CN109165643A - 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
- CN109165643A CN109165643A CN201810955796.4A CN201810955796A CN109165643A CN 109165643 A CN109165643 A CN 109165643A CN 201810955796 A CN201810955796 A CN 201810955796A CN 109165643 A CN109165643 A CN 109165643A
- Authority
- CN
- China
- Prior art keywords
- license plate
- indicate
- characters
- candidate
- character
- 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.)
- Granted
Links
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
- G06V10/23—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 based on positionally close patterns or neighbourhood relationships
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/245—Aligning, centring, orientation detection or correction of the image by locating a pattern; Special marks for positioning
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Character Input (AREA)
- Character Discrimination (AREA)
Abstract
The invention discloses a kind of licence plate recognition methods based on deep learning, include the following steps: step 1: the training one depth convolutional neural networks model M for characters on license plate detection;Construct characters on license plate tag set B;Step 2: the license plate image I that positioning obtains being input to characters on license plate detection network M, exports candidate license plate character set H;Step 3: the candidate license plate character set H that step 2 is obtained, by the upper left corner abscissa h of candidate license plate character minimum circumscribed rectanglei.x ascending that H reorders, obtain set C;Step 4: screening operation further being executed to the set C that step 3 obtains, step 6: the set E obtained to step 5 sequentially traverses set E;Step 7: returning and license plate recognition result L is obtained by step 6.The influence for situations such as the beneficial effects of the present invention are: effectively inhibiting characters on license plate adhesion, fracture, deformation, license plate is stained, and shade is remained on license plate sloped and license plate.
Description
Technical field
The present invention relates to field of intelligent transportation technology, and in particular to a kind of licence plate recognition method based on deep learning.
Background technique
Over the last couple of decades, license plate recognition technology is all greatly improved on accuracy of identification and efficiency of algorithm.
With being constantly progressive for wisdom traffic system the relevant technologies, automatic license plate image recognition is considered as one and possesses mature solution party
The settled problem of case.Traffic flow analysis, vehicle speed measuring, vehicle violation detection are many answering based on license plate recognition technology
Representative.However, the standard specification of license plate is very more in practice, license plate font, color have notable difference, and characters on license plate is long
It spends different.Only just there are blue bottom wrongly written or mispronounced character uniline license plate, yellow bottom black word uniline license plate, yellow bottom black word duplicate rows license plate, black matrix wrongly written or mispronounced character in China
The multinomial license plate type of uniline license plate etc. ten.In addition, licence plate recognition method is easy by illumination, resolution ratio, imaged viewing angle, shade etc.
The interference of environmental factor.
Computer mould apery identifies license plate, which includes positioning and identification.Therefore, complete automatic Car license recognition system
System is made of License Plate and Car license recognition two parts.License Plate is not in the range of this method discussion.So far, it has deposited
In many licence plate recognition methods.Method can be classified as to conventional method and deep learning method two major classes.Traditional Car license recognition
Process includes two step of License Plate Character Segmentation and Recognition of License Plate Characters.Process can be divided into Character segmentation and word by deep learning method
Symbol two steps of identification, end-to-end can also settle at one go.
Traditional licence plate recognition method is Task-decomposing at two subtasks of License Plate Character Segmentation and Recognition of License Plate Characters.License plate
Character segmentation is the premise of Car license recognition, Character segmentation it is accurate whether directly influence recognition result.With image procossing skill
The continuous development of art, a variety of registration number character dividing methods are suggested, such as: character sciagraphy, and Character mother plate matching method and character connect
Logical domain method.Common license plate character recognition method is mainly based upon the OCR of machine learning, such as: the multiword based on support vector machines
It accords with classifying identification method and the multiword based on artificial neural network accords with classifying identification method.OCR method based on machine learning is first
The feature of character picture is first extracted, then training obtains Classification and Identification model, is eventually used for the identification mission of single character.But
Situations such as being that, due to characters on license plate adhesion, fracture, deformation, license plate is stained, shade is remained on license plate sloped and license plate, is very normal
See, traditional licence plate recognition method is limited by the selection and calculating of validity feature, cannot reach ideal recognition effect.
Deep learning method effectively inhibits characters on license plate adhesion, fracture, deformation, license plate is dirty by calculating convolution feature
Damage, the license plate sloped influence with situations such as remaining shade on license plate.Existing one stream licence plate recognition method is all deep learning side
Method.Deep learning method is mainly divided to method and end-to-end two class of method based on segmentation.
Deep learning method based on segmentation still continues to use the process of traditional licence plate recognition method, by Car license recognition task point
For two subtasks of Character segmentation and character recognition.By deep learning learn convolution to convolution feature come handle two sons appoint
Business.Document (Y.Yang, D.Li and Z.Duan, " Chinese vehicle license plate recognition
using kernel-based extreme learning machine with deep convolutional
features,"in IET Intelligent Transport Systems,vol.12,no.3,pp.213-219,42018.)
The learning machine that transfinites (kernel-based extreme learning machine, KELM) based on core is applied to vehicle for the first time
In board character recognition, Softmax is replaced with it to obtain classification results.Author is extracting the KELM classification 31 of depth convolution feature
A chinese character (31 Chinese provinces are referred to as), although training speed is than very fast, low in the recognition efficiency of model, resource is accounted for
With height.Document (license plate recognition technology research [D] Qingdao of the Zhao Zhenxing based on deep learning: Qingdao University of Science and Technology, 2017) base
Complete Car license recognition task in two stages in SSD: the first stage is characters on license plate detection, i.e. segmentation simultaneously and identification license plate word
Symbol;Second stage is that characters on license plate process of aggregation is obtained by the characters on license plate set permutation and combination for obtaining identification that reorders
Final license plate recognition result.Although this method can handle uniline license plate and duplicate rows license plate simultaneously, license plate resolution ratio not
Gao Shi, this method first stage, there are erroneous detection and missing inspection, the permutation and combination of second stage can not solve the problems, such as the first stage.Text
Offer (Hefei research [D] of Recognition Algorithm of License Plate under soup jade Yao complex background: China Science & Technology University, 2016.) it is being based on
Connected domain method is divided after obtaining characters on license plate, after obtaining disaggregated model with deep learning training, one by one to the character divided
Identification, obtains final recognition result.This method has a defect that traditional Character segmentation based on connected domain is ineffective, into
And influence subsequent identification.
End-to-end licence plate recognition method no longer needs individual License Plate Character Segmentation, inputs a license plate image, output
License plate recognition result.Document (Hong-Hyun Kim, Je-Kang Park, Joo-Hee Oh, Dong-Joong
Kang.Multi-task convolutional neural network system for license plate
recognition[J].International Journal ofControl,Automation and Systems,2017,15
(6): 2942-2949 the end-to-end Car license recognition of single phase) is realized based on depth convolutional neural networks, but the network mechanism
It is fixed, it can only identify South Korea's license plate of fixed format, computational efficiency is not high.Document (Jakub Spanhel, Jakub Sochor,
Roman Juranek,Adam Herout,Lukas Marsik,Pavel Zemcik.Holistic recognition of
low quality license plates by CNN using track annotated data.14th IEEE
International Conference on Advanced Video and Signal Based Surveillance
(AVSS), Lecce, Italy, Aug 29-Sep 01,2017.) propose the neural network comprising 8 full connection branches, energy
Enough identification contains up to the low resolution license plate of 8 characters, and each branch is responsible for character machining and the identification in a region.But
It is that the network only considered uniline license plate, does not account for the license plate of other standards.
Deep learning method based on segmentation has that error accumulation, the accuracy of Character segmentation affect subsequent
Character recognition, and the diversity of license plate standard increases the difficulty of Character segmentation.Although end-to-end licence plate recognition method exists
It is dominant simultaneously in efficiency and effect, but this kind of network is high to the required precision of License Plate, and depth network design difficulty limits
The license plate standard that can be applicable in is made.
In conclusion there are the following shortcomings for licence plate recognition method at present: 1) license plate standard is varied, and Car license recognition is difficult
To take into account the license plate of all standards;2) there is mistake in the method based on segmentation (including conventional method and based on the method for deep learning)
The segmentation error that poor accumulating questions, i.e. segmentation stage generate can accumulate the subsequent Recognition of License Plate Characters stage, reduce whole knowledge
Not rate;3) although end-to-end Car license recognition network can have been able to the multiple types of processing variable character length in design
License plate, but the license plate with similar standard can only be still handled, the network suitable for uniline license plate cannot take on duplicate rows license plate
Ideal effect is obtained, and such network needs premised on accurate License Plate, condition is harsh;4) a variety of vehicles can be handled simultaneously
The licence plate recognition method of board standard requires license plate imaging definition high, it is difficult to successfully handle low-resolution image;5) environment
Factor (such as: illumination, noise) has a certain impact to the identification of license plate image.
Summary of the invention
For the disadvantages mentioned above for overcoming the prior art, the present invention proposes a kind of licence plate recognition method based on deep learning,
After deep learning detects characters on license plate, first sorts to set element, then set element is screened, finally reconfigure
To license plate recognition result character string.
Technical scheme is as follows:
A kind of licence plate recognition method based on deep learning, which comprises the steps of:
Step 1: the training one depth convolutional neural networks model M for characters on license plate detection;Construct characters on license plate mark
Sign set B={ bi| i=1,2 ..., n, n=67 }=' 0 ', ' 1 ', ' 2 ', ' 3 ', ' 4 ', ' 5 ', ' 6 ', ' 7 ', ' 8 ', ' 9 ',
‘A’,‘B’,‘C’,‘D’,‘E’,‘F’,‘G’,‘H’,‘I’,‘J’,‘K’,‘L’,‘M’,‘N’,‘O’,‘P’,‘Q’,‘R’,‘S’,
' T ', ' U ', ' V ', ' W ', ' X ', ' Y ', ' Z ', ' capital ', ' saliva ', ' Ji ', ' Shanxi ', ' illiteracy ', ' the Liao Dynasty ', ' Ji ', ' black ', ' Shanghai ',
' Soviet Union ', ' Zhejiang ', ' Anhui ', ' Fujian ', ' Jiangxi ', ' Shandong ', ' Henan ', ' Hubei Province ', ' Hunan ', ' Guangdong ', ' osmanthus ', ' fine jade ', ' Chongqing ', ' river ', ' expensive ',
' cloud ', ' hiding ', ' Shan ', ' sweet ', ' blueness ', ' peaceful ', ' new ' };
Step 2: the license plate image I that positioning obtains being input to characters on license plate detection network M, exports candidate license plate character set
Close H={ hi| i=1,2,3 ..., nH, wherein nHIndicate the element number of set H, hiIndicate i-th of candidate license plate of set H
Character, hiIt is the triple being made of (b, q, r), b indicates that the class label of candidate characters, b ∈ B, q indicate setting for candidate characters
Reliability, q ∈ [0,1], r are the four-tuples being made of (x, y, w, h), and r indicates that the boundary rectangle frame of candidate characters, x, y, w and h divide
Not Biao Shi rectangle frame upper left corner abscissa, upper left corner ordinate, width and height;
Step 3: the candidate license plate character set H that step 2 is obtained, by the upper left of candidate license plate character minimum circumscribed rectangle
Angle abscissa hi.x ascending that H reorders, obtain set C={ ci| i=1,2,3 ..., nC, ciIndicate the i-th of set C
A candidate license plate character, ciIt is the triple being made of (b, q, r), nCIndicate the element number of set C;
Step 4: screening operation, specific steps are further executed to the set C that step 3 obtains are as follows:
Step 4.1: C is sequentially traversed, it is rightI=1,2 ..., nCIf ci.q<Tq, then by ciSet D is added, and will
ciIt is deleted from set C;Otherwise by ciStay in set C;Finally, obtaining D={ di| i=1,2 ..., nD, diIt is by (b, q, r) structure
At triple, nDIndicate the quantity of element in set D;Wherein, ci.q candidate characters c is indicatediConfidence level, TqIndicate confidence
Spend threshold value;
Step 4.2: the set C that sequentially traversal step 4.1 obtains is rightI=1,2 ..., nC- 1, by formula (1)
A is calculated to delete element a from set C if a ≠ φ;Otherwise, operation is not executed;
Wherein, ci.r candidate characters c is indicatediBoundary rectangle frame, ci+1.r candidate characters c is indicatedi+1Boundary rectangle frame,
ToverIndicate the anti-eclipse threshold of rectangle frame, | | indicate the area of rectangle frame, ∩ operation indicates to seek the intersection of two rectangle frames;
Step 4.3: the set D that the set C and step 4.1 obtain to step 4.2 is obtained is rightI=1,2 ...,
nC- 1, ifMeet formula (2), records so that djMeet the i value of formula (2), and by djSet C is inserted into as the
(i+1) a element;Wherein, djIndicate j-th of element of set D, j=1,2 ..., nD;
Wherein, ci.r.x、ci.r.y、ciAnd c .r.wi.r.h candidate characters c is respectively indicatediThe upper left corner of boundary rectangle frame is horizontal
Coordinate, upper left corner ordinate, width and height;dj.r.x、dj.r.y、djAnd d .r.wj.r.h candidate characters d is respectively indicatedjIt is external
Upper left corner abscissa, upper left corner ordinate, width and the height of rectangle frame;
Step 5: the set C that step 4.3 is obtained, firstly, set C is sequentially traversed, ifI=1,2 ..., nC,
Meet formula (3), by ciSet E is added, and by ciIt is deleted from C;Then, the surplus element for traversing set C, is added sequentially to
Set E;Finally obtain E={ ei| i=1,2,3 ..., nE, eiIndicate i-th of candidate license plate character of set E, nEIndicate set E
Element number;
Step 6: the set E obtained to step 5 sequentially traverses set E, the label of all elements in set E is connected into
License plate recognition resultWherein, ei.b ∈ B, ei.b the label of i-th of candidate license plate character of set E, Σ table are indicated
Show character serial operation;
Step 7: algorithm terminates.
The beneficial effects of the present invention are:
1) the present invention is based on deep learning realize characters on license plate detection and identification, effectively inhibit characters on license plate adhesion,
The influence for situations such as fracture, deformation, license plate are stained, and shade is remained on license plate sloped and license plate.
2) present invention completes Car license recognition by subsequent screening operation and permutation and combination, does not limit the license plate that can be identified
Character length is not high to license plate resolution requirement suitable for the uniline license plate and duplicate rows license plate of a variety of license plate standards.
3) compared to the licence plate recognition method based on segmentation, the present invention is avoided segmentation step bring error accumulation to knowledge
Other step, accuracy of identification are higher.
4) compared to end-to-end licence plate recognition method, the present invention is lower to the required precision of License Plate, and not receiving end
Limitation in correspondent network design, can identify the license plate of more multi-standard.
Detailed description of the invention
Fig. 1 is the license plate image of input of the invention.
Fig. 2 is the testing result visual image of the invention by depth convolutional neural networks to input license plate.
Fig. 3 is the result visualization image of the invention handled by step 4.
Fig. 4 is the final result visual image of the invention handled by step 7.
Specific embodiment
The specific implementation of the licence plate recognition method of the invention based on deep learning is elaborated below with reference to embodiment
Mode.
Step 1: the training one depth convolutional neural networks model M for characters on license plate detection;Construct characters on license plate mark
Sign set B={ bi| i=1,2 ..., n, n=67 }=' 0 ', ' 1 ', ' 2 ', ' 3 ', ' 4 ', ' 5 ', ' 6 ', ' 7 ', ' 8 ', ' 9 ',
‘A’,‘B’,‘C’,‘D’,‘E’,‘F’,‘G’,‘H’,‘I’,‘J’,‘K’,‘L’,‘M’,‘N’,‘O’,‘P’,‘Q’,‘R’,‘S’,
' T ', ' U ', ' V ', ' W ', ' X ', ' Y ', ' Z ', ' capital ', ' saliva ', ' Ji ', ' Shanxi ', ' illiteracy ', ' the Liao Dynasty ', ' Ji ', ' black ', ' Shanghai ',
' Soviet Union ', ' Zhejiang ', ' Anhui ', ' Fujian ', ' Jiangxi ', ' Shandong ', ' Henan ', ' Hubei Province ', ' Hunan ', ' Guangdong ', ' osmanthus ', ' fine jade ', ' Chongqing ', ' river ', ' expensive ',
' cloud ', ' hiding ', ' Shan ', ' sweet ', ' blueness ', ' peaceful ', ' new ' };
Step 2: the license plate image I that positioning obtains being input to characters on license plate detection network M, exports candidate license plate character set
Close H={ hi| i=1,2,3 ..., nH, wherein nHIndicate the element number of set H, hiIndicate i-th of candidate license plate of set H
Character, hiIt is the triple being made of (b, q, r), b indicates that the class label of candidate characters, b ∈ B, q indicate setting for candidate characters
Reliability, q ∈ [0,1], r are the four-tuples being made of (x, y, w, h), and r indicates that the boundary rectangle frame of candidate characters, x, y, w and h divide
Not Biao Shi rectangle frame upper left corner abscissa, upper left corner ordinate, width and height;In this example, license plate image is tested such as
Shown in Fig. 1, deep neural network M testing result is as shown in Figure 2;
Step 3: the candidate license plate character set H that step 2 is obtained, by the upper left of candidate license plate character minimum circumscribed rectangle
Angle abscissa hi.x ascending that H reorders, obtain set C={ ci| i=1,2,3 ..., nC, ciIndicate the i-th of set C
A candidate license plate character, ciIt is the triple being made of (b, q, r), nCIndicate the element number of set C;
Step 4: screening operation, specific steps are further executed to the set C that step 3 obtains are as follows:
Step 4.1: C is sequentially traversed, it is rightI=1,2 ..., nCIf ci.q<Tq, then by ciSet D is added, and will
ciIt is deleted from set C;Otherwise by ciStay in set C;Finally, obtaining D={ di| i=1,2 ..., nD, diIt is by (b, q, r) structure
At triple, nDIndicate the quantity of element in set D;Wherein, ci.q candidate characters c is indicatediConfidence level, TqIndicate confidence
Spend threshold value;In this example, Tq=0.5;
Step 4.2: the set C that sequentially traversal step 4.1 obtains is rightI=1,2 ..., nC- 1, by formula (1)
A is calculated to delete element a from set C if a ≠ φ;Otherwise, operation is not executed;
Wherein, ci.r candidate characters c is indicatediBoundary rectangle frame, ci+1.r candidate characters c is indicatedi+1Boundary rectangle frame,
ToverIndicate the anti-eclipse threshold of rectangle frame, | | indicate the area of rectangle frame, ∩ operation indicates to seek the intersection of two rectangle frames;
In this example, Tover=0.6;
Step 4.3: the set D that the set C and step 4.1 obtain to step 4.2 is obtained is rightI=1,2 ...,
nC- 1, ifMeet formula (2), records so that djMeet the i value of formula (2), and by djSet C is inserted into as the
(i+1) a element;Wherein, djIndicate j-th of element of set D, j=1,2 ..., nD;In this example, the knot of final set C
Fruit visual image is as shown in Figure 3;
Wherein, ci.r.x、ci.r.y、ciAnd c .r.wi.r.h candidate characters c is respectively indicatediThe upper left corner of boundary rectangle frame is horizontal
Coordinate, upper left corner ordinate, width and height;dj.r.x、dj.r.y、djAnd d .r.wj.r.h candidate characters d is respectively indicatedjIt is external
Upper left corner abscissa, upper left corner ordinate, width and the height of rectangle frame;
Step 5: the set C that step 4.3 is obtained, firstly, set C is sequentially traversed, ifI=1,2 ..., nC,
Meet formula (3), by ciSet E is added, and by ciIt is deleted from C;Then, the surplus element for traversing set C, is added sequentially to
Set E;Finally obtain E={ ei| i=1,2,3 ..., nE, eiIndicate i-th of candidate license plate character of set E, nEIndicate set E
Element number;
Step 6: the set E obtained to step 5 sequentially traverses set E, the label of all elements in set E is connected into
License plate recognition resultWherein, ei.b ∈ B, ei.b the label of i-th of candidate license plate character of set E, Σ table are indicated
Show character serial operation;In this example, L is printed upon in original image and obtains final result visual image, as shown in Figure 4;
Step 7: algorithm terminates.
Claims (1)
1. a kind of licence plate recognition method based on deep learning, which comprises the steps of:
Step 1: the training one depth convolutional neural networks model M for characters on license plate detection;Construct characters on license plate tally set
Close B={ bi| i=1,2 ..., n, n=67 }=' 0 ', ' 1 ', ' 2 ', ' 3 ', ' 4 ', ' 5 ', ' 6 ', ' 7 ', ' 8 ', ' 9 ', ' A ',
‘B’,‘C’,‘D’,‘E’,‘F’,‘G’,‘H’,‘I’,‘J’,‘K’,‘L’,‘M’,‘N’,‘O’,‘P’,‘Q’,‘R’,‘S’,‘T’,
' U ', ' V ', ' W ', ' X ', ' Y ', ' Z ', ' capital ', ' saliva ', ' Ji ', ' Shanxi ', ' illiteracy ', ' the Liao Dynasty ', ' Ji ', ' black ', ' Shanghai ', ' Soviet Union ',
' Zhejiang ', ' Anhui ', ' Fujian ', ' Jiangxi ', ' Shandong ', ' Henan ', ' Hubei Province ', ' Hunan ', ' Guangdong ', ' osmanthus ', ' fine jade ', ' Chongqing ', ' river ', ' expensive ', ' cloud ',
' hiding ', ' Shan ', ' sweet ', ' blueness ', ' peaceful ', ' new ' };
Step 2: the license plate image I that positioning obtains being input to characters on license plate detection network M, exports candidate license plate character set H
={ hi| i=1,2,3 ..., nH, wherein nHIndicate the element number of set H, hiIndicate i-th of candidate license plate word of set H
Symbol, hiIt is the triple being made of (b, q, r), b indicates that the class label of candidate characters, b ∈ B, q indicate the confidence of candidate characters
Degree, q ∈ [0,1], r are the four-tuples being made of (x, y, w, h), and r indicates the boundary rectangle frame of candidate characters, x, y, w and h difference
Indicate upper left corner abscissa, upper left corner ordinate, width and the height of rectangle frame;
Step 3: the candidate license plate character set H that step 2 is obtained, it is horizontal by the upper left corner of candidate license plate character minimum circumscribed rectangle
Coordinate hi.x ascending that H reorders, obtain set C={ ci| i=1,2,3 ..., nC, ciIndicate i-th of time of set C
Select characters on license plate, ciIt is the triple being made of (b, q, r), nCIndicate the element number of set C;
Step 4: screening operation, specific steps are further executed to the set C that step 3 obtains are as follows:
Step 4.1: C is sequentially traversed, it is rightIf ci.q<Tq, then by ciSet D is added, and by ciFrom
It is deleted in set C;Otherwise by ciStay in set C;Finally, obtaining D={ di| i=1,2 ..., nD, diIt is to be made of (b, q, r)
Triple, nDIndicate the quantity of element in set D;Wherein, ci.q candidate characters c is indicatediConfidence level, TqIndicate confidence level
Threshold value;
Step 4.2: the set C that sequentially traversal step 4.1 obtains is rightA is calculated by formula (1),
If a ≠ φ, element a is deleted from set C;Otherwise, operation is not executed;
Wherein, ci.r candidate characters c is indicatediBoundary rectangle frame, ci+1.r candidate characters c is indicatedi+1Boundary rectangle frame, Tover
Indicate the anti-eclipse threshold of rectangle frame, | | indicate the area of rectangle frame, ∩ operation indicates to seek the intersection of two rectangle frames;
Step 4.3: the set D that the set C and step 4.1 obtained according to step 4.2 is obtained is right
IfMeet formula (2), records so that djMeet the i value of formula (2), and by djSet C is inserted into as (i+1)
A element;Wherein, djIndicate j-th of element of set D, j=1,2 ..., nD;
Wherein, ci.r.x、ci.r.y、ciAnd c .r.wi.r.h candidate characters c is respectively indicatediThe horizontal seat in the upper left corner of boundary rectangle frame
Mark, upper left corner ordinate, width and height;dj.r.x、dj.r.y、djAnd d .r.wj.r.h candidate characters d is respectively indicatedjExternal square
Upper left corner abscissa, upper left corner ordinate, width and the height of shape frame;
Step 5: the set C that step 4.3 is obtained, firstly, set C is sequentially traversed, if Meet public
Formula (3), by ciSet E is added, and by ciIt is deleted from C;Then, the surplus element for traversing set C, is added sequentially to set E;
Finally obtain E={ ei| i=1,2,3 ..., nE, eiIndicate i-th of candidate license plate character of set E, nEIndicate the element of set E
Number;
Step 6: the set E obtained to step 5 sequentially traverses set E, the label of all elements in set E is connected into license plate
Recognition resultWherein, ei.b ∈ B, ei.b indicate that the label of i-th of candidate license plate character of set E, Σ indicate word
Accord with serial operation;
Step 7: algorithm terminates.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810955796.4A CN109165643B (en) | 2018-08-21 | 2018-08-21 | License plate recognition method based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810955796.4A CN109165643B (en) | 2018-08-21 | 2018-08-21 | License plate recognition method based on deep learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109165643A true CN109165643A (en) | 2019-01-08 |
CN109165643B CN109165643B (en) | 2021-11-16 |
Family
ID=64896372
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810955796.4A Active CN109165643B (en) | 2018-08-21 | 2018-08-21 | License plate recognition method based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109165643B (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN110287959A (en) * | 2019-06-27 | 2019-09-27 | 浙江工业大学 | A kind of licence plate recognition method based on recognition strategy again |
CN110288031A (en) * | 2019-06-27 | 2019-09-27 | 浙江工业大学 | A kind of licence plate recognition method based on Sequence Learning |
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 |
CN110717493A (en) * | 2019-09-16 | 2020-01-21 | 浙江大学 | License plate recognition method containing stacked characters based on deep learning |
CN110852324A (en) * | 2019-08-23 | 2020-02-28 | 上海撬动网络科技有限公司 | Deep neural network-based container number detection method |
CN111461112A (en) * | 2020-03-03 | 2020-07-28 | 华南理工大学 | License plate character recognition method based on double-cycle transcription network |
CN113723408A (en) * | 2021-11-02 | 2021-11-30 | 上海仙工智能科技有限公司 | License plate recognition method and system and readable storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102663377A (en) * | 2012-03-15 | 2012-09-12 | 华中科技大学 | Character recognition method based on template matching |
WO2015147764A1 (en) * | 2014-03-28 | 2015-10-01 | Kisa Mustafa | A method for vehicle recognition, measurement of relative speed and distance with a single camera |
US20150307278A1 (en) * | 2014-04-29 | 2015-10-29 | Flextronics Ap, Llc | Method and system for order fulfillment |
CN106650731A (en) * | 2016-12-23 | 2017-05-10 | 中山大学 | Robust license plate and logo recognition method |
CN107292898A (en) * | 2017-05-04 | 2017-10-24 | 浙江工业大学 | A kind of car plate shadow Detection and minimizing technology based on HSV |
-
2018
- 2018-08-21 CN CN201810955796.4A patent/CN109165643B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102663377A (en) * | 2012-03-15 | 2012-09-12 | 华中科技大学 | Character recognition method based on template matching |
WO2015147764A1 (en) * | 2014-03-28 | 2015-10-01 | Kisa Mustafa | A method for vehicle recognition, measurement of relative speed and distance with a single camera |
US20150307278A1 (en) * | 2014-04-29 | 2015-10-29 | Flextronics Ap, Llc | Method and system for order fulfillment |
CN106650731A (en) * | 2016-12-23 | 2017-05-10 | 中山大学 | Robust license plate and logo recognition method |
CN107292898A (en) * | 2017-05-04 | 2017-10-24 | 浙江工业大学 | A kind of car plate shadow Detection and minimizing technology based on HSV |
Non-Patent Citations (3)
Title |
---|
JIAN W. FANG: ""The research for license plate recognition using sub-image fast independent component analysis"", 《2011 CHINESE CONTROL AND DECISION CONFERENCE (CCDC)》 * |
YASUKO TAKAHASHI等: ""License plate recognition using gray image template matching with noise reduction filters and character alignment"", 《SYSTEMS AND COMPUTERS IN JAPAN》 * |
王洪元等: ""基于多特征组合的车牌识别算法"", 《常州大学学报(自然科学版)》 * |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN110020650B (en) * | 2019-03-26 | 2021-08-03 | 武汉大学 | Inclined license plate recognition method and device based on deep learning recognition model |
CN110287959B (en) * | 2019-06-27 | 2021-06-29 | 浙江工业大学 | License plate recognition method based on re-recognition strategy |
CN110287959A (en) * | 2019-06-27 | 2019-09-27 | 浙江工业大学 | A kind of licence plate recognition method based on recognition strategy again |
CN110288031A (en) * | 2019-06-27 | 2019-09-27 | 浙江工业大学 | A kind of licence plate recognition method based on Sequence Learning |
CN110288031B (en) * | 2019-06-27 | 2021-07-27 | 浙江工业大学 | License plate recognition method based on sequence learning |
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 |
CN110852324A (en) * | 2019-08-23 | 2020-02-28 | 上海撬动网络科技有限公司 | Deep neural network-based container number detection method |
CN110717493A (en) * | 2019-09-16 | 2020-01-21 | 浙江大学 | License plate recognition method containing stacked characters based on deep learning |
CN110717493B (en) * | 2019-09-16 | 2022-04-01 | 浙江大学 | License plate recognition method containing stacked characters based on deep learning |
CN111461112A (en) * | 2020-03-03 | 2020-07-28 | 华南理工大学 | License plate character recognition method based on double-cycle transcription network |
CN111461112B (en) * | 2020-03-03 | 2023-02-14 | 华南理工大学 | License plate character recognition method based on dual-cycle transcription network |
CN113723408A (en) * | 2021-11-02 | 2021-11-30 | 上海仙工智能科技有限公司 | License plate recognition method and system and readable storage medium |
CN113723408B (en) * | 2021-11-02 | 2022-02-25 | 上海仙工智能科技有限公司 | License plate recognition method and system and readable storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN109165643B (en) | 2021-11-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109165643A (en) | A kind of licence plate recognition method based on deep learning | |
CN105046196B (en) | Front truck information of vehicles structuring output method based on concatenated convolutional neutral net | |
CN105608456B (en) | A kind of multi-direction Method for text detection based on full convolutional network | |
WO2017031716A1 (en) | Method for analyzing and recognizing handwritten mathematical formula structure in natural scene image | |
CN100357988C (en) | Identification method of mobile number plate based on three-channel parallel artificial nerve network | |
CN113344857B (en) | Defect detection network training method, defect detection method and storage medium | |
CN104778470B (en) | Text detection based on component tree and Hough forest and recognition methods | |
CN105205488B (en) | Word area detection method based on Harris angle points and stroke width | |
CN103577817A (en) | Method and device for identifying forms | |
CN109190625B (en) | Large-angle perspective deformation container number identification method | |
CN102880859A (en) | Method for recognizing number plate | |
CN112381175A (en) | Circuit board identification and analysis method based on image processing | |
CN111967313A (en) | Unmanned aerial vehicle image annotation method assisted by deep learning target detection algorithm | |
CN113096085A (en) | Container surface damage detection method based on two-stage convolutional neural network | |
CN111860570A (en) | Cloud particle image extraction and classification method | |
CN111340032A (en) | Character recognition method based on application scene in financial field | |
CN110516673A (en) | Ancient Books in Yi Language character detection method based on connected component and regression equation character segmentation | |
CN114387592B (en) | Character positioning and identifying method under complex background | |
CN107392115B (en) | Traffic sign identification method based on hierarchical feature extraction | |
CN110991374A (en) | Fingerprint singular point detection method based on RCNN | |
CN111414938A (en) | Target detection method for bubbles in plate heat exchanger | |
CN109508714B (en) | Low-cost multi-channel real-time digital instrument panel visual identification method and system | |
Sharma et al. | Pincode detection using deep CNN for postal automation | |
CN112884741B (en) | Printing apparent defect detection method based on image similarity comparison | |
CN116862883A (en) | Concrete slump detection method based on image semantic segmentation |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |