遮挡掩码图像进行组合。
B3：基于构建的样本遮挡车牌图像，生成样本遮挡车牌图像集。
本申请实施例中，通过图像像素覆盖操作，利用样本遮挡掩码图像对样本车牌图进行遮挡，构造样本遮挡车牌图像，由于样本遮挡掩码图像、样本车牌图像以及遮挡类型可以任意组合，从而可自动构造大量的样本遮挡车牌图像。
S104：基于所述样本遮挡车牌图像集与所述样本车牌图像构成的样本车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练。
在本申请实施例中，在上述步骤S103中，将样本遮挡掩码图像、样本车牌图像以及遮挡类型任意组合，构建了包括大量的样本遮挡车牌图像的样本遮挡车牌图像集，将所述样本遮挡车牌图像集与所述样本车牌图像构成的样本车牌图像集作为输入，对用于遮挡车牌字符识别的深度学习网络模型进行训练，大大丰富了所述深度学习网络模型的训练样本，从而可增强所述深度学习网络模型的训练效果。
作为本申请的一个实施例，在上述步骤S104之前，所述遮挡车牌字符识别方法还包括对构建的所述样本遮挡车牌图像集中的样本遮挡车牌图像进行验证。图4示出了本申请实施例提供的遮挡车牌字符识别方法中对样本遮挡车牌图像的验证的具体实现流程，详述如下：
C1：将所述述样本遮挡车牌图像集中的样本遮挡车牌图像作为预先训练好的车牌遮挡识别模型的输入，所述车牌遮挡识别模型用于识别车牌是否被遮挡。具体地，所述车牌遮挡识别模型是指用于识别车牌的模型，在本申请实施例中，所述车牌遮挡识别模型的训练步骤，包括：
c1：构建车牌遮挡识别模型，所述车牌遮挡识别模型包括卷积层、循环神经网络层；
c2：将已标注的正负样本车牌图像作为所述车牌遮挡识别模型的输入，其中，正样本车牌图像为未遮挡的正常车牌图像，负样本车牌图像为遮挡车牌图像；
c3：在所述卷积层中进行特征提取，得到特征序列；
c4：将所述特征序列输入至所述循环神经网络层中进行遮挡识别，输出遮挡识别结果；
c5：根据识别结果与误差函数，调整所述车牌遮挡识别模型的模型参数至最优，直至所述识别结果与所述样本遮挡车牌图像的标注一致。
具体地，在本申请实施例中，所述车牌遮挡识别模型采用CRNN卷积循环神经网络算法，所述车牌遮挡识别模型的训练过程包括将已标注的正负样本车牌图像作为输入，其中，正样本车牌图像为未遮挡的正常车牌图像，负样本车牌图像为遮挡车牌图像，经过卷积层进行特征提取，在最后一层逐列拼接，得到特征序列，把得到的特征序列放入循环神经网络层中进行遮挡识别，最后通过转译层将各重复字符进行合并后输出遮挡识别结果，根据识别结果与误差函数，调整所述车牌遮挡识别模型的模型参数至最优。
C2：根据所述车牌遮挡识别模型的输出结果，验证所述样本遮挡车牌图像是否为有效遮挡的样本车牌图像。在验证过程中，将所述述样本遮挡车牌图像集中的样本遮挡车牌图像作为所述车牌遮挡识别模型的输入，根据所述车牌遮挡识别模型的识别结果，验证所述样本遮挡车牌图像是否遮挡有效。本实施例中，由于遮挡车牌存在字符缺失，可根据识别的字符字段是否完整判断车牌遮挡是否有效，若识别的字符字段完整，则确定该车牌遮挡图像中的车牌遮挡无效；若识别的字符字段不完整，则确定该车牌遮挡图像中的车牌遮挡有效；
C3：根据验证为有效遮挡的样本车牌图像，构建目标样本遮挡车牌图像集。具体地，标记验证为遮挡无效的样本遮挡车牌图像，将遮挡无效的样本遮挡车牌图像从所述样本遮挡车牌图像集中删除，进而构建包含验证为遮挡有效的样本遮挡车牌图像的目标样本遮挡车牌图像集。
此时，所述基于所述样本遮挡车牌图像集与包括样本车牌图像的车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练的步骤，包括：
基于所述目标样本遮挡车牌图像集与包括样本车牌图像的车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练。
在本申请实施例中，为了保证所述样本遮挡车牌图像的中的遮挡车牌样本被遮挡的部分缺失遮挡有效，对所述样本遮挡车牌图像集中的样本遮挡车牌图像进行验证，将覆盖了样本遮挡掩码图像但依然能识别出车牌字符的遮挡车牌样本剔除，从而提高构造的样本数据的有效性。
S105：利用训练好的所述深度学习网络模型进行遮挡车牌字符识别。
在本申请实施例中，具体地，将待识别的遮挡车牌图像作为训练好的所述深度学习网络模型中进行遮挡车牌字符识别。
本申请实施例中，通过获取样本车牌图像以及样本扩充策略，基于所述样本扩充策略，从预设的样本遮挡掩码图像集中抽取遮挡样本掩码图像，然后利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集，实现对已有的样本遮挡图像进行扩充，有效增加样本遮挡图像的数量和种类，再基于所述样本遮挡车牌图像集与所述样本车牌图像构成的样本车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练，丰富的样本车牌图像集可增强用于遮挡车牌字符识别的深度学习网络模型的训练效果，最后利用训练好的所述深度学习网络模型进行遮挡车牌字符识别，从而可提高遮挡车牌字符识别的精度和效率。
应理解，上述实施例中各步骤的序号的大小并不意味着执行顺序的先后，各过程的执行顺序应以其功能和内在逻辑确定，而不应对本申请实施例的实施过程构成任何限定。
对应于上文实施例所述的遮挡车牌字符识别方法，图5示出了本申请实施例提供的遮挡车牌字符识别装置的结构框图，为了便于说明，仅示出了与本申请实施例相关的部分。
参照图5，该遮挡车牌字符识别装置包括：图像及策略获取单元51，掩码图像抽取单元52，遮挡图像构建单元53，深度学习模型训练单元54，遮挡识别单元55，其中：
图像及策略获取单元51，用于获取样本车牌图像以及样本扩充策略；
掩码图像抽取单元52，用于基于所述样本扩充策略，从预设的样本遮挡掩码图像集中抽取遮挡样本掩码图像；
遮挡图像构建单元53，用于利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集；
深度学习模型训练单元54，用于基于所述样本遮挡车牌图像集与所述样本车牌图像构成的样本车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练；
遮挡识别单元55，用于利用训练好的所述深度学习网络模型进行遮挡车牌字符识别。
可选地，所述图像及策略获取单元包括：
车辆图像抽取模块，用于从车辆图像集合中，随机抽取车辆图像；
第一样本车牌图像获取模块，用于从抽取的所述车辆图像中提取车牌图像作为样本车牌图像；
车辆图像提取模块，用于从获取的道路监控视频中提取车辆图像；
车牌图像检测模块，用于利用预先训练好的车牌检测模型对所述车辆图像进行检测，获取车牌图像；
第二样本车牌图像获取模块，用于将获取的所述车牌图像作为样本车牌图像。
可选地，所述遮挡车牌字符识别装置还包括：
图像校正单元，用于根据所述车牌坐标与预设校正算法，对所述样本车牌图像进行图像校正，获取校正后的标准样本车牌图像；
此时，所述遮挡图像构建单元还用于：
利用抽取的样本遮挡掩码图像对所述标准样本车牌图像进行遮挡，构建样本遮挡车牌图像集。
可选地，所述遮挡图像构建单元包括：
图像组合模块，用于将所述样本车牌图像与所述样本遮挡掩码图像进行随机组合；
图像融合模块，用于将每一组合中的样本车牌图像与样本遮挡掩码图像进行图像融合，生成多个样本遮挡车牌图像；
遮挡图像构建模块，用于基于构建的样本遮挡车牌图像，生成样本遮挡车牌图像集。
可选地，所述遮挡车牌字符识别装置还包括：
验证模型输入确定单元，用于将所述述样本遮挡车牌图像集中的样本遮挡车牌图像作为预先训练好的车牌遮挡识别模型的输入，所述车牌遮挡识别模型用于识别车牌是否被遮挡；
遮挡验证单元，用于根据所述车牌遮挡识别模型的输出结果，验证所述样本遮挡车牌图像是否为有效遮挡的样本车牌图像；
目标图像集构建单元，用于根据验证为有效遮挡的样本车牌图像，构建目标样本遮挡车牌图像集；
所述深度学习模型训练单元54，还用于基于所述目标样本遮挡车牌图像集与包括样本车牌图像的车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练。
可选地，所述遮挡车牌字符识别装置还包括：
遮挡识别模型构建单元，用于构建车牌遮挡识别模型，所述车牌遮挡识别模型包括卷积层、循环神经网络层；
样本输入确定单元，用于将已标注的正负样本车牌图像作为所述车牌遮挡识别模型的输入，其中，正样本车牌图像为未遮挡的正常车牌图像，负样本车牌图像为遮挡车牌图像；
特征提取单元，用于在所述卷积层中进行特征提取，得到特征序列；
字符识别单元，用于将所述特征序列输入至所述循环神经网络层中进行遮挡识别，输出遮挡识别结果；
模型优化单元，用于根据识别结果与误差函数，调整所述车牌遮挡识别模型的模型参数至最优。
本申请实施例中，通过获取样本车牌图像以及样本扩充策略，基于所述样本扩充策略，从预设的样本遮挡掩码图像集中抽取遮挡样本掩码图像，然后利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集，实现对已有的样本遮挡图像进行扩充，有效增加样本遮挡图像的数量和种类，再基于所述样本遮挡车牌图像集与所述样本车牌图像构成的样本车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练，丰富的样本车牌图像集可增强用于遮挡车牌字符识别的深度学习网络模型的训练效果，最后利用训练好的所述深度学习网络模型进行遮挡车牌字符识别，从而可提高遮挡车牌字符识别的精度和效率。
本申请实施例还提供一种计算机可读存储介质，所述计算机可读存储介质存储有计算机可读指令，所述计算机可读指令被处理器执行时实现如图1至图4表示的任意一种遮挡车牌字符识别方法的步骤。
本申请实施例还提供一种计算机可读指令产品，当该计算机可读指令产品在智能设备上运行时，使得智能设备执行实现如图1至图4表示的任意一种遮挡车牌字符识别方法的步骤。
本申请实施例还提供一种智能设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令，所述处理器执行所述计算机可读指令时实现如图1至图4表示的任意一种遮挡车牌字符识别方法的步骤。
图6是本申请一实施例提供的智能设备的示意图。如图6所示，该实施例的智能设备6包括：处理器60、存储器61以及存储在所述存储器61中并可在所述处理器60上运行的计算机可读指令62。所述处理器60执行所述计算机可读指令62时实现上述各个遮挡车牌字符识别方法实施例中的步骤，例如图1所示的步骤101至105。或者，所述处理器60执行所述计算机可读指令62时实现上述各装置实施例中各模块/单元的功能，例如图5所示 单元51至55的功能。
示例性的，所述计算机可读指令62可以被分割成一个或多个模块/单元，所述一个或者多个模块/单元被存储在所述存储器61中，并由所述处理器60执行，以完成本申请。所述一个或多个模块/单元可以是能够完成特定功能的一系列计算机可读指令段，该指令段用于描述所述计算机可读指令62在所述智能设备6中的执行过程。
所述智能设备6可以是服务器及云端智能设备等计算设备。所述智能设备6可包括，但不仅限于，处理器60、存储器61。本领域技术人员可以理解，图6仅仅是智能设备6的示例，并不构成对智能设备6的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件，例如所述智能设备6还可以包括输入输出设备、网络接入设备、总线等。
所述处理器60可以是中央处理单元(Central Processing Unit，CPU)，还可以是其他通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现成可编程门阵列(Field-Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
所述存储器61可以是所述智能设备6的内部存储单元，例如智能设备6的硬盘或内存。所述存储器61也可以是所述智能设备6的外部存储设备，例如所述智能设备6上配备的插接式硬盘，智能存储卡(Smart Media Card,SMC)，安全数字(Secure Digital,SD)卡，闪存卡(Flash Card)等。进一步地，所述存储器61还可以既包括所述智能设备6的内部存储单元也包括外部存储设备。所述存储器61用于存储所述计算机可读指令以及所述智能设备所需的其他程序和数据。所述存储器61还可以用于暂时地存储已经输出或者将要输出的数据。
需要说明的是，上述装置/单元之间的信息交互、执行过程等内容，由于与本申请方法实施例基于同一构思，其具体功能及带来的技术效果，具体可参见方法实施例部分，此处不再赘述。
所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，仅以上述各功能单元、模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能单元、模块完成，即将所述装置的内部结构划分成不同的功能单元或模块，以完成以上描述的全部或者部分功能。实施例中的各功能单元、模块可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中，上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。另外，各功能单元、模块的具体名称也只是为了便于相互区分，并不用于限制本申请的保护范围。上述系统中单元、模块的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中，所述计算机可读存储介质可以是非易失性，也可以是易失性。基于这样的理解，本申请实现上述实施例方法中的全部或部分流程，可以通过计算机可读指令来指令相关的硬件来完成，所述的计算机可读指令可存储于一计算机可读存储介质中，该计算机可读指令在被处理器执行时，可实现上述各个方法实施例的步骤。其中，所述计算机可读指令包括计算机可读指令代码，所述计算机可读指令代码可以为源代码形式、对象代码形式、可执行文件或某些中间形式等。所述计算机可读介质至少可以包括：能够将计算机可读指令代码携带到装置/终端设备的任何实体或装置、记录介质、计算机存储器、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、电载波信号、电信信号以及软件分发介质。例如U盘、移动硬盘、磁碟或者光盘等。在某些司法管辖区，根据立法和专利实践，计算机可读介质不可以是电载波信号和电信信号。
在上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述或记载的部分，可以参见其它实施例的相关描述。
以上所述实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围，均应包含在本申请的保护范围之内。

Claims (20)

  1. 一种遮挡车牌字符识别方法，其中，包括：
    获取样本车牌图像以及样本扩充策略；
    基于所述样本扩充策略，从预设的样本遮挡掩码图像集中抽取遮挡样本掩码图像；
    利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集；
    基于所述样本遮挡车牌图像集与所述样本车牌图像构成的样本车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练；
    利用训练好的所述深度学习网络模型进行遮挡车牌字符识别。
  2. 根据权利要求1所述的遮挡车牌字符识别方法，其中，所述获取用于构造遮挡样本车牌图像的样本车牌图像的步骤，包括：
    从车辆图像集合中，随机抽取车辆图像；
    从抽取的所述车辆图像中提取车牌图像作为样本车牌图像；
    和/或，
    从获取的道路监控视频中提取车辆图像；
    利用预先训练好的车牌检测模型对所述车辆图像进行检测，获取车牌图像；
    将获取的所述车牌图像作为样本车牌图像。
  3. 根据权利要求1所述的遮挡车牌字符识别方法，其中，所述车牌数据包括车牌坐标，在所述利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集的步骤之前，还包括：
    根据所述车牌坐标与预设校正算法，对所述样本车牌图像进行图像校正，获取校正后的标准样本车牌图像；
    此时，所述利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集的步骤，包括：
    利用抽取的样本遮挡掩码图像对所述标准样本车牌图像进行遮挡，构建样本遮挡车牌图像集。
  4. 根据权利要求1所述的遮挡车牌字符识别方法，其中，所述利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集的步骤，包括：
    将所述样本车牌图像与所述样本遮挡掩码图像进行随机组合；
    将每一组合中的样本车牌图像与样本遮挡掩码图像进行图像融合，生成多个样本遮挡车牌图像；
    基于构建的样本遮挡车牌图像，生成样本遮挡车牌图像集。
  5. 根据权利要求1所述的遮挡车牌字符识别方法，其中，在所述基于所述样本遮挡车牌图像集与包括样本车牌图像的车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练的步骤之前，还包括：
    将所述样本遮挡车牌图像集中的样本遮挡车牌图像作为预先训练好的车牌遮挡识别模型的输入，所述车牌遮挡识别模型用于识别车牌是否被遮挡；
    根据所述车牌遮挡识别模型的输出结果，验证所述样本遮挡车牌图像是否为有效遮挡的样本车牌图像；
    根据验证为有效遮挡的样本车牌图像，构建目标样本遮挡车牌图像集；
    所述基于所述样本遮挡车牌图像集与包括样本车牌图像的车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练的步骤，包括：
    基于所述目标样本遮挡车牌图像集与包括样本车牌图像的车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练。
  6. 根据权利要求5所述的遮挡车牌字符识别方法，其中，所述车牌遮挡识别模型 的训练步骤，包括：
    构建车牌遮挡识别模型，所述车牌遮挡识别模型包括卷积层、循环神经网络层；
    将已标注的正负样本车牌图像作为所述车牌遮挡识别模型的输入，其中，正样本车牌图像为未遮挡的正常车牌图像，负样本车牌图像为遮挡车牌图像；
    在所述卷积层中进行特征提取，得到特征序列；
    将所述特征序列输入至所述循环神经网络层中进行遮挡识别，输出遮挡识别结果；
    根据识别结果与误差函数，调整所述车牌遮挡识别模型的模型参数至最优。
  7. 一种智能设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令，其中，所述处理器执行所述计算机可读指令时实现如下步骤：
    获取样本车牌图像以及样本扩充策略；
    基于所述样本扩充策略，从预设的样本遮挡掩码图像集中抽取遮挡样本掩码图像；
    利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集；
    基于所述样本遮挡车牌图像集与所述样本车牌图像构成的样本车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练；
    利用训练好的所述深度学习网络模型进行遮挡车牌字符识别。
  8. 根据权利要求7所述的智能设备，其中，所述获取用于构造遮挡样本车牌图像的样本车牌图像的步骤，包括：
    从车辆图像集合中，随机抽取车辆图像；
    从抽取的所述车辆图像中提取车牌图像作为样本车牌图像；
    和/或，
    从获取的道路监控视频中提取车辆图像；
    利用预先训练好的车牌检测模型对所述车辆图像进行检测，获取车牌图像；
    将获取的所述车牌图像作为样本车牌图像。
  9. 根据权利要求7所述的智能设备，其中，所述车牌数据包括车牌坐标，在所述利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集的步骤之前，还包括：
    根据所述车牌坐标与预设校正算法，对所述样本车牌图像进行图像校正，获取校正后的标准样本车牌图像；
    此时，所述利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集的步骤，包括：
    利用抽取的样本遮挡掩码图像对所述标准样本车牌图像进行遮挡，构建样本遮挡车牌图像集。
  10. 根据权利要求7所述的智能设备，其中，所述利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集的步骤，包括：
    将所述样本车牌图像与所述样本遮挡掩码图像进行随机组合；
    将每一组合中的样本车牌图像与样本遮挡掩码图像进行图像融合，生成多个样本遮挡车牌图像；
    基于构建的样本遮挡车牌图像，生成样本遮挡车牌图像集。
  11. 根据权利要求7所述的智能设备，其中，在所述基于所述样本遮挡车牌图像集与包括样本车牌图像的车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练的步骤之前，还包括：
    将所述样本遮挡车牌图像集中的样本遮挡车牌图像作为预先训练好的车牌遮挡识别模型的输入，所述车牌遮挡识别模型用于识别车牌是否被遮挡；
    根据所述车牌遮挡识别模型的输出结果，验证所述样本遮挡车牌图像是否为有效 遮挡的样本车牌图像；
    根据验证为有效遮挡的样本车牌图像，构建目标样本遮挡车牌图像集；
    所述基于所述样本遮挡车牌图像集与包括样本车牌图像的车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练的步骤，包括：
    基于所述目标样本遮挡车牌图像集与包括样本车牌图像的车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练。
  12. 根据权利要求11所述的智能设备，其中，所述车牌遮挡识别模型的训练步骤，包括：
    构建车牌遮挡识别模型，所述车牌遮挡识别模型包括卷积层、循环神经网络层；
    将已标注的正负样本车牌图像作为所述车牌遮挡识别模型的输入，其中，正样本车牌图像为未遮挡的正常车牌图像，负样本车牌图像为遮挡车牌图像；
    在所述卷积层中进行特征提取，得到特征序列；
    将所述特征序列输入至所述循环神经网络层中进行遮挡识别，输出遮挡识别结果；
    根据识别结果与误差函数，调整所述车牌遮挡识别模型的模型参数至最优。
  13. 一种计算机可读存储介质，所述计算机可读存储介质存储有计算机可读指令，其中，所述计算机可读指令被处理器执行时实现如下步骤：
    获取样本车牌图像以及样本扩充策略；
    基于所述样本扩充策略，从预设的样本遮挡掩码图像集中抽取遮挡样本掩码图像；
    利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集；
    基于所述样本遮挡车牌图像集与所述样本车牌图像构成的样本车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练；
    利用训练好的所述深度学习网络模型进行遮挡车牌字符识别。
  14. 根据权利要求13所述的计算机可读存储介质，其中，所述获取用于构造遮挡样本车牌图像的样本车牌图像的步骤，包括：
    从车辆图像集合中，随机抽取车辆图像；
    从抽取的所述车辆图像中提取车牌图像作为样本车牌图像；
    和/或，
    从获取的道路监控视频中提取车辆图像；
    利用预先训练好的车牌检测模型对所述车辆图像进行检测，获取车牌图像；
    将获取的所述车牌图像作为样本车牌图像。
  15. 根据权利要求13所述的计算机可读存储介质，其中，所述车牌数据包括车牌坐标，在所述利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集的步骤之前，还包括：
    根据所述车牌坐标与预设校正算法，对所述样本车牌图像进行图像校正，获取校正后的标准样本车牌图像；
    此时，所述利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集的步骤，包括：
    利用抽取的样本遮挡掩码图像对所述标准样本车牌图像进行遮挡，构建样本遮挡车牌图像集。
  16. 根据权利要求13所述的计算机可读存储介质，其中，所述利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集的步骤，包括：
    将所述样本车牌图像与所述样本遮挡掩码图像进行随机组合；
    将每一组合中的样本车牌图像与样本遮挡掩码图像进行图像融合，生成多个样本遮挡车牌图像；
    基于构建的样本遮挡车牌图像，生成样本遮挡车牌图像集。
  17. 根据权利要求13所述的计算机可读存储介质，其中，在所述基于所述样本遮挡车牌图像集与包括样本车牌图像的车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练的步骤之前，还包括：
    将所述样本遮挡车牌图像集中的样本遮挡车牌图像作为预先训练好的车牌遮挡识别模型的输入，所述车牌遮挡识别模型用于识别车牌是否被遮挡；
    根据所述车牌遮挡识别模型的输出结果，验证所述样本遮挡车牌图像是否为有效遮挡的样本车牌图像；
    根据验证为有效遮挡的样本车牌图像，构建目标样本遮挡车牌图像集；
    所述基于所述样本遮挡车牌图像集与包括样本车牌图像的车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练的步骤，包括：
    基于所述目标样本遮挡车牌图像集与包括样本车牌图像的车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练。
  18. 根据权利要求13所述的计算机可读存储介质，其中，所述车牌遮挡识别模型的训练步骤，包括：
    构建车牌遮挡识别模型，所述车牌遮挡识别模型包括卷积层、循环神经网络层；
    将已标注的正负样本车牌图像作为所述车牌遮挡识别模型的输入，其中，正样本车牌图像为未遮挡的正常车牌图像，负样本车牌图像为遮挡车牌图像；
    在所述卷积层中进行特征提取，得到特征序列；
    将所述特征序列输入至所述循环神经网络层中进行遮挡识别，输出遮挡识别结果；
    根据识别结果与误差函数，调整所述车牌遮挡识别模型的模型参数至最优。
  19. 一种遮挡车牌字符识别装置，其中，包括：
    图像及策略获取单元，用于获取样本车牌图像以及样本扩充策略；
    掩码图像抽取单元，用于基于所述样本扩充策略，从预设的样本遮挡掩码图像集中抽取遮挡样本掩码图像；
    遮挡图像构建单元，用于利用抽取的样本遮挡掩码图像对所述样本车牌图像进行遮挡，构建样本遮挡车牌图像集；
    深度学习模型训练单元，用于基于所述样本遮挡车牌图像集与所述样本车牌图像构成的样本车牌图像集，对用于遮挡车牌字符识别的深度学习网络模型进行训练；
    遮挡识别单元，用于利用训练好的所述深度学习网络模型进行遮挡车牌字符识别。
  20. 根据权利要求19所述的遮挡车牌字符识别装置，其中，所述遮挡图像构建单元包括：
    图像组合模块，用于将所述样本车牌图像与所述样本遮挡掩码图像进行随机组合；
    图像融合模块，用于将每一组合中的样本车牌图像与样本遮挡掩码图像进行图像融合，生成多个样本遮挡车牌图像；
    遮挡图像构建模块，用于基于构建的样本遮挡车牌图像，生成样本遮挡车牌图像集。
PCT/CN2020/135510 2020-03-17 2020-12-11 一种遮挡车牌字符识别方法、装置、存储介质和智能设备 WO2021184847A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010185281.8A CN111476234B (zh) 2020-03-17 2020-03-17 一种遮挡车牌字符识别方法、装置、存储介质和智能设备
CN202010185281.8 2020-03-17

Publications (1)

Publication Number Publication Date
WO2021184847A1 true WO2021184847A1 (zh) 2021-09-23

Family

ID=71747630

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/135510 WO2021184847A1 (zh) 2020-03-17 2020-12-11 一种遮挡车牌字符识别方法、装置、存储介质和智能设备

Country Status (2)

Country Link
CN (1) CN111476234B (zh)
WO (1) WO2021184847A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113657354A (zh) * 2021-10-19 2021-11-16 深圳市菁优智慧教育股份有限公司 基于深度学习的答题卡识别方法及系统
CN115063757A (zh) * 2022-06-24 2022-09-16 杭州鸿泉物联网技术股份有限公司 车辆落水识别方法和装置
CN116503614A (zh) * 2023-04-27 2023-07-28 杭州食方科技有限公司 餐盘形状特征提取网络训练方法和餐盘形状信息生成方法

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111476234B (zh) * 2020-03-17 2024-05-24 平安科技（深圳）有限公司 一种遮挡车牌字符识别方法、装置、存储介质和智能设备
CN112149542A (zh) * 2020-09-15 2020-12-29 北京字节跳动网络技术有限公司 训练样本生成方法、图像分类方法、装置、设备和介质
CN112488204A (zh) * 2020-11-30 2021-03-12 北京字跳网络技术有限公司 训练样本生成方法、图像分割方法、装置、设备和介质
CN113052174B (zh) * 2021-03-26 2023-08-11 北京百度网讯科技有限公司 车牌数据样本生成方法、装置、电子设备和存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170249524A1 (en) * 2016-02-25 2017-08-31 Xerox Corporation Method and system for detection-based segmentation-free license plate recognition
CN109670528A (zh) * 2018-11-14 2019-04-23 中国矿业大学 面向行人重识别任务的基于成对样本随机遮挡策略的数据扩充方法
CN109753914A (zh) * 2018-12-28 2019-05-14 安徽清新互联信息科技有限公司 一种基于深度学习的车牌字符识别方法
CN109902577A (zh) * 2019-01-25 2019-06-18 华中科技大学 一种轻量级手势检测卷积神经网络模型的构建方法及应用
CN111476234A (zh) * 2020-03-17 2020-07-31 平安科技（深圳）有限公司 一种遮挡车牌字符识别方法、装置、存储介质和智能设备

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108108746B (zh) * 2017-09-13 2021-04-09 湖南理工学院 基于Caffe深度学习框架的车牌字符识别方法
WO2019169532A1 (zh) * 2018-03-05 2019-09-12 深圳前海达闼云端智能科技有限公司 车牌识别方法及云系统
JP7219023B2 (ja) * 2018-06-22 2023-02-07 日立造船株式会社 情報処理装置および対象物判定プログラム
CN110287960B (zh) * 2019-07-02 2021-12-10 中国科学院信息工程研究所 自然场景图像中曲线文字的检测识别方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170249524A1 (en) * 2016-02-25 2017-08-31 Xerox Corporation Method and system for detection-based segmentation-free license plate recognition
CN109670528A (zh) * 2018-11-14 2019-04-23 中国矿业大学 面向行人重识别任务的基于成对样本随机遮挡策略的数据扩充方法
CN109753914A (zh) * 2018-12-28 2019-05-14 安徽清新互联信息科技有限公司 一种基于深度学习的车牌字符识别方法
CN109902577A (zh) * 2019-01-25 2019-06-18 华中科技大学 一种轻量级手势检测卷积神经网络模型的构建方法及应用
CN111476234A (zh) * 2020-03-17 2020-07-31 平安科技（深圳）有限公司 一种遮挡车牌字符识别方法、装置、存储介质和智能设备

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113657354A (zh) * 2021-10-19 2021-11-16 深圳市菁优智慧教育股份有限公司 基于深度学习的答题卡识别方法及系统
CN115063757A (zh) * 2022-06-24 2022-09-16 杭州鸿泉物联网技术股份有限公司 车辆落水识别方法和装置
CN116503614A (zh) * 2023-04-27 2023-07-28 杭州食方科技有限公司 餐盘形状特征提取网络训练方法和餐盘形状信息生成方法

Also Published As

Publication number Publication date
CN111476234B (zh) 2024-05-24
CN111476234A (zh) 2020-07-31

Similar Documents

Publication Publication Date Title
WO2021184847A1 (zh) 一种遮挡车牌字符识别方法、装置、存储介质和智能设备
CN110851835B (zh) 图像模型检测方法、装置、电子设备及存储介质
WO2018233038A1 (zh) 基于深度学习的车牌识别方法、装置、设备及存储介质
WO2023082784A1 (zh) 一种基于局部特征注意力的行人重识别方法和装置
CN111461170A (zh) 车辆图像检测方法、装置、计算机设备及存储介质
WO2021151317A1 (zh) 活体检测方法、装置、电子设备及存储介质
CN111191649A (zh) 一种识别弯曲多行文本图像的方法与设备
WO2020146980A1 (zh) 车道线识别方法、车道线识别装置以及非易失性存储介质
CN109472193A (zh) 人脸检测方法及装置
JP2022133378A (ja) 顔生体検出方法、装置、電子機器、及び記憶媒体
CN109948420B (zh) 人脸比对方法、装置及终端设备
CN110378328B (zh) 一种证件图像处理方法及装置
CN113158773B (zh) 一种活体检测模型的训练方法及训练装置
WO2021147437A1 (zh) 证卡边缘检测方法、设备及存储介质
WO2022206680A1 (zh) 图像处理方法、装置、计算机设备和存储介质
CN111753766B (zh) 一种图像处理方法、装置、设备及介质
CN112329602A (zh) 人脸标注图像的获取方法及装置、电子设备、存储介质
CN114758145B (zh) 一种图像脱敏方法、装置、电子设备及存储介质
CN116091781A (zh) 一种用于图像识别的数据处理方法及装置
US20240193987A1 (en) Face liveness detection method, terminal device and non-transitory computer-readable storage medium
CN107862259A (zh) 人像采集方法及装置、终端装置和计算机可读存储介质
CN114359352A (zh) 图像处理方法、装置、设备、存储介质及计算机程序产品
WO2021139169A1 (zh) 卡片识别方法、装置、设备及存储介质
CN116229406B (zh) 车道线检测方法、系统、电子设备及存储介质
WO2020244076A1 (zh) 人脸识别方法、装置、电子设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20926234

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20926234

Country of ref document: EP

Kind code of ref document: A1