CN100452079C - Vehicle mode identifying method in whole-automatic vehicle-cleaning - Google Patents

Abstract

The present invention provides a vehicle mode identifying method in full-automatic vehicle cleaning, which comprises the following steps: (1) light beams are emitted to different positions of the surface of a measured vehicle by a laser; (2) laser light reflected on different light-spot positions of the surface of the measured vehicle is collected by a CCD camera head, and then the depth information of the vehicle is calculated by using a conventional method; (3) the measured vehicle is shot by the CCD camera head, and a plane image of the landmark characteristic part of the measured vehicle is obtained; (4) the similarity between standard moulding board images in the measured vehicle and a database is calculated according to the plane image of the landmark characteristic part of the measured vehicle; (5) whether the measured vehicle is an existing vehicle model or not or the vehicle model does not exist is judged by comparing whether the depth information of the vehicle is fitted with the depth information corresponding to the standard moulding board in the database or not if the similarity of the plane images is higher. The technical scheme of the present invention is applied. The present invention has the advantages of high identifying correction rate and easy realization, and is helpful to improve work efficiency.

Description

The method of vehicle mode identification during a kind of full automatic car cleans

Technical field

The present invention relates to a kind of vehicle mode recognition methods, relate in particular to the method for vehicle mode identification in a kind of full automatic car cleaning.

Background technology

In present various automatic car washing installations,, need not import car shape, type information owing to all be based on the roller hairbrush.In novel cleaning method, have that to need outside vehicle profile information, general method be the instant vehicle model of measuring or manually import, from database, to obtain existing profile wheel information.

The method of existing object identification adopts plane picture to handle and identification substantially, limited by environmental baseline, and recognition correct rate is not high.

In actual applications, most outside vehicle profile informations all are stored in the database, therefore need a kind of method of utilizing existing database that vehicle model is accurately discerned.

Summary of the invention

Goal of the invention of the present invention can be passed through car plane image information and steric information combination, and related data realizes in the comparison database.

The invention provides vehicle mode recognition methods in a kind of full automatic car cleaning, may further comprise the steps:

(1) launches light beam by a common low power laser to tested vehicle figuratrix diverse location;

(2), utilize the distance between conventional method calculating tested vehicle surface diverse location and the CCD camera then, i.e. the depth information of vehicle by the different light spot position laser light reflected in CCD camera collection tested vehicle surface;

(3) this CCD camera is also made a video recording to tested vehicle, obtains to have the symbolic characteristic plane picture partly of tested vehicle;

(4), calculate tested vehicle and store similarity between the standard form image in the database of standard vehicle characteristic according to the plane picture of the symbolic characteristic of tested vehicle part;

(5) if certain the standard form image correlation in tested vehicle symbolic characteristic part and this database is higher, calculate the similarity between the depth information of this standard form correspondence in the depth information of this vehicle and this database, if these two kinds of depth informations are also similar substantially, judge that then tested vehicle is certain existing vehicle in the database, otherwise do not have this vehicle in the judgment data storehouse.

Preferably, in step (3), tested vehicle afterbody or head are made a video recording.

Preferably, after step (3) obtains plane picture, also comprise to this plane picture carry out smoothly, contrast strengthens, the step of edge extracting processing.

Preferably, utilize the method for linear interpolation method computed range to obtain depth information in the step (2), its computing formula is for working as x _i＜x＜x _I+1The time (1≤i≤n),

z＝z _i+(z _i+1-z _i)(x-x _i)/(x _i+1-x _i)，

X wherein, x ₁, x ₂..., x _nFor laser spots in the value of this CCD camera lateral attitude with make z, z ₁, z ₂..., z _nFor these laser spots correspondences and this CCD camera between range data.

Preferably, the computing formula of similarity is in the step (4):

$R(i,j)=\frac{\underset{m=1}{\overset{M}{\mathrm{\Σ}}}\underset{n=1}{\overset{M}{\mathrm{\Σ}}}[S^{i,j}(m,n)\×T(m,n)]}{\underset{m=1}{\overset{M}{\mathrm{\Σ}}}\underset{n=1}{\overset{M}{\mathrm{\Σ}}}{\left[S^{i,j}(m,n)\right]}^2}$

Wherein (i j) is similarity measure between tested vehicle and the standard form image, S to R ^{I, j}(i, j represents the position) represents feature image, and T represents template image, pixel place row, column sequence number in m, the n representing images.

Preferably, the determination methods of similarity is in the step (5):

At first calculated difference D (h, k),

D(h，k)＝G(h，k)-Q(h，k)；

H=1 wherein, 2 ..., H; K=1,2 ..., K; H, k are the sequence number of range data in level and vertical direction; (h k) represents vehicle characteristics case depth data to G, and (h k) represents the higher standard form depth data of similarity measure in the step (4) to Q;

Then, (h seeks the maximum data of the value of being equal to each other in k), if the number of these data accounts for the number percent of overall data number more than 98% at hk difference D, and the maximum difference between other data just thinks that less than permissible error vehicle mode is similar to the standard form depth information; Otherwise, just think vehicle mode and standard form depth information dissmilarity.

Use technical scheme of the present invention, the recognition correct rate height is realized easily, helps to increase work efficiency.

Description of drawings

Below in conjunction with accompanying drawing, concrete enforcement of the present invention is described in detail.

The FB(flow block) of Fig. 1 vehicle mode identification of the present invention.

Embodiment

When as shown in Figure 1, carrying out vehicle mode identification:

At first, to tested vehicle special characteristic sector scanning formula emission light beam, CCD is every a determining deviation sampled light point, and the calculating light spot position by a common low power laser.This laser instrument is the laser instrument that the prior art laser ranging is used, as long as can satisfy distance measurement function.The general 440-550nm of the wavelength of used laser instrument.Adopt the laser instrument of 480nm at a preferred embodiment.

Secondly, by a CCD (Charge Coupled Device, charge-coupled image sensor) the different light spot position laser light reflected in camera collection tested vehicle surface are utilized the distance between conventional method calculating tested vehicle surface diverse location and the CCD camera, i.e. the depth information of vehicle then.The conventional computing method of the depth information of vehicle comprise trigonometry or method of interpolation etc.In general, with light spot position and actual range having been carried out the demarcation of enough multiple spots before the laser ranging, utilize the light spot position of demarcating right, utilize trigonometry or method of interpolation, just can obtain the measured distance value of enough accuracy according to the actual measurement light spot position with the data of corresponding actual demarcation distance.Adopt the linear interpolation method to calculate the depth information of vehicle in a preferred embodiment.

In calibration process, can obtain laser spot at horizontal (being X-direction) the position calibration coordinate figure x of CCD camera ₁, x ₂..., x _nWith the range data z corresponding with it ₁, z ₂..., z _nThese data are preserved after demarcation is finished.When the actual measurement vehicle, be x if obtain surface of vehicle laser spot p actual measurement coordinate figure of lateral attitude on the CCD camera, in order to obtain the numerical value of surface of vehicle laser spots p, can calculate with following formula to the distance z of CCD camera:

Work as x _i＜x＜x _I+1The time (1≤i≤n),

z＝z _i+(z _i+1-z _i)(x-x _i)/(x _i+1-x _i)；

If x equals x _i, then directly make z=z _iGet final product.As long as x ₁＜x＜x _n, can calculate the depth information of vehicle afterbody point according to following formula.

Then, this CCD camera is also made a video recording to tested vehicle, obtains to have the symbolic characteristic plane picture partly of tested vehicle.In general, the symbolic characteristic of tested vehicle partly is positioned at the afterbody and the head of vehicle, and for example vehicle afterbody/head is represented the metal word of car model, or the afterbody first half represent vehicle and car series font size (as Citroen zx, ZX, 1.6L) etc.Conventional processing such as the plane picture for acquisition preferably also carries out smoothly, contrast enhancing, edge extracting.Present embodiment adopts the camera of 1240 * 800 pixels.

Then, according to the plane picture of the symbolic characteristic of tested vehicle part, calculate the similarity between the standard form image in the database that tested vehicle and stores the standard vehicle characteristic.Wherein, in a preferred embodiment, the computing formula of plane picture similarity is:

$R(i,j)=\frac{\underset{m=1}{\overset{M}{\mathrm{\Σ}}}\underset{n=1}{\overset{M}{\mathrm{\Σ}}}[S^{i,j}(m,n)\×T(m,n)]}{\underset{m=1}{\overset{M}{\mathrm{\Σ}}}\underset{n=1}{\overset{M}{\mathrm{\Σ}}}{\left[S^{i,j}(m,n)\right]}^2}$

Wherein (i j) is simple crosscorrelation similarity measure between tested vehicle and the standard form image, S to R ^{I, j}Represent feature image, i, j represents the position, T represents template image, m, pixel place row, column sequence number in the n representing images.

At last, if certain the standard form image similarity in tested vehicle symbolic characteristic part and this database is higher, be that cross-correlation coefficient R is greater than preset threshold value, for example its cross-correlation coefficient R thinks and the template basically identical greater than 0.95, then, judge the similarity between the depth information of this standard form correspondence in the depth information of this vehicle and this database, if both depth information similaritys, judge that then tested vehicle is certain existing vehicle in the database, otherwise do not have this vehicle in the judgment data storehouse.Wherein, in a preferred embodiment, the determination methods of depth information similarity is:

At first calculated difference D (h, k),

D(h，k)＝G(h，k)-Q(h，k)；

H=1 wherein, 2 ..., H; K=1,2 ..., K; H, k are the sequence number of range data in level and vertical direction; (h k) represents vehicle characteristics case depth data to G, and (h k) represents R greater than 0.95 standard form depth data to Q;

Then, (h seeks the maximum data of the value of being equal to each other in k), if the number of these data accounts for the number percent of overall data number more than 98% at hk difference D, and the maximum difference between other data just thinks that less than permissible error vehicle mode is similar to the standard form depth information; Otherwise, just think vehicle mode and standard form depth information dissmilarity.In general, if distance value is unit with the millimeter, two values differ less than 2, think that promptly two values equate.Permissible error is generally 5 millimeters.

In general, just the model of vehicle can have been discerned with the simple crosscorrelation similarity measure of plane picture.But under special circumstances, because car owner or outside, possible car mark is complete or the car mark (the car mark that runs quickly being installed as the Xiali car) of other car has been installed, and will cause difficulty to identification.At this moment the big and less situation of R ' of corresponding R value, computing machine can be pointed out the staff, carries out artificial selection from two result of calculations.The recognition methods of combination of above plane picture correlation data and depth information can improve the reliability of automatic identification greatly, avoid because operating personnel are unfamiliar with the efficient that vehicle causes reduces or input error.

Use vehicle mode recognition methods of the present invention, at vehicle mark just often, recognition accuracy can effectively be increased work efficiency up to 98%.The present invention also is with a wide range of applications, can be in order to the identification of many objects with solid geometry feature.

Claims (7)

1. vehicle mode recognition methods during a full automatic car cleans is characterized in that, may further comprise the steps:

(1) launches light beam by a common low power laser to tested vehicle surface diverse location;

(2) by the different light spot position laser light reflected in CCD camera collection tested vehicle surface, calculate the distance between tested vehicle surface diverse location and the CCD camera then, i.e. the depth information of vehicle;

(3) this CCD camera is also made a video recording to tested vehicle, obtains to have the symbolic characteristic plane picture partly of tested vehicle;

(4), calculate tested vehicle and store similarity between the standard form image in the database of standard vehicle characteristic according to the plane characteristic image of the symbolic characteristic of tested vehicle part;

(5) if the cross-correlation coefficient between certain the standard form image in tested vehicle symbolic characteristic part and this database greater than preset threshold value, calculate the similarity between the depth information of this standard form correspondence in the depth information of this vehicle and this database, if similarity is also higher between these two depth informations, judge that then tested vehicle is certain existing vehicle in the database, otherwise do not have this vehicle in the judgment data storehouse.

2. vehicle mode recognition methods as claimed in claim 1 is characterized in that, in step (3) tested vehicle afterbody or head is made a video recording.

3. vehicle mode recognition methods as claimed in claim 1 is characterized in that, also comprises the step to this plane picture carries out smoothly, contrast strengthens, edge extracting is handled after step (3) obtains plane picture.

4. as the described vehicle mode recognition methods of one of claim 1-3, it is characterized in that utilize the method for linear interpolation method computed range to obtain depth information in the step (2), its computing formula is, works as x _i＜x＜x _I+1The time (1≤i≤n),

z＝z _i+(z _i+1-z _i)(x-x _i)/(x _i+1-x _i)，

Wherein x is the actual measurement coordinate figure of laser spot in this CCD camera lateral attitude, x ₁, x ₂..., x _nBe the demarcation coordinate figure of laser spot in this CCD camera lateral attitude, z, z ₁, z ₂..., z _nFor these laser spots correspondences and this CCD camera between range data.

5. as the described vehicle mode recognition methods of one of claim 1-3, it is characterized in that the computing formula of similarity is in the step (4):

$R(i,j)=\frac{\underset{m=1}{\overset{M}{\mathrm{\Σ}}}\underset{n=1}{\overset{M}{\mathrm{\Σ}}}[S^{i,j}(m,n)\×T(m,n)]}{\underset{m=1}{\overset{M}{\mathrm{\Σ}}}\underset{n=1}{\overset{M}{\mathrm{\Σ}}}{\left[S^{i,j}(m,n)\right]}^2}$

Wherein (i j) is simple crosscorrelation similarity measure between tested vehicle and the standard form image, S to R ^{I, j}Represent feature image, i, j represents the position, T represents template image, m, pixel place row, column sequence number in the n representing images.

6. as the described vehicle mode recognition methods of one of claim 1-3, it is characterized in that the determination methods of similarity is in the step (5):

At first calculated difference D (h, k),

D(h，k)＝G(h，k)-Q(h，k)；

H=1 wherein, 2 ..., H; K=1,2 ..., K; H, k are the sequence number of range data in level and vertical direction; (h k) represents vehicle characteristics surface actual measurement depth data to G, and (h k) represents the corresponding vehicle figuratrix of the higher template of plane picture similarity depth data to Q;

Then, at h * k difference D (h, k) seek the maximum data of the value of being equal to each other in, if the number of these data accounts for the number percent of overall data number more than 98%, and the maximum difference between other data just thinks that less than permissible error vehicle mode is similar to the standard form depth information; Otherwise, just think vehicle mode and standard form depth information dissmilarity.

7. vehicle mode recognition methods as claimed in claim 1 is characterized in that, preset threshold value is 0.95 in the step (5).

Images