Invention content
Based on this, in view of the above technical problems, a kind of automobile logo identification method and its identifying system are provided.
In order to solve the above technical problems, the present invention adopts the following technical scheme that:
A kind of automobile logo identification method, includes the following steps:
Picture pre-processes:Binary conversion treatment is carried out to the video pictures comprising vehicle, obtains bianry image;
Vehicle-logo recognition region coarse positioning:Sobel edge detections are carried out to the bianry image, obtain including vehicle target area
g′(x,y);In the region g ' (x, y), the x for meeting formula (1) is chosenmin,xmax,ymin,ymax:
α is taken, the value of β is 1, and passes through formula (2) coarse positioning vehicle-logo recognition region (x, y, α W, β H):
W is the width in vehicle-logo recognition region, and H is the height in vehicle-logo recognition region, and α, β are respectively width, height parameter,
(x, y) is the coordinate of each pixel in vehicle-logo recognition region;
Vehicle-logo recognition:When identifying logo major class, HOG and PCA is carried out in the vehicle-logo recognition region of α=0.2, β=0.2
The extraction of feature identifies logo major class by logo grader;When identifying logo major class and subclass, in α=1.3, β=0.2
Vehicle-logo recognition region carry out the extractions of HOG and PCA features, pass through logo grader and identify logo major class and subclass;Know
It other logo major class, subclass and prescribes a time limit in year, carrying for HOG and PCA features is carried out in the vehicle-logo recognition region of α=1.3, β=0.6
It takes, logo major class, subclass and the time limit is identified by logo grader.
The logo grader is trained by following steps and is obtained:
Logo samples pictures are pre-processed by the picture pre-treatment step;
Vehicle-logo recognition region is positioned in the logo samples pictures by the vehicle-logo recognition region coarse positioning step;
The extraction of HOG and PCA features is carried out in the vehicle-logo recognition region;
The logo grader is trained with HOG the and PCA features.
This programme further relates to a kind of vehicle-logo recognition system, including:
Picture pretreatment unit obtains bianry image for carrying out binary conversion treatment to the video pictures comprising vehicle;
Vehicle-logo recognition region coarse positioning unit obtains including vehicle for carrying out sobel edge detections to the bianry image
Target area g ' (x, y);In the region g ' (x, y), the x for meeting formula (1) is chosenmin,xmax,ymin,ymax:
α is taken, the value of β is 1, and passes through formula (2) coarse positioning vehicle-logo recognition region (x, y, α W, β H):
W is the width in vehicle-logo recognition region, and H is the height in vehicle-logo recognition region, and α, β are respectively width, height parameter,
(x, y) is the coordinate of each pixel in vehicle-logo recognition region;
For identification when logo major class, HOG is carried out in the vehicle-logo recognition region of α=0.2, β=0.2 for vehicle-logo recognition unit
And the extraction of PCA features, logo major class is identified by logo grader;Identify logo major class and when subclass, in α=1.3,
The vehicle-logo recognition region of β=0.2 carries out the extraction of HOG and PCA features, and logo major class and son are identified by logo grader
Class;It identifies logo major class, subclass and prescribes a time limit in year, HOG and PCA spies are carried out in the vehicle-logo recognition region of α=1.3, β=0.6
The extraction of sign identifies logo major class, subclass and the time limit by logo grader.
The logo grader is trained by following steps and is obtained:
Logo samples pictures are pre-processed by the picture pre-treatment step;
Vehicle-logo recognition region is positioned in the logo samples pictures by the vehicle-logo recognition region coarse positioning step;
The extraction of HOG and PCA features is carried out in the vehicle-logo recognition region;
The logo grader is trained with HOG the and PCA features.
The present invention has the following advantages:
1, identification is accurate:Logo major class can be identified according to identification demand, logo major class and subclass at identification, or
Logo major class, subclass and the time limit at person's identification;
2, identification is comprehensive:It can realize the vehicle-logo recognition of preceding board and realize the vehicle-logo recognition of tail board.
Specific implementation mode
As shown in Figure 1, a kind of automobile logo identification method, includes the following steps:
S110, picture pretreatment:Priori is used for reference, logo is normally above the level of car plate and logo is general
In the vertical axis of symmetry of the front/rear face of vehicle, binary conversion treatment is carried out to the video pictures comprising vehicle, obtains bianry image
g(x,y);
S120, vehicle-logo recognition region coarse positioning:Sobel edge detections are carried out to the bianry image, obtain including logo
Region g ' (x, y);In region g ' (x, y), the x for meeting formula (1) is chosenmin,xmax,ymin,ymax:
α is taken, the value of β is 1, and passes through formula (2) coarse positioning vehicle-logo recognition region (x, y, α W, β H):
W is the width in vehicle-logo recognition region, and H is the height in vehicle-logo recognition region, and α, β are respectively width, height parameter,
(x, y) is the coordinate of each pixel in vehicle-logo recognition region.
S130, vehicle-logo recognition:
It, can be secondary by adjusting width and height parameter α, β according to specific detection depth requirements after the completion of coarse positioning
Vehicle-logo recognition region is positioned, realizes the identification output of the logo of different depth.It is specific as follows:
When identifying logo major class, in α=0.2, the vehicle-logo recognition region of β=0.2 carries out the extraction of HOG and PCA features,
Logo major class is identified by logo grader, such as:Masses, Audi, BMW, Buick etc..
When identifying logo major class and subclass, HOG and PCA spies are carried out in the vehicle-logo recognition region of α=1.3, β=0.2
The extraction of sign identifies logo major class and subclass, such as by logo grader:Masses-Santana, masses-Jetta, BMW-Q5,
Buick-is triumphant more etc..
Identification logo major class, subclass and prescribe a time limit in year, in α=1.3, the vehicle-logo recognition region of β=0.6 carry out HOG and
The extraction of PCA features identifies logo major class, subclass and the time limit, such as by logo grader:Masses-Santana -2006-
2007 sections, Audi's-AL4-2004 moneys etc..
Above-mentioned logo grader is trained by following steps and is obtained:
Logo samples pictures are pre-processed by picture pre-treatment step S110.
Vehicle-logo recognition region is positioned in logo samples pictures by vehicle-logo recognition region coarse positioning step S120.
The extraction of HOG and PCA features is carried out in vehicle-logo recognition region.
Logo grader is trained with HOG and PCA features.
It is found in (the including preceding board and tail board) collection of a large amount of vehicle pictures and analytic process, the vehicle pictures of preceding board
Logo LOGO is relatively apparent, and is respectively positioned on the vertical centre position of detection vehicle, therefore can be directly by being based on logo LOGO extractions
Textural characteristics realize the detection and identification of logo, and the positions logo LOGO of tail board are not fixed, and include the vertical centre of vehicle
Position (above car plate, below car plate), the horizontal position (left or right side) of vehicle, thus it is traditional based on positioning logo LOGO extractions
The method that textural characteristics realize vehicle-logo recognition realizes that difficulty is larger in being identified to the detection of tail board logo.And it is of the invention,
In the case of navigating to logo test position information for the first time, use can be effectively expanded by adjusting width and height parameter α, β
In the logo detection zone of feature extraction, the same of the feature weight of logo differentiation in reduction unification is carried out based on logo LOGO
When, it adds based on multiple characteristic features weights for characterizing logo attribute such as car light, wagon flow line style so that the identification of logo
It is characterized in combination with to multiple characteristic synthetics, the final differentiation realized to logo, therefore the method for discrimination of the logo proposed in the present invention,
Effectively logo feature type is expanded so that the identification of logo no longer solely relies on logo LOGO, efficiently solves and is based on
Logo LOGO carries out logo and differentiates there is vehicle-logo recognition problem under the conditions of blocking and damage in logo, while also efficiently solving
Tail board logo position is not fixed caused vehicle-logo recognition problem.Board vehicle-logo recognition before the present invention can not only realize, but also can realize tail
Board vehicle-logo recognition, and it can be directed to actual demand, select major class, major class+subclass style or major class+subclass style+time limit
Logo exports.
As shown in Fig. 2, a kind of vehicle-logo recognition system, including picture pretreatment unit 110, vehicle-logo recognition region coarse positioning list
Member 120 and vehicle-logo recognition unit 130.
Picture pretreatment unit 110 is normally at horizontal top and the logo of car plate for using for reference priori, logo
It is normally in the vertical axis of symmetry of the front/rear face of vehicle, binary conversion treatment is carried out to the video pictures comprising vehicle, obtains two-value
Image g (x, y);
Vehicle-logo recognition region coarse positioning unit 120 is used to carry out sobel edge detections to the bianry image, including
Vehicle target area g ' (x, y);In region g ' (x, y), the x for meeting formula (1) is chosenmin,xmax,ymin,ymax:
α is taken, the value of β is 1, and passes through formula (2) coarse positioning vehicle-logo recognition region (x, y, α W, β H):
W is the width in vehicle-logo recognition region, and H is the height in vehicle-logo recognition region, and α, β are respectively width, height parameter,
(x, y) is the coordinate of each pixel in vehicle-logo recognition region.
Vehicle-logo recognition unit 130 is used for:
When identifying logo major class, in α=0.2, the vehicle-logo recognition region of β=0.2 carries out the extraction of HOG and PCA features,
Logo major class is identified by logo grader, such as:Masses, Audi, BMW, Buick etc..
When identifying logo major class and subclass, HOG and PCA spies are carried out in the vehicle-logo recognition region of α=1.3, β=0.2
The extraction of sign identifies logo major class and subclass, such as by logo grader:Masses-Santana, masses-Jetta, BMW-Q5,
Buick-is triumphant more etc..
Identification logo major class, subclass and prescribe a time limit in year, in α=1.3, the vehicle-logo recognition region of β=0.6 carry out HOG and
The extraction of PCA features identifies logo major class, subclass and the time limit, such as by logo grader:Masses-Santana -2006-
2007 sections, Audi's-AL4-2004 moneys etc..
Above-mentioned logo grader is trained by following steps and is obtained:
Logo samples pictures are pre-processed by picture pre-treatment step S110.
Vehicle-logo recognition region is positioned in logo samples pictures by vehicle-logo recognition region coarse positioning step S120.
The extraction of HOG and PCA features is carried out in vehicle-logo recognition region.
Logo grader is trained with HOG and PCA features.
But those of ordinary skill in the art it should be appreciated that more than embodiment be intended merely to illustrate this
Invention, and be not used as limitation of the invention, as long as in the spirit of the present invention, to embodiment described above
Variation, modification will all fall within the scope of claims of the present invention.