CN108596277A - A kind of testing vehicle register identification method, apparatus and storage medium - Google Patents

A kind of testing vehicle register identification method, apparatus and storage medium Download PDF

Info

Publication number
CN108596277A
CN108596277A CN201810444371.7A CN201810444371A CN108596277A CN 108596277 A CN108596277 A CN 108596277A CN 201810444371 A CN201810444371 A CN 201810444371A CN 108596277 A CN108596277 A CN 108596277A
Authority
CN
China
Prior art keywords
vehicle
image
similarity
identified
sample
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
Application number
CN201810444371.7A
Other languages
Chinese (zh)
Other versions
CN108596277B (en
Inventor
彭湃
张有才
余宗桥
郭晓威
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tencent Technology Shenzhen Co Ltd
Original Assignee
Tencent Technology Shenzhen Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tencent Technology Shenzhen Co Ltd filed Critical Tencent Technology Shenzhen Co Ltd
Priority to CN201810444371.7A priority Critical patent/CN108596277B/en
Publication of CN108596277A publication Critical patent/CN108596277A/en
Application granted granted Critical
Publication of CN108596277B publication Critical patent/CN108596277B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/08—Detecting or categorising vehicles

Abstract

The embodiment of the invention discloses a kind of testing vehicle register identification method, apparatus and storage mediums；The embodiment of the present invention can obtain at least one and refer to vehicle image when needing that vehicle image to be identified is identified, and then, calculate the vehicle image to be identified and the similarity with reference to vehicle image, obtain global similarity；And respectively from the vehicle image to be identified and with reference to the image block for extracting default marker region in vehicle image, it obtains topography to be identified and refers to topography, and calculate the topography to be identified and with reference to the similarity of topography according to twin neural network model is preset, obtain local similarity, subsequently, obtain the identity information corresponding to the reference vehicle image of global similarity and the default first condition of local similarity satisfaction, the identity information as vehicle to be identified；The program can greatly improve the validity and accuracy of identification.

Description

A kind of testing vehicle register identification method, apparatus and storage medium
Technical field
The present invention relates to fields of communication technology, and in particular to a kind of testing vehicle register identification method, apparatus and storage medium.
Background technology
In recent years, with the development of economic technology, the quantity of vehicle is also increased substantially, at the same time, with vehicle The relevant all kinds of violation cases of violation of law also increase year by year, and the real name based on vehicle registers characteristic, accurately carry out identity to vehicle Identification ensures that social security has positive meaning for scouting case.
In the prior art, typically the identity of vehicle is determined by the way that license board information is identified, for example, specifically Vehicle image can be obtained, license board information is extracted from vehicle image and identifies the license board information, is then based on recognition result Determine the identity, etc. of vehicle.But in the research and practice process to the prior art, it was found by the inventors of the present invention that Under some scenes, license board information is also likely to be present fraud, missing, fuzzy or the case where be difficult to, and most dislikes for example, existing Vehicle is doubted, the true identity, etc. for often hiding vehicle by the way of deck, therefore, existing scheme identify effective Property and accuracy are not satisfactory.
Invention content
A kind of testing vehicle register identification method, apparatus of offer of the embodiment of the present invention and storage medium, can improve having for identification Effect property and accuracy.
The embodiment of the present invention provides a kind of testing vehicle register identification method, including：
It obtains vehicle image to be identified and at least one refers to vehicle image；
The vehicle image to be identified and the similarity with reference to vehicle image are calculated, global similarity is obtained；
Respectively from the vehicle image to be identified and with reference to the figure for extracting default marker region in vehicle image As block, obtains topography to be identified and refer to topography；
The similarity of the topography to be identified and reference topography is calculated according to default twin neural network model, Obtain local similarity；
Obtain the identity corresponding to the reference vehicle image of global similarity and the default first condition of local similarity satisfaction Information, the identity information as vehicle to be identified.
Correspondingly, the embodiment of the present invention also provides a kind of vehicle identification device, including：
Acquiring unit refers to vehicle image for obtaining vehicle image to be identified and at least one；
Global calculation unit, for calculating the vehicle image to be identified and with reference to the similarity of vehicle image, obtaining complete Office's similarity；
Extraction unit, for extracting default marker from the vehicle image to be identified and reference vehicle image respectively The image block of region obtains topography to be identified and refers to topography；
Local calculation unit, for calculating the topography to be identified and reference according to default twin neural network model The similarity of topography, obtains local similarity；
Recognition unit meets the reference vehicle image of default first condition for obtaining global similarity and local similarity Corresponding identity information, the identity information as vehicle to be identified.
Optionally, in some embodiments, the recognition unit may include operation subelement and determination subelement, such as Under：
The operation subelement can be used for being weighted fortune to the global similarity and corresponding local similarity It calculates, obtains comprehensive similarity；
The determination subelement can be used for obtaining comprehensive similarity and meet the reference vehicle image institute for presetting second condition Corresponding identity information, the identity information as vehicle to be identified.
Optionally, in some embodiments, the vehicle identification device can also include setting unit, as follows：
The setting unit can be used for obtaining the true identity letter of each reference vehicle with reference to corresponding to vehicle image Breath, the identity information includes license board information and owner information, is established each with reference to vehicle image identity information corresponding with its Mapping relations, and preserve the mapping relations；
Then at this point, the determination subelement, specifically can be used for meeting comprehensive similarity into the reference for presetting second condition Vehicle image obtains the corresponding identity information of target vehicle image as target vehicle image, according to the mapping relations, as The identity information of vehicle to be identified.
Optionally, in some embodiments, the extraction unit specifically can be used for obtaining default marker information；Root According to the default marker information determine marker the vehicle image to be identified first position information, according to described first Location information intercepts the image block of default marker region from the vehicle image to be identified, obtains Local map to be identified Picture；Marker is determined in the second position information with reference to vehicle image, according to described according to the default marker information Second position information, with reference to the image block for presetting marker region is intercepted in vehicle image, obtains referring to Local map from described Picture.
Optionally, in some embodiments, the acquiring unit specifically can be used for obtaining vehicle image to be identified, with And Candidate Set is obtained, the Candidate Set includes that multiple refer to vehicle image；By in Candidate Set reference vehicle image with it is to be identified Vehicle image is matched；The reference vehicle image for being less than setting value to matching degree is filtered, Candidate Set after being filtered；From At least one is obtained after the filtering in Candidate Set and refers to vehicle image.
Optionally, in some embodiments, the vehicle identification device can also include collecting unit, combination list Member, combining unit and training unit, it is as follows：
The collecting unit, can be used for acquiring multiple vehicle sample images, and the vehicle sample image has true Identity information；
The assembled unit can be used for carrying out combination of two to multiple described vehicle sample images, to establish sample pair；
The combining unit is added to trained sample after can be used for each sample to merging into a multichannel image This concentration；
The training unit can be used for being trained according to the default initial twin model of training sample set pair, obtain twin Raw neural network model.
Optionally, in some embodiments, the sample is to including positive sample pair and negative sample pair, the assembled unit, The specific vehicle sample image that can be used for selecting to belong to from multiple described vehicle sample images same vehicle, belongs to described The vehicle sample image of same vehicle carries out combination of two, to establish positive sample pair；And from multiple described vehicle sample images The middle vehicle sample image for selecting to be not belonging to same vehicle, the vehicle sample image for being not belonging to same vehicle is carried out two-by-two Combination, to establish negative sample pair.
Optionally, in some embodiments, the combining unit is specifically determined for the vehicle of each sample centering The Color Channel is added by the Color Channel of sample image, obtains each sample to a corresponding multichannel image, Obtained multichannel image is added to training sample to concentrate.
Optionally, in some embodiments, the training unit may include trained subelement and restrain subelement, such as Under：
The trained subelement can be used for according to the training sample set respectively to the upper half of default initial twin model It is trained in branching networks and lower branch network, obtains the training sample and concentrate the corresponding sample of every multichannel image To similarity predicted value；
The convergence subelement can be used for obtaining the similarity actual value of each sample pair, true to the similarity Value and similarity predicted value are restrained, and twin neural network model is obtained.
Optionally, in some embodiments, the trained subelement specifically can be used for concentrating from the training sample and select A multichannel image is selected, as current training sample；Current training sample is directed respectively into the upper of default initial twin model It is trained in half branching networks and lower branch network, obtains upper half branching networks output vector and the output of lower branch network Vector；Dimension is carried out to upper half branching networks output vector and lower branch network output vector and connects operation entirely, is worked as The similarity predicted value of the corresponding sample pair of preceding training sample；It returns to execute from the training sample and concentrates one multichannel of selection Image, the step of as current training sample, until the multichannel image training that the training sample is concentrated finishes.
In addition, the embodiment of the present invention also provides a kind of storage medium, the storage medium is stored with a plurality of instruction, the finger Order is loaded suitable for processor, to execute the step in any testing vehicle register identification method that the embodiment of the present invention is provided Suddenly.
The embodiment of the present invention can obtain at least one and refer to vehicle when needing that vehicle image to be identified is identified Then on the one hand image calculates the vehicle image to be identified and the similarity with reference to vehicle image, obtains global similarity；Separately On the one hand, respectively from the vehicle image to be identified and with reference to the image for extracting default marker region in vehicle image Block obtains topography to be identified and refers to topography, and calculates the office to be identified according to default twin neural network model The similarity of portion's image and reference topography, obtains local similarity, subsequently, obtains global similarity and local similarity Meet the identity information corresponding to the reference vehicle image for presetting first condition, the identity information as vehicle to be identified；Due to The program can be by calculating global similarity and local similarity to match automatically and vehicle to be identified is most like has The vehicle of true identity information, and then identify the identity of the vehicle to be identified, it can only be believed by car plate accordingly, with respect to existing It, can be caused to avoid faking, lacking or obscuring because of situations such as license board information for ceasing the scheme to carry out identification The generation that can not identify or identify the situation of mistake, can reduce the dependence to vehicle license plate information to be identified, greatly improve The validity and accuracy of identification.
Description of the drawings
To describe the technical solutions in the embodiments of the present invention more clearly, make required in being described below to embodiment Attached drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for For those skilled in the art, without creative efforts, it can also be obtained according to these attached drawings other attached Figure.
Fig. 1 a are the schematic diagram of a scenario of testing vehicle register identification method provided in an embodiment of the present invention；
Fig. 1 b are the flow charts of testing vehicle register identification method provided in an embodiment of the present invention；
Fig. 1 c are the structural schematic diagrams of twin neural network model provided in an embodiment of the present invention；
Fig. 2 a are the training process structure figures of CNN provided in an embodiment of the present invention；
Fig. 2 b are that training sample set establishes schematic diagram in vehicle identification method provided in an embodiment of the present invention；
Fig. 2 c are the training process structure figures of twin neural network model provided in an embodiment of the present invention；
Fig. 2 d are another flow diagrams of testing vehicle register identification method provided in an embodiment of the present invention；
Fig. 2 e are vehicle local shape factor schematic diagrames in testing vehicle register identification method provided in an embodiment of the present invention；
Fig. 2 f are the identification process Organization Charts of twin neural network model provided in an embodiment of the present invention；
Fig. 3 a are the structural schematic diagrams of vehicle identification device provided in an embodiment of the present invention；
Fig. 3 b are another structural schematic diagrams of vehicle identification device provided in an embodiment of the present invention；
Fig. 4 is the structural schematic diagram of the network equipment provided in an embodiment of the present invention.
Specific implementation mode
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete Site preparation describes, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on Embodiment in the present invention, the every other implementation that those skilled in the art are obtained without creative efforts Example, shall fall within the protection scope of the present invention.
A kind of testing vehicle register identification method, apparatus of offer of the embodiment of the present invention and storage medium.
This vehicle identification device can be specifically integrated in the network equipment, such as the equipment such as terminal or server.
It, can be to be identified by this when user is when needing the identity to vehicle to be identified to be identified for example, with reference to Fig. 1 a The image (vehicle image i.e. to be identified) of vehicle is supplied to the network equipment, and at the same time, the network equipment can obtain at least one With reference to vehicle image, then, the vehicle image to be identified and the similarity with reference to vehicle image is calculated, global similarity is obtained, For example convolutional neural networks model may be used and calculate the vehicle image to be identified and the similarity with reference to vehicle image, it obtains complete Office's similarity etc.；And respectively from the vehicle image to be identified and with reference to extracting default marker region in vehicle image Image block, obtain topography to be identified and refer to topography, and this is calculated according to default twin neural network model and is waited for It identifies topography and the similarity with reference to topography, obtains local similarity, subsequently, obtain global similarity and part Similarity meets the identity information corresponding to the reference vehicle image for presetting first condition, and the identity as vehicle to be identified is believed Breath, and then achieve the purpose that identify vehicle identification.
It is described in detail separately below.It should be noted that the sequence of following embodiment is not as preferably suitable to embodiment The restriction of sequence.
Embodiment one,
In the present embodiment, it will be described from the angle of vehicle identification device, which specifically may be used To be integrated in the network equipment, such as the equipment such as terminal or server.
The embodiment of the present invention provides a kind of testing vehicle register identification method, including：Obtain vehicle image to be identified and at least One refers to vehicle image, calculates the vehicle image to be identified and the similarity with reference to vehicle image, obtains global similarity, with And it respectively from the vehicle image to be identified and with reference to the image block for extracting default marker region in vehicle image, obtains Topography to be identified and refer to topography, the topography to be identified and ginseng are calculated according to default twin neural network model The similarity for examining topography obtains local similarity, obtains global similarity and local similarity meets default first condition Reference vehicle image corresponding to identity information, the identity information as vehicle to be identified.
As shown in Figure 1 b, the detailed process of the testing vehicle register identification method can be as follows：
101, it obtains vehicle image to be identified and at least one refers to vehicle image.
In embodiments of the present invention, the vehicle for referring mainly to have confirmed that car owner's true identity with reference to vehicle, such as car plate letter Breath shows normal vehicle, and the vehicle to be identified refers mainly to the vehicle for needing to carry out identification, is for example car owner unconfirmed The vehicle of identity, such as show abnormal vehicle without license board information or license board information.
This is referred to reference to vehicle image comprising the image with reference to vehicle, and vehicle image to be identified refers to including to wait knowing The image of other vehicle.This can be the image that vehicle entirety is referred to comprising this with reference to vehicle image, can also be comprising the reference Vehicle presets the image of regional area, and similarly, which can be the image for including the vehicle entirety to be identified, It can also be the image that regional area is preset comprising the vehicle to be identified；The regional area can be some specified object institute on vehicle Region, which needs have distinct personal feature, for example pastes annual test mark on glass for vehicle window, interior The specified object is known as " default marker ", refers mainly to annual test mark by pendant and decoration etc. in embodiments of the present invention.Institute Annual test mark is called, refers to verification of conformity acquired when vehicle passes through relevant departments' detection within the prescribed time-limit, the year The next annual test time has been indicated on inspection mark.In general, the time of annual test for the first time of vehicle gets the time depending on licence plate, it is fixed to need later Phase checked that different automobile types inspection cycle is different, for example operation passenger car is examined 1 time every year within 5 years, more than 5 years, every 6 It examines 1 time within a month.Cargo vehicle and large-scale, medium-sized non-operation passenger car are examined 1 time every year within 10 years, more than 10 years, Examine 1 time within every 6 months, etc., the annual test time on different vehicle annual test mark is typically different.
Wherein, obtain vehicle image to be identified and at least one with reference to vehicle image mode can there are many, example Such as, the testing vehicle register identification request of user's triggering can specifically be received, wherein carried in testing vehicle register identification request to be identified Vehicle image, then, according to the testing vehicle register identification acquisition request, at least one refers to vehicle image.
It wherein, specifically can be by being shot with reference to vehicle, being intercepted from monitoring video or carried out from other picture libraries The approach such as extraction refer to vehicle image to obtain this.
Optionally, in order to reduce subsequent calculation amount, treatment effeciency is improved, it, can be with when obtaining with reference to vehicle image Preliminary screening is carried out with reference to vehicle image to these, " is obtained with the apparent inconsistent image of vehicle to be identified, i.e. step with filtering out At least one is taken to refer to vehicle image " may include：
Obtain Candidate Set, the Candidate Set include multiple refer to vehicle image, by Candidate Set reference vehicle image with wait for Identification vehicle image is matched, and the reference vehicle image that setting value is less than to matching degree is filtered, candidate after being filtered Collection obtains at least one from Candidate Set after the filtering and refers to vehicle image.
Wherein, matching way can be configured according to the demand of practical application, for example, can from vehicle ornament, The information such as interior trim, vehicle frontal, and/or the vehicle back side are compared, and using obtained similarity as matching degree.Wherein, vehicle The information such as ornament and interior trim in can be obtained by detection means, and the front of vehicle and the vehicle back side can pass through detection Vehicle key point obtains, specific detection mode can there are many, therefore not to repeat here.
102, the vehicle image to be identified and the similarity with reference to vehicle image are calculated, global similarity is obtained.
Wherein, calculate the vehicle image to be identified and with reference to vehicle image similarity mode can there are many, for example, can To be calculated using common convolutional neural networks (CNN, Convolutional Neural Network), i.e., step " calculates The similarity of the vehicle image to be identified and reference vehicle image, obtains global similarity " may include：
The vehicle image to be identified is calculated using default CNN and refers to vehicle image similarity, obtains global similarity.
Optionally, can also the overall situation similarity be calculated using another twin neural network model.I.e. step " calculates The similarity of the vehicle image to be identified and reference vehicle image, obtains global similarity " may include：
The vehicle image to be identified is calculated using default twin neural network model and refers to vehicle image similarity, is obtained Global similarity.
Wherein, what the training method with the embodiment of the present invention of another twin neural network model were provided is used to calculate The twin neural network model of local similarity is similar, for example, a large amount of vehicle general image can be acquired as vehicle sample Then image carries out combination of two to multiple vehicle sample images, to establish sample pair, such as the vehicle that same vehicle will be belonged to Sample image subsequently, utilizes this as positive sample pair using the vehicle sample image for belonging to different vehicle as negative sample pair Positive sample pair and negative sample are trained default initial twin model, obtain the twin nerve net for calculating global similarity Network model, later, can using vehicle image to be identified and with reference to vehicle image as one " image to " (i.e. image combination, with Sample is to similar), which is input in the twin neural network model for calculating global similarity, to calculate The global similarity of the vehicle image to be identified and reference vehicle image, subsequently will be described in more detail, therefore not to repeat here.
103, default marker region is extracted from the vehicle image to be identified and reference vehicle image respectively Image block obtains topography to be identified and refers to topography.
For example, can specifically obtain default marker information, presetting marker information according to this determines that marker is waited at this The first position information for identifying vehicle image intercepts default mark according to the first position information from the vehicle image to be identified The image block of object region obtains topography to be identified；And marker information is preset according to this and determines marker at this With reference to the second position information of vehicle image, marker is preset with reference to interception in vehicle image from this according to the second position information The image block of region obtains referring to topography.
Wherein, which can be depending on the demand of practical application, which, which generally requires, has Distinct personal feature, for example paste the annual test mark on glass for vehicle window, interior pendant and decoration etc., for convenience, In the present invention is implemented, it will be illustrated so that the marker is annual test mark as an example.
Wherein, the first position information and second position information specifically can be with coordinate informations.
104, basis presets twin neural network model and calculates the topography to be identified and with reference to the similar of topography Degree, obtains local similarity.
For example, specifically combination of two can be carried out by the topography to be identified and with reference to topography, multiple figures are obtained As right, a multichannel image is merged by the topography to be identified of each image pair and with reference to topography, for example 6 lead to Then the multichannel image is inputted the upper half branching networks of the twin neural network model and lower half point by the image in road respectively Branch network obtains upper half branching networks output vector and lower branch network output vector, hereafter, can calculate upper half branched network Manhatton distance between network output vector and lower branch network output vector, and according to the manhatton distance being calculated into Row dimension connects operation entirely, obtains the similarity of the corresponding image pair of the multichannel image, which is that this is to be identified The similarity of topography and reference topography, i.e. local similarity.
Optionally, it when multichannel image is inputted lower branch network, can also be pre-processed, for example, being cut The operations such as sanction, down-sampling and/or rotation, to obtain the multichannel image of the smaller scale of data enhancing, such as smaller scale Image the image of the multichannel image of the smaller scale such as smaller scale is just then imported into the twin neural network model Lower branch network in calculated.That is, upper half branching networks can handle the multichannel image of archeus, and under Half branching networks can handle the multichannel image of smaller scale.
Wherein, which can in advance be established by operation maintenance personnel, can also be by the vehicle identification Identification device is voluntarily established, i.e., step " according to default twin neural network model calculate the topography to be identified and With reference to the similarity of topography, local similarity is obtained " before, which can also include：
(1) multiple vehicle sample images are acquired, which has true identity information, such as with normal License board information and true owner information etc..
For example, specifically can be by shooting the way such as the image of a large amount of vehicle and multiple images of the same vehicle of shooting Diameter acquires multiple vehicle sample images；Alternatively, can also be by searching on the internet or from vehicle pictures database To obtain multiple vehicle sample images, etc..
Wherein, which includes the image of multiple different vehicles, also includes the different figures of same vehicle As (such as image in the shooting of different location, different time or different angle), which can be the general image of vehicle, Can be the image of vehicle regional area, due in this step, primarily to twin nerve of the training about local feature Therefore network model is mainly illustrated by taking the image of vehicle regional area as an example herein, it should be noted that, if collected Image is the general image of vehicle, then can obtain the image of vehicle regional area by cutting.The regional area can be Region on vehicle where some default marker, which needs have distinct personal feature, for example pastes in glass for vehicle window On annual test mark, interior pendant and decoration etc., in embodiments of the present invention, can mainly refer to annual test mark.
(2) combination of two is carried out to multiple vehicle sample images, to establish sample pair.
Wherein, sample to refer to combined by two vehicle sample images at set, the sample is to that can be positive sample It is right, can also be negative sample pair, positive sample is to the vehicle sample image that refers to belonging to same vehicle, for example can be by right Two images that the annual test mark of same vehicle is shot, and negative sample is to the vehicle sample graph that refers to belonging to different vehicle Picture, such as two images, etc. that can be shot by the annual test mark to different vehicle.
If the sample is to including positive sample pair and negative sample pair, step is " two-by-two to multiple vehicle sample images progress Combination, with establish sample to ", may include：
Selection belongs to the vehicle sample image of same vehicle from multiple vehicle sample images, this is belonged to same vehicle Vehicle sample image carry out combination of two, to establish positive sample pair；Selection is not belonging to same from multiple vehicle sample images The vehicle sample image of one vehicle, the vehicle sample image that this is not belonging to same vehicle carries out combination of two, to establish negative sample This is right.
(3) after by each sample to merging into a multichannel image, it is added to training sample concentration；For example, specifically may be used With as follows：
The Color Channel for determining the vehicle sample image of each sample centering, which is added, and is obtained every A sample is added to training sample to a corresponding multichannel image, by obtained multichannel image and concentrates.
For example, if each sample centering includes vehicle sample image A and B, wherein the color of vehicle sample image A and B are logical Road is 3 channels, i.e. red channel (R, Red), green channel (G, Green) and blue channel (B, Blue), then can be by vehicle Sample image A and B merge into the image in 6 channels, that is, it includes two red channels, two green channels to merge into one Then the image in 6 channel is added to training sample and is concentrated with the image of two blue channels.
Due to by each sample to merging into a multichannel image, subsequently in training pattern, calculation amount and Required computing resource can also greatly reduce, and can improve the efficiency of training pattern.
(4) it is trained according to the default initial twin model of training sample set pair, obtains twin neural network model.
Wherein, this can set default initial twin model according to the demand of practical application, for example, such as Fig. 1 c institutes Show, which may include upper half branching networks and lower branch network, wherein upper half branching networks are under Half branch's network structure is identical but does not share weight.
By taking the structure is convolutional neural networks as an example, then as illustrated in figure 1 c, which may include four convolutional layers (Convolution) and a full articulamentum (FC, Fully Connected Layers), as follows：
Convolutional layer：It is mainly used for carrying out feature to the image of input (such as training sample or need the image identified) carrying It takes, wherein convolution kernel size can be depending on practical application, for example, from first layer convolutional layer to the volume of the 4th layer of convolutional layer Product core size can be (7,7), (5,5), (3,3), (3,3) successively；Optionally, in order to reduce the complexity of calculating, meter is improved Efficiency is calculated, the convolution kernel size of this four layers of convolutional layers can also be both configured to (3,3)；Optionally, in order to improve the expression of model Non-linear factor can also be added by the way that activation primitive is added in ability, and in embodiments of the present invention, which is " relu (line rectification function, Rectified Linear Unit) ", and fill that (padding refers to attribute definition element frame Space between element content) mode is " same ", and " same " filling mode can be simply interpreted as with 0 filling edge, Number and the right that the left side (top) mends 0 (following) mend as 0 number or lack one；Optionally, it is counted to be further reduced Calculation amount, can also in second to the 4th layer of convolutional layer all layers or it is arbitrary 1~2 layer carry out down-sampling (pooling) operate, Down-sampling operation is essentially identical with the operation of convolution, and only the convolution kernel of down-sampling is the maximum value for only taking corresponding position (max pooling) or average value (average pooling) etc., for convenience, in embodiments of the present invention, will with Down-sampling operation is carried out in second layer convolutional layer and third time convolutional layer, and down-sampling operation is specially max pooling For illustrate.
It should be noted that for convenience, in embodiments of the present invention, by layer where activation primitive and down-sampling layer (also referred to as pond layer) is included into convolutional layer, it should be appreciated that it is also assumed that the structure includes convolutional layer, activation primitive Place layer, down-sampling layer (i.e. pond layer) and full articulamentum, it is, of course, also possible to include the input layer for input data and be used for The output layer of output data, therefore not to repeat here.
Full articulamentum：Can be by the Feature Mapping acquired to sample labeling space, the master in entire convolutional neural networks Play the role of " grader ", each node and last layer (the down-sampling layer in such as convolutional layer) of full articulamentum export All nodes be connected, wherein a node of full articulamentum is a neuron being known as in full articulamentum, in full articulamentum The quantity of neuron can be depending on the demand of practical application, for example, in the upper half branched network of the twin neural network model In network and lower branch network, the neuronal quantity of full articulamentum can be disposed as 512, alternatively, can also be disposed as 128, etc..It is similar with convolutional layer, it optionally, can also be non-thread to be added by the way that activation primitive is added in full articulamentum Sexual factor, for example, activation primitive sigmoid (S type functions) can be added.
Since the upper half branching networks and lower branch network of the initial twin model can export multiple vectors, and to The quantity of amount is consistent with the quantity of neuron, if for example, the nerve of the full articulamentum of upper half branching networks and lower branch network First quantity is disposed as 512, then upper half branching networks and lower branch network can export 512 vectors respectively；For another example, If the neuronal quantity of the full articulamentum of upper half branching networks and lower branch network is disposed as 128, upper half branched network Network and lower branch network can export 128 vectors, etc. respectively, therefore, as illustrated in figure 1 c, can also be arranged one it is one-dimensional The full articulamentum of degree connects entirely to carry out dimension to upper half branching networks output vector and lower branch network output vector These output vectors (are mapped as one-dimensional data) by operation by connecting entirely, obtain the corresponding similarity of input picture, than Such as the similarity, etc. between the corresponding sample pair of certain training sample.
Based on the structure of above-mentioned default initial twin model, step is " according to the default initial twin model of training sample set pair It is trained, obtains twin neural network model " it specifically can be as follows：
S1, according to the training sample set respectively to the upper half branching networks and lower branch network of default initial twin model In be trained, obtain the similarity predicted value that the training sample concentrates the corresponding sample pair of every multichannel image.
For example, can specifically concentrate one multichannel image of selection from the training sample, (i.e. should as current training sample Current training sample is a multichannel image, and the multichannel image corresponds to a sample pair, that is to say, that the multichannel figure As corresponding two vehicle sample images), current training sample is directed respectively into the upper half branching networks of default initial twin model It is trained in lower branch network, upper half branching networks output vector and lower branch network output vector is obtained, to upper Half branching networks output vector and lower branch network output vector carry out dimension and connect operation entirely, obtain current training sample The similarity predicted value of corresponding sample pair returns to execute from the training sample and concentrates one multichannel image of selection, as working as The step of preceding training sample, until the multichannel image training that the training sample is concentrated finishes.
For example, specifically can by current training sample import preset twin neural network model upper half branching networks in into Row training obtains upper half branching networks output vector, and carries out default processing to current training sample, will currently be instructed after processing Practice sample import preset twin neural network model lower branch network in be trained, obtain lower branch network export to Amount, wherein the default processing can depending on the demand of practical application, for example, current training sample can be cut out, The operations such as down-sampling and/or rotation, to obtain the current training sample of the smaller scale of data enhancing, that is to say, that on Half branching networks can handle the training sample of archeus, and lower branch network can handle the training sample of smaller scale. Hereafter, the manhatton distance between upper half branching networks output vector and lower branch network output vector, and root can be calculated Dimension is carried out according to the manhatton distance being calculated and connects operation entirely, obtains the similar of the corresponding sample pair of current training sample Spend predicted value.
Optionally, operation result can also be connected entirely to the dimension using activation primitive to handle, i.e., on obtaining After half branching networks output vector and lower branch network output vector, upper half branching networks output vector can be calculated under Manhatton distance (L between half branching networks output vector1Distance), and it is one-dimensional according to the manhatton distance progress being calculated The full connection operation of degree, the result for being connected operation entirely to dimension using default activation primitive are calculated, obtain currently training sample The similarity predicted value of this corresponding sample pair.
Wherein, which can be depending on the demand of practical application, for example, being specifically as follows sigmoid。
S2, the similarity actual value for obtaining each sample pair, receive the similarity actual value and similarity predicted value It holds back, obtains twin neural network model.
The similarity actual value and similarity predicted value are restrained for example, default loss function specifically may be used, Obtain twin neural network model.
Wherein, which can be flexibly arranged according to practical application request, for example, loss function J can be selected It is as follows for cross entropy：
Wherein, C is class number, and whether the different values representative of C=2, k ∈ (1,2), k belong to same vehicle,It is defeated The similarity predicted value gone out, ykFor similarity actual value.By reducing between network similarity predicted value and similarity actual value Error, carry out constantly train, to adjust weight to appropriate value, the twin neural network model can be obtained.
105, it obtains global similarity and local similarity meets corresponding to the reference vehicle image for presetting first condition Identity information, the identity information as vehicle to be identified.
Wherein, which can be configured according to the demand of practical application, for example, can by global similarity and After local similarity is merged according to a certain percentage, vehicle image conduct is suitably referred to screen based on the fusion results Target vehicle image, i.e., for example, step " obtains global similarity and local similarity meets the reference vehicle for presetting first condition Identity information corresponding to image, the identity information as vehicle to be identified " specifically can be as follows：
(1) overall situation similarity and corresponding local similarity are weighted, obtain comprehensive similarity.For example, Being formulated can be as follows：
Sim=(1- μ) simglobal+μsimlocal；
Wherein, sim is comprehensive similarity, simglobalFor global similarity, simlocalFor local similarity, μ is weight, μ In (0,1) range, the specific value of μ can be depending on the demand of practical application, and details are not described herein.
(2) identity information corresponding to the reference vehicle image of the default second condition of comprehensive similarity satisfaction is obtained, as The identity information of vehicle to be identified, that is, comprehensive similarity is met into the reference vehicle image for presetting second condition as target carriage Image, and using the identity information corresponding to target vehicle image as the identity information of vehicle to be identified.
Wherein, which can be " being higher than predetermined threshold value ", can also be " the highest preceding N of comprehensive similarity It is a ", the value of the predetermined threshold value and N can be depending on the demands of practical application, and N is positive integer, for example, by taking N is 10 as an example, Obtained multiple comprehensive similarities can be then ranked up, then, higher first 10 of comprehensive similarity be selected to refer to vehicle Image is as target vehicle image, etc., and therefore not to repeat here.
Wherein, the identity information with reference to corresponding to vehicle image, can be based on default mapping relations (with reference to vehicle image and With reference to the mapping relations between the true identity information of vehicle) obtain, you can choosing, obtain comprehensive similarity meet it is default Before identity information of the identity information as vehicle to be identified corresponding to the reference vehicle image of second condition, the vehicle identification Recognition methods can also include：
Each true identity information of reference vehicle with reference to corresponding to vehicle image is obtained, which may include License board information and owner information etc.；Each mapping relations with reference to vehicle image identity information corresponding with its are established, and are preserved The mapping relations.
Then at this point, step " obtains the identity corresponding to the reference vehicle image of the default second condition of comprehensive similarity satisfaction Information, the identity information as vehicle to be identified ", is specifically as follows：Comprehensive similarity is met to the reference for presetting second condition Vehicle image obtains the corresponding identity information of target vehicle image as target vehicle image, according to the mapping relations, as waiting for Identify the identity information of vehicle.
From the foregoing, it will be observed that the present embodiment when needing that vehicle image to be identified is identified, can obtain at least one ginseng Vehicle image is examined, then, on the one hand calculates the vehicle image to be identified and the similarity with reference to vehicle image, is obtained global similar Degree；On the other hand, default marker region is extracted from the vehicle image to be identified and reference vehicle image respectively Image block obtains topography to be identified and refers to topography, and calculates this according to default twin neural network model and wait knowing The similarity of other topography and reference topography, obtains local similarity, subsequently, obtains global similarity and Local Phase Meet the identity information corresponding to the reference vehicle image for presetting first condition, the identity information as vehicle to be identified like degree； Since the program can be most like with vehicle to be identified to match automatically by calculating global similarity and local similarity Vehicle with true identity information, and then identify the identity of the vehicle to be identified, it can only pass through vehicle accordingly, with respect to existing For scheme of the board information to carry out identification, it can be led to avoid faking, lacking or obscuring because of situations such as license board information The generation of the situation that can not identify or identify mistake of cause, can reduce the dependence to vehicle license plate information to be identified, significantly Improve the validity and accuracy of identification.
Embodiment two,
According to method described in preceding embodiment, citing is described in further detail below.
In the present embodiment, will with the vehicle identifier specifically it is integrated in the network device, and by common CNN come Calculate vehicle image to be identified and with reference to vehicle image global similarity for illustrate.
(1) training of model.
For example, first, the network equipment can acquire a large amount of vehicle sample image, multiple vehicle samples by multiple approach This image may include the image of multiple different vehicles, also include the different images of same vehicle, such as in different location, difference Time or different angle shoot obtained image to same vehicle, which can be the general image of vehicle, can also be The image of vehicle regional area can obtain vehicle office if the image collected is the general image of vehicle by cutting The image in portion region, for example the topography of annual test mark region can be therefrom extracted (usually in front windshield The upper right corner), etc..
Secondly, the CNN that this can be trained common using the general image of these vehicles is (by carrying out based on global similarity Calculate), and the twin neural network model (for carrying out local similarity calculating) is trained using the image of vehicle regional area, It specifically can be as follows：
(1) training of common CNN；
After the network equipment is screened multiple vehicle sample images comprising vehicle entirety, for example carries out duplicate removal or go After falling the fuzzy vehicle sample image of some displays, the training sample that remaining vehicle sample image is added to the CNN is concentrated, Then, the network equipment can preset initial model according to the training sample set pair and be trained, and obtain CNN.
Wherein, which may include four convolutional layers and a full articulamentum.In order to drop The complexity of low calculating improves computational efficiency, and in the present embodiment, the convolution kernel size of this four layers of convolutional layers can be both configured to (3,3), activation primitive is all made of " relu ", and padding modes are disposed as " same ".
Optionally, it in order to be further reduced calculation amount, can also be carried out in second layer convolutional layer and third time convolutional layer Down-sampling operates, such as maxpooling.
After carrying out maxpooling operations, the output after being operated to maxpooling by full articulamentum carries out Mapping, the neuronal quantity of full articulamentum could be provided as 512 (or being disposed as 128, etc.), and may be used Sigmoid is as activation primitive.
When needing to carry out model training, the network equipment can concentrate two vehicle sample images of selection from the training sample As current training sample；Then, as shown in Figure 2 a, which is imported into the initial pattern, is currently trained The corresponding similarity predicted value of sample obtains the similarity actual value of the current training sample, and using default loss function pair The similarity actual value and similarity predicted value are restrained, to adjust the parameters in the twin neural network model to conjunction Suitable numerical value subsequently can return to execution and select two vehicle sample images as current training sample from training sample concentration This step of, is calculated and is restrained with the similarity predicted value between other vehicle sample images for being concentrated to training sample, Until all vehicle sample images that the training sample is concentrated calculate and convergence finishes, you can obtain required CNN.
Wherein, which can be not limited thereto depending on the demand of practical application.
(2) training of twin neural network model；
The network equipment can include that the vehicle sample image of vehicle regional area carries out combination of two to multiple, to establish sample This is right, for example, the vehicle sample image for belonging to same vehicle (including vehicle regional area) positive sample pair can be used as, will belong to It is used as negative sample pair in the vehicle sample image (including vehicle regional area) of different vehicle, then, it is determined that each sample centering Vehicle sample image Color Channel, which is added, obtains each sample to a corresponding multichannel Image, and by obtained multichannel image be added to twin neural network model training sample concentrate.
For example, referring to Fig. 2 b, if vehicle sample image A1, vehicle sample image A2 and vehicle sample image A3 etc. are vehicle The different images of A, vehicle sample image B1 ... and the different images that vehicle sample image Bn is vehicle B, vehicle sample Image C is the image of vehicle C, and these vehicle sample images are the image in 3 channels (Color Channel RGB), then network is set It is standby to make following combination to these vehicle sample images and merge：
Vehicle sample image A1 and vehicle sample image A2 are combined, as positive sample pair, and merge into 6 channels The multichannel image 1 of (two red channels, two green channels and two blue channels), and the multichannel image 1 that will be obtained The training sample for being added to twin neural network model is concentrated；
Vehicle sample image A1 and vehicle sample image A3 are combined, as positive sample pair, and merge into 6 channels The multichannel image 2 of (two red channels, two green channels and two blue channels), and the multichannel image 2 that will be obtained The training sample for being added to twin neural network model is concentrated；
Vehicle sample image A1 and vehicle sample image B1 are combined, as negative sample pair, and merge into 6 channels The multichannel image 3 of (two red channels, two green channels and two blue channels), and the multichannel image 3 that will be obtained The training sample for being added to twin neural network model is concentrated；
……
Vehicle sample image A2 and vehicle sample image Bn are combined, as negative sample pair, and merge into 6 channels The multichannel image n-1 of (two red channels, two green channels and two blue channels), and the multichannel image that will be obtained The training sample that n-1 is added to twin neural network model is concentrated；
Vehicle sample image A2 and vehicle sample image C are combined, as negative sample pair, and merge into 6 channels (two A red channel, two green channels and two blue channels) multichannel image n, and the multichannel image n addition that will be obtained Training sample to twin neural network model is concentrated.
After the training sample set for obtaining twin neural network model, the network equipment can be according to the training sample set pair Default initial twin model is trained, and obtains twin neural network model.
Wherein, which may include upper half branching networks With lower branch network, the identical CNN of structure, but not shared power may be used in the upper half branching networks and lower branch network Weight, that is to say, that the twin neural network model includes two CNN, wherein each CNN may include four convolutional layers with One full articulamentum.In order to reduce the complexity of calculating, computational efficiency, in the present embodiment, the volume of this four layers of convolutional layers are improved Product core size can be both configured to (3,3), and activation primitive is all made of " relu ", and padding modes are disposed as " same "；It can Choosing, in order to be further reduced calculation amount, down-sampling operation can also be carried out in second layer convolutional layer and third time convolutional layer, Such as maxpooling.After carrying out maxpooling operations, after being operated to maxpooling by full articulamentum Output is mapped, wherein in the present embodiment, either upper half branching networks or lower branch network, full articulamentum Neuronal quantity could be provided as 512 (or being disposed as 128, etc.), and may be used sigmoid as swash Function living.
In addition, as shown in Figure 2 c, the default initial twin model is in addition to may include upper half branching networks and lower branch Except network, the full articulamentum of a dimension can also be included, for by the defeated of upper half branching networks and lower branch network Outgoing vector is mapped as one-dimensional data；Wherein, the neuronal quantity of the full articulamentum of the dimension is 1, and activation primitive can be adopted Use sigmoid.
When needing to carry out model training, the network equipment can be concentrated from the training sample of the twin neural network model and be selected A multichannel image (multichannel image corresponds to a sample pair, that is, corresponds to two vehicle sample images) is selected, as working as Preceding training sample；Then, as shown in Figure 2 c, on the one hand, the current training sample can be imported in advance according to original scale size If the upper half branching networks of initial twin model, obtain upper half branching networks output vector, on the other hand, can be to currently training The operations such as sample is cut out, down-sampling and/or rotation, to obtain the training sample of the smaller scale of data enhancing, so Afterwards, the training sample of the smaller scale is imported in the lower branch network of default initial twin model and is trained, obtained down Half branching networks output vector；Hereafter, can calculate upper half branching networks output vector and lower branch network output vector it Between manhatton distance (L1 distances), dimension carried out according to the manhatton distance that is calculated connect operation entirely (to connect entirely One neuron), and the result for using activation primitive sigmoid to connect operation entirely to dimension calculates, and is currently instructed The similarity predicted value for practicing the corresponding sample pair of sample obtains the similarity actual value of the sample pair, and using default loss letter It is several that the similarity actual value and similarity predicted value are restrained, to adjust the parameters in the initial twin model to conjunction Suitable numerical value subsequently can return to execution and select a multichannel image as current training sample from training sample concentration The step of, it is calculated and is restrained with the similarity predicted value for other multichannel images concentrated to training sample, until the instruction All multichannel images in white silk sample set calculate and convergence finishes, you can the model after being trained, i.e., required is twin Neural network model.
Wherein, loss function J can be selected as cross entropy, as follows：
Wherein, C is class number, and whether the different values representative of C=2, k ∈ (1,2), k belong to same vehicle, for output Similarity predicted value, be similarity actual value.
It should be noted that when hands-on, which can be without pre-training, and direct normal distribution is initial Change weight, since the number of plies is shallower, convergence rate is very fast, for example, being restrained after about 40 epoch, that is to say, that the present invention is implemented It is less (lightweight) that the twin neural network model that is provided of example does not only take up computing resource, and recognition speed is fast, efficiency It is higher.
In addition, it should be noted that, in order to ensure the accuracy of the common CNN and the identification of twin neural network model, It, can be with the new vehicle sample of timing acquiring other than it can be trained offline to the CNN and twin neural network model Image is updated with the training sample concentrated to each training sample, and based on training sample set pair CNN and twin after update Raw neural network model is updated, that is, the CNN and twin neural network model is allow constantly to be learnt.
(2) it establishes and refers to vehicle identity information library.
It acquires multiple and refers to vehicle image, establish Candidate Set, in addition, the network equipment can also obtain each reference vehicle figure As the corresponding true identity information of reference vehicle, such as license board information and owner information etc.；Then, it establishes each with reference to vehicle The mapping relations of image identity information corresponding with its, and the mapping relations are preserved into reference to vehicle identity information library.
For example, the network equipment specifically can be by shooting with reference to vehicle, or carried from registered databases such as vehicle administration offices The approach such as take to obtain with reference to vehicle image, wherein refer mainly to have confirmed that the vehicle of car owner's true identity, such as vehicle with reference to vehicle The normal vehicle of board presentation of information.
(3) testing vehicle register identification.
As shown in Figure 2 d, based on model after above-mentioned training, the detailed process of the vehicle identification method can be as follows：
201, the network equipment obtains testing vehicle register identification request, wherein is carried in testing vehicle register identification request to be identified Vehicle image.
For example, specifically can be by user by being shot to vehicle to be identified or the approach such as extracting from other picture libraries It obtains the vehicle image to be identified, and is supplied to the network equipment.
Wherein, vehicle to be identified refers mainly to the vehicle for needing to identify identity, for example is that car owner's body is confirmed in monitoring video The vehicle of part, such as show abnormal vehicle without license board information or license board information.
202, the network equipment can obtain Candidate Set, wherein the Candidate Set after receiving testing vehicle register identification request It may include that multiple refer to vehicle image, then execute step 203.
Wherein, it refers to, comprising the image with reference to vehicle, referring mainly to have confirmed that car owner is true with reference to vehicle with reference to vehicle image The vehicle of real identity, such as license board information show normal vehicle.
203, the network equipment matches the reference vehicle image in Candidate Set with vehicle image to be identified, to Candidate Set The reference vehicle image that middle matching degree is less than setting value is filtered, Candidate Set after being filtered, and then executes step 204.
Wherein, for setting value to be configured according to the demand of practical application, matching way can also be according to practical application Demand is configured, for example, can be compared from information such as ornament, interior trim, vehicle frontal, and/or the vehicle back sides in vehicle It is right, and using obtained similarity as matching degree, you can the reference vehicle image of apparent dissmilarity to be filtered out.Wherein, vehicle The information such as ornament and interior trim in can be obtained by detection means, and the front of vehicle and the vehicle back side can pass through detection Vehicle key point obtains, specific detection mode can there are many, therefore not to repeat here.
204, the network equipment determines current to be treated with reference to vehicle image from Candidate Set after the filtering.
205, the network equipment calculates vehicle image to be identified using the CNN after training and this is current to be treated with reference to vehicle The similarity of image, obtains global similarity.
206, the network equipment extracts the image block of default marker region from vehicle image to be identified, is waited for It identifies topography, and from reference to the image block for extracting default marker region in vehicle image, obtains with reference to office Then portion's image executes step 207.
For example, the network equipment can specifically obtain default marker information, presetting marker information according to this determines mark Object is intercepted according to the first position information from the vehicle image to be identified in the first position information of the vehicle image to be identified The image block of default marker region, obtains topography to be identified；And marker information is preset according to this and determines mark Will object refers to the second position information of vehicle image at this, pre- with reference to being intercepted in vehicle image from this according to the second position information If the image block of marker region, obtain with reference to topography, etc..
Wherein, which can be depending on the demand of practical application, which, which generally requires, has Distinct personal feature, for example paste the annual test mark on glass for vehicle window, interior pendant and decoration etc., in the present embodiment, Mainly presets for marker is specially annual test mark and illustrate by this.
For example, Fig. 2 e are referred to, which includes ' inspection ' word, and searching shows next below or above In the time (such as 2010) of secondary inspection vehicle, be the Arabic numerals of 1-12 around searching, one of them can be perforated, and beat that of hole A Arabic numerals just represent the month (for example the number punched in Fig. 2 e is 4) for examining vehicle next time, before being normally at vehicle The windshield upper right corner, and since the size of 80 × 80 pixel values (pixels) is enough to cover a complete annual test mark, therefore The tile size of extraction generally could be provided as being no more than 80*80pixels, and certainly, the size in the extraction region can basis Practical application scene is flexibly adjusted, and is not limited thereto.
Optionally, it since global similarity is bigger, represents target vehicle and vehicle to be identified is more alike in appearance, because This, in order to reduce local feature matching, (namely annual test tag match, the i.e. data processing amount of step 207) can be selected only global Highest first M of similarity carries out annual test sign image extraction with reference to vehicle image, so that it is guaranteed that being carried for annual test sign image Reference vehicle in the reference vehicle image taken is roughly the same with vehicle appearance to be identified, for example belongs to same vehicle, same Color, same brand etc..Wherein, M is positive integer, and specific value can be depending on the demand of practical application.
Wherein, step 205 and 206 execution can be in no particular order.
207, the network equipment (is trained by (one) model training part twin according to the twin neural network model Raw neural network model) topography to be identified and the similarity with reference to topography are calculated, obtain local similarity.
For example, as shown in figure 2f, the computational methods of the similarity of the topography to be identified and reference topography are specific It can be as follows：
First, the network equipment can be combined by the topography to be identified and with reference to topography, as " an image It is right ", and by " image to " the topography to be identified and with reference to topography merge into a multichannel image, such as 6 logical The image K in road.
Secondly, on the one hand, it is twin that the image K of for example original scale size of the multichannel image can be inputted this by the network equipment It is calculated in the upper half branching networks of neural network model, upper half branching networks vector is obtained, on the other hand, to the multichannel The operations such as image is cut out, down-sampling and/or rotation, to obtain the multichannel image of the smaller scale of data enhancing, As then the image K of the multichannel image of the smaller scale such as smaller scale is imported the twin god by the image K of smaller scale It is calculated in lower branch network through network model, obtains lower branch network output vector.
Hereafter, the network equipment can calculate between upper half branching networks output vector and lower branch network output vector Manhatton distance (L1Distance), dimension is carried out according to the manhatton distance being calculated and connects operation (i.e. full connection one entirely Neuron), and the result of operation connects dimension using activation primitive sigmoid entirely and calculates, is somebody's turn to do " image to " Similarity predicted value, wherein should the similarity predicted value of " image to " be the vehicle image to be identified and refer to vehicle figure The local similarity of picture.
And so on, Local Phase of the vehicle image to be identified with other with reference to vehicle image can be obtained according to aforesaid way Like degree.
208, the network equipment is similar with obtained part in step 207 to obtained overall situation similarity in step 205 Degree is weighted, and obtains comprehensive similarity, then executes step 209.It can be as follows for example, being formulated：
Sim=(1- μ) simglobal+μsimlocal；
Wherein, sim is comprehensive similarity, simglobalFor global similarity, simlocalFor local similarity, μ is weight, μ In (0,1) range, the specific value of μ can be depending on the demand of practical application, and details are not described herein.
209, comprehensive similarity is met the reference vehicle image for presetting second condition as target vehicle figure by the network equipment Then picture executes step 210.
Wherein, which can be " being higher than predetermined threshold value ", can also be " the highest preceding N of comprehensive similarity It is a ", the value of the predetermined threshold value and N can be depending on the demands of practical application, and N is positive integer, for example, by taking N is 2 as an example, Obtained multiple comprehensive similarities can be then ranked up, then, higher first 2 of comprehensive similarity be selected to refer to vehicle figure As being used as target vehicle image, and so on, etc..
210, the network equipment obtains the corresponding identity information of target vehicle image from reference to vehicle identity information library, for example obtains Take the corresponding license board information of target vehicle image and owner information etc., the identity information as vehicle to be identified.
If for example, the target vehicle image is with reference to vehicle image A, corresponding license board information is " Guangdong B0000 ", car owner For " Zhang San "；Then at this point it is possible to determine that the identity information of the vehicle to be identified may be " Guangdong B0000 ", and car owner is " Zhang San ".
In another example if the target vehicle image is with reference to vehicle image B and with reference to vehicle image C.Wherein, with reference to vehicle figure The corresponding license board informations of picture B are " Guangdong B0001 ", and car owner is " Li Si "；It is " Guangdong with reference to the corresponding license board informations of vehicle image C B0002 ", car owner are " king five "；Then at this point it is possible to which the identity information of the determining vehicle to be identified may be " Guangdong B0001 ", and vehicle Main is " Li Si ", and, it is also possible to for " Guangdong B0002 ", car owner is " king five ", and then again by other means, such as artificial sieve It selects to determine final vehicle identity information.
Optionally, except the identity information that the vehicle to be identified identified is provided, corresponding synthesis can also be provided The numerical value of similarity, alternatively, corresponding global similarity, each number of local similarity and comprehensive similarity can also be provided Value, so that user judges the confidence level of the recognition result accordingly, and then can also be by manually being made into one based on the recognition result Step screening, therefore not to repeat here.
From the foregoing, it will be observed that the present embodiment when needing that vehicle image to be identified is identified, can obtain at least one ginseng Vehicle image is examined, then, on the one hand calculates the vehicle image to be identified and the similarity with reference to vehicle image, is obtained global similar Degree；On the other hand, default marker region is extracted from the vehicle image to be identified and reference vehicle image respectively Image block obtains topography to be identified and refers to topography, and calculates this according to default twin neural network model and wait knowing The similarity of other topography and reference topography, obtains local similarity, subsequently, similar with part to global similarity Degree is weighted, and to obtain comprehensive similarity, and comprehensive similarity is met to the reference vehicle image for presetting second condition Corresponding identity information, the identity information as vehicle to be identified；Due to the program can by calculate global similarity and Local similarity quickly and accurately matches the vehicle with true identity information most like with vehicle to be identified, in turn Identify the identity of the vehicle to be identified, for example car plate is how many, whose, etc. car owner be, can only be passed through accordingly, with respect to existing It, can be to avoid faking, lack or obscure institute because of situations such as license board information for scheme of the license board information to carry out identification The generation of the caused situation that can not identify or identify mistake, can reduce the dependence to vehicle license plate information to be identified, greatly The big validity and accuracy for improving identification.
Embodiment three,
In order to preferably implement above method, the embodiment of the present invention also provides a kind of vehicle identification device, the vehicle Identification device can be specifically integrated in the network equipment, such as the equipment such as terminal or server.
For example, as shown in Figure 3a, the vehicle identification device, including acquiring unit 301, global calculation unit 302, carry Unit 303, local calculation unit 304 and recognition unit 305 are taken, it is as follows：
(1) acquiring unit 301；
Acquiring unit 301 refers to vehicle image for obtaining vehicle image to be identified and at least one.
For example, acquiring unit 301, specifically can be used for receiving the testing vehicle register identification request of user's triggering, wherein the vehicle Vehicle image to be identified is carried in identification request, then, according at least one ginseng of the testing vehicle register identification acquisition request Examine vehicle image.
It wherein, specifically can be by being shot with reference to vehicle, being intercepted from monitoring video or carried out from other picture libraries The approach such as extraction refer to vehicle image to obtain this.
Optionally, in order to reduce subsequent calculation amount, treatment effeciency is improved, it, can be with when obtaining with reference to vehicle image Preliminary screening is carried out with reference to vehicle image to these, with filter out with the apparent inconsistent image of vehicle to be identified, i.e.,：
The acquiring unit 301 specifically can be used for obtaining vehicle image to be identified, and obtain Candidate Set, the Candidate Set Vehicle image is referred to including multiple, the reference vehicle image in Candidate Set is matched with vehicle image to be identified, to matching Degree is filtered less than reference the vehicle image of setting value, Candidate Set after being filtered, from Candidate Set after the filtering acquisition to Few one refers to vehicle image.
Wherein, matching way can be configured according to the demand of practical application, for example, can from vehicle ornament, The information such as interior trim, vehicle frontal, and/or the vehicle back side are compared, and using obtained similarity as matching degree.Wherein, vehicle The information such as ornament and interior trim in can be obtained by detection means, and the front of vehicle and the vehicle back side can pass through detection Vehicle key point obtains, specific detection mode can there are many, therefore not to repeat here.
(2) global calculation unit 302；
Global calculation unit 302, for calculating the vehicle image to be identified and with reference to the similarity of vehicle image, obtaining complete Office's similarity.
Wherein, calculate the vehicle image to be identified and with reference to vehicle image similarity mode can there are many, for example, can To be calculated using common CNN, i.e.,：
Global calculation unit 302 specifically can be used for calculating the vehicle image to be identified using default CNN and refer to vehicle Image similarity obtains global similarity.
Optionally, can also the overall situation similarity be calculated using another twin neural network model.I.e.:
Global calculation unit 302 specifically can be used for calculating the vehicle to be identified using default twin neural network model Image and reference vehicle image similarity, obtain global similarity.
Wherein, which specifically may refer to the embodiment of front, and details are not described herein.
(3) extraction unit 303；
Extraction unit 303, for extracting default mark from the vehicle image to be identified and reference vehicle image respectively The image block of object region obtains topography to be identified and refers to topography；
For example, the extraction unit 303, specifically can be used for obtaining default marker information, marker letter is preset according to this Breath determine marker the vehicle image to be identified first position information, according to the first position information from the vehicle to be identified The image block for presetting marker region is intercepted in image, obtains topography to be identified；Marker information is preset according to this Determine that marker refers to the second position information of vehicle image at this, according to the second position information from this with reference in vehicle image The image block of marker region is preset in interception, obtains referring to topography.
Wherein, which can be depending on the demand of practical application, which, which generally requires, has Distinct personal feature, for example paste the annual test mark on glass for vehicle window, interior pendant and decoration etc., and the first position is believed Breath and second position information then specifically can be with coordinate informations.
(4) local calculation unit 304；
Local calculation unit 304, for calculating the topography to be identified and ginseng according to default twin neural network model The similarity for examining topography, obtains local similarity.
For example, the local calculation unit 304, specifically can be used for by the topography to be identified and with reference to topography into Row combination of two, obtains multiple images pair, merges by the topography to be identified of each image pair and with reference to topography For a multichannel image, the multichannel image is inputted into the upper half branching networks of the twin neural network model and lower half point respectively Branch network, obtains upper half branching networks output vector and lower branch network output vector, calculate upper half branching networks export to Manhatton distance between amount and lower branch network output vector, and dimension is carried out according to the manhatton distance being calculated Full connection operation, obtains local similarity, for details, reference can be made to the embodiments of front, and details are not described herein.
(5) recognition unit 305；
Recognition unit 305 meets the reference vehicle of default first condition for obtaining global similarity and local similarity Identity information corresponding to image, the identity information as vehicle to be identified.
For example, the recognition unit 305 may include operation subelement and determination subelement, it is as follows：
The operation subelement obtains comprehensive for the overall situation similarity and corresponding local similarity to be weighted Close similarity.It can be as follows for example, being formulated：
Sim=(1- μ) simglobal+μsimlocal；
Wherein, sim is comprehensive similarity, simglobalFor global similarity, simlocalFor local similarity, μ is weight, μ In (0,1) range, the specific value of μ can be depending on the demand of practical application, and details are not described herein.
The determination subelement is preset for obtaining comprehensive similarity satisfaction corresponding to the reference vehicle image of second condition Identity information, the identity information as vehicle to be identified.
Wherein, which can be " being higher than predetermined threshold value ", can also be " the highest preceding N of comprehensive similarity It is a ", the value of the predetermined threshold value and N can be depending on the demands of practical application, and N is positive integer, for example, by taking N is 10 as an example, Obtained multiple comprehensive similarities can be then ranked up, then, higher first 10 of comprehensive similarity be selected to refer to vehicle Image is as target vehicle image, etc., and therefore not to repeat here.
Wherein, the identity information with reference to corresponding to vehicle image, can be based on default mapping relations (with reference to vehicle image and With reference to the mapping relations between the true identity information of vehicle) it obtains, i.e., as shown in Figure 3b, the vehicle identification device is also May include setting unit 306, it is as follows：
The setting unit 306 can be used for obtaining each true identity of reference vehicle with reference to corresponding to vehicle image Information, the identity information include license board information and owner information, are established each with reference to vehicle image identity information corresponding with its Mapping relations, and preserve the mapping relations.
Then at this point, the determination subelement, specifically can be used for meeting comprehensive similarity into the reference vehicle for presetting second condition Image obtains the corresponding identity information of target vehicle image as target vehicle image, according to the mapping relations, as waiting knowing The identity information of other vehicle.
Wherein, which can in advance be established by operation maintenance personnel, can also be by the vehicle identification Identification device is voluntarily established, i.e., as shown in Figure 3b, which can also include collecting unit 307, group Unit 308, combining unit 309 and training unit 310 are closed, it is as follows：
The collecting unit 307 can be used for acquiring multiple vehicle sample images, which has true body Part information.
The assembled unit 308 can be used for carrying out combination of two to multiple vehicle sample images, to establish sample pair.
Wherein, sample to refer to combined by two vehicle sample images at set, the sample is to that can be positive sample It is right, can also be negative sample pair, positive sample is to the vehicle sample image that refers to belonging to same vehicle, for example can be by right Two images that the annual test mark of same vehicle is shot, and negative sample is to the vehicle sample graph that refers to belonging to different vehicle Picture, such as two images, etc. that can be shot by the annual test mark to different vehicle.
If the sample, to including positive sample pair and negative sample pair, which specifically can be used for from this multiple Selection belongs to the vehicle sample image of same vehicle in vehicle sample image, by this belong to the vehicle sample image of same vehicle into Row combination of two, to establish positive sample pair；Selection is not belonging to the vehicle sample of same vehicle from multiple vehicle sample images Image, the vehicle sample image that this is not belonging to same vehicle carries out combination of two, to establish negative sample pair.
The combining unit 309 is added to trained sample after can be used for each sample to merging into a multichannel image This concentration.
For example, the combining unit 309, the color for being specifically determined for the vehicle sample image of each sample centering is logical The Color Channel is added by road, obtains each sample to a corresponding multichannel image, the multichannel image that will be obtained It is added to training sample concentration.
The training unit 310 can be used for being trained according to the default initial twin model of training sample set pair, obtain twin Raw neural network model.
For example, the training unit 310 may include trained subelement and convergence subelement, it is as follows：
The training subelement can be used for according to the training sample set respectively to the upper half branch of default initial twin model It is trained in network and lower branch network, obtains the phase that the training sample concentrates the corresponding sample pair of every multichannel image Like degree predicted value.
For example, the training subelement, specifically can be used for concentrating one multichannel image of selection from the training sample, as Current training sample；Current training sample is directed respectively into the upper half branching networks and lower branch net of default initial twin model It is trained in network, obtains upper half branching networks output vector and lower branch network output vector；It is defeated to upper half branching networks Outgoing vector and lower branch network output vector carry out dimension and connect operation entirely, obtain the corresponding sample pair of current training sample Similarity predicted value；It returns to execute from the training sample and concentrates one multichannel image of selection, as current training sample Step, until the multichannel image training that the training sample is concentrated finishes.
The convergence subelement can be used for obtaining the similarity actual value of each sample pair, to the similarity actual value and Similarity predicted value is restrained, and twin neural network model is obtained.
For example, the convergence subelement, specifically can be used for using default loss function to the similarity actual value and similar Degree predicted value is restrained, and twin neural network model is obtained.
Wherein, which specifically can be depending on the demand of practical application, and therefore not to repeat here.
When it is implemented, above each unit can be realized as independent entity, arbitrary combination can also be carried out, is made It is realized for same or several entities, the specific implementation of above each unit can be found in the embodiment of the method for front, herein not It repeats again.
From the foregoing, it will be observed that the vehicle identification device of the present embodiment is when needing that vehicle image to be identified is identified, At least one being obtained by acquiring unit 301 and referring to vehicle image, then, on the one hand being calculated by global calculation unit 302 should The similarity of vehicle image to be identified and reference vehicle image, obtains global similarity；On the other hand, divided by extraction unit 303 Not from the vehicle image to be identified and with reference to the image block for extracting default marker region in vehicle image, obtain waiting knowing Other topography and topography is referred to, and this is calculated according to default twin neural network model by local calculation unit 304 and is waited for It identifies topography and the similarity with reference to topography, obtains local similarity, subsequently, obtained entirely by recognition unit 305 Office's similarity and local similarity meet the identity information corresponding to the reference vehicle image for presetting first condition, as to be identified The identity information of vehicle；Since the program can match and wait to know automatically by calculating global similarity and local similarity The most like vehicle with true identity information of other vehicle, and then identify the identity of the vehicle to be identified, accordingly, with respect to It is existing can only be by scheme of the license board information to carry out identification for, can to avoid faked due to license board information, lack or The generation of the caused situation that can not identify or identify mistake of situations such as obscuring, can reduce to vehicle license plate information to be identified Dependence, greatly improve the validity and accuracy of identification.
Example IV,
The embodiment of the present invention also provides a kind of network equipment, which can be the equipment such as server or terminal, It is integrated with any vehicle identification device that the embodiment of the present invention is provided.As shown in figure 4, it illustrates the present invention to implement The structural schematic diagram of the network equipment involved by example, specifically：
The network equipment may include one or more than one processing core processor 401, one or more The components such as memory 402, power supply 403 and the input unit 404 of computer readable storage medium.Those skilled in the art can manage It solves, network equipment infrastructure does not constitute the restriction to the network equipment shown in Fig. 4, may include more more or fewer than illustrating Component either combines certain components or different components arrangement.Wherein：
Processor 401 is the control centre of the network equipment, utilizes various interfaces and connection whole network equipment Various pieces by running or execute the software program and/or module that are stored in memory 402, and are called and are stored in Data in reservoir 402 execute the various functions and processing data of the network equipment, to carry out integral monitoring to the network equipment. Optionally, processor 401 may include one or more processing cores；Preferably, processor 401 can integrate application processor and tune Demodulation processor processed, wherein the main processing operation system of application processor, user interface and application program etc., modulatedemodulate is mediated Reason device mainly handles wireless communication.It is understood that above-mentioned modem processor can not also be integrated into processor 401 In.
Memory 402 can be used for storing software program and module, and processor 401 is stored in memory 402 by operation Software program and module, to perform various functions application and data processing.Memory 402 can include mainly storage journey Sequence area and storage data field, wherein storing program area can storage program area, the application program (ratio needed at least one function Such as sound-playing function, image player function) etc.；Storage data field can be stored uses created number according to the network equipment According to etc..In addition, memory 402 may include high-speed random access memory, can also include nonvolatile memory, such as extremely A few disk memory, flush memory device or other volatile solid-state parts.Correspondingly, memory 402 can also wrap Memory Controller is included, to provide access of the processor 401 to memory 402.
The network equipment further includes the power supply 403 powered to all parts, it is preferred that power supply 403 can pass through power management System and processor 401 are logically contiguous, to realize management charging, electric discharge and power managed etc. by power-supply management system Function.Power supply 403 can also include one or more direct current or AC power, recharging system, power failure monitor The random components such as circuit, power supply changeover device or inverter, power supply status indicator.
The network equipment may also include input unit 404, which can be used for receiving the number or character of input Information, and generate keyboard related with user setting and function control, mouse, operating lever, optics or trace ball signal Input.
Although being not shown, the network equipment can also be including display unit etc., and details are not described herein.Specifically in the present embodiment In, the processor 401 in the network equipment can correspond to the process of one or more application program according to following instruction Executable file be loaded into memory 402, and the application program being stored in memory 402 is run by processor 401, It is as follows to realize various functions：
It obtains vehicle image to be identified and at least one refers to vehicle image, calculate the vehicle image to be identified and ginseng The similarity for examining vehicle image obtains global similarity, and respectively from the vehicle image to be identified and with reference in vehicle image The image block for extracting default marker region obtains topography to be identified and refers to topography, according to default twin Raw neural network model calculates the topography to be identified and the similarity with reference to topography, obtains local similarity, obtains Global similarity and local similarity meet the identity information corresponding to the reference vehicle image for presetting first condition, as waiting knowing The identity information of other vehicle.
Wherein, which can in advance be established by operation maintenance personnel, can also be by the vehicle identification Identification device is voluntarily established, i.e., processor 401 can also run the application program being stored in memory 402, to real Now following function：
Multiple vehicle sample images are acquired, which has true identity information, such as with normal License board information and true owner information etc.；Combination of two is carried out to multiple vehicle sample images, to establish sample pair； By each sample to merging into a multichannel image after, be added to training sample concentration；It is default just according to training sample set pair The twin model that begins is trained, and obtains twin neural network model.
The specific implementation of above each operation can be found in the embodiment of front, and details are not described herein.
From the foregoing, it will be observed that the network equipment of the present embodiment when needing that vehicle image to be identified is identified, can obtain At least one refers to vehicle image, then, on the one hand calculates the vehicle image to be identified and the similarity with reference to vehicle image, obtains To global similarity；On the other hand, default marker is extracted from the vehicle image to be identified and reference vehicle image respectively The image block of region obtains topography to be identified and refers to topography, and according to default twin neural network model The topography to be identified and the similarity with reference to topography are calculated, local similarity is obtained, subsequently, is obtained global similar Degree and local similarity meet the identity information corresponding to the reference vehicle image for presetting first condition, as vehicle to be identified Identity information；Since the program can be matched and vehicle to be identified automatically by calculating global similarity and local similarity The most like vehicle with true identity information, and then identify the identity of the vehicle to be identified, accordingly, with respect to existing It, can be to avoid being faked, lacked or fuzzy etc. due to license board information for can be by scheme of the license board information to carry out identification The generation of the situation of can not identify caused by situation or identification mistake, can reduce the dependence to vehicle license plate information to be identified Property, greatly improve the validity and accuracy of identification.
Embodiment five,
It will appreciated by the skilled person that all or part of step in the various methods of above-described embodiment can be with It is completed by instructing, or controls relevant hardware by instructing and complete, which can be stored in one and computer-readable deposit In storage media, and is loaded and executed by processor.
For this purpose, the embodiment of the present invention provides a kind of storage medium, wherein being stored with a plurality of instruction, which can be handled Device is loaded, to execute the step in any testing vehicle register identification method that the embodiment of the present invention is provided.For example, this refers to Order can execute following steps：
It obtains vehicle image to be identified and at least one refers to vehicle image, calculate the vehicle image to be identified and ginseng The similarity for examining vehicle image obtains global similarity, and respectively from the vehicle image to be identified and with reference in vehicle image The image block for extracting default marker region obtains topography to be identified and refers to topography, according to default twin Raw neural network model calculates the topography to be identified and the similarity with reference to topography, obtains local similarity, obtains Global similarity and local similarity meet the identity information corresponding to the reference vehicle image for presetting first condition, as waiting knowing The identity information of other vehicle.
Wherein, which can in advance be established by operation maintenance personnel, can also be by the vehicle identification Identification device is voluntarily established, and specific method for building up can be found in the embodiment of front.
The specific implementation of above each operation can be found in the embodiment of front, and details are not described herein.
Wherein, which may include：Read-only memory (ROM, Read Only Memory), random access memory Body (RAM, Random Access Memory), disk or CD etc..
By the instruction stored in the storage medium, any vehicle body that the embodiment of the present invention is provided can be executed Step in part recognition methods, it is thereby achieved that any testing vehicle register identification method institute that the embodiment of the present invention is provided The advantageous effect that can be realized, refers to the embodiment of front, details are not described herein.
It is provided for the embodiments of the invention a kind of testing vehicle register identification method, apparatus above and storage medium has carried out in detail Thin to introduce, principle and implementation of the present invention are described for specific case used herein, and above example is said The bright method and its core concept for being merely used to help understand the present invention；Meanwhile for those skilled in the art, according to this hair Bright thought, there will be changes in the specific implementation manner and application range, in conclusion the content of the present specification should not manage Solution is limitation of the present invention.

Claims (15)

1. a kind of testing vehicle register identification method, which is characterized in that including：
It obtains vehicle image to be identified and at least one refers to vehicle image；
The vehicle image to be identified and the similarity with reference to vehicle image are calculated, global similarity is obtained；
The image block of default marker region is extracted from the vehicle image to be identified and reference vehicle image respectively, It obtains topography to be identified and refers to topography；
The topography to be identified is calculated according to default twin neural network model and with reference to the similarity of topography, is obtained Local similarity；
The identity information corresponding to the reference vehicle image of global similarity and the default first condition of local similarity satisfaction is obtained, Identity information as vehicle to be identified.
2. according to the method described in claim 1, it is characterized in that, acquisition overall situation similarity and local similarity meet in advance If the identity information corresponding to the reference vehicle image of condition, as the identity information of vehicle to be identified, including：
The global similarity and corresponding local similarity are weighted, comprehensive similarity is obtained；
The identity information corresponding to the reference vehicle image of the default second condition of comprehensive similarity satisfaction is obtained, as vehicle to be identified Identity information.
3. according to the method described in claim 2, it is characterized in that, the comprehensive similarity that obtains meets default second condition Identity information with reference to corresponding to vehicle image before the identity information as vehicle to be identified, further includes：
Each true identity information of reference vehicle with reference to corresponding to vehicle image is obtained, the identity information includes car plate letter Breath and owner information；
Each mapping relations with reference to vehicle image identity information corresponding with its are established, and preserve the mapping relations；
The comprehensive similarity that obtains meets identity information corresponding to the reference vehicle image for presetting second condition, as waiting knowing The identity information of other vehicle, specially：Comprehensive similarity is met into the reference vehicle image for presetting second condition as target carriage Image obtains the corresponding identity information of target vehicle image according to the mapping relations, and the identity as vehicle to be identified is believed Breath.
4. according to the method described in claim 1, it is characterized in that, described respectively from the vehicle image to be identified and with reference to vehicle The image block that default marker region is extracted in image obtains topography to be identified and refers to topography, packet It includes：
It obtains and presets marker information；
According to the default marker information determine marker the vehicle image to be identified first position information, according to institute The image block that first position information intercepts default marker region from the vehicle image to be identified is stated, is obtained to be identified Topography；
Marker is determined in the second position information with reference to vehicle image, according to described according to the default marker information Second position information, with reference to the image block for presetting marker region is intercepted in vehicle image, obtains referring to Local map from described Picture.
5. method according to any one of claims 1 to 4, which is characterized in that obtain at least one and refer to vehicle image, packet It includes：
Candidate Set is obtained, the Candidate Set includes that multiple refer to vehicle image；
Reference vehicle image in Candidate Set is matched with vehicle image to be identified；
The reference vehicle image for being less than setting value to matching degree is filtered, Candidate Set after being filtered；
At least one is obtained from Candidate Set after the filtering refers to vehicle image.
6. method according to any one of claims 1 to 4, which is characterized in that the basis presets twin neural network mould Type calculates the topography to be identified and the similarity with reference to topography, before obtaining local similarity, further includes：
Multiple vehicle sample images are acquired, the vehicle sample image has true identity information；
Combination of two is carried out to multiple described vehicle sample images, to establish sample pair；
By each sample to merging into a multichannel image after, be added to training sample concentration；
It is trained according to the default initial twin model of training sample set pair, obtains twin neural network model.
7. according to the method described in claim 6, it is characterized in that, the sample is to including positive sample pair and negative sample pair, institute It states and combination of two is carried out to multiple described vehicle sample images, to establish sample pair, including：
Selection belongs to the vehicle sample image of same vehicle from multiple described vehicle sample images, belongs to same vehicle by described Vehicle sample image carry out combination of two, to establish positive sample pair；
Selection is not belonging to the vehicle sample image of same vehicle from multiple described vehicle sample images, by it is described be not belonging to it is same The vehicle sample image of vehicle carries out combination of two, to establish negative sample pair.
8. according to the method described in claim 6, it is characterized in that, it is described by each sample to merging into a multichannel image Afterwards, it is added to training sample concentration, including：
Determine the Color Channel of the vehicle sample image of each sample centering；
The Color Channel is added, obtains each sample to a corresponding multichannel image；
Obtained multichannel image is added to training sample to concentrate.
9. according to the method described in claim 6, it is characterized in that, described according to the default initial twin model of training sample set pair It is trained, obtains twin neural network model, including：
According in the training sample set respectively the upper half branching networks to default initial twin model and lower branch network into Row training, obtains the similarity predicted value that the training sample concentrates the corresponding sample pair of every multichannel image；
The similarity actual value for obtaining each sample pair is restrained the similarity actual value and similarity predicted value, is obtained To twin neural network model.
10. according to the method described in claim 9, it is characterized in that, described first to presetting respectively according to the training sample set Begin twin model upper half branching networks and lower branch network in be trained, obtain the training sample concentrate every it is mostly logical The similarity predicted value of the corresponding sample pair of road image, including：
One multichannel image of selection is concentrated from the training sample, as current training sample；
Current training sample is directed respectively into default initially the upper half branching networks of twin model and lower branch network and is carried out Training, obtains upper half branching networks output vector and lower branch network output vector；
Dimension is carried out to upper half branching networks output vector and lower branch network output vector and connects operation entirely, is obtained current The similarity predicted value of the corresponding sample pair of training sample；
It returns to execute from the training sample and concentrates one multichannel image of selection, the step of as current training sample, until The multichannel image that the training sample is concentrated training finishes.
11. a kind of vehicle identification device, which is characterized in that including：
Acquiring unit refers to vehicle image for obtaining vehicle image to be identified and at least one；
Global calculation unit, for calculating the vehicle image to be identified and with reference to the similarity of vehicle image, obtaining global phase Like degree；
Extraction unit, for respectively from the vehicle image to be identified and with reference to where extracting default marker in vehicle image The image block in region obtains topography to be identified and refers to topography；
Local calculation unit, for calculating the topography to be identified according to default twin neural network model and with reference to part The similarity of image, obtains local similarity；
Recognition unit, the reference vehicle image institute for obtaining global similarity and the default first condition of local similarity satisfaction are right The identity information answered, the identity information as vehicle to be identified.
12. according to the devices described in claim 11, which is characterized in that the recognition unit includes operation subelement and determines sub Unit；
The operation subelement obtains comprehensive for the global similarity and corresponding local similarity to be weighted Close similarity；
The determination subelement meets the body corresponding to the reference vehicle image of default second condition for obtaining comprehensive similarity Part information, the identity information as vehicle to be identified.
13. device according to claim 12, which is characterized in that further include setting unit；
The setting unit, it is described for obtaining each true identity information of reference vehicle with reference to corresponding to vehicle image Identity information includes license board information and owner information, establishes each mapping with reference to vehicle image identity information corresponding with its and closes System, and preserve the mapping relations；
The determination subelement is specifically used for comprehensive similarity meeting the reference vehicle image for presetting second condition as target Vehicle image obtains the corresponding identity information of target vehicle image, the identity as vehicle to be identified according to the mapping relations Information.
14. according to claim 11 to 13 any one of them device, which is characterized in that further include collecting unit, assembled unit, Combining unit and training unit；
The collecting unit, for acquiring multiple vehicle sample images, the vehicle sample image has true identity information；
The assembled unit, for carrying out combination of two to multiple described vehicle sample images, to establish sample pair；
The combining unit is added to training sample concentration after by each sample to merging into a multichannel image；
The training unit obtains twin nerve net for being trained according to the default initial twin model of training sample set pair Network model.
15. a kind of storage medium, which is characterized in that the storage medium is stored with a plurality of instruction, and described instruction is suitable for processor It is loaded, the step in 1 to 10 any one of them testing vehicle register identification method is required with perform claim.
CN201810444371.7A 2018-05-10 2018-05-10 Vehicle identity recognition method and device and storage medium Active CN108596277B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810444371.7A CN108596277B (en) 2018-05-10 2018-05-10 Vehicle identity recognition method and device and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810444371.7A CN108596277B (en) 2018-05-10 2018-05-10 Vehicle identity recognition method and device and storage medium

Publications (2)

Publication Number Publication Date
CN108596277A true CN108596277A (en) 2018-09-28
CN108596277B CN108596277B (en) 2020-07-07

Family

ID=63637152

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810444371.7A Active CN108596277B (en) 2018-05-10 2018-05-10 Vehicle identity recognition method and device and storage medium

Country Status (1)

Country Link
CN (1) CN108596277B (en)

Cited By (30)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109509355A (en) * 2019-01-07 2019-03-22 黄慧 Wisdom traffic information taken terminal
CN109635825A (en) * 2018-12-19 2019-04-16 苏州市科远软件技术开发有限公司 Vehicle attribute detection method, device and storage medium
CN109815953A (en) * 2019-01-30 2019-05-28 电子科技大学 One kind being based on vehicle annual test target vehicle identification matching system
CN110070092A (en) * 2019-03-15 2019-07-30 平安科技（深圳）有限公司 Model recognizing method, device, computer equipment and storage medium
CN110135517A (en) * 2019-05-24 2019-08-16 北京百度网讯科技有限公司 For obtaining the method and device of vehicle similarity
CN110348393A (en) * 2019-07-12 2019-10-18 上海眼控科技股份有限公司 Vehicle characteristics extract model training method, vehicle identification method and equipment
CN110390291A (en) * 2019-07-18 2019-10-29 北京字节跳动网络技术有限公司 Data processing method, device and electronic equipment
CN110598019A (en) * 2019-09-11 2019-12-20 腾讯科技（深圳）有限公司 Repeated image identification method and device
CN110610191A (en) * 2019-08-05 2019-12-24 深圳优地科技有限公司 Elevator floor identification method and device and terminal equipment
CN110826484A (en) * 2019-11-05 2020-02-21 上海眼控科技股份有限公司 Vehicle weight recognition method and device, computer equipment and model training method
CN111242986A (en) * 2020-01-07 2020-06-05 北京百度网讯科技有限公司 Cross-camera obstacle tracking method, device, equipment, system and medium
CN111401326A (en) * 2020-04-21 2020-07-10 招商局金融科技有限公司 Target identity recognition method based on picture recognition, server and storage medium
CN111476101A (en) * 2020-03-11 2020-07-31 咪咕文化科技有限公司 Video shot switching detection method and device and computer readable storage medium
CN111611414A (en) * 2019-02-22 2020-09-01 杭州海康威视数字技术股份有限公司 Vehicle retrieval method, device and storage medium
CN111639970A (en) * 2020-05-28 2020-09-08 深圳壹账通智能科技有限公司 Method for determining price of article based on image recognition and related equipment
CN111652087A (en) * 2020-05-15 2020-09-11 泰康保险集团股份有限公司 Car checking method and device, electronic equipment and storage medium
CN111696217A (en) * 2020-05-25 2020-09-22 上海金亥通信设备有限公司 Park parking management system
CN111914921A (en) * 2020-07-24 2020-11-10 山东工商学院 Similarity image retrieval method and system based on multi-feature fusion
CN112633281A (en) * 2020-12-30 2021-04-09 广东赛诺科技股份有限公司 Vehicle identity authentication method and system based on Hash algorithm
CN112749326A (en) * 2019-11-15 2021-05-04 腾讯科技（深圳）有限公司 Information processing method, information processing device, computer equipment and storage medium
CN113159039A (en) * 2021-02-09 2021-07-23 北京市商汤科技开发有限公司 Image recognition method and device, electronic equipment and storage medium
CN113361300A (en) * 2020-03-04 2021-09-07 阿里巴巴集团控股有限公司 Identification information identification method, device, equipment and storage medium
CN113569911A (en) * 2021-06-28 2021-10-29 北京百度网讯科技有限公司 Vehicle identification method and device, electronic equipment and storage medium
CN113569912A (en) * 2021-06-28 2021-10-29 北京百度网讯科技有限公司 Vehicle identification method and device, electronic equipment and storage medium
CN113655348A (en) * 2021-07-28 2021-11-16 国网湖南省电力有限公司 Power equipment partial discharge fault diagnosis method based on deep twin network, system terminal and readable storage medium
CN113821674A (en) * 2021-11-23 2021-12-21 北京中超伟业信息安全技术股份有限公司 Intelligent cargo supervision method and system based on twin neural network
WO2022015236A1 (en) * 2020-07-17 2022-01-20 Hitachi, Ltd. Method of image processing for object identification and system thereof
CN114429649A (en) * 2022-04-07 2022-05-03 青岛美迪康数字工程有限公司 Target image identification method and device
CN114445980A (en) * 2022-01-28 2022-05-06 成都商汤科技有限公司 Commodity checkout method and device, electronic device and storage medium
CN115497633A (en) * 2022-10-19 2022-12-20 联仁健康医疗大数据科技股份有限公司 Data processing method, device, equipment and storage medium

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050135679A1 (en) * 2003-12-17 2005-06-23 Yoon Ho S. Face detecting system and method using symmetric axis
JP2010176706A (en) * 2003-12-25 2010-08-12 Fujitsu Ltd Vehicle retrieval system and vehicle retrieval method
CN105069089A (en) * 2015-08-04 2015-11-18 小米科技有限责任公司 Picture detection method and device
CN106250870A (en) * 2016-08-16 2016-12-21 电子科技大学 A kind of pedestrian's recognition methods again combining local and overall situation similarity measurement study
CN106469299A (en) * 2016-08-31 2017-03-01 北京邮电大学 A kind of vehicle search method and device
CN107562805A (en) * 2017-08-08 2018-01-09 浙江大华技术股份有限公司 It is a kind of to scheme to search the method and device of figure
CN108388888A (en) * 2018-03-23 2018-08-10 腾讯科技（深圳）有限公司 A kind of vehicle identification method, device and storage medium

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050135679A1 (en) * 2003-12-17 2005-06-23 Yoon Ho S. Face detecting system and method using symmetric axis
JP2010176706A (en) * 2003-12-25 2010-08-12 Fujitsu Ltd Vehicle retrieval system and vehicle retrieval method
CN105069089A (en) * 2015-08-04 2015-11-18 小米科技有限责任公司 Picture detection method and device
CN106250870A (en) * 2016-08-16 2016-12-21 电子科技大学 A kind of pedestrian's recognition methods again combining local and overall situation similarity measurement study
CN106469299A (en) * 2016-08-31 2017-03-01 北京邮电大学 A kind of vehicle search method and device
CN107562805A (en) * 2017-08-08 2018-01-09 浙江大华技术股份有限公司 It is a kind of to scheme to search the method and device of figure
CN108388888A (en) * 2018-03-23 2018-08-10 腾讯科技（深圳）有限公司 A kind of vehicle identification method, device and storage medium

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
PEI LI ET AL.: "Learning Face Similarity for Re-identification from Real Surveillance Video: A Deep Metric Solution", 《2017 IEEE INTERNATIONAL JOINT CONFERENCE ON BIOMETRICS (IJCB)》 *

Cited By (41)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109635825A (en) * 2018-12-19 2019-04-16 苏州市科远软件技术开发有限公司 Vehicle attribute detection method, device and storage medium
CN109509355A (en) * 2019-01-07 2019-03-22 黄慧 Wisdom traffic information taken terminal
CN109815953A (en) * 2019-01-30 2019-05-28 电子科技大学 One kind being based on vehicle annual test target vehicle identification matching system
CN111611414B (en) * 2019-02-22 2023-10-24 杭州海康威视数字技术股份有限公司 Vehicle searching method, device and storage medium
CN111611414A (en) * 2019-02-22 2020-09-01 杭州海康威视数字技术股份有限公司 Vehicle retrieval method, device and storage medium
CN110070092A (en) * 2019-03-15 2019-07-30 平安科技（深圳）有限公司 Model recognizing method, device, computer equipment and storage medium
CN110135517B (en) * 2019-05-24 2023-04-07 北京百度网讯科技有限公司 Method and device for obtaining vehicle similarity
CN110135517A (en) * 2019-05-24 2019-08-16 北京百度网讯科技有限公司 For obtaining the method and device of vehicle similarity
CN110348393A (en) * 2019-07-12 2019-10-18 上海眼控科技股份有限公司 Vehicle characteristics extract model training method, vehicle identification method and equipment
CN110390291A (en) * 2019-07-18 2019-10-29 北京字节跳动网络技术有限公司 Data processing method, device and electronic equipment
CN110390291B (en) * 2019-07-18 2021-10-08 北京字节跳动网络技术有限公司 Data processing method and device and electronic equipment
CN110610191A (en) * 2019-08-05 2019-12-24 深圳优地科技有限公司 Elevator floor identification method and device and terminal equipment
CN110598019A (en) * 2019-09-11 2019-12-20 腾讯科技（深圳）有限公司 Repeated image identification method and device
CN110826484A (en) * 2019-11-05 2020-02-21 上海眼控科技股份有限公司 Vehicle weight recognition method and device, computer equipment and model training method
CN112749326B (en) * 2019-11-15 2023-10-03 腾讯科技（深圳）有限公司 Information processing method, information processing device, computer equipment and storage medium
CN112749326A (en) * 2019-11-15 2021-05-04 腾讯科技（深圳）有限公司 Information processing method, information processing device, computer equipment and storage medium
CN111242986A (en) * 2020-01-07 2020-06-05 北京百度网讯科技有限公司 Cross-camera obstacle tracking method, device, equipment, system and medium
CN111242986B (en) * 2020-01-07 2023-11-24 阿波罗智能技术(北京)有限公司 Cross-camera obstacle tracking method, device, equipment, system and medium
CN113361300A (en) * 2020-03-04 2021-09-07 阿里巴巴集团控股有限公司 Identification information identification method, device, equipment and storage medium
CN111476101A (en) * 2020-03-11 2020-07-31 咪咕文化科技有限公司 Video shot switching detection method and device and computer readable storage medium
CN111401326B (en) * 2020-04-21 2023-04-18 招商局金融科技有限公司 Target identity recognition method based on picture recognition, server and storage medium
CN111401326A (en) * 2020-04-21 2020-07-10 招商局金融科技有限公司 Target identity recognition method based on picture recognition, server and storage medium
CN111652087A (en) * 2020-05-15 2020-09-11 泰康保险集团股份有限公司 Car checking method and device, electronic equipment and storage medium
CN111696217A (en) * 2020-05-25 2020-09-22 上海金亥通信设备有限公司 Park parking management system
CN111696217B (en) * 2020-05-25 2022-01-07 上海金亥通信设备有限公司 Park parking management system
CN111639970A (en) * 2020-05-28 2020-09-08 深圳壹账通智能科技有限公司 Method for determining price of article based on image recognition and related equipment
WO2022015236A1 (en) * 2020-07-17 2022-01-20 Hitachi, Ltd. Method of image processing for object identification and system thereof
CN111914921A (en) * 2020-07-24 2020-11-10 山东工商学院 Similarity image retrieval method and system based on multi-feature fusion
CN112633281A (en) * 2020-12-30 2021-04-09 广东赛诺科技股份有限公司 Vehicle identity authentication method and system based on Hash algorithm
CN112633281B (en) * 2020-12-30 2022-11-22 广东赛诺科技股份有限公司 Vehicle identity authentication method and system based on Hash algorithm
CN113159039A (en) * 2021-02-09 2021-07-23 北京市商汤科技开发有限公司 Image recognition method and device, electronic equipment and storage medium
CN113569912A (en) * 2021-06-28 2021-10-29 北京百度网讯科技有限公司 Vehicle identification method and device, electronic equipment and storage medium
CN113569911A (en) * 2021-06-28 2021-10-29 北京百度网讯科技有限公司 Vehicle identification method and device, electronic equipment and storage medium
CN113655348A (en) * 2021-07-28 2021-11-16 国网湖南省电力有限公司 Power equipment partial discharge fault diagnosis method based on deep twin network, system terminal and readable storage medium
CN113655348B (en) * 2021-07-28 2023-12-08 国网湖南省电力有限公司 Power equipment partial discharge fault diagnosis method, system terminal and readable storage medium based on deep twin network
CN113821674A (en) * 2021-11-23 2021-12-21 北京中超伟业信息安全技术股份有限公司 Intelligent cargo supervision method and system based on twin neural network
CN114445980A (en) * 2022-01-28 2022-05-06 成都商汤科技有限公司 Commodity checkout method and device, electronic device and storage medium
CN114445980B (en) * 2022-01-28 2023-11-28 成都商汤科技有限公司 Commodity checkout method and device, electronic equipment and storage medium
CN114429649A (en) * 2022-04-07 2022-05-03 青岛美迪康数字工程有限公司 Target image identification method and device
CN115497633A (en) * 2022-10-19 2022-12-20 联仁健康医疗大数据科技股份有限公司 Data processing method, device, equipment and storage medium
CN115497633B (en) * 2022-10-19 2024-01-30 联仁健康医疗大数据科技股份有限公司 Data processing method, device, equipment and storage medium

Also Published As

Publication number Publication date
CN108596277B (en) 2020-07-07

Similar Documents

Publication Publication Date Title
CN108596277A (en) A kind of testing vehicle register identification method, apparatus and storage medium
CN108388888A (en) A kind of vehicle identification method, device and storage medium
CN106650806B (en) A kind of cooperating type depth net model methodology for pedestrian detection
CN108319907A (en) A kind of vehicle identification method, device and storage medium
CN104166841B (en) The quick detection recognition methods of pedestrian or vehicle is specified in a kind of video surveillance network
CN108875600A (en) A kind of information of vehicles detection and tracking method, apparatus and computer storage medium based on YOLO
CN108446612A (en) vehicle identification method, device and storage medium
CN109376603A (en) A kind of video frequency identifying method, device, computer equipment and storage medium
CN107273836A (en) A kind of pedestrian detection recognition methods, device, model and medium
CN108269250A (en) Method and apparatus based on convolutional neural networks assessment quality of human face image
CN110110617A (en) Medical image dividing method, device, electronic equipment and storage medium
CN109447169A (en) The training method of image processing method and its model, device and electronic system
CN110263681A (en) The recognition methods of facial expression and device, storage medium, electronic device
CN109978918A (en) A kind of trajectory track method, apparatus and storage medium
CN111898523A (en) Remote sensing image special vehicle target detection method based on transfer learning
CN107085696A (en) A kind of vehicle location and type identifier method based on bayonet socket image
CN108986075A (en) A kind of judgment method and device of preferred image
CN110464366A (en) A kind of Emotion identification method, system and storage medium
CN110490238A (en) A kind of image processing method, device and storage medium
CN106022285A (en) Vehicle type identification method and vehicle type identification device based on convolutional neural network
CN108491827A (en) A kind of vehicle checking method, device and storage medium
TW200842733A (en) Object image detection method
CN104504408A (en) Human face identification comparing method and system for realizing the method
CN110119726A (en) A kind of vehicle brand multi-angle recognition methods based on YOLOv3 model
CN109684922A (en) A kind of recognition methods based on the multi-model of convolutional neural networks to finished product dish

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