CN107038442A - A kind of car plate detection and global recognition method based on deep learning - Google Patents
A kind of car plate detection and global recognition method based on deep learning Download PDFInfo
- Publication number
- CN107038442A CN107038442A CN201710187289.6A CN201710187289A CN107038442A CN 107038442 A CN107038442 A CN 107038442A CN 201710187289 A CN201710187289 A CN 201710187289A CN 107038442 A CN107038442 A CN 107038442A
- Authority
- CN
- China
- Prior art keywords
- car plate
- detection
- deep learning
- plate
- confidence level
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/63—Scene text, e.g. street names
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
- G06V20/54—Surveillance or monitoring of activities, e.g. for recognising suspicious objects of traffic, e.g. cars on the road, trains or boats
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/625—License plates
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Biology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention provides a kind of car plate detection and global recognition method based on deep learning, including:Step a:Vehicle detection is carried out, target vehicle is obtained, whole car plate detection region is determined by target vehicle;Step b:By the car plate detection region division into n fritter, partly overlapped between each fritter;Step c:The license plate area obtained substantially is fitted using the first deep neural network learning model, while obtaining the confidence level that the region is true car plate;Step d:According to the position of the license plate area and the relation of confidence level, fusion obtains final license plate area;And step e:Overall identification is carried out to the number-plate number of the license plate area by the second deep neural network structure.The present invention reduces the time of car plate detection using the license plate area regression technique of piecemeal;By introducing deep learning model, car plate detection discrimination is improved, overcoming traditional licence plate recognition method needs to carry out the dependence brought of License Plate Character Segmentation and the misrecognition problem caused by result uncertainty.
Description
Technical field
The present invention relates to intelligent transportation system and digital safety-security area, and in particular to a kind of car plate inspection based on deep learning
Survey and global recognition method.
Background technology
Car license recognition is the foundation and important step in intelligent transportation system (ITS) field, to intelligent transportation system
The other parts such as analysis of vehicle-logo recognition, body color, model analysis, bayonet socket detection, electronic police, and other traffic events are related
The processing quality of module has important influence.
Traditional Vehicle License Plate Recognition System, is roughly divided into car plate detection, License Plate, Character segmentation and character recognition etc. several
Submodule.Its car plate detection link is scanned classification using detector in full figure or area-of-interest (ROI), to consider chi
The problems such as degree and transverse and longitudinal ratio, amount of calculation is than larger;License Plate is as above determined due to by light, picture quality, car plate below
The influence of the factors such as size is difficult to accurately, and successive character is split and recognizes that influence is very big.Before the heavy dependence of Character segmentation part
The result of positioning, its used binarization method is for some special circumstances such as negative and positive board, and car plate light is unbalanced and other effects
It is not good.Character recognition then uses the single character recognition mode of traditional shallow-layer grader, though it is not bad under normal operation, entirely
The stability of weather and the ga s safety degree of similar character are difficult to ensure that.
Therefore, there is problems with traditional Vehicle License Plate Recognition System:
1st, conventional method car plate detection needs to travel through sliding window, than relatively time-consuming;
2nd, conventional method needs to be accurately positioned car plate and split again, and dependence is strong, easy segmentation errors;
3rd, conventional method light difference, resolution ratio is low, be stained, tilt big situations such as stability it is bad;
4th, single model is difficult to ensure that preferable distinguishing indexes.
The content of the invention
The problem of present invention exists for prior art, proposes a kind of car plate detection based on deep learning and overall identification
Method.
A kind of car plate detection and global recognition method based on deep learning of the present invention, comprises the following steps:
Step a:Vehicle detection is carried out, target vehicle is obtained, whole car plate detection region is determined by target vehicle;
Step b:By the car plate detection region division into n fritter, partly overlapped between each fritter;
Step c:The license plate area obtained substantially is fitted using the first deep neural network model, while obtaining the region is
The confidence level of true car plate;
Step d:According to the position of the license plate area and the relation of confidence level, fusion obtains final license plate area;With
And
Step e:Overall identification is carried out to the number-plate number of the license plate area by the second deep neural network model.
Preferably, in step b, n=5 or 9.
Preferably, in step c, the number range of the confidence level is 0~1,0 to represent that the confidence level of true car plate is 0,1 table
The confidence level for showing true car plate is 100%.
Preferably, first deep neural network model includes 3 convolutional layers and 3 full articulamentums being sequentially connected,
Last full articulamentum include Liang Ge branches, one branch into prediction license plate area rectangle the upper left corner and the lower right corner sit
Mark, another branches into the classification of true and false car plate.
Preferably, in step c, the method for fusion further comprises:
Step d1:Result is ranked up from big to small by the confidence level of true car plate;
Step d2:Binarization operation is done to ranking results, you can reliability is compared with selected threshold value, if higher than threshold value
True car plate is then considered, otherwise it is assumed that being false car plate;
Step d3:Obtain license plate candidate area in true car plate set, as truly car plate collection be combined into sky if the target vehicle
Regional determination is without car plate.
Preferably, in step d3, by Max strategies in true car plate set, selection confidence level highest testing result is
License plate candidate area;Or be weighted by Avg strategies in the car plate detection position of true car plate set according to its confidence level flat
Obtain license plate candidate area.
Preferably, car plate position optimization step d ' is also included between step d, e, using deep learning model to license plate area
Optimize, the license plate area after being optimized.
Preferably, the second depth network model includes 4 convolutional layers and 2 full articulamentums being sequentially connected, finally
One full articulamentum includes seven branches, and each branch corresponds to Chinese character, letter, numeral or mark on car plate respectively.
Preferably, step f is also included after step e, by the way of multiple model integrateds, synthesis is identified result:
Step f1:Count the number of identical number-plate number recognition result and calculate its confidence level average value;
Step f2:How much sorted from big to small according to number, can by it if the different number-plate numbers is the same number of
Reliability sorts from high to low;
Step f3:It is recognition result to select the confidence level highest number-plate number.
Preferably, in step f, by the way of 3 model integrateds.
The present invention has the advantages that:
1st, the present invention reduces the time of car plate detection using the license plate area regression technique of piecemeal;
2nd, present invention introduces deep learning model, car plate detection discrimination is improved;
3rd, the present invention overcomes traditional licence plate recognition method to need to carry out dependence and result that License Plate Character Segmentation is brought
Misrecognition problem caused by uncertainty;
4th, car plate detection identification of the present invention is in various robustness in particular cases.
Brief description of the drawings
Fig. 1 is the car plate detection based on deep learning of the present invention and the schematic flow sheet of global recognition method.
Fig. 2 is 5 pieces of sub-zone dividing schematic diagrames of car plate detection of the invention.
Fig. 3 is car plate detection depth network structure of the invention.
Fig. 4 is sub-block car plate detection result schematic diagram of the invention.
Fig. 5 is car plate detection position optimization schematic diagram of the invention.
Fig. 6 is overall license plate recognition result schematic diagram of the invention.
Embodiment
Below by embodiment, the invention will be further described, and its purpose is only that the research for more fully understanding the present invention
The protection domain that content is not intended to limit the present invention.
With reference to Fig. 1, of the invention is a kind of based on deep learning, the particularly car plate detection of convolutional neural networks and overall knowledge
Other method, comprises the following steps a~e.
Step a:Vehicle detection is carried out, target vehicle is obtained, whole car plate detection region is determined by target vehicle.
The purpose of vehicle detection is that real vehicle region is detected in whole camera fields of view field, exclude background area and
The part of interference, so as to reduce flase drop, improves the accuracy rate of car plate detection, and reduce the time loss of subsequent treatment.This hair
The bright method detected using existing conventional vehicles, for example, can use motion analysis technique, background model be set up, before extraction
Scape, can also use wagon detector to carry out Static Detection, and carry out target association.Or both combine, really felt emerging
The target vehicle of interest.
Step b:By the car plate detection region division into n fritter, partly overlapped between each fritter.
Traditional car plate detection technique, by the way of sliding window, travels through each subwindow in car plate detection region, finds side
Edge is abundant, gradient energy is changed greatly, and the region of similar vehicle license plate characteristic is used as license plate candidate area.Disadvantage of this is that consumption
Shi Duo, detection efficiency is not high.The present invention carries out car plate detection using the license plate area regression technique based on piecemeal.Specific practice
It is that, by whole car plate detection region division into a few fritter, a preferred embodiment is between 5 fritters, fritter
Divide overlapping.
Specifically, car plate detection region R (w, h) can be obtained by vehicle detection region, or true from artificial or program
Fixed area-of-interest, and some fritters are divided into, concurrently detected respectively, then result is merged.This hair
Bright segment partition scheme is using 5 pieces of (2 × 2+1) methods, or 9 palace lattice (3 × 3), and more fine granularity is then not necessarily to, especially with 5 pieces of methods
Based on.5 pieces divide as shown in Fig. 2 each sub-block locations are respectively:Upper left, upper right, lower-left, bottom right, the center in car plate detection region
It is located at car plate detection region R center, the high width per sub-block is respectively w*2/3, h*2/3.Overlapped each other between sub-block, be
In order to handle the situation that car plate target is located at sub-block boundaries.
Step c:License plate area substantially is obtained using deep learning models fitting to each fritter, while obtaining the region
It is the confidence level of true car plate.
Specifically, appropriate deep learning model is used each sub-block, and fitting obtains license plate area substantially, depth
Practise model and provide the confidence level that the region is true car plate simultaneously.The number range of confidence level represents the credible of true car plate from 0~1,0
Spend for 0,1 represents 100% firmly believes to be true car plate.
Consider effect and efficiency, the deep neural network model that the present invention is used is convolutional Neural as shown in Figure 3
Network (CNN).In the convolutional neural networks, including 3 convolutional layers (Conv) being sequentially connected and 3 full articulamentums (IP),
Last full articulamentum include Liang Ge branches, one branch into prediction car plate rectangle the upper left corner and bottom right angular coordinate, make
Lost with L2, another branches into the classification of true and false car plate, is lost using SoftMaxLoss.Each one pond of convolutional layer heel
Change layer (Max Pool) (not showing pond layer in Fig. 3).Conv layers and IP layers of parameter is shown in shown in Fig. 3, is come for convolutional layer
Say, numeral represents width height, the port number of feature map (i.e. input picture) and convolution kernel, and for IP layers, numeral represents it
Neuron number.
As the first convolutional layer (as figure 1. → 2.) five parameters (30,30,3,3,3), represent the input picture of this layer
Wide height is all 30, and has 3 passages (R, G, B), and this layer of convolution operation core size is 3 × 3.Second convolutional layer (as figure 2. → 3.)
In three parameters (10,3,3), represent that the layer has 10 passages (i.e. the feature map of preceding layer are output as 10), layer volume
The size of product operation core is 3 × 3.3rd convolutional layer (as figure 3. → 4.) in three parameters (20,2,2), represent that the layer has 20
Individual passage, the size of this layer of convolution operation core is 2 × 2, finally obtains the layer that port number is 40 (as shown in 4.).5. first is represented
Full articulamentum, neuron number is 120,6. represents the second full articulamentum, neuron number is 60,7. represents the 3rd full articulamentum
That is last articulamentum, with Liang Ge branches, the neuron number of a branch is 4, another branch for 2.In network
Conv layers of stride (step-length) are that 1, Pool layers of core size are that 2 × 2, stride is 2.In addition to last IP layers, each Conv
ReLU activation primitives are all used with IP layers.
As described above, being that the fitting of car plate position and confidence level computing function can be achieved by the network shown in Fig. 3, that is, input
One artwork, calculates by the network (being obtained by training), is that can obtain confidence value and car plate square in last output layer
Shape region upper left, bottom right angular coordinate.
Step d:According to the position of the license plate area and the relation of confidence level, fusion obtains final license plate area.This
Sample one, no matter detection zone size be it is how many, no matter yardstick and transverse and longitudinal ratio change number, because block count is fixed, car plate
The time complexity of detection is constant, i.e. O (1).The purpose of car plate is returned used here as depth model, is for utilization depth
The advantage of the self-teaching of habit, the depth characteristic stablized in all cases.
The fusion of car plate detection result above-mentioned, fusion method includes Max, the strategy such as Avg.
Specifically, this step further comprises:
Step d1:Result is ranked up from big to small by the confidence level of true car plate;
Step d2:Binarization operation is done to ranking results, you can reliability is compared with selected threshold value, if higher than threshold value
True car plate is then considered, otherwise it is assumed that being false car plate;
Step d3:Obtain license plate candidate area in true car plate set, as truly car plate collection be combined into sky if the target vehicle
Regional determination is without car plate.Wherein, by Max strategies in true car plate set, selection confidence level highest testing result is car
Board candidate region;Or be weighted by Avg strategies in the car plate detection position of true car plate set according to its confidence level average
Obtain license plate candidate area.
Step e is described below:Overall identification is carried out to the number-plate number of the license plate area by depth network structure.
Either traditional license plate recognition technology, is still currently based on some Vehicle License Plate Recognition Systems of deep learning, substantially
Character recognition one by one again after will being split to characters on license plate, due to by location accuracy and car plate surface image quality
Influence, the effect of Character segmentation is often unsatisfactory, so as to greatly reduce final number-plate number discrimination.
Overall character recognition technologies based on deep learning model have had some practical applications, such as Google engineer
Using the continuous number in deep learning Model Identification streetscape, Oxonian researcher uses CNN Network Recognition natural scenes
In English character sequence.The overall character identifying method of deep learning model is applied to intelligent transportation scene by the present invention, can
Mitigate it and the characteristics of stability requirement is big split to conventional characters, the powerful feature learning ability of deep learning can also be utilized,
Raising includes number-plate number discrimination and the robustness under special case.
The present invention is similar to Fig. 3 for the depth network structure that the number-plate number is recognized, preferred embodiment includes connecting successively
Conv layers and 2 IP layers of four connect.Sample input size is 64 × 64,3 passages, and sample here is that above step is obtained
The image of the license plate area obtained.Conv1 cores size 5 × 5, feature map are output as 64, Conv2 cores size 5 × 5,
Feature map are output as 128, Conv3 cores size 3 × 3, and feature map are output as 256, Conv4 cores size 3 × 3,
Feature map are output as 512, and all Conv layers of stride are 1, are activated using ReLU, each one Max of Conv layers of heel
Pool layers, core size 2 × 2, stride is 2.IP1 neuron number (or being output) is 4096, is activated using ReLU,
IP2 layers have 7 branches, and its neuron number is respectively 58,26,36,36,36,36,41, is lost using SoftMaxLoss.
Here IP2 uses 7 branches, and Chinese character, letter, numeral and mark etc. on car plate are corresponded to respectively.For example, first point
Branch correspondence Chinese character, including 31 provinces such as Shanghai, Anhui, Jiangxi referred to as and some special board " army ", " sky ", " seas " etc. totally 58 it is defeated
Go out, the second branch correspondence English alphabet, totally 26 output, the three~six branch correspondence " numeral+letter ", thus totally 36 it is defeated
Go out, the 7th branch correspondence " letter+numeral+special board mark " (special board mark includes trailer mark etc.), totally 41 output.Often
The result of each output of individual branch is the probability of the output item, and the probability sum of each output of each branch is 1, is taken general
The maximum item of rate as the branch final output.For example, the first branch has 58 output items, the probability in such as " Shanghai " is
98%, the probability in " Anhui " is 1%, and the probability in " Jiangxi " is 1%, and other output items are 0, then maximum probability is 98% here
" Shanghai ", then the output of first branch is " Shanghai ".By regarding the license plate area image of foregoing acquisition as deep neural network
Input, by calculate, then can integrally recognize the acquisition number-plate number.
Except customized CNN, some conventional networks such as GoogLeNet, VGGNet etc. can also be applied to the present invention
Method.
Further, method of the invention can also include car plate position optimization step d ' between step d, e, using depth
Degree learning model is optimized to license plate area, the license plate area after being optimized.Car plate is obtained by above-mentioned steps d fusions to examine
Survey behind region, in order to further improve the position of car plate, be easy to follow-up Car license recognition module, the present invention is right after car plate detection
Car plate predicted position is optimized, and optimization is same to use deep learning model, and general simple network can be used, not special herein
Do not mentionlet alone bright.Optimize schematic diagram as shown in Figure 5.
Meanwhile, also include step f after discrimination, above-mentioned steps e of the invention to further improve, using multiple moulds
The integrated mode of type, synthesis is identified result, it is preferred that can be specifically included by the way of 3 model integrateds:
Step f1:Count the number of identical number-plate number recognition result and calculate its confidence level average value;
Step f2:How much sorted from big to small according to number, can by it if the different number-plate numbers is the same number of
Reliability sorts from high to low;
Step f3:It is recognition result to select the confidence level highest number-plate number.
For example there are 5 recognition results, Shanghai A12345, Shanghai A12345, Shanghai A72345, Shanghai A72345, Shanghai A72344, confidence level
Respectively 0.998,0.996,0.945,0.941,0.9, then ranking results be:Shanghai A12345, Shanghai A72345, Shanghai A72344, can
Certainty value is:0.997,0.943,0.9, Shanghai A12345 is finally taken to predict the outcome.
Obviously, those of ordinary skill in the art is it should be appreciated that the embodiment of the above is intended merely to explanation originally
Invention, and be not used as limitation of the invention, as long as in the spirit of the present invention, to embodiment described above
Change, modification will all fall in claims of the present invention model.
Claims (10)
1. a kind of car plate detection and global recognition method based on deep learning, it is characterised in that comprise the following steps:
Step a:Vehicle detection is carried out, target vehicle is obtained, whole car plate detection region is determined by target vehicle;
Step b:By the car plate detection region division into n fritter, partly overlapped between each fritter;
Step c:The license plate area obtained substantially is fitted using the first deep neural network model, while it is true car to obtain the region
The confidence level of board;
Step d:According to the position of the license plate area and the relation of confidence level, fusion obtains final license plate area;And
Step e:Overall identification is carried out to the number-plate number of the license plate area by the second deep neural network model.
2. a kind of car plate detection and global recognition method based on deep learning according to claim 1, it is characterised in that
In step b, n=5 or 9.
3. a kind of car plate detection and global recognition method based on deep learning according to claim 1, it is characterised in that
In step c, the number range of the confidence level is that the confidence level of 0~1, the 0 true car plate of expression is the confidence level of 0, the 1 true car plate of expression
For 100%.
4. a kind of car plate detection and global recognition method based on deep learning according to claim 1, it is characterised in that
First deep neural network model includes 3 convolutional layers and 3 full articulamentums being sequentially connected, last full articulamentum
Comprising Liang Ge branches, one branch into prediction license plate area rectangle the upper left corner and bottom right angular coordinate, another is branched into very
The classification of false car plate.
5. a kind of car plate detection and global recognition method based on deep learning according to claim 1, it is characterised in that
In step c, the method for fusion further comprises:
Step d1:Result is ranked up from big to small by the confidence level of true car plate;
Step d2:Binarization operation is done to ranking results, you can reliability is compared with selected threshold value, recognized if threshold value is higher than
To be true car plate, otherwise it is assumed that being false car plate;
Step d3:Obtain license plate candidate area in true car plate set, as truly car plate collection be combined into sky if the target vehicle region
It is determined as no car plate.
6. a kind of car plate detection and global recognition method based on deep learning according to claim 5, it is characterised in that
In step d3, by Max strategies in true car plate set, selection confidence level highest testing result is license plate candidate area;Or
Person is weighted average acquisition car plate candidate regions in the car plate detection position of true car plate set by Avg strategies according to its confidence level
Domain.
7. a kind of car plate detection and global recognition method based on deep learning according to claim 1, it is characterised in that
Also include car plate position optimization step d ' between step d, e, license plate area is optimized using deep learning model, obtain excellent
License plate area after change.
8. a kind of car plate detection and global recognition method based on deep learning according to claim 1, it is characterised in that
The second depth network model includes 4 convolutional layers and 2 full articulamentums being sequentially connected, and last full articulamentum is included
Seven branches, each branch corresponds to Chinese character, letter, numeral or mark on car plate respectively.
9. a kind of car plate detection and global recognition method based on deep learning according to claim 1, it is characterised in that
Also include step f after step e, by the way of multiple model integrateds, synthesis is identified result:
Step f1:Count the number of identical number-plate number recognition result and calculate its confidence level average value;
Step f2:How much sorted from big to small according to number, if the different number-plate numbers is the same number of, by its confidence level
Sort from high to low;
Step f3:It is recognition result to select the confidence level highest number-plate number.
10. a kind of car plate detection and global recognition method based on deep learning according to claim 9, its feature exist
In in step f, by the way of 3 model integrateds.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710187289.6A CN107038442A (en) | 2017-03-27 | 2017-03-27 | A kind of car plate detection and global recognition method based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710187289.6A CN107038442A (en) | 2017-03-27 | 2017-03-27 | A kind of car plate detection and global recognition method based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107038442A true CN107038442A (en) | 2017-08-11 |
Family
ID=59533482
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710187289.6A Pending CN107038442A (en) | 2017-03-27 | 2017-03-27 | A kind of car plate detection and global recognition method based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107038442A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107679483A (en) * | 2017-09-27 | 2018-02-09 | 北京小米移动软件有限公司 | Number plate recognition methods and device |
CN108509954A (en) * | 2018-04-23 | 2018-09-07 | 合肥湛达智能科技有限公司 | A kind of more car plate dynamic identifying methods of real-time traffic scene |
WO2019174405A1 (en) * | 2018-03-14 | 2019-09-19 | 台达电子工业股份有限公司 | License plate identification method and system thereof |
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 |
WO2020155300A1 (en) * | 2019-01-31 | 2020-08-06 | 网宿科技股份有限公司 | Model prediction method and device |
CN111915897A (en) * | 2019-05-10 | 2020-11-10 | 北京万集科技股份有限公司 | Method and device for determining position of identification area, storage medium and electronic device |
CN112560608A (en) * | 2020-12-05 | 2021-03-26 | 江苏爱科赛尔云数据科技有限公司 | Vehicle license plate recognition method |
CN113313115A (en) * | 2021-06-11 | 2021-08-27 | 浙江商汤科技开发有限公司 | License plate attribute identification method and device, electronic equipment and storage medium |
CN114677500A (en) * | 2022-05-25 | 2022-06-28 | 松立控股集团股份有限公司 | Weak surveillance video license plate recognition method based on eye tracker point annotation information |
CN112560608B (en) * | 2020-12-05 | 2024-05-24 | 江苏爱科赛尔云数据科技有限公司 | Vehicle license plate recognition method |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6373962B1 (en) * | 1998-04-24 | 2002-04-16 | Ume Tech Yugen Kaisha | License plate information reader device for motor vehicles |
CN104182728A (en) * | 2014-07-26 | 2014-12-03 | 佳都新太科技股份有限公司 | Vehicle logo automatic location and recognition method based on pattern recognition |
CN105160292A (en) * | 2015-07-07 | 2015-12-16 | 深圳市捷顺科技实业股份有限公司 | Vehicle identification recognition method and system |
CN105956626A (en) * | 2016-05-12 | 2016-09-21 | 成都新舟锐视科技有限公司 | Deep learning based vehicle license plate position insensitive vehicle license plate recognition method |
CN106056086A (en) * | 2016-06-02 | 2016-10-26 | 江苏华通晟云科技有限公司 | Vehicle brand and model identification method based on fast learning framework |
-
2017
- 2017-03-27 CN CN201710187289.6A patent/CN107038442A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6373962B1 (en) * | 1998-04-24 | 2002-04-16 | Ume Tech Yugen Kaisha | License plate information reader device for motor vehicles |
CN104182728A (en) * | 2014-07-26 | 2014-12-03 | 佳都新太科技股份有限公司 | Vehicle logo automatic location and recognition method based on pattern recognition |
CN105160292A (en) * | 2015-07-07 | 2015-12-16 | 深圳市捷顺科技实业股份有限公司 | Vehicle identification recognition method and system |
CN105956626A (en) * | 2016-05-12 | 2016-09-21 | 成都新舟锐视科技有限公司 | Deep learning based vehicle license plate position insensitive vehicle license plate recognition method |
CN106056086A (en) * | 2016-06-02 | 2016-10-26 | 江苏华通晟云科技有限公司 | Vehicle brand and model identification method based on fast learning framework |
Non-Patent Citations (1)
Title |
---|
HUI LI 等: "Reading Car License Plates Using Deep Convolutional Neural Networks and LSTMs", 《ARXIV:1601.05610V1 [CS.CV]》 * |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107679483A (en) * | 2017-09-27 | 2018-02-09 | 北京小米移动软件有限公司 | Number plate recognition methods and device |
WO2019174405A1 (en) * | 2018-03-14 | 2019-09-19 | 台达电子工业股份有限公司 | License plate identification method and system thereof |
US11443535B2 (en) | 2018-03-14 | 2022-09-13 | Delta Electronics, Inc. | License plate identification method and system thereof |
CN108509954A (en) * | 2018-04-23 | 2018-09-07 | 合肥湛达智能科技有限公司 | A kind of more car plate dynamic identifying methods of real-time traffic scene |
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 |
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 |
WO2020155300A1 (en) * | 2019-01-31 | 2020-08-06 | 网宿科技股份有限公司 | Model prediction method and device |
CN111915897A (en) * | 2019-05-10 | 2020-11-10 | 北京万集科技股份有限公司 | Method and device for determining position of identification area, storage medium and electronic device |
CN111915897B (en) * | 2019-05-10 | 2021-11-16 | 北京万集科技股份有限公司 | Method and device for determining position of identification area, storage medium and electronic device |
CN112560608A (en) * | 2020-12-05 | 2021-03-26 | 江苏爱科赛尔云数据科技有限公司 | Vehicle license plate recognition method |
CN112560608B (en) * | 2020-12-05 | 2024-05-24 | 江苏爱科赛尔云数据科技有限公司 | Vehicle license plate recognition method |
CN113313115A (en) * | 2021-06-11 | 2021-08-27 | 浙江商汤科技开发有限公司 | License plate attribute identification method and device, electronic equipment and storage medium |
CN113313115B (en) * | 2021-06-11 | 2023-08-04 | 浙江商汤科技开发有限公司 | License plate attribute identification method and device, electronic equipment and storage medium |
CN114677500A (en) * | 2022-05-25 | 2022-06-28 | 松立控股集团股份有限公司 | Weak surveillance video license plate recognition method based on eye tracker point annotation information |
CN114677500B (en) * | 2022-05-25 | 2022-08-23 | 松立控股集团股份有限公司 | Weak surveillance video license plate recognition method based on eye tracker point annotation information |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107085723A (en) | A kind of characters on license plate global recognition method based on deep learning model | |
CN107038442A (en) | A kind of car plate detection and global recognition method based on deep learning | |
CN110321923B (en) | Target detection method, system and medium for fusion of different-scale receptive field characteristic layers | |
CN105160309B (en) | Three lanes detection method based on morphological image segmentation and region growing | |
CN110084292A (en) | Object detection method based on DenseNet and multi-scale feature fusion | |
CN109934163A (en) | A kind of aerial image vehicle checking method merged again based on scene priori and feature | |
CN102609686B (en) | Pedestrian detection method | |
CN108830171B (en) | Intelligent logistics warehouse guide line visual detection method based on deep learning | |
CN107871126A (en) | Model recognizing method and system based on deep-neural-network | |
CN106845487A (en) | A kind of licence plate recognition method end to end | |
CN109284669A (en) | Pedestrian detection method based on Mask RCNN | |
CN109472298A (en) | Depth binary feature pyramid for the detection of small scaled target enhances network | |
CN104809443A (en) | Convolutional neural network-based license plate detection method and system | |
CN107463892A (en) | Pedestrian detection method in a kind of image of combination contextual information and multi-stage characteristics | |
CN113420607A (en) | Multi-scale target detection and identification method for unmanned aerial vehicle | |
CN109726717A (en) | A kind of vehicle comprehensive information detection system | |
CN110119726A (en) | A kind of vehicle brand multi-angle recognition methods based on YOLOv3 model | |
CN110097044A (en) | Stage car plate detection recognition methods based on deep learning | |
CN106408030A (en) | SAR image classification method based on middle lamella semantic attribute and convolution neural network | |
CN105005764A (en) | Multi-direction text detection method of natural scene | |
CN109492596A (en) | A kind of pedestrian detection method and system based on K-means cluster and region recommendation network | |
CN108985145A (en) | The Opposite direction connection deep neural network model method of small size road traffic sign detection identification | |
CN111738300A (en) | Optimization algorithm for detecting and identifying traffic signs and signal lamps | |
CN110210418A (en) | A kind of SAR image Aircraft Targets detection method based on information exchange and transfer learning | |
CN109815802A (en) | A kind of monitor video vehicle detection and recognition method based on convolutional neural networks |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170811 |
|
RJ01 | Rejection of invention patent application after publication |