CN107506759A - A kind of motor vehicle brand identification method and device - Google Patents
A kind of motor vehicle brand identification method and device Download PDFInfo
- Publication number
- CN107506759A CN107506759A CN201610417425.1A CN201610417425A CN107506759A CN 107506759 A CN107506759 A CN 107506759A CN 201610417425 A CN201610417425 A CN 201610417425A CN 107506759 A CN107506759 A CN 107506759A
- Authority
- CN
- China
- Prior art keywords
- motor vehicle
- image
- sample
- identified
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/63—Scene text, e.g. street names
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
Abstract
The embodiments of the invention provide a kind of motor vehicle brand identification method and device, applied to electronic equipment, methods described includes：The characteristic area of described first image is identified in the first image comprising motor vehicle to be identified, wherein, the characteristic area is the region where the car plate of the motor vehicle to be identified；According to the characteristic area of described first image, and default extension rule, it is determined that the second image comprising the characteristic area；By in the network model of second image input training in advance, the characteristic pattern of second image is obtained；Wherein, the network model trains what is obtained for the second sample motor vehicle image comprising its characteristic area according to each first sample motor vehicle image, and the brand message of motor vehicle corresponding with each second sample motor vehicle image；By in the disaggregated model of characteristic pattern input training in advance, brand message corresponding to the motor vehicle to be identified is obtained.The embodiment of the present invention can improve the accuracy of motor vehicle brand recognition.
Description
Technical field
The present invention relates to technical field of intelligent traffic, more particularly to a kind of motor vehicle brand identification method and device.
Background technology
Motor vehicle, refer to drive or draw with power set, upper road driving supplies personnel riding or for transporting thing
Product and the wheeled vehicle for carrying out engineering special project operation.
Motor vehicle brand refers to the mark and title of motor vehicle manufacturers, and it represents the attribute and given price of vehicle
Value.Wherein, motor vehicle brand is typically made up of main brand, subtype number and a year money, for example, masses-Santana -2000 is motor-driven
Car brand, its main brand for masses, sub- model Santana, year money be 2000.
The identification of motor vehicle brand has very important significance in actual life, for example, for being related to criminal offence
Motor vehicle, the application of the identification of motor vehicle brand in bayonet socket and electronic police system, contained to a certain extent various
The generation of illegal activities；In field of video monitoring, the identification of motor vehicle brand also plays very in road vehicle safety precaution
Big progradation.
Existing motor vehicle brand identification method, the predominantly characteristic pattern based on artificial extraction motor vehicle image, then will
Characteristic pattern is sent into the recognizer of multi-class support vector machine.SVMs is obtained using substantial amounts of sample training in advance,
When carrying out motor vehicle brand recognition, motor vehicle image is gathered first, by manually determining the characteristic pattern of the motor vehicle image, and will be true
The characteristic pattern of fixed motor vehicle image is input in the SVMs trained, so as to obtain the brand message of the motor vehicle.
But the characteristic pattern that this method uses in identification is usually binary image, i.e., gray value is 0 or 255 figure
Picture, it is necessary to artificial empirically determined threshold value when determining binary image, and then obtained according to the threshold value by motor vehicle image
Binary image.And be difficult manually to accurately determine threshold value, when the threshold value inaccuracy manually determined, by the standard of effect characteristicses figure
True property, further influence the accuracy of final recognition result.
The content of the invention
The purpose of the embodiment of the present invention is to provide a kind of motor vehicle brand identification method and device, to improve motor vehicle product
The accuracy of board identification.Concrete technical scheme is as follows：
In a first aspect, the embodiments of the invention provide a kind of motor vehicle brand identification method, it is described applied to electronic equipment
Method includes：
The characteristic area of described first image is identified in the first image comprising motor vehicle to be identified, wherein, the spy
Region is levied as the region where the car plate of the motor vehicle to be identified；
According to the characteristic area of described first image, and default extension rule, it is determined that including the characteristic area
Second image；
By in the network model of second image input training in advance, the characteristic pattern of second image is obtained；Wherein,
The network model is the second sample motor vehicle image comprising its characteristic area according to each first sample motor vehicle image, with
And the brand message of motor vehicle corresponding with each second sample motor vehicle image trains what is obtained；
By in the disaggregated model of characteristic pattern input training in advance, obtain brand corresponding to the motor vehicle to be identified and believe
Breath.
Alternatively, the characteristic area according to described first image, and default extension rule, it is determined that comprising described
Second image of characteristic area includes：
Identify the target length and target width of the characteristic area；
According to the target length and target width of the characteristic area, determine that length is the target length first is predetermined
Multiple, width is the second prearranged multiple of the target width, and the image comprising the characteristic area is second image.
Alternatively, the disaggregated model is convolutional neural networks CNN models or support vector machines.
Alternatively, when the disaggregated model is CNN models, the classification mould that the characteristic pattern is inputted to training in advance
In type, obtain the motor vehicle to be identified corresponding to brand message include：
By the full linking layer of the CNN models of characteristic pattern input training in advance, it is corresponding to obtain the motor vehicle to be identified
Brand message.
Alternatively, the network model is：Principal component analysis network PCANET；The training process bag of the PCANET models
Include：
First sample motor vehicle image is obtained, and determines second of the characteristic area comprising each first sample motor vehicle image
Sample motor vehicle image；
By the second sample motor vehicle image, and the brand of motor vehicle corresponding with each second sample motor vehicle image
Information obtains the PCANET models as training sample, training.
Alternatively, after the training obtains the PCANET models, methods described also includes：
According to each second sample motor vehicle image, and the PCANET models that training obtains, each second is obtained
Characteristic pattern corresponding to sample motor vehicle image；
The training process of the CNN models includes：
By characteristic pattern corresponding to each second sample motor vehicle image, and the brand of motor vehicle corresponding to each characteristic pattern
Information obtains the CNN models as training sample, training.
Alternatively, methods described also includes：
Obtain the confidence level of brand message corresponding to the motor vehicle to be identified.
Alternatively, it is described obtain the motor vehicle to be identified corresponding to brand message include：
According to the confidence level of various brands information corresponding to the motor vehicle to be identified, according to confidence level from big to small suitable
Sequence, obtain brand message corresponding to the motor vehicle to be identified of predetermined quantity.
Second aspect, it is described applied to electronic equipment the embodiments of the invention provide a kind of motor vehicle brand recognition device
Device includes：
Identification module, for identifying the characteristic area of described first image in the first image comprising motor vehicle to be identified
Domain, wherein, the characteristic area is the region where the car plate of the motor vehicle to be identified；
Determining module, for the characteristic area according to described first image, and default extension rule, it is determined that including institute
State the second image of characteristic area；
First obtains module, for second image to be inputted in the network model of training in advance, obtains described second
The characteristic pattern of image；Wherein, the network model is according to the of each first sample motor vehicle image comprising its characteristic area
Two sample motor vehicle images, and the brand message of motor vehicle corresponding with each second sample motor vehicle image train what is obtained；
Second obtains module, for the characteristic pattern to be inputted in the disaggregated model of training in advance, obtains described to be identified
Brand message corresponding to motor vehicle.
Alternatively, the determining module includes：
Submodule is identified, for identifying the target length and target width of the characteristic area；
Determination sub-module, for the target length and target width according to the characteristic area, it is the mesh to determine length
The first prearranged multiple of length is marked, width is the second prearranged multiple of the target width, and includes the figure of the characteristic area
As being second image.
Alternatively, the disaggregated model is convolutional neural networks CNN models or support vector machines.
Alternatively, when the disaggregated model is CNN models, described second obtains module, specifically for by the feature
The full linking layer of the CNN models of figure input training in advance, obtains brand message corresponding to the motor vehicle to be identified.
Alternatively, the network model is：Principal component analysis network PCANET；Described device also includes：
Acquisition module, for obtaining first sample motor vehicle image, and determine comprising each first sample motor vehicle image
Second sample motor vehicle image of characteristic area；
First training module, for by the second sample motor vehicle image, and with each second sample motor vehicle image
The brand message of corresponding motor vehicle obtains the PCANET models as training sample, training.
Alternatively, described device also includes：
3rd obtains module, for according to each second sample motor vehicle image, and training obtain it is described
PCANET models, obtain characteristic pattern corresponding to each second sample motor vehicle image；
Second training module, for by characteristic pattern corresponding to each second sample motor vehicle image, and each characteristic pattern
The brand message of corresponding motor vehicle obtains the CNN models as training sample, training.
Alternatively, described device also includes：
4th obtains module, for obtaining the confidence level of brand message corresponding to the motor vehicle to be identified.
Alternatively, described second module is obtained, specifically for the various brands information according to corresponding to the motor vehicle to be identified
Confidence level, according to the order of confidence level from big to small, obtain brand corresponding to the motor vehicle to be identified of predetermined quantity and believe
Breath.
, can be according to including to be identified motor-driven the embodiments of the invention provide a kind of motor vehicle brand identification method and device
The image of car car plate, the characteristic pattern of the image is obtained using the network model of training in advance, and then according to this feature figure, by advance
The disaggregated model of training determines the brand message of motor vehicle to be identified, and compared with other regions of motor vehicle, the car plate of motor vehicle is attached
Near field has preferable identification, also, the network model is to include it previously according to each first sample motor vehicle image
Second sample motor vehicle image of car plate region, and the brand of motor vehicle corresponding with each second sample motor vehicle image
Information trains what is obtained, therefore, using the network model of the training in advance, can obtain the figure for including automobile license plate to be identified
The more accurately characteristic pattern of picture, and then, when identifying the brand message of motor vehicle to be identified according to this feature figure, it is possible to increase machine
The accuracy of motor-car brand message identification.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the required accompanying drawing used in technology description to be briefly described, it should be apparent that, drawings in the following description are only this
Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, can be with
Other accompanying drawings are obtained according to these accompanying drawings.
Fig. 1 is a kind of flow chart of motor vehicle brand identification method provided in an embodiment of the present invention；
Fig. 2 is a kind of another flow chart of motor vehicle brand identification method provided in an embodiment of the present invention；
Fig. 3 is the schematic diagram of the second sample motor vehicle image corresponding to first sample motor vehicle image；
Fig. 4 is a kind of another flow chart of motor vehicle brand identification method provided in an embodiment of the present invention；
Fig. 5 (a), 5 (b) are the second sample motor vehicle image and corresponding characteristic pattern；
Fig. 6 is a kind of structural representation of motor vehicle brand recognition device provided in an embodiment of the present invention；
Fig. 7 is a kind of another structural representation of motor vehicle brand recognition device provided in an embodiment of the present invention；
Fig. 8 is a kind of another structural representation of motor vehicle brand recognition device provided in an embodiment of the present invention.
Embodiment
In order to improve the accuracy of motor vehicle brand recognition, the embodiments of the invention provide a kind of motor vehicle brand recognition side
Method and device.
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation describes, it is clear that described embodiment is only part of the embodiment of the present invention, rather than whole embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained every other under the premise of creative work is not made
Embodiment, belong to the scope of protection of the invention.
It should be noted that in the case where not conflicting, the feature in embodiment and embodiment in the present invention can phase
Mutually combination.Describe the present invention in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
In order to improve the accuracy of motor vehicle brand recognition, the embodiments of the invention provide a kind of motor vehicle brand recognition side
Method process, applied to electronic equipment, as shown in figure 1, the process may comprise steps of：
S101, the characteristic area of described first image is identified in the first image comprising motor vehicle to be identified, wherein, institute
State the region where the car plate that characteristic area is the motor vehicle to be identified.
In embodiments of the present invention, electronic equipment can identify that this is treated according to the first image comprising motor vehicle to be identified
Identify the brand message of motor vehicle.
Wherein, electronic equipment can obtain the first image comprising motor vehicle to be identified first, and then according to first figure
Picture, identify the brand message of motor vehicle to be identified.
It should be noted that in embodiments of the present invention, electronic equipment obtains the first image for including motor vehicle to be identified
Process can use prior art, for example, in field of video monitoring, should first image comprising motor vehicle to be identified can be by
Monitoring device gathers, and the embodiment of the present invention is to this process without repeating.
After getting the first image comprising motor vehicle to be identified, electronic equipment can is according to first image, identification
The brand message of the motor vehicle to be identified.
Specifically, electronic equipment can identify the characteristic area of first image in the first image, wherein, this feature area
Domain is the region where the car plate of the motor vehicle to be identified.
For example, electronic equipment can use image-recognizing method, the car plate for the motor vehicle that the first image of identification includes, enter
And the region where determining car plate is the characteristic area of the first image.
It should be noted that in embodiments of the present invention, electronic equipment identifies the automobile license plate that the first image includes
Process, existing any image-recognizing method can be used, the embodiment of the present invention is to this process without repeating.
S102, according to the characteristic area of described first image, and default extension rule, it is determined that including the characteristic area
Second image in domain.
Electronic equipment identifies the characteristic area of the first image, that is, identifies the automobile license plate that the first image includes
Afterwards, can be further according to the characteristic area of first image, and default extension rule, it is determined that including this feature region
Second image.
For example, electronic equipment can outwards be expanded up and down centered on the characteristic area of the first image, obtain predetermined
The region of size, and the region is defined as the second image.
S103, second image is inputted in the network model of training in advance, obtain the characteristic pattern of second image；
Wherein, the network model is the second sample motor vehicle figure comprising its characteristic area according to each first sample motor vehicle image
Picture, and the brand message of motor vehicle corresponding with each second sample motor vehicle image train what is obtained.
In embodiments of the present invention, electronic equipment can include its feature previously according to each first sample motor vehicle image
The second sample motor vehicle image in region, and the brand message training of motor vehicle corresponding with each second sample motor vehicle image
Obtain network model.Using the network model, when second image of the input comprising the first image characteristic region, the model can be with
Export the characteristic pattern of second image.
Therefore, in embodiments of the present invention, when carrying out motor vehicle brand recognition, when electronic equipment obtains the second image
Afterwards, it can input second image in the network model of training in advance, obtain the characteristic pattern of second image, should with basis
Characteristic pattern, identify the brand message of motor vehicle to be identified.
S104, the characteristic pattern is inputted in the disaggregated model of training in advance, obtained corresponding to the motor vehicle to be identified
Brand message.
In embodiments of the present invention, electronic equipment can obtain disaggregated model with training in advance.Using the disaggregated model, when defeated
When entering the characteristic pattern of the second image, the disaggregated model can export the brand message of motor vehicle corresponding with this feature figure.
Therefore, in embodiments of the present invention, when carrying out motor vehicle brand recognition, when electronic equipment obtains the second image
After characteristic pattern, it can input this feature figure in the disaggregated model of training in advance, obtain the brand letter of the motor vehicle to be identified
Breath.
, can be according to including automobile license plate to be identified the embodiments of the invention provide a kind of motor vehicle brand identification method
Image, the characteristic pattern of the image is obtained using the network model of training in advance, and then according to this feature figure, by training in advance
Disaggregated model determines the brand message of motor vehicle to be identified, compared with other regions of motor vehicle, the car plate near zone of motor vehicle
With preferable identification, also, the network model is to include its car plate institute previously according to each first sample motor vehicle image
The second sample motor vehicle image in region, and the brand message instruction of motor vehicle corresponding with each second sample motor vehicle image
Get, therefore, using the network model of the training in advance, the ratio of image for including automobile license plate to be identified can be obtained
Accurate characteristic pattern, and then, when identifying the brand message of motor vehicle to be identified according to this feature figure, it is possible to increase motor vehicle product
The accuracy of board information identification.
Further, in embodiments of the present invention, electronic equipment is according to the characteristic area of the first image, it is determined that including the spy
When levying second image in region, the target length and target width in this feature region can be identified first；And then according to this feature
The target length and target width in region, determine the first prearranged multiple that length is the target length, such as 3,4,5, width is
Second prearranged multiple of the target width, such as 4,5,6, and the image comprising this feature region is the second image.
It is appreciated that the region near automobile license plate generally has a preferable identification, the motor vehicle of different brands, its
Car plate near zone generally has larger difference.Therefore, vehicle information as much as possible is contained in the second image, and is wrapped
Containing ambient noise as few as possible.
Therefore, by the second image for determining to include characteristic area, further, wait to know to identify according to second image
The brand message of other motor vehicle, it is possible to increase the accuracy of motor vehicle brand recognition.
Further, in embodiments of the present invention, for identifying that the disaggregated model of motor vehicle brand message to be identified can be with
It is convolutional neural networks CNN models or support vector machines.
When disaggregated model is CNN models, electronic equipment, can be by this feature figure after the characteristic pattern of the second image is obtained
The full linking layer of the CNN models of training in advance is inputted, and then obtains brand message corresponding to motor vehicle to be identified.
Further, in embodiments of the present invention, when electronic equipment identifies brand message corresponding to motor vehicle to be identified, also
The confidence level of brand message corresponding to the motor vehicle to be identified can be obtained.Such as, set according to user, motor vehicle to be identified is corresponding
Brand message confidence level can be between 0-1 any number, such as 0.10,0.82,0.90,0.98.When confidence level is higher,
Show that the confidence level of the corresponding brand message of the motor vehicle to be identified is higher；When confidence level is relatively low, show the machine to be identified
The confidence level of the corresponding brand message of motor-car is relatively low.
By obtaining the confidence level of brand message corresponding to motor vehicle to be identified, user can be judged by the confidence level
Whether brand message corresponding to the motor vehicle to be identified is credible, therefore, it is possible to improve Consumer's Experience.
Accordingly, when electronic equipment can obtain the confidence level of the corresponding brand message of motor vehicle to be identified, electricity
Sub- equipment when obtaining brand message corresponding to motor vehicle to be identified, can according to corresponding to motor vehicle to be identified various brands information
Confidence level, according to the order of confidence level from big to small, obtain brand message corresponding to the motor vehicle to be identified of predetermined quantity.
For example, user can be configured in advance, corresponding to the motor vehicle to be identified for making electronic equipment output predetermined quantity
Brand message, such as 2,5,8.Electronic equipment, can basis when obtaining brand message corresponding to motor vehicle to be identified
User is set, and according to the confidence level of various brands information corresponding to motor vehicle to be identified, according to the order of confidence level from big to small, obtains
Brand message corresponding to the motor vehicle to be identified of the predetermined quantity set to user.
Further, in embodiments of the present invention, electronic equipment can be obtained to be identified motor-driven for obtaining with training in advance
The network model of the characteristic pattern of the image of car second.Wherein, the network model can be：Principal component analysis network PCANET.
Specifically, as shown in Fig. 2 motor vehicle brand identification method provided in an embodiment of the present invention, can also include following
Step：
S201, first sample motor vehicle image is obtained, and determine the characteristic area for including each first sample motor vehicle image
The second sample motor vehicle image.
In embodiments of the present invention, it is motor-driven can to obtain first sample first when training PCANET models for electronic equipment
Car image.
For example, electronic equipment can obtain first sample motor vehicle image as much as possible, the first sample motor vehicle figure
Motor vehicle as every kind of brand can be included.
Further, after electronic equipment gets first sample motor vehicle image, can also determine to include each first sample
Second sample motor vehicle image of the characteristic area of this motor vehicle image.
For example, electronic equipment can use image-recognizing method, the machine that each first sample motor vehicle image includes is identified
The car plate of motor-car, and then, the region where determining car plate is the characteristic area of first sample motor vehicle image.
It should be noted that in embodiments of the present invention, electronic equipment identification first sample motor vehicle image includes
The process of automobile license plate, can use existing any image-recognizing method, the embodiment of the present invention to this process without
Repeat.
Further, electronic equipment can be according to the characteristic area of each first sample motor vehicle image, it is determined that respectively comprising spy
Levy the second sample motor vehicle image in region.
In one implementation, electronic equipment can centered on the characteristic area of each first sample motor vehicle image,
Outwards expand up and down, obtain the region of predefined size, and the region is defined as the second sample motor vehicle image.
In another implementation, electronic equipment can according to the characteristic area of first sample motor vehicle image, it is determined that
During the second sample motor vehicle image comprising this feature region, it can identify that the target length in this feature region and target are wide first
Degree；And then according to the target length and target width in this feature region, the first prearranged multiple that length is the target length is determined,
Such as 3,4,5, width is the second prearranged multiple of the target width, such as 4,5,6, and the image comprising this feature region is the
Two sample motor vehicle images.
Fig. 3 is refer to, it illustrates showing for the second sample motor vehicle image corresponding to multiple first sample motor vehicle images
It is intended to.As shown in figure 3, include the first sample image of motor vehicle corresponding to multiple brand messages in figure, such as the system of BMW -5, treasured
The system of horse -7, BMW-X1, BMW-X3, benz-E levels, benz-GLK levels, benz-R levels, benz-Viano etc..
S202, by the second sample motor vehicle image, and motor vehicle corresponding with each second sample motor vehicle image
Brand message as training sample, training obtains the PCANET models.
After obtaining each second sample motor vehicle image, electronic equipment can be by each second sample motor vehicle image, Yi Jiyu
The brand message of motor vehicle corresponding to each second sample motor vehicle image obtains PCANET models as training sample, training.
It should be noted that in embodiments of the present invention, the training process of PCANET models can use prior art, this
Inventive embodiments are to this process without repeating.
The scheme that the present embodiment provides, can be according to each first sample motor vehicle image comprising its car plate region
Second sample motor vehicle image, and the brand message of motor vehicle corresponding with each second sample motor vehicle image train to obtain
PCANET models, when carrying out motor vehicle brand recognition, it can obtain more accurately including using the PCANET models and wait to know
The characteristic pattern of second image of other automobile license plate, and then, when identifying the brand message of motor vehicle to be identified according to this feature figure,
The accuracy of motor vehicle brand message identification can be improved.
Further, in embodiments of the present invention, after training obtains PCANET models, electronic equipment can be further
The PCANET models that ground obtains according to training, training obtain disaggregated model.The disaggregated model can be CNN models.
Specifically, as shown in figure 4, motor vehicle brand identification method provided in an embodiment of the present invention, can also include following
Step：
S203, according to each second sample motor vehicle image, and the PCANET models that training obtains, obtain each
Characteristic pattern corresponding to second sample motor vehicle image.
After training obtains PCANET models, electronic equipment can be according to each second sample motor vehicle image, and is somebody's turn to do
PCANET models, obtain characteristic pattern corresponding to each second sample motor vehicle image.
Specifically, each second sample motor vehicle image can be inputted in the PCANET models, so as to obtain each second sample
Characteristic pattern corresponding to this motor vehicle image.
As shown in Fig. 5 (a), it illustrates any second sample motor vehicle image；Fig. 5 (b) is the shown in Fig. 5 (a)
Characteristic pattern corresponding to two sample motor vehicle images.
S204, by characteristic pattern corresponding to each second sample motor vehicle image, and motor vehicle corresponding to each characteristic pattern
Brand message as training sample, training obtains the CNN models.
After obtaining characteristic pattern corresponding to each second sample motor vehicle image, electronic equipment can be by each second sample motor vehicle
Characteristic pattern corresponding to image, and the brand message of motor vehicle corresponding to each characteristic pattern obtain CNN moulds as training sample, training
Type.
It should be noted that in embodiments of the present invention, the training process of CNN models can use prior art, this hair
Bright embodiment is to this process without repeating.
The scheme that the present embodiment provides, can according to corresponding to each second sample motor vehicle image characteristic pattern, and each spy
The brand message of motor vehicle corresponding to figure is levied, training obtains CNN models, can be according to bag when carrying out motor vehicle brand recognition
The characteristic pattern of the second image containing automobile license plate to be identified, the brand message of motor vehicle to be identified is obtained using the CNN models,
The accuracy of motor vehicle brand message identification can be improved.
Corresponding to above method embodiment, the embodiment of the present invention additionally provides corresponding device embodiment.
Fig. 6 is a kind of motor vehicle brand recognition device provided in an embodiment of the present invention, applied to electronic equipment, described device
Including：
Identification module 610, for identifying the feature of described first image in the first image comprising motor vehicle to be identified
Region, wherein, the characteristic area is the region where the car plate of the motor vehicle to be identified；
Determining module 620, for the characteristic area according to described first image, and default extension rule, it is determined that bag
The second image containing the characteristic area；
First obtains module 630, for that will obtain described the in the network model of second image input training in advance
The characteristic pattern of two images；Wherein, the network model is comprising its characteristic area according to each first sample motor vehicle image
Second sample motor vehicle image, and the brand message of motor vehicle corresponding with each second sample motor vehicle image train to obtain
's；
Second obtains module 640, for that will obtain described waiting to know in the disaggregated model of characteristic pattern input training in advance
Brand message corresponding to other motor vehicle.
, can be according to including automobile license plate to be identified the embodiments of the invention provide a kind of motor vehicle brand recognition device
Image, the characteristic pattern of the image is obtained using the network model of training in advance, and then according to this feature figure, by training in advance
Disaggregated model determines the brand message of motor vehicle to be identified, compared with other regions of motor vehicle, the car plate near zone of motor vehicle
With preferable identification, also, the network model is to include its car plate institute previously according to each first sample motor vehicle image
The second sample motor vehicle image in region, and the brand message instruction of motor vehicle corresponding with each second sample motor vehicle image
Get, therefore, using the network model of the training in advance, the ratio of image for including automobile license plate to be identified can be obtained
Accurate characteristic pattern, and then, when identifying the brand message of motor vehicle to be identified according to this feature figure, it is possible to increase motor vehicle product
The accuracy of board information identification.
Further, the determining module 620 includes：
Submodule (not shown) is identified, for identifying the target length and target width of the characteristic area；
Determination sub-module (not shown), for the target length and target width according to the characteristic area, it is determined that
Length is the first prearranged multiple of the target length, and width is the second prearranged multiple of the target width, and comprising described
The image of characteristic area is second image.
Further, the disaggregated model is convolutional neural networks CNN models or support vector machines.
Further, when the disaggregated model is CNN models, described second obtains module 640, specifically for by described in
The full linking layer of the CNN models of characteristic pattern input training in advance, obtains brand message corresponding to the motor vehicle to be identified.
Further, described device also includes：
4th obtains module (not shown), for obtaining the confidence of brand message corresponding to the motor vehicle to be identified
Degree.
Further, described second module 640 is obtained, specifically for the various brands according to corresponding to the motor vehicle to be identified
The confidence level of information, according to the order of confidence level from big to small, obtain product corresponding to the motor vehicle to be identified of predetermined quantity
Board information.
Further, in embodiments of the present invention, the network model is：Principal component analysis network PCANET；Such as Fig. 7 institutes
Show, described device also includes：
Acquisition module 710, for obtaining first sample motor vehicle image, and determine to include each first sample motor vehicle image
Characteristic area the second sample motor vehicle image；
First training module 720, for by the second sample motor vehicle image, and with each second sample motor vehicle figure
The brand message of motor vehicle obtains the PCANET models as training sample, training as corresponding to.
Further, as shown in figure 8, described device also includes：
3rd obtains module 730, for according to each second sample motor vehicle image, and training obtain it is described
PCANET models, obtain characteristic pattern corresponding to each second sample motor vehicle image；
Second training module 740, for by characteristic pattern corresponding to each second sample motor vehicle image, and each feature
The brand message of motor vehicle corresponding to figure obtains the CNN models as training sample, training.
, can be according to including to be identified motor-driven the embodiments of the invention provide a kind of motor vehicle brand identification method and device
The image of car car plate, the characteristic pattern of the image is obtained using the network model of training in advance, and then according to this feature figure, by advance
The disaggregated model of training determines the brand message of motor vehicle to be identified, and compared with other regions of motor vehicle, the car plate of motor vehicle is attached
Near field has preferable identification, also, the network model is to include it previously according to each first sample motor vehicle image
Second sample motor vehicle image of car plate region, and the brand of motor vehicle corresponding with each second sample motor vehicle image
Information trains what is obtained, therefore, using the network model of the training in advance, can obtain the figure for including automobile license plate to be identified
The more accurately characteristic pattern of picture, and then, when identifying the brand message of motor vehicle to be identified according to this feature figure, it is possible to increase machine
The accuracy of motor-car brand message identification.
It should be noted that herein, such as first and second or the like relational terms are used merely to a reality
Body or operation make a distinction with another entity or operation, and not necessarily require or imply and deposited between these entities or operation
In any this actual relation or order.Moreover, term " comprising ", "comprising" or its any other variant are intended to
Nonexcludability includes, so that process, method, article or equipment including a series of elements not only will including those
Element, but also the other element including being not expressly set out, or it is this process, method, article or equipment also to include
Intrinsic key element.In the absence of more restrictions, the key element limited by sentence "including a ...", it is not excluded that
Other identical element also be present in process, method, article or equipment including the key element.
Each embodiment in this specification is described by the way of related, identical similar portion between each embodiment
Divide mutually referring to what each embodiment stressed is the difference with other embodiment.It is real especially for system
For applying example, because it is substantially similar to embodiment of the method, so description is fairly simple, related part is referring to embodiment of the method
Part explanation.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the scope of the present invention.It is all
Any modification, equivalent substitution and improvements made within the spirit and principles in the present invention etc., are all contained in protection scope of the present invention
It is interior.
Claims (16)
1. a kind of motor vehicle brand identification method, it is characterised in that applied to electronic equipment, methods described includes：
The characteristic area of described first image is identified in the first image comprising motor vehicle to be identified, wherein, the characteristic area
Domain is the region where the car plate of the motor vehicle to be identified；
According to the characteristic area of described first image, and default extension rule, it is determined that including the second of the characteristic area
Image；
By in the network model of second image input training in advance, the characteristic pattern of second image is obtained；Wherein, it is described
Network model is according to the second sample motor vehicle image comprising its characteristic area of each first sample motor vehicle image, Yi Jiyu
The brand message of motor vehicle trains what is obtained corresponding to each second sample motor vehicle image；
By in the disaggregated model of characteristic pattern input training in advance, brand message corresponding to the motor vehicle to be identified is obtained.
2. according to the method for claim 1, it is characterised in that the characteristic area according to described first image, and
Default extension rule, it is determined that the second image comprising the characteristic area includes：
Identify the target length and target width of the characteristic area；
According to the target length and target width of the characteristic area, determine that length is the target length first makes a reservation for again
Number, width is the second prearranged multiple of the target width, and the image comprising the characteristic area is second image.
3. according to the method for claim 1, it is characterised in that the disaggregated model be convolutional neural networks CNN models or
Support vector machines.
4. according to the method for claim 3, it is characterised in that described by described in when the disaggregated model is CNN models
Characteristic pattern input training in advance disaggregated model in, obtain the motor vehicle to be identified corresponding to brand message include：
By the full linking layer of the CNN models of characteristic pattern input training in advance, product corresponding to the motor vehicle to be identified are obtained
Board information.
5. according to the method for claim 3, it is characterised in that the network model is：Principal component analysis network PCANET；
The training process of the PCANET models includes：
First sample motor vehicle image is obtained, and determines the second sample of the characteristic area comprising each first sample motor vehicle image
Motor vehicle image；
By the second sample motor vehicle image, and the brand message of motor vehicle corresponding with each second sample motor vehicle image
As training sample, training obtains the PCANET models.
6. according to the method for claim 5, it is characterised in that described after the training obtains the PCANET models
Method also includes：
According to each second sample motor vehicle image, and the PCANET models that training obtains, each second sample is obtained
Characteristic pattern corresponding to motor vehicle image；
The training process of the CNN models includes：
By characteristic pattern corresponding to each second sample motor vehicle image, and the brand message of motor vehicle corresponding to each characteristic pattern
As training sample, training obtains the CNN models.
7. according to any described methods of claim 1-6, it is characterised in that methods described also includes：
Obtain the confidence level of brand message corresponding to the motor vehicle to be identified.
8. according to the method for claim 7, it is characterised in that described to obtain brand letter corresponding to the motor vehicle to be identified
Breath includes：
According to the confidence level of various brands information corresponding to the motor vehicle to be identified, according to the order of confidence level from big to small, obtain
To brand message corresponding to the motor vehicle to be identified of predetermined quantity.
9. a kind of motor vehicle brand recognition device, it is characterised in that applied to electronic equipment, described device includes：
Identification module, for identifying the characteristic area of described first image in the first image comprising motor vehicle to be identified, its
In, the characteristic area is the region where the car plate of the motor vehicle to be identified；
Determining module, for the characteristic area according to described first image, and default extension rule, it is determined that including the spy
Levy second image in region；
First obtains module, for second image to be inputted in the network model of training in advance, obtains second image
Characteristic pattern；Wherein, the network model is the second sample comprising its characteristic area according to each first sample motor vehicle image
This motor vehicle image, and the brand message of motor vehicle corresponding with each second sample motor vehicle image train what is obtained；
Second obtains module, for the characteristic pattern to be inputted in the disaggregated model of training in advance, obtains described to be identified motor-driven
Brand message corresponding to car.
10. device according to claim 9, it is characterised in that the determining module includes：
Submodule is identified, for identifying the target length and target width of the characteristic area；
Determination sub-module, for the target length and target width according to the characteristic area, determine that length is grown for the target
First prearranged multiple of degree, width is the second prearranged multiple of the target width, and the image comprising the characteristic area is
Second image.
11. device according to claim 9, it is characterised in that the disaggregated model be convolutional neural networks CNN models or
Support vector machines.
12. device according to claim 11, it is characterised in that when the disaggregated model is CNN models, described second
Module is obtained, the full linking layer of the CNN models specifically for the characteristic pattern to be inputted to training in advance, obtains the machine to be identified
Brand message corresponding to motor-car.
13. device according to claim 11, it is characterised in that the network model is：Principal component analysis network
PCANET；Described device also includes：
Acquisition module, for obtaining first sample motor vehicle image, and determine the feature for including each first sample motor vehicle image
The second sample motor vehicle image in region；
First training module, for by the second sample motor vehicle image, and it is corresponding with each second sample motor vehicle image
Motor vehicle brand message as training sample, training obtains the PCANET models.
14. device according to claim 13, it is characterised in that described device also includes：
3rd obtains module, for according to each second sample motor vehicle image, and the PCANET moulds that training obtains
Type, obtain characteristic pattern corresponding to each second sample motor vehicle image；
Second training module, for characteristic pattern corresponding to each second sample motor vehicle image, and each characteristic pattern is corresponding
Motor vehicle brand message as training sample, training obtains the CNN models.
15. according to any described devices of claim 9-14, it is characterised in that described device also includes：
4th obtains module, for obtaining the confidence level of brand message corresponding to the motor vehicle to be identified.
16. device according to claim 15, it is characterised in that described second obtains module, specifically for according to
The confidence level of various brands information corresponding to motor vehicle to be identified, according to the order of confidence level from big to small, obtain predetermined quantity
Brand message corresponding to the motor vehicle to be identified.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610417425.1A CN107506759A (en) | 2016-06-14 | 2016-06-14 | A kind of motor vehicle brand identification method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610417425.1A CN107506759A (en) | 2016-06-14 | 2016-06-14 | A kind of motor vehicle brand identification method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107506759A true CN107506759A (en) | 2017-12-22 |
Family
ID=60679098
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610417425.1A Pending CN107506759A (en) | 2016-06-14 | 2016-06-14 | A kind of motor vehicle brand identification method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107506759A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111046891A (en) * | 2018-10-11 | 2020-04-21 | 杭州海康威视数字技术股份有限公司 | Training method of license plate recognition model, and license plate recognition method and device |
| CN112036421A (en) * | 2019-05-16 | 2020-12-04 | 搜狗（杭州）智能科技有限公司 | Image processing method and device and electronic equipment |
| CN115880565A (en) * | 2022-12-06 | 2023-03-31 | 江苏凤火数字科技有限公司 | Neural network-based scraped vehicle identification method and system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103488973A (en) * | 2013-09-12 | 2014-01-01 | 上海依图网络科技有限公司 | Method and system for recognizing vehicle brand based on image |
| CN105512684A (en) * | 2015-12-09 | 2016-04-20 | 江苏大为科技股份有限公司 | Vehicle logo automatic identification method based on principal component analysis convolutional neural network |
| CN105574543A (en) * | 2015-12-16 | 2016-05-11 | 武汉烽火众智数字技术有限责任公司 | Vehicle brand and model identifying method and system based on deep learning |
| US20160148072A1 (en) * | 2012-03-13 | 2016-05-26 | Massachusetts Institute Of Technology | Assisted surveillance of vehicles-of-interest |
-
2016
- 2016-06-14 CN CN201610417425.1A patent/CN107506759A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160148072A1 (en) * | 2012-03-13 | 2016-05-26 | Massachusetts Institute Of Technology | Assisted surveillance of vehicles-of-interest |
| CN103488973A (en) * | 2013-09-12 | 2014-01-01 | 上海依图网络科技有限公司 | Method and system for recognizing vehicle brand based on image |
| CN105512684A (en) * | 2015-12-09 | 2016-04-20 | 江苏大为科技股份有限公司 | Vehicle logo automatic identification method based on principal component analysis convolutional neural network |
| CN105574543A (en) * | 2015-12-16 | 2016-05-11 | 武汉烽火众智数字技术有限责任公司 | Vehicle brand and model identifying method and system based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 张顺利: "基于视觉的汽车品牌分类研究", 《中国优秀硕士学位论文全文数据库-信息科技辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111046891A (en) * | 2018-10-11 | 2020-04-21 | 杭州海康威视数字技术股份有限公司 | Training method of license plate recognition model, and license plate recognition method and device |
| CN112036421A (en) * | 2019-05-16 | 2020-12-04 | 搜狗（杭州）智能科技有限公司 | Image processing method and device and electronic equipment |
| CN115880565A (en) * | 2022-12-06 | 2023-03-31 | 江苏凤火数字科技有限公司 | Neural network-based scraped vehicle identification method and system |
| CN115880565B (en) * | 2022-12-06 | 2023-09-05 | 江苏凤火数字科技有限公司 | Neural network-based scraped vehicle identification method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107967806B (en) | Vehicle fake-license detection method, device, readable storage medium storing program for executing and electronic equipment | |
| CN105938560B (en) | Convolutional-neural-network-based vehicle model refined classification system | |
| CN109766769A (en) | A kind of road target detection recognition method based on monocular vision and deep learning | |
| CN108090423A (en) | A kind of depth detection method of license plate returned based on thermodynamic chart and key point | |
| CN110473418B (en) | Dangerous road segment identification method, dangerous road segment identification device, server and storage medium | |
| CN104463241A (en) | Vehicle type recognition method in intelligent transportation monitoring system | |
| CN105574550A (en) | Vehicle identification method and device | |
| CN105335710A (en) | Fine vehicle model identification method based on multi-stage classifier | |
| CN110533950A (en) | Detection method, device, electronic equipment and the storage medium of parking stall behaviour in service | |
| CN106781503A (en) | Method and apparatus for monitoring driving behavior | |
| CN104853972A (en) | Augmenting ADAS features of vehicle with image processing support in on-board vehicle platform | |
| CN108470458A (en) | Speed limit suggestion device and method | |
| CN110991221B (en) | Dynamic traffic red light running recognition method based on deep learning | |
| CN107506759A (en) | A kind of motor vehicle brand identification method and device | |
| CN110516633A (en) | A kind of method for detecting lane lines and system based on deep learning | |
| CN107944425A (en) | The recognition methods of road sign and device | |
| CN107967773A (en) | A kind of supermarket self-help purchase method of view-based access control model identification | |
| CN113723528A (en) | Vehicle-mounted voice-video fusion multi-mode interaction method, system, device and storage medium | |
| CN114495060B (en) | Road traffic marking recognition method and device | |
| CN112613434A (en) | Road target detection method, device and storage medium | |
| CN109948419A (en) | A kind of illegal parking automatic auditing method based on deep learning | |
| CN111008554B (en) | Deep learning-based method for identifying pedestrians without giving away in dynamic traffic zebra stripes | |
| KR20100064140A (en) | Method for detecting traffic sign board in intelligent vehicle and system for executing the method | |
| CN115908263A (en) | Road pavement damage intelligent detection system and detection method and storage medium thereof | |
| CN114463755A (en) | Automatic sensitive information detection desensitization method in high-precision map-based acquired picture |
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: 20171222 |
|
| RJ01 | Rejection of invention patent application after publication |