CN109145900A - A kind of licence plate recognition method based on deep learning - Google Patents
A kind of licence plate recognition method based on deep learning Download PDFInfo
- Publication number
- CN109145900A CN109145900A CN201810853164.7A CN201810853164A CN109145900A CN 109145900 A CN109145900 A CN 109145900A CN 201810853164 A CN201810853164 A CN 201810853164A CN 109145900 A CN109145900 A CN 109145900A
- Authority
- CN
- China
- Prior art keywords
- license plate
- character
- picture
- convolutional neural
- neural networks
- 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
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/14—Image acquisition
- G06V30/148—Segmentation of character regions
- G06V30/153—Segmentation of character regions using recognition of characters or words
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Abstract
The invention discloses a kind of licence plate recognition methods based on deep learning, including license plate rectification module, Character segmentation module and character recognition module；The first depth convolutional neural networks positioning licence plate coordinate position of license plate rectification module simultaneously operates correction license plate shape by affine transformation；Character segmentation module neural network prediction Character segmentation position is simultaneously cut into single character picture；Character recognition module carries out character classification with the second depth convolutional neural networks.Through the above way, licence plate recognition method based on deep learning of the invention, it is improved and has been optimized around the network structure and parameter of each neural network, greatly reduce the constraint to license plate sloped angle, can tolerate the license plate sloped angle within 50 degree.
Description
Technical field
The present invention relates to Car license recognition fields, more particularly to a kind of licence plate recognition method based on deep learning.
Background technique
With the fast development of transportation industry, infrastructure construction speed lags behind vehicle growth rate, and infrastructure is short
It lacks and the traffic congestion of low, as big and medium-sized cities the universal phenomenon of utilization rate, the traffic safety a series of traffic such as the situation is tense
Problem has to be solved.In order to solve these problems, the whole world is reached an agreement target --- construction intelligent transportation system (ITS), not
It is disconnected to emerge new intelligent transportation system.Wherein automatic Car license recognition (Automatic license plate
Recognition, ALPR) be most basic and crucial one of the core technology of intelligent transportation system, and expressway tol lcollection,
Traffic administration, vehicle monitoring, parking lot fee collection management etc. have produced extensive practical application meaning.
Nowadays, it is increased sharply with China's vehicle guaranteeding organic quantity with astonishing speed, results in disobey and stop phenomenon and spread unchecked.Road
Not only road occupying influences city appearance to side violation parking motor vehicles, but also the traffic environment of urban road is unable to get promotion, hands over
Logical congestion makes one mood agitation, quality of life of the tangible effect to us.Curb parking toll administration is each in recent years
Big city is carried out, and is gradually being developed to intelligent direction, its function mode is mainly held by special toll collector
Terminal device is recorded and is returned to the progress charging of background service program to driving into or out of for vehicle, is driven into record vehicle
Cheng Zhong, license plate number must be recorded as the identity ID of vehicle, this will necessarily introduce automatic license plate recognition technology.
The curb parking toll administration application scenarios new as one, the place of it and previous application scenarios main difference
Be: the randomness that handheld device obtains vehicle photo causes the inclined degree of license plate big.Therefore, directly traditional license plate is known
Other method, which is applied to this, cannot obtain satisfactory recognition effect, and the factor for causing recognition effect bad is exactly inclined degree
The license plate of larger (more than 15 degree).Therefore, license plate correction seems most important.
The research of existing related license plate correction, mainly according to graph image, such as: color, texture etc., but in reality
Find that the limitation of such methods is very big in the application of border.
Summary of the invention
The invention mainly solves the technical problem of providing a kind of licence plate recognition methods based on deep learning, have Gao Lu
Stick and efficient advantage can be applied for the biggish license plate of inclined degree.
In order to solve the above technical problems, one technical scheme adopted by the invention is that: it provides a kind of based on deep learning
Licence plate recognition method, including license plate rectification module, Character segmentation module and character recognition module；
The first depth convolutional neural networks positioning licence plate coordinate position of the license plate rectification module is simultaneously operated by affine transformation
Correct license plate shape；
The Character segmentation module neural network prediction Character segmentation position is simultaneously cut into single character picture；
The character recognition module carries out character classification with the second depth convolutional neural networks.
In a preferred embodiment of the present invention, the coordinate position of the license plate is the coordinate bit on four vertex of license plate
It sets.
In a preferred embodiment of the present invention, the correction license plate process of the license plate rectification module is rolled up including the first depth
The building of product neural network, the training of the first depth convolutional neural networks and affine transformation operation；The first depth convolution mind
Building through network includes building multilayer convolutional network layer and the full articulamentum of multilayer, wherein each convolutional network layer is all made of
Conditional nonlinear activation function ReLU and batch regularization；The training of the first depth convolutional neural networks includes artificial mark
Caravan board coordinate generates training image, inputs training image and the corresponding license plate coordinate of the training image to the first depth is rolled up
Product neural network, training the first depth convolutional neural networks, until the vehicle of the first depth convolutional neural networks output
Error between board coordinate license plate coordinate corresponding with the training image does not continue to the amplitude for becoming smaller or becoming smaller and is less than in advance
If threshold value until；The affine transformation operation is using the license plate coordinate of the first depth convolutional neural networks output as imitative
The parameter of map function is penetrated, the jamming pattern information in input picture is removed, realizes the correction of license plate.
In a preferred embodiment of the present invention, the cutting character process of the Character segmentation module includes neural network
Building, the training of neural network and OpenCV image cutting operation；The building of the neural network includes constructing multiple full connections
Layer, the last one full articulamentum export floating number；The training of the neural network includes manually to the license plate picture mark after correction
Caravan board cutting position, resets the size of the license plate picture, and is converted into grayscale image, then for each grayscale image each
Capable and each column gray value of images, which sum to obtain one, contains 328(256+72) one-dimensional vector of a value, by it is described it is one-dimensional to
The cutting position of amount and the license plate is input to the neural network, until neural network output cutting position and the figure
Error between the true cutting position of piece does not continue to the amplitude for becoming smaller or becoming smaller less than until preset threshold value；It is described
OpenCV image cutting operation includes the floating number that exports the neural network multiplied by the cutting position coordinate of characters on license plate, is cut
Cut the mathematic(al) representation of picture approach are as follows:, wherein Image indicates license plate original image,
Image_i indicates i-th of character picture after cutting.
In a preferred embodiment of the present invention, the character recognition process of the character recognition module is with monocase picture
With picture alphanumeric tag as inputting, it is input to the second depth convolutional neural networks.
The beneficial effects of the present invention are: the licence plate recognition method of the invention based on deep learning, surrounds each nerve net
The network structure and parameter of network are improved and have been optimized, and the constraint to license plate sloped angle is greatly reduced, and can tolerate five
License plate sloped angle within ten degree.
Detailed description of the invention
To describe the technical solutions in the embodiments of the present invention more clearly, make required in being described below to embodiment
Attached drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for
For those of ordinary skill in the art, without creative efforts, it can also be obtained according to these attached drawings other
Attached drawing, in which:
Fig. 1 is the overall flow figure of licence plate recognition method one preferred embodiment of the invention based on deep learning；
Fig. 2 is the structural framing figure of the first depth convolutional neural networks in the licence plate recognition method based on deep learning；
Fig. 3 is that characters on license plate cutting flow chart is predicted in the licence plate recognition method based on deep learning.
Specific embodiment
The technical scheme in the embodiments of the invention will be clearly and completely described below, it is clear that described implementation
Example is only a part of the embodiments of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, this field is common
Technical staff's all other embodiment obtained without making creative work belongs to the model that the present invention protects
It encloses.
Fig. 1, Fig. 2 and Fig. 3 are please referred to, a kind of licence plate recognition method based on deep learning is provided, including license plate corrects mould
Block, Character segmentation module and character recognition module.Fig. 1 shows the licence plate recognition method of the present invention based on deep learning
Specific structure frame.
Bring background interference when the license plate rectification module can remove car plate detection by the shape of correction license plate, can mention
The accuracy rate of high Character segmentation and character classification.The first depth convolutional neural networks positioning licence plate of the license plate rectification module
The coordinate position on four vertex simultaneously operates correction license plate shape by affine transformation.The first depth convolutional neural networks can be pre-
The coordinate position on four vertex of measuring car board, and using the coordinate as the parameter of affine transformation, to realize that license plate is corrected.
The correction license plate process of the license plate rectification module includes the building of the first depth convolutional neural networks, the first depth
The training of convolutional neural networks and affine transformation operation.
The building of the first depth convolutional neural networks includes building multilayer convolutional network layer and the full articulamentum of multilayer.Institute
It states each convolutional network layer and is all made of conditional nonlinear activation function ReLU and batch regularization.In the present embodiment, described
Convolutional network layer includes 5 convolutional network layers, is respectively as follows: conv1 convolutional network layer, conv2 convolutional network layer, conv3 convolution
Network layer, conv4 convolutional network layer, conv5 convolutional network layer；The full articulamentum includes 3 full articulamentums, is respectively as follows: fc6
Full articulamentum, the full articulamentum of fc7 and the full articulamentum of fc8.
In 5 convolutional network layers, conv1 convolutional network layer is made of the filter of 64 3*3, conv2 convolution net
Network layers are made of the filter of 192 3*3, and conv3 convolutional network layer is made of the filter of 384 3*3, conv4 convolution net
Network layers are made of the filter of 256 3*3, and conv5 convolutional network layer is made of the filter of 256 3*3.In conv1 convolution
It has been done respectively after network layer and conv3 convolutional network layer once to down-sampling, for carrying out dimensionality reduction to characteristic pattern.
In described 3 full articulamentums, 256 Feature Mapping figures that the full articulamentum of fc6 exports conv5 convolutional network layer turn
One-dimensional feature vector is turned to, the parameter of the full articulamentum of fc6 is reduced to the full articulamentum of 100, fc8 and exports 8 by the full articulamentum of fc7
Floating number.Floating number indicates that the transverse and longitudinal coordinate on clockwise four vertex since license plate bottom right vertex is former relative to input
The high ratio of figure width, therefore, it is coordinate position of the license plate vertex on picture that floating number, which is multiplied by corresponding wide higher position,.Finally,
Using the location information on four vertex, the license plate after correction can be obtained by an affine transformation.
The training of the first depth convolutional neural networks includes handmarking's license plate vertex position, generates training image；
The apex coordinate of picture of the input comprising license plate and the picture trains first depth to the first depth convolutional neural networks
Convolutional neural networks, until the true top of the license plate apex coordinate of the first depth convolutional neural networks output and the picture
Error between point coordinate does not continue to the amplitude for becoming smaller or becoming smaller less than until preset threshold value.
Wherein be input to having for the first depth convolutional neural networks: (1) size is the RGB picture of 128*64, for training
Neural network；(2) coordinate (V0 (x0, y0), V1 (x1, y1), V2 (x2, y2), V3 (x3, y3)) on four vertex of license plate is used
In calculating loss function.In the training process, the first depth convolutional neural networks use classical forward direction, backpropagation
It is trained, the initial weight of network is initialized using random number, and loss function uses manhatton distance, mathematical expression
Formula is as follows:
Wherein, W and H is respectively the width height of original image, and xi and yi are the coordinates for the Vi that the neural network forecast goes out, and Xi and Yi are the true of Vi
Coordinate, and loss function is minimized using backpropagation and stochastic gradient descent.
In order to train the first depth convolutional neural networks, has from Hefei city curb parking fee collecting system and have collected 10000
The motor vehicles original picture (size 720*1028) for being labelled with license plate number is opened, and is sat by handmarking license plate vertex
Cursor position and license plate area.In addition, in order to exponentially increase data set, it can be by amplifying certain ratio for every original picture
Example produces 2~4 small pictures comprising license plate area.
The affine transformation operation are as follows: the output of the first depth convolutional neural networks is eight floating numbers, respectively table
Show that the cross/ordinate on four vertex of the clock-wise order since license plate bottom right vertex inputs wide/high ratio of picture belonging to accounting for,
Therefore, the wide higher position for being multiplied by the input picture is the coordinate value on four vertex of license plate；It is grasped the coordinate value as affine transformation
The parameter of work can remove the jamming pattern information in the input picture, to complete the correction of license plate.
The Character segmentation module neural network prediction Character segmentation position is simultaneously cut into single character picture.Character is cut
The accuracy cut is affected to the accuracy of character recognition, such as: when cutting letter P, cutting position is to the left to will lead to knowledge
Do not become larger at the probability of F, therefore, cutting position can be predicted using the neural network.The cutting word of the Character segmentation module
Symbol process includes the building of neural network, the training of neural network and OpenCV image cutting operation.
The building of the neural network includes constructing multiple full articulamentums, the last one full articulamentum exports floating number.?
In the present embodiment, the neural network is made of four full articulamentums, and it is (shared that the last one full articulamentum exports six floating numbers
7 characters on license plate), that is, the position of six Character segmentations.
The training of the neural network: manually to the license plate picture indicia license plate cutting position after correction, its size is reset
For 256*72 and it is converted into grayscale image.Then each grayscale image sums the gray value of image of every a line and each column to obtain
One contains 328(256+72) one-dimensional vector of a value；The cutting position of this one-dimensional vector and the license plate is input to described
Neural network, until the neural network output cutting position and the true cutting position of the picture between error not followed by
Until the continuous amplitude for becoming smaller or becoming smaller is less than preset threshold value.
The neural network uses manhatton distance as loss function, and mathematic(al) representation is as follows:
, wherein W indicates that the width of input picture, zi indicate the net
I-th of cutting position that network predicts, Zi indicate i-th of true cutting position of the network.
The OpenCV image cutting operation: the output of the neural network is six floating numbers (with seven character license plates
For), the coordinate for respectively indicating six cutting positions from left to right accounts for the ratio of picture width；Therefore, it is multiplied by the input
The width of picture is exactly the cutting position coordinate (z1 ~ z6) of characters on license plate；Indicate that picture left end, z7 indicate right end with z0；
The method of cutting picture can be expressed as follows with mathematic(al) representation:, wherein Image table
Show license plate original image, Image_i indicates i-th of character picture after cutting.It, can be by license plate area cutting by cutting position information
For 7 character pictures.
The character recognition module carries out character classification with the second depth convolutional neural networks.The character recognition module
Purpose is classified to each character picture, using the second depth convolutional neural networks that can be carried out character picture classification.
In character recognition, using monocase picture and picture alphanumeric tag as input, it is input to AlexNet depth convolutional Neural net
Network.
By taking the license plate of Chinese blue bottom as an example, totally 7 characters, first character are province Chinese character (common totally 34 provinces),
Second character is letter (removing alphabetical ' I ' and ' O ' totally 24 letters), and rear five characters are letter or number (removal letter
' I ' and ' O ', letter and number are worth for 34 totally).In order to improve the accuracy rate of character recognition, the identical neural network of the present invention
Three models (being expressed as M1, M2, M3) have been respectively trained in structure, are respectively used to first character, second character and five latter
The character picture classification of character.
Second depth convolutional neural networks structure uses AlexNet depth convolutional neural networks structure, only in the last layer
Full articulamentum is modified, and is specifically amended as follows:
For M1, the output dimension of the last one full articulamentum is changed to 34；For M2, the output of the last one full articulamentum
Dimension is changed to 24；For M3, the output dimension of the last one full articulamentum is changed to 34.
In the training process, in order to accelerate fit procedure, present invention employs trained on ImageNet data set
Model parameter initializes the parameters of the second depth convolutional neural networks, loss function SoftMax.
Method proposed by the present invention is not related to the process of car plate detection, the process of detection can with Faster-RCNN, SSD,
The well-known target detection frame such as YOLO is realized.
The present invention is for the motor vehicle shot by man-hour manually hand-held terminal device in roadside under city curb parking charge scene
Picture, devises a kind of licence plate recognition method that can tolerate larger license plate sloped angle, and this method supports automatic identification license plate
Tilt angle is the small license plate picture within 50 degree.The present invention can solve the licence plate recognition method of prior art offer, not be related to
License plate is corrected and leads to can only to identify tilt angle and correct in 15 degree license plates below, by relational graph calculating realization license plate
Lead to problems such as time-consuming discrimination again low.
The above description is only an embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalent structure or equivalent flow shift made by bright description is applied directly or indirectly in other relevant technology necks
Domain is included within the scope of the present invention.
Claims (5)
1. a kind of licence plate recognition method based on deep learning, which is characterized in that including license plate rectification module, Character segmentation module
And character recognition module；
The first depth convolutional neural networks positioning licence plate coordinate position of the license plate rectification module is simultaneously operated by affine transformation
Correct license plate shape；
The Character segmentation module neural network prediction Character segmentation position is simultaneously cut into single character picture；
The character recognition module carries out character classification with the second depth convolutional neural networks.
2. the licence plate recognition method according to claim 1 based on deep learning, which is characterized in that the coordinate of the license plate
Position is the coordinate position on four vertex of license plate.
3. the licence plate recognition method according to claim 1 based on deep learning, which is characterized in that the license plate corrects mould
The correction license plate process of block includes the building of the first depth convolutional neural networks, the training of the first depth convolutional neural networks and imitates
Penetrate map function；
The building of the first depth convolutional neural networks includes constructing multilayer convolutional network layer and the full articulamentum of multilayer, wherein institute
It states each convolutional network layer and is all made of conditional nonlinear activation function ReLU and batch regularization；
The training of the first depth convolutional neural networks includes handmarking's license plate coordinate, generates training image, input training
To the first depth convolutional neural networks, training the first depth convolution is refreshing for image and the corresponding license plate coordinate of the training image
Through network, until the license plate coordinate license plate coordinate corresponding with the training image of the first depth convolutional neural networks output
Between error do not continue to the amplitude for becoming smaller or becoming smaller less than until preset threshold value；
The affine transformation operation is to grasp the license plate coordinate of the first depth convolutional neural networks output as affine transformation
The parameter of work removes the jamming pattern information in input picture, realizes the correction of license plate.
4. the licence plate recognition method according to claim 1 based on deep learning, which is characterized in that the Character segmentation mould
The cutting character process of block includes the building of neural network, the training of neural network and OpenCV image cutting operation；
The building of the neural network includes constructing multiple full articulamentums, the last one full articulamentum exports floating number；
The training of the neural network includes manually resetting the license plate to the license plate picture indicia license plate cutting position after correction
The size of picture, and it is converted into grayscale image, then every a line and the gray value of image of each column are summed for each grayscale image
Obtain one and contain 328(256+72) one-dimensional vector of a value, the cutting position of the one-dimensional vector and the license plate is inputted
To the neural network, until the error between neural network output cutting position and the true cutting position of the picture
The amplitude for becoming smaller or becoming smaller is not continued to less than until preset threshold value；
The OpenCV image cutting operation includes the cleavage of the floating number that exports the neural network multiplied by characters on license plate
Coordinate is set, the mathematic(al) representation of picture approach is cut are as follows:, wherein Image table
Show license plate original image, Image_i indicates i-th of character picture after cutting.
5. the licence plate recognition method according to claim 1 based on deep learning, which is characterized in that the character recognition mould
The character recognition process of block is to be input to the second depth convolutional Neural net using monocase picture and picture alphanumeric tag as input
Network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810853164.7A CN109145900A (en) | 2018-07-30 | 2018-07-30 | A kind of licence plate recognition method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810853164.7A CN109145900A (en) | 2018-07-30 | 2018-07-30 | A kind of licence plate recognition method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109145900A true CN109145900A (en) | 2019-01-04 |
Family
ID=64799120
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810853164.7A Pending CN109145900A (en) | 2018-07-30 | 2018-07-30 | A kind of licence plate recognition method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109145900A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109784342A (en) * | 2019-01-24 | 2019-05-21 | 厦门商集网络科技有限责任公司 | A kind of OCR recognition methods and terminal based on deep learning model |
| CN110020650A (en) * | 2019-03-26 | 2019-07-16 | 武汉大学 | A kind of construction method, recognition methods and the device of the deep learning identification model for inclination license plate |
| CN110059683A (en) * | 2019-04-15 | 2019-07-26 | 广州广电银通金融电子科技有限公司 | A kind of license plate sloped antidote of wide-angle based on end-to-end neural network |
| CN110136449A (en) * | 2019-06-17 | 2019-08-16 | 珠海华园信息技术有限公司 | Traffic video frequency vehicle based on deep learning disobeys the method for stopping automatic identification candid photograph |
| CN110427937A (en) * | 2019-07-18 | 2019-11-08 | 浙江大学 | A kind of correction of inclination license plate and random length licence plate recognition method based on deep learning |
| CN110659638A (en) * | 2019-09-24 | 2020-01-07 | 北京精英路通科技有限公司 | License plate recognition method and device, computer equipment and storage medium |
| CN110717907A (en) * | 2019-10-06 | 2020-01-21 | 浙江大学 | Intelligent hand tumor detection method based on deep learning |
| CN110837832A (en) * | 2019-11-08 | 2020-02-25 | 深圳市深视创新科技有限公司 | Rapid OCR recognition method based on deep learning network |
| CN110969135A (en) * | 2019-12-05 | 2020-04-07 | 中南大学 | Vehicle logo recognition method in natural scene |
| CN111178357A (en) * | 2019-12-31 | 2020-05-19 | 松立控股集团股份有限公司 | License plate recognition method, system, device and storage medium |
| CN111310754A (en) * | 2019-12-31 | 2020-06-19 | 创泽智能机器人集团股份有限公司 | Method for segmenting license plate characters |
| CN112164227A (en) * | 2020-08-26 | 2021-01-01 | 深圳奇迹智慧网络有限公司 | Illegal parking vehicle warning method and device, computer equipment and storage medium |
| CN113095320A (en) * | 2021-04-01 | 2021-07-09 | 湖南大学 | License plate recognition method and system and computing device |
| CN113392837A (en) * | 2021-07-09 | 2021-09-14 | 超级视线科技有限公司 | License plate recognition method and device based on deep learning |
| CN114677502A (en) * | 2022-05-30 | 2022-06-28 | 松立控股集团股份有限公司 | License plate detection method with any inclination angle |
| CN115100639A (en) * | 2022-08-08 | 2022-09-23 | 松立控股集团股份有限公司 | License plate recognition method by utilizing single character correction |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103530600A (en) * | 2013-06-06 | 2014-01-22 | 东软集团股份有限公司 | License plate recognition method and system under complicated illumination |
| CN106022232A (en) * | 2016-05-12 | 2016-10-12 | 成都新舟锐视科技有限公司 | License plate detection method based on deep learning |
| CN106709486A (en) * | 2016-11-11 | 2017-05-24 | 南京理工大学 | Automatic license plate identification method based on deep convolutional neural network |
-
2018
- 2018-07-30 CN CN201810853164.7A patent/CN109145900A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103530600A (en) * | 2013-06-06 | 2014-01-22 | 东软集团股份有限公司 | License plate recognition method and system under complicated illumination |
| CN106022232A (en) * | 2016-05-12 | 2016-10-12 | 成都新舟锐视科技有限公司 | License plate detection method based on deep learning |
| CN106709486A (en) * | 2016-11-11 | 2017-05-24 | 南京理工大学 | Automatic license plate identification method based on deep convolutional neural network |
Non-Patent Citations (3)
| Title |
|---|
| 刘丰: "机器学习和卷积神经网络在车牌识别的研究和应用", 《中国优秀硕士学位论文全文数据库（电子期刊）信息科技辑》 * |
| 贺赛娜: "复杂场景下的中国车牌识别研究", 《中国优秀硕士学位论文全文数据库（电子期刊）工程科技Ⅱ辑》 * |
| 陈利: "基于深度学习的车牌识别系统设计", 《计算机技术与发展》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109784342B (en) * | 2019-01-24 | 2021-03-12 | 厦门商集网络科技有限责任公司 | OCR (optical character recognition) method and terminal based on deep learning model |
| CN109784342A (en) * | 2019-01-24 | 2019-05-21 | 厦门商集网络科技有限责任公司 | A kind of OCR recognition methods and terminal based on deep learning model |
| CN110020650A (en) * | 2019-03-26 | 2019-07-16 | 武汉大学 | A kind of construction method, recognition methods and the device of the deep learning identification model for inclination license plate |
| CN110059683A (en) * | 2019-04-15 | 2019-07-26 | 广州广电银通金融电子科技有限公司 | A kind of license plate sloped antidote of wide-angle based on end-to-end neural network |
| CN110136449A (en) * | 2019-06-17 | 2019-08-16 | 珠海华园信息技术有限公司 | Traffic video frequency vehicle based on deep learning disobeys the method for stopping automatic identification candid photograph |
| CN110136449B (en) * | 2019-06-17 | 2022-05-20 | 珠海华园信息技术有限公司 | Deep learning-based traffic video vehicle illegal parking automatic identification snapshot method |
| CN110427937A (en) * | 2019-07-18 | 2019-11-08 | 浙江大学 | A kind of correction of inclination license plate and random length licence plate recognition method based on deep learning |
| CN110427937B (en) * | 2019-07-18 | 2022-03-22 | 浙江大学 | Inclined license plate correction and indefinite-length license plate identification method based on deep learning |
| CN110659638A (en) * | 2019-09-24 | 2020-01-07 | 北京精英路通科技有限公司 | License plate recognition method and device, computer equipment and storage medium |
| CN110717907A (en) * | 2019-10-06 | 2020-01-21 | 浙江大学 | Intelligent hand tumor detection method based on deep learning |
| CN110837832A (en) * | 2019-11-08 | 2020-02-25 | 深圳市深视创新科技有限公司 | Rapid OCR recognition method based on deep learning network |
| CN110969135A (en) * | 2019-12-05 | 2020-04-07 | 中南大学 | Vehicle logo recognition method in natural scene |
| CN110969135B (en) * | 2019-12-05 | 2023-04-18 | 中南大学 | Vehicle logo recognition method in natural scene |
| CN111178357A (en) * | 2019-12-31 | 2020-05-19 | 松立控股集团股份有限公司 | License plate recognition method, system, device and storage medium |
| CN111310754A (en) * | 2019-12-31 | 2020-06-19 | 创泽智能机器人集团股份有限公司 | Method for segmenting license plate characters |
| CN111178357B (en) * | 2019-12-31 | 2023-09-29 | 松立控股集团股份有限公司 | License plate recognition method, system, device and storage medium |
| CN112164227A (en) * | 2020-08-26 | 2021-01-01 | 深圳奇迹智慧网络有限公司 | Illegal parking vehicle warning method and device, computer equipment and storage medium |
| CN113095320A (en) * | 2021-04-01 | 2021-07-09 | 湖南大学 | License plate recognition method and system and computing device |
| CN113392837A (en) * | 2021-07-09 | 2021-09-14 | 超级视线科技有限公司 | License plate recognition method and device based on deep learning |
| CN114677502A (en) * | 2022-05-30 | 2022-06-28 | 松立控股集团股份有限公司 | License plate detection method with any inclination angle |
| CN115100639A (en) * | 2022-08-08 | 2022-09-23 | 松立控股集团股份有限公司 | License plate recognition method by utilizing single character correction |
| CN115100639B (en) * | 2022-08-08 | 2024-04-09 | 松立控股集团股份有限公司 | License plate recognition method using single character correction |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109145900A (en) | A kind of licence plate recognition method based on deep learning | |
| Wang et al. | RENet: Rectangular convolution pyramid and edge enhancement network for salient object detection of pavement cracks | |
| CN105354568A (en) | Convolutional neural network based vehicle logo identification method | |
| CN112016556B (en) | Multi-type license plate recognition method | |
| CN106934374B (en) | Method and system for identifying traffic signboard in haze scene | |
| CN1932838A (en) | Vehicle plate extracting method based on skiagraphy and mathematical morphology | |
| CN111695448A (en) | Roadside vehicle identification method based on visual sensor | |
| CN107977677A (en) | A kind of multi-tag pixel classifications method in the reconstruction applied to extensive city | |
| CN102254152A (en) | License plate location method based on color change points and color density | |
| CN109886147A (en) | A kind of more attribute detection methods of vehicle based on the study of single network multiple-task | |
| CN104978567A (en) | Vehicle detection method based on scenario classification | |
| Wang et al. | Vehicle detection based on drone images with the improved faster R-CNN | |
| Xiang et al. | Lightweight fully convolutional network for license plate detection | |
| CN103164697B (en) | The licence plate recognition method of a kind for the treatment of time and accuracy of identification self-adaptation | |
| Jin et al. | A semi-automatic annotation technology for traffic scene image labeling based on deep learning preprocessing | |
| Riaz et al. | YOLO based recognition method for automatic license plate recognition | |
| CN105404859A (en) | Vehicle type recognition method based on pooling vehicle image original features | |
| CN111259950B (en) | Method for training YOLO neural network based on 3D model | |
| Liu et al. | A large-scale benchmark for vehicle logo recognition | |
| Lin et al. | A lightweight, high-performance multi-angle license plate recognition model | |
| CN100385450C (en) | Vehicle license plate extraction method based on high-hat switch and wavelet switch | |
| CN109740554A (en) | A kind of road edge line recognition methods and system | |
| Zhang et al. | License plate recognition model based on CNN+ LSTM+ CTC | |
| CN108734170B (en) | License plate character segmentation method based on machine learning and template | |
| Rahmani et al. | IR-LPR: A Large Scale Iranian License Plate Recognition Dataset |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190104 |