CN108009548A - A kind of Intelligent road sign recognition methods and system - Google Patents
A kind of Intelligent road sign recognition methods and system Download PDFInfo
- Publication number
- CN108009548A CN108009548A CN201810019360.4A CN201810019360A CN108009548A CN 108009548 A CN108009548 A CN 108009548A CN 201810019360 A CN201810019360 A CN 201810019360A CN 108009548 A CN108009548 A CN 108009548A
- Authority
- CN
- China
- Prior art keywords
- image
- road sign
- module
- text
- intelligent road
- 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
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/14—Image acquisition
- G06V30/148—Segmentation of character regions
- G06V30/153—Segmentation of character regions using recognition of characters or words
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Character Discrimination (AREA)
Abstract
The invention discloses a kind of Intelligent road sign recognition methods and system, this method is to gather guideboard image by image capture module, by 3G communication modules by the image transmitting of collection to Cloud Server, Cloud Server pre-processes the image of collection, character area extracts, Text segmentation handles and Text region, and recognition result is sent to central control module by 3G communication modules, the text message of recognition result is changed into voice messaging finally by voice broadcast module, user is informed by earphone or loudspeaker；Intelligent road sign identifying system, including central control module, central control module are connected to voice broadcast module and 3G communication modules, and 3G communication modules are connected with Cloud Server, and central control module is also associated with image capture module.The present invention can carry out intelligent recognition to guideboard, and driver will be avoided to divert one's attention when driving and cause traffic jam and traffic safety, promoted the development of intelligent transportation by text information by voice broadcast.
Description
Technical field
The present invention relates to intelligent transportation field, particularly a kind of Intelligent road sign recognition methods and system.
Background technology
In life of today, the vehicles of the automobile as the main force, the trip being convenient for people to, improves the life matter of the people
Amount.But growing with automobile quantity, the demand of road is also continuously increased, the renewal of road and build so that transportation network
It is increasingly complicated various.When people drive a car in strange environment, the information such as road conditions thick as hail, Sign Board, which easily allows, drives
The person's of sailing increase uncertainty that pressure and then increase drive at heart, thus easily leads to various the unlawful practices even generation of accident,
Traffic is allowed to become more severe.Resource loss of the China every year caused by traffic jam issue reaches tens billion of members,
And fast and effectively solution, this loss are still constantly rising due to lacking.
Intelligent transportation system (Intelligent Transport System, ITS), as a kind of innovative transit systems
A variety of advanced technologies are combined, the much informations such as people, car, road are subjected to intensive processing, can be made in a wide range of, comprehensive performance
With, and congested traffic condition can be alleviated, guarantee driving safety, improves traffic efficiency.It is initially by U.S.'s intelligent transportation
Meeting (CITS America) proposes, is referred to as Intelligent road Vehicular system (IVHS, Intelligent Vehicle- in early days
Highway System).Nowadays the navigation system used is based on GPS (Global Positioning System, global location
System) and storing map, easily lead to navigation error when GPS failures or map fail to upgrade in time.Traffic based on intelligent recognition
Navigation system is the developing direction of following navigation, and guideboard identification is one of part.Guideboard is that common traffic is led
To mark, its feature easily identifies, can submit necessary information for driver, guiding driver arrives at.Guideboard one
As be placed in the top or both sides of road, need wholeheartedly two to use since driver reads guideboard information in driving procedure, and by
In reasons such as speed, weather, road conditions, the time that driver can obtain information is shorter, is often needed in the case where being unfamiliar with road
By vehicle deceleration with reading information, security risk is thus brought.Therefore guideboard information is effectively obtained based on intelligence system and incited somebody to action
Information is conveyed to the research direction that driver is intelligent automobile.These information also can be used in nowadays studying being closed extensively
The pilotless automobile neighborhood of note.
In guideboard character identification system, working-flow is mainly that system receiving front-end collecting device collects
Picture, is pre-processed.Then the segmentation of guideboard character area is carried out, the image with Chinese character is carried out to the positioning of chinese character
And identification.Oneself identification word is played back in a voice form finally by read module.User is allowed also can so as to reach
" listening " arrives the effect of picture.The chief component of system is Text region algorithm, its main object is Chinese character.Chinese character is China
The representative and crystallization of culture, undergo the development to thousands of years of history of modern Chinese character by the inscriptions on bones or tortoise shells, are unique in four major acids ancient country
Letter symbol so far can be used.Relative to the character of Romance, a big feature of Chinese character is exactly the diversity of its character.
And since Chinese character is the pictograph of structuring, even if a Chinese character is split, the quantity of its local form is still very
It is more.In nowadays this age of knowledge explosion, how the word in real world effectively to be imported into computer and is carried out accurate
Identifying processing, is a good problem to study.
The content of the invention
It is an object of the present invention to provide a kind of Intelligent road sign recognition methods and system.The present invention can carry out intelligence to guideboard
It can identify, and driver will be avoided to divert one's attention when driving and cause traffic jam and traffic to pacify by text information by voice broadcast
Entirely, the development of intelligent transportation has been promoted.
Technical scheme：A kind of Intelligent road sign recognition methods, guideboard image is gathered by image capture module, by
The natural scene character image of collection is transmitted to Cloud Server by 3G communication modules, and Cloud Server locates the image of collection in advance
Reason, the extraction of character area obtain the character image of background, finally carry out Text segmentation processing and Text region, and will identification
As a result central control module is sent to by 3G communication modules with text formatting, finally by voice broadcast module by recognition result
Text message change into voice messaging, user is informed by earphone or loudspeaker.
In a kind of foregoing Intelligent road sign recognition methods, the Cloud Server is by image analysis processing system, is used
Traditional image processing method completes the pretreatment to natural scene character image and the region segmentation of image, utilizes deep learning
Neural network algorithm, identifies the text information in cut zone.
In a kind of foregoing Intelligent road sign recognition methods, the Text region, is made using deep learning neutral net
Model for word training is completed, and is produced Chinese character image training dataset using computer programming, is existed using the data set of generation
The network model of training design on server.
Intelligent road sign identifying system used in a kind of foregoing Intelligent road sign recognition methods, including central control module, in
Centre control module is connected to voice broadcast module and 3G communication modules, and 3G communication modules are connected with Cloud Server, center control
Molding block is also associated with image capture module.
In foregoing Intelligent road sign identifying system, the voice broadcast module includes phonetic synthesis unit and and phonetic synthesis
The voice broadcast unit of unit connection.
The image capture module of the present invention, the image in road wing board is established in collection, and the image of collection is carried out
MJPEG format compressions（MJPEG refers to Motion JPRG, i.e. motion jpeg）.
Speech processing module, is connected with central control module, for reporting word recognition result on guideboard to driver.
3G communication modules, establish and are in communication with each other in real time for front end guideboard Image Acquisition and Cloud Server；
Central control module, the data collected to each functional unit and information carry out Macro or mass analysis, and anti-according to data institute
The scene reflected, corresponding control instruction is sent to each function module；The guideboard image collected is sent out using 3G communication modules
Send to Cloud Server；
Cloud Server, image analysis system is established by building software platform on Cloud Server, builds deep learning convolution god
Through network, and to complete to train, the Chinese character picture of segmentation is sent into trained network model is identified, and by Text region knot
Fruit feeds back to front center control module；
The literary Chinese characters recognition method of the present invention, is that deep learning nerve is utilized on Cloud Server by Intelligent road sign identifying system
The model that network is trained as word, produces Chinese character image training dataset using computer programming, uses the data set of generation
The network model of training design on the server.The text image for being collected image capture module by 3G communication modules uploads
To Cloud Server, the image of collection is pre-processed using digital figure treatment technology, the processing such as Text segmentation, then by having instructed
The model perfected completes character identification function, and recognition result is sent back front center control module with text formatting, finally
The text message of recognition result is changed into by voice messaging by voice broadcast module, use is informed by earphone or loudspeaker
Person.
Compared with prior art, present invention incorporates machine vision, Digital Image Processing, scene Text region, depth
The cutting edge technologies such as habit, computer network, realize the image segmentation to complex background, the identification to Chinese character.Design convolutional Neural
Network, identifies Chinese character, the large increase accuracy rate of Chinese Character Recognition, effectively can feed back to user by guideboard information in real time.
This has very wide application prospect and application value in intelligent transportation in future and unpiloted application.In conclusion
The present invention can carry out guideboard intelligent recognition, and will by text information by voice broadcast, avoid driver divert one's attention when driving and
Traffic jam and traffic safety are caused, has promoted the development of intelligent transportation.
Brief description of the drawings
Fig. 1 is the overall structure block diagram of Intelligent road sign identifying system；
Fig. 2 is the operational flow diagram of Intelligent road sign identifying system front end；
Fig. 3 is the operational flow diagram of Intelligent road sign identifying system server；
Fig. 4 is front-end collection and the figure of region guideboard extraction；
Fig. 5 is guideboard image level direction projection experimental result picture；
Fig. 6 is guideboard image vertical direction projection experiments result figure；
Fig. 7 is guideboard pictograph segmentation result figure；
Fig. 8 is training set and test set part figure；
Fig. 9 is CNN implementation model simplification figures；
Figure 10 is verbal model training network illustraton of model；
Figure 11 is SoftmaxWithLoss layers of schematic diagram；
Figure 12 is accuracy and the relation of iterations.
Embodiment
The present invention is further illustrated with reference to the accompanying drawings and examples, but be not intended as to the present invention limit according to
According to.
Embodiment.A kind of Intelligent road sign recognition methods, as shown in Figures 1 to 12：Guideboard is gathered by image capture module
Image, is at this time the image under natural scene, by 3G communication modules by the image transmitting of collection to Cloud Server, Cloud Server pair
The image of collection is pre-processed, the extraction of character area obtains depositing text image, carries out Text segmentation processing and Text region,
And recognition result is sent to central control module with text formatting by 3G communication modules, will finally by voice broadcast module
The text message of recognition result changes into voice messaging, and user is informed by earphone or loudspeaker.
The Cloud Server is by image analysis processing system, is completed with traditional image processing method to image
Pretreatment and the region segmentation of image, using deep learning neural network algorithm, identify the text information in cut zone；It is described
Text region, be model completion train using deep learning neutral net as word, use the computer programming generation Chinese
Word image training dataset, uses the data set network model that training designs on the server of generation.
Intelligent road sign identifying system used in above-mentioned Intelligent road sign recognition methods, including central control module, center control
Module is connected to voice broadcast module and 3G communication modules, and 3G communication modules are connected with Cloud Server, central control module
It is also associated with image capture module.
The voice broadcast module includes phonetic synthesis unit and the voice broadcast unit being connected with phonetic synthesis unit.
Operation principle：The Software for Design part of the system mainly includes programming and Intelligent road sign knowledge on Cloud Server
The programming of other system front end.Software for Design on Cloud Server be mainly the image of front-end collection is pre-processed, area
Domain is extracted and text segmentation, and the identification function of word is finally completed using trained model, finally sends out recognition result again
Send to front end main control unit.The programming of Intelligent road sign identifying system front end mainly including camera guideboard Image Acquisition,
Report and keys interrupt configuration of communication connection, voice broadcast module of the 3G communication modules to front end and server to recognition result
Program etc. function module is write.The overall structure block diagram of the system is as shown in Figure 1, specific workflow such as Fig. 2 and Fig. 3 institutes
Show.
1 natural scene Text region algorithm designs
The core of Intelligent road sign identifying system is the character image that image capture module collects is identified.Because collection
Image belong to image under natural scene, pre-processed first, the extraction of guideboard character area, obtain character image into style of writing
Word is split has whole identifying system critically important influence with link, the quality of each link processing such as Text regions.
Since present guideboard scene species is various, the design for typical scene-i.e. green bottom wrongly written or mispronounced character and and indigo plant bottom wrongly written or mispronounced character,
Words direction is from left from right guideboard scene.In the image such as Fig. 4 of front-end collection shown in (a), server is sent to front end
Image, first passes through Canny operator edge detections coarse positioning and MSER algorithm fine positionings, using SVM algorithm to determine whether containing
Guideboard, guideboard extracted region is carried out after judging guideboard mark to image.
1.1 slant correction
Take pictures or it is other acquisition picture during, can more or less cause the part of picture character to tilt, in order to improve
Word carries out Slant Rectify, by studying and testing this kind of the comprising only found for OCR identifications by the accuracy of row segmentation
The image of the characters such as word, numeral, letter is relatively good using the method effect of Hough transform.
Hough takes a series of processing step such as the superiors' image and then extraction image border by generating image pyramid
Then the angle of inclination for detecting image and direction that can be more accurate carry out rotation correction, test result shows this algorithm essence
Exactness is higher.
1.2 Contrast enhanced
The purpose handled using Contrast enhanced is sought to the difference of the gray value of enhancing literal line and line space and reduces its own
Gray difference, so as to improve recognition effect.
For the average image of grey value profile, its visual effect is better than the image of other distributions, gray value
Uneven embodiment be background colour may gray value between a very big scope, and character color equally also has very big model
Enclose, for this phenomenon, we can carry out processing so as to highlight the contrast of character and background by the method for histogram equalization
Degree.
1.3 guideboard character areas extract
Using the method based on HSI color spaces.In view of used in life mostly using green bottom and blue bottom as guideboard background
H components in HSI spaces come out guideboard extracted region.By H parameter settings green bottom and blue bottom can identified section, obtain
Go out in experimental result such as Fig. 4 shown in (b).
1.4 edge detection
Edge detection, is the obvious point of brightness change in reference numerals image, extracts the edge of word, this is also image
Basic problem in processing and computer vision, therefore edge detection also plays critically important effect in Text region.
The design carries out edge detection using Canny operators, and Image Edge-Detection is eliminated with identifying incoherent information,
Recognition time is saved by significantly reducing data volume, and remains the important structure attribute of image, improves system
The efficiency by row segmentation is carried out when importing word picture.
2 Text segmentations
2.1 read gray-scale maps and by its binaryzation
Picture mainly stores with a matrix type in a computer.Two gray-scale maps be single channel matrix he preserve the color of picture
Pixel value is changed into 0 and 255 with binarization method in 0-255., represents the color value of background and font respectively, be by information, numerical value
Segmentation is prepared below.
2.2 Text segmentation
Using sciagraphy, split according to the feature of pixel value.After binary conversion treatment, floor projection pixel is first carried out
Value is added, and is split into every trade.Segmentation effect is as shown in Figure 5.Then vertical direction is projected and carries out pixel value addition, it is vertical to carry out
Direction is split, and segmentation effect is as shown in Figure 6.Last entirety segmentation effect is as shown in Figure 7.
3 Text regions
After above-mentioned figure pretreatment, Text segmentation operation is completed, next need that the word of segmentation is identified.
The acquisition of 3.1 training datas
In Intelligent road sign identifying system, we select daily common 3000 Chinese characters and are trained, based on national standard
Regulation --- road guideboard generates respective image data in experiment using upright black matrix form Chinese character by computer.Utilize
C# language, generates the Chinese character picture of 40*40 pixels as test set and training set, as shown in Figure 8.
Various interference occur in actual acquisition picture in view of front end acquisition module, such as hardware fever, external environment
Interference, bright dark etc. the factor of light, can make it that image carry noise spot, picture blur, writing it is unclear or produce rotate and
The situations such as distortion.Therefore, each Chinese character image of generation need to be subjected to image procossing, carries out various random noises to it respectively
Point generation, corrosion expansion and the rotation and distortion of different angle, so produce 30 different figures to each Chinese character picture
Piece, obtains more data volumes, and such test set and training set one share 90000 data.Pass through the convolution god shown in Fig. 7
It is trained through network.
The selection of 3.2 training patterns
Convolutional neural networks（CNN）It is an important algorithm in deep learning field, the effect of brilliance is above shown in many applications
Fruit.Character recognition algorithm has a many kinds at present, but to the identification of Chinese character, and the Text region of particularly natural scene has certain
Limitation.By a variety of document character recognition algorithms compared with CNN, it is found that convolutional neural networks algorithm than other algorithms effect all
It is good.In the Intelligent road sign identifying system of this paper, CNN is mainly utilized（Fig. 9）To 3000 Chinese characters common in daily life into
Row training and then identification.CNN is to improve to obtain in the structure of BP neural network, they are to employ to propagate forward to calculate net
The output valve of network, by error calculation formula, back-propagation corrects weight and the value of biasing.CNN and traditional feature extraction side
The improvement of method maximum is exactly that convolutional neural networks are to carry out feature extraction using convolution kernel, is not to connect entirely between adjacent layer, and
Simply partly it is attached, so as to obtain local feature.Weights shared mechanism is used in a characteristic plane, very big reduces
The quantity of weights.Another aspect CNN models have more preferable generalization ability, even if when image deforms or there are during noise
Recognition result will not be caused significantly to influence；Another further aspect its net reduced by the methods of the wild and shared weights of local sensing
The complexity of network model and parameter, and than the accuracy higher of conventional model.
Training network model uses for reference the network of MNIST Handwritten Digit Recognitions, but Chinese character have it is increasingly complex and fine
Structure, therefore the adjustment of the network parameters such as parameter setting, network model configuration is carried out on this basis, make it more suitable for Chinese character
Training and identification.Training network structure is as shown in Figure 10.
This identifying system uses ReLU activation primitives, it belongs to unsaturation activation primitive, is missed in convolutional neural networks
During poor backpropagation, derivation is carried out, there are problems that gradient disappearance during derivation, i.e., every layer will be with activation primitive
First derivativeIt is multiplied, when the network number of plies increases, gradient G will constantly decay until disappearing.Activation primitive
For,
Compared with traditional sigmoid, tanh function, expression effect is most preferable in this experiment for ReLu functions.
Pooling layers in network structure carry out pond using the pattern of max, choose and the region is chosen in the region of 2*2
In maximum as result.Record position of the maximum in each zonule, during backpropagation, residual error is delivered to this
Maximum value position, other positions zero setting.
When network is too deep, network may over-fitting.So after convolutional layer iteration, over-fitting in order to prevent, rear several
The pooling layers of layer and ReLU layer add dropout layers, dropout layers at random allow node weights zero setting so that export result
Also be 0, thus avoid that network some features only just produce pair under fixed combination as a result, so as to allow network to go to learn
Universal general character, rather than some characteristics of some training samples.
Last output layer selects SoftmaxWithLoss graders, as shown in figure 11.SoftmaxWithLoss is actual
It is Multinomial Logistic Loss Layer（Cross entropy cost function）With the combination of Softmax Layer.Assuming that sample
This quantity has m, and each sample characteristics quantity is b, calculates this probability of m sample in n class, and calculation formula is：
By the real vector of k dimensionsIt is mapped as, thenAccording to size carry out classify more
Task（Weighting weighs the one-dimensional of maximum）.
After Text segmentation, Chinese character image to be identified is sent into trained model, by convolutional neural networks into
Row propagated forward, propagated forward basic principle are as follows：
It is calculated as from input unit to first hidden layer H1：, wherein all inputs of k values traversal
Node layer,It is the weighted sum to all nodes of preceding layer,For nonlinear function, rear layer and so on.
Input information is handled from input layer through hidden layer, and is transmitted to output layer, one under the influence of the state of each layer of neuron
The state of layer neuron, is acted on using nonlinear function, calculates the error of network output and desired output.Arrived for input layer
The weights of hidden layerThen still using BP algorithm renewal weight.Input information is handled from input layer through hidden layer, and processing mode is：If
Initial weight and thresholding are put, they are all set to less random number.
In the propagated forward stage, data source arises from digital independent layer, by some process layers, reaches last layer（Loss
Layer either characteristic layer）, in this stage, the weights in network do not change, and network path is a directed acyclic graph
（DAG）, the node since most, by some process layers, there is no loop structure, therefore data flow can go ahead and push away
Into until terminal.
Propagated forward process is studied with dataflow analysis method, i.e.,：Concentrated from input data and take a sample（X,
Y）, wherein X is data, and Y is label.X is sent into network, is successively calculated, obtains corresponding network processes output O, network performs
Calculating be formulated as：
Wherein,Represent nonlinear transformation,Represent each weights layer weights.It is defeated for network
Go out, Ke Yiyong（Y,O）Network quality is assessed, preferable network meets Y==O.
In the training process of character data collection, propagated forward is first passed through by information is inputted and is transferred to output layer, to network
Output and error are modified, last layer relatively obtains loss function with object function, and calculation error updated value, adjusts hidden layer
Connection weight to output layer is：,For learning rate.
Input layer is adjusted to the connection weight of hidden layer, wherein
Since image feature information is more, data are numerous and diverse, therefore training process needs the training that iterates, until loss restrain,
During training sample, uniformly input is kept, can finally realize comparatively ideal discrimination.
In text categorization task, 3000 Chinese characters in common use are classified as, into after training up excessively, network structure model
When iterations reaches 1000 times, accuracy rate can reach more than 99% accuracy（As shown in figure 12）, illustrate this network mould
Type structure meets the design needs.
Claims (5)
- A kind of 1. Intelligent road sign recognition methods, it is characterised in that：Guideboard image is gathered by image capture module, is communicated mould by 3G Block is by the image transmitting of collection to Cloud Server, and Cloud Server pre-processes the image of collection, character area extracts To the word for going background and then Text segmentation processing and Text region are carried out, and recognition result is communicated with text formatting by 3G Module is sent to central control module, and the text message of recognition result is changed into voice letter finally by voice broadcast module Breath, user is informed by earphone or loudspeaker.
- A kind of 2. Intelligent road sign recognition methods according to claim 1, it is characterised in that：The Cloud Server is to pass through figure As analysis process system, the pretreatment to natural scene character image and the area of image are completed with traditional image processing method Regional partition, using deep learning neural network algorithm, identifies the text information in cut zone.
- A kind of 3. Intelligent road sign recognition methods according to claim 1, it is characterised in that：The Text region, is to adopt The model trained by the use of deep learning neutral net as word is completed, and Chinese character image training data is produced using computer programming Collection, uses the data set network model that training designs on the server of generation.
- 4. the Intelligent road sign identification used in a kind of Intelligent road sign recognition methods described according to claim 1-3 any claims System, it is characterised in that：Including central control module, central control module is connected to voice broadcast module and 3G communication moulds Block, 3G communication modules are connected with Cloud Server, and central control module is also associated with image capture module.
- 5. Intelligent road sign identifying system according to claim 1, it is characterised in that：The voice broadcast module includes voice Synthesis unit and the voice broadcast unit being connected with phonetic synthesis unit.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810019360.4A CN108009548A (en) | 2018-01-09 | 2018-01-09 | A kind of Intelligent road sign recognition methods and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810019360.4A CN108009548A (en) | 2018-01-09 | 2018-01-09 | A kind of Intelligent road sign recognition methods and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108009548A true CN108009548A (en) | 2018-05-08 |
Family
ID=62050456
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810019360.4A Pending CN108009548A (en) | 2018-01-09 | 2018-01-09 | A kind of Intelligent road sign recognition methods and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108009548A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108960200A (en) * | 2018-07-31 | 2018-12-07 | 北京微播视界科技有限公司 | A kind of data processing method and electronic equipment based on intelligent interaction |
| CN109284678A (en) * | 2018-08-17 | 2019-01-29 | 宽凳（北京）科技有限公司 | Guideboard method for recognizing semantics and system |
| CN110322495A (en) * | 2019-06-27 | 2019-10-11 | 电子科技大学 | A kind of scene text dividing method based on Weakly supervised deep learning |
| CN110689039A (en) * | 2019-08-19 | 2020-01-14 | 浙江工业大学 | Trunk texture identification method based on four-channel convolutional neural network |
| CN112101128A (en) * | 2020-08-21 | 2020-12-18 | 东南大学 | Unmanned formula racing car perception planning method based on multi-sensor information fusion |
| CN112232354A (en) * | 2020-11-23 | 2021-01-15 | 中国第一汽车股份有限公司 | Character recognition method, device, equipment and storage medium |
| CN112700458A (en) * | 2020-12-31 | 2021-04-23 | 南京太司德智能电气有限公司 | Electric power SCADA warning interface character segmentation and processing method |
| CN112766263A (en) * | 2021-01-21 | 2021-05-07 | 西安理工大学 | Identification method for multi-layer stock control relation share graph |
| CN116309494A (en) * | 2023-03-23 | 2023-06-23 | 宁波斯年智驾科技有限公司 | Method, device, equipment and medium for determining interest point information in electronic map |
| CN117668326A (en) * | 2024-01-30 | 2024-03-08 | 深圳柯赛标识智能科技有限公司 | Intelligent identification data processing method, system and equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101584624A (en) * | 2009-06-18 | 2009-11-25 | 上海交通大学 | Guideboard recognition blind-guide device and method thereof based on DSP |
| CN106096602A (en) * | 2016-06-21 | 2016-11-09 | 苏州大学 | A kind of Chinese licence plate recognition method based on convolutional neural networks |
-
2018
- 2018-01-09 CN CN201810019360.4A patent/CN108009548A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101584624A (en) * | 2009-06-18 | 2009-11-25 | 上海交通大学 | Guideboard recognition blind-guide device and method thereof based on DSP |
| CN106096602A (en) * | 2016-06-21 | 2016-11-09 | 苏州大学 | A kind of Chinese licence plate recognition method based on convolutional neural networks |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108960200A (en) * | 2018-07-31 | 2018-12-07 | 北京微播视界科技有限公司 | A kind of data processing method and electronic equipment based on intelligent interaction |
| CN109284678A (en) * | 2018-08-17 | 2019-01-29 | 宽凳（北京）科技有限公司 | Guideboard method for recognizing semantics and system |
| CN110322495A (en) * | 2019-06-27 | 2019-10-11 | 电子科技大学 | A kind of scene text dividing method based on Weakly supervised deep learning |
| CN110689039A (en) * | 2019-08-19 | 2020-01-14 | 浙江工业大学 | Trunk texture identification method based on four-channel convolutional neural network |
| CN112101128A (en) * | 2020-08-21 | 2020-12-18 | 东南大学 | Unmanned formula racing car perception planning method based on multi-sensor information fusion |
| CN112232354A (en) * | 2020-11-23 | 2021-01-15 | 中国第一汽车股份有限公司 | Character recognition method, device, equipment and storage medium |
| CN112700458A (en) * | 2020-12-31 | 2021-04-23 | 南京太司德智能电气有限公司 | Electric power SCADA warning interface character segmentation and processing method |
| CN112766263A (en) * | 2021-01-21 | 2021-05-07 | 西安理工大学 | Identification method for multi-layer stock control relation share graph |
| CN112766263B (en) * | 2021-01-21 | 2024-02-02 | 西安理工大学 | Identification method for multi-layer control stock relationship share graphs |
| CN116309494A (en) * | 2023-03-23 | 2023-06-23 | 宁波斯年智驾科技有限公司 | Method, device, equipment and medium for determining interest point information in electronic map |
| CN116309494B (en) * | 2023-03-23 | 2024-01-23 | 宁波斯年智驾科技有限公司 | Method, device, equipment and medium for determining interest point information in electronic map |
| CN117668326A (en) * | 2024-01-30 | 2024-03-08 | 深圳柯赛标识智能科技有限公司 | Intelligent identification data processing method, system and equipment |
| CN117668326B (en) * | 2024-01-30 | 2024-04-30 | 深圳柯赛标识智能科技有限公司 | Intelligent identification data processing method, system and equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108009548A (en) | A kind of Intelligent road sign recognition methods and system | |
| Kamal et al. | Automatic traffic sign detection and recognition using SegU-Net and a modified Tversky loss function with L1-constraint | |
| CN110163187B (en) | F-RCNN-based remote traffic sign detection and identification method | |
| CN110033002B (en) | License plate detection method based on multitask cascade convolution neural network | |
| CN111368886B (en) | Sample screening-based label-free vehicle picture classification method | |
| CN107563372B (en) | License plate positioning method based on deep learning SSD frame | |
| CN111414906B (en) | Data synthesis and text recognition method for paper bill pictures | |
| CN102693432B (en) | Use reliable partial model more to newly arrive and regulate clear path to detect | |
| CN105354568A (en) | Convolutional neural network based vehicle logo identification method | |
| CN106682569A (en) | Fast traffic signboard recognition method based on convolution neural network | |
| CN112395951B (en) | Complex scene-oriented domain-adaptive traffic target detection and identification method | |
| CN102682301B (en) | Adaptation for clear path detection with additional classifiers | |
| CN106096602A (en) | A kind of Chinese licence plate recognition method based on convolutional neural networks | |
| US11481576B2 (en) | Subject-object interaction recognition model | |
| CN110852358A (en) | Vehicle type distinguishing method based on deep learning | |
| CN113743417B (en) | Semantic segmentation method and semantic segmentation device | |
| CN112069900A (en) | Bill character recognition method and system based on convolutional neural network | |
| CN106599848A (en) | Depth visual feature and support vector machine-based terrain texture recognition algorithm | |
| CN113052215A (en) | Sonar image automatic target identification method based on neural network visualization | |
| Imaduddin et al. | Indonesian vehicle license plate number detection using deep convolutional neural network | |
| CN111274863B (en) | Text prediction method based on text mountain probability density | |
| KR102026280B1 (en) | Method and system for scene text detection using deep learning | |
| CN116189130A (en) | Lane line segmentation method and device based on image annotation model | |
| Al Awaimri et al. | Automatic number plate recognition system for Oman | |
| Tumuluru et al. | SMS: SIGNS MAY SAVE–Traffic Sign Recognition and Detection using CNN |
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: 20180508 |
|
| RJ01 | Rejection of invention patent application after publication |