CN103473761A - Automobile chassis three-dimensional elevation determination method based on binocular linear array CCD - Google Patents

Abstract

The invention discloses an automobile chassis three-dimensional elevation determination method based on a binocular linear array CCD. The method is specifically implemented according to the following steps: 1. performing gray scale closing operation processing on an automobile chassis image acquired by the binocular linear array CCD; 2. extracting the characteristics of a gray scale closing operation result obtained in step one; 3. performing characteristic matching on a left side view and a right side view; and 4. obtaining the elevation information of a stereo object. According to the method provided by the invention, by searching appropriate matching characteristics in terms of the properties of the objects of an automobile chassis and by relying on an automobile chassis safety real-time monitoring system, the detection result of the system is displayed in a three-dimensional mode, and sparse three-dimensional elevation data is obtained on the basis of binocular characteristic correct matching, so that dense three-dimensional elevation data can be further obtained accordingly, thus the restrictions based on an orientation map method are overcome, and geometrical constraint relations of different views are recovered by only depending on the characteristic matching among images, thereby the provided method has good application prospects.

Description

The three-dimensional elevation of automobile chassis based on the binocular line array CCD is determined method

Technical field

The invention belongs to binocular line array CCD 3 Dimension Image Technique field, be specifically related to the three-dimensional elevation of a kind of automobile chassis based on the binocular line array CCD and determine method.

Background technology

The detection of automobile chassis Environmental security is an important research direction of intelligent traffic safety management system.Existing vehicle safety check place, detection mode for automobile chassis, be mainly to rely on manual detection and manual instrument and equipment to be detected for a long time, the part place also needs the staff to utilize mirror image observation automobile chassis whether to carry contraband goods secretly, also has department to be banned by police dog.This just level method is wasted time and energy, inefficiency.

Summary of the invention

The purpose of this invention is to provide the three-dimensional elevation of a kind of automobile chassis based on the binocular line array CCD and determine method, solved the manual type observation automobile chassis existed in the prior art, waste time and energy, ineffective problem.

The technical solution adopted in the present invention is that the three-dimensional elevation of a kind of automobile chassis based on the binocular line array CCD is determined method, according to following steps, specifically implements:

Step 1, the automobile chassis image that the binocular line array CCD is collected carry out gray scale closed operation processing;

The feature of the gray scale closed operation result that step 2, extraction step 1 obtain;

Step 3, left and right view is carried out to characteristic matching;

Step 4, obtain the stereoscopic article elevation information.

The invention has the beneficial effects as follows, object characteristic for automobile chassis is searched suitable matching characteristic, rely on the automobile chassis safety real time monitoring system, testing result to system is carried out three-dimensional display, obtain sparse three-dimensional altitude figures on the basis of the correct coupling of binocular feature, further obtain thus dense three-dimensional altitude figures.

The accompanying drawing explanation

Fig. 1 is the binocular camera imaging model schematic diagram in the inventive method.

Embodiment

Below in conjunction with the drawings and specific embodiments, the present invention is described in detail.

The three-dimensional elevation of automobile chassis based on the binocular line array CCD of the present invention is determined method, according to following steps, specifically implements:

Step 1, the automobile chassis image that the binocular line array CCD is collected carry out gray scale closed operation processing

If the image that the binocular line array CCD collects size is m * n, by left and right view difference called after [I _l(x, y)] _{m * n}and [I _r(x, y)] _{m * n}, to utilize Gauss's template respectively left and right view to be done to the gray scale closed operation and process, the Gauss's template arranged in order to do the closed operation processing is [G (x, y)] _{m * n}, computing formula is:

$G(x,y)=K\·\frac{1}{2{\mathrm{\π\σ}}^2}\exp (-\frac{x^2+y^2}{2{\mathrm{\σ}}^2}),$ x＝1,2,…m;y＝1,2,…,n （1）

Wherein, K is scale-up factor, and for the image of 256 GTGs, span is K ∈ [80,100]; σ gets empirical value, and span is σ ∈ [10,20];

Gauss's template of utilizing formula (1) to obtain, respectively to left view [I _l(x, y)] _{m * n}and right view [I _r(x, y)] _{m * n}carry out the gray scale expansion process,

Above-mentioned [G (x, y)] _{m * n}what mean with G (x, y) is not same concept, [G (x, y)] _{m * n}what mean is the matrix of a m * n size, and that in formula (1), provide is all elements G (x, y) in this matrix, x=1, and 2 ..., m; Y=1,2 ..., the computing formula of n, at first, according to formula (2) and template [G (x, y)] _{m * n}carry out gray scale and computing, obtain result and be

with

$\widetilde{I_k}(x,y)=I_k(x,y)+G(x,y),$ x＝1,2,…,m;y＝1,2,…,n；k＝L,R （2）

Afterwards, according to with

obtain the gray scale expansion results according to formula (3)

with

x＝1,2,…,m;y＝1,2,…,n；k＝L,R （3）

Left and right view after the gray scale expansion process is carried out to the gray scale corrosion treatment again,

At first, according to formula (4), formula (3) is obtained

with carry out and template [G (x, y)] _{m * n}the gray scale difference computing, obtain result and be

with

${\stackrel{\‾}{I}}_k(x,y)={\hat{I}}_k(x,y)-G(x,y),$ x＝1,2,…,m;y＝1,2,…,n；k＝L,R， （4）

Afterwards, according to

with obtain the result [I of gray scale closed operation according to formula (5) _l ^close(x, y)] _{m * n}[I _r ^close(x, y)] _{m * n}:

x＝1,2,…,m;y＝1,2,…,n；k＝L,R； （5）

The feature of the gray scale closed operation result that step 2, extraction step 1 obtain

Step 2a, the image [I after step 1 is processed _l ^close(x, y)] _{m * n}[I _r ^close(x, y)] _{m * n}extract the canny edge, obtain edge image [I _l ^edge(x, y)] _{m * n}[I _r ^edge(x, y)] _{m * n}, afterwards, remove the interference of small size, (the canny edge extracting method all has detailed introduction in general textbook, at this, no longer repeats),

To image [I _l ^edge(x, y)] _{m * n}[I _r ^edge(x, y)] _{m * n}carry out the processing of labelling of eight connected domains, (labeling method all has detailed introduction in general textbook, at this, no longer repeats), establish after connected domain is labelled image [I _l ^edge(x, y)] _{m * n}obtain N _lindividual connected domain, the area of each connected domain is Area _i ^l, i is tag number, i=1, and 2 ..., N _l; Image [I _r ^edge(x, y)] _{m * n}obtain N _rindividual connected domain, the area of each connected domain is Area _i ^r, i is tag number, i=1, and 2 ..., N _r;

Set afterwards the threshold value A rea_Th that eliminates noise, (size of this threshold value is according to the size of the resolution of image, and Image Acquisition quality, the size of the noise that may occur is determined), all connected domains are processed, when the area of connected domain is greater than Area_Th, retain this connected domain; If while being less than Area_Th, think noise, remove this connected domain;

Step 2b, remove the transverse edge in edge image

Due to when carrying out three-dimensional modeling, only need to know left and right parallax, so remove transverse edge, be in order to resist the interference of transverse edge,

The rectangle template [M (x, y)] that the effect template is a p * q size is set _{p * q}, M (x, y)=1, x=1,2 ..., p; Y=1,2 ..., q; The p value is odd number, and in order to judge the thickness of transverse edge, span is p ∈ { 3,5,7}; The q value is odd number, in order to judge the minimum length of transverse edge, span be q ∈ 5,7,9,11},

By template [M (x, y)] _{p * q}act on two images that step 2a obtains

[I _r ^edge(x, y)] _{m * n}, on the position of the center of template corresponding to the pending pixel of image, carry out following computing, obtained removing the characteristic image [MI of transverse edge _l ^edge(x, y)] _{m * n}and [MI _r ^edge(x, y)] _{m * n}:

${{\mathrm{MI}}_k}^{\mathrm{edge}}(x,y)=\left\{\begin{array}{cc}0& \underset{i=x-(p-1)/2}{\overset{x+(p-1)/2}{\mathrm{\Π}}}\underset{j=y-(q-1)/2}{\overset{y+(q-1)/2}{\mathrm{\Σ}}}{I_k}^{\mathrm{edge}}(x+i,y+j)\·M(i,j)=0\\ 1& \underset{i=x-(p-1)/2}{\overset{x+(p-1)/2}{\mathrm{\Π}}}\underset{j=y-(q-1)/2}{\overset{y+(q-1)/2}{\mathrm{\Σ}}}{I_k}^{\mathrm{edge}}(x+i,y+j)\·M(i,j)\≠0\end{array}\right.,---\left(6\right)$

X=wherein (p+1)/2, (p+1)/2+1 ..., m-(p+1)/2,

y＝(q+1)/2,(q+1)/2+1,…,n-(q+1)/2，k＝L,R

MI _k ^edge(x,y)＝0， （7）

X=1 wherein, 2 ..., (p+1)/2-1,

y＝1,2,…,(q+1)/2-1，k＝L,R；

Step 2c, with reference to step 2a, remove image [MI _l ^edge(x, y)] _{m * n}[MI _r ^edge(x, y)] _{m * n}in the small size zone, for convenience of description for the purpose of, will remove the image of surface area still with [MI _l ^edge(x, y)] _{m * n}[MI _r ^edge(x, y)] _{m * n}meaned;

Step 2d, remove unreliable marginal information

Because the depth information of the stereoscopic article in image is reflected in left view [I _l(x, y)] _{m * n}and right view [I _r(x, y)] _{m * n}parallax in, therefore obtain its parallax with the mark stereoscopic article,

Suppose, left and right view is carried out to registration, (method for registering of two images all has detailed introduction in textbook, at this, no longer repeat), because what adopt here is the line array CCD imaging, so there is not rotation relationship in the same target in left and right view, therefore, the alternate position spike of reference field after the establishing standard in left and right view is Δ x and Δ y, disparity map [the ▽ I in order to judge stereoscopic article _l,R(x, y)] _{m * n}computing formula be:

$\&dtri;I_{L,R}(x,y)=\left\{\begin{array}{cc}0& |I_L(x,y)-I_R(x+\mathrm{\&Delta;x},y+\mathrm{\&Delta;y})|<\mathrm{\&epsiv;}\\ 1& |I_L(x,y)-I_R(x+\mathrm{\&Delta;x},y+\mathrm{\&Delta;y})|\&GreaterEqual;\mathrm{\&epsiv;}\end{array},\right.,$ x＝1,2,…,m;y＝1,2,…,n （8）

Wherein, ε is the adjustment parameter of parallax value, considers the impact that is subject in actual applications photoenvironment, for the parallax of non-stereoscopic article, certain difference is also arranged, and its value is taken as empirical value according to environment is set, and span is ε ∈ [15,30];

Afterwards, at disparity map

upper will in the edge feature obtained in step 2, do not belong to the part of stereoscopic article remove;

Because be distributed in left and right disparity map [▽ I on the edge line principle of stereoscopic article _l,R(x, y)] _{m * n}be 1 position, so by [MI _l ^edge(x, y)] _{m * n}[MI _r ^edge(x, y)] _{m * n}respectively with

carry out AND operation, obtain the left and right boundary characteristic image [I of stereoscopic article _l ^chara(x, y)] _{m * n}[I _r ^chara(x, y)] _{m * n}:

X=1 wherein, 2 ..., m; Y=1,2 ..., n, k=L, R;

Step 3, left and right view is carried out to characteristic matching

Step 3a, ask the left and right characteristic image [I that looks _l ^chara(x, y)] _{m * n}[I _r ^chara(x, y)] _{m * n}feature between correlativity

To the left and right characteristic image [I that looks _l ^chara(x, y)] _{m * n}[I _r ^chara(x, y)] _{m * n}the processing of being labelled, and order is from left to right arranged tag number according to connected domain central series coordinate, by [I _l ^chara(x, y)] _{m * n}tag number to sort successively be 1,2 ..., Lab _l, by [I _r ^chara(x, y)] _{m * n}tag number to sort successively be 1,2 ..., Lab _r; Afterwards, obtain the left and right feature correlation of looking characteristic image,

Feature correlation comprises following four correlation parameters:

1) characteristic image [I is looked on a left side _l ^chara(x, y)] _{m * n}characteristic curve i and the right side look characteristic image [I _r ^chara(x, y)] _{m * n}the common row number N of characteristic curve j ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _r,

At first, calculate the capable terminal point coordinate b (i, j) between the left and right characteristic curve of looking characteristic image, i=1,2 according to formula (10) ... Lab _l, j=1,2 ... Lab _r:

i＝1,2…Lab _L，j＝1,2,…Lab _R， （10）

Wherein, b _l(i) be that characteristic image [I is looked on a left side _l ^chara(x, y)] _{m * n}the capable terminal point coordinate of characteristic curve i, b _r(j) be that characteristic image [I is looked on the right side _r ^chara(x, y)] _{m * n}the capable terminal point coordinate of characteristic curve j;

Afterwards, calculate the beginning-of-line coordinate t (i, j) between the left and right characteristic curve of looking characteristic image, i=1,2 according to formula (11) ... Lab _l, j=1,2 ... Lab _r:

i＝1,2…Lab _L，j＝1,2,…Lab _R， （11）

T _l(i) be that characteristic image [I is looked on a left side _l ^chara(x, y)] _{m * n}the beginning-of-line coordinate of characteristic curve i, t _r(j) be that characteristic image [I is looked on the right side _r ^chara(x, y)] _{m * n}the beginning-of-line coordinate of characteristic curve j;

The beginning-of-line coordinate t (i, j) obtained according to formula (10) and formula (11), i=1,2 ... Lab _l, j=1,2 ... Lab _r, and row terminal point coordinate b (i, j), i=1,2 ... Lab _l, j=1,2 ... Lab _r, obtain a left side according to formula (12) and look characteristic image [I _l ^chara(x, y)] _{m * n}characteristic curve i, characteristic image [I is looked on the right side _r ^chara(x, y)] _{m * n}the common row number N of characteristic curve j ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _r:

N ^Match(i,j)=b(i,j)-t(i,j)，i＝1,2…Lab _L，j＝1,2,…Lab _R， （12）

2) characteristic image [I is looked on the left side obtained according to formula (12) _l ^chara(x, y)] _{m * n}characteristic curve i and the right side look characteristic image [I _r ^chara(x, y)] _{m * n}the common row number N of characteristic curve j ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _r, obtain a left side according to formula (13) and look characteristic image [I _l ^chara(x, y)] _{m * n}characteristic curve i and the right side look characteristic image the common row of characteristic curve j account for the number ratio L of short characteristic curve ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _r,

$L^{\mathrm{Match}}(i,j)=\frac{N^{\mathrm{Match}}(i,j)}{L(i,j)},$ i＝1,2,…,Lab _L，j＝1,2,…,Lab _R， （13）

Wherein:

i＝1,2,…,Lab _L，j＝1,2,…,Lab _R， （14)

3) calculate a left side according to formula (15) and look characteristic image [I _l ^chara(x, y)] _{m * n}characteristic curve i and the right side look characteristic image [I _r ^chara(x, y)] _{m * n}the slope differences of characteristic curve j

i=1,2 ..., Lab _l, j=1,2 ..., Lab _r,

If (x _t ^l(i), y _t ^l(i)) for a left side, look characteristic image the starting point coordinate of characteristic curve i, (x _b ^l(i), y _b ^l(i)) be the terminal point coordinate of its characteristic curve i; (x _t ^r(j), y _t ^r(j)) for the right side, look characteristic image [I _r ^chara(x, y)] _{m * n}the starting point coordinate of characteristic curve j, (x _b ^r(j), y _b ^r(j)) be the terminal point coordinate of its characteristic curve j:

$S^{\mathrm{Match}}(i,j)=|\frac{{y_b}^L\left(i\right)-{y_t}^L\left(i\right)}{{x_b}^L\left(i\right)-{x_t}^L\left(i\right)}-\frac{{y_b}^R\left(j\right)-{y_t}^R\left(j\right)}{{x_b}^R\left(j\right)-{y_j}^R\left(j\right)}|,$ i＝1,2,…,Lab _L，j＝1,2,…,Lab _R， （15）

4), according to formula (16), calculate a left side and look characteristic image [I _l ^chara(x, y)] _{m * n}the corresponding left view [I of characteristic curve i _l(x, y)] _{m * n}, and characteristic image [I is looked on the right side _r ^chara(x, y)] _{m * n}the corresponding right view [I of characteristic curve j _r(x, y)] _{m * n}around characteristic block gray scale difference accumulative total and

i=1,2 ..., Lab _l, j=1,2 ..., Lab _r:

$R^{\mathrm{Match}}(i,j)=\underset{(x,y)\&Element;{\mathrm{\&Omega;}}_{i,j}}{\mathrm{\&Sigma;}}|I_L(x,y)-I_R(x,y)|,$ I=1,2 ..., Lab _l, j=1,2 ..., Lab _r, (16) wherein, Ω _i,jfor looking characteristic image [I in a left side _l ^chara(x, y)] _{m * n}characteristic curve i expand to the left and right sides Δ i, Δ i is spreading range, the experience value, span is Δ i ∈ [10,30]; Characteristic image [I is looked on the right side _r ^chara(x, y)] _{m * n}characteristic curve j expand to the left and right sides Δ j, Δ j is spreading range, experience value, the zone that span is Δ j ∈ [10,30];

Step 3b, the relative coefficient of obtaining according to step 3a find optimum matching pair

1), according to formula (17), obtain the left side obtained according to formula (12) and look characteristic image [I _l ^chara(x, y)] _{m * n}characteristic curve i and the right side look characteristic image [I _r ^chara(x, y)] _{m * n}the common row number N of characteristic curve j ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _rmaximal value N ^match _max:

N ^Match _max＝max{N ^Match(i,j)|i＝1,2,…,Lab _L;j＝1,2,…,Lab _R}， （17）

According to formula (18), obtain the left side obtained according to formula (13) and look characteristic image [I again _l ^chara(x, y)] _{m * n}characteristic curve i and the right side look characteristic image [I _r ^chara(x, y)] _{m * n}the common row of characteristic curve j account for the number ratio L of short characteristic curve ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _r, maximal value L ^match _max:

L ^Match _max＝max{L ^Match(i,j)|i＝1,2,…,Lab _L;j＝1,2,…,Lab _R}， （18）

2), according to formula (19), obtain the left side obtained according to formula (15) and look characteristic image [I _l ^chara(x, y)] _{m * n}characteristic curve i and the right side look characteristic image [I _r ^chara(x, y)] _{m * n}the slope differences of characteristic curve j

i=1,2 ..., Lab _l, j=1,2 ..., Lab _rminimum value S ^match _min:

S ^Match _min＝min{S ^Match(i,j)|i＝1,2,…,Lab _L;j＝1,2,…,Lab _R}， （19）

Build the matrix that mark can not make left and right view feature pairing

its value is:

i＝1,2,…,Lab _L，j＝1,2,…,Lab _R，

Wherein, k _max, k _minbe characteristic curve matching degree parameter, the similarity degree assignment according between acceptable characteristic curve, be generally the experience value, and span is k _max∈ [0.5,0.7], k _min∈ [2,6];

3) according to matching the mark matrix with according to formula (16), characteristic image [I is looked on the left side obtained _l ^chara(x, y)] _{m * n}the corresponding left view [I of characteristic curve i _l(x, y)] _{m * n}look characteristic image [I with the right side _r ^chara(x, y)] _{m * n}the corresponding left view [I of characteristic curve j _r(x, y)] _{m * n}around gray scale difference accumulative total and the R of characteristic block ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _r, obtain the right sequence number (i*, j*) of pairing characteristic curve of looking the same target in characteristic image when front left and right, computing formula is:

$(i*,j*)\underset{i,j}{\arg }\min \left\{R^{\mathrm{Match}}(i,j)\right|i=\mathrm{1,2},...,{\mathrm{Lab}}_L;j=1,2,...,{\mathrm{Lab}}_R;a(i,j)=0\},---(21)$

Step 3c, repeating step 3b, until all characteristic curves have matched;

Step 4, obtain the stereoscopic article elevation information

Step 4a, obtain the characteristic curve elevation

Characteristic image [I is looked on a traversal left side _l ^chara(x, y)] _{m * n}, find successively the right side to look characteristic image [I _r ^chara(x, y)] _{m * n}in the characteristic curve of all successful matchings to resequencing as (i, i), i=1,2 ..., Lab _lR, Lab wherein _lRthe right number of characteristic curve of successful matching,

If i is matched line to (i, i), i=1,2 ..., Lab _lRcharacteristic curve, look characteristic image on a left side

row-coordinate be x _i ^l, the row coordinate is y _i ^l; Look characteristic image [I on the right side _r ^chara(x, y)] _{m * n}in, the characteristic curve matched with it, on a left side, looking in characteristic image (is x on the identical row-coordinate of characteristic curve _i ^r=x _i ^l) the row coordinate be y _i ^r, claim e _i=y _i ^r-y _i ^lfor its parallax, i=1,2 ..., Lab _lR,

The elevation of three-dimensional target is obtained by disparity computation, establishes elevation matrix [h (x, y)] _{m * n}, be initialized as full null matrix,

With reference to Fig. 1, be binocular camera imaging model schematic diagram, wherein O _lleft camera, O _rbe right camera, establish and be imaged on a left side and look characteristic image [I _l ^chara(x, y)] _{m * n}look characteristic image [I with the right side _r ^chara(x, y)] _{m * n}the pairing characteristic curve on aerial image vegetarian refreshments (x, y, z), look characteristic image [I on a left side _l ^chara(x, y)] _{m * n}the row coordinate of middle imaging point is yl, looks characteristic image [I on the right side _r ^chara(x, y)] _{m * n}in the row coordinate be y _r, corresponding to its i, match line to (i, i), y _l=y _i ^l, y _r=y _i ^r, f is camera focus, and z is the distance of spatial point from camera plane, and b is the distance of left and right camera, by imaging model, is obtained:

$\frac{b}{z}=\frac{(b+y_r)-y_l}{z-f},$ That is: $z=\frac{b\&CenterDot;f}{y_l-y_r},---\left(22\right)$

h(x,y)＝C-z， （23）

Wherein, C is the fixing camera plane distance to the automobile chassis reference field, and h (x, y) is exactly the elevation on the characteristic curve that detects of required automobile chassis stereoscopic article;

Step 4b, find paired elevation characteristic curve

After step 2 has been removed transverse edge, three-dimensional target is just formed by two vertical surrounded by edges, therefore, according to step 4a, the elevation matrix [h (x, y)] by the characteristic curve of step 3 successful matching to obtaining respectively its characteristic curve _{m * n}, then look characteristic image [I according to a left side _l ^chara(x, y)] _{m * n}characteristic curve i, i=1,2 ..., Lab _lR, find it at matrix [h (x, y)] _{m * n}elevation on upper relevant position, according in the elevation matrix, the elevation of two characteristic curves is the most approaching, and the row-coordinate of two characteristic curves is all maximum principle mutually, detect in elevation map two characteristic curves that belong to same target, these two characteristic curves are just as the left and right boundary line of same three-dimensional target;

Step 4c, repeating step 4b, the elevation characteristic curve that meets determining step 4b until all has matched;

Step 4d, will be judged to be the altitude traverse that belongs to same target included center section will be defined as to target area, and give paired altitude traverse height value by it, complete the foundation to the three-dimensional elevation map of three-dimensional object.

By above step, just realize accurately that the elevation of automobile chassis 3 D stereo target detects.The elevation that method of the present invention also can extend to other objectives based on the imaging of binocular line array CCD detects.

Claims (5)

1. the three-dimensional elevation of the automobile chassis based on the binocular line array CCD is determined method, it is characterized in that, according to following steps, specifically implements:

Step 1, the automobile chassis image that the binocular line array CCD is collected carry out gray scale closed operation processing;

The feature of the gray scale closed operation result that step 2, extraction step 1 obtain;

Step 3, left and right view is carried out to characteristic matching;

Step 4, obtain the stereoscopic article elevation information.

2. the three-dimensional elevation of automobile chassis based on the binocular line array CCD according to claim 1 is determined method, it is characterized in that: the detailed step of described step 1 is,

If the image that the binocular line array CCD collects size is m * n, by left and right view difference called after [I _l(x, y)] _{m * n}and [I _r(x, y)] _{m * n}, to utilize Gauss's template respectively left and right view to be done to the gray scale closed operation and process, the Gauss's template arranged in order to do the closed operation processing is [G (x, y)] _{m * n}, computing formula is:

$G(x,y)=K\&CenterDot;\frac{1}{2\mathrm{\&pi;}{\mathrm{\&sigma;}}^2}\exp (-\frac{x^2+y^2}{2{\mathrm{\&sigma;}}^2}),$ x＝1,2,…,m;y＝1,2,…,n （1）

Wherein, K is scale-up factor, for the image of 256 GTGs, and K ∈ [80,100], σ ∈ [10,20];

Gauss's template of utilizing formula (1) to obtain, respectively to left view [I _l(x, y)] _{m * n}and right view [I _r(x, y)] _{m * n}carry out the gray scale expansion process,

According to formula (2) and template [G (x, y)] _{m * n}carry out gray scale and computing, obtain result and be

with

$\widetilde{I_k}(x,y)=I_k(x,y)+G(x,y),$ x＝1,2,…,m;y＝1,2,…,n；k＝L,R （2）

Afterwards, according to

with

obtain the gray scale expansion results according to formula (3)

with

x＝1,2,…,m;y＝1,2,…,n；k＝L,R （3）

Left and right view after the gray scale expansion process is carried out to the gray scale corrosion treatment again,

According to formula (4), formula (3) is obtained

with carry out and template [G (x, y)] _{m * n}the gray scale difference computing, obtain result and be

with

${\stackrel{\_}{I}}_k(x,y)={\hat{I}}_k(x,y)-G(x,y),$ x＝1,2,…,m;y＝1,2,…,n；k＝L,R， （4）

Afterwards, according to

with

obtain the result [I of gray scale closed operation according to formula (5) _l ^close(x, y)] _{m * n}[I _r ^close(x, y)] _{m * n}:

x＝1,2,…,m;y＝1,2,…,n；k＝L,R。（5）

3. the three-dimensional elevation of automobile chassis based on the binocular line array CCD according to claim 1 is determined method, it is characterized in that: the detailed step of described step 2 is,

Step 2a, the image [I after step 1 is processed _l ^close(x, y)] _{m * n}[I _r ^close(x, y)] _{m * n}extract the canny edge, obtain edge image [I _l ^edge(x, y)] _{m * n}[I _r ^edge(x, y)] _{m * n}, afterwards, remove the interference of small size,

To image [I _l ^edge(x, y)] _{m * n}[I _r ^edge(x, y)] _{m * n}carry out the processing of labelling of eight connected domains, establish after connected domain is labelled image [I _l ^edge(x, y)] _{m * n}obtain N _lindividual connected domain, the area of each connected domain is Area _i ^l, i is tag number, i=1, and 2 ..., N _l; Image [I _r ^edge(x, y)] _{m * n}obtain N _rindividual connected domain, the area of each connected domain is Area _i ^r, i is tag number, i=1, and 2 ..., N _r;

Set afterwards the threshold value A rea_Th that eliminates noise, all connected domains are processed, when the area of connected domain is greater than Area_Th, retain this connected domain; If while being less than Area_Th, think noise, remove this connected domain;

Step 2b, remove the transverse edge in edge image

The rectangle template [M (x, y)] that the effect template is a p * q size is set _{p * q}, M (x, y)=1, x=1,2 ..., p; Y=1,2 ..., q; The p value is odd number, and span is p ∈ { 3,5,7}; The q value is odd number, span be q ∈ 5,7,9,11},

By template [M (x, y)] _{p * q}act on two image [I that step 2a obtains _l ^edge(x, y)] _{m * n}[I _r ^edge(x, y)] _{m * n}, on the position of the center of template corresponding to the pending pixel of image, carry out following computing, obtained removing the characteristic image [MI of transverse edge _l ^edge(x, y)] _{m * n}and [MI _r ^edge(x, y)] _{m * n}:

${{\mathrm{MI}}_k}^{\mathrm{edge}}(x,y)=\left\{\begin{array}{cc}0& \underset{i=x-(p-1)/2}{\overset{x+(p-1)/2}{\mathrm{\&Pi;}}}\underset{j=y-(q-1)/2}{\overset{y+(q-1)/2}{\mathrm{\&Sigma;}}}{I_k}^{\mathrm{edge}}(x+i,y+j)\&CenterDot;M(i,j)=0\\ 1& \underset{i=x-(p-1)/2}{\overset{x+(p-1)/2}{\mathrm{\&Pi;}}}\underset{j=y-(q-1)/2}{\overset{y+(q-1)/2}{\mathrm{\&Sigma;}}}{I_k}^{\mathrm{edge}}(x+i,y+j)\&CenterDot;M(i,j)\&NotEqual;0\end{array},---\left(6\right)\right.$

X=wherein (p+1)/2, (p+1)/2+1 ..., m-(p+1)/2,

y＝(q+1)/2,(q+1)/2+1,…,n-(q+1)/2，k＝L,R

MI _k ^edge(x,y)＝0， （7）

X=1 wherein, 2 ..., (p+1)/2-1,

y＝1,2,…,(q+1)/2-1，k＝L,R；

Step 2c, with reference to step 2a, remove image [MI _l ^edge(x, y)] _{m * n}[MI _r ^edge(x, y)] _{m * n}in the small size zone, will remove the image of surface area still with [MI _l ^edge(x, y)] _{m * n}[MI _r ^edge(x, y)] _{m * n}meaned;

Step 2d, remove unreliable marginal information

Suppose, left and right view is carried out to registration, the alternate position spike of the reference field after the establishing standard in left and right view is Δ x and Δ y, disparity map [the ▽ I in order to judge stereoscopic article _l,R(x, y)] _{m * n}computing formula be:

$\&dtri;I_{L,R}(x,y)=\left\{\begin{array}{cc}0& |I_L(x,y)-I_R(x+\mathrm{\&Delta;x},y+\mathrm{\&Delta;y})|<\mathrm{\&epsiv;}\\ 1& |I_L(x,y)-I_R(x+\mathrm{\&Delta;x},y+\mathrm{\&Delta;y})|\&GreaterEqual;\mathrm{\&epsiv;}\end{array},\right.$ x＝1,2,…,m;y＝1,2,…,n （8）

Wherein, ε is the adjustment parameter of parallax value, and span is ε ∈ [15,30];

Afterwards, at disparity map [▽ I _l,R(x, y)] _{m * n}upper will in the edge feature obtained in step 2, do not belong to the part of stereoscopic article remove;

By [MI _l ^edge(x, y)] _{m * n}[MI _r ^edge(x, y)] _{m * n}respectively with [▽ I _l,R(x, y)] _{m * n}carry out AND operation, obtain the left and right boundary characteristic image [I of stereoscopic article _l ^chara(x, y)] _{m * n}[I _r ^chara(x, y)] _{m * n}:

X=1 wherein, 2 ..., m; Y=1,2 ..., n, k=L, R.

4. the three-dimensional elevation of automobile chassis based on the binocular line array CCD according to claim 1 is determined method, it is characterized in that: the detailed step of described step 3 is,

Step 3a, ask the left and right characteristic image [I that looks _l ^chara(x, y)] _{m * n}[I _r ^chara(x, y)] _{m * n}feature between correlativity

To the left and right characteristic image [I that looks _l ^chara(x, y)] _{m * n}[I _r ^chara(x, y)] _{m * n}the processing of being labelled, and order is from left to right arranged tag number according to connected domain central series coordinate, by [I _l ^chara(x, y)] _{m * n}tag number to sort successively be 1,2 ..., Lab _l, by [I _r ^chara(x, y)] _{m * n}tag number to sort successively be 1,2 ..., Lab _r; Afterwards, obtain the left and right feature correlation of looking characteristic image,

Feature correlation comprises following four correlation parameters:

1) characteristic image [I is looked on a left side _l ^chara(x, y)] _{m * n}characteristic curve i and the right side look characteristic image [I _r ^chara(x, y)] _{m * n}the common row number N of characteristic curve j ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _r,

At first, calculate the capable terminal point coordinate b (i, j) between the left and right characteristic curve of looking characteristic image, i=1,2 according to formula (10) ... Lab _l, j=1,2 ... Lab _r:

i＝1,2…Lab _L，j＝1,2,…Lab _R， （10）

Wherein, b _l(i) be that characteristic image [I is looked on a left side _l ^chara(x, y)] _{m * n}the capable terminal point coordinate of characteristic curve i, b _r(j) be that characteristic image [I is looked on the right side _r ^chara(x, y)] _{m * n}the capable terminal point coordinate of characteristic curve j;

Afterwards, calculate the beginning-of-line coordinate t (i, j) between the left and right characteristic curve of looking characteristic image, i=1,2 according to formula (11) ... Lab _l, j=1,2 ... Lab _r:

i=1,2 ... Lab _l, j=1,2 ... Lab _r, (11) t _l(i) be that characteristic image [I is looked on a left side _l ^chara(x, y)] _{m * n}the beginning-of-line coordinate of characteristic curve i, t _r(j) be that characteristic image [I is looked on the right side _r ^chara(x, y)] _{m * n}the beginning-of-line coordinate of characteristic curve j;

The beginning-of-line coordinate t (i, j) obtained according to formula (10) and formula (11), i=1,2 ... Lab _l, j=1,2 ... Lab _r, and row terminal point coordinate b (i, j), i=1,2 ... Lab _l, j=1,2 ... Lab _r, obtain a left side according to formula (12) and look characteristic image [I _l ^chara(x, y)] _{m * n}characteristic curve i, characteristic image [I is looked on the right side _r ^chara(x, y)] _{m * n}the common row number N of characteristic curve j ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _r:

N ^Match(i,j)=b(i,j)-t(i,j)，i＝1,2…Lab _L，j＝1,2,…Lab _R， （12）

2) characteristic image [I is looked on the left side obtained according to formula (12) _l ^chara(x, y)] _{m * n}characteristic curve i and the right side look characteristic image [I _r ^chara(x, y)] _{m * n}the common row number N of characteristic curve j ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _r, obtain a left side according to formula (13) and look characteristic image [I _l ^chara(x, y)] _{m * n}characteristic curve i and the right side look characteristic image [I _r ^chara(x, y)] _{m * n}the common row of characteristic curve j account for the number ratio L of short characteristic curve ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _r,

$L^{\mathrm{Match}}(i,j)=\frac{N^{\mathrm{match}}(i,j)}{L(i,j)},$ i＝1,2,…,Lab _L，j＝1,2,…,Lab _R， （13）

Wherein:

i＝1,2,…,Lab _L，j＝1,2,…,Lab _R， （14)

3) calculate a left side according to formula (15) and look characteristic image [I _l ^chara(x, y)] _{m * n}characteristic curve i and the right side look characteristic image [I _r ^chara(x, y)] _{m * n}the slope differences S of characteristic curve j ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _r,

If (x _t ^l(i), y _t ^l(i)) for a left side, look characteristic image [I _l ^chara(x, y)] _{m * n}the starting point coordinate of characteristic curve i, (x _b ^l(i), y _b ^l(i)) be the terminal point coordinate of its characteristic curve i; (x _t ^r(j), y _t ^r(j)) for the right side, look characteristic image [I _r ^chara(x, y)] _{m * n}the starting point coordinate of characteristic curve j, (x _b ^r(j), y _b ^r(j)) be the terminal point coordinate of its characteristic curve j:

$S^{\mathrm{Match}}(i,j)=|\frac{{y_b}^L\left(i\right)-{y_t}^L\left(i\right)}{{x_b}^L\left(i\right)-{x_t}^L\left(i\right)}-\frac{{y_b}^R\left(j\right)-{y_t}^R\left(j\right)}{{x_b}^R\left(j\right)-{y_t}^R\left(j\right)}|,$ i＝1,2,…,Lab _L，j＝1,2,…,Lab _R， （15）

4), according to formula (16), calculate a left side and look characteristic image [I _l ^chara(x, y)] _{m * n}the corresponding left view [I of characteristic curve i _l(x, y)] _{m * n}look characteristic image [I with the right side _r ^chara(x, y)] _{m * n}the corresponding right view [I of characteristic curve j _r(x, y)] _{m * n}around gray scale difference accumulative total and the R of characteristic block ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _r:

$R^{\mathrm{Match}}(i,j)=\underset{(x,y)\&Element;{\mathrm{\&Omega;}}_{i,j}}{\mathrm{\&Sigma;}}|I_L(x,y)-I_R(x,y)|,$ i＝1,2,…,Lab _L，j＝1,2,…,Lab _R， （16）

Wherein, Ω _i,jfor looking characteristic image [I in a left side _l ^chara(x, y)] _{m * n}characteristic curve i expand to the left and right sides Δ i, Δ i is spreading range, span is Δ i ∈ [10,30]; Characteristic image [I is looked on the right side _r ^chara(x, y)] _{m * n}characteristic curve j expand to the left and right sides Δ j, Δ j is spreading range, span is Δ j ∈ [10,30];

Step 3b, the relative coefficient of obtaining according to step 3a find optimum matching pair

1), according to formula (17), obtain the left side obtained obtained according to formula (12) and look characteristic image [I _l ^chara(x, y)] _{m * n}characteristic curve i and the right side look characteristic image [I _r ^chara(x, y)] _{m * n}the common row number N of characteristic curve j ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _rmaximal value N ^match _max:

N ^Match _max＝max{N ^Match(i,j)|i＝1,2,…,Lab _L;j＝1,2,…,Lab _R}， （17）

Again according to formula (18), obtain and obtain a left side according to formula (13) and look characteristic image [I _l ^chara(x, y)] _{m * n}characteristic curve i and the right side look characteristic image [I _r ^chara(x, y)] _{m * n}the common row of characteristic curve j account for the number ratio L of short characteristic curve ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _r, maximal value L ^match _max:

L ^Match _max＝max{L ^Match(i,j)|i＝1,2,…,Lab _L;j＝1,2,…,Lab _R}， （18）

2), according to formula (19), obtain the left side obtained according to formula (15) and look characteristic image [I _l ^chara(x, y)] _{m * n}characteristic curve i and the right side look characteristic image [I _r ^chara(x, y)] _{m * n}the slope differences S of characteristic curve j ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _r, the minimum value S of this slope differences ^match _minbe:

S ^Match _min＝min{S ^Match(i,j)|i＝1,2,…,Lab _L;j＝1,2,…,Lab _R}， （19）

Build the matrix that mark can not make left and right view feature pairing

its value is:

i＝1,2,…,Lab _L，j＝1,2,…,Lab _R，

K _max, k _minbe characteristic curve matching degree parameter, span is k _max∈ [0.5,0.7], k _min∈ [2,6];

3) according to matching the mark matrix with according to formula (16), characteristic image [I is looked on the left side obtained _l ^chara(x, y)] _{m * n}the corresponding left view [I of characteristic curve i _l(x, y)] _{m * n}look characteristic image [I with the right side _r ^chara(x, y)] _{m * n}the corresponding left view [I of characteristic curve j _r(x, y)] _{m * n}around gray scale difference accumulative total and the R of characteristic block ^match(i, j), i=1,2 ..., Lab _l, j=1,2 ..., Lab _r, obtain the right sequence number (i*, j*) of pairing characteristic curve of looking the same target in characteristic image when front left and right, computing formula is:

$(i*,j*)=\underset{i,j}{\arg }\min \left\{R^{\mathrm{Match}}(i,j)\right|i=\mathrm{1,2},...,{\mathrm{Lab}}_L;j=\mathrm{1,2},...,{\mathrm{Lab}}_R;a(i,j)=0\},---\left(21\right)$

Step 3c, repeating step 3b, until all characteristic curves have matched.

5. the three-dimensional elevation of automobile chassis based on the binocular line array CCD according to claim 1 is determined method, it is characterized in that: the detailed step of described step 4 is,

Step 4a, obtain the characteristic curve elevation

Characteristic image [I is looked on a traversal left side _l ^chara(x, y)] _{m * n}, find successively the right side to look characteristic image [I _r ^chara(x, y)] _{m * n}in the characteristic curve pair of all successful matchings, resequence as (i, i), i=1,2 ..., Lab _lR, Lab wherein _lRthe right number of characteristic curve of successful matching,

If i is matched line to (i, i), i=1,2 ..., Lab _lRcharacteristic curve, look characteristic image [I on a left side _l ^chara(x, y)] _{m * n}row-coordinate be x _i ^l, the row coordinate is y _i ^l; Look characteristic image [I on the right side _r ^chara(x, y)] _{m * n}in, the characteristic curve matched with it, on a left side, looking in characteristic image (is x on the identical row-coordinate of characteristic curve _i ^r=x _i ^l) the row coordinate be y _i ^r, claim e _i=y _i ^r-y _i ^lfor its parallax, i=1,2 ..., Lab _lR,

The elevation of three-dimensional target is obtained by disparity computation, establishes elevation matrix [h (x, y)] _{m * n}, be initialized as full null matrix,

If O _lleft camera, O _rbe right camera, establish and be imaged on a left side and look characteristic image [I _l ^chara(x, y)] _{m * n}look characteristic image [I with the right side _r ^chara(x, y)] _{m * n}the pairing characteristic curve on aerial image vegetarian refreshments (x, y, z), look characteristic image [I on a left side _l ^chara(x, y)] _{m * n}the row coordinate of middle imaging point is y _l, look characteristic image [I on the right side _r ^chara(x, y)] _{m * n}in the row coordinate be y _r, corresponding to its i, match line to (i, i), y _l=y _i ^l, y _r=y _i ^r, f is camera focus, and z is the distance of spatial point from camera plane, and b is the distance of left and right camera, by imaging model, is obtained:

$\frac{b}{z}=\frac{(b+y_r)-y_l}{z-f},$ That is: $z=\frac{b\&CenterDot;f}{y_l-y_r},---\left(22\right)$

h(x,y)＝C-z， （23）

Wherein, C is the fixing camera plane distance to the automobile chassis reference field, and h (x, y) is exactly the elevation on the characteristic curve that detects of required automobile chassis stereoscopic article;

Step 4b, find paired elevation characteristic curve

After step 2 has been removed transverse edge, three-dimensional target is just formed by two vertical surrounded by edges, therefore, according to step 4a, the elevation matrix [h (x, y)] by the characteristic curve of step 3 successful matching to obtaining respectively its characteristic curve _{m * n}, then look characteristic image [I according to a left side _l ^chara(x, y)] _{m * n}characteristic curve i, i=1,2 ..., Lab _lR, find it at matrix [h (x, y)] _{m * n}elevation on upper relevant position, according in the elevation matrix, the elevation of two characteristic curves is the most approaching, and the row-coordinate of two characteristic curves is all maximum principle mutually, detect in elevation map two characteristic curves that belong to same target, these two characteristic curves are just as the left and right boundary line of same three-dimensional target;

Step 4c, repeating step 4b, until the elevation characteristic curve of all satisfied judgements has matched;

Step 4d, will be judged to be the altitude traverse that belongs to same target included center section will be defined as to target area, and give paired altitude traverse height value by it, complete the foundation of the three-dimensional elevation map of three-dimensional object.

Images