CN106845341A - A kind of unlicensed vehicle identification method based on virtual number plate - Google Patents
A kind of unlicensed vehicle identification method based on virtual number plate Download PDFInfo
- Publication number
- CN106845341A CN106845341A CN201611156309.5A CN201611156309A CN106845341A CN 106845341 A CN106845341 A CN 106845341A CN 201611156309 A CN201611156309 A CN 201611156309A CN 106845341 A CN106845341 A CN 106845341A
- Authority
- CN
- China
- Prior art keywords
- virtual number
- number plate
- vehicle
- image
- feature
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/584—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Abstract
The present invention discloses a kind of unlicensed vehicle identification method based on virtual number plate, it is characterised in that methods described includes：Step 1 is virtual number plate construction step, including carries out vehicle detection based on video or image, obtains vehicle close up view, and feature extraction is carried out to vehicle image, generates virtual number plate, builds virtual number plate storehouse；Step 2 is virtual number plate storehouse identification step, including carries out target vehicle detection based on video or image, obtains target vehicle close up view, and feature extraction is carried out to target vehicle image, virtual number plate is generated, with the virtual number plate storehouse comparison result.The method is on the basis of based on image recognition, for each car, generate unique vehicle virtual number plate, target vehicle is retrieved so as to also effectively being searched without Car license recognition, is made a dash across the red light for highway fee evasion inspection, unlicensed car, vehicle is quickly searched etc., and various traffic applications have positive effect.
Description
Technical field
The present invention relates to computer vision field, more particularly to a kind of unlicensed vehicle identification method based on virtual number plate,
It is an important application in traffic application field.
Background technology
With the fast-developing extensive increase with civilian vehicle of China's communication, driving vehicle is entered by image
Row Classification Management is particularly important, the identification for unlicensed car and management work when special.There is presently no a kind of maturation
Unlicensed car recognizes solution, because the program needs accurate position size, vehicle brand, annual test mark, the suspension member for obtaining vehicle
Etc. information, same vehicle is uniquely determined with this.Unlicensed car test rope and management only in this way could be effectively carried out, but it is conventional
Method often due to feature description not comprehensively, cause this work launch be not fine.
The content of the invention
In order to effective lookup retrieves target vehicle, virtual number plate storehouse is built by image procossing, thus improve unlicensed
The identification and management of car.The purpose of the present invention is achieved through the following technical solutions.
A kind of unlicensed vehicle identification method based on virtual number plate, it is characterised in that methods described includes：Step 1 is void
Intend number plate construction step, including vehicle detection is carried out based on video or image, obtain vehicle close up view, spy is carried out to vehicle image
Extraction is levied, virtual number plate is generated, virtual number plate storehouse is built；Step 2 is virtual number plate storehouse identification step, including based on video or figure
As carrying out target vehicle detection, target vehicle close up view is obtained, feature extraction is carried out to target vehicle image, generate virtual number
Board, with the virtual number plate storehouse comparison result.
Preferably, the step 1 includes：
Step 1-1, vehicle detection is carried out based on video or image, extract vehicle image；
Step 1-2, feature extraction is carried out to vehicle image, the feature extraction is based on the image described by deep learning
Feature, described image feature includes some dimension global characteristics of the full car image of description, and/or describes some of vehicle regional area
Dimension local feature；
Step 1-3, the global characteristics of gained and/or local feature are quantified, and stored, generate virtual number plate, built empty
Intend number plate storehouse.
Preferably, the step 2 includes：
Step 2-1, target vehicle detection is carried out based on video or image, extract target vehicle image；
Step 2-2, feature extraction is carried out to target vehicle image, the feature extraction is based on described by deep learning
Characteristics of image, described image feature includes some dimension global characteristics of the full car image of description, and/or describes vehicle regional area
Some dimension local features；
Step 2-3, the global characteristics of the target vehicle of gained and/or local feature are quantified, and stored, generate target carriage
Virtual number plate；
Step 2-4, the virtual number plate of target vehicle is compared one by one with the virtual number plate in virtual number plate storehouse, be identified
As a result.
Preferably, in step 1-2 or step 2-2, the deep learning includes using fast area depth convolutional Neural
Network objectives detection algorithm.
Preferably, in step 1-2 or step 2-2, the method for the deep learning is not used, but use cascade nature
Object detection method.
Preferably, the fast area depth convolutional neural networks algorithm of target detection is included in the close up view for obtaining vehicle
Afterwards, using depth convolution method, using the close up view as depth convolutional neural networks input, by feedforward neural network
Some dimension global characteristics of vehicle image are calculated, while determine regional area according to geometry site, based on this
Obtain corresponding some dimension local features.
Preferably, the step 1-3 includes for the global characteristics and local feature of acquisition forming complete characteristic vector
And the lossless virtual number plate storehouse for being stored in feature database, vehicle being built with this, or according to the training of great amount of samples data, will
Feature forms Hash codes according to training threshold value, and characterizing algorithm according to corresponding Hash forms Quick Response Code, and the void of vehicle is built with this
Intend number plate storehouse.
Preferably, the step 2-4 includes by the way of Distributed Parallel Computing, using reciprocal proportion distance and cosine away from
Virtual number plate in the virtual number plate and data storehouse by target vehicle carries out Distance conformability degree matching, so as to carry out vehicle comparison
Analysis.
Preferably, the step 2-4 includes that judgement is recognized successfully when similarity is higher than certain threshold value, and exports institute
The virtual number plate result matched somebody with somebody.
The advantage of the invention is that：On the basis of based on image recognition, overall situation and partial situation's feature of vehicle is made full use of,
Include but are not limited to the characteristics of image based on deep learning or cascade goal description.And then generate virtual number plate and compare, it is comprehensive
Conjunction considers entirety and local message, based on this, for each car, can generate unique vehicle virtual number
Board, target vehicle is retrieved so as to also can effectively be searched without Car license recognition, is rushed for highway fee evasion inspection, unlicensed car
Quickly various traffic applications such as lookup have positive effect for red light, vehicle.
Brief description of the drawings
By reading the detailed description of hereafter preferred embodiment, various other advantages and benefit is common for this area
Technical staff will be clear understanding.Accompanying drawing is only used for showing the purpose of preferred embodiment, and is not considered as to the present invention
Limitation.And in whole accompanying drawing, identical part is denoted by the same reference numerals.In the accompanying drawings：
Fig. 1 shows a kind of unlicensed vehicle identification method flow chart based on virtual number plate according to embodiments of the present invention；
Fig. 2 shows the structure chart of multilayer convolutional neural networks model according to embodiments of the present invention；
Fig. 3 shows the structure chart of improved multilayer convolutional neural networks model according to embodiments of the present invention.
Fig. 4 shows that DPM+ depth targets according to embodiments of the present invention detect the method flow diagram for vehicle identification.
Fig. 5 shows depth recognition neural network framework figure according to embodiments of the present invention.
Specific embodiment
The illustrative embodiments of the disclosure are more fully described below with reference to accompanying drawings.Although showing this public affairs in accompanying drawing
The illustrative embodiments opened, it being understood, however, that may be realized in various forms the disclosure without the reality that should be illustrated here
The mode of applying is limited.Conversely, there is provided these implementation methods are able to be best understood from the disclosure, and can be by this public affairs
The scope opened it is complete convey to those skilled in the art.
Embodiments in accordance with the present invention, disclose a kind of unlicensed vehicle identification method based on virtual number plate, methods described
Comprise the following steps：
Step 1-1, vehicle detection is carried out based on video or image, extract vehicle image；
Step 1-2, feature extraction is carried out to vehicle image, the feature extraction be based on deep learning (for example：Quick area
Domain depth convolutional neural networks algorithm of target detection) described by characteristics of image, described image feature includes the full car image of description
Some dimensions (such as 4096 dimension) global characteristics, it is also possible to some dimension local features including describing vehicle regional area, such as
Some dimensions (such as 4096 dimension) local feature of the regional areas such as vehicle glazing, annual test mark, exhaust grille, pendant；
Step 1-3, the global characteristics of gained and/or local feature are quantified, and stored, generate virtual number plate, built empty
Intend number plate storehouse；
Step 2-1, target vehicle detection is carried out based on video or image, extract target vehicle image；
Step 2-2, feature extraction is carried out to target vehicle image, the feature extraction be based on deep learning (for example：Hurry up
Fast regional depth convolutional neural networks algorithm of target detection) described by characteristics of image, described image feature includes the full car of description
Some dimensions (such as 4096 dimension) global characteristics of image, and/or some dimension local features of vehicle regional area, such as car are described
Some dimensions (such as 4096 dimension) local feature of the regional areas such as vehicle window, annual test mark, exhaust grille, pendant；
Step 2-3, the global characteristics of the target vehicle of gained and/or local feature are quantified, and stored, generate target carriage
Virtual number plate；
Step 2-4, the virtual number plate of target vehicle is compared one by one with the virtual number plate in virtual number plate storehouse, be identified
As a result.
Embodiments in accordance with the present invention, vehicle close up view belongs to natural image, the statistical property of a part for its image with
Other parts are the same, and the feature for thus learning in a part also can be with another part, so on this image
All positions, it is possible to use same learning characteristic.
Preferably, in step 1-2, fast area depth convolutional neural networks algorithm of target detection is included in and obtains vehicle
Close up view after, using depth convolution method, using the close up view as depth convolutional neural networks input, by feedforward
Neural computing obtains some dimension global characteristics of vehicle image, while determine regional area according to geometry site, with
Corresponding some dimension local features are obtained based on this.Specifically, based on convolutional neural networks obtain it is some dimension global characteristics or
The some dimension local features of person comprise the following steps：
For the vehicle close up view for being gathered, from vehicle feature global image (or from vehicle glazing, annual test mark, row
In the local area images such as gas grid, pendant) randomly select a fritter, such as 3x3 as sample, from this fritter sample middle school
Practise some features, can using from this 3x3 samples learning to feature as detector, be applied to appointing for vehicle close-up image
In meaning place.Preferably, with the feature learnt from 3x3 samples with script vehicle close up view (including vehicle feature
The local area image such as global image or vehicle glazing, annual test mark, exhaust grille, pendant) make convolution, so as to appointing on figure
One position obtains an activation value for different characteristic.Embodiments in accordance with the present invention, first from a vehicle close up view of 96x96
As the feature that learning has to the sample of its 3x3, it is assumed that this is by there is 100 own codings of implicit unit to complete
's.Small images region to each 3x3 of the image of 96x96 carries out convolution algorithm, to obtain convolution feature.Namely
Say, extract 3x3 pocket, and since origin coordinates successively be labeled as (1,1), (1,2) ..., until (94,
94), then the region extracted is run the sparse own coding trained to obtain the activation value of feature one by one.In this embodiment
In, 100 set can be obtained, each set contains 89x89 convolution feature.According to the embodiment of the present invention, process of convolution mistake
Cheng Wei：Assuming that given the vehicle close-up image of r*c, being defined as Xlarge.Taken out by from vehicle close-up image first
The small-sized image sample Xsmall of the a*b for taking trains sparse own coding, and (k is hidden layer neuron number to have obtained k feature
Amount), then for the block of each the a*b size in Xlarge, activation value fs is sought, convolution then is carried out to these fs.Such
Eigenmatrix to after (r-a+1) * (c-b+1) * k convolution.
After the feature for obtaining vehicle close up view by convolution, classification is done using these features.In the prior art
Use all feature associations for parsing a to grader, such as softmax graders, amount of calculation is very big.
For example：For an image for 96X96 pixels, 400 features are obtained by 3x3 input study, and each convolution
A result set for (96-3+1) * (96-3+1)=8836 can be obtained, due to having been obtained for 400 features, so for every
The size of individual example result collection is just up to million grades of feature.And learn a classification for the input for having more than million features
Easily there is the phenomenon of overfitting in device.In order to solve this problem and describe big image, in an embodiment of the present invention,
Feature to the diverse location of vehicle close-up image carries out aggregate statistics, it is preferable that certain calculated on one region of image is special
Determine the average value (or maximum) of feature.Not only there is these summary statistics features much lower dimension (to compare and be carried using all
The feature for obtaining), while can also improve result (being not easy overfitting).Average (or maximum) feature in these regions is come
Convolution feature the most improved, classifies for doing.
Normally, vehicle close up view uses the cromogram, cromogram to have 3 passages, individually enter for each passage
Row convolution and pond, each value of hidden layer are that 3 passages for corresponding to a width figure wear, so point 3 passages enter
To be added up after row convolution, just can just corresponded on a neuron for hidden layer, that is, correspond to a feature.
Further, in a particular embodiment, using multilayer convolution, then reuse full articulamentum and be trained,
The purpose of multilayer convolution is that the feature that one layer of convolution is acquired is often local, and the number of plies is higher, feature more globalization acquired.
Convolution can be carried out to vehicle close up view using the model shown in Fig. 2.The model employ 2-GPU parallel organizations, i.e., the 1st, 2,
4th, model parameter is all divided into 2 parts and is trained by 5 convolutional layers.Herein, further, parallel organization is divided into data simultaneously
Row is parallel with model.Data parallel refers to that model structure is identical on different GPU, but training data is carried out into cutting, respectively
Training obtains different models, is then again merged model.And model is then parallel, if the model parameter of dried layer is carried out
Cutting, is trained on different GPU using identical data, and the result for obtaining is directly connected to the input as next layer.
The basic parameter of Fig. 2 institutes representation model is：
Input：The vehicle close up view of 224 × 224 sizes, 3 passages
Ground floor convolution：The convolution kernel of 5 × 5 sizes 96, upper 48 of each GPU.
Ground floor max-pooling：2 × 2 core.
Second layer convolution：3 × 3 convolution kernels 256, upper 128 of each GPU.
Second layer max-pooling：2 × 2 core.
Third layer convolution：It is to be connected entirely with last layer, the convolution kernel of 3*3 384.Assign to two last 192 of GPU.
4th layer of convolution：3 × 3 convolution kernel 384, two each 192 of GPU.The layer is connected without process with last layer
Pooling layers.
Layer 5 convolution：3 × 3 convolution kernel 256, two last 128 of GPU.
Layer 5 max-pooling：2 × 2 core.
Ground floor is connected entirely：4096 dimensions, an one-dimensional vector is connected to become by the output of layer 5 max-pooling, is made
It is the input of this layer.
The second layer is connected entirely：4096 dimensions
Softmax layers：It is output as 1000, the every one-dimensional of output is all probability that vehicle close up view belongs to the category.
Above-mentioned model is improved it is possible to further the structure using Fig. 3.In figure 3, it is complete at last only one layer
Articulamentum, is then exactly softmax layers, wherein, using full articulamentum the representing as image.In full articulamentum, with the 4th
Thus the output of layer convolution and third layer max-pooling learn to vehicle close up view part as the input of full articulamentum
With global feature.
In step 1-3, the global characteristics and local feature of acquisition are formed into complete characteristic vector and lossless is deposited
It is stored in feature database, the virtual number plate storehouse of vehicle is built with this, or according to the training of great amount of samples data, by feature according to instruction
Practice threshold value and form Hash codes, characterizing algorithm according to corresponding Hash forms Quick Response Code, and the virtual number plate storehouse of vehicle is built with this.
Preferably, the virtual number plate storehouse that vehicle is built in the step can be using one of following three kinds of algorithms：
The step of algorithm one is：
1. picture unification is zoomed to 8*8 according to foregoing convolutional neural networks algorithm, totally 64 pictures of pixel.2. turn
Turn to gray-scale map：Picture after scaling is converted into the gray-scale map of 256 ranks.
Gray-scale map related algorithm is following (R=red, G=green, B=blue)：
1) floating-point arithmetics：Gray=R*0.3+G*0.59+B*0.11
2) integers method：Gray=(R*30+G*59+B*11)/100
3) displacement methods：Gray=(R*76+G*151+B*28)>>8；
4) mean value methods：Gray=(R+G+B)/3；
5) only takes green：Gray=G；
3. average value is calculated：Calculating carries out the average value of all pixels point of picture after gray proces.
4. compared pixels gray value：Traversal gray scale picture each pixel, 1 is recorded as if greater than average value, otherwise for
0。
5. information fingerprint is obtained：Combination 64 bit, sequentially random being consistent property.
6. compared using distance, the more big then explanation picture of distance is more inconsistent, conversely, apart from smaller, illustrating that picture gets over phase
Seemingly, when distance is 0, illustrate identical.
The step of algorithm two is：
1. picture unification is zoomed to according to foregoing convolutional neural networks algorithm the size of 32*32, facilitate DCT to calculate.
2. gray-scale map is converted into：Picture after scaling is converted into the gray-scale map of 256 ranks.(specific algorithm is shown in algorithm one
Step)
3. DCT is calculated:Set of the DCT picture separated component rate.
4. DCT is reduced：DCT is 32*32, retains the 8*8 in the upper left corner, the low-limit frequency of these pictures for representing.5. calculate
Average value：Calculate the average value of all pixels point after reducing DCT.
6. DCT is further reduced：1 is recorded as more than average value, otherwise is recorded as 0.
7. information fingerprint is obtained：64 information bits of combination, random the being consistent property of order.
8. compared using distance, the more big then explanation picture of distance is more inconsistent, conversely, apart from smaller, illustrating that picture gets over phase
Seemingly, when distance is 0, illustrate identical.
The step of algorithm three is：
1. picture unification is zoomed to according to foregoing convolutional neural networks algorithm the size of 9*8, totally 72 pixels.
2. gray-scale map is converted into：Picture after scaling is converted into the gray-scale map of 256 ranks.(specific algorithm is shown in algorithm one
Step)
3. difference value is calculated：Algorithm three be work between adjacent pixels, so often generate 8 between 9 pixels of row
Different difference, 8 row, then generate 64 difference values altogether.
4. fingerprint is obtained：If brighter on the right of the pixel ratio on the left side, 1 is recorded as, is otherwise 0.
5. information fingerprint is obtained：64 information bits of combination, random the being consistent property of order.
6. compared using distance, the more big then explanation picture of distance is more inconsistent, conversely, apart from smaller, illustrating that picture gets over phase
Seemingly, when distance is 0, illustrate identical.
In step 2-2, using the method similar with step 1-2, target vehicle is obtained based on convolutional neural networks
Some dimension global characteristics or some dimension local features.
In step 2-3, by the global characteristics of the target vehicle obtained by step 2-2 and/or office with step 1-3 similarly
Portion's characteristic quantification, and store, generate the virtual number plate of target vehicle.
In step 2-4, by the way of Distributed Parallel Computing, using reciprocal proportion distance and COS distance by target carriage
Virtual number plate and data storehouse in virtual number plate carry out Distance conformability degree matching, so as to carry out vehicle compare analysis, work as phase
During like degree higher than certain threshold value, judgement is recognized successfully, and exports the virtual number plate result for being matched.
In the particular embodiment, it is using two vector folders in vector space to carry out judgement using Distance conformability degree matching
Cosine of an angle value is used as two measurements of the size of interindividual variation of measurement.Vector is directive line segment in hyperspace, such as
Really two directions of vector are consistent, i.e. angle is close to zero, then the two vectors are just close.And to determine that two vector directions are
No consistent, this will use the angle that the cosine law calculates vector.It is assumed that three sides of a triangle are a, b and c, corresponding three
Angle is A, B and C, then the cosine of angle A is：
It is above-mentioned if regarding the both sides b and c of triangle as two vectors
Formula is equivalent to：
Wherein denominator represents two vector b
With the length of c, molecule represents two inner products of vector.
In an embodiment of the present invention, feature vector, X and step 2-3 are obtained in the virtual number plate storehouse that step 1-3 is obtained
The characteristic vector Y of the virtual number plate of target vehicle is respectively：
X1, x2 ..., x6400 and y1, y2 ..., y6400
Then, the COS distance between them can be represented with the cosine value of angle between them：
When co sinus vector included angle is more bordering on 1, characteristic vector levels off to and repeat completely.When co sinus vector included angle is higher than one
When determining threshold value, judgement is recognized successfully, and exports the virtual number plate result in the virtual number plate storehouse for being matched.
In another preferred embodiment of the invention, in step 1-2 or step 2-2, single depth is not used
The method of habit, but use cascade nature object detection method.Specifically, using DPM (deformable member model) and depth targets
The method that is combined of detection, wherein, DPM is briefly summarized as algorithm of target detection, the step of the algorithm：
1 masterplate cascades Shape Feature Extraction；
2 global images cascade Shape Feature Extraction；
The 3 SVM target detections based on the matching of quick sliding window are fitted with position.
Depth targets detection is that a kind of cascade target designed by using the good feature representation ability of deep learning is examined
Workflow is surveyed, its synthesis weighed the relation between hardware cost and computing accuracy rate, be briefly summarized as the step of the algorithm：
1 image multi-resolution pyramid builds；
The feedforward of 2 hierarchy depths is calculated, and obtains multi-layer characteristic pattern；
3 target detections based on the quick sliding window matchings of SOFTMAX；
The regional location NMS fittings of 4 multistage voting mechanisms.
In the preferred embodiments of the present invention, the algorithm process for being used is to be combined DPM and depth targets detection.Generally
Ground, depth targets detection algorithm shows excellent performance in the verification and measurement ratio of target, but due to not accounting in object
Geometric relationship, especially for the clearer and more definite object of the rigid structures such as vehicle, bicycle, pedestrian, therefore this hair
Spring deformation model in the bright algorithm by DPM is incorporated into deep learning model.DPM+ depth targets are detected for vehicle identification
Method flow diagram it is as shown in Figure 4.
Depth convolutional network structure of the invention is as shown in figure 5, the network takes full advantage of the part and global letter of image
Various depth characteristics such as breath feature, texture, color, gradient comprising image, it is bottom-up specific from being abstracted into, it is truly realized
Comprehensive accurate description.In specific convolution operation, next layer of convolution operation does not rely on the convolutional layer of last layer just, also according to
Rely upper several layers of convolution, so can sufficiently consider region consistency, the description of exactly this approximate redundancy causes algorithm
Robustness is very high.
The present invention is no to use of the prior art hundred layers of even thousand layers of neutral net, but peculiar with reference to actually proposing
Depth convolutional neural networks structure, the network structure is clearly easy, and operability is very high with scalability.
The above, the only present invention preferably specific embodiment, but protection scope of the present invention is not limited thereto,
Any one skilled in the art the invention discloses technical scope in, the change or replacement that can be readily occurred in,
Should all be included within the scope of the present invention.Therefore, protection scope of the present invention should be with the protection model of the claim
Enclose and be defined.
Claims (9)
1. a kind of unlicensed vehicle identification method based on virtual number plate, it is characterised in that methods described includes：
Step 1 is virtual number plate construction step, including carries out vehicle detection based on video or image, obtains vehicle close up view, right
Vehicle image carries out feature extraction, generates virtual number plate, builds virtual number plate storehouse；
Step 2 is virtual number plate storehouse identification step, including carries out target vehicle detection based on video or image, obtains target vehicle
Close up view, feature extraction is carried out to target vehicle image, virtual number plate is generated, with the virtual number plate storehouse comparison result.
2. a kind of unlicensed vehicle identification method based on virtual number plate according to claim 1, it is characterised in that the step
Rapid 1 includes：
Step 1-1, vehicle detection is carried out based on video or image, extract vehicle image；
Step 1-2, feature extraction is carried out to vehicle image, the feature extraction is special based on the image described by deep learning
Levy, described image feature includes some dimension global characteristics of the full car image of description, and/or some dimensions for describing vehicle regional area
Local feature；
Step 1-3, the global characteristics of gained and/or local feature are quantified, and stored, generate virtual number plate, build virtual number
Board storehouse.
3. the unlicensed vehicle identification method based on virtual number plate according to claim 1 and 2, it is characterised in that：The step
Rapid 2 include：
Step 2-1, target vehicle detection is carried out based on video or image, extract target vehicle image；
Step 2-2, feature extraction is carried out to target vehicle image, the feature extraction is based on the image described by deep learning
Feature, described image feature includes some dimension global characteristics of the full car image of description, and/or describes some of vehicle regional area
Dimension local feature；
Step 2-3, the global characteristics of the target vehicle of gained and/or local feature are quantified, and stored, generation target vehicle
Virtual number plate；
Step 2-4, the virtual number plate of target vehicle is compared one by one with the virtual number plate in virtual number plate storehouse, be identified knot
Really.
4. the unlicensed vehicle identification method based on virtual number plate according to Claims 2 or 3, it is characterised in that：Step 1-2
Or in step 2-2, the deep learning includes using fast area depth convolutional neural networks algorithm of target detection.
5. the unlicensed vehicle identification method based on virtual number plate according to Claims 2 or 3, it is characterised in that：Step 1-2
Or in step 2-2, the method that the deep learning is not used, but use cascade nature object detection method.
6. the unlicensed vehicle identification method based on virtual number plate according to claim 4, it is characterised in that：The quick area
Domain depth convolutional neural networks algorithm of target detection is included in after the close up view for obtaining vehicle, using depth convolution method, will
The close up view is calculated some dimensions of vehicle image by feedforward neural network as the input of depth convolutional neural networks
Global characteristics, while determining regional area according to geometry site, obtain corresponding some dimension local features based on this.
7. a kind of unlicensed vehicle identification method based on virtual number plate according to claim 3, it is characterised in that
The step 1-3 includes for the global characteristics and local feature of acquisition forming complete characteristic vector and lossless storage
In feature database, the virtual number plate storehouse of vehicle is built with this, or according to the training of great amount of samples data, by feature according to training
Threshold value forms Hash codes, and characterizing algorithm according to corresponding Hash forms Quick Response Code, and the virtual number plate storehouse of vehicle is built with this.
8. a kind of unlicensed vehicle identification method based on virtual number plate according to claim 3, it is characterised in that the step
Rapid 2-4 includes by the way of Distributed Parallel Computing, using reciprocal proportion distance and COS distance by the virtual number of target vehicle
Virtual number plate in board and database carries out Distance conformability degree matching, and analysis is compared so as to carry out vehicle.
9. a kind of unlicensed vehicle identification method based on virtual number plate according to claim 8, it is characterised in that the step
Rapid 2-4 includes that judgement is recognized successfully when similarity is higher than certain threshold value, and exports the virtual number plate result for being matched.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611156309.5A CN106845341B (en) | 2016-12-15 | 2016-12-15 | Unlicensed vehicle identification method based on virtual number plate |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611156309.5A CN106845341B (en) | 2016-12-15 | 2016-12-15 | Unlicensed vehicle identification method based on virtual number plate |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106845341A true CN106845341A (en) | 2017-06-13 |
| CN106845341B CN106845341B (en) | 2020-04-10 |
Family
ID=59139259
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611156309.5A Active CN106845341B (en) | 2016-12-15 | 2016-12-15 | Unlicensed vehicle identification method based on virtual number plate |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106845341B (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107748895A (en) * | 2017-10-29 | 2018-03-02 | 北京工业大学 | UAV Landing landforms image classification method based on DCT CNN models |
| CN108960412A (en) * | 2018-06-29 | 2018-12-07 | 北京京东尚科信息技术有限公司 | Image-recognizing method, device and computer readable storage medium |
| CN109190649A (en) * | 2018-07-02 | 2019-01-11 | 北京陌上花科技有限公司 | A kind of optimization method and device of deep learning network model server |
| CN109241349A (en) * | 2018-08-14 | 2019-01-18 | 中国电子科技集团公司第三十八研究所 | A kind of monitor video multiple target classification retrieving method and system based on deep learning |
| CN109992690A (en) * | 2019-03-11 | 2019-07-09 | 中国华戎科技集团有限公司 | A kind of image search method and system |
| CN110490272A (en) * | 2019-09-05 | 2019-11-22 | 腾讯音乐娱乐科技（深圳）有限公司 | Image content similarity analysis method, apparatus and storage medium |
| CN110555125A (en) * | 2018-05-14 | 2019-12-10 | 桂林远望智能通信科技有限公司 | Vehicle retrieval method based on local features |
| CN111007761A (en) * | 2019-11-28 | 2020-04-14 | 上海蓝色帛缔智能工程有限公司 | Automatic monitoring and management system of data center |
| CN112990136A (en) * | 2021-04-29 | 2021-06-18 | 成都深蓝思维信息技术有限公司 | Target detection method and device |
| CN113157641A (en) * | 2021-02-07 | 2021-07-23 | 北京卓视智通科技有限责任公司 | Unlicensed vehicle archiving and querying method, device, system, equipment and storage medium |
| CN113706390A (en) * | 2021-10-29 | 2021-11-26 | 苏州浪潮智能科技有限公司 | Image conversion model training method, image conversion method, device and medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102902957A (en) * | 2012-09-05 | 2013-01-30 | 佳都新太科技股份有限公司 | Video-stream-based automatic license plate recognition method |
| CN103000028A (en) * | 2011-09-14 | 2013-03-27 | 上海宝康电子控制工程有限公司 | Vehicle registration plate recognition system and method |
| CN103258213A (en) * | 2013-04-22 | 2013-08-21 | 中国石油大学(华东) | Vehicle model dynamic identification method used in intelligent transportation system |
| CN103530640A (en) * | 2013-11-07 | 2014-01-22 | 沈阳聚德视频技术有限公司 | Unlicensed vehicle detection method based on AdaBoost and SVM (support vector machine) |
| CN104951784A (en) * | 2015-06-03 | 2015-09-30 | 杨英仓 | Method of detecting absence and coverage of license plate in real time |
| CN105354533A (en) * | 2015-09-28 | 2016-02-24 | 江南大学 | Bag-of-word model based vehicle type identification method for unlicensed vehicle at gate |
| CN105488099A (en) * | 2015-11-03 | 2016-04-13 | 杭州全实鹰科技有限公司 | Vehicle retrieval method based on similarity learning |
| CN105512662A (en) * | 2015-06-12 | 2016-04-20 | 北京卓视智通科技有限责任公司 | Detection method and apparatus for unlicensed vehicle |
-
2016
- 2016-12-15 CN CN201611156309.5A patent/CN106845341B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103000028A (en) * | 2011-09-14 | 2013-03-27 | 上海宝康电子控制工程有限公司 | Vehicle registration plate recognition system and method |
| CN102902957A (en) * | 2012-09-05 | 2013-01-30 | 佳都新太科技股份有限公司 | Video-stream-based automatic license plate recognition method |
| CN103258213A (en) * | 2013-04-22 | 2013-08-21 | 中国石油大学(华东) | Vehicle model dynamic identification method used in intelligent transportation system |
| CN103530640A (en) * | 2013-11-07 | 2014-01-22 | 沈阳聚德视频技术有限公司 | Unlicensed vehicle detection method based on AdaBoost and SVM (support vector machine) |
| CN104951784A (en) * | 2015-06-03 | 2015-09-30 | 杨英仓 | Method of detecting absence and coverage of license plate in real time |
| CN105512662A (en) * | 2015-06-12 | 2016-04-20 | 北京卓视智通科技有限责任公司 | Detection method and apparatus for unlicensed vehicle |
| CN105354533A (en) * | 2015-09-28 | 2016-02-24 | 江南大学 | Bag-of-word model based vehicle type identification method for unlicensed vehicle at gate |
| CN105488099A (en) * | 2015-11-03 | 2016-04-13 | 杭州全实鹰科技有限公司 | Vehicle retrieval method based on similarity learning |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107748895A (en) * | 2017-10-29 | 2018-03-02 | 北京工业大学 | UAV Landing landforms image classification method based on DCT CNN models |
| CN107748895B (en) * | 2017-10-29 | 2021-06-25 | 北京工业大学 | Unmanned aerial vehicle landing landform image classification method based on DCT-CNN model |
| CN110555125A (en) * | 2018-05-14 | 2019-12-10 | 桂林远望智能通信科技有限公司 | Vehicle retrieval method based on local features |
| CN108960412A (en) * | 2018-06-29 | 2018-12-07 | 北京京东尚科信息技术有限公司 | Image-recognizing method, device and computer readable storage medium |
| CN109190649A (en) * | 2018-07-02 | 2019-01-11 | 北京陌上花科技有限公司 | A kind of optimization method and device of deep learning network model server |
| CN109190649B (en) * | 2018-07-02 | 2021-10-01 | 北京陌上花科技有限公司 | Optimization method and device for deep learning network model server |
| CN109241349A (en) * | 2018-08-14 | 2019-01-18 | 中国电子科技集团公司第三十八研究所 | A kind of monitor video multiple target classification retrieving method and system based on deep learning |
| CN109241349B (en) * | 2018-08-14 | 2022-03-25 | 中国电子科技集团公司第三十八研究所 | Monitoring video multi-target classification retrieval method and system based on deep learning |
| CN109992690A (en) * | 2019-03-11 | 2019-07-09 | 中国华戎科技集团有限公司 | A kind of image search method and system |
| CN110490272A (en) * | 2019-09-05 | 2019-11-22 | 腾讯音乐娱乐科技（深圳）有限公司 | Image content similarity analysis method, apparatus and storage medium |
| CN110490272B (en) * | 2019-09-05 | 2022-10-21 | 腾讯音乐娱乐科技（深圳）有限公司 | Image content similarity analysis method and device and storage medium |
| CN111007761A (en) * | 2019-11-28 | 2020-04-14 | 上海蓝色帛缔智能工程有限公司 | Automatic monitoring and management system of data center |
| CN113157641A (en) * | 2021-02-07 | 2021-07-23 | 北京卓视智通科技有限责任公司 | Unlicensed vehicle archiving and querying method, device, system, equipment and storage medium |
| CN113157641B (en) * | 2021-02-07 | 2023-07-04 | 北京卓视智通科技有限责任公司 | Method, device, system, equipment and storage medium for archiving and inquiring non-license vehicle |
| CN112990136A (en) * | 2021-04-29 | 2021-06-18 | 成都深蓝思维信息技术有限公司 | Target detection method and device |
| CN113706390A (en) * | 2021-10-29 | 2021-11-26 | 苏州浪潮智能科技有限公司 | Image conversion model training method, image conversion method, device and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106845341B (en) | 2020-04-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106845341A (en) | A kind of unlicensed vehicle identification method based on virtual number plate | |
| Spencer Jr et al. | Advances in computer vision-based civil infrastructure inspection and monitoring | |
| US10607362B2 (en) | Remote determination of containers in geographical region | |
| CN110363104B (en) | Detection method of diesel black smoke vehicle | |
| Zorzi et al. | Polyworld: Polygonal building extraction with graph neural networks in satellite images | |
| Zhou et al. | Automated residential building detection from airborne LiDAR data with deep neural networks | |
| CN108875600A (en) | A kind of information of vehicles detection and tracking method, apparatus and computer storage medium based on YOLO | |
| CN106446150B (en) | A kind of method and device of vehicle precise search | |
| CN110599537A (en) | Mask R-CNN-based unmanned aerial vehicle image building area calculation method and system | |
| CN111626295B (en) | Training method and device for license plate detection model | |
| Shang et al. | Facilitating efficient mars terrain image classification with fuzzy-rough feature selection | |
| CN113449784B (en) | Image multi-classification method, device, equipment and medium based on priori attribute map | |
| EP3553700A2 (en) | Remote determination of containers in geographical region | |
| CN108734200A (en) | Human body target visible detection method and device based on BING features | |
| CN111639697B (en) | Hyperspectral image classification method based on non-repeated sampling and prototype network | |
| Elmannai et al. | Classification using semantic feature and machine learning: Land-use case application | |
| Chawla et al. | Classification of computer generated images from photographic images using convolutional neural networks | |
| CN112613354A (en) | Heterogeneous remote sensing image change detection method based on sparse noise reduction self-encoder | |
| Guili et al. | A man-made object detection algorithm based on contour complexity evaluation | |
| CN115601684A (en) | Emergency early warning method and device, electronic equipment and storage medium | |
| Rodríguez-Cuenca et al. | A spatial contextual postclassification method for preserving linear objects in multispectral imagery | |
| CN114783042A (en) | Face recognition method, device, equipment and storage medium based on multiple moving targets | |
| Altaei et al. | Satellite image classification using multi features based descriptors | |
| JP4963824B2 (en) | Change area extraction apparatus and change area extraction method | |
| Sun et al. | Contextual models for automatic building extraction in high resolution remote sensing image using object-based boosting method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |