CN105205444A - Vehicle logo identification method based on dot pair characteristics - Google Patents

Abstract

The invention discloses a vehicle logo identification method based on dot pair characteristics. The method comprises steps that, pre-processing on a vehicle logo pattern is carried out, and determination, binaryzation and normalization for back and forth backgrounds are carried out to form a standard vehicle logo pattern; a framework area in the back and forth backgrounds is extracted based on the standard vehicle logo pattern, characteristic dot pair extraction is carried out through a random dot mode to form standard dot pairs; vehicle logo samples are intercepted from a practical bayonet image to carry out characteristic dot pair validity determination, valid dot pairs are stored in a database to form a characteristic dot pair database, and determination thresholds of vehicle logos in various types are further calculated; characteristic dot pair templates in the database are extracted to be in multi-scale matching with candidate vehicle logo areas to accomplish vehicle logo identification. Through the method, a specific identification scheme is proposed for vehicle logo identification in the bayonet image, a correct rate of the identification result is high, and requirements of a practical intelligent traffic system can be satisfied.

Description

A kind of based on the automobile logo identification method of point to feature

Technical field

The present invention relates to target identification technology field, specifically a kind of based on the automobile logo identification method of point to feature.

Background technology

Vehicle-logo recognition is an important branch in the application of intelligent transportation system computer vision, be widely used in traffic flow analysis, the management of the vehicle that hits out against theft, specification traffic order, large parking lot, the field such as highway automatic charging.Road vehicle intelligent monitoring and recording system (also known as Gate System) has a lot of application, the candid photograph image from then on obtained in system in monitoring overspeed of vehicle, traffic hazard detection etc., is called bayonet socket image.Vehicle-logo recognition technology in the present invention is exactly a kind of recognition methods proposed for the vehicle-logo recognition in bayonet socket image.

When carrying out vehicle-logo recognition, crucial technology point is " extraction of car mark feature ".No matter be the automobile logo identification method based on edge histogram, the automobile logo identification method based on edge invariant moment, the automobile logo identification method based on pixel distribution, the automobile logo identification method based on template matches, or based on the automobile logo identification method of pattern-recognition, its core is all " extraction of feature ".In said method, have adopt edge histogram, edge invariant moment, pixel distribution be feature, also have and carry out vehicle-logo recognition based on Corner Feature, SIFT feature, HOG feature, LBP feature etc., also have based on grey value characteristics, adopt PCA, ICA method to carry out feature extraction.

But in the complex scene of reality, due to the impact by illumination, weather, these features can not carry out unique description to car mark, and its robustness has limitation.Therefore, so far, the automobile logo identification method carrying out applying in actual card port system is few, and vehicle-logo recognition problem remains a challenging Research Challenges.

Summary of the invention

The object of this invention is to provide a kind of based on the automobile logo identification method of point to feature, to solve the vehicle-logo recognition problem in bayonet socket image.

In order to achieve the above object, the technical solution adopted in the present invention is:

Based on the automobile logo identification method of point to feature, it is characterized in that: comprise the following steps:

(1) pre-service of car logo image:

To mark on a map case based on standard vehicle, in conjunction with car target imaging contexts in actual bayonet socket image, form car logo image, pre-service is carried out to car logo image, form car mark standard picture;

(2) extraction of feature point pairs:

Point is adopted (, to put form value by the coordinates of 2 feature, point describes the magnitude relationship of gray-scale value between 2 to value) car mark is described, based on car mark standard picture, carry out a little to the extraction of feature, the result extracted is feature point pairs, is called standard point pair;

(3) foundation of feature point pairs database:

In conjunction with the car logo image in actual bayonet socket image, the validity right to standard point judges, extracts effective standard point pair, stored in database, morphogenesis characters point is to database, and effective standard point corresponding to often kind of car mark, to formation template, is called feature point pairs template;

(4) car target identification:

Based on position and the size of car plate in bayonet socket image, carry out coarse positioning to car mark, the region claiming car mark to occur is candidate region, in candidate region, utilizes the feature point pairs template in feature point pairs database to carry out Based on Multiscale Matching, to realize the identification of car target.

Described is a kind of based on the automobile logo identification method of point to feature, and it is characterized in that: in step (1), first carrying out pretreated process to car logo image is:

The car of selection standard is marked on a map case, forms the car logo image of rectangle according to its height and width.Regard the pattern part in car mark as prospect, other parts regard background as, and can find out according to car target imaging contexts actual in bayonet socket image, prospect is generally partially bright, and background is partially dark; Therefore carry out binaryzation according to prospect and background to car mark, prospect all represents by white, and background black represents, car logo image size be normalized, the length and width after normalization is respectively L _standard, W _standard, form car mark standard picture.

Described is a kind of based on the automobile logo identification method of point to feature, and it is characterized in that: in step (2), the leaching process carrying out feature point pairs based on the car mark standard picture after pre-service is as follows:

Based on black, white two parts of car mark standard picture, extract inside and outside contour respectively, inside and outside contour is shunk respectively to centre, each contraction pixel, until the region after shrinking is 1/4 of former area size, claim the region after shrinking to be backbone region.

Respectively the point in black, white part backbone region is formed set, be designated as set A and set B, random extracts respectively a bit from set A, B, morphogenesis characters point pair, with (x _a, y _a, x _b, y _b, R) represent, wherein, (x _a, y _a) be the coordinate of A point, (x _b, y _b) be the coordinate of B point, R represents the grey scale pixel value magnitude relationship between A, B at 2, is called a little to value, puts and has 1 ,-1 two kind of possibility to value value, 1 represents that the gray-scale value of A point is more than or equal to the gray-scale value of B point, and-1 represents that the gray-scale value of A point is less than the gray-scale value of B point;

For often kind of car mark, according to the method described above extract minutiae pair, repeat to extract m time, the set that morphogenesis characters point is right, is called standard point pair, is designated as P _{standPotntMatches}, some logarithm is m, and these feature point pairs, for describing car target feature, claim this being characterized as a little to feature.

Described is a kind of based on the automobile logo identification method of point to feature, it is characterized in that: in step (3), sets up car target feature point pairs data base procedure as follows:

Based on bayonet socket image, the manual car mark intercepting various vehicle, set up car mark database, require that often kind of car target number is no less than 400, its size be normalized, the length and width after normalization is consistent with the length and width of car mark standard picture respectively;

For the n width car logo image of a certain type intercepted from bayonet socket image, by the standard point extracted in previous step to being superimposed upon in n width car logo image respectively, calculate often group point to put value n that obtains in n width car logo image, as fruit dot is more than or equal to 86% to the ratio that value is consistent, then think that this point is to being available point pair, retains; As fruit dot is less than 86% to the ratio that value is consistent, then think that this point is to being point of instability pair, deletes this point right.The available point finally retained, to being effective standard point pair, remembers that its number is m _used;

Then, for jth width image, calculate each point to the point that the car logo image of correspondence is put to value sum, that is: calculate n width car logo image mid point to maximal value Sum _maxwith minimum value Sum _min, then Sum _maxand Sum _minbecome and judge the type car target threshold value;

For the car mark of every type, obtain final effective standard point according to the method described above to, Sum _maxand Sum _min, morphogenesis characters point to database, effective standard point corresponding to often kind of car mark to morphogenesis characters point to template, for follow-up to the identification of car target.

Described is a kind of based on the automobile logo identification method of point to feature, and it is characterized in that: in step (4), carry out Based on Multiscale Matching with the feature point pairs template in feature point pairs database, to realize the identification of car target, process is as follows:

Based on the position of car plate, coarse positioning is carried out to car mark, extract the image in coarse positioning region, be called the candidate image of car mark, be designated as I _candidate;

Based on size and the proportionate relationship of car plate, determine car target size in image, note is long and wide is respectively L, W, and based on car target size, determine and range scale during feature point pairs template matches, the scope of yardstick S is $\left(\min \right(\frac{L}{L_{s\tan dard}},\frac{W}{W_{s\tan dard}})-\mathrm{\α},\max (\frac{L}{L_{s\tan dard}},\frac{W}{W_{s\tan dard}})+\mathrm{\α}),$ Wherein α is scale parameter, can adjust according to actual conditions;

According to range scale, carry out convergent-divergent to the size of template, the length and width of convergent-divergent rear pattern plate is respectively S*L _standards*W _standard; Utilize the template under different scale scope to carry out scan matching in candidate image, if match, then think that car mark type corresponding to car mark and this template is identical, the end of scan; If do not matched, then again carry out Based on Multiscale Matching with other feature point pairs template.

Compared with the prior art, beneficial effect of the present invention is embodied in:

1, in recognition methods:

(1) point that the present invention proposes well can describe car mark feature, not easily by the impact of illumination, weather, has very strong robustness.

(2) the feature point pairs extracting method based on backbone region adopted in the present invention, effectively make use of the fact that car logo image backbone region gray-value variation is minimum, simultaneously, to effective standard point to when choosing, make use of the car obtained from actual bayonet socket image and be designated as reference, the feature point pairs extracted in this way can effectively be described the different car marks in bayonet socket image.

2, on recognition effect, based on car target identification in bayonet socket image, the present invention can obtain very high discrimination, can meet the needs of actual intelligent transportation system.

Accompanying drawing explanation

Fig. 1 is a kind of based on the process flow diagram of point to the automobile logo identification method of feature.

Fig. 2 is the schematic diagram of extracted feature point pairs.

Embodiment

As shown in Figure 1, a kind of based on the automobile logo identification method of point to feature, comprise the following steps:

(1) pre-service of car logo image:

To mark on a map case based on standard vehicle, in conjunction with car target imaging contexts in actual bayonet socket image, form car logo image, pre-service is carried out to car logo image, form car mark standard picture;

(2) extraction of feature point pairs:

Point is adopted (, to put form value by the coordinates of 2 feature, point describes the magnitude relationship of gray-scale value between 2 to value) car mark is described, based on car mark standard picture, carry out a little to the extraction of feature, the result extracted is feature point pairs, is called standard point pair;

(3) foundation of feature point pairs database:

In conjunction with the car logo image in actual bayonet socket image, the validity right to standard point judges, extracts effective standard point pair, stored in database, morphogenesis characters point is to database, and effective standard point corresponding to often kind of car mark, to formation template, is called feature point pairs template;

(4) car target identification:

Based on position and the size of car plate in bayonet socket image, carry out coarse positioning to car mark, the region claiming car mark to occur is candidate region, in candidate region, utilizes the feature point pairs template in feature point pairs database to carry out Based on Multiscale Matching, to realize the identification of car target.

In step (1), carrying out pretreated process to car logo image is:

The car of selection standard is marked on a map case, forms the car logo image of rectangle according to its height and width.Regard the pattern part in car mark as prospect, other parts regard background as, and can find out according to car target imaging contexts actual in bayonet socket image, prospect is generally partially bright, and background is partially dark; Therefore carry out binaryzation according to prospect and background to car mark, prospect all represents by white, and background black represents, car logo image size be normalized, the length and width after normalization is respectively L _standard, W _standard, form car mark standard picture.

In step (2), the leaching process carrying out feature point pairs based on the car mark standard picture after pre-service is as follows:

Based on black, white two parts of car mark standard picture, extract inside and outside contour respectively, inside and outside contour is shunk respectively to centre, each contraction pixel, until the region after shrinking is 1/4 of former area size, claim the region after shrinking to be backbone region.

Respectively the point in black, white part backbone region is formed set, be designated as set A and set B, random extracts respectively a bit from set A, B, morphogenesis characters point pair, with (x _a, y _a, x _b, y _b, R) represent, wherein, (x _a, y _a) be the coordinate of A point, (x _b, y _b) be the coordinate of B point, R represents the grey scale pixel value magnitude relationship between A, B at 2, is called a little to value, puts and has 1 ,-1 two kind of possibility to value value, 1 represents that the gray-scale value of A point is more than or equal to the gray-scale value of B point, and-1 represents that the gray-scale value of A point is less than the gray-scale value of B point;

For often kind of car mark, according to the method described above extract minutiae pair, repeat to extract m time, the set that morphogenesis characters point is right, is called standard point pair, is designated as P _{standPotntMatches}, some logarithm is m, and these feature point pairs, for describing car target feature, claim this being characterized as a little to feature.

In step (3), set up car target feature point pairs data base procedure as follows:

Based on bayonet socket image, the manual car mark intercepting various vehicle, set up car mark database, require that often kind of car target number is no less than 400, its size be normalized, the length and width after normalization is consistent with the length and width of car mark standard picture respectively;

For the n width car logo image of a certain type intercepted from bayonet socket image, by the standard point extracted in previous step to being superimposed upon in n width car logo image respectively, calculate often group point to put value n that obtains in n width car logo image, as fruit dot is more than or equal to 86% to the ratio that value is consistent, then think that this point is to being available point pair, retains; As fruit dot is less than 86% to the ratio that value is consistent, then think that this point is to being point of instability pair, deletes this point right.The available point finally retained, to being effective standard point pair, remembers that its number is m _used;

Then, for jth width image, calculate each point to the point that the car logo image of correspondence is put to value sum, that is: calculate n width car logo image mid point to maximal value Sum _maxwith minimum value Sum _min, then Sum _maxand Sum _minbecome and judge the type car target threshold value;

For the car mark of every type, obtain final effective standard point according to the method described above to, Sum _maxand Sum _min, morphogenesis characters point to database, effective standard point corresponding to often kind of car mark to morphogenesis characters point to template, for follow-up to the identification of car target.

In step (4), carry out Based on Multiscale Matching with the feature point pairs template in feature point pairs database, to realize the identification of car target, process is as follows:

Based on the position of car plate, coarse positioning is carried out to car mark, extract the image in coarse positioning region, be called the candidate image of car mark, be designated as I _candidate;

Based on size and the proportionate relationship of car plate, determine car target size in image, note is long and wide is respectively L, W, and based on car target size, determine and range scale during feature point pairs template matches, the scope of yardstick S is $\left(\min \right(\frac{L}{L_{s\tan dard}},\frac{W}{W_{s\tan dard}})-\mathrm{\α},\max (\frac{L}{L_{s\tan dard}},\frac{W}{W_{s\tan dard}})+\mathrm{\α}),$ Wherein α is scale parameter, can adjust according to actual conditions;

According to range scale, carry out convergent-divergent to the size of template, the length and width of convergent-divergent rear pattern plate is respectively S*L _standard, S*W _standard; Utilize the template under different scale scope to carry out scan matching in candidate image, if match, then think that car mark type corresponding to car mark and this template is identical, the end of scan; If do not matched, then again carry out Based on Multiscale Matching with other feature point pairs template.

The present invention, when carrying out vehicle-logo recognition, first needs from Gate System, obtain various car target sample, requires that every type car target sample number is no less than 400.Then, this method is mainly carried out (as shown in Figure 1) according to following four steps:

Step 1: to mark on a map case based on standard vehicle, in conjunction with car target imaging contexts in actual bayonet socket image, forms car logo image, carries out pre-service to car logo image, forms car mark standard picture;

The car of selection standard is marked on a map case, forms the car logo image of rectangle according to its height and width.Regard the pattern part in car mark as prospect, other parts regard background as, and can find out according to car target imaging contexts actual in bayonet socket image, prospect is generally partially bright, and background is partially dark.Therefore, carry out binaryzation according to prospect and background to car mark, prospect all represents by white, and background black represents, car logo image size be normalized, the length and width after normalization is respectively L _standard, W _standard, form car mark standard picture.

Step 2: the extraction based on car mark standard picture, carrying out " putting feature ".

Based on black, white two parts of car mark standard picture, extract inside and outside contour respectively, inside and outside contour is shunk respectively to centre, each contraction pixel, until the region after shrinking is 1/4 of former area size, the region after shrinking is claimed to be " backbone region ".

Respectively the point in black, white part backbone region is formed set, be designated as set A and set B.Random extracts respectively a bit from set A, B, morphogenesis characters point pair, with (x _a, y _a, x _b, y _b, R) represent, wherein, (x _a, y _a) be the coordinate of A point, (x _b, y _b) be the coordinate of B point, R represents the grey scale pixel value magnitude relationship between A, B at 2, is called " putting value ", and its value has 1 ,-1 two kind of possibility, 1 represents that the gray-scale value of A point is more than or equal to the gray-scale value of B point, and-1 represents that the gray-scale value of A point is less than the gray-scale value of B point.

For often kind of car target standard picture, extract minutiae pair according to the method described above, repeats to extract m time, the set that morphogenesis characters point is right, is called and " standard point to " is designated as P _{standPotntMatches}, some logarithm is m (as shown in Figure 2).

These feature point pairs, for describing car target feature, claim this being characterized as " putting feature ".

Step 3: in conjunction with the car logo image in actual bayonet socket image, judges the validity of " standard point to ", extracts effective standard point pair, is formed " feature point pairs database ".

Based on bayonet socket image, the manual car mark intercepting various vehicle, set up car mark database, require that often kind of car target number is no less than 400, its size be normalized, the length and width after normalization is consistent with the length and width of car mark standard picture respectively.

For the n width car logo image of a certain type intercepted from bayonet socket image, by the standard point extracted in previous step to being superimposed upon in n width car logo image respectively, calculate often group point to put value n that obtains in n width car logo image, as fruit dot is more than or equal to 86% to the ratio that value is consistent, then think that this point is to being available point pair, retains; As fruit dot is less than 86% to the ratio that value is consistent, then think that this point is to being point of instability pair, deletes this point right.The available point finally retained, to being effective standard point pair, remembers that its number is m _used.

Then, for jth width image, calculate each point to the point that the car logo image of correspondence is put to value sum, that is: calculate n width car logo image mid point to maximal value Sum _maxwith minimum value Sum _min, then Sum _maxand Sum _minbecome and judge the type car target threshold value.

For the car mark of every type, obtain final effective standard point according to the method described above to, Sum _maxand Sum _min, morphogenesis characters point is to database.Effective standard point corresponding to often kind of car mark to morphogenesis characters point to template, for follow-up to the identification of car target.

Step 4: in " candidate region " that car mark may occur, utilizes the feature point pairs template in feature point pairs database to carry out Based on Multiscale Matching, to realize the identification of car target.

Based on the position of car plate, coarse positioning is carried out to car mark, extract the image in coarse positioning region, be called the candidate image of car mark, be designated as I _candidate;

Based on size and the proportionate relationship of car plate, determine car target size in image, note is long and wide is respectively L, W.Based on car target size, determine and range scale during feature point pairs template matches, the scope of yardstick S is $\left(\min \right(\frac{L}{L_{s\tan dard}},\frac{W}{W_{s\tan dard}})-\mathrm{\α},\max (\frac{L}{L_{s\tan dard}},\frac{W}{W_{s\tan dard}})+\mathrm{\α}),$ Wherein α is scale parameter, can adjust according to actual conditions.

According to range scale, carry out convergent-divergent to the size of template, the length and width of convergent-divergent rear pattern plate is respectively S*L _standard, S*W _standard.Utilize the template under different scale scope to carry out scan matching in candidate image, if match, then think that car mark type corresponding to car mark and this template is identical, the end of scan; If do not matched, then again carry out Based on Multiscale Matching with other feature point pairs template.

Unique distinction of the present invention is embodied in:

1. show by experiment, due to the impact by illumination, weather, the imaging of car mark is not the same, but in whole car logo image, the gray-scale value magnitude relationship of car mark pattern part and background parts is not easily influenced, based on this priori, propose the thought of " putting feature ", with point, the description of car target is carried out to feature.

2. show by experiment, the edge of car mark pattern part is subject to the impact of illumination, weather, its gray-scale value situation instability (this result also in the instability at background parts edge), the core of case but car is marked on a map, its affected degree is minimum, therefore, the present invention's employing gets method a little at random from pattern center part and background core, morphogenesis characters point pair.

To sum up, the present invention utilize a little to feature to carry out the description of car target, by the extraction to feature point pairs, in car mark candidate region, carry out Based on Multiscale Matching, to realize the identification of car target.

Claims (5)

1., based on the automobile logo identification method of point to feature, it is characterized in that: comprise the following steps:

(1) pre-service of car logo image:

To mark on a map case based on standard vehicle, in conjunction with car target imaging contexts in actual bayonet socket image, form car logo image, pre-service is carried out to car logo image, form car mark standard picture;

(2) extraction of feature point pairs:

Adopt point to be described car mark feature, put to feature by 2 coordinate, put value formed, put magnitude relationship value being described to gray-scale value between 2, based on car mark standard picture, carry out a little to the extraction of feature, the result of extraction is feature point pairs, is called standard point pair;

(3) foundation of feature point pairs database:

In conjunction with the car logo image in actual bayonet socket image, the validity right to standard point judges, extracts effective standard point pair, stored in database, morphogenesis characters point is to database, and effective standard point corresponding to often kind of car mark, to formation template, is called feature point pairs template;

(4) car target identification:

Based on position and the size of car plate in bayonet socket image, carry out coarse positioning to car mark, the region claiming car mark to occur is candidate region, in candidate region, utilizes the feature point pairs template in feature point pairs database to carry out Based on Multiscale Matching, to realize the identification of car target.

2. according to claim 1 a kind of based on the automobile logo identification method of point to feature, it is characterized in that: in step (1), carrying out pretreated process to car logo image is:

The car of selection standard is marked on a map case, and form the car logo image of rectangle according to its height and width, regard the pattern part in car mark as prospect, other parts regard background as, and can find out according to car target imaging contexts actual in bayonet socket image, prospect is generally partially bright, and background is partially dark; Therefore carry out binaryzation according to prospect and background to car mark, prospect all represents by white, and background black represents, car logo image size be normalized, the length and width after normalization is respectively L _standard, W _standard, form car mark standard picture.

3. according to claim 1 a kind of based on the automobile logo identification method of point to feature, it is characterized in that: in step (2), the leaching process carrying out feature point pairs based on the car mark standard picture after pre-service is as follows:

Based on black, white two parts of car mark standard picture, extract inside and outside contour respectively, inside and outside contour is shunk respectively to centre, each contraction pixel, until the region after shrinking is 1/4 of former area size, claim the region after shrinking to be backbone region;

Respectively the point in black, white part backbone region is formed set, be designated as set A and set B, random extracts respectively a bit from set A, B, morphogenesis characters point pair, with (x _a, y _a, x _b, y _b, R) represent, wherein, (x _a, y _a) be the coordinate of A point, (x _b, y _b) be the coordinate of B point, R represents the grey scale pixel value magnitude relationship between A, B at 2, is called a little to value, puts and has 1 ,-1 two kind of possibility to value value, 1 represents that the gray-scale value of A point is more than or equal to the gray-scale value of B point, and-1 represents that the gray-scale value of A point is less than the gray-scale value of B point;

For often kind of car mark, according to the method described above extract minutiae pair, repeat to extract m time, the set that morphogenesis characters point is right, is called standard point pair, is designated as P _{standPotntMatches}, some logarithm is m, and these feature point pairs, for describing car target feature, claim this being characterized as a little to feature.

4. according to claim 1 a kind of based on the automobile logo identification method of point to feature, it is characterized in that: in step (3), set up car target feature point pairs data base procedure as follows:

Based on bayonet socket image, the manual car mark intercepting various vehicle, set up car mark database, require that often kind of car target number is no less than 400, its size be normalized, the length and width after normalization is consistent with the length and width of car mark standard picture respectively;

For the n width car logo image of a certain type intercepted from bayonet socket image, by the standard point extracted in previous step to being superimposed upon in n width car logo image respectively, calculate often group point to put value n that obtains in n width car logo image, as fruit dot is more than or equal to 86% to the ratio that value is consistent, then think that this point is to being available point pair, retains; As fruit dot is less than 86% to the ratio that value is consistent, then think that this point is to being point of instability pair, deletes this point right, the available point finally retained, to being effective standard point pair, remembers that its number is m _used;

Then, for jth width image, calculate each point to the point that the car logo image of correspondence is put to value sum, that is: calculate n width car logo image mid point to maximal value Sum _maxwith minimum value Sum _min, then Sum _maxand Sum _minbecome and judge the type car target threshold value;

For the car mark of every type, obtain final effective standard point according to the method described above to, Sum _maxand Sum _min, morphogenesis characters point to database, effective standard point corresponding to often kind of car mark to morphogenesis characters point to template, for follow-up to the identification of car target.

5. according to claim 1 a kind of based on the automobile logo identification method of point to feature, it is characterized in that: in step (4), carry out Based on Multiscale Matching with the feature point pairs template in feature point pairs database, to realize the identification of car target, process is as follows:

Based on the position of car plate, coarse positioning is carried out to car mark, extract the image in coarse positioning region, be called the candidate image of car mark, be designated as I _candldate;

Based on size and the proportionate relationship of car plate, determine car target size in image, note is long and wide is respectively L, W, and based on car target size, determine and range scale during feature point pairs template matches, the scope of yardstick S is wherein α is scale parameter, can adjust according to actual conditions;

According to range scale, carry out convergent-divergent to the size of template, the length and width of convergent-divergent rear pattern plate is respectively S*L _standard, S*W _standard; Utilize the template under different scale scope to carry out scan matching in candidate image, if match, then think that car mark type corresponding to car mark and this template is identical, the end of scan; If do not matched, then again carry out Based on Multiscale Matching with other feature point pairs template.