CN110348393A - Vehicle characteristics extract model training method, vehicle identification method and equipment - Google Patents
Vehicle characteristics extract model training method, vehicle identification method and equipment Download PDFInfo
- Publication number
- CN110348393A CN110348393A CN201910632120.6A CN201910632120A CN110348393A CN 110348393 A CN110348393 A CN 110348393A CN 201910632120 A CN201910632120 A CN 201910632120A CN 110348393 A CN110348393 A CN 110348393A
- Authority
- CN
- China
- Prior art keywords
- sample image
- vehicle
- image
- negative sample
- positive 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
Links
Classifications
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of vehicle characteristics and extracts model training method, vehicle identification method and equipment, by obtaining the multiple image in vehicle monitoring video flowing, two vehicle images of same vehicle are intercepted from two field pictures and are spliced into positive sample image, and two vehicle images of different vehicle are intercepted from two field pictures and are spliced into negative sample image；A part is chosen from each positive sample image and negative sample image is used as test data；Design include two branches twin network model, by test data positive sample image or negative sample image split two vehicle images respectively；Based on twin network model, obtain the feature of two vehicle images of the positive sample image or negative sample image in test data, judge that the twin network model is that training is completed also to be to continue with training based on obtained feature, the accuracy rate of the feature of the vehicle image obtained by twin network model can be improved, and then improve the accuracy rate of the feature identification vehicle based on vehicle image.
Description
Technical field
The present invention relates to computer fields more particularly to a kind of vehicle characteristics to extract model training method, vehicle identification side
Method and equipment.
Background technique
The vehicle identification method of vehicle at present, in use, accuracy rate is lower before engineering.At the same time, before engineering
To when finding certain same or different vehicle match mistake, it is difficult to solve a problem promptly.
Summary of the invention
Model training method, vehicle identification method are extracted it is an object of the present invention to provide a kind of vehicle characteristics and are set
It is standby.
According to an aspect of the invention, there is provided a kind of vehicle characteristics extract model training method, this method comprises:
The multiple image in vehicle monitoring video flowing is obtained, two vehicle images of same vehicle are intercepted from two field pictures
And it is spliced into positive sample image, two vehicle images of different vehicle are intercepted from two field pictures and is spliced into negative sample image；
A part is chosen from each positive sample image and negative sample image is used as test data；
Design includes the twin network model of two branches, by the positive sample image or negative sample figure in the test data
As splitting two vehicle images respectively；
Based on the twin network model, two of the positive sample image or negative sample image in the test data are obtained
The feature of vehicle image judges that the twin network model is that training is completed also to be to continue with training based on obtained feature.
Further, in the above method, it is based on the twin network model, obtains the positive sample figure in the test data
The feature of two vehicle images of picture or negative sample image, judges that the twin network model is trained complete based on obtained feature
At being also to continue with training, comprising:
Step S41, by from the test data positive sample image or negative sample image split obtain two vehicles respectively
The current twin network model of image input carries out feature extraction, obtains the positive sample image or negative sample in the test data
The feature of two vehicle images of this image；
Step S42 calculates the spy of two vehicle images of the positive sample image or negative sample image in the test data
The similarity of sign judges two of the positive sample image or negative sample image in the test data based on the similarity being calculated
Vehicle image is opened, judges that two vehicle images are to belong to positive sample image to still fall within negative sample image, obtains judging result,
The actual result of positive sample image or negative sample image that the judging result and two vehicle images belong to is compared,
If comparing unanimously, step S43 will not be chosen for test data in each positive sample image and negative sample image
Remainder using the training data as input, and continues to train as training data by way of being fitted loss function
After twin network model, step S41 is returned to；
If comparing unanimously, step S44, twin network model training terminates.
Further, in the above method, two of the positive sample image or negative sample image in the test data are calculated
The similarity of the feature of vehicle image, comprising:
Pass through norm layers of positive sample image or negative sample figure calculated in the test data of normalization in caffe frame
The denominator part of the cosine similarity of the feature of two vehicle images of picture；
The element_wise layers of positive sample calculated in the test data are calculated by the element in the caffe frame
The molecular moiety of the cosine similarity of the feature of two vehicle images of image or negative sample image；
It will be calculated by first link InnerProduct layers in the caffe frame and in conjunction with concat layers remaining
The denominator part of string similarity and molecular moiety are converted to neural network, and are exported the neural network with softmax.
Further, in the above method, using the training data as input, and by way of being fitted loss function after
The continuous twin network model of training, comprising:
By the result for using softmax to export the neural network and the training data as input, and pass through
The mode of fitting loss function continues to train twin network model.
Further, in the above method, the multiple image in vehicle monitoring video flowing is obtained, is cut from two field pictures every time
It takes two vehicle images of same vehicle and is spliced into positive sample image, intercept two of different vehicle from two field pictures every time
Vehicle image is simultaneously spliced into negative sample image, comprising:
All vehicles of different frame image in vehicle monitoring video flowing are found by the detection algorithm of deep learning；
Based on the vehicle found, same color identifier is arranged to the same vehicle in different frame image；
Based on the color identifier, two vehicle images of same color identifier are intercepted from two field pictures, and are carried out
Splicing is used as positive sample image, two vehicle images of color identifier not of the same race is intercepted from two field pictures, and carry out splicing work
Be negative sample image.
Further, in the above method, by the test data positive sample image or negative sample image split respectively
Two vehicle images, comprising:
The size of positive sample image or negative sample image in the test data is uniformly adjusted to pre-set dimension；
By the slice layer of caffe frame, by the unified test data adjusted after size positive sample image or
Negative sample image cutting is two vehicle images.
Further, in the above method, by from the test data positive sample image or negative sample image tear open respectively
Get two current twin network models of vehicle image input and carry out feature extraction, comprising:
Every two vehicle images segmented are inputted to two GoogLenet of the twin network model respectively
Inception-V2 perhaps two GoogLenet Inception-V2 of carry out feature extraction of two ResNet50 branches or
Two ResNet50 branches share it is all can learning parameter, wherein first GoogLenet Inception-V2 or
ResNet50 branch can learning parameter be iterated, Article 2 GoogLenet Inception-V2 or ResNet50 branch
Can learning parameter directly learn from after the iteration of first GoogLenet Inception-V2 or ResNet50 branch
Parameter copy.
According to another aspect of the present invention, a kind of vehicle identification method is also provided, this method comprises:
The image of two cars to be matched is inputted in the twin network model that training described in any of the above embodiments is completed；
The feature of the image of the two cars of branch output therein is intercepted from the twin network model；
Calculate the similarity of the feature of the image for the two cars being truncated to, the spy of the image based on the two cars being calculated
The similarity of sign and preset similarity threshold judge that two cars are same vehicle or different vehicles.
According to another aspect of the present invention, a kind of equipment for handling in network equipment client information, the equipment are also provided
Including the memory for storing computer program instructions and the processor for executing program instructions, wherein when the computer
When program instruction is executed by the processor, the vehicle characteristics extraction model training method that the equipment executes any of the above-described is triggered.
According to another aspect of the present invention, a kind of equipment for handling in network equipment client information, the equipment are also provided
Including the memory for storing computer program instructions and the processor for executing program instructions, wherein when the computer
When program instruction is executed by the processor, triggers the equipment and execute above-mentioned vehicle identification method.
Compared with prior art, the present invention is by obtaining the multiple image in vehicle monitoring video flowing, from two field pictures
It intercepts two vehicle images of same vehicle and is spliced into positive sample image, two vehicles of different vehicle are intercepted from two field pictures
Image is simultaneously spliced into negative sample image；A part is chosen from each positive sample image and negative sample image as test number
According to；Design includes the twin network model of two branches, by the positive sample image or negative sample image point in the test data
It Chai Fen not two vehicle images；Based on the twin network model, the positive sample image or negative sample in the test data are obtained
The feature of two vehicle images of this image, based on obtained feature judge the twin network model be training complete or after
Continuous training can be improved the accuracy rate of the feature of the vehicle image obtained by twin network model, and then improve and be based on vehicle
The accuracy rate of the feature identification vehicle of image.
Detailed description of the invention
By reading a detailed description of non-restrictive embodiments in the light of the attached drawings below, of the invention other
Feature, objects and advantages will become more apparent upon:
The vehicle characteristics that Fig. 1 shows one embodiment of the invention extract the flow chart of model training method；
Fig. 2 shows the flow charts that the vehicle characteristics of another embodiment of the present invention extract model training method；
Fig. 3 shows the schematic diagram of the positive sample image of one embodiment of the invention；
Fig. 4 shows the schematic diagram of the negative sample image of one embodiment of the invention；
Fig. 5 shows the schematic diagram of one embodiment of the invention；
Fig. 6 shows the application scenarios schematic diagram of one embodiment of the invention；
Fig. 7 shows the flow chart of the vehicle identification method of one embodiment of the invention.
The same or similar appended drawing reference represents the same or similar component in attached drawing.
Specific embodiment
Present invention is further described in detail with reference to the accompanying drawing.
In a typical configuration of this application, terminal, the equipment of service network and trusted party include one or more
Processor (CPU), input/output interface, network interface and memory.
Memory may include the non-volatile memory in computer-readable medium, random access memory (RAM) and/or
The forms such as Nonvolatile memory, such as read-only memory (ROM) or flash memory (flash RAM).Memory is computer-readable medium
Example.
Computer-readable medium includes permanent and non-permanent, removable and non-removable media can be by any method
Or technology come realize information store.Information can be computer readable instructions, data structure, the module of program or other data.
The example of the storage medium of computer includes, but are not limited to phase change memory (PRAM), static random access memory (SRAM), moves
State random access memory (DRAM), other kinds of random access memory (RAM), read-only memory (ROM), electric erasable
Programmable read only memory (EEPROM), flash memory or other memory techniques, read-only disc read only memory (CD-ROM) (CD-ROM),
Digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or
Any other non-transmission medium, can be used for storage can be accessed by a computing device information.As defined in this article, computer
Readable medium does not include non-temporary computer readable media (transitory media), such as the data-signal and carrier wave of modulation.
As shown in Figure 1, the application provides a kind of vehicle characteristics extraction model training method, which comprises
Step S1 obtains the multiple image in vehicle monitoring video flowing, intercepts same vehicle from two field pictures every time
Two vehicle images are simultaneously spliced into positive sample image, intercept two vehicle images of different vehicle from two field pictures every time and spell
It is connected in negative sample image；
Here, available front-end collection equipment collects a few frame images in vehicle monitoring video flowing, number can use
Identical or different vehicle is collected according to sampling instrument, and is spliced respectively as positive sample image or negative sample image；
As shown in figure 3, two images of same vehicle can be intercepted every time from two field pictures and be spliced into positive sample figure
Picture, as shown in figure 4, and intercepting two images of different vehicle every time from two field pictures and being spliced into negative sample image；
Step S2 chooses a part from each positive sample image and negative sample image and is used as test data；
Here, such as positive sample image and negative sample image share 1000,600 therein can be chosen as training
Data are used as test data for remaining 400；
Step S3, design include two branches twin network model, by the test data positive sample image or
Negative sample image splits two vehicle images respectively；
Step S4 is based on the twin network model, obtains the positive sample image or negative sample figure in the test data
The feature of two vehicle images of picture judges that the twin network model is that training is completed also to be to continue with instruction based on obtained feature
Practice.
Here, the application is based on the twin network model, the positive sample image or negative sample in the test data are obtained
The feature of two vehicle images of this image judges whether to continue to train the twin network model based on obtained feature, can
To improve the accuracy rate of the feature of the vehicle image obtained by twin network model, and then improve the feature based on vehicle image
Identify the accuracy rate of vehicle.
As shown in Fig. 2, the vehicle characteristics of the application extract in the implementation of model training method one, step S4, based on described twin
Raw network model, obtains the feature of two vehicle images of the positive sample image or negative sample image in the test data, base
Judge that the twin network model is that training is completed also to be to continue with training in obtained feature:
Step S41, by from the test data positive sample image or negative sample image split obtain two vehicles respectively
The current twin network model of image input carries out feature extraction, obtains the positive sample image or negative sample in the test data
The feature of two vehicle images of this image；
Step S42 calculates the spy of two vehicle images of the positive sample image or negative sample image in the test data
The similarity of sign judges two of the positive sample image or negative sample image in the test data based on the similarity being calculated
Vehicle image is opened, judges that two vehicle images are to belong to positive sample image to still fall within negative sample image, obtains judging result,
The actual result of positive sample image or negative sample image that the judging result and two vehicle images belong to is compared,
If comparison is inconsistent, step S43 will not be chosen for test data in each positive sample image and negative sample image
Remainder as training data, using the training data as input, and continue to instruct by way of being fitted loss function
After practicing twin network model, step S41 is returned to；
Here, the loss function can use cross entropy loss function；
If comparing unanimously, step S44, twin network model training terminates.
Here, the application proposes a kind of reid algorithm of identification again independent of certain vehicle data amount, and work as engineering
Forward direction is found can be directly by continuing to train twin network mould to matching error data are added in training data when matching error
Type, to obtain reliable twin network model.
As shown in figure 5, the vehicle characteristics of the application extract in the implementation of model training method one, in step S42, described in calculating
The similarity of the feature of two vehicle images of positive sample image or negative sample image in test data, comprising:
Step S411 passes through norm layers of positive sample image calculated in the test data of normalization in caffe frame
Or the denominator part of the cosine similarity of the feature of two vehicle images of negative sample image, that is, realize first vehicle image
The mould of feature vector A is multiplied with the mould of the feature vector B of second vehicle image, and wherein A and B is by twin network mould respectively
Feature vector after type feature extraction；
Step S412 calculates element_wise layers by the element in the caffe frame and calculates the test data
In positive sample image or negative sample image two vehicle images feature cosine similarity molecular moiety, that is, realize special
It levies vector A and feature vector B and carries out dot product；
Step S413 is incited somebody to action by first link InnerProduct layers in the caffe frame and in conjunction with concat layers
The denominator part and molecular moiety that cosine similarity is calculated are converted to neural network, and with softmax by the nerve net
Network is exported.
Here, the post-processing network of loss is realized in design, due to being judged to middle using cosine similarity algorithm before engineering
Whether two cars are identical, therefore use element_wise, norm, InnerProduct and concat layer in caffe frame
Realize the calculating of cosine similarity algorithm.
The vehicle characteristics of the application extract during model training method one implemented, in step S43, using the training data as
Input, and continue to train twin network model by way of being fitted loss function, comprising:
By the result for using softmax to export the neural network and the training data as input, and pass through
The mode of fitting loss function continues to train twin network model, obtains more reliable twin network model with training.
The vehicle characteristics of the application extract in the implementation of model training method one, and step S1 is obtained in vehicle monitoring video flowing
Multiple image, two vehicle images of same vehicle are intercepted from two field pictures every time and are spliced into positive sample image, every time
Two vehicle images of different vehicle are intercepted from two field pictures and are spliced into negative sample image, comprising:
Step S11 finds all vehicles of different frame image in vehicle monitoring video flowing by the detection algorithm of deep learning
?；
Same color identifier is arranged to the same vehicle in different frame image based on the vehicle found in step S12；
Step S13 is based on the color identifier, intercepts two vehicles of same color identifier from two field pictures every time
Image, and splicing is carried out as positive sample image, two vehicle figures of color identifier not of the same race are intercepted from two field pictures every time
Picture, and splicing is carried out as negative sample image.
Here, can more reliable, efficiently obtain positive and negative sample image and negative sample by the color identifier of setting vehicle
Image.
The vehicle characteristics of the application extract in the implementation of model training method one, will be in the test data in step S3
Positive sample image or negative sample image split two vehicle images respectively, comprising:
The size of positive sample image or negative sample image in the test data is uniformly adjusted to default by step S31
Size；
Here, the size of positive sample image or negative sample image in the test data can be uniformly adjusted to 400*
200 pixels；
Step S32, by the slice layer of caffe frame, by the positive sample in the test data after unified adjustment size
This image or negative sample image cutting are two vehicle images.
Here, can be by the slice layer of caffe frame to the positive sample in the test data after unified adjustment size
This image or negative sample image carry out the image for being cut into two 200*200 pixels up and down.
Here, the sample spliced can be split using slice layers with the backbone network of project training network, and
It uses using GoogLenet Inception-V2 or ResNet50 network as the oviparity network of frame, two will split respectively
Image, which is input in twin network model, carries out feature extraction.
The vehicle characteristics of the application extract in the implementation of model training method one, and step S41 will be from the test data
Positive sample image or negative sample image are split respectively obtains the current twin network models progress feature of two vehicle image inputs
It extracts, comprising:
Every two vehicle images segmented are inputted to two GoogLenet of the twin network model respectively
Inception-V2 perhaps two GoogLenet Inception-V2 of carry out feature extraction of two ResNet50 branches or
Two ResNet50 branches share it is all can learning parameter, wherein first GoogLenet Inception-V2 or
ResNet50 branch can learning parameter be iterated, Article 2 GoogLenet Inception-V2 or ResNet50 branch
Can learning parameter directly learn from after the iteration of first GoogLenet Inception-V2 or ResNet50 branch
Parameter copy.
As shown in fig. 7, the application also provides a kind of vehicle identification method, which comprises
Step S5 inputs the image of two cars to be matched in the twin network model that training above-mentioned is completed；
Step S6 intercepts the spy of the image of the two cars of branch output therein from the twin network model
Sign；
Step S7 calculates the similarity of the feature of the image for the two cars being truncated to, based on the two cars being calculated
The similarity of the feature of image and preset similarity threshold judge that two cars are same vehicle or different vehicles.
Here, to reduce feedforward network burden, and similarity threshold such as cosine similarity threshold can be adjusted with scene change
Value, therefore can only intercept the characteristic extraction part of a twin network model branch therein, the i.e. the last layer of feedforward network
It is the full articulamentum of n dimension；
Before engineering in, twin network model is read, inputs the image of two cars to be matched, after feature extraction, into
The similar calculating of row cosine sets a cosine similarity threshold value, as shown in FIG. 6 if cosine similarity is more than or equal to threshold value
In image, two cars belong to same vehicle, are less than threshold value, then two cars belong to different vehicle.
According to another aspect of the present invention, a kind of equipment for handling in network equipment client information, the equipment are also provided
Including the memory for storing computer program instructions and the processor for executing program instructions, wherein when the computer
When program instruction is executed by the processor, the vehicle characteristics extraction model training method that the equipment executes any of the above-described is triggered.
According to another aspect of the present invention, a kind of equipment for handling in network equipment client information, the equipment are also provided
Including the memory for storing computer program instructions and the processor for executing program instructions, wherein when the computer
When program instruction is executed by the processor, triggers the equipment and execute above-mentioned vehicle identification method.
The detailed content of each equipment and storage medium embodiment of the invention, for details, reference can be made to the correspondences of each method embodiment
Part, here, repeating no more.
Obviously, those skilled in the art can carry out various modification and variations without departing from the essence of the application to the application
Mind and range.In this way, if these modifications and variations of the application belong to the range of the claim of this application and its equivalent technologies
Within, then the application is also intended to include these modifications and variations.
It should be noted that the present invention can be carried out in the assembly of software and/or software and hardware, for example, can adopt
With specific integrated circuit (ASIC), general purpose computer or any other realized similar to hardware device.In one embodiment
In, software program of the invention can be executed to implement the above steps or functions by processor.Similarly, of the invention
Software program (including relevant data structure) can be stored in computer readable recording medium, for example, RAM memory,
Magnetic or optical driver or floppy disc and similar devices.In addition, some of the steps or functions of the present invention may be implemented in hardware, example
Such as, as the circuit cooperated with processor thereby executing each step or function.
In addition, a part of the invention can be applied to computer program product, such as computer program instructions, when its quilt
When computer executes, by the operation of the computer, it can call or provide according to the method for the present invention and/or technical solution.
And the program instruction of method of the invention is called, it is possibly stored in fixed or moveable recording medium, and/or pass through
Broadcast or the data flow in other signal-bearing mediums and transmitted, and/or be stored according to described program instruction operation
In the working storage of computer equipment.Here, according to one embodiment of present invention including a device, which includes using
Memory in storage computer program instructions and processor for executing program instructions, wherein when the computer program refers to
When enabling by processor execution, method and/or skill of the device operation based on aforementioned multiple embodiments according to the present invention are triggered
Art scheme.
It is obvious to a person skilled in the art that invention is not limited to the details of the above exemplary embodiments, Er Qie
In the case where without departing substantially from spirit or essential attributes of the invention, the present invention can be realized in other specific forms.Therefore, no matter
From the point of view of which point, the present embodiments are to be considered as illustrative and not restrictive, and the scope of the present invention is by appended power
Benefit requires rather than above description limits, it is intended that all by what is fallen within the meaning and scope of the equivalent elements of the claims
Variation is included in the present invention.Any reference signs in the claims should not be construed as limiting the involved claims.This
Outside, it is clear that one word of " comprising " does not exclude other units or steps, and odd number is not excluded for plural number.That states in device claim is multiple
Unit or device can also be implemented through software or hardware by a unit or device.The first, the second equal words are used to table
Show title, and does not indicate any particular order.
Claims (10)
1. a kind of vehicle characteristics extract model training method, which is characterized in that this method comprises:
The multiple image in vehicle monitoring video flowing is obtained, two vehicle images of same vehicle are intercepted from two field pictures and are spelled
It is connected in positive sample image, two vehicle images of different vehicle are intercepted from two field pictures and is spliced into negative sample image；
A part is chosen from each positive sample image and negative sample image is used as test data；
Design includes the twin network model of two branches, by the positive sample image or negative sample image point in the test data
It Chai Fen not two vehicle images；
Based on the twin network model, two vehicles of the positive sample image or negative sample image in the test data are obtained
The feature of image judges that the twin network model is that training is completed also to be to continue with training based on obtained feature.
2. obtaining the test number the method according to claim 1, wherein being based on the twin network model
The feature of two vehicle images of positive sample image or negative sample image in judges the twin net based on obtained feature
Network model is that training is completed also to be to continue with training, comprising:
Step S41, by from the test data positive sample image or negative sample image split obtain two vehicle figures respectively
As the current twin network model progress feature extraction of input, the positive sample image or negative sample figure in the test data are obtained
The feature of two vehicle images of picture；
Step S42 calculates the feature of two vehicle images of the positive sample image or negative sample image in the test data
Similarity judges two vehicles of the positive sample image or negative sample image in the test data based on the similarity being calculated
Image, judges that two vehicle images are to belong to positive sample image to still fall within negative sample image, judging result is obtained, by institute
The actual result for stating positive sample image or negative sample image that judging result and two vehicle images belong to is compared,
If comparing unanimously, step S43 will not be chosen for the residue of test data in each positive sample image and negative sample image
Part is used as training data, using the training data as inputting, and by way of being fitted loss function continues to train twin
After network model, step S41 is returned to；
If comparing unanimously, step S44, twin network model training terminates.
3. according to the method described in claim 2, it is characterized in that, calculating the positive sample image or negative sample in the test data
The similarity of the feature of two vehicle images of this image, comprising:
Pass through positive sample images in norm layer calculating test data of normalization in caffe frame or negative sample image
The denominator part of the cosine similarity of the feature of two vehicle images；
The element_wise layers of positive sample image calculated in the test data are calculated by the element in the caffe frame
Or the molecular moiety of the cosine similarity of the feature of two vehicle images of negative sample image；
Cosine phase will be calculated by first link InnerProduct layers in the caffe frame and in conjunction with concat layers
It is converted to neural network like the denominator part of degree and molecular moiety, and is exported the neural network with softmax.
4. according to the method described in claim 3, it is characterized in that, using the training data as input, and by fitting damage
The mode for losing function continues to train twin network model, comprising:
By the result for using softmax to export the neural network and the training data as input, and pass through fitting
The mode of loss function continues to train twin network model.
5. the method according to claim 1, wherein obtaining the multiple image in vehicle monitoring video flowing, every time
Two vehicle images of same vehicle are intercepted from two field pictures and are spliced into positive sample image, are intercepted from two field pictures every time
Two vehicle images of different vehicle are simultaneously spliced into negative sample image, comprising:
All vehicles of different frame image in vehicle monitoring video flowing are found by the detection algorithm of deep learning；
Based on the vehicle found, same color identifier is arranged to the same vehicle in different frame image；
Based on the color identifier, two vehicle images of same color identifier are intercepted from two field pictures, and are spliced
As positive sample image, two vehicle images of color identifier not of the same race are intercepted from two field pictures, and carry out splicing as negative
Sample image.
6. the method according to claim 1, wherein by positive sample image or negative sample in the test data
Image splits two vehicle images respectively, comprising:
The size of positive sample image or negative sample image in the test data is uniformly adjusted to pre-set dimension；
By the slice layer of caffe frame, by the positive sample image or negative sample in the test data after unified adjustment size
The cutting of this image is two vehicle images.
7. according to the method described in claim 2, it is characterized in that, will be from the positive sample image or negative sample in the test data
This image is split respectively obtains the current twin network models progress feature extraction of two vehicle image inputs, comprising:
Every two vehicle images segmented are inputted to two GoogLenet of the twin network model respectively
Inception-V2 perhaps two GoogLenet Inception-V2 of carry out feature extraction of two ResNet50 branches or
Two ResNet50 branches share it is all can learning parameter, wherein first GoogLenetInception-V2 or
ResNet50 branch can learning parameter be iterated, Article 2 GoogLenetInception-V2 or ResNet50 branch
Can learning parameter directly learn to join from after the iteration of first GoogLenetInception-V2 ResNet50 branch
Number copy.
8. a kind of vehicle identification method, which is characterized in that this method comprises:
The image of two cars to be matched is inputted into the twin network mould that training as described in any one of claims 1 to 7 is completed
In type；
The feature of the image of the two cars of branch output therein is intercepted from the twin network model；
The similarity of the feature of the image for the two cars being truncated to is calculated, the feature of the image based on the two cars being calculated
Similarity and preset similarity threshold judge that two cars are same vehicle or different vehicles.
9. a kind of equipment for handling in network equipment client information, which includes for storing depositing for computer program instructions
Reservoir and processor for executing program instructions, wherein when the computer program instructions are executed by the processor, triggering should
Method described in any one of equipment perform claim requirement 1 to 7.
10. a kind of equipment for handling in network equipment client information, the equipment include for storing computer program instructions
Memory and processor for executing program instructions, wherein when the computer program instructions are executed by the processor, triggering
Method described in equipment perform claim requirement 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910632120.6A CN110348393B (en) | 2019-07-12 | 2019-07-12 | Vehicle feature extraction model training method, vehicle identification method and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910632120.6A CN110348393B (en) | 2019-07-12 | 2019-07-12 | Vehicle feature extraction model training method, vehicle identification method and equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110348393A true CN110348393A (en) | 2019-10-18 |
| CN110348393B CN110348393B (en) | 2020-11-20 |
Family
ID=68176088
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910632120.6A Expired - Fee Related CN110348393B (en) | 2019-07-12 | 2019-07-12 | Vehicle feature extraction model training method, vehicle identification method and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110348393B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN111353580A (en) * | 2020-02-03 | 2020-06-30 | 中国人民解放军国防科技大学 | Training method of target detection network, electronic device and storage medium |
| CN111612820A (en) * | 2020-05-15 | 2020-09-01 | 北京百度网讯科技有限公司 | Multi-target tracking method, and training method and device of feature extraction model |
| CN111881791A (en) * | 2020-07-16 | 2020-11-03 | 北京宙心科技有限公司 | Security identification method and system |
| CN112184640A (en) * | 2020-09-15 | 2021-01-05 | 中保车服科技服务股份有限公司 | Image detection model construction method and device and image detection method and device |
| CN113256992A (en) * | 2021-07-15 | 2021-08-13 | 智道网联科技(北京)有限公司 | Processing method and device based on vehicle road cloud |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106228142A (en) * | 2016-07-29 | 2016-12-14 | 西安电子科技大学 | Face verification method based on convolutional neural networks and Bayesian decision |
| 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 |
| CN108446612A (en) * | 2018-03-07 | 2018-08-24 | 腾讯科技（深圳）有限公司 | vehicle identification method, device and storage medium |
| CN108596277A (en) * | 2018-05-10 | 2018-09-28 | 腾讯科技（深圳）有限公司 | A kind of testing vehicle register identification method, apparatus and storage medium |
| US20190147320A1 (en) * | 2017-11-15 | 2019-05-16 | Uber Technologies, Inc. | "Matching Adversarial Networks" |
| CN109767456A (en) * | 2019-01-09 | 2019-05-17 | 上海大学 | A kind of method for tracking target based on SiameseFC frame and PFP neural network |
| CN109800624A (en) * | 2018-11-27 | 2019-05-24 | 上海眼控科技股份有限公司 | A kind of multi-object tracking method identified again based on pedestrian |
| CN109886141A (en) * | 2019-01-28 | 2019-06-14 | 同济大学 | A kind of pedestrian based on uncertainty optimization discrimination method again |
-
2019
- 2019-07-12 CN CN201910632120.6A patent/CN110348393B/en not_active Expired - Fee Related
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106228142A (en) * | 2016-07-29 | 2016-12-14 | 西安电子科技大学 | Face verification method based on convolutional neural networks and Bayesian decision |
| US20190147320A1 (en) * | 2017-11-15 | 2019-05-16 | Uber Technologies, Inc. | "Matching Adversarial Networks" |
| CN108446612A (en) * | 2018-03-07 | 2018-08-24 | 腾讯科技（深圳）有限公司 | vehicle identification method, device and storage medium |
| 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 |
| CN108596277A (en) * | 2018-05-10 | 2018-09-28 | 腾讯科技（深圳）有限公司 | A kind of testing vehicle register identification method, apparatus and storage medium |
| CN109800624A (en) * | 2018-11-27 | 2019-05-24 | 上海眼控科技股份有限公司 | A kind of multi-object tracking method identified again based on pedestrian |
| CN109767456A (en) * | 2019-01-09 | 2019-05-17 | 上海大学 | A kind of method for tracking target based on SiameseFC frame and PFP neural network |
| CN109886141A (en) * | 2019-01-28 | 2019-06-14 | 同济大学 | A kind of pedestrian based on uncertainty optimization discrimination method again |
Non-Patent Citations (7)
| Title |
|---|
| QIAN ZHANG ET AL.: "Vehicle Verification Based on Deep Siamese Network with Similarity Metric", 《LECTURE NOTES IN COMPUTER SCIENCE, 773–782. DOI:10.1007/978-3-319-77380-3_74》 * |
| 中国航天科工集团第三研究员三一〇所: "《自助系统与人工智能领域技术发展报告》", 30 April 2017, 国防工业出版社 * |
| 唐怡冬: "基于信息融合的车辆跟踪技术研究", 《中国优秀博硕士学位论文全文数据库（硕士） 工程科技Ⅱ辑》 * |
| 李洁樱: "基于孪生卷积神经网络的车辆一致性判别方法", 《中国交通信息化》 * |
| 杨清夙: "车辆视频检测与跟踪系统的研究与实现", 《中国优秀博硕士学位论文全文数据库（硕士） 工程科技辑》 * |
| 甘小楚: "基于深度学习的图片文字相关性计算算法研究与实现", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 赵作升: "基于HSV颜色空间的视频车辆检测与跟踪算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN111353580A (en) * | 2020-02-03 | 2020-06-30 | 中国人民解放军国防科技大学 | Training method of target detection network, electronic device and storage medium |
| CN111353580B (en) * | 2020-02-03 | 2023-06-20 | 中国人民解放军国防科技大学 | Training method of target detection network, electronic equipment and storage medium |
| CN111612820A (en) * | 2020-05-15 | 2020-09-01 | 北京百度网讯科技有限公司 | Multi-target tracking method, and training method and device of feature extraction model |
| CN111612820B (en) * | 2020-05-15 | 2023-10-13 | 北京百度网讯科技有限公司 | Multi-target tracking method, training method and device of feature extraction model |
| CN111881791A (en) * | 2020-07-16 | 2020-11-03 | 北京宙心科技有限公司 | Security identification method and system |
| CN111881791B (en) * | 2020-07-16 | 2021-10-15 | 北京宙心科技有限公司 | Security identification method and system |
| CN112184640A (en) * | 2020-09-15 | 2021-01-05 | 中保车服科技服务股份有限公司 | Image detection model construction method and device and image detection method and device |
| CN113256992A (en) * | 2021-07-15 | 2021-08-13 | 智道网联科技(北京)有限公司 | Processing method and device based on vehicle road cloud |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110348393B (en) | 2020-11-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110348393A (en) | Vehicle characteristics extract model training method, vehicle identification method and equipment | |
| CN108846355B (en) | Image processing method, face recognition device and computer equipment | |
| EP3690742A1 (en) | Method for auto-labeling training images for use in deep learning network to analyze images with high precision, and auto-labeling device using the same | |
| EP3486838A1 (en) | System and method for semi-supervised conditional generative modeling using adversarial networks | |
| CN108460415B (en) | Language identification method | |
| KR102042168B1 (en) | Methods and apparatuses for generating text to video based on time series adversarial neural network | |
| US20120155766A1 (en) | Patch description and modeling for image subscene recognition | |
| EP3989158A1 (en) | Method, apparatus and device for video similarity detection | |
| CN114241505B (en) | Method and device for extracting chemical structure image, storage medium and electronic equipment | |
| CN113177630B (en) | Data memory elimination method and device for deep learning model | |
| US20200151458A1 (en) | Apparatus and method for video data augmentation | |
| US8204889B2 (en) | System, method, and computer-readable medium for seeking representative images in image set | |
| CN110348392B (en) | Vehicle matching method and device | |
| Li et al. | Image manipulation localization using attentional cross-domain CNN features | |
| CN112906631A (en) | Dangerous driving behavior detection method and detection system based on video | |
| CN111461211B (en) | Feature extraction method for lightweight target detection and corresponding detection method | |
| CN111241873A (en) | Image reproduction detection method, training method of model thereof, payment method and payment device | |
| CN110766077A (en) | Method, device and equipment for screening sketch in evidence chain image | |
| CN113298015A (en) | Video character social relationship graph generation method based on graph convolution network | |
| CN109522921A (en) | Statement similarity method of discrimination and equipment | |
| CN115984949A (en) | Low-quality face image recognition method and device with attention mechanism | |
| CN113805977B (en) | Test evidence obtaining method, model training method, device, equipment and storage medium | |
| Song et al. | Text Siamese network for video textual keyframe detection | |
| CN115620083A (en) | Model training method, face image quality evaluation method, device and medium | |
| CN115705758A (en) | Living body identification method, living body identification device, electronic device, and storage medium |
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 | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: Vehicle feature extraction model training method, vehicle recognition method and equipment Effective date of registration: 20220211 Granted publication date: 20201120 Pledgee: Shanghai Bianwei Network Technology Co.,Ltd. Pledgor: SHANGHAI EYE CONTROL TECHNOLOGY Co.,Ltd. Registration number: Y2022310000023 |
|
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20201120 |