CN107590492A - A kind of vehicle-logo location and recognition methods based on convolutional neural networks - Google Patents
A kind of vehicle-logo location and recognition methods based on convolutional neural networks Download PDFInfo
- Publication number
- CN107590492A CN107590492A CN201710752742.3A CN201710752742A CN107590492A CN 107590492 A CN107590492 A CN 107590492A CN 201710752742 A CN201710752742 A CN 201710752742A CN 107590492 A CN107590492 A CN 107590492A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- mtd
- msubsup
- msup
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of vehicle-logo location based on convolutional neural networks and recognition methods.Its present invention uses computer vision technique, is classified by logo rough localization method, logo fine positioning method and the logo based on convolutional neural networks, realizes vehicle-logo location and identification.The present invention solves the problems, such as vehicle-logo recognition, possess preferable accuracy rate and efficiency simultaneously, improve the traditional working mode that car is distinguished with car plate, unique vehicle can be determined more accurately, admissible evidence is provided for illegal activities such as the investigation of car plate deck, escapes violating the regulations, intelligentized traffic administration is better achieved.
Description
Technical field
The invention belongs to computer vision technique and technical field of image processing, and in particular to one kind is based on convolutional Neural net
The vehicle-logo location of network and recognition methods.
Background technology
As national economy high speed development and living standards of the people improve constantly, automobile is commonly used as people's go off daily
Walking-replacing tool.However, automobile quantity is skyrocketed through bringing new challenge to urban traffic safety management, testing vehicle register is known
Other technology has turned into the important topic of intelligent transportation research field.In recent years, the orders of blocking traffic such as car plate, automobile deck are blocked
Illegal activities emerge in an endless stream, only rely on Car license recognition and do not adapt to current current situation of traffic, therefore, vehicle-logo recognition technology
Become even more important, it can make up the deficiency of Car license recognition, so as to further improve the reliability of intelligent transportation system.
The vehicle-logo recognition of moving vehicle is the research direction for comparing concern in Intelligent traffic management systems in recent years in video
One of.At present, the method for some domestic existing vehicle-logo recognitions, wherein including with the more similar technical scheme of the present invention：Patent
It is (full of leaves.Automobile mark sample training and recognition methods based on air-inlet grille positioning.CN104156692A[P].2014.) use
The method identification logo of vehicle intake grid is identified, use direction histogram of gradients (HOG) algorithm is carried out to headstock air inlet gate part
Feature extraction is simultaneously trained and classified with SVM, because the car of different brands may have a similar air inlet grill texture, and same brand
The air inlet grill of car is also possible to difference, therefore the discrimination of this method is relatively low；Patent (Lu Hui, Jiang Lianhua, Zhang Renhui.Vehicle-logo location
With recognition methods.CN103310231A[P].2013.) using Sobel operators progress coarse positioning, and car is directed to using HOG algorithms
Logo image extracts characteristic value and characteristic vector input BP neural network is identified, but this method will to vehicle-logo location algorithm
Ask higher, and HOG plays that ability to express is limited, and causing it, discrimination is not in more classification problems as the operator of engineer
It is high；Document (peaceful sparkling, Li Wenju, Wang Xinnian.Vehicle-logo recognition [J] based on principal component analysis and BP neural network.Liaoning is pedagogical
College journal (natural science edition), 2010,33 (2):Feature 179-184) is extracted and by BP god by principal component analysis (PCA)
Logo is trained and identified through network, the algorithm comparison relies on vehicle-logo location algorithm, if do not navigated to more completely
Logo picture can produce considerable influence to recognition result, in addition, interference of this method to noise is also more sensitive, BP neural network
In training, convergence is relatively slow can not even restrain.
In summary, when logo is identified, there is following deficiency in current method：(1) can not be properly positioned sometimes
Logo；(2) SVM or BP neural network are used, speed is slower in training or classification；(3) noise can to recognition result produce compared with
Big influence.The present invention proposes a kind of vehicle-logo location based on convolutional neural networks and recognition methods for these problems.
The content of the invention
To solve the above problems, the invention provides a kind of vehicle-logo location based on convolutional neural networks and recognition methods.
Described a kind of vehicle-logo location and recognition methods based on convolutional neural networks, it is characterised in that specific steps are such as
Under：
Step 1：Define logo species collection and be combined into C={ Ci| i=1 ..., t }, wherein t is the sum of logo, and establishes phase
The logo data set answered；
Step 2：Build the convolutional neural networks for logo classification and be trained with the logo data set in step 1,
Obtain convolutional neural networks；
Step 3：RGB image is gathered using the monitoring camera of intersection and the image to collecting uses medium filtering
Processing, extracts to obtain vehicle image I using automobile detecting following algorithm；
Step 4：Utilize the car plate rectangular area R=(x, y, w, h), wherein image I in Recognition Algorithm of License Plate extraction image I
The upper left corner be pixel coordinate origin, (x, y) is the coordinate in the car plate rectangular area upper left corner, and h and w is respectively car plate rectangular area
Height and width, unit is pixel, and obtains the coarse positioning region D of logo according to formula (2)：
Wherein, ρ1,ρ2For proportionality coefficient, (xd,yd) be the rectangular area D upper left corners coordinate, hdWith wdRespectively rectangle region
Domain D height and width；
Step 5：The air inlet gate region for including logo is filtered out, obtains vehicle-logo location essence region D'；
Step 6：D' is normalized to N*N pixel sizes, the convolutional neural networks of step 2 training gained is passed to, obtains defeated
Outgoing vector set U=(u1,u2,...,ut), ukLogo C is corresponded to for D'kProbability, k=1,2 ..., 10；
Step 7：The probability u of maximum is obtained according to formula (14)q, q=1,2 ..., 10, then D' vehicle-logo recognition result is Cq,
Cq∈ C, complete the identification of logo：
uq=max (u1,u2,...,ut) (14)
Wherein uqFor the value of maximum probability, q uqSubscript position.
Described a kind of vehicle-logo location and recognition methods based on convolutional neural networks, it is characterised in that specific training process
It is as follows：
Step 2.1：Step is built containing 7 layers of convolutional neural networks, and 7 layers are convolutional layer Conv1 successively, pond layer
Pool2, convolutional layer Conv3, pond layer Pool4, full articulamentum Fc5, full articulamentum Fc6, classify layer Softmax7, wherein convolution
Layer Conv1 input size is N*N, and classification layer Softmax7 output vector size is t；
Step 2.2：Random initializtion convolutional neural networks, the logo data set built using step 1 is to convolutional Neural net
Network is trained, and according to formula (1) counting loss function L, and according to the reverse error of chain rule step by step calculation, update each layer
Weight parameter value, until loss function L≤θ of output, complete training：
Wherein, YiWithThe respectively value of i-th of neuron of reality output and true tag, θ are the threshold value of setting.
Described a kind of vehicle-logo location and recognition methods based on convolutional neural networks, it is characterised in that the sieve in step 5
The air inlet gate region for including logo is selected, obtains vehicle-logo location essence region D', detailed process is：
Step 5.1：Rectangular area D is gone under hsv color space from RGB color, and is divided into equal-sized
Nrow*NcolIndividual rectangle super-pixel block, NrowWith NcolRespectively super-pixel block row sum with row sum, each super-pixel block it is big
Small is N=width*height, and filters out effective super-pixel block set D* according to formula (3), (4), (5), (6)：
Wherein, λ is standard deviation threshold method, DijThe super-pixel block arranged for the i-th row jth,For DijPixel criterion it is poor,WithRespectively DijStandard deviation and average gray on k passages, k=1,2,3,For the weight coefficient of k-th of passage,
Span be [0,360],Span is [0,1],Span is [0,1],Represent DijKth passage
Image,RepresentIn coordinate points (x, y) place grey scale pixel value；
Step 5.2：Effective sample set D* is classified according to formula (7), obtains the set L={ L of classification seti|
I=1,2 ..., 10 }, wherein LiFor the sample set of the i-th class：
Step 5.3：Counted according to formula (8) in D* per the classification results of a line super-pixel rowR=1,2 ..., 10, and
Calculated according to formula (9) and (10) whether effective per a line：If FiFor 1, then it represents that the effective row of the i-th behavior, be inactive line otherwise：
Wherein, μ1For proportionality coefficient, NLimaxFor quantity at most a kind of in the i-th row；
Step 5.4：Single effective row is filtered, i.e., effective row is all up and down inactive line, obtains new classification results Fi, and
According to formula (11), (12) and (13) obtain vehicle-logo location essence region D'：
imax=max (i | Fi=1) (11)
imin=min (i | Fi=1) (12)
By using above-mentioned technology, compared with existing automobile logo identification method, beneficial effects of the present invention are：The present invention makes
With computer vision technique, pass through logo rough localization method, logo fine positioning method and the logo based on convolutional neural networks point
Class, solve the problems, such as vehicle-logo recognition, while possess preferable accuracy rate and efficiency, improve the conventional operation mould that car is distinguished with car plate
Formula, unique vehicle can be determined more accurately, reliable card is provided for illegal activities such as the investigation of car plate deck, escapes violating the regulations
According to intelligentized traffic administration is better achieved.
Brief description of the drawings
Fig. 1 is the flow chart of the present invention；
Fig. 2 is convolutional neural networks model schematic；
Fig. 3 schemes for specific embodiment example；
Fig. 4 is the vehicle moved in the Fig. 3 extracted using detecting and tracking algorithm；
Fig. 5 is the Car license recognition schematic diagram of Fig. 4 vehicles, is noted with yellow rectangle collimation mark；
Fig. 6 is the logo coarse positioning schematic diagram obtained on the basis of Fig. 5, is marked with green rectangle frame；
Fig. 7 is super-pixel region division schematic diagram；
Fig. 8 is the whether effective statistical result of each super-pixel row；
Fig. 9 is logo fine positioning result figure.
Embodiment
A kind of specific implementation of the automobile logo identification method based on convolutional neural networks is elaborated with reference to embodiment
Method.It should be appreciated that instantiation described herein is only used for explaining the present invention, it is not intended to limit the present invention.
As shown in figure 1, a kind of detailed process of automobile logo identification method based on convolutional neural networks of the present invention, specific step
It is rapid as follows：
Step 1：Define logo species collection and be combined into C={ Ci| i=1 ..., t }, wherein t is the sum of logo, and establishes phase
The logo data set answered, in the present embodiment, t takes 10, C={ Ci|i=1,2 ..., t }={ Audi, BMW, benz, Kai Dila
Gram, Ford, Chevrolet is modern, popular, Buick, Volvo }；
Step 2：Build the convolutional neural networks for logo classification and be trained with the logo data set in step 1,
Specially：
Step 2.1：Step is built containing 7 layers of convolutional neural networks, and 7 layers are convolutional layer Conv1 successively, pond layer
Pool2, convolutional layer Conv3, pond layer Pool4, full articulamentum Fc5, full articulamentum Fc6, classify layer Softmax7, wherein convolution
Layer Conv1 input size is N*N, and classification layer Softmax7 output vector size is t, and in the present embodiment, N takes 44；
Step 2.2：Convolutional neural networks model uses step 1 structure as shown in Fig. 2 random initializtion convolutional neural networks
The logo data set built is trained to convolutional neural networks, is comprised the following steps that：
Step 2.2.1：Input picture is zoomed into 44*44 pixel sizes and inputs convolutional layer Conv1, is 5*5 with size
Convolution kernel the convolution operation that step-length is 1 is carried out to it, 20 convolution kernels altogether, it is big for 40*40 pixels to obtain 20 resolution ratio
Small characteristic pattern；
Step 2.2.2：By gained characteristic pattern input pond layer Pool2 in step 2.1.1, carrying out pond block size to it is
2*2, the maximum pondization that step-length is 2 operate, and obtain the characteristic pattern that 20 resolution ratio are 20*20 pixel sizes；
Step 2.2.3：By in previous step gained characteristic pattern input convolutional layer Conv3, with size be 5*5 with size volume
It carries out the convolution operation that step-length is 1 for product verification, altogether 50 convolution kernels, obtains 50 resolution ratio as 16*16 pixel sizes
Characteristic pattern；
Step 2.2.4：By gained characteristic pattern input pond layer Pool4 in previous step, it is 2* that pond block size is carried out to it
2nd, the average pondization that step-length is 2 operates, and obtains the characteristic pattern that 50 resolution ratio are 8*8 pixel sizes；
Step 2.2.5：64 characteristic patterns obtained by the layer Pool4 of pond are lined up into 3200 dimensional vectors with the order arranged, it is defeated
Go out to full articulamentum Fc5, Fc5 output is 500 dimensional vectors；
Step 2.2.6：By the vector input Fc6 of full articulamentum Fc5 outputs, its output is 10 dimensional vectors, Fc5 and Fc6
Form general neutral net；
Step 2.2.7：By the characteristic vector input classification layer Softmax7 of full articulamentum Fc6 outputs, input picture is obtained
The probability of corresponding 10 kinds of logos, and the label of maximum probability is exported according to formula (1) loss function L, and according to chain rule
The reverse error of step by step calculation, update the weight parameter value of each layer：
Wherein, YiWithThe respectively value of i-th of neuron of reality output and true tag, it is in the present embodiment, described
Convolutional neural networks training method it is disclosed in Application No. CN201610861396.8 file, will not be described in detail herein；
Step 2.2.8：Repeat step 2.2.1 to 2.2.7, until loss function L≤θ of output, wherein θ is the threshold of setting
Value, training is completed, in the present embodiment, θ takes 0.001；
Step 3：RGB image is gathered using the monitoring camera of intersection and the image to collecting uses medium filtering
Processing, is extracted to obtain vehicle image I using automobile detecting following algorithm, and in the present embodiment, described automobile detecting following is calculated
Method is disclosed in Application No. CN201510831439.3 file, will not be described in detail herein, result reference picture 3, Fig. 4；
Step 4：Utilize the car plate rectangular area R=(x, y, w, h), wherein image I in Recognition Algorithm of License Plate extraction image I
The upper left corner be pixel coordinate origin, (x, y) is the coordinate in the car plate rectangular area upper left corner, and h and w is respectively car plate rectangular area
Height and width, unit is pixel, and obtains the coarse positioning region D of logo according to formula (2)：
Wherein, ρ1,ρ2For proportionality coefficient, (xd,yd) be the rectangular area D upper left corners coordinate, hdWith wdRespectively rectangle region
Domain D height and width, in the present embodiment, ρ1Take 1, ρ25 are taken, described Recognition Algorithm of License Plate is in Application No.
It is disclosed in CN201510937041.8 file, it will not be described in detail herein, result reference picture 5, Fig. 6；
Step 5：The air inlet gate region for including logo is filtered out, is specially：
Step 5.1：D is gone under hsv color space from RGB color, and is divided into equal-sized Nrow*NcolIt is individual
Rectangle super-pixel block, NrowWith NcolRespectively the row sum of super-pixel block and row sum, the size of each super-pixel block is N=
Width*height, and filter out effective super-pixel block set D* according to formula (3), (4), (5), (6)：
Wherein, λ is standard deviation threshold method, DijThe super-pixel block arranged for the i-th row jth,For DijPixel criterion it is poor,WithRespectively DijStandard deviation and average gray on k passages, k=1,2,3,For the weight coefficient of k-th of passage,
Span be [0,360],Span is [0,1],Span is [0,1],Represent DijKth passage figure
Picture,RepresentIn coordinate points (x, y) place grey scale pixel value, in the present embodiment, it is 4 to select width and height,
It is 200 to select λ,Respectively 1,85,255, result reference picture 7；
Step 5.2：Effective sample set D* is classified according to formula (7), obtains the set L={ L of classification seti
I=1,2 ..., 10 }, wherein LiFor the sample set of the i-th class：
Step 5.3：Counted according to formula (8) in D* per the classification results of a line super-pixel rowR=1,2 ..., 10, and
Calculated according to formula (9) and (10) whether effective per a line：If FiFor 1, then it represents that the effective row of the i-th behavior, it is inactive line otherwise,
Result reference picture 8：
Wherein, μ1For proportionality coefficient, NLimaxFor quantity at most a kind of in the i-th row, in the present embodiment, μ is selected1For 0.5；
Step 5.4：Single effective row is filtered, i.e., effective row is all up and down inactive line, obtains new classification results Fi, and
According to formula (11), (12) and (13) obtain vehicle-logo location essence region D'：
imax=max (i | Fi=1) (11)
imin=min (i | Fi=1) (12)
Step 6：D' is normalized to N*N pixel sizes, the convolutional neural networks of step 2 training gained is passed to, obtains defeated
Outgoing vector set U=(u1,u2,...,ut), ukLogo C is corresponded to for D'kProbability, k=1,2 ..., 10；
Step 7：The probability u of maximum is obtained according to formula (14)q, q=1,2 ..., 10, then D' vehicle-logo recognition result is Cq,
Cq∈ C, complete the identification of logo：
uq=max (u1,u2,...,ut) (14)
Wherein uqFor the value of maximum probability, q uqSubscript position.
Content described in this specification embodiment is only enumerating to the way of realization of inventive concept, protection of the invention
The concrete form for being not construed as being only limitted to embodiment and being stated of scope, protection scope of the present invention is also and in this area skill
Art personnel according to present inventive concept it is conceivable that equivalent technologies mean.
Claims (3)
1. a kind of vehicle-logo location and recognition methods based on convolutional neural networks, it is characterised in that comprise the following steps that：
Step 1：Define logo species collection and be combined into C={ Ci| i=1 ..., t }, wherein t is the sum of logo, and establishes corresponding car
Mark data set；
Step 2：Build the convolutional neural networks for logo classification and be trained with the logo data set in step 1, obtained
Convolutional neural networks；
Step 3：RGB image is gathered using the monitoring camera of intersection and the image to collecting is used at medium filtering
Reason, extracts to obtain vehicle image I using automobile detecting following algorithm；
Step 4：Utilize the car plate rectangular area R=(x, y, w, h) in Recognition Algorithm of License Plate extraction image I, wherein an image I left side
Upper angle is pixel coordinate origin, and (x, y) is the coordinate in the car plate rectangular area upper left corner, and h and w is respectively the height of car plate rectangular area
And width, unit are pixel, and the coarse positioning region D of logo is obtained according to formula (2)：
<mrow>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mi>D</mi>
<mo>=</mo>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>d</mi>
</msub>
<mo>,</mo>
<msub>
<mi>y</mi>
<mi>d</mi>
</msub>
<mo>,</mo>
<msub>
<mi>w</mi>
<mi>d</mi>
</msub>
<mo>,</mo>
<msub>
<mi>h</mi>
<mi>d</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>x</mi>
<mi>d</mi>
</msub>
<mo>=</mo>
<mi>x</mi>
<mo>-</mo>
<mfrac>
<mrow>
<msub>
<mi>&rho;</mi>
<mn>1</mn>
</msub>
<mo>-</mo>
<mn>1</mn>
</mrow>
<mn>2</mn>
</mfrac>
<mo>*</mo>
<mi>w</mi>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>y</mi>
<mi>d</mi>
</msub>
<mo>=</mo>
<mi>y</mi>
<mo>-</mo>
<msub>
<mi>&rho;</mi>
<mn>2</mn>
</msub>
<mo>*</mo>
<mi>h</mi>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>w</mi>
<mi>d</mi>
</msub>
<mo>=</mo>
<msub>
<mi>&rho;</mi>
<mn>1</mn>
</msub>
<mo>*</mo>
<mi>w</mi>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>h</mi>
<mi>d</mi>
</msub>
<mo>=</mo>
<msub>
<mi>&rho;</mi>
<mn>2</mn>
</msub>
<mo>*</mo>
<mi>h</mi>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, ρ1,ρ2For proportionality coefficient, (xd,yd) be the rectangular area D upper left corners coordinate, hdWith wdRespectively rectangular area D
Height and width；
Step 5：The air inlet gate region for including logo is filtered out, obtains vehicle-logo location essence region D'；
Step 6：D' is normalized to N*N pixel sizes, is passed to the convolutional neural networks of step 2 training gained, obtain exporting to
Duration set U=(u1,u2,...,ut), ukLogo C is corresponded to for D'kProbability, k=1,2 ..., 10；
Step 7：The probability u of maximum is obtained according to formula (14)q, q=1,2 ..., 10, then D' vehicle-logo recognition result is Cq, Cq∈
C, complete the identification of logo：
uq=max (u1,u2,...,ut) (14)
Wherein uqFor the value of maximum probability, q uqSubscript position.
2. a kind of vehicle-logo location and recognition methods based on convolutional neural networks according to claim 1, it is characterised in that
Specific training process is as follows：
Step 2.1：Step is built containing 7 layers of convolutional neural networks, and 7 layers are convolutional layer Conv1, pond layer Pool2 successively, are rolled up
Lamination Conv3, pond layer Pool4, full articulamentum Fc5, full articulamentum Fc6, classify layer Softmax7, wherein convolutional layer Conv1
Input size be N*N, classification layer Softmax7 output vector size is t；
Step 2.2：Random initializtion convolutional neural networks, the logo data set built using step 1 are entered to convolutional neural networks
Row training, and according to formula (1) counting loss function L, and according to the reverse error of chain rule step by step calculation, update the weight of each layer
Parameter value, until loss function L≤θ of output, complete training：
<mrow>
<mi>L</mi>
<mo>=</mo>
<mo>-</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>t</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</munderover>
<mo>&lsqb;</mo>
<msubsup>
<mi>Y</mi>
<mi>G</mi>
<mi>i</mi>
</msubsup>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<msup>
<mi>Y</mi>
<mi>i</mi>
</msup>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<msubsup>
<mi>Y</mi>
<mi>G</mi>
<mi>i</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<msup>
<mi>Y</mi>
<mi>i</mi>
</msup>
<mo>)</mo>
</mrow>
<mo>&rsqb;</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, YiWithThe respectively value of i-th of neuron of reality output and true tag, θ are the threshold value of setting.
3. a kind of vehicle-logo location and recognition methods based on convolutional neural networks according to claim 1, it is characterised in that
The air inlet gate region filtered out comprising logo in step 5, obtains vehicle-logo location essence region D', and detailed process is：
Step 5.1：Rectangular area D is gone under hsv color space from RGB color, and is divided into equal-sized Nrow*
NcolIndividual rectangle super-pixel block, NrowWith NcolRespectively the row sum of super-pixel block is with row sum, the size of each super-pixel block
N=width*height, and filter out effective super-pixel block set D* according to formula (3), (4), (5), (6)：
<mrow>
<mi>D</mi>
<mo>*</mo>
<mo>=</mo>
<mo>{</mo>
<msub>
<mi>D</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>|</mo>
<mover>
<msub>
<mi>&sigma;</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>&le;</mo>
<mi>&lambda;</mi>
<mo>}</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>3</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mover>
<msubsup>
<mi>A</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
<mi>k</mi>
</msubsup>
<mo>&OverBar;</mo>
</mover>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mi>N</mi>
</mfrac>
<munder>
<mo>&Sigma;</mo>
<mrow>
<msubsup>
<mi>A</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
<mo>,</mo>
<mi>x</mi>
<mi>y</mi>
</mrow>
<mi>k</mi>
</msubsup>
<mo>&Element;</mo>
<msubsup>
<mi>D</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
<mi>k</mi>
</msubsup>
</mrow>
</munder>
<msubsup>
<mi>A</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
<mo>,</mo>
<mi>x</mi>
<mi>y</mi>
</mrow>
<mi>k</mi>
</msubsup>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>5</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msubsup>
<mi>&sigma;</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
<mi>k</mi>
</msubsup>
<mo>=</mo>
<msqrt>
<mrow>
<mfrac>
<mn>1</mn>
<mi>N</mi>
</mfrac>
<munder>
<mo>&Sigma;</mo>
<mrow>
<msubsup>
<mi>A</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
<mo>,</mo>
<mi>x</mi>
<mi>y</mi>
</mrow>
<mi>k</mi>
</msubsup>
<mo>&Element;</mo>
<msubsup>
<mi>D</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
<mi>k</mi>
</msubsup>
</mrow>
</munder>
<msup>
<mrow>
<mo>(</mo>
<mover>
<msubsup>
<mi>A</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
<mi>k</mi>
</msubsup>
<mo>&OverBar;</mo>
</mover>
<mo>-</mo>
<msubsup>
<mi>A</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
<mo>,</mo>
<mi>x</mi>
<mi>y</mi>
</mrow>
<mi>k</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</msqrt>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>6</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, λ is standard deviation threshold method, DijThe super-pixel block arranged for the i-th row jth,For DijPixel criterion it is poor,WithPoint
Wei not DijStandard deviation and average gray on k passages, k=1,2,3,For the weight coefficient of k-th of passage,Take
It is [0,360] to be worth scope,Span is [0,1],Span is [0,1],Represent DijKth channel image,RepresentIn coordinate points (x, y) place grey scale pixel value；
Step 5.2：Effective sample set D* is classified according to formula (7), obtains the set L={ L of classification seti| i=1,
2 ..., 10 }, wherein LiFor the sample set of the i-th class：
Step 5.3：Counted according to formula (8) in D* per the classification results of a line super-pixel rowAnd according to formula
(9) and whether (10) calculating is effective per a line：If FiFor 1, then it represents that the effective row of the i-th behavior, be inactive line otherwise：
<mrow>
<msubsup>
<mi>NL</mi>
<mi>i</mi>
<mi>r</mi>
</msubsup>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<msub>
<mi>N</mi>
<mrow>
<mi>c</mi>
<mi>o</mi>
<mi>l</mi>
</mrow>
</msub>
</munderover>
<mfenced open = "(" close = ")">
<mtable>
<mtr>
<mtd>
<mrow>
<mn>1</mn>
<mo>,</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<mi>i</mi>
<mi>f</mi>
<mi> </mi>
<msub>
<mi>D</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>&Element;</mo>
<msub>
<mi>L</mi>
<mi>r</mi>
</msub>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mn>0</mn>
<mo>,</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<mi>e</mi>
<mi>l</mi>
<mi>s</mi>
<mi>e</mi>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>8</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msub>
<mi>NL</mi>
<mrow>
<mi>i</mi>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
</msub>
<mo>=</mo>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>NL</mi>
<mi>i</mi>
<mn>1</mn>
</msubsup>
<mo>,</mo>
<msubsup>
<mi>NL</mi>
<mi>i</mi>
<mn>2</mn>
</msubsup>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<msubsup>
<mi>NL</mi>
<mi>i</mi>
<mn>10</mn>
</msubsup>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>9</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msub>
<mi>F</mi>
<mi>i</mi>
</msub>
<mo>=</mo>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mrow>
<msub>
<mi>NL</mi>
<mrow>
<mi>i</mi>
<mi>max</mi>
</mrow>
</msub>
<mo><</mo>
<msub>
<mi>&mu;</mi>
<mn>1</mn>
</msub>
<mo>*</mo>
<msub>
<mi>N</mi>
<mrow>
<mi>c</mi>
<mi>o</mi>
<mi>l</mi>
</mrow>
</msub>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mrow>
<msub>
<mi>NL</mi>
<mrow>
<mi>i</mi>
<mi>max</mi>
</mrow>
</msub>
<mo>&GreaterEqual;</mo>
<msub>
<mi>&mu;</mi>
<mn>1</mn>
</msub>
<mo>*</mo>
<msub>
<mi>N</mi>
<mrow>
<mi>c</mi>
<mi>o</mi>
<mi>l</mi>
</mrow>
</msub>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>10</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, μ1For proportionality coefficient, NLimaxFor quantity at most a kind of in the i-th row；
Step 5.4：Single effective row is filtered, i.e., effective row is all up and down inactive line, obtains new classification results Fi, and according to
Formula (11), (12) and (13) obtain vehicle-logo location essence region D'：
imax=max (i | Fi=1) (11)
imin=min (i | Fi=1) (12)
<mrow>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<msup>
<mi>D</mi>
<mo>&prime;</mo>
</msup>
<mo>=</mo>
<mrow>
<mo>(</mo>
<msup>
<mi>x</mi>
<mo>&prime;</mo>
</msup>
<mo>,</mo>
<msup>
<mi>y</mi>
<mo>&prime;</mo>
</msup>
<mo>,</mo>
<msup>
<mi>w</mi>
<mo>&prime;</mo>
</msup>
<mo>,</mo>
<msup>
<mi>h</mi>
<mo>&prime;</mo>
</msup>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msup>
<mi>x</mi>
<mo>&prime;</mo>
</msup>
<mo>=</mo>
<msub>
<mi>x</mi>
<mi>d</mi>
</msub>
<mo>+</mo>
<mfrac>
<mrow>
<msub>
<mi>w</mi>
<mi>d</mi>
</msub>
<mo>-</mo>
<msup>
<mi>w</mi>
<mo>&prime;</mo>
</msup>
</mrow>
<mn>2</mn>
</mfrac>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msup>
<mi>y</mi>
<mo>&prime;</mo>
</msup>
<mo>=</mo>
<msub>
<mi>y</mi>
<mi>d</mi>
</msub>
<mo>+</mo>
<msub>
<mi>i</mi>
<mi>min</mi>
</msub>
<mo>*</mo>
<mi>h</mi>
<mi>e</mi>
<mi>i</mi>
<mi>g</mi>
<mi>h</mi>
<mi>t</mi>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msup>
<mi>w</mi>
<mo>&prime;</mo>
</msup>
<mo>=</mo>
<msup>
<mi>h</mi>
<mo>&prime;</mo>
</msup>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msup>
<mi>h</mi>
<mo>&prime;</mo>
</msup>
<mo>=</mo>
<mrow>
<mo>(</mo>
<msub>
<mi>i</mi>
<mi>max</mi>
</msub>
<mo>-</mo>
<msub>
<mi>i</mi>
<mi>min</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>*</mo>
<mi>h</mi>
<mi>e</mi>
<mi>i</mi>
<mi>g</mi>
<mi>h</mi>
<mi>t</mi>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>13</mn>
<mo>)</mo>
</mrow>
<mo>.</mo>
</mrow>
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710752742.3A CN107590492B (en) | 2017-08-28 | 2017-08-28 | A kind of vehicle-logo location and recognition methods based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710752742.3A CN107590492B (en) | 2017-08-28 | 2017-08-28 | A kind of vehicle-logo location and recognition methods based on convolutional neural networks |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107590492A true CN107590492A (en) | 2018-01-16 |
| CN107590492B CN107590492B (en) | 2019-11-19 |
Family
ID=61050153
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710752742.3A Active CN107590492B (en) | 2017-08-28 | 2017-08-28 | A kind of vehicle-logo location and recognition methods based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107590492B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108898087A (en) * | 2018-06-22 | 2018-11-27 | 腾讯科技（深圳）有限公司 | Training method, device, equipment and the storage medium of face key point location model |
| CN108960240A (en) * | 2018-07-12 | 2018-12-07 | 浙江工业大学 | A kind of vehicle intake grid localization method based on color analysis |
| CN109215358A (en) * | 2018-08-16 | 2019-01-15 | 武汉元鼎创天信息科技有限公司 | City signal crossing safety guidance method and system based on line holographic projections technology |
| CN109948612A (en) * | 2019-03-19 | 2019-06-28 | 苏州怡林城信息科技有限公司 | Detection method of license plate, storage medium and detection device based on convolutional network |
| CN110032991A (en) * | 2019-04-23 | 2019-07-19 | 福州大学 | A kind of logo detection and recognition methods based on logo repositioning |
| CN110543838A (en) * | 2019-08-19 | 2019-12-06 | 上海光是信息科技有限公司 | Vehicle information detection method and device |
| CN110852358A (en) * | 2019-10-29 | 2020-02-28 | 中国科学院上海微系统与信息技术研究所 | Vehicle type distinguishing method based on deep learning |
| CN111091547A (en) * | 2019-12-12 | 2020-05-01 | 哈尔滨市科佳通用机电股份有限公司 | Railway wagon brake beam strut fracture fault image identification method |
| CN111259777A (en) * | 2020-01-13 | 2020-06-09 | 天地伟业技术有限公司 | End-to-end multitask vehicle brand identification method |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111460894B (en) * | 2020-03-03 | 2021-09-03 | 温州大学 | Intelligent car logo detection method based on convolutional neural network |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104156692A (en) * | 2014-07-07 | 2014-11-19 | 叶茂 | Automobile logo sample training and recognition method based on air-inlet grille positioning |
| CN104657748A (en) * | 2015-02-06 | 2015-05-27 | 中国石油大学(华东) | Vehicle type recognition method based on convolutional neural network |
| CN105205486A (en) * | 2015-09-15 | 2015-12-30 | 浙江宇视科技有限公司 | Vehicle logo recognition method and device |
| CN105354568A (en) * | 2015-08-24 | 2016-02-24 | 西安电子科技大学 | Convolutional neural network based vehicle logo identification method |
| CN105868774A (en) * | 2016-03-24 | 2016-08-17 | 西安电子科技大学 | Selective search and convolutional neural network based vehicle logo recognition method |
| CN105938560A (en) * | 2016-03-23 | 2016-09-14 | 吉林大学 | Convolutional-neural-network-based vehicle model refined classification system |
| EP3147799A1 (en) * | 2015-09-22 | 2017-03-29 | Xerox Corporation | Similarity-based detection of prominent objects using deep cnn pooling layers as features |
| CN106778745A (en) * | 2016-12-23 | 2017-05-31 | 深圳先进技术研究院 | A kind of licence plate recognition method and device, user equipment |
-
2017
- 2017-08-28 CN CN201710752742.3A patent/CN107590492B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104156692A (en) * | 2014-07-07 | 2014-11-19 | 叶茂 | Automobile logo sample training and recognition method based on air-inlet grille positioning |
| CN104657748A (en) * | 2015-02-06 | 2015-05-27 | 中国石油大学(华东) | Vehicle type recognition method based on convolutional neural network |
| CN105354568A (en) * | 2015-08-24 | 2016-02-24 | 西安电子科技大学 | Convolutional neural network based vehicle logo identification method |
| CN105205486A (en) * | 2015-09-15 | 2015-12-30 | 浙江宇视科技有限公司 | Vehicle logo recognition method and device |
| EP3147799A1 (en) * | 2015-09-22 | 2017-03-29 | Xerox Corporation | Similarity-based detection of prominent objects using deep cnn pooling layers as features |
| CN105938560A (en) * | 2016-03-23 | 2016-09-14 | 吉林大学 | Convolutional-neural-network-based vehicle model refined classification system |
| CN105868774A (en) * | 2016-03-24 | 2016-08-17 | 西安电子科技大学 | Selective search and convolutional neural network based vehicle logo recognition method |
| CN106778745A (en) * | 2016-12-23 | 2017-05-31 | 深圳先进技术研究院 | A kind of licence plate recognition method and device, user equipment |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108898087A (en) * | 2018-06-22 | 2018-11-27 | 腾讯科技（深圳）有限公司 | Training method, device, equipment and the storage medium of face key point location model |
| CN108960240A (en) * | 2018-07-12 | 2018-12-07 | 浙江工业大学 | A kind of vehicle intake grid localization method based on color analysis |
| CN108960240B (en) * | 2018-07-12 | 2022-02-15 | 浙江工业大学 | Vehicle air inlet grid positioning method based on color analysis |
| CN109215358A (en) * | 2018-08-16 | 2019-01-15 | 武汉元鼎创天信息科技有限公司 | City signal crossing safety guidance method and system based on line holographic projections technology |
| CN109948612A (en) * | 2019-03-19 | 2019-06-28 | 苏州怡林城信息科技有限公司 | Detection method of license plate, storage medium and detection device based on convolutional network |
| CN110032991A (en) * | 2019-04-23 | 2019-07-19 | 福州大学 | A kind of logo detection and recognition methods based on logo repositioning |
| CN110032991B (en) * | 2019-04-23 | 2022-07-08 | 福州大学 | Car logo detection and identification method based on car logo relocation |
| CN110543838A (en) * | 2019-08-19 | 2019-12-06 | 上海光是信息科技有限公司 | Vehicle information detection method and device |
| CN110852358A (en) * | 2019-10-29 | 2020-02-28 | 中国科学院上海微系统与信息技术研究所 | Vehicle type distinguishing method based on deep learning |
| CN111091547A (en) * | 2019-12-12 | 2020-05-01 | 哈尔滨市科佳通用机电股份有限公司 | Railway wagon brake beam strut fracture fault image identification method |
| CN111259777A (en) * | 2020-01-13 | 2020-06-09 | 天地伟业技术有限公司 | End-to-end multitask vehicle brand identification method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107590492B (en) | 2019-11-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107590492A (en) | A kind of vehicle-logo location and recognition methods based on convolutional neural networks | |
| CN107316007B (en) | Monitoring image multi-class object detection and identification method based on deep learning | |
| CN102663357A (en) | Color characteristic-based detection algorithm for stall at parking lot | |
| CN106682586A (en) | Method for real-time lane line detection based on vision under complex lighting conditions | |
| CN107798335A (en) | A kind of automobile logo identification method for merging sliding window and Faster R CNN convolutional neural networks | |
| CN103824081B (en) | Method for detecting rapid robustness traffic signs on outdoor bad illumination condition | |
| CN106022285A (en) | Vehicle type identification method and vehicle type identification device based on convolutional neural network | |
| CN105740886B (en) | A kind of automobile logo identification method based on machine learning | |
| CN104112122A (en) | Vehicle logo automatic identification method based on traffic video | |
| CN105678813A (en) | Skin color detection method and device | |
| CN107292933A (en) | A kind of vehicle color identification method based on BP neural network | |
| CN105825203A (en) | Ground arrowhead sign detection and identification method based on dotted pair matching and geometric structure matching | |
| CN101127076A (en) | Human eye state detection method based on cascade classification and hough circle transform | |
| CN106446792A (en) | Pedestrian detection feature extraction method in road traffic auxiliary driving environment | |
| CN101976338B (en) | Method for detecting judgment type visual saliency based on gradient direction histogram | |
| CN102999749A (en) | Intelligent safety belt regulation violation event detecting method based on face detection | |
| CN109949593A (en) | A kind of traffic lights recognition methods and system based on crossing priori knowledge | |
| CN103530608A (en) | Vehicle type judgment method and vehicle type judgment device | |
| CN103390167A (en) | Multi-characteristic layered traffic sign identification method | |
| CN107480585A (en) | Object detection method based on DPM algorithms | |
| CN111915583B (en) | Vehicle and pedestrian detection method based on vehicle-mounted thermal infrared imager in complex scene | |
| CN106127812A (en) | A kind of passenger flow statistical method of non-gate area, passenger station based on video monitoring | |
| CN106919902A (en) | A kind of vehicle identification and trajectory track method based on CNN | |
| CN106778646A (en) | Model recognizing method and device based on convolutional neural networks | |
| CN104036250A (en) | Video pedestrian detecting and tracking method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |