CN109993058A - The recognition methods of road signs based on multi-tag classification - Google Patents
The recognition methods of road signs based on multi-tag classification Download PDFInfo
- Publication number
- CN109993058A CN109993058A CN201910144912.9A CN201910144912A CN109993058A CN 109993058 A CN109993058 A CN 109993058A CN 201910144912 A CN201910144912 A CN 201910144912A CN 109993058 A CN109993058 A CN 109993058A
- Authority
- CN
- China
- Prior art keywords
- road signs
- tag
- classification
- label
- 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
- 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
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/582—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of traffic signs
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of recognition methods of road signs based on multi-tag classification, extract all labels of each road signs；It is predicted to obtain the multi-tag template of road signs, for judging whether road signs image to be identified belongs to the road signs；The prediction uses convolutional neural networks as learner, classified using multi-tag classifier, by calculating the matching degree between the multi-tag template of road signs image to be identified and the multi-tag template of standard road traffic sign, differentiate whether road signs image to be identified belongs to the road signs.The present invention is able to solve the identification problem to road signs classification, has interpretation, improves the accuracy rate of convolutional neural networks model identification road signs, and recognition methods has high robust.
Description
Technical field
The invention belongs to technical field of computer vision, are related to traffic sign recognition method more particularly to a kind of base
In the recognition methods of the road signs of multi-tag classification.
Background technique
In the fields such as autonomous driving vehicle and high-precision map, traffic sign recognition is wherein essential ring
Section, also therefore, it has become a research hotspots of field of image recognition for traffic sign recognition.Traffic sign recognition
The resolution ratio that difficult point essentially consists in road signs image is relatively low, the visual angle of road signs Image Acquisition and scene
Illumination, weather condition change greatly, while similitude is very big between road signs, and system is easy to misidentify image, influence
The application of reality scene.The method of traffic sign recognition is broadly divided into two major classes, and one kind is calculated using traditional-handwork feature
Son reuses to extract feature and classifies classifiers to identify classification belonging to each road signs image more.It is another kind of to be
By deep learning, directly original image is identified using every class road signs as individual one kind.The first kind is adopted
Adaptive capacity to environment with the method for traditional-handwork feature operator, algorithm is poor, and in the second class method, deep learning model is being instructed
With certain robustness when white silk collection sample is enough.However, these existing methods all not can guarantee the character symbol learnt
The semantic knowledge that the mankind carry out traffic sign recognition is closed, so that these traffic sign recognition system mistakes is general
Rate increases, and has Unpredictability.
Summary of the invention
In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides a kind of road traffic marks based on multi-tag classification
The recognition methods of will passes through the deep learning model based on convolutional neural networks model using the method classified based on multi-tag
(hereinafter referred to as convolutional neural networks model) learns the mankind and carries out knowledge information used when traffic sign recognition, makes
Obtaining convolutional neural networks model has interpretation to the identification of road signs classification, i.e., it is special only to meet template multi-tag
The image of sign just belongs to the road signs classification, thus filters out the sample of certain tag misses, to improve convolution mind
Accuracy rate through network model identification road signs, and recognition methods has higher robustness.
In the present invention, using element included in road signs image as a label or tag class；Pass through mark
The probability of multiple labels of input picture is predicted in label classification using classifier.The method of the present invention is different from existing based on multiclass
The convolutional neural networks model of other classification task is not to use a road signs as a classification, but use
Foundation of the composed structure knowledge of road signs as convolutional neural networks Model checking images to be recognized generic, is adopted
Make the convolutional neural networks model learning mankind used when carrying out traffic sign recognition with based on the method that multi-tag is classified
The knowledge information arrived, so that the identification of road signs has better robustness.
Present invention provide the technical scheme that
A kind of recognition methods of the road signs based on multi-tag classification, passes through the structure to each road signs
At being analyzed, all labels of the road signs, then the institute according to the road signs extracted are manually extracted
There is label to obtain the template of road signs, the template be used for judge final image whether belong to the road signs according to
According to.Wherein in order to predict the road signs template for acquiring image, use convolutional neural networks as learner, using more
Label classifier is classified.By calculating the multi-tag template of acquisition image and the multi-tag template of standard road traffic sign
Between matching degree, can differentiate whether road signs image to be identified belongs to the road signs.Specifically
Steps are as follows:
1) composition and whole labels of standard road traffic sign, the whole labels that will be extracted are obtained by manually extracting
Template as road signs.
Wherein for road traffic prohibitory sign (ginseng Fig. 1), according to its Shape Classification are as follows: whether be round, apex angle downward
Equilateral triangle and octagonal, whether be divided into color is white background red circle, blue bottom red circle, white background black circle and red bottom red circle, figure
Classify in case are as follows: whether have red vertical bar, red slash, forbid slash and black thin slash, classify on direction arrow are as follows: whether containing straight
Row arrow, to the left arrow, right-hand arrow, turn around arrow and arrow of overtaking other vehicles, and classifies on composition are as follows: whether contains motorcycle, non-machine
Motor-car, manpower passenger tricycle, manpower shipping tricycle, rickshaw, animal-drawn vehicle, electro-tricycle, pedestrian, dangerous goods vehicle,
Motor vehicle, minibus, car, cargo vehicle, trailer, three-wheeled motor car, loudspeaker and tractor.For example, for forbidding to from left to right
For turning mark, the template that can extract the road signs is circle, white background red circle, to the left arrow and arrow to the right
Head.
When it is implemented, being by each road signs t template formal definitions generated (quantity of 1≤i≤m, m=road signs).
2) according to the corresponding template of each road signs, multi-tag categorized data set, i.e. multi-tag classification shape are constructed
Formula template.
It whether there is the available instruction in the template of the road signs according to each label of road signs
The multi-tag mask practiced and used when predicting, the multi-tag mask are made of the binary string of fixed length, wherein 0 represents road friendship
Be free of the label in the template of logical mark, 1 represents and contains the label in the template of the road signs.The multi-tag mask is straight
Foundation when connecing as convolutional neural networks training and prediction, is the formalization representation of the template constructed in the first step.All roads
Traffic sign corresponding multi-tag mask in road constitutes multi-tag categorized data set；
3) using the multi-tag data set training convolutional neural networks of above-mentioned construction.
The back bone network of convolutional neural networks can use various networks, will wherein the last one full articulamentum make
For multi-tag classification layer, wherein the number of neuron is the number of label in multi-tag data set, for every in full articulamentum
For a neuron, the value of neuron output is the score that the corresponding label of the neuron whether there is.The score passes through
After the activation of sigmoid activation primitive, the probability that whether there is the label in the acquisition image is obtained, that is, is passed throughBy probability of the score normalization of convolutional neural networks prediction between 0-1.Carrying out neural network
When training, need to calculate the loss of training set, model uses binary cross entropy loss function, is expressed as follows the public affairs in face
Formula:
L (x, y)=L={ l1,…,ln}ln=-(yn·logxn+(1-yn)·log(1-xn))
Wherein, x, y are respectively tensor composed by tag template that convolutional neural networks predict input picture and its right
Tensor composed by the true multi-tag mask answered；L (x, y) is binary cross entropy loss function；l1,…,lnFor each input
The corresponding loss function of image pattern；xnThe probability institute of each label of n-th of the sample come is predicted for convolutional neural networks
The vector of composition；ynFor multi-tag mask vector corresponding to true tag；Log is the nature truth of a matter, i.e. ln in formula.
4) carry out the classification of road signs belonging to Prediction and Acquisition image using above-mentioned trained convolutional neural networks.
Predict in images to be recognized that there are the probability of each label (to pass through with the method in step 3) first
After the activation of sigmoid activation primitive, the probability for the label that whether there is in the acquisition image is obtained), use following formula
Calculate the Probability p that acquisition image i matches j-th of road signsij:
Wherein, k is the number of label, xinIt include the probability of label n, y for convolutional neural networks forecast image ijnFor jth
A road signs include the mask value of label n.
From all pijThe road signs r of maximum probability is chosen in (1≤j≤m) as Mode Road traffic sign
Identification as a result, wherein m be road signs number.
Compared with prior art, the beneficial effects of the present invention are:
Existing identification technology only judges that label classification, classification are determined by the image in training set with simple single sorting technique
Justice, and lack semantic information.The present invention uses the multi-tag sorting technique with semantic knowledge to judge based on convolutional neural networks
Label classification, providing the traffic sign recognition method based on multi-tag classification will be known by manual construction label data collection
Other semantic information fusion enters in convolutional neural networks model, and convolutional neural networks can be made to extract institute when meeting mankind inference
The feature used, enhances the interpretation and robustness of model, while also increasing the prison of convolutional neural networks training and prediction
Superintend and direct information, i.e., only the probability distribution of each label that predicts of convolutional neural networks and meanwhile meet road signs when
It waits, model can just be predicted as the road signs, thus can filter out and certain not meet the defeated of road signs composition
Enter image, the accuracy of lift scheme.
Detailed description of the invention
Fig. 1 is existing road traffic sign prohibitory sign.
Fig. 2 is the flow diagram of the method for the present invention.
Fig. 3 is the flow diagram that the method for the present invention uses convolutional neural networks to be trained and predict.
Specific embodiment
With reference to the accompanying drawing, the present invention, the model of but do not limit the invention in any way are further described by embodiment
It encloses.
As shown in Fig. 2, specific implementation step of the invention are as follows:
1) the multi-tag template of road signs is constructed.
Fig. 1 shows existing road traffic sign prohibitory signs.Wherein, for road traffic prohibitory sign, according to its shape
Whether it is round, apex angle equilateral triangle directed downwardly and octagonal that shape is divided into, and whether be divided into color is that white background red circle, blue bottom are red
Whether circle, white background black circle and red bottom red circle, being divided on pattern has red vertical bar, red slash, forbids slash and black thin slash, direction arrow
Head on be divided into whether containing straight trip arrow, to the left arrow, right-hand arrow, turn around arrow and arrow of overtaking other vehicles, be divided on composition whether
Contain motorcycle, non-motor vehicle, manpower passenger tricycle, manpower shipping tricycle, rickshaw, animal-drawn vehicle, electro-tricycle, row
People, dangerous goods vehicle, motor vehicle, minibus, car, cargo vehicle, trailer, three-wheeled motor car, loudspeaker and tractor.For example, right
For mark of forbidding bending to right to the left, the template that can extract the road signs is circle, white background red circle, to the left
Arrow and right-hand arrow.Table 1 illustrates the multi-tag template building result of all road signs prohibitory signs.
1 multi-tag template of table constructs result
| Road signs | Mark name | One label two of label | Label three | Label four | Label five |
| p1 | Overtaking prohibited sign | Circle white background red circle | Red vertical bar | It overtakes other vehicles arrow | Straight trip arrow |
| p2 | Animal-drawn vehicle is forbidden to enter mark | Circle white background red circle | Forbid slash | Animal-drawn vehicle | |
| p3 | Motorbus is forbidden to drive into mark | Circle white background red circle | Forbid slash | Car | |
| p4 | Electro-tricycle is forbidden to drive into mark | Circle white background red circle | Forbid slash | Electro-tricycle | |
| p5 | No turns indicates | Circle white background red circle | Forbid slash | Turn around arrow | |
| p6 | Non-motor vehicle is forbidden to enter mark | Circle white background red circle | Forbid slash | Non-motor vehicle | |
| p7 | Cargo vehicle is forbidden to turn left | Circle white background red circle | Forbid slash | Arrow to the left | Cargo vehicle |
| p8 | Trailer, semitrailer is forbidden to drive into mark | Circle white background red circle | Forbid slash | Trailer | |
| p9 | Pedestrian is forbidden to enter mark | Circle white background red circle | Forbid slash | Pedestrian | |
| p10 | Motor vehicle is forbidden to drive into mark | Circle white background red circle | Forbid slash | Motor vehicle | |
| p11 | No horn indicates | Circle white background red circle | Forbid slash | Loudspeaker | |
| p12 | Motorcycle is forbidden to drive into mark | Circle white background red circle | Forbid slash | Motorcycle | Pedestrian |
| p13 | Certain two kinds of vehicle is forbidden to drive into mark | Circle white background red circle | Forbid slash | Cargo vehicle | Three-wheeled motor car |
| p14 | Straight trip is forbidden to indicate | Circle white background red circle | Forbid slash | Straight trip arrow | |
| p15 | Rickshaw is forbidden to enter mark | Circle white background red circle | Forbid slash | Rickshaw | |
| p16 | Manpower shipping tricycle is forbidden to enter mark | Circle white background red circle | Forbid slash | Manpower shipping tricycle | Pedestrian |
| p17 | Manpower passenger tricycle is forbidden to enter mark | Circle white background red circle | Forbid slash | Manpower passenger tricycle | Pedestrian |
| p18 | Tractor is forbidden to enter mark | Circle white background red circle | Forbid slash | Tractor | |
| p19 | No right turn mark | Circle white background red circle | Forbid slash | Right-hand arrow | |
| p20 | Forbid the mark that bends to right to the left | Circle white background red circle | Forbid slash | Arrow to the left | Right-hand arrow |
| p21 | Forbid mark of keeping straight on and bend to right | Circle white background red circle | Forbid slash | Straight trip arrow | Right-hand arrow |
| p22 | Three-wheeled motor car, low-speed truck is forbidden to drive into mark | Circle white background red circle | Forbid slash | Three-wheeled motor car | |
| p23 | No left turn indicates | Circle white background red circle | Forbid slash | Arrow to the left | |
| p24 | Forbid minibus to turn right to indicate | Circle white background red circle | Forbid slash | Right-hand arrow | Minibus |
| p25 | Station wagon is forbidden to drive into mark | Circle white background red circle | Forbid slash | Minibus | |
| p26 | Cargo vehicle is forbidden to drive into mark | Circle white background red circle | Forbid slash | Cargo vehicle | |
| p27 | Transport of dangerous goods vehicle is forbidden to drive into mark | Circle white background red circle | Forbid slash | Motor vehicle | Dangerous goods vehicle |
| p28 | Forbid mark of keeping straight on and bend to right | Circle white background red circle | Forbid slash | Arrow to the left | Straight trip arrow |
| pd | Customs's mark | Circle white background red circle | Black horizontal line | Customs | |
| pc | Stop sign | Circle white background red circle | Black horizontal line | It checks | |
| pn | No parking indicates | Circle indigo plant bottom red circle | Forbid slash | Red slash | |
| pnl | Stop sign when forbidding long | Circle indigo plant bottom red circle | Forbid slash | ||
| ps | Stop sign | The red bottom red circle of octagon | |||
| pg | Give way mark | Triangle white background red circle | |||
| pb | Traffic prohibited sign | Circle white background red circle | |||
| pe | Traffic has priority over oncoming vehicle indicates | Circle white background red circle | Straight trip arrow | Turn around arrow | |
| pne | No entry sign | The red bottom red circle of circle | White horizontal line | ||
| pm* | Limit quality mark | Circle white background red circle | * | t | |
| pa* | Axle load limited mark | Circle white background red circle | * | t | Axis indicates again |
| pl* | Limit speed marker | Circle white background red circle | * | ||
| pr* | Lift restrictions speed marker | Circle white background black circle | * | Black thin slash | |
| ph* | Maximum height limit mark | Circle white background red circle | * | m | Bench margin |
| pw* | Max. Clearance _M. mark | Circle white background red circle | * | m | Width indicator |
2) according to the road signs multi-tag template of building, the formalization template of each road signs is generated,
The rule of the formalization template generation is that all labels being likely to occur sort by lexicographic ordering, and then generating a length is mark
01 string of quantity is signed, wherein the corresponding position of label occurred in all road signs is set as 1, the mark not occurred
It signs corresponding position and is set as 0, be by each road signs t formalization template definition generated(quantity of 1≤i≤m, m=road signs).
3) training convolutional neural networks (ginseng Fig. 3).
There are many kinds of the back bone network structures of current convolutional neural networks, including VGG, ResNet, Inception,
DenseNet, MobileNet and ShuffleNet etc..Back bone network of the invention can be with any one of the above or their change
Kind, but scope of the invention is still fallen within using other back bone networks, the present embodiment has selected VGG19 as the skeleton of model
Network.For the sake of simplicity it is assumed that the amount of images of input convolutional neural networks is only 1, actual conditions may be 16,32 etc..It should
The input picture size of the input layer of network is 32 × 32, and port number is 3.Wherein input picture is to have already passed through road traffic mark
The image that will detector has detected cuts road signs image according to detection block from the image of acquisition.Figure
It may include all road signs that model needs to identify as in, made in the present embodiment using all road prohibitory signs
To need the road signs identified.Output layer is full articulamentum, and the number of output is different labels in multi-tag data set
Number, be denoted as n here.Therefore, output can be defined as outputi(1≤i≤n) is made using softmax function normalization
The probability occurred for each label is denoted asAssuming that the real classification of the input picture is y, it is defeated to calculate this
After entering the probability that each label of image occurs, calculated using binary cross entropy loss function Wherein masky,iTo be generated in step 2
Road signs template.Gradient (gradient calculating of each parameter of convolutional neural networks relative to loss function is calculated later
Process is provided by neural network framework, such as pytorch, tensorflow even depth learning framework), decline optimization using gradient and calculates
Method updates the weights of convolutional neural networks, and wherein the selection of gradient descent algorithm the present embodiment is SGD (stochastic gradient descent)
Algorithm, other optional gradient descent algorithms include Adam scheduling algorithm.Wherein in order to enable convolutional neural networks training result more
Add robust, the overturning of image Random Level and random cropping by input, then be input in convolutional neural networks and be trained.
4) using the classification of trained convolutional neural networks prediction input picture (ginseng Fig. 2), i.e., which input picture belongs to
Class road signs.Wherein input picture needs to first pass through the detection of road signs detector, is carried out using testing result
It cuts, send the image of cutting as input picture into convolutional neural networks prediction, i.e., only include road traffic in the image
The background of sign image and its place scene.Trained convolutional neural networks parameter is loaded into the memory of computer, it
Input picture is compressed to 32 × 32 afterwards.The image for being compressed to fixed size is input to convolutional neural networks, wherein convolution mind
Parameter through network is that step 3) training obtains, and the full articulamentum of the output layer of final convolutional neural networks exports point of each label
Number obtains the probability that each label occurs, with the x in step 3) similarly after softmax is normalizedi.Use convolution mind
The each label probability come out through neural network forecast, calculating input image and each matched Probability p of road signs tt, wherein
madkt,iFor the template mask of corresponding i-th of the label of obtained t-th of the road signs of step 2, calculation formula is
Wherein, for the stability that numerical value calculates, log is taken simultaneously on formula both sides, is obtained:
Select so that the maximum road signs of matching probability as model prediction as a result, i.e.For model prediction result.
It should be noted that the purpose for publicizing and implementing example is to help to further understand the present invention, but the skill of this field
Art personnel, which are understood that, not to be departed from the present invention and spirit and scope of the appended claims, and various substitutions and modifications are all
It is possible.Therefore, the present invention should not be limited to embodiment disclosure of that, and the scope of protection of present invention is with claim
Subject to the range that book defines.
Claims (6)
1. a kind of recognition methods of the road signs based on multi-tag classification, extracts all of each road signs
Label；It is predicted to obtain the multi-tag template of road signs, for judging that road signs image to be identified is
It is no to belong to the road signs；The prediction uses convolutional neural networks as learner, is carried out using multi-tag classifier
Classification, by calculating the multi-tag template of road signs image to be identified and the multi-tag mould of standard road traffic sign
Matching degree between plate, differentiates whether road signs image to be identified belongs to the road signs；Including as follows
Step:
1) composition for obtaining standard road traffic sign and whole labels are extracted, using the whole labels extracted as road traffic
The template of mark；
2) according to the template of each road signs, multi-tag categorized data set, i.e. multi-tag classification formalization template are constructed；
It performs the following operations:
21) according to each label of road signs whether there is in the template of the road signs, obtain training and
The multi-tag mask used when prediction, the multi-tag mask are made of the binary string of fixed length, wherein 0 represents the road traffic mark
The label is free of in the template of will, 1 represents in the template of the road signs containing the label；Multi-tag mask is used for convolution
Neural metwork training and prediction；
The template definition that each road signs t is generated is multi-tag mask:
Wherein, m is the quantity of road signs；
22) the corresponding multi-tag mask of all road signs is constituted into multi-tag classification model；
3) the multi-tag classification model training convolutional neural networks constructed using step 2), obtain trained convolutional Neural net
Network；Specifically perform the following operations:
31) using the last one full articulamentum of convolutional neural networks as multi-tag classification layer, wherein the number of neuron is more
The number of label in labeling template；
32) neural metwork training is carried out, the loss of training set is calculated using loss function；Loss function indicates are as follows:
L (x, y)=L={ l1,…,ln}ln=-(yn·logxn+(1-yn)·log(1-xn))
Wherein, x, y are respectively tensor composed by tag template that convolutional neural networks predict input picture and its corresponding
Tensor composed by true multi-tag mask；L (x, y) is binary cross entropy loss function；l1,…,lnFor each input picture
The corresponding loss function of sample；xnIt is made of the probability that convolutional neural networks predict each label of n-th of the sample come
Vector；ynFor multi-tag mask vector corresponding to true tag；
33) value of each neuron output in full articulamentum is the score that the corresponding label of the neuron whether there is；
34) it after the score is activated by activation primitive, obtains in road signs image to be identified with the presence or absence of the label
Probability；
4) road signs belonging to road signs image to be identified are predicted using trained convolutional neural networks
Classification；It performs the following operations:
41) prediction obtains in road signs image to be identified that there are the probability of each label；
42) Probability p that image i matches j-th of road signs is calculated using following formulaij:
Wherein, k is the number of label, xinIt include the probability of label n, y for convolutional neural networks forecast image ijnFor j-th of road
Traffic sign includes the mask value of label n；
43) from all pijResult of the middle road signs r for choosing maximum probability as Mode Road Traffic Sign Recognition；
Through the above steps, the identification for the road signs classified based on multi-tag is realized.
2. the recognition methods of the road signs as described in claim 1 based on multi-tag classification, characterized in that step 1) mentions
The composition and whole labels for obtaining standard road traffic sign, using the whole labels extracted as the mould of road signs
Plate；Specific to road traffic prohibitory sign, Shape Classification is included: whether as round, apex angle equilateral triangle directed downwardly and eight
It is angular；Color classification includes: whether as white background red circle, blue bottom red circle, white background black circle and red bottom red circle；Pattern classification includes: to be
It is no to have red vertical bar, red slash, forbid slash and black thin slash；Direction arrow classification is included: whether containing straight trip arrow, to the left arrow
Head, right-hand arrow, turn around arrow and arrow of overtaking other vehicles；Group constituent class is included: whether containing motorcycle, non-motor vehicle, manpower passenger traffic
Tricycle, manpower shipping tricycle, rickshaw, animal-drawn vehicle, electro-tricycle, pedestrian, dangerous goods vehicle, motor vehicle, minibus,
Car, cargo vehicle, trailer, three-wheeled motor car, loudspeaker and tractor.
3. the recognition methods of the road signs as described in claim 1 based on multi-tag classification, characterized in that the convolution
Neural network includes but is not limited to VGG, ResNet, Inception, DenseNet, MobileNet or ShuffleNet.
4. the recognition methods of the road signs as described in claim 1 based on multi-tag classification, characterized in that the activation
Function is sigmoid activation primitive.
5. the recognition methods of the road signs as described in claim 1 based on multi-tag classification, characterized in that step 34)
Especially by activation primitiveThe score normalization that convolutional neural networks are predicted is general between 0-1
Rate.
6. the recognition methods of the road signs as described in claim 1 based on multi-tag classification, characterized in that the convolution
Neural network is preferably VGG19.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910144912.9A CN109993058A (en) | 2019-02-27 | 2019-02-27 | The recognition methods of road signs based on multi-tag classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910144912.9A CN109993058A (en) | 2019-02-27 | 2019-02-27 | The recognition methods of road signs based on multi-tag classification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109993058A true CN109993058A (en) | 2019-07-09 |
Family
ID=67130266
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910144912.9A Pending CN109993058A (en) | 2019-02-27 | 2019-02-27 | The recognition methods of road signs based on multi-tag classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109993058A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110705378A (en) * | 2019-09-12 | 2020-01-17 | 创新奇智(合肥)科技有限公司 | Algorithm for counting quantity of articles by using multi-label network |
| CN111008672A (en) * | 2019-12-23 | 2020-04-14 | 腾讯科技（深圳）有限公司 | Sample extraction method, sample extraction device, computer-readable storage medium and computer equipment |
| CN111275107A (en) * | 2020-01-20 | 2020-06-12 | 西安奥卡云数据科技有限公司 | Multi-label scene image classification method and device based on transfer learning |
| CN112598076A (en) * | 2020-12-29 | 2021-04-02 | 北京易华录信息技术股份有限公司 | Motor vehicle attribute identification method and system |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102982344A (en) * | 2012-11-12 | 2013-03-20 | 浙江大学 | Support vector machine sorting method based on simultaneously blending multi-view features and multi-label information |
| WO2016130719A2 (en) * | 2015-02-10 | 2016-08-18 | Amnon Shashua | Sparse map for autonomous vehicle navigation |
| CN105956524A (en) * | 2016-04-22 | 2016-09-21 | 北京智芯原动科技有限公司 | Method and device for identifying traffic signs |
| CN106951911A (en) * | 2017-02-13 | 2017-07-14 | 北京飞搜科技有限公司 | A kind of quick multi-tag picture retrieval system and implementation method |
| CN107067042A (en) * | 2017-05-17 | 2017-08-18 | 江苏本能科技有限公司 | Vehicle electron identifying classification processing method and system |
| CN108416270A (en) * | 2018-02-06 | 2018-08-17 | 南京信息工程大学 | A kind of traffic sign recognition method based on more attribute union features |
| CN108664924A (en) * | 2018-05-10 | 2018-10-16 | 东南大学 | A kind of multi-tag object identification method based on convolutional neural networks |
| CN109165674A (en) * | 2018-07-19 | 2019-01-08 | 南京富士通南大软件技术有限公司 | A kind of certificate photo classification method based on multi-tag depth convolutional network |
| CN109214410A (en) * | 2018-07-10 | 2019-01-15 | 上海斐讯数据通信技术有限公司 | A kind of method and system promoting multi-tag classification accuracy rate |
-
2019
- 2019-02-27 CN CN201910144912.9A patent/CN109993058A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102982344A (en) * | 2012-11-12 | 2013-03-20 | 浙江大学 | Support vector machine sorting method based on simultaneously blending multi-view features and multi-label information |
| WO2016130719A2 (en) * | 2015-02-10 | 2016-08-18 | Amnon Shashua | Sparse map for autonomous vehicle navigation |
| CN105956524A (en) * | 2016-04-22 | 2016-09-21 | 北京智芯原动科技有限公司 | Method and device for identifying traffic signs |
| CN106951911A (en) * | 2017-02-13 | 2017-07-14 | 北京飞搜科技有限公司 | A kind of quick multi-tag picture retrieval system and implementation method |
| CN107067042A (en) * | 2017-05-17 | 2017-08-18 | 江苏本能科技有限公司 | Vehicle electron identifying classification processing method and system |
| CN108416270A (en) * | 2018-02-06 | 2018-08-17 | 南京信息工程大学 | A kind of traffic sign recognition method based on more attribute union features |
| CN108664924A (en) * | 2018-05-10 | 2018-10-16 | 东南大学 | A kind of multi-tag object identification method based on convolutional neural networks |
| CN109214410A (en) * | 2018-07-10 | 2019-01-15 | 上海斐讯数据通信技术有限公司 | A kind of method and system promoting multi-tag classification accuracy rate |
| CN109165674A (en) * | 2018-07-19 | 2019-01-08 | 南京富士通南大软件技术有限公司 | A kind of certificate photo classification method based on multi-tag depth convolutional network |
Non-Patent Citations (3)
| Title |
|---|
| YUNCHAO WEI 等: ""CNN: Single-label to Multi-label"", 《JOURNAL OF LATEX CLASS FILES》 * |
| 孙振华 等: ""基于卷积神经网络的多标签飞机识别算法"", 《计算机应用与软件》 * |
| 陈玉婷: ""基于交叉熵的CNN交通标志识别方法研究"", 《软件导刊》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110705378A (en) * | 2019-09-12 | 2020-01-17 | 创新奇智(合肥)科技有限公司 | Algorithm for counting quantity of articles by using multi-label network |
| CN111008672A (en) * | 2019-12-23 | 2020-04-14 | 腾讯科技（深圳）有限公司 | Sample extraction method, sample extraction device, computer-readable storage medium and computer equipment |
| CN111275107A (en) * | 2020-01-20 | 2020-06-12 | 西安奥卡云数据科技有限公司 | Multi-label scene image classification method and device based on transfer learning |
| CN112598076A (en) * | 2020-12-29 | 2021-04-02 | 北京易华录信息技术股份有限公司 | Motor vehicle attribute identification method and system |
| CN112598076B (en) * | 2020-12-29 | 2023-09-19 | 北京易华录信息技术股份有限公司 | Motor vehicle attribute identification method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109977812B (en) | Vehicle-mounted video target detection method based on deep learning | |
| AU2019101142A4 (en) | A pedestrian detection method with lightweight backbone based on yolov3 network | |
| CN107133570B (en) | A kind of vehicle/pedestrian detection method and system | |
| WO2022083784A1 (en) | Road detection method based on internet of vehicles | |
| CN109993058A (en) | The recognition methods of road signs based on multi-tag classification | |
| Najjar et al. | Combining satellite imagery and open data to map road safety | |
| CN109800736A (en) | A kind of method for extracting roads based on remote sensing image and deep learning | |
| CN107633220A (en) | A kind of vehicle front target identification method based on convolutional neural networks | |
| CN107886073A (en) | A kind of more attribute recognition approaches of fine granularity vehicle based on convolutional neural networks | |
| CN108509954A (en) | A kind of more car plate dynamic identifying methods of real-time traffic scene | |
| CN106372577A (en) | Deep learning-based traffic sign automatic identifying and marking method | |
| CN105844257A (en) | Early warning system based on machine vision driving-in-fog road denoter missing and early warning method | |
| CN105868700A (en) | Vehicle type recognition and tracking method and system based on monitoring video | |
| CN105354568A (en) | Convolutional neural network based vehicle logo identification method | |
| CN110097145A (en) | One kind being based on CNN and the pyramidal traffic contraband recognition methods of feature | |
| CN103996041A (en) | Vehicle color identification method and system based on matching | |
| CN110232316A (en) | A kind of vehicle detection and recognition method based on improved DSOD model | |
| CN107239730A (en) | The quaternary number deep neural network model method of intelligent automobile Traffic Sign Recognition | |
| CN109993138A (en) | A kind of car plate detection and recognition methods and device | |
| CN110807485B (en) | Method for fusing two-classification semantic segmentation maps into multi-classification semantic map based on high-resolution remote sensing image | |
| CN110321897A (en) | Divide the method for identification non-motor vehicle abnormal behaviour based on image, semantic | |
| CN109871789A (en) | Vehicle checking method under a kind of complex environment based on lightweight neural network | |
| CN108875803A (en) | A kind of detection of harmful influence haulage vehicle and recognition methods based on video image | |
| CN111523415A (en) | Image-based two-passenger one-dangerous vehicle detection method and device | |
| CN110069982A (en) | A kind of automatic identifying method of vehicular traffic and pedestrian |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190709 |