CN104361343B - Vehicle type recognition method and its device - Google Patents
Vehicle type recognition method and its device Download PDFInfo
- Publication number
- CN104361343B CN104361343B CN201410604350.9A CN201410604350A CN104361343B CN 104361343 B CN104361343 B CN 104361343B CN 201410604350 A CN201410604350 A CN 201410604350A CN 104361343 B CN104361343 B CN 104361343B
- Authority
- CN
- China
- Prior art keywords
- vehicle
- image
- feature
- logo
- face
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of vehicle type recognition method and its device based on video image analysis, the method includes：Collection vehicle image；It identifies area-of-interest, and extracts logo and vehicle face feature；Type of vehicle is sorted out according to the logo and vehicle face feature.The present invention is combined by identifying vehicle face with two kinds of technologies of vehicle-logo recognition, can improve the accuracy of vehicle type recognition.
Description
Technical field
The present invention relates to image recognition and traffic safety technology fields, in particular to one kind based on video image point
The vehicle type recognition method and its device of analysis.
Background technology
It is usually tested the speed either to be captured to vehicle and be needed using hardware such as velocity radar or ground induction coils, but these
Method can not provide intuitive traffic scene information.With ITS (Intelligent Transport System, intelligent transportation
System) technology constantly brings forth new ideas and develops, and the monitoring system based on video sensor has obtained significant progress, specifically
Ground, the vehicle cab recognition based on video image have become a critically important research branch, can be more by video sensor
Intuitively to obtain the information in traffic scene.
The existing vehicle cab recognition based on video image can only by identifying the size of vehicle, and by vehicle identification be it is big,
Medium and small vehicle.Or in certain existing technical solutions, vehicle is classified as Audi, masses, mark only by vehicle-logo recognition
The serial brand vehicle such as cause, the sub-brand name under various brands type can not be finely divided, to vehicle classification accuracy and
Availability is relatively low.Vehicle-logo recognition technology refers to using digital picture or video signal flow as research object, by image procossing and certainly
Dynamic recognition methods, obtains a kind of practical technique of motor vehicles brand message.Vehicle-logo recognition technology is as in intelligent transportation system
An important link, application be increasingly valued by people, application field is related to highway toll, vehicle management, public affairs
It deploys to ensure effective monitoring and control of illegal activities on road.Such as University Of Qingdao's journal (engineering technology version) discloses a kind of vehicle-logo location algorithm based on Laws template convolutions
【The 3rd phase of volume 27, in September, 2012】, logo coarse positioning is carried out according to the priori of car plate and logo, further according to logo week
The texture information of radiator-grid is enclosed, and removes the interference of radiator-grid around logo using the filtering of Laws masterplates, realizes the positioning of logo.
Or in certain existing other technologies schemes, vehicle is sorted out by the identification of vehicle face, but it is equally existed
The relatively low problem of accuracy.Such as the patent document that China Patent Publication No. is CN102411710A discloses one kind and being based on vehicle face
The vehicle type recognition method of feature comprising following steps：(1) it by monitor camera collection vehicle image, and carries out pre-
Processing, then detects and is partitioned into the vehicle face image that can characterize type of vehicle, specifically comprise the following steps：(1-1) uses monitoring
The vehicle image of camera acquisition various；(1-2) image preprocessing enhances picture quality using homomorphic filtering；(1-3) is adopted
With based on the vehicle face region detection of car plate location information and segmentation, being partitioned into the vehicle face region that can characterize vehicle feature；(2) to vehicle
Image carries out Curvelet wavelet transformations, and the vehicle face eigenmatrix of vehicle feature can be characterized with extraction；(3) supporting vector is used
Machine grader classifies to the Curvelet wavelet-based attribute vectors of the vehicle face image of extraction, to identify vehicle.
Invention content
In order to which specific type of vehicle is recognized accurately, being designed to provide for the embodiment of the present invention is a kind of based on video figure
As the vehicle type recognition method and its device of analysis.
The embodiment of the present invention is realized using following technical scheme：
A kind of vehicle type recognition method, including：
Collection vehicle image；
It identifies area-of-interest, and extracts logo and vehicle face feature；
Type of vehicle is sorted out according to the logo and vehicle face feature.
Preferably, the step of extraction vehicle face feature includes：
Into driving face positioning in the area-of-interest；
Image after being positioned to vehicle face carries out eliminating illumination effect pretreatment；
Extract the textural characteristics and edge feature of vehicle face respectively from through the elimination pretreated image of illumination effect,
And after textural characteristics to vehicle face and edge feature are combined, vehicle face feature is formed.
Preferably, from the method through eliminating the textural characteristics for extracting vehicle face in the pretreated image of illumination effect respectively
For：
To carrying out size normalization processing through eliminating the pretreated image of illumination effect, extraction Gabor wavelet feature is made
For the textural characteristics of vehicle face.
Preferably, include into the step of driving face positioning in the area-of-interest：
License Plate is carried out in the area-of-interest, and coarse positioning is carried out to vehicle face；
Target to the image data in the channels yuv format vehicle face image extraction Y after coarse positioning as image procossing, meter
Calculate vehicle face image gradient map；
Vehicle face image gradient map is projected to vertical direction, to determine the up-and-down boundary of vehicle face, and by vehicle face image
Gradient map is projected to horizontal direction, to determine the right boundary of vehicle face.
Preferably, the step of extracting vehicle face feature further includes carrying out principal component analysis PCA to obtained vehicle face feature
The processing step of (Principal Component Analysis, principal component analysis) dimensionality reduction：
The covariance matrix for calculating vehicle face feature samples, calculates the characteristic value and feature vector of the covariance matrix, and right
Characteristic value and feature vector carry out orthonomalization processing；
Descending to characteristic value to be ranked up, then the corresponding feature vector of k characteristic value and composition matrix before taking will
Characteristic value after sequence and the matrix multiple, obtain the compression vehicle face feature of k dimensions, wherein k >=1.
Preferably, the step of extraction logo feature includes：
Vehicle-logo location is carried out in the area-of-interest；
Size normalization and histogram equalization processing carried out to the image after vehicle-logo location, and to treated image
Feature extraction is carried out to logo using improved edge feature, wherein operation is weighted with prominent to the characteristic value among image
Go out the edge feature among logo, realizes the improvement to edge feature.
Preferably, include the step of progress vehicle-logo location in the area-of-interest：
Coarse positioning is carried out to logo in the area-of-interest；
Using the filter group based on texture to carrying out horizontal direction filtering to the logo image after coarse positioning, use later
Medium filtering carries out denoising to image, is carrying out binary conversion treatment to image and after obtaining binary image, to image into
The processing of row morphology opening operation, and the image after calculation process is projected on vertical direction, longest company after being projected
Continuous region and the up-and-down boundary as logo；
Horizontal filtering and vertical filtering are carried out respectively to the logo image after coarse positioning, use medium filtering to image later
Denoising is carried out, after carrying out binaryzation and the processing of morphology opening operation to image, the image after calculation process is carried out
Connected component labeling obtains horizontal filtering treated the picture fv (x, y) after picture fh (x, y) and vertical filtering, compares picture
The quantity of fh (x, y) and picture fv (x, y) white pixel, and the picture of white pixel negligible amounts is thrown to horizontal direction
Shadow, to determine the right boundary of logo.
Preferably, the step of extraction logo feature further includes：
PCA dimension-reduction treatment is carried out to obtained logo feature.
Preferably, the step of sorting out to type of vehicle according to the logo and vehicle face feature include：
Classification is trained to vehicle face feature and logo feature respectively using support vector machines, to respectively obtain
One classification results and the second classification results；
First classification results and the second classification results are matched, if including in the first classification results
When the type of vehicle for the logo that two classification results are sorted out, select the type of vehicle as type of vehicle categorization results；Otherwise, it selects
The most similar categorization results are selected in multiple first classification results as type of vehicle categorization results.
A kind of vehicle type recognition device, including：
Collecting unit is used for collection vehicle image；
Area-of-interest recognition unit, goes out area-of-interest for identification；
Feature extraction unit, for extracting logo and vehicle face feature；
Vehicle type recognition unit, for sorting out to type of vehicle according to the logo and vehicle face feature.
The present invention is combined by identifying vehicle face with two kinds of technologies of vehicle-logo recognition, can improve the standard of vehicle type recognition
True property.
Description of the drawings
Fig. 1 is vehicle type recognition method flow schematic diagram provided in an embodiment of the present invention；
Fig. 2 is vehicle face positioning flow schematic diagram provided in an embodiment of the present invention；
Fig. 3 is vehicle face feature recognition flow diagram provided in an embodiment of the present invention；
Fig. 4 is vehicle-logo location flow diagram provided in an embodiment of the present invention.
The realization, functional characteristics and excellent effect of the object of the invention, below in conjunction with specific embodiment and attached drawing do into
The explanation of one step.
Specific implementation mode
Technical solution of the present invention is described in further detail in the following with reference to the drawings and specific embodiments, so that this
The technical staff in field can be better understood from the present invention and can be practiced, but illustrated embodiment is not as the limit to the present invention
It is fixed.
As shown in Figure 1, an embodiment of the present invention provides a kind of vehicle type recognition method, include the following steps：
S10, collection vehicle image；
S20, it identifies area-of-interest, and extracts logo and vehicle face feature；
S30, type of vehicle is sorted out according to the logo and vehicle face feature.
In the present invention, vehicle image is acquired by front end camera first, then detects that needs identify
ROI (Region Of Interest, area-of-interest) region, the logo to vehicle and the areas Che Lian using edge and textural characteristics
Domain carries out feature extraction, finally utilizes support vector machines by graphic collection to such as 2006 sections of Bora, BMW X6, Magotan 2008
200 kinds of vehicle style classifications under 32 brands such as money, 2012 sections of Magotan improve the accuracy of type of vehicle classification with this.
In the present embodiment, extract vehicle face feature the step of include：
S201, in the area-of-interest into driving face positioning；
S202, the image after the positioning of vehicle face is carried out to eliminate illumination effect pretreatment；
S203, the textural characteristics and edge for extracting vehicle face respectively from through the elimination pretreated image of illumination effect
Feature, and after textural characteristics to vehicle face and edge feature are combined, form vehicle face feature.
Preferably, in the step S203, vehicle is extracted respectively from through eliminating in the pretreated image of illumination effect
The method of the textural characteristics of face is：To carrying out size normalization processing, extraction through eliminating the pretreated image of illumination effect
Textural characteristics of the Gabor wavelet feature as vehicle face.
Preferably, in the step S201, include into the step of driving face positioning in the area-of-interest：
S2011, License Plate is carried out in the area-of-interest, and coarse positioning is carried out to vehicle face；
S2012, the image data in the channels Y is extracted to the yuv format vehicle face image after coarse positioning as image procossing
Target calculates vehicle face image gradient map；
S2013, vehicle face image gradient map is projected to vertical direction, to determine the up-and-down boundary of vehicle face, and by vehicle
Face image gradient map is projected to horizontal direction, to determine the right boundary of vehicle face.
In the specific implementation mode of the present invention, as shown in Fig. 2, the step of executing the positioning of vehicle face includes：
1) License Plate is carried out to collected vehicle image, coarse positioning then is carried out to vehicle face.
2) vehicle face is accurately positioned in the vehicle face image of coarse positioning, since collected data are from video camera
The image data of yuv format, due to can only need gray level image, so only extracting the image data conduct in the channels Y herein
The target of image procossing calculates vehicle face image gradient map.
3) gradient map is projected to vertical direction, gradient image is separated among the image, each abscissa and its
Value in symmetric coordinates be the projection with its symmetric position projection gray level and minimum value, the gray scale for calculating projected image is average
Value, traverses all projections, finds first and the last one is more than average value intersection point, then the part between the two points is vehicle
The height component of face.
4) after the height of vehicle face determines, it is thus necessary to determine that the width segments of vehicle face.Specific method is：By gradient map to level
Direction projection, it is similar with the method for vehicle face height is determined, determine the right boundary of vehicle face, the program flow diagram of vehicle face positioning.
Preferably, in the present embodiment, extract vehicle face feature the step of further include to obtained vehicle face feature carry out it is main at
The processing step of analysis PCA dimensionality reductions：
S204, the covariance matrix for calculating vehicle face feature samples, calculate the characteristic value and feature vector of the covariance matrix,
And orthonomalization processing is carried out to characteristic value and feature vector；
S205, it is descending to characteristic value be ranked up, the corresponding feature vector of k characteristic value and composition matrix before taking,
Then by the characteristic value and the matrix multiple after sequence, the compression vehicle face feature of k dimensions is obtained, wherein k >=1.
In the specific implementation mode of the present invention, as shown in figure 3, the step of executing the identification of vehicle face includes：
1) image preprocessing is carried out to the picture f (x, y) after positioning, eliminates influence of the illumination to image.Using histogram
The method of equalization pre-processes image, and expression formula is as follows：
Wherein, T (k) is the grey scale mapping value corresponding to gray level k, and n is f (x, y) image pixel summation, and ni is gray level
For the pixel summation of i.
2) size normalization is carried out to the image of input, extracts Gabor wavelet feature.Gabor characteristic can be very good to retouch
The textural characteristics of image are stated, the expression formula of Two-Dimensional Gabor Wavelets is as follows：
X'=x cos θ+y sin θs；
Y'=-x sin θ+y cos θ；
Wherein, λ indicates that small echo length, θ indicate the direction of small echo,Indicate phase offset, σ is standard deviation, and γ is space
Than.In order to improve the calculating speed of convolution, by Fast Fourier Transform (FFT), image is transformed into frequency domain from image area,
It is equivalent to the product in frequency domain as convolution in the spatial domain, after making product in frequency domain, needs to pass through fast Flourier
Inverse transformation transforms to result in spatial domain.A series of images can be generated by different small echo length and small echo direction, it will
It is divided into several regions per piece image, counts the average value and variance of each area grayscale, forms Gabor characteristic.
3) in order to preferably describe vehicle face feature, herein in addition to the textural characteristics of extraction vehicle face, also it is extracted the side of vehicle face
Texture and edge feature, are then combined by edge feature, generate new vehicle face feature, wherein combination textural characteristics and side
The known technology that edge feature and the technological means for forming new vehicle face feature are grasped by those skilled in the art.
4) in order to reduce calculation amount, need to feature carry out PCA (Principal Component Analysis, it is main at
Analysis) dimensionality reduction.It is as follows：The covariance matrix for calculating sample first, calculates the characteristic value and feature of the matrix
Vector, and orthonomalization processing is carried out to characteristic value and feature vector, it is then descending to characteristic value to be ranked up, before taking
The corresponding feature vector of k characteristic value forms matrix, and then assemblage characteristic and the matrix multiple, obtain the compressive features of k dimensions.
In the present embodiment, it is preferable that extraction logo feature the step of include：
S206, vehicle-logo location is carried out in the area-of-interest；
S207, size normalization and histogram equalization processing are carried out to the image after vehicle-logo location, and to processing after
Image using improved edge feature to logo carry out feature extraction, wherein behaviour is weighted to the characteristic value among image
Make, to protrude the edge feature among logo, to realize the improvement to edge feature.
Preferably for the step S206, the step of vehicle-logo location is carried out in the area-of-interest, includes：
S2061, coarse positioning is carried out to logo in the area-of-interest；
S2062, using the filter group based on texture to after coarse positioning logo image carry out horizontal direction filtering, it
Medium filtering is used to carry out denoising to image afterwards, it is right after carrying out binary conversion treatment to image and obtaining binary image
Image carries out morphology opening operation processing, and the image after calculation process is projected on vertical direction, after being projected most
Long continuum and the up-and-down boundary as logo；
S2063, horizontal filtering and vertical filtering are carried out respectively to the logo image after coarse positioning, uses medium filtering later
Denoising is carried out to image, after carrying out binaryzation and the processing of morphology opening operation to image, to the figure after calculation process
As carrying out connected component labeling, horizontal filtering is obtained treated the picture fv (x, y) after picture fh (x, y) and vertical filtering, than
Compared with the quantity of picture fh (x, y) and picture fv (x, y) white pixel, and by the picture of white pixel negligible amounts to horizontal direction
It projects, to determine the right boundary of logo.
According to the specific implementation mode of the present invention, as shown in figure 4, the step of executing vehicle-logo location includes：
1) in order to pick out final result from the similar vehicle face result of top n, vehicle mark is needed to be assisted in identifying.
Logo is positioned firstly the need of from the vehicle face picture after positioning, naturally it is also possible into driving directly from area-of-interest
Demarcate position.In order to reduce detection range, coarse positioning is carried out to logo, i.e., the up-and-down boundary of logo coarse positioning and vehicle face picture is upper
Lower boundary is identical, and right boundary is the right boundary of car plate.
2) in order to obtain the up-and-down boundary of logo, use a kind of filter group based on texture to the vehicle after coarse positioning herein
Piece of marking on a map carries out horizontal direction filtering, then medium filtering is used to carry out denoising to image, is carried out at binaryzation to image
Reason after obtaining binary image, carries out morphology opening operation.By treated, image is projected on vertical direction, and statistics is every
The number of pixels that one-line pixel value is 255 then looks for longest continuum after projection, which is the height portion of image
Point.
3) determine that logo right boundary is similar with the method for logo up-and-down boundary is determined, since the background texture of logo may
It is horizontal texture it could also be possible that vertical or reticular texture, so needing to carry out horizontal filtering and vertical filtering respectively, in
Value filtering carries out denoising, carries out binaryzation to image and morphology opening operation is handled.Connected domain is carried out to the image after processing
Label, filters out the small connected domain that area is crossed, and obtains horizontal filtering treated the figure after picture fh (x, y) and vertical filtering
Piece fv (x, y) compares fh (x, y) and fv (x, y) white pixel quantity, chooses the few picture of white pixel quantity alternatively most
Terminate fruit.
4) it by the picture after selection, is projected to horizontal direction, the characteristics of according to logo symmetry, logo is to horizontal direction
After projecting, if there is pixel in the left survey somewhere of logo axis, pixel is also will appear at its symmetric coordinates, it is no
Then it is believed that the position is not the boundary position of logo.The specific practice is as follows：In order to ensure calculating speed, by picture centre position
As the position of logo symmetry axis, x-axis is traversed from left to right up to centre position, if finding to contain at this on the axis left side white
Colour vegetarian refreshments, then finding whether also contain white pixel point around its symmetrical place, if containing white pixel point, explanation has
Symmetry, then right boundary determines, if do not contained, continues traversal and finds until axis..
Preferably for the step S206, the step of extracting logo feature, further includes：
S2064, PCA dimension-reduction treatment is carried out to obtained logo feature.
Specifically, according to the specific implementation mode of the present invention, the specific steps for executing vehicle-logo recognition include：
1) picture after positioning needs to carry out size normalization and histogram equalization, eliminates the extraneous factors such as illumination
It influences, feature extraction is carried out to logo using improved edge feature and is identified.It makes discovery from observation, the main side of logo
Edge feature concentrates on the middle section of logo, thus using weighting by the way of edge feature is improved, i.e., to image among
Characteristic value be weighted operation, to protrude the edge feature among logo.
2) PCA dimensionality reductions are carried out to logo feature.
Preferably, in the step S30, the step of classification to type of vehicle according to the logo and vehicle face feature
Including：
S301, classification is trained to vehicle face feature and logo feature respectively using support vector machines, to obtain respectively
To the first classification results and the second classification results；
S302, first classification results and the second classification results are matched, if wrapped in the first classification results
When the type of vehicle for the logo sorted out containing the second classification results, select the type of vehicle as type of vehicle categorization results；
Otherwise, select in multiple first classification results the most similar categorization results as type of vehicle categorization results.
Specifically, it in one embodiment of the present invention, executes logo and is matched with vehicle face to accurately determine type of vehicle
Step includes：
1) classification is trained to compressed vehicle face feature using support vector machines (SVM), can directly known after classification
Style type belonging to other vehicle, such as public 2006 sections of Bora, 2007 sections of Toyota's Corolla, Honda CRV2012 moneys.SVM is
A kind of machine learning method based on statistical theory, this method can be very good to solve small sample training problem.Classification is main
Depending on the selection of SVM kernel function K (x, y), classified herein to sample using RBF kernel functions, expression is：
Wherein x is characterized, and y is to calculate the feature vector generated, and g is threshold value.Since there may be vehicle faces under different brands
The very much like situation of feature needs to return the most like vehicle face of top n in case most to improve the accuracy rate of identification after classification
It is compared afterwards with the result of vehicle-logo recognition, returns to final result.
2) use support vector machines (SVM) to carry out PCA dimensionality reductions logo feature be supported vector machine (SVM) training and
Classification, obtains final classification results.
3) obtained logo classification results are matched with the most like top n result that vehicle face is classified, if
Contain result corresponding with vehicle-logo recognition in the top n result of vehicle face identification：As vehicle-logo recognition result is：Masses, and vehicle face
Recognition result is：2007 sections of Toyota's Corolla, public 2007 sections of Passat, Honda CRV2012 moneys, then final type of vehicle point
Class result is：2007 sections of public Passat.If it is matched not with logo as a result, if final result be vehicle face identify top n
As a result most like vehicle in.
The embodiment of the present invention has also correspondingly provided a kind of vehicle type recognition device, including：
Collecting unit is used for collection vehicle image；
Area-of-interest recognition unit, goes out area-of-interest for identification；
Feature extraction unit, for extracting logo and vehicle face feature；
Vehicle type recognition unit, for sorting out to type of vehicle according to the logo and vehicle face feature.
The vehicle type recognition device can be based on the vehicle type recognition method that above-described embodiment provides and to numerous vapour
Specific vehicle below vehicle brand carries out effective and accurate Classification and Identification, can be with specific reference to upper for its specific implementation process
Described in text.
The foregoing is merely the preferred embodiment of the present invention, are not intended to limit the scope of the invention, every utilization
Equivalent structure or equivalent flow shift made by description of the invention and accompanying drawing content is applied directly or indirectly in other correlations
Technical field, be included within the scope of the present invention.
Claims (8)
1. a kind of vehicle type recognition method, which is characterized in that including：
Collection vehicle image；
It identifies area-of-interest, and extracts logo and vehicle face feature；
The step of extraction vehicle face feature includes：
Into driving face positioning in the area-of-interest；
Image after being positioned to vehicle face carries out eliminating illumination effect pretreatment；
Extract the textural characteristics and edge feature of vehicle face respectively from through the elimination pretreated image of illumination effect, and right
After the textural characteristics and edge feature of vehicle face are combined, vehicle face feature is formed；
The step of extraction logo feature includes：
Vehicle-logo location is carried out in the area-of-interest；
Size normalization and histogram equalization processing are carried out to the image after vehicle-logo location, and image uses to treated
Improved edge feature carries out feature extraction to logo, wherein is weighted operation to the characteristic value among image to protrude vehicle
Intermediate edge feature is marked, realizes the improvement to edge feature；
Type of vehicle is sorted out according to the logo and vehicle face feature, by graphic collection to vehicle style classification.
2. vehicle type recognition method as described in claim 1, which is characterized in that from through eliminate illumination effect it is pretreated
The method for extracting the textural characteristics of vehicle face in image respectively is：
To carrying out size normalization processing through eliminating the pretreated image of illumination effect, extraction Gabor wavelet feature is as vehicle
The textural characteristics of face.
3. vehicle type recognition method as described in claim 1, which is characterized in that into driving face in the area-of-interest
The step of positioning includes：
License Plate is carried out in the area-of-interest, and coarse positioning is carried out to vehicle face；
Target to the image data in the channels yuv format vehicle face image extraction Y after coarse positioning as image procossing, calculates vehicle
Face image gradient map；
Vehicle face image gradient map is projected to vertical direction, to determine the up-and-down boundary of vehicle face, and by vehicle face image gradient
Figure is projected to horizontal direction, to determine the right boundary of vehicle face.
4. vehicle type recognition method as described in claim 1, which is characterized in that further include being carried out to obtained vehicle face feature
The processing step of principal component analysis PCA dimensionality reductions：
The covariance matrix for calculating vehicle face feature samples, calculates the characteristic value and feature vector of the covariance matrix, and to feature
Value and feature vector carry out orthonomalization processing；
Descending to characteristic value to be ranked up, the corresponding feature vector of k characteristic value and composition matrix before taking then will sequences
Characteristic value afterwards and the matrix multiple, obtain the compression vehicle face feature of k dimensions, wherein k >=1.
5. vehicle type recognition method as described in claim 1, which is characterized in that carry out logo in the area-of-interest
The step of positioning includes：
Coarse positioning is carried out to logo in the area-of-interest；
Using the filter group based on texture to carrying out horizontal direction filtering to the logo image after coarse positioning, intermediate value is used later
Filtering carries out denoising to image, and after carrying out binary conversion treatment to image and obtaining binary image, shape is carried out to image
The processing of state opening operation, and the image after calculation process is projected on vertical direction, longest continuum after being projected
Domain and up-and-down boundary as logo；
Horizontal filtering and vertical filtering are carried out respectively to the logo image after coarse positioning, use medium filtering to carry out image later
Denoising is connected to the image after calculation process after carrying out binaryzation and the processing of morphology opening operation to image
Field mark obtains horizontal filtering treated the picture fv (x, y) after picture fh (x, y) and vertical filtering, compare picture fh (x,
Y) and the quantity of picture fv (x, y) white pixel, and the picture of white pixel negligible amounts is projected to horizontal direction, with
Determine the right boundary of logo.
6. vehicle type recognition method as described in claim 1, which is characterized in that further include：
PCA dimension-reduction treatment is carried out to obtained logo feature.
7. vehicle type recognition method as described in claim 1, which is characterized in that according to the logo and vehicle face feature to vehicle
The step of type is sorted out include：
Classification is trained to vehicle face feature and logo feature respectively using support vector machines, to respectively obtain first point
Class result and the second classification results；
First classification results and the second classification results are matched, if in the first classification results including second point
When the type of vehicle for the logo that class result is sorted out, select the type of vehicle as type of vehicle categorization results；Otherwise, it selects more
The most similar categorization results are as type of vehicle categorization results in a first classification results.
8. a kind of vehicle type recognition device, which is characterized in that including：
Collecting unit is used for collection vehicle image；
Area-of-interest recognition unit, goes out area-of-interest for identification；
Feature extraction unit, for extracting logo and vehicle face feature；
The feature extraction unit is specifically used for：
Into driving face positioning in the area-of-interest；
Image after being positioned to vehicle face carries out eliminating illumination effect pretreatment；
Extract the textural characteristics and edge feature of vehicle face respectively from through the elimination pretreated image of illumination effect, and right
After the textural characteristics and edge feature of vehicle face are combined, vehicle face feature is formed；
The feature extraction unit is specifically additionally operable to：
Vehicle-logo location is carried out in the area-of-interest；
Size normalization and histogram equalization processing are carried out to the image after vehicle-logo location, and image uses to treated
Improved edge feature carries out feature extraction to logo, wherein is weighted operation to the characteristic value among image to protrude vehicle
Intermediate edge feature is marked, realizes the improvement to edge feature；
Vehicle type recognition unit, for sorting out to type of vehicle according to the logo and vehicle face feature, by graphic collection
To vehicle style classification.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410604350.9A CN104361343B (en) | 2014-10-30 | 2014-10-30 | Vehicle type recognition method and its device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410604350.9A CN104361343B (en) | 2014-10-30 | 2014-10-30 | Vehicle type recognition method and its device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104361343A CN104361343A (en) | 2015-02-18 |
| CN104361343B true CN104361343B (en) | 2018-08-31 |
Family
ID=52528601
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410604350.9A Active CN104361343B (en) | 2014-10-30 | 2014-10-30 | Vehicle type recognition method and its device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104361343B (en) |
Families Citing this family (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104680149B (en) * | 2015-03-10 | 2018-07-03 | 苏州科达科技股份有限公司 | A kind of object type recognition methods and system |
| CN104820831A (en) * | 2015-05-13 | 2015-08-05 | 沈阳聚德视频技术有限公司 | Front vehicle face identification method based on AdaBoost license plate location |
| CN106650550A (en) * | 2015-10-28 | 2017-05-10 | 中通服公众信息产业股份有限公司 | Vehicle model identification method and vehicle model identification system through image characteristics of vehicle mark and vehicle head |
| CN105551046B (en) * | 2015-12-17 | 2019-04-30 | 浙江宇视科技有限公司 | Vehicle face-positioning method and device |
| CN105652332B (en) * | 2016-02-24 | 2019-02-12 | 北京君和信达科技有限公司 | A kind of method for controlling radiation source and fast general formula safe examination system |
| CN107451511B (en) * | 2016-05-31 | 2020-12-15 | 优信拍（北京）信息科技有限公司 | System and method for extracting OBD data based on vehicle type information |
| CN106529471B (en) * | 2016-11-08 | 2019-08-23 | 广东安居宝数码科技股份有限公司 | Vehicle face-positioning method and system |
| CN106557579B (en) * | 2016-11-28 | 2020-08-25 | 中通服公众信息产业股份有限公司 | Vehicle model retrieval system and method based on convolutional neural network |
| CN106650660A (en) * | 2016-12-19 | 2017-05-10 | 深圳市华尊科技股份有限公司 | Vehicle type recognition method and terminal |
| CN107301388A (en) * | 2017-06-16 | 2017-10-27 | 重庆交通大学 | A kind of automatic vehicle identification method and device |
| CN108090429B (en) * | 2017-12-08 | 2020-07-24 | 浙江捷尚视觉科技股份有限公司 | Vehicle type recognition method for graded front face bayonet |
| CN108254748A (en) * | 2018-01-15 | 2018-07-06 | 武汉理工大学 | Inland navigation craft superelevation alarm system and method based on laser ranging and radar image |
| CN110889867B (en) * | 2018-09-10 | 2022-11-04 | 浙江宇视科技有限公司 | Method and device for detecting damaged degree of car face |
| CN111723601A (en) * | 2019-03-19 | 2020-09-29 | 杭州海康威视数字技术股份有限公司 | Image processing method and device |
| CN110135318B (en) * | 2019-05-08 | 2023-05-12 | 佳都科技集团股份有限公司 | Method, device, equipment and storage medium for determining passing record |
| CN111222409A (en) * | 2019-11-26 | 2020-06-02 | 北京迈格威科技有限公司 | Vehicle brand labeling method, device and system |
| CN111062396B (en) * | 2019-11-29 | 2022-03-25 | 深圳云天励飞技术有限公司 | License plate number recognition method and device, electronic equipment and storage medium |
| CN111199228B (en) * | 2019-12-26 | 2023-03-28 | 深圳市芊熠智能硬件有限公司 | License plate positioning method and device |
| CN112465816B (en) * | 2020-12-17 | 2021-08-06 | 哈尔滨市科佳通用机电股份有限公司 | Method, system and device for detecting damage fault of windshield of railway motor car |
| CN113569927A (en) * | 2021-07-13 | 2021-10-29 | 安徽绿舟科技有限公司 | Man-machine interaction system of unmanned power station |
| CN113963214A (en) * | 2021-10-27 | 2022-01-21 | 广东利通科技投资有限公司 | Vehicle type recognition method, device, system, computer device and storage medium |
| CN115880565B (en) * | 2022-12-06 | 2023-09-05 | 江苏凤火数字科技有限公司 | Neural network-based scraped vehicle identification method and system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101196979A (en) * | 2006-12-22 | 2008-06-11 | 四川川大智胜软件股份有限公司 | Method for recognizing vehicle type by digital picture processing technology |
| CN104112122A (en) * | 2014-07-07 | 2014-10-22 | 叶茂 | Vehicle logo automatic identification method based on traffic video |
-
2014
- 2014-10-30 CN CN201410604350.9A patent/CN104361343B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101196979A (en) * | 2006-12-22 | 2008-06-11 | 四川川大智胜软件股份有限公司 | Method for recognizing vehicle type by digital picture processing technology |
| CN104112122A (en) * | 2014-07-07 | 2014-10-22 | 叶茂 | Vehicle logo automatic identification method based on traffic video |
Non-Patent Citations (4)
| Title |
|---|
| 基于PCA和SVM的车标识别方法研究;童雯;《中国优秀硕士学位论文全文数据库》;20120915(第9期);第1.2.2节第2段以及第3.2.3节 * |
| 车牌检测及汽车类型分类方法研究;姜谊;《中国优秀硕士学位论文全文数据库》;20101015(第10期);第3页倒数第1段 * |
| 车辆检测及汽车类型分类方法研究;姜谊;《中国优秀硕士学位论文全文数据库》;20101015(第10期);第3页倒数第1段 * |
| 运动车辆车标定位的四阶段法研究;周宇;《重庆交通学院学报》;20061231;第25卷(第6期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104361343A (en) | 2015-02-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104361343B (en) | Vehicle type recognition method and its device | |
| CN103198332B (en) | A kind of far infrared vehicle-mounted pedestrian detection method of real-time robust | |
| CN107358258B (en) | SAR image target classification based on NSCT double CNN channels and selective attention mechanism | |
| Huang et al. | Vehicle detection and inter-vehicle distance estimation using single-lens video camera on urban/suburb roads | |
| Derpanis et al. | Classification of traffic video based on a spatiotemporal orientation analysis | |
| Ardianto et al. | Real-time traffic sign recognition using color segmentation and SVM | |
| CN109918971B (en) | Method and device for detecting number of people in monitoring video | |
| Sharma et al. | Indian vehicle license plate extraction and segmentation | |
| CN111915583B (en) | Vehicle and pedestrian detection method based on vehicle-mounted thermal infrared imager in complex scene | |
| CN103544487B (en) | Front truck recognition methods based on monocular vision | |
| CN109635733B (en) | Parking lot and vehicle target detection method based on visual saliency and queue correction | |
| CN107704833A (en) | A kind of front vehicles detection and tracking based on machine learning | |
| Rabee et al. | License plate detection and recognition in complex scenes using mathematical morphology and support vector machines | |
| CN105354547A (en) | Pedestrian detection method in combination of texture and color features | |
| Soumya et al. | License plate detection and character recognition using contour analysis | |
| CN108921147A (en) | A kind of black smoke vehicle recognition methods based on dynamic texture and transform domain space-time characteristic | |
| CN106971402A (en) | A kind of SAR image change detection aided in based on optics | |
| Kanchev et al. | Blurred image regions detection using wavelet-based histograms and SVM | |
| Dalka et al. | Vehicle classification based on soft computing algorithms | |
| Sommer et al. | A comprehensive study on object proposals methods for vehicle detection in aerial images | |
| Chen et al. | Context-aware lane marking detection on urban roads | |
| Tayo et al. | Vehicle license plate recognition using edge detection and neural network | |
| Ktata et al. | License plate localization using Gabor filters and neural networks | |
| Mejía-Iñigo et al. | Color-based texture image segmentation for vehicle detection | |
| Chen et al. | License plate location for vehicles passing through a gate |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information |
Address after: 518000, Shenzhen, Guangdong, Nanshan District Nanshan commercial and cultural center Tianli Central Business Plaza (phase two) C-3005 Applicant after: SHENZHEN UNIHZ TECHNOLOGIES CO., LTD. Address before: 518000, Shenzhen, Guangdong, Nanshan District Nanshan commercial and cultural center Tianli Central Business Plaza (phase two) C-3005 Applicant before: UNIHZ Technologies Corp. |
|
| COR | Change of bibliographic data | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |