CN101986322B - Structured light stripe central point reliability evaluation method - Google Patents

Abstract

The invention relates to a structured light stripe central point reliability evaluation method. An evaluation object is a light stripe central point obtained from a light stripe image by adopting a method, and structured light stripe central point evaluation basis is energy sum of a Gauss normalization model which takes the central point as the center on the cross section in the normal direction of the light stripe. The method includes: a light stripe central point is extracted by adopting a method, a point sequence of cross section in the normal direction of light stripe at the central point is acquired, and Gauss model normalization is carried out on the point sequence of the cross direction in the normal direction of the light stripe; energy sum of normalization Gauss model of the point sequence of cross section in the normal direction of the light stripe is calculated, and reliability evaluation index of the point is obtained; grey scale sum of the calculated light stripe central point cross section normalization Gauss model sequence is calculated; normalization of global image is carried out on the obtained central point reliability evaluation index, reliability is normalized according to reliability maximum of the image, and finally normalized reliability evaluation is obtained. The evaluation on light stripe central point in the invention accords with actual evaluation requirement.

Description

Light strip center of structured light point reliability evaluation method

Technical field

The invention belongs to the machine vision technique field, relate to the reliability evaluation method of the structure striation central point that extraction is obtained.

Background technology

Along with the development of measuring technique, dimensional visual measurement is widely used in precision measurements scenes such as industry, traffic owing to have characteristics such as noncontact, dynamic response is fast, system flexibility is good.In structured light vision detection, the high precision image coordinate that obtains structure striation center is structured light vision sensor calibration and the committed step that obtains the testee three-D profile, also is the first-hand information of measuring.

During the structure striation was measured, optical strip image finally caused structure striation gray scale, width on the piece image inconsistent owing to environment, perspective imaging, object material such as deposit at reason.For the three-dimensional structure photo measure; Usually can run into condition of severe more; Index of actual needs is estimated the credibility that extracts the striation central point obtain, obtains " threshold value " of a central point comprehensive evaluation, the existence of judging signal whether, distinguish signal and noise.Under the situation of the overall target that keeps original point, check that which point meets signal standards more, guarantees the reliability of data.Simultaneously, the result to the evaluation of striation central point confidence level also can be used as the foundation of estimating the good and bad of optical strip image and estimating the quality of striation method for distilling.

Common light stripe centric line method for distilling, like extremum method, grey scale centre of gravity method, direction template method etc., though simple, real-time, precision is not high, and because these methods are the basis with the Threshold Segmentation, mostly to noise-sensitive; And the robustness of curve fitting method, Steger algorithm is good, precision is high, but algorithm complex is bigger.The striation central point that existing method obtains all is that the binaryzation of striation central point is represented, promptly only is the position of finding structure striation central point.And the structured light three-dimensional vision measurement is faced with complicated situation; Particularly outdoor measurement; Often run into that the speckle that brings like strong surround lighting, unlike material, color and vein absorb, the abominable situation of reflection interference striation or the like; Make the structure striation that distortion greatly take place, it is inconsistent on an auxiliary structure optical strip image, can to demonstrate the striation width, and striation central point gray scale is inconsistent.Carry out the three-dimensional structure photo measure in these cases, need know the credibility of measurement result.For obtaining three-dimensional structure photo measure result's confidence level, the first step is to obtain the credibility that the striation central point extracts.The confidence level evaluation of the striation central point that therefore said method is obtained is very important, is the key issue that needs to be resolved hurrily in the three-dimensional structure light vision measurement.

Summary of the invention

The present invention seeks to: propose a kind of striation central point reliability evaluation method that structure striation extraction with certain width is obtained.The credibility that the reliability definition of structure striation central point is the striation central point for this point.Owing to reasons such as environment, perspective imaging, object material cause structure striation noise, gray scale, width on the width of cloth optical strip image inconsistent; Under different situations and demand; Utilize the extraction of the method for different principle algorithm, the striation central point that extraction is obtained departs from the real center point.To this situation; Reliability for the striation central point that guarantees to obtain; The present invention provides a kind of evaluation of programme to striation central point confidence level, and extracting the central point that obtains with evaluation is the confidence level of striation real center point, satisfies the reliability assessment demand that final measurement obtains simultaneously.

Technical scheme of the present invention is: light strip center of structured light point reliability evaluation method; Its evaluation object is the striation central point that obtains through a kind of method on the optical strip image, the evaluation of structure striation central point according to for the energy of the Gaussian normalization model that on striation normal direction xsect, is the center with this point with.This method comprises:

To extracting the striation central point that obtains through a kind of method, obtain the striation normal direction xsect point sequence at this some place, the central point that obtains with extraction is the center, and the point sequence of striation normal direction xsect is carried out the normalization of Gauss model;

The energy that calculates striation normal direction xsect point sequence normalization Gauss model with, obtain the reliability evaluation index of this point; To the striation central point xsect normalization Gauss model sequence of trying to achieve ask gray scale with:

For C _jCentral point: E _j=∑ g (X _i), be the reliability index of needed this point;

Central point reliability evaluation index to obtaining is carried out the normalization of global image, carries out normalization by the reliability maximal value of this image and handles reliability: R _j=E _j/ E _Jmax, E _Max=max (E _j), j=1～K, K finally obtains the evaluation of normalization reliability for extracting the striation central point sum that obtains.

Wherein, to the striation normal direction xsect point sequence of structure striation central spot, and be the center with the central point that extraction obtains, the point sequence of xsect is carried out the normalization of Gauss model, specifically comprise:

To the striation normal direction xsect point sequence of the striation central spot obtained, estimating should the normal direction cross-sectional width w of place _j: estimate this place's local loop back noise and remove; The best-fit Gauss model obtains final normalization model; Estimate in the above-mentioned normalization Gauss model method that the xsect effective width is: be the amplitude of Gauss model with the central point gray-scale value; Center position is the center of Gauss model; Extend the point of 5%～20% the gray scale of searching about amplitude successively along striation normal both sides and locate the starting point and the terminating point of striation normal direction xsect as this; Sequence with between 2 of starting point and the terminating points is the striation ordered sequence, and its length is for estimating live width; Perhaps locate the effective width w of striation cross-sectional width for this with the average live width of global image _j

Estimate in the above-mentioned normalization Gauss model method that local loop back noise and the method for removing are:

At this standardized width in striation xsect left and right sides, place is w _jRegional area is done on average with the estimated noise substrate gray-scale value in this scope; Also can locate the standardized width in the striation xsect left and right sides at this is w _jRegional area carries out statistics of histogram in this part, from gray-scale value low to the gray-scale value height ask quantity with, get 1/2～2/3 the point of always counting and do average estimated noise substrate;

The original gray scale of xsect sequence deducts noise floor, the sequence gray-scale value behind the noise that promptly is removed, and this gray-scale value is not for negative simultaneously.

The method of best-fit Gauss model comprises in the above-mentioned normalization Gauss model method:

The best-fit Gauss model obtains final normalization model: for the gray scale sequence of removing denoising, central point C _j(=X _c) denoising after gray scale be p _0cj, carry out Gauss model normalization match.Wherein, with C _jBe Gauss model center position μ _x, be the Gauss model peak A with gray-scale value after the denoising of this point:

${\mathrm{<figure-callout\; id="0"\; label="p"\; filenames="HDA0000033518570000051.png"\; state="\{\{state\}\}">p</figure-callout>}}_{0\mathrm{ij}}={\mathrm{<figure-callout\; id="0"\; label="p"\; filenames="HDA0000033518570000051.png"\; state="\{\{state\}\}">p</figure-callout>}}_{0\mathrm{cj}}\exp [-\frac{{(X_i-X_c)}^2}{{2\mathrm{\&sigma;}}^2}],$ (1≤a-w _j≤i≤b+w _j≤N _j)

Then has only a unknown quantity σ in the Gauss model;

The method of best-fit Gauss model comprises in the above-mentioned normalization Gauss model method:

For the gray scale sequence of removing denoising, central point C _j(=X _c) denoising after gray scale be p _0cj, carry out Gauss model normalization match.Wherein, with C _jBe Gauss model center position μ _x, be the Gauss model peak A with gray-scale value after the denoising of this point:

${\mathrm{<figure-callout\; id="0"\; label="p"\; filenames="HDA0000033518570000051.png"\; state="\{\{state\}\}">p</figure-callout>}}_{0\mathrm{ij}}={\mathrm{<figure-callout\; id="0"\; label="p"\; filenames="HDA0000033518570000051.png"\; state="\{\{state\}\}">p</figure-callout>}}_{0\mathrm{cj}}\exp [-\frac{{(X_i-X_c)}^2}{{2\mathrm{\&sigma;}}^2}],$ (1≤a-w _j≤i≤b+w _j≤N _j)

Then has only a unknown quantity σ in the Gauss model;

Taken the logarithm in the modular form both sides:

$\mathrm{<figure-callout\; id="0"\; label="ln\; p"\; filenames="HDA0000033518570000051.png"\; state="\{\{state\}\}">ln</figure-callout>}{p}_{}{}_{0\mathrm{ij}}=\mathrm{<figure-callout\; id="0"\; label="ln\; p"\; filenames="HDA0000033518570000051.png"\; state="\{\{state\}\}">ln</figure-callout>}{p}_{}{}_{0\mathrm{cj}}+[--\frac{{(X_i-X_c)}^2}{{2\mathrm{\&sigma;}}^2}]$

Make k=1/ σ ², α=X _i-X _c, T=2 [lnp _0cj-lnp _0ij], following formula turns to T=k α ²

Utilize least square method that the quadratic expression after substituting is carried out linear fit, obtain best k _c, be in the Gaussian normalization model unique

Then obtain this central point normalization Gauss model sequence:

$g\left(X_i\right)=\left\{\begin{array}{cc}{\mathrm{<figure-callout\; id="0"\; label="p"\; filenames="HDA0000033518570000051.png"\; state="\{\{state\}\}">p</figure-callout>}}_{0\mathrm{cj}}\exp [-\frac{{(X_i-X_c)}^2}{{{2\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="c"\; filenames="HDA0000033518570000021.png"\; state="\{\{state\}\}">c</figure-callout>}}}^2},& (1\&le;a-w_j\&le;i\&le;b+w_j\&le;N_j)\\ 0& ,\mathrm{else}\end{array}\right..$

The energy that calculates striation cross section normalization Gauss model in the striation central point reliability evaluation method with, concrete grammar is for to ask gray scale and E to the striation central point xsect normalization Gauss model sequence of trying to achieve _j=∑ g (X _i), be the reliability index of our needed this point.

Beneficial effect of the present invention: the present invention is directed to existing structure striation central point method for distilling, original proposition is to extracting the reliability evaluation of programme that obtains the striation central point.Like this, in the process of three-dimensional structure photo measure,, also can utilize this appraisement system to estimate the credibility of striation central point even comparatively under the condition of severe (like strong surround lighting etc.).Also can judge whether extract the central point that obtains really exists, to distinguish signaling point and noise spot to the standardized threshold value of this evaluation result.Simultaneously, this evaluation method result can be follow-up three-dimensional model recovery and measures the original reliability evaluating data that confidence level is provided, to guarantee the confidence level of whole three-dimensional measurement.

On the other hand, to the reliability evaluation that the whole bag of tricks extracts the striation central point that obtains, also can be used as the evaluation good and bad to various method for distilling.If promptly to same width of cloth optical strip image, carry out central point with method A and B respectively and extract, higher than the reliability of another kind of method B if A extracts the reliability of the striation central point that obtains, then A is superior to B surely.

In addition, the reliability evaluation of striation central point is based on digital picture, so also can be used for the quality of optical strip image is estimated.Suppose that promptly (like different images inductor or different background light or the like) obtains optical strip image F and G under the different situations; Use with a kind of method two images are carried out the central point extraction; And carry out the reliability evaluation with the present invention; If the reliability of comprehensive evaluation result F is higher than G, can think that then image F is more excellent than image G quality.

Description of drawings

Fig. 1 is reliability evaluation object optical strip image f of the present invention

Fig. 2 is the striation central point sequence of Fig. 1 optical strip image through obtaining someway.

Fig. 3 is the gray scale xsect and the Gauss curve fitting curve of four kinds of situation striations.

Fig. 4 is the process flow diagram of reliability evaluation method of the present invention.

Fig. 5 is the concrete grammar process flow diagram of normalization best-fit Gauss model in the reliability evaluation method flow process of the present invention.

Fig. 6 is for asking for the synoptic diagram of striation width and estimation ground noise in the normalization fitted gaussian model of the present invention.

Fig. 7 utilizes the synoptic diagram of statistics with histogram estimation local loop back noise for the present invention.

Fig. 8 is the comparison of the inventive method with the (a) and (b) that utilize gray-scale value evaluation, the evaluation of Hessian matrix second order Grad finally to obtain, (c), (d), (e) five kinds of situation.Wherein (I) obtains the reliability evaluation result for the inventive method, (II) for the gray-scale value standard evaluation obtains the result, (III) is Hessian matrix second order Grad evaluation result.

Among Fig. 1,3,8; The striation that potential striation in (a) and (b), (c), (d) corresponding respectively background environment, different colours texture absorb, the striation speckle that unlike material causes and because four kinds of situation of interference striation that reflection produces (e) are form striation preferably.

Embodiment

Reliability evaluation object of the present invention is that optical strip image f (height of f image and width size are M, N) goes up the striation central point sequence C through obtaining someway _jReliability:

C _j＝(x _cj，y _cj)

(j=1～K, K is for extracting the striation central point sum that obtains, (x _Cj, y _Cj) be the coordinate of central point on image f) as shown in Figure 1.

Explain further details in the face of the present invention down.

At first, introduce structure striation model:

In the actual measurement, the ambient image a when image that obtains through a frame projective structure light and a frame non-structure striation, frame subtracts each other the structure optical strip image of the ground unrest that is removed before and after carrying out.The structure striation is to have certain width, and its normal section shows as the strip characteristic.Some scholar is approximately the striation xsect square or secondary parabola type.According to the mechanism that line laser produces, the light source laser beam profiles all is Gaussian distribution, and therefore, the tangent plane of optical strip image also must meet Gaussian distribution.

In actual measurement,, guarantee structured light vertical scanning for improving measuring accuracy as far as possible; Make the striation that obtains vertical on image; Striation normal direction xsect level of approximation is calculated so be combined into simplification in most of fields, comes approximate substitution striation normal direction xsect with the level cross-sectionn of striation.The various situation of actual observation structure striation normal section, as shown in Figure 3.

In the three-dimensional structure photo measure, diffuse reflection takes place on the testee surface in line-structured light.According to factors such as the material of testee, environmental impacts, can striation be divided into four kinds of situation: the potential striation Fig. 3 (a) in the background environment, striation Fig. 3 (b) that the different colours texture absorbs, striation speckle pattern 3 (c) and because interference striation Fig. 3 (d) that reflection produces that unlike material causes.Visible by Fig. 3, the striation cross section of several kinds of situation, its both sides are that single order slowly changes and is tending towards smooth, are not that simple secondary is changed to zero, so Gauss model meets more.And have certain noise around the actual striation, actual observation gets, and the noise of this moment is the random noise with certain expectation value.Noise source is in the fluctuation of environment, the fluctuation of imageing sensor.Simultaneously, can know that by Fig. 2 the striation each several part has different ground noises, but can think that noise is stable in the part.So one of usefulness that can be more definite has the Gauss model G of ground noise the striation xsect is described:

$G(x,y)=\mathrm{Aexp}[-\frac{{\left({x-\mathrm{\&mu;}}_x\right)}^2+{(y-{\mathrm{\&mu;}}_y)}^2}{{2\mathrm{\&sigma;}}^2}]+n---\left(1\right)$

Wherein, G (x, y) be point on the optical strip image (x, gray-scale value y), x are horizontal ordinate, y is an ordinate, (x, y) on striation normal y=kx+b, A is the Gauss model amplitude to point, (μ _x, μ _y) be the center point coordinate of Gauss model, σ is the standard deviation of Gauss model---the span of reflection distribution curve, n is the substrate of Gauss model.

About striation normal direction (n _x, n _y) ask for; Existing existing ripe algorithm; Like Steger (STEGER C, " AnUnbiased Detector of Curvilinear Structures ", IEEE Transactions on Pattern Analysis andmachine Intelligence.1998; 20 (2): 113-125.) propose fringe center method for distilling or the like based on the Hessian matrix, this does not give unnecessary details.Be combined into simplification in most of fields and calculate, come approximate substitution striation normal direction xsect with the level cross-sectionn of striation.

On the other hand, under the unsaturated situation of imageing sensors such as CCD, the gray scale that obtains the structure optical strip image is directly proportional with incident intensity, and studying carefully its essence is that the striation gray scale is directly proportional with the energy of structure striation incident.The gray-scale value of the optical strip image that promptly obtains is high more, and the energy of representative structure striation is also high more, and this place exists the degree of reliability of structure striation big more on the image, and confidence level is big more.

Actual line laser is to have striation certain width, the Gaussian cross section, so should and be true striation energy with the energy in this striation cross section.Cross section energy and big more, this place's structure striation existence is big more, and it is high more to extract the central point reliability that obtains.On image, correspondence shows as, the gray-scale value of this place's striation xsect and big more, and then this place exists the confidence level of striation central point high more.

Because the shape in actual striation cross section is comparatively complicated, so, can't under same standard, unify comparison, and have certain random noise on the image for piece image, can not be simply with the gray-scale value of light sliver transvers section and as the energy in cross section and.So the present invention carries out the normalization of Gauss model with the striation cross section of various different situations, estimates then.Since The present invention be directed to extract central point estimate, so the normalization Gauss model is the center with striation central point to be evaluated.

To sum up state, the reliability of striation central point is estimated according to be on striation normal direction xsect, with the energy of this some Gaussian normalization model that is the center and.

According to Gaussian mode pattern (1), the striation model has gaussian basis back noise n, and this is not the true energy of striation, need before clearing striation true energy, remove, and the striation model of removing noise floor is:

$G_0(x,y)=\mathrm{Aexp}[-\frac{{\left({x-\mathrm{\&mu;}}_x\right)}^2+{(y-{\mathrm{\&mu;}}_y)}^2}{{2\mathrm{\&sigma;}}^2}]---\left(2\right)$

Wherein, G ₀(x is except that point (x, gray-scale value y) on the optical strip image behind the denoising y); X is a horizontal ordinate, and y is an ordinate, and (x is y) on striation normal y=kx+b for point; A is the Gauss model amplitude; (μ _x, μ _y) be the center point coordinate of Gauss model, in the normalization process with central point C to be evaluated _j(x _Cj, y _Cj) be the center; σ is the standard deviation of Gauss model---the span of reflection distribution curve.

By above-mentioned, the concrete reliability evaluation of programme of the present invention is following:

The reliability appraisement system is input as optical strip image f to be evaluated (height of f image and width size are M, N) and the striation central point sequence C through obtaining someway _j:

C _j＝(x _cj，y _cj)

(j=1～K establishes and extracts the striation central point obtain and add up to K.Subscript c representes that c point of the sequence of the point on the striation xsect is the central point that obtains, (x _Cj, y _Cj) be the horizontal ordinate of central point on image.)

Because the striation xsect is the one-dimensional discrete sequence, so convenient calculating is converted to the enterprising row processing of a dimension coordinate with these points.The striation central point that obtains with extraction is a true origin, and the striation normal is horizontal ordinate X.Then for C _jThe xsect sequence that obtains is:

L _j＝X _i

(i=1～N _jN _jBe always counting on the xsect, X during i=c _cBe the position of central point under this coordinate)

Its corresponding gray on image f is:

P _j(X _i)=p _Ij(i=1～N _j, p during i=c _CjThe central point gray scale that obtains for extraction)

Then the straight line on the two dimensional image is converted to one dimension and handles, striation modular form (2) then converts into:

$G_0\left(X\right)=\mathrm{Aexp}[-\frac{{\left({X-\mathrm{\&mu;}}_x\right)}^2}{{2\mathrm{\&sigma;}}^2}]---\left(3\right)$

Wherein, G ₀(X) be the corresponding grey scale value, X is a horizontal ordinate, and A is the Gauss model amplitude, μ _xFor the center point coordinate of Gauss model (is center point coordinate X to be evaluated _c), σ is the standard deviation of Gauss model---the span of reflection distribution curve.

The reliability appraisement system is output as the corresponding reliability of striation central point that this method obtains on this image and estimates R _j

Whole evaluation method process flow diagram is seen Fig. 4, and concrete steps are:

1, to being that the xsect at center carries out the normalization of Gauss model with the central point

This step idiographic flow is seen Fig. 5.

Actual striation xsect is because a variety of causes have certain ground noise, and the ground noise of diverse location is different.According to the striation cross-sectional model, think the noise floor that is outside the striation width.In addition, in the structural light strip image, the structural light strip zone accounts for the ratio of entire image often less than the background pixel zone, brings non-striation zone into calculating, comes down to redundant computation.Then match striation xsect is that the method for Gaussian normalization model is:

1.1 estimating should the cross-sectional width w of place _j

This step purpose is to extract effective width striation, avoids redundant computation.For obtaining the width of striation xsect, three kinds of methods can be arranged:

A kind of method is the width w of this place's striation normal direction xsect of adaptive estimation _j

According to the relation between the striation xsect gray-scale value, estimation width w _jAs shown in Figure 6, establish striation central point C _jGray-scale value is p _Cj, with central point C _j(=X _c) be the center, extend along the normal direction both sides and search about p successively _CjThe point of 5%～20% gray scale as the starting point A of this place's striation normal direction xsect _j(=X _a) and terminating point B (=X _b), (subscript a, b represent, a of the sequence of the point on the striation xsect, a b point, a, b=1～N _j).Then can obtain effective striation sequence:

L _0j＝X _i(i＝a～b，a、b＝1～N _j)

L then _0jThe length of sequence is the live width w of estimation _j=length (L _0j).

Another kind method is, with the average live width w=mean (w of global image _j) locate the width w of striation normal direction xsect for this _j

Because when striation projected testee, whole striation width should be uniformly, consistent, so can characterize the live width w of striation with a value w _jA kind of algorithm of method obtains the live width w ' of each point in the utilization _j, to w ' _jAsk average, promptly get w _j=w=mean (w ' _j).

1.2 estimation local loop back noise

is also removed

According to Gaussian mode pattern (1), the striation model has gaussian basis back noise n, and this is not the true energy of striation, removes before the clearing striation true energy that needs.Actual observation gets, and the noise of this moment is the random noise with certain expectation value.Noise source is in the fluctuation of environment, the fluctuation of imageing sensor.Simultaneously, can know that by Fig. 2 the ground noise of striation each several part is different, but can think that noise is stable in the part.

So in order accurately to estimate the local loop back noise, the present invention is w at this standardized width in striation xsect left and right sides, place _jRegional area, as shown in Figure 6.

In this scope, gray-scale value is done on average with estimated noise substrate

$\stackrel{\&OverBar;}{n_j}=\frac{\underset{i=a-w_j}{\overset{i=a}{\mathrm{\&Sigma;}}}{p}_{\mathrm{ij}}+\underset{i=b}{\overset{i=b+w_j}{\mathrm{\&Sigma;}}}{p}_{\mathrm{ij}}}{{2w}_j},j=1~K,a-w_j\&GreaterEqual;1,b+w_j\&le;N_j---\left(5\right)$

Also can come the estimated noise substrate through statistics of histogram is carried out in the part, as shown in Figure 7.Since striation partly account for always count 1/3; Noise spot account for always count 2/3; According to the xsect grey value profile; The gray-scale value of noise spot is the lower part of gray scale in the histogram, do on average to make grey level histogram so can get the lower point of 2/3 gray scale generally to get tonal range 0～255 in estimated noise substrate

experiment, statistical series nbar (i); I＜=0～255, i is an integer.

From gray-scale value low to the gray-scale value height ask quantity with, get 1/2～2/3 the point of always counting and do average estimated noise substrate

$\stackrel{\&OverBar;}{n_j}=\frac{\underset{i=1}{\overset{i=i_c}{\mathrm{\&Sigma;}}}i\&CenterDot;\mathrm{Nbar}\left(i\right)}{\underset{i=1}{\overset{i=i_c}{\mathrm{\&Sigma;}}}\mathrm{Nbar}\left(i\right)},0\&le;i\&le;255,$ i _cFor getting 1/2～2/3 o'clock the sequence number (6) of always counting

Then, except that the striation xsect gray scale sequence behind the denoising be:

$P_{0j}\left(X_i\right)={\mathrm{<figure-callout\; id="0"\; label="p"\; filenames="HDA0000033518570000051.png"\; state="\{\{state\}\}">p</figure-callout>}}_{0\mathrm{ij}}=p_{\mathrm{ij}}-\stackrel{\&OverBar;}{n_j},{\mathrm{<figure-callout\; id="0"\; label="p"\; filenames="HDA0000033518570000051.png"\; state="\{\{state\}\}">p</figure-callout>}}_{0\mathrm{ij}}\&GreaterEqual;0---\left(7\right)$

1.3 fitted gaussian model

Actual striation effective coverage be (a b), calculates for simplifying, below all at C _jRegional area (a-w on the striation xsect of central point _j≤i≤b+w _j) scope in calculate.

For the gray scale sequence P that removes denoising _0j, central point C _j(=X _c) denoising after gray scale be p _0cj, carry out Gauss model normalization match to formula (3).Wherein, with C _jBe Gauss model center position μ _x, with the gray-scale value p of this place's point _0cjBe peak A, then formula (3) turns to:

${\mathrm{<figure-callout\; id="0"\; label="p"\; filenames="HDA0000033518570000051.png"\; state="\{\{state\}\}">p</figure-callout>}}_{0\mathrm{ij}}={\mathrm{<figure-callout\; id="0"\; label="p"\; filenames="HDA0000033518570000051.png"\; state="\{\{state\}\}">p</figure-callout>}}_{0\mathrm{cj}}\exp [-\frac{{(X_i-X_c)}^2}{{2\mathrm{\&sigma;}}^2}],$ (1≤a-w _j≤i≤b+w _j≤N _j)(8)

Wherein has only a unknown quantity σ.

The present invention utilizes least square method to come best-fit σ.

Taken the logarithm in formula (4) both sides:

$\mathrm{<figure-callout\; id="0"\; label="ln\; p"\; filenames="HDA0000033518570000051.png"\; state="\{\{state\}\}">ln</figure-callout>}{p}_{}{}_{0\mathrm{ij}}=\mathrm{<figure-callout\; id="0"\; label="ln\; p"\; filenames="HDA0000033518570000051.png"\; state="\{\{state\}\}">ln</figure-callout>}{p}_{}{}_{0\mathrm{cj}}+[--\frac{{(X_i-X_c)}^2}{{2\mathrm{\&sigma;}}^2}]---\left(9\right)$

T＝kα ² (10)

K=1/ σ in the formula (14) ², α=X _i-X _c, T=2 [lnp _0cj-lnp _0ij].

Utilize least square method that formula (10) is carried out linear fit, obtain best k _c, be in the Gaussian normalization model unique

Then the normalization Gauss model is:

$g\left(X_i\right)=\left\{\begin{array}{cc}{\mathrm{<figure-callout\; id="0"\; label="p"\; filenames="HDA0000033518570000051.png"\; state="\{\{state\}\}">p</figure-callout>}}_{0\mathrm{cj}}\exp [-\frac{{(X_i-X_c)}^2}{{{2\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="c"\; filenames="HDA0000033518570000021.png"\; state="\{\{state\}\}">c</figure-callout>}}}^2},& (1\&le;a-w_j\&le;i\&le;b+w_j\&le;N_j)\\ 0& ,\mathrm{else}\end{array}\right.---\left(11\right)$

2, the reliability evaluation index is found the solution, the energy that calculates striation cross section normalization Gauss model with

By aforesaid evaluation foundation---on the striation xsect, with the energy of this some Gaussian normalization model that is the center with, be on the image on the striation cross section gray-scale value of Gaussian normalization model with.,, this a part of gaussian shape is carried out the integration summation here, asks the energy E of striation xsect by optimum matching model sequence (11):

For C _jCentral point: E _j=∑ g (X _i) (12)

Be the reliability index of our needed this point.

3, carry out the normalization of global image.

The three-dimensional structure photo measure is a raw data by the multiframe optical strip image, so need unify the central point reliability of different frame optical strip image.To the some reliability evaluation of the piece image central point that obtains, the present invention carries out normalization by the maximal value of this image and handles:

Reliability R _j=E _j/ E _Jmax, E _Max=max (E _j), 1≤j≤K (13)

Embodiment:

Utilize above-mentioned evaluation method optical strip image to carry out the reliability evaluation:

As shown in Figure 8, be example with the optical strip image of Fig. 1, adopt evaluation method of the present invention respectively, and the second order Grad of original gray-scale value of central point and Hessian matrix comes the striation central point that obtains on the evaluation map 2 as the reliability evaluation criterion.Wherein Fig. 8 (a) and (b), (c), (d) distinguish (a) and (b), (c), (d) four kinds of situation in the corresponding diagram 1, and Fig. 8 (e) is a form Gaussian curve preferably, is that reliability is the most reliable, and with this reference as other situation, table 1 is seen in the comparative result qualitative analysis:

The various situation The qualitative analysis of table 1

Annotate: zero expression evaluation result comparatively accurately tallies with the actual situation, and △ representes to take second place, and * expression is the poorest.

For (a); Laser projections is to background object far away; On optical strip image, this striation is submerged in ground unrest, and three kinds of evaluation methods all will be evaluated as very low ((a) curve among Fig. 8) here; And the reliability value of the inventive method and Hessian matrix second order gradient method more is lower than the gray-scale value evaluation method, is different from the striation of non-noise more.

And (b) situation, because color and vein is when producing absorption in various degree, the gray-scale value of striation is lower, and reliability is lower, and like (b) curve among Fig. 8, three kinds of methods have all truly showed this situation.

For (c), because the big speckle that material produces, make regional area; The striation gray-scale value is closely saturated, and ambient noise also increases simultaneously, is not good so locate reliability; But because energy is higher here, data are comparatively reliable, and also being has certain confidence level; Visible by Fig. 8 (c) curve, evaluation method of the present invention has more truly showed this situation, and the evaluation result of gray-scale value method can't be distinguished here and striation form striation central point preferably; And the evaluation of Hessian matrix second order Grad is on the low side, and comparatively chaotic.

For (d), because the existence of reflecting interface produces and disturbs striation, reliability is on the low side.Can be known that by Fig. 8 (d) curve the reliability that evaluation method of the present invention obtains is estimated minimum, and in fact, this section interference is incredible, such threshold value just can be cut apart the point of high central point of reliability and noise.

For situation (e), the inventive method and gray-scale value evaluation have all showed the high characteristics of this place's reliability really, and the evaluation of Hessian matrix second order gradient is then relatively more chaotic.

It is thus clear that the inventive method can accurately be estimated each central point situation, and meets reality.

Claims (1)

1. light strip center of structured light point reliability evaluation method; Its evaluation object is the striation normal direction xsect point sequence of the striation central spot that obtains through a kind of method on the optical strip image, light strip center of structured light point estimate according to for the energy of the Gaussian normalization model that on striation normal direction xsect, is the center with this point with; This method comprises:

Obtain the striation normal direction xsect point sequence at light strip center of structured light point place, the central point C that obtains with extraction _jBe the center, the point sequence of striation normal direction xsect is carried out the normalization of Gauss model;

The energy that calculates striation normal direction xsect point sequence normalization Gauss model with, obtain the reliability evaluation index of this point;

To the striation central point C that tries to achieve _jXsect normalization Gauss model sequence g (X _i) ask gray scale with:

For striation central point C _j: E _j=∑ g (X _i), be the reliability index of needed this point;

Central point reliability evaluation index to obtaining is carried out the normalization of global image, carries out normalization by the reliability maximal value of this image and handles reliability: R _j=E _j/ E _Max, E _Max=max (E _j), finally obtain the evaluation of normalization reliability;

The described striation normal direction xsect point sequence that obtains light strip center of structured light point place, and be the center with the central point that extraction obtains, the point sequence of xsect is carried out the normalization of Gauss model, specifically comprise:

To the striation normal direction xsect point sequence of the striation central spot obtained, estimating should the normal direction cross-sectional width w of place _j: estimate this place's local loop back noise and remove; The best-fit Gauss model obtains final normalization model;

The estimation technique is to cross-sectional width w in the above-mentioned normalization Gauss model method _jFor: with the central point gray-scale value is the amplitude of Gauss model, and center position is the center of Gauss model, extends the point of 5%～20% the gray scale of searching about amplitude successively along striation normal both sides and locates the starting point A of striation normal direction xsect as this _jWith terminating point B _j, A _j, B _jCoordinate be X _a, X _b, with starting point A _jWith terminating point B _jSequence between 2 is the striation ordered sequence, and its length is normal direction cross-sectional width w _jPerhaps the method for average with global image is the striation normal direction cross-sectional width w of this place to cross-sectional width _j

Estimate in the above-mentioned normalization Gauss model method that local loop back noise and the method for removing are:

At this standardized width in striation xsect both sides, place is w _jRegional area is done on average with the estimated noise substrate gray-scale value in this scope; Or be w at this standardized width in striation xsect left and right sides, place _jRegional area carries out statistics of histogram in this part, from gray-scale value low to the gray-scale value height ask quantity with, get 1/2～2/3 the point of always counting and do average estimated noise substrate;

The original gray scale of xsect sequence deducts noise floor, the sequence gray-scale value behind the noise that promptly is removed, and this gray-scale value is not for negative simultaneously;

The method of best-fit Gauss model is in the above-mentioned normalization Gauss model method:

The best-fit Gauss model obtains final normalization model: for the gray scale sequence of removing denoising, central point C _jDenoising after gray-scale value be p _0cj, carry out Gauss model normalization match; Wherein, with C _jCoordinate X _cBe the Gauss model center position, with gray-scale value p after the denoising of this point _0cjBe the Gauss model peak value:

$p_{0\mathrm{ij}}=p_{0\mathrm{ij}}\exp [-\frac{{(X_i-X_c)}^2}{{2\mathrm{\&sigma;}}^2}],$ (1≤a-w _j≤i≤b+w _j≤N _j)

Then has only a unknown quantity---the standard deviation sigma of Gauss model in the Gauss model;

Taken the logarithm in the modular form both sides:

$\ln p_{0\mathrm{ij}}=\ln p_{0\mathrm{cj}}+[-\frac{{(X_i-X_c)}^2}{{2\mathrm{\&sigma;}}^2}]$

Make k=1/ σ ², α=X _i-X _c, T=2 [lnp _0cj-lnp _{) 0ij}], following formula turns to T=k α ²Utilize least square method that the quadratic expression after substituting is carried out linear fit, obtain the optimum estimate k of parameter k _c, the standard deviation sigma of the Gauss model in the then corresponding Gaussian normalization model is unique

Then obtain this central point normalization Gauss model sequence:

$g\left(X_i\right)=\left\{\begin{array}{cc}{p}_{0\mathrm{cj}}\exp [-\frac{{(X_i-X_c)}^2}{{{2\mathrm{\&sigma;}}_c}^2}],& (1\&le;a-w_j\&le;i\&le;b+w_j\&le;N_j)\\ 0,& \mathrm{else}\end{array}\right..$

Images