CN106096602A - Chinese license plate recognition method based on convolutional neural network - Google Patents
Chinese license plate recognition method based on convolutional neural network Download PDFInfo
- Publication number
- CN106096602A CN106096602A CN201610445168.2A CN201610445168A CN106096602A CN 106096602 A CN106096602 A CN 106096602A CN 201610445168 A CN201610445168 A CN 201610445168A CN 106096602 A CN106096602 A CN 106096602A
- Authority
- CN
- China
- Prior art keywords
- convolutional neural
- neural networks
- character
- license plate
- chinese
- 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
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 51
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000012545 processing Methods 0.000 claims abstract description 15
- 230000008569 process Effects 0.000 claims description 15
- 230000011218 segmentation Effects 0.000 claims description 14
- 238000013461 design Methods 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 11
- 230000005284 excitation Effects 0.000 claims description 6
- 239000000284 extract Substances 0.000 claims description 6
- 238000012706 support-vector machine Methods 0.000 claims description 6
- 230000008878 coupling Effects 0.000 claims description 5
- 238000010168 coupling process Methods 0.000 claims description 5
- 238000005859 coupling reaction Methods 0.000 claims description 5
- 238000001514 detection method Methods 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 5
- 238000005457 optimization Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 230000007935 neutral effect Effects 0.000 description 4
- 244000174681 Michelia champaca Species 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 230000001537 neural effect Effects 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 238000012937 correction Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 210000004205 output neuron Anatomy 0.000 description 2
- 230000003252 repetitive effect Effects 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 206010011469 Crying Diseases 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000007795 chemical reaction product Substances 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 230000001276 controlling effect Effects 0.000 description 1
- 230000007850 degeneration Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
-
- 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
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a Chinese license plate recognition method based on a convolutional neural network, which comprises the following steps of S1: positioning the license plate, namely combining a color template matching algorithm and a contour searching algorithm to position the license plate; s2: dividing characters, namely obtaining an external rectangle after binarizing, closing and contouring the license plate rectangular blocks after graying, and accordingly intercepting character image blocks; s3: designing and training a convolutional neural network, wherein the convolutional neural network is arranged into a 10-layer structure; s4: and (4) character recognition, namely inputting the character to be recognized by using the convolutional neural network trained in advance in S3 as a classifier to obtain a classification result and a confidence rate of the classification result. Through the mode, the Chinese license plate recognition method based on the convolutional neural network has the advantages of extremely high accuracy, extremely strong universality, short processing time and the like, and has wide application prospects in the scenes of modern intelligent traffic systems, parking lot management, highway toll stations and the like.
Description
Technical field
The present invention relates to computer vision, Digital Image Processing and degree of depth learning areas, particularly relate to based on convolution god
Chinese licence plate recognition method through network.
Background technology
Car license recognition is one of important component part in modern intelligent transportation system, applies quite varied.It is with numeral
Based on the technology such as image procossing, computer vision, machine learning, vehicle image or the video sequence to shot by camera
It is analyzed, obtains each the unique number-plate number of automobile, thus complete identification process.Can by some subsequent treatment means
To realize parking lot fee collection management, traffic flow Con trolling index is measured, and vehicle location, automobile burglar, high way super speed is automatic
Change supervision, electronic eye used for catching red light runner, toll station etc. function.For safeguarding traffic safety and urban public security, prevent traffic
Blocking, it is achieved traffic automation management has the meaning of reality.
The hardware foundation of Vehicle License Plate Recognition System generally comprises triggering equipment (whether monitoring vehicle enters the visual field), shooting sets
Standby, luminaire, image capture device, the datatron (such as computer) etc. of the identification number-plate number, and its software kernels includes car
Board location algorithm, Character Segmentation of License Plate and character recognition algorithm.
License Plate and License Plate Character Segmentation part mainly use computer vision and image processing techniques.By to quiet
The morphological operations such as state image carries out rotating, converts, gray processing, fuzzy, closed operation, burn into contouring, statistics with histogram, can
To extract part required for us from image.
Conventional character identifying method, substantially can be divided into two classes: OCR (OCR) and ANN
Network algorithm (ANN).OCR mainly uses algorithm based on template matching, and method is as follows: first by the character binaryzation after segmentation,
And its size is scaled the size of template in character database, then mate with all of template, select optimal
Coupling is as result.But the method has two significant drawbacks, on the one hand, it is limited to the size of template database, if data
Storehouse is the least and separating character deformation big, then be likely to produce erroneous matching result；On the other hand, matching process ratio is time-consuming,
For requiring that the application scenarios of real-time is unacceptable.
ANN algorithm is one of most popular algorithm of machine learning circle, can complete the classification task of complexity.One simplification
ANN can be divided into input layer, hidden layer and input layer.Input layer is responsible for receiving data, and hidden layer is responsible for decomposing data
And process, integrate end product to output layer.Middle hidden layer is the most, and the sign ability of ANN is the strongest, more can extract data
Feature.In theory, the ANN of 3 hidden layers can represent any function, i.e. can process any classification task.But, along with
The increase of ANN hidden layers numbers, ANN model training time and complexity index rise, and its development enters the bottleneck phase.
2006, machine learning field banker Geoffrey Hinton published an article at " Science ", demonstrates two
Viewpoint: (1) many hidden layers neutral net has the feature learning ability of excellence, learn to feature data are had more essential quarter
Draw, thus be conducive to visualization or classification；
(2) deep neural network difficulty in training, effectively can be overcome by " successively initializing ".
This demonstration, not only solves neutral net difficulty computationally, also illustrate that deep-neural-network is being learned simultaneously
Superiority in habit.Therefore, the neutral net with multiple hidden layer is referred to as deep neural network, based on degree of depth nerve net
The Learning Studies of network is referred to as degree of depth study.It addition, along with computer capacity improve and GPU calculate, Distributed Calculation send out
Exhibition, deep neural network model can train out in the range of acceptable time.
Convolutional neural networks (CNN), as the one of deep neural network, has become as current speech analysis and image is known
The research in other field is popular.CNN directly can identify visual pattern from original image, extracts and learning characteristic, and can
Identify the pattern changed, there is the robustness to simple geometry deformation.
CNN is more general, and ANN has the following advantages in terms of image procossing: (1) input picture and topology of networks can be very
Good coincide；(2) feature extraction and pattern classification are carried out simultaneously；(3) weights are shared and can be made god with the training parameter of less network
Simpler through network structure, adaptability is higher.
In sum, in Chinese Vehicle License Plate Recognition System, the CNN model in degree of depth study is used to replace traditional OCR side
Method and ANN algorithm, can be greatly improved recognition accuracy, reduces recognition time, has extremely bright in Chinese Car license recognition field
Application prospect.
Summary of the invention
Technical problem is how of present invention mainly solves provides one to have that accuracy rate is high, universality is extremely strong, processes
The advantages such as the time is the shortest, have widely in the scenes such as modern intelligent transportation system, parking lot management, freeway toll station
The Chinese licence plate recognition method of the convolutional neural networks of application prospect.
For solving above-mentioned technical problem, the technical scheme that the present invention uses is: provide a kind of based on convolutional Neural net
The Chinese licence plate recognition method of network, including comprising the steps:
S1: License Plate step, combines color template matching algorithm and profile lookup algorithm, orients car plate；
S2: Character segmentation step, by the car plate rectangular block after gray processing, after binaryzation, closed operation, contouring, outside can obtaining
Connect rectangle, thus intercept out character segment；
The design of S3: convolutional neural networks and training step, wherein, convolutional neural networks is set to 10 Rotating fields, defeated including 1
Enter layer, 1 output layer, 2 convolutional layers for feature extraction, 2 sampling layers chosen for characteristic optimization, 2 for table
Show that the full articulamentum of feature, 1 excitation layer for Fast Convergent and 1 are for calculating output and the loss of target loss value
Layer；
S4: character recognition step, uses the convolutional neural networks that in S3, training in advance is good, as grader, the word that will identify
Symbol input, obtains classification results and its confidence rate.
In a preferred embodiment, S1 uses color lookup algorithm, to Image semantic classification and carry out color template
Coupling, and whether the match is successful to judge car plate.
In a preferred embodiment, if color lookup algorithm cannot positioning licence plate, the then lookup of use profile in S1
Algorithm, after image gray processing, carries out angle judgement and size detection, obtains car plate rectangular block, uses training in advance simultaneously
Support vector machine classifier, filters out real car plate.
In a preferred embodiment, both can apply to still image at License Plate described in S1, can apply again
In dynamic video, if during dynamic video, the most first extract key frame, be re-used as still image and process.
In a preferred embodiment, use the character split in S2, as data set, be iterated training, will
Data set is divided into training set and checking collection in 8:2 ratio, through 100,000 iteration, trains convolutional neural networks model.
The invention has the beneficial effects as follows: there is the advantages such as accuracy rate is high, universality is extremely strong, the process time is the shortest, existing
Have a wide range of applications in the scenes such as intelligent transportation system, parking lot management, freeway toll station.
Accompanying drawing explanation
For the technical scheme being illustrated more clearly that in the embodiment of the present invention, in embodiment being described below required for make
Accompanying drawing be briefly described, it should be apparent that, below describe in accompanying drawing be only some embodiments of the present invention, for
From the point of view of those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to obtain other according to these accompanying drawings
Accompanying drawing, wherein:
Fig. 1 is the car plate of the first stage of the present invention Chinese licence plate recognition method one based on convolutional neural networks preferred embodiment
Positioning flow figure;
Fig. 2 is the character of the second stage of the present invention Chinese licence plate recognition method one based on convolutional neural networks preferred embodiment
Segmentation flow chart;
Fig. 3 is the design of the phase III of the present invention Chinese licence plate recognition method one based on convolutional neural networks preferred embodiment
Convolutional neural networks PRNet illustraton of model;
Convolution in the phase III that Fig. 4 is the present invention Chinese licence plate recognition method one based on convolutional neural networks preferred embodiment
Operation chart;
Fig. 5 is in the phase III of the present invention Chinese licence plate recognition method one based on convolutional neural networks preferred embodiment
PRNet model training schematic diagram；
Fig. 6 is the overall flow figure of the present invention Chinese licence plate recognition method one based on convolutional neural networks preferred embodiment.
Detailed description of the invention
Technical scheme in the embodiment of the present invention will be clearly and completely described below, it is clear that described enforcement
Example is only a part of embodiment of the present invention rather than whole embodiments.Based on the embodiment in the present invention, this area is common
All other embodiments that technical staff is obtained under not making creative work premise, broadly fall into the model of present invention protection
Enclose.
Refer to Fig. 1-6, a kind of Chinese based on convolutional neural networks is provided in one particular embodiment of the present invention
Licence plate recognition method, described based on convolutional neural networks Chinese licence plate recognition method includes comprising the steps:
S1: License Plate step, combines color template matching algorithm and profile lookup algorithm, orients car plate；
S2: Character segmentation step, by the car plate rectangular block after gray processing, after binaryzation, closed operation, contouring, outside can obtaining
Connect rectangle, thus intercept out character segment；
The design of S3: convolutional neural networks and training step, wherein, convolutional neural networks is set to 10 Rotating fields, defeated including 1
Enter layer, 1 output layer, 2 convolutional layers for feature extraction, 2 sampling layers chosen for characteristic optimization, 2 for table
Show that the full articulamentum of feature, 1 excitation layer for Fast Convergent and 1 are for calculating output and the loss of target loss value
Layer；
S4: character recognition step, uses the convolutional neural networks that in S3, training in advance is good, as grader, the word that will identify
Symbol input, obtains classification results and its confidence rate.
In S1, use color lookup algorithm, to Image semantic classification the coupling that carries out color template, and judge that car plate is
It is no that the match is successful.If color lookup algorithm cannot positioning licence plate, then use profile lookup algorithm, after image gray processing, enter
Row angle judges and size detection, obtains car plate rectangular block, uses the support vector machine classifier of training in advance simultaneously, filter out
Real car plate.Described License Plate both can apply to still image, can apply to again dynamic video, if dynamic video
Time, the most first extract key frame, be re-used as still image and process.Use the character split in S2, as data set, carry out
Repetitive exercise, is divided into training set and checking collection by data set in 8:2 ratio, through 100,000 iteration, trains convolutional Neural net
Network model.
In a specific embodiment: a kind of Chinese licence plate recognition method based on convolutional neural networks, use traditional
Computer vision and image processing techniques, from dynamic video or static images, orient car plate, and cut into character；Use
Convolutional neural networks (CNN) the classification character that sign ability that training in advance is good is powerful, is combined into characters on license plate by classification results
String；For the indiscernible problem of Chinese character of Chinese car plate, use CNN can significantly improve discrimination.
Comprise the steps: specifically
(1) License Plate: color template matching algorithm and profile lookup algorithm are combined, orients car plate.First by face
Color lookup algorithm, after Image semantic classification, carries out the color template coupling of champac dichromatism, if success, is probably car plate.
If color lookup algorithm cannot positioning licence plate, then use profile lookup algorithm, after image gray processing, use Sobel operator
Contouring, carries out angle judgement and size detection, and obtaining may car plate rectangular block.In order to improve accuracy rate further, use pre-
Support vector machine (SVM) grader first trained, filters out real car plate.
(2) Character segmentation: by the car plate rectangular block after gray processing, after binaryzation, closed operation, contouring, outside can obtaining
Connect rectangle, thus intercept out character segment.Chinese character is interval owing to there is stroke, contouring when, can produce fracture
Phenomenon, is divided into multiple profile.For this problem, use position Backstipping design, by second city codes character, counter release
Chinese character position.
(3) design of CNN and training: the abbreviation PRNet of the CNN of present invention design, has 10 Rotating fields, defeated including 1
Enter layer, 1 output layer, 2 convolutional layers, 2 sampling layers, 2 full articulamentums, 1 excitation layer and 1 loss layer.Wherein, convolution
Layer, for feature extraction；Sampling layer, chooses for characteristic optimization；Full layer in succession, is used for representing feature；Excitation layer, for quickly
Convergence；Loss layer, is used for calculating output and target loss value.With traditional OCR (OCR) and artificial neuron
Network algorithm (ANN) is compared, and this PRNet model sign ability is higher, and serious forgiveness is higher, can preferably learn to Chinese car plate
Characteristic information, reaches more preferable character recognition effect.In order to train this PRNet model, need to use splitting of step (2)
Character, as data set, is iterated training.Data set is divided into training set and checking collection in 8:2 ratio, through 100,000 times repeatedly
In generation, train PRNet model.Owing to this PRNet model is to drive by minimizing output and the loss (loss) of target
Practise, in continuous repetitive exercise, can be with self-correcting learning effect.
(4) character recognition: using the PRNet that step (3) training in advance is good, as grader, the character that will identify is defeated
Enter, classification results and its confidence rate can be obtained.7 character classification results of combination car plate, just obtain license plate recognition result.
Described license plate locating method in step (1) both can apply to still image, can apply to again dynamic video,
Dynamic video needs first to extract key frame, is re-used as still image and processes.
The training of the PRNet in step (3) is the degree of depth based on Berkeley University's vision and learning center (BVLC) study
Framework caffe.
The PRNet used in step (4) has only to through once training, and can repeatedly use many places.
Technical solution of the present invention is broadly divided into four-stage, respectively: License Plate, License Plate Character Segmentation, convolutional Neural
The design of network and training, application convolutional neural networks are identified.Wherein License Plate and License Plate Character Segmentation mainly use
Computer vision and image processing techniques.The design of convolutional neural networks can be optimized for particular case；And trained
But journey needs a longer acceptable time.This convolutional neural networks has only to once train, and just can repeatedly make many places
With.So in actual application scenarios, embodiment of the present invention only has three parts, i.e. License Plate, Character segmentation and use convolution
Neural network recognization.
One, the License Plate stage, as it is shown in figure 1, combine color template matching algorithm and profile lookup algorithm, accurately
Orient car plate.Mostly it is in RGB color due to input picture, and RGB is one and takes in proportion based on redgreenblue
Joining the space producing different colours, this brings the biggest difficulty to use champac dichromatism template matching.It is thus desirable to input is schemed
As being transformed into suitable hsv color space, between 200-280 and S value and V-value fall between 0.35-1.00 to find H-number
Region, the biggest probability in these regions is car plate.
If color template matching algorithm cannot positioning licence plate, or positioning licence plate quantity is less than predetermined threshold value, then continue
Use profile lookup algorithm positioning licence plate.First Gaussian Blur, to image denoising, then gray processing, picture is converted into gray-scale map
Sheet.Use the vertical edge in Soble operator detection image, carry out closed operation after binarization, allow car plate letter be linked to be one
Connected domain, in order to contouring.Contouring is obtained rectangle, carries out size and judge and angle judgement, get rid of non-car plate rectangular block.
In order to follow-up License Plate Character Segmentation obtains more preferable effect, need car plate to be rotated and normalization.In order to improve further
License Plate rate, uses the support vector machine classifier of training in advance, filters out real car plate.
Two, the Character segmentation stage, as in figure 2 it is shown, this single stepping is relatively simple.Car plate was successfully oriented in the last stage
On the basis of picture, by the most separated for all car plate words, form single character block.First will export on last stage
Car plate picture transfers gray scale picture to.Owing to, character contouring when, the binaryzation parameter for champac car plate is different, needs
Here carry out color judgement.Then after using adaptive thresholding algorithm (Otsu threshold method) to carry out binarization operation, contouring
Obtain rear 6 characters.The reason the most not taking whole 7 characters is, the first Chinese character is interval due to stroke, as
" reviving " word, contouring can be divided into two connected regions.In order to solve this problem, by deputy city codes letter forward
Push away 1.15 times of distances of character duration, be Chinese character.Through this stage, Character segmentation process completes, and obtains 7 of car plate
Character block picture.
Three, convolutional neural networks design and the training stage.The Car license recognition convolutional neural networks of present invention design is called for short
PRNet, its structure is as it is shown on figure 3, one has 10 layers.The most successively explain its effect and mentality of designing.
Input layer is the character of input, a size of 32x32.
Convolutional layer is made up of 20 width characteristic patterns, each neuron in every width characteristic pattern and front layer input picture same area
Being connected, extract the non-vision features such as the edge in image, flex point, entered convolution operation, output picture size is 28*28.So-called
Convolution operation be exactly just in image qualified part screen, feature is namely proposed.As shown in Figure 4, a 5x5
Gradation of image value matrix, entered a 3x3 convolution kernels operation after, obtain the image array of a 3x3.With middle 9
As a example by individual cell, its convolution weights are respectively 1,0,1,0,1,0,1,0,1, obtain the single lattice that value is 4 after convolution.?
In the PRNet of present invention design, convolutional layer kernel is all 5x5.
Sampling layer (being also pond layer) is come feature double sampling by secondary sampling, and is that the feature that convolution obtains has space not
Degeneration.This layer of characteristic pattern number is consistent with characteristic pattern number in preceding layer convolutional layer, and kernel is 2x2.
In order to improve sampling precision, extracting more features details, PRNet constructs twice convolutional layer-sampling Rotating fields, and
And second section convolutional layer-sampling layer feature map sheet number becomes 50.
Full linking layer (being also connecting layer) is used for representing feature, learns more information.First full linking layer has 500
Output neuron, can represent more characteristic details.Second full linking layer converges to 65 output neurons, and this is with final
Output number is consistent.
Excitation layer is used for Fast Convergent, and 500 outputs are converged to 65.65 of these 65 output also the most corresponding car plates
Character (including 31 municipalities directly under the Central Government of provinces and cities, 24 letters and 10 numerals).
Loss layer is used for helping PRNet model automatically to correct learning effect, by minimizing the loss of output and target
(loss) study is driven.
As it is shown in figure 5, PRNet model training has two processes, it is to propagate forward and back-propagation respectively.Propagate forward and be
One process from bottom to top layer, by layer each in model all as function, calculates output, eventually arrives at loss of top layer.
And back-propagation (also crying back propagation), it is process from top to bottom, penalty values calculates the derivative of each layer, finally arrive
Reach bottom and find key parameter, the learning parameter of correction model, reach the function of self-correcting learning effect.
Data set is divided into training set and test set in 8:2 ratio, stores with lmdb form, form data Layer.?
In the degree of depth of increasing income learning framework caffe, define model structure defined above with protobuf form, preset initial ginseng
Number, carries out 100,000 iteration.In continuous self-teaching and correcting, finally give a highest Car license recognition convolution of discrimination
Neutral net PRNet.
Four, the character recognition stage, the PRNet model training out on last stage is used, to the character segment needing identification
Classifying, classification results is the actual value of this character segment.7 character segment classification results of combination car plate, the most permissible
Obtain this license plate recognition result.
Therefore, the invention have the advantages that
One, by color template matching algorithm and profile lookup algorithm being combined, License Plate accuracy rate is substantially increased；
Two, by based on the car plate after normalization and positive twist, use the anti-pushing manipulation in position, position Chinese character well；
Three, by using convolutional neural networks directly from artwork identification visual pattern, self-teaching and correction, the highest knowledge is completed
Not in the case of rate, it is to avoid substantial amounts of pretreatment work；
Four, convolutional neural networks can once be trained, and is used for multiple times, and single recognition time is in millisecond rank, it is possible to is competent at and needs
Want the scene of Real-time Vehicle License Plate identification.
The foregoing is only embodiments of the invention, not thereby limit the scope of the claims of the present invention, every utilize this
Equivalent structure or equivalence flow process that bright description is made convert, or are directly or indirectly used in other relevant technology neck
Territory, is the most in like manner included in the scope of patent protection of the present invention.
Claims (5)
1. a Chinese licence plate recognition method based on convolutional neural networks, it is characterised in that include comprising the steps:
S1: License Plate step, combines color template matching algorithm and profile lookup algorithm, orients car plate；
S2: Character segmentation step, by the car plate rectangular block after gray processing, after binaryzation, closed operation, contouring, outside can obtaining
Connect rectangle, thus intercept out character segment；
The design of S3: convolutional neural networks and training step, wherein, convolutional neural networks is set to 10 Rotating fields, defeated including 1
Enter layer, 1 output layer, 2 convolutional layers for feature extraction, 2 sampling layers chosen for characteristic optimization, 2 for table
Show that the full articulamentum of feature, 1 excitation layer for Fast Convergent and 1 are for calculating output and the loss of target loss value
Layer；
S4: character recognition step, uses the convolutional neural networks that in S3, training in advance is good, as grader, the word that will identify
Symbol input, obtains classification results and its confidence rate.
Chinese licence plate recognition method based on convolutional neural networks the most according to claim 1, it is characterised in that in S1
Using color lookup algorithm, to Image semantic classification the coupling that carries out color template, and whether the match is successful to judge car plate.
Chinese licence plate recognition method based on convolutional neural networks the most according to claim 1, it is characterised in that in S1
If color lookup algorithm cannot positioning licence plate, then use profile lookup algorithm, after image gray processing, carry out angle judge and
Size detection, obtains car plate rectangular block, uses the support vector machine classifier of training in advance simultaneously, filter out real car plate.
Chinese licence plate recognition method based on convolutional neural networks the most according to claim 1, it is characterised in that in S1
Described License Plate both can apply to still image, can apply to again dynamic video, if during dynamic video, the most first extracts
Go out key frame, be re-used as still image and process.
Chinese licence plate recognition method based on convolutional neural networks the most according to claim 1, it is characterised in that in step
In S3, use the character split in S2, as data set, be iterated training, data set is divided into training in 8:2 ratio
Collection and checking collection, through 100,000 iteration, train convolutional neural networks model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610445168.2A CN106096602A (en) | 2016-06-21 | 2016-06-21 | Chinese license plate recognition method based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610445168.2A CN106096602A (en) | 2016-06-21 | 2016-06-21 | Chinese license plate recognition method based on convolutional neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106096602A true CN106096602A (en) | 2016-11-09 |
Family
ID=57237935
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610445168.2A Pending CN106096602A (en) | 2016-06-21 | 2016-06-21 | Chinese license plate recognition method based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106096602A (en) |
Cited By (34)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106557768A (en) * | 2016-11-25 | 2017-04-05 | 北京小米移动软件有限公司 | The method and device is identified by word in picture |
| CN106570565A (en) * | 2016-11-21 | 2017-04-19 | 中国科学院计算机网络信息中心 | Depth learning method and system for big data |
| CN106650721A (en) * | 2016-12-28 | 2017-05-10 | 吴晓军 | Industrial character identification method based on convolution neural network |
| 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 |
| CN106778765A (en) * | 2016-11-22 | 2017-05-31 | 深圳市捷顺科技实业股份有限公司 | A kind of method and device of Car license recognition |
| CN106886778A (en) * | 2017-04-25 | 2017-06-23 | 福州大学 | A kind of car plate segmentation of the characters and their identification method under monitoring scene |
| CN106934396A (en) * | 2017-03-09 | 2017-07-07 | 深圳市捷顺科技实业股份有限公司 | A kind of license number search method and system |
| CN107292307A (en) * | 2017-07-21 | 2017-10-24 | 华中科技大学 | One kind is inverted Chinese character identifying code automatic identifying method and system |
| CN107590774A (en) * | 2017-09-18 | 2018-01-16 | 北京邮电大学 | A kind of car plate clarification method and device based on generation confrontation network |
| CN107609485A (en) * | 2017-08-16 | 2018-01-19 | 中国科学院自动化研究所 | The recognition methods of traffic sign, storage medium, processing equipment |
| CN107679452A (en) * | 2017-08-28 | 2018-02-09 | 中国电子科技集团公司第二十八研究所 | Goods train license number real-time identifying system based on convolutional neural networks under big data |
| CN108009548A (en) * | 2018-01-09 | 2018-05-08 | 贵州大学 | A kind of Intelligent road sign recognition methods and system |
| CN108009547A (en) * | 2017-12-26 | 2018-05-08 | 深圳供电局有限公司 | Method and device for identifying nameplate of substation equipment |
| CN108052866A (en) * | 2017-11-17 | 2018-05-18 | 克立司帝控制系统（上海）有限公司 | Car license recognition learning method and system based on artificial neural network |
| CN108108738A (en) * | 2017-11-28 | 2018-06-01 | 北京达佳互联信息技术有限公司 | Image processing method, device and terminal |
| CN108156130A (en) * | 2017-03-27 | 2018-06-12 | 上海观安信息技术股份有限公司 | Network attack detecting method and device |
| WO2018112900A1 (en) * | 2016-12-23 | 2018-06-28 | 深圳先进技术研究院 | License plate recognition method and apparatus, and user equipment |
| CN108416348A (en) * | 2018-01-29 | 2018-08-17 | 重庆邮电大学 | Plate location recognition method based on support vector machines and convolutional neural networks |
| CN108460772A (en) * | 2018-02-13 | 2018-08-28 | 国家计算机网络与信息安全管理中心 | Harassing of advertisement facsimile signal detecting system based on convolutional neural networks and method |
| CN108491866A (en) * | 2018-03-06 | 2018-09-04 | 平安科技（深圳）有限公司 | Porny identification method, electronic device and readable storage medium storing program for executing |
| CN108734170A (en) * | 2018-05-25 | 2018-11-02 | 电子科技大学 | Registration number character dividing method based on machine learning and template |
| CN109583451A (en) * | 2018-11-28 | 2019-04-05 | 上海鹰觉科技有限公司 | Automatic identifying method and system based on warship ship side number |
| CN109635637A (en) * | 2018-10-30 | 2019-04-16 | 深圳市航天华拓科技有限公司 | A kind of licence plate recognition method, device and calculate equipment |
| CN109784334A (en) * | 2019-01-24 | 2019-05-21 | 合肥视展光电科技有限公司 | Round-the-clock round-the-clock licence plate recognition method, system, device and path identification method |
| CN110348396A (en) * | 2019-07-15 | 2019-10-18 | 南京信息工程大学 | A kind of road top text traffic sign and device based on deep learning |
| CN110569836A (en) * | 2018-06-06 | 2019-12-13 | 北京深鉴智能科技有限公司 | variable-length character string identification method and device |
| CN110569833A (en) * | 2019-09-06 | 2019-12-13 | 上海应用技术大学 | License plate positioning method |
| CN110858306A (en) * | 2018-08-22 | 2020-03-03 | 西门子（中国）有限公司 | License plate character recognition apparatus, method and computer-readable storage medium |
| CN111881914A (en) * | 2020-06-23 | 2020-11-03 | 安徽清新互联信息科技有限公司 | License plate character segmentation method and system based on self-learning threshold |
| CN111915025A (en) * | 2017-05-05 | 2020-11-10 | 英特尔公司 | Immediate deep learning in machine learning for autonomous machines |
| CN114463757A (en) * | 2022-01-28 | 2022-05-10 | 上海电机学院 | Industrial scene character end-side reasoning training device and method based on machine vision |
| CN114612730A (en) * | 2022-04-06 | 2022-06-10 | 哈尔滨工业大学 | Method and device for detecting household garbage classification generation proportion |
| CN114913515A (en) * | 2021-12-31 | 2022-08-16 | 北方工业大学 | End-to-end license plate recognition network construction method |
| CN115375626A (en) * | 2022-07-25 | 2022-11-22 | 浙江大学 | Medical image segmentation method, system, medium, and apparatus based on physical resolution |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103136528A (en) * | 2011-11-24 | 2013-06-05 | 同济大学 | Double-edge detection based vehicle license plate identification method |
| CN103824066A (en) * | 2014-03-18 | 2014-05-28 | 厦门翼歌软件科技有限公司 | Video stream-based license plate recognition method |
| CN105335743A (en) * | 2015-10-28 | 2016-02-17 | 重庆邮电大学 | Vehicle license plate recognition method |
-
2016
- 2016-06-21 CN CN201610445168.2A patent/CN106096602A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103136528A (en) * | 2011-11-24 | 2013-06-05 | 同济大学 | Double-edge detection based vehicle license plate identification method |
| CN103824066A (en) * | 2014-03-18 | 2014-05-28 | 厦门翼歌软件科技有限公司 | Video stream-based license plate recognition method |
| CN105335743A (en) * | 2015-10-28 | 2016-02-17 | 重庆邮电大学 | Vehicle license plate recognition method |
Non-Patent Citations (3)
| Title |
|---|
| YING-NONG CHEN等: "《18th International Conference on Pattern Recognition (ICPR"06)》", 24 August 2006 * |
| YUE HUANG等: ""Vehicle Logo Recognition System Based on Convolutional Neural Networks With a Pretraining Strategy"", 《IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS》 * |
| 张立等: ""基于卷积神经网络SLeNet_5的车牌识别方法"", 《信息技术》 * |
Cited By (48)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN106570565A (en) * | 2016-11-21 | 2017-04-19 | 中国科学院计算机网络信息中心 | Depth learning method and system for big data |
| CN106778765A (en) * | 2016-11-22 | 2017-05-31 | 深圳市捷顺科技实业股份有限公司 | A kind of method and device of Car license recognition |
| CN106557768A (en) * | 2016-11-25 | 2017-04-05 | 北京小米移动软件有限公司 | The method and device is identified by word in picture |
| CN106557768B (en) * | 2016-11-25 | 2021-07-06 | 北京小米移动软件有限公司 | Method and device for recognizing characters in picture |
| US10984289B2 (en) | 2016-12-23 | 2021-04-20 | Shenzhen Institute Of Advanced Technology | License plate recognition method, device thereof, and user equipment |
| WO2018112900A1 (en) * | 2016-12-23 | 2018-06-28 | 深圳先进技术研究院 | License plate recognition method and apparatus, and user equipment |
| CN106650721A (en) * | 2016-12-28 | 2017-05-10 | 吴晓军 | Industrial character identification method based on convolution neural network |
| CN106650721B (en) * | 2016-12-28 | 2019-08-13 | 吴晓军 | A kind of industrial character identifying method based on convolutional neural networks |
| CN106934396A (en) * | 2017-03-09 | 2017-07-07 | 深圳市捷顺科技实业股份有限公司 | A kind of license number search method and system |
| CN108156130A (en) * | 2017-03-27 | 2018-06-12 | 上海观安信息技术股份有限公司 | Network attack detecting method and device |
| CN106886778A (en) * | 2017-04-25 | 2017-06-23 | 福州大学 | A kind of car plate segmentation of the characters and their identification method under monitoring scene |
| CN106886778B (en) * | 2017-04-25 | 2020-02-07 | 福州大学 | License plate character segmentation and recognition method in monitoring scene |
| CN111915025B (en) * | 2017-05-05 | 2024-04-30 | 英特尔公司 | Instant deep learning in machine learning for autonomous machines |
| CN111915025A (en) * | 2017-05-05 | 2020-11-10 | 英特尔公司 | Immediate deep learning in machine learning for autonomous machines |
| CN107292307A (en) * | 2017-07-21 | 2017-10-24 | 华中科技大学 | One kind is inverted Chinese character identifying code automatic identifying method and system |
| CN107292307B (en) * | 2017-07-21 | 2019-12-17 | 华中科技大学 | Automatic identification method and system for inverted Chinese character verification code |
| CN107609485A (en) * | 2017-08-16 | 2018-01-19 | 中国科学院自动化研究所 | The recognition methods of traffic sign, storage medium, processing equipment |
| CN107679452A (en) * | 2017-08-28 | 2018-02-09 | 中国电子科技集团公司第二十八研究所 | Goods train license number real-time identifying system based on convolutional neural networks under big data |
| CN107590774A (en) * | 2017-09-18 | 2018-01-16 | 北京邮电大学 | A kind of car plate clarification method and device based on generation confrontation network |
| CN108052866A (en) * | 2017-11-17 | 2018-05-18 | 克立司帝控制系统（上海）有限公司 | Car license recognition learning method and system based on artificial neural network |
| CN108108738A (en) * | 2017-11-28 | 2018-06-01 | 北京达佳互联信息技术有限公司 | Image processing method, device and terminal |
| CN108108738B (en) * | 2017-11-28 | 2018-11-16 | 北京达佳互联信息技术有限公司 | Image processing method, device and terminal |
| CN108009547A (en) * | 2017-12-26 | 2018-05-08 | 深圳供电局有限公司 | Method and device for identifying nameplate of substation equipment |
| CN108009548A (en) * | 2018-01-09 | 2018-05-08 | 贵州大学 | A kind of Intelligent road sign recognition methods and system |
| CN108416348A (en) * | 2018-01-29 | 2018-08-17 | 重庆邮电大学 | Plate location recognition method based on support vector machines and convolutional neural networks |
| CN108460772A (en) * | 2018-02-13 | 2018-08-28 | 国家计算机网络与信息安全管理中心 | Harassing of advertisement facsimile signal detecting system based on convolutional neural networks and method |
| WO2019169767A1 (en) * | 2018-03-06 | 2019-09-12 | 平安科技（深圳）有限公司 | Pornographic picture identification method, electronic device, and readable storage medium |
| CN108491866A (en) * | 2018-03-06 | 2018-09-04 | 平安科技（深圳）有限公司 | Porny identification method, electronic device and readable storage medium storing program for executing |
| CN108734170B (en) * | 2018-05-25 | 2022-05-03 | 电子科技大学 | License plate character segmentation method based on machine learning and template |
| CN108734170A (en) * | 2018-05-25 | 2018-11-02 | 电子科技大学 | Registration number character dividing method based on machine learning and template |
| CN110569836A (en) * | 2018-06-06 | 2019-12-13 | 北京深鉴智能科技有限公司 | variable-length character string identification method and device |
| CN110569836B (en) * | 2018-06-06 | 2022-07-12 | 赛灵思电子科技(北京)有限公司 | Variable-length character string identification method and device |
| CN110858306A (en) * | 2018-08-22 | 2020-03-03 | 西门子（中国）有限公司 | License plate character recognition apparatus, method and computer-readable storage medium |
| CN109635637A (en) * | 2018-10-30 | 2019-04-16 | 深圳市航天华拓科技有限公司 | A kind of licence plate recognition method, device and calculate equipment |
| CN109583451A (en) * | 2018-11-28 | 2019-04-05 | 上海鹰觉科技有限公司 | Automatic identifying method and system based on warship ship side number |
| CN109784334A (en) * | 2019-01-24 | 2019-05-21 | 合肥视展光电科技有限公司 | Round-the-clock round-the-clock licence plate recognition method, system, device and path identification method |
| CN110348396B (en) * | 2019-07-15 | 2022-02-11 | 南京信息工程大学 | Deep learning-based method and device for recognizing character traffic signs above roads |
| CN110348396A (en) * | 2019-07-15 | 2019-10-18 | 南京信息工程大学 | A kind of road top text traffic sign and device based on deep learning |
| CN110569833A (en) * | 2019-09-06 | 2019-12-13 | 上海应用技术大学 | License plate positioning method |
| CN111881914A (en) * | 2020-06-23 | 2020-11-03 | 安徽清新互联信息科技有限公司 | License plate character segmentation method and system based on self-learning threshold |
| CN111881914B (en) * | 2020-06-23 | 2024-02-13 | 安徽清新互联信息科技有限公司 | License plate character segmentation method and system based on self-learning threshold |
| CN114913515A (en) * | 2021-12-31 | 2022-08-16 | 北方工业大学 | End-to-end license plate recognition network construction method |
| CN114913515B (en) * | 2021-12-31 | 2024-04-02 | 北方工业大学 | End-to-end license plate recognition network construction method |
| CN114463757A (en) * | 2022-01-28 | 2022-05-10 | 上海电机学院 | Industrial scene character end-side reasoning training device and method based on machine vision |
| CN114612730A (en) * | 2022-04-06 | 2022-06-10 | 哈尔滨工业大学 | Method and device for detecting household garbage classification generation proportion |
| CN114612730B (en) * | 2022-04-06 | 2023-08-29 | 哈尔滨工业大学 | Method and device for detecting household garbage classification generation proportion |
| CN115375626A (en) * | 2022-07-25 | 2022-11-22 | 浙江大学 | Medical image segmentation method, system, medium, and apparatus based on physical resolution |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106096602A (en) | Chinese license plate recognition method based on convolutional neural network | |
| Yang et al. | Deep detection network for real-life traffic sign in vehicular networks | |
| Du et al. | Weak and occluded vehicle detection in complex infrared environment based on improved YOLOv4 | |
| Qian et al. | Robust Chinese traffic sign detection and recognition with deep convolutional neural network | |
| Sun et al. | Traffic sign detection and recognition based on convolutional neural network | |
| CN106022300B (en) | Traffic sign recognition method and system based on cascade deep study | |
| CN104166841B (en) | The quick detection recognition methods of pedestrian or vehicle is specified in a kind of video surveillance network | |
| CN102043945B (en) | License plate character recognition method based on real-time vehicle tracking and binary index classification | |
| CN111767882A (en) | Multi-mode pedestrian detection method based on improved YOLO model | |
| CN109543606A (en) | A kind of face identification method that attention mechanism is added | |
| CN105956626A (en) | Deep learning based vehicle license plate position insensitive vehicle license plate recognition method | |
| CN106845487A (en) | A kind of licence plate recognition method end to end | |
| CN107945153A (en) | A kind of road surface crack detection method based on deep learning | |
| CN109598268A (en) | A kind of RGB-D well-marked target detection method based on single flow depth degree network | |
| CN113221655B (en) | Face spoofing detection method based on feature space constraint | |
| CN104517103A (en) | Traffic sign classification method based on deep neural network | |
| CN104504395A (en) | Method and system for achieving classification of pedestrians and vehicles based on neural network | |
| CN109543632A (en) | A kind of deep layer network pedestrian detection method based on the guidance of shallow-layer Fusion Features | |
| Wang et al. | Traffic sign detection using a cascade method with fast feature extraction and saliency test | |
| CN108681735A (en) | Optical character recognition method based on convolutional neural networks deep learning model | |
| CN110334703B (en) | Ship detection and identification method in day and night image | |
| CN111274886B (en) | Deep learning-based pedestrian red light running illegal behavior analysis method and system | |
| CN110956158A (en) | Pedestrian shielding re-identification method based on teacher and student learning frame | |
| CN110781882A (en) | License plate positioning and identifying method based on YOLO model | |
| CN107220598A (en) | Iris Texture Classification based on deep learning feature and Fisher Vector encoding models |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into 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: 20161109 |