CN101504770A - Structural light strip center extraction method - Google Patents

Abstract

The invention discloses a method for extracting the light strip center of structured light, which comprises: acquiring an initial light strip center point range; acquiring the light strip center of the structured light through the iteration for times between light strip center points and the normal directions on the light strip center points; and smoothing the acquired light strip center of the structured light, and extracting the final light strip center of the structured light. The method for extracting the light strip center of the structured light is based on the extraction of a one-dimensional light strip center, realizes high-precision extraction of the light strip center through the iteration for times between the light strip center point range and the normal directions on the light strip center points, and is not related to massive addition-multiplication operations of image filtering, so the method can reduce the calculation and improve the processing speed when guaranteeing the detection precision simultaneously. Moreover, the normal directions of various light strip center points can be well calculated through iteration and the influence of the shape of light strip images on the extraction precision is small, so that the method has good universality and strong interference resistance.

Description

A kind of extracting method of light strip center of structured light

Technical field

The present invention relates to the structure light vision measuring technique, relate in particular to a kind of extracting method of light strip center of structured light.

Background technology

During structure light vision was measured, quick, the extracted with high accuracy of structural light strip were one of gordian techniquies.High speed dynamic structure light vision measurement system requires striation center extraction rate in several milliseconds, and in order to guarantee accuracy of measurement system and stability, the precision that requires the extraction of striation center is at sub-pixel.

Existing sub-pixel striation center extraction algorithm mainly contains: striation cross section fitting process (M.A.G., Sub-pixel measurement of 3D surfaces by laser scanning, Sensors and Actuators A:Physical, 1999,76 (1-3): 1～8; McIvor A M, Substripe localisation for improvedstructured light system performance, Proc Of DICTA/IVCNZ97.New Zealand:Massey University, 1997,309～314), (Hu Bin is based on the structural light stripes center detection method of direction template for multi-direction template, computer engineering and application, 2002,38 (11): 59-60,109; Thunder naval, a kind of structural light stripes center method for quick, Central China University of Science and Technology's journal (natural science edition), 2003,31 (1): 74-76) with Hessian matrix method (Carsten Steger, Unbiased Extraction ofCurvilinear Structures from 2D and 3DImages, Germany:Technische

M ü nchen, 1998.27～81; D.Eberly, R.Gardner, Ridges for image analysis.Journal ofMathematical Imaging and Vision, 1994,4:353～373; Hu Kun, a kind of quick structural light stripes center sub-pixel precision extracting method, Chinese journal of scientific instrument, 2006,27 (10), 1326～1329; Zhou Fuqiang, the vision-mix disposal route that structural light strip extracts, photoelectron and laser, 2008,19 (11), 1534～1537).

Above-mentioned striation cross section fitting process changes little image for the striation normal direction, on the striation cross section, carry out Gauss or fitting of parabola, by asking its extreme point to obtain the sub-pixel location at striation center, this method depends on the shape of striation again, and versatility is bad; The striation that changes greatly for normal direction can adopt above-mentioned multi-direction template, detects its sub-pix center, but adopts this method operand big; Above-mentioned Hessian matrix method belongs to the match interpolation algorithm in essence, has utilized the Hessian matrix of image, and good versatility is arranged, but need carry out convolution algorithm repeatedly to image, and processing speed is slow.

Summary of the invention

In view of this, fundamental purpose of the present invention is to provide a kind of extracting method of light strip center of structured light, can reduce operand, improve processing speed, and versatility be good, antijamming capability is strong when guarantee accuracy of detection.

For achieving the above object, technical scheme of the present invention is achieved in that

A kind of extracting method of light strip center of structured light comprises:

Obtain the initial striation central point row of structural light strip;

By striation central point in the described initial striation central point row and the repeatedly iteration between the described striation central spot normal direction, obtain light strip center of structured light;

The described light strip center of structured light of obtaining is carried out smooth operation, extract final light strip center of structured light.

Describedly obtain initial striation central point and classify as: after determining the initial normal direction of described structural light strip, obtain initial striation central point row according to the initial normal direction of determined striation.

The initial normal direction of described definite striation is:

The binaryzation optical strip image;

Calculate the one dimension projection matrix of bianry image respectively in level and vertical direction;

The direction of determining to have the one dimension projection matrix correspondence of less nonzero value is the initial normal direction of striation.

The initial normal direction of the striation that described basis is determined is obtained initial striation central point and is classified as:

Initial normal direction is a horizontal direction, obtains initial striation central point row by line scanning; Perhaps, initial normal direction is a vertical direction, obtains initial striation central point row by column scan.

Described by the repeatedly iteration between striation central point and the described striation central spot normal direction, extract light strip center of structured light and comprise:

A, extract a plurality of striation central points in the initial striation central point row obtained, calculate the normal direction of described each striation central point;

B, calculate new striation central point according to the normal direction of described striation central point;

C, judge whether to stop iteration, if then carry out described smooth operation; Otherwise the normal direction of the described new striation central point of calculation procedure b is returned step b.

Describedly judge whether to stop iteration and be: judge $\sqrt{\underset{k=0}{\overset{k=n}{\mathrm{\&Sigma;}}}({(x_k^t-x_k^{t-1})}^2+{(y_k^t-y_k^{t-1})}^2)/n}<\mathrm{\&epsiv;}$ Whether set up, wherein, n, ε is little positive number if being the number that described step a extracts the striation central point.

This method further comprises: the upper limit T that iterations is set;

Accordingly, describedly judge whether to stop iteration and be: judge whether iterations t=T sets up.

The extracting method of light strip center of structured light of the present invention, extracting method based on one dimension striation center, by the repeatedly iteration between striation center point range and the striation normal direction, realize the extraction at high-precision striation center, because the present invention does not relate to adding in a large number of image filtering and takes advantage of operation, so, can when guarantee accuracy of detection, reduce operand, improve processing speed.

In addition, because the present invention has adopted iterative idea, can calculate the normal direction of striation each point preferably, so the shape of optical strip image is very little to extraction precision influence of the present invention, in other words, versatility of the present invention is good, antijamming capability is strong.

Description of drawings

Fig. 1 is the process flow diagram of light strip center of structured light extracting method of the present invention;

Fig. 2 is under the perfect condition, striation intensity profile function synoptic diagram;

Fig. 3 is in the practical application, striation intensity profile function synoptic diagram;

Fig. 4 obtains the method flow diagram of the striation center point range of degree of precision for the present invention;

Fig. 5 has a desirable optical strip image synoptic diagram as Fig. 3 intensity profile for what computing machine generated;

The binaryzation optical strip image synoptic diagram of Fig. 6 for adopting the adaptive threshold method to obtain;

Fig. 7 for emulation striation shown in Figure 5 extracting section effect synoptic diagram.

Embodiment

Basic thought of the present invention is: based on the extracting method at one dimension striation center, the repeatedly iteration by between striation center point range and the striation normal direction realizes the extraction at high-precision striation center.

For making the purpose, technical solutions and advantages of the present invention clearer, by the following examples and with reference to accompanying drawing, the present invention is described in more detail.

Fig. 1 is the process flow diagram of light strip center of structured light extracting method of the present invention, and as shown in Figure 1, the extracting method of light strip center of structured light of the present invention mainly may further comprise the steps:

Step 11: the initial normal direction of determining structural light strip.

The initial normal direction of striation has only two kinds of possibilities: level or vertical, its depend on optical strip image in the horizontal direction with the projected length of vertical direction, long in a certain direction projection length, think that then this direction more approaches the trend of striation, thereby determine that perpendicular other direction is the initial normal direction of striation.Such as, optical strip image projected length in the horizontal direction is longer, thinks that then the trend of striation is similar to level more, and its initial normal direction is a vertical direction.So, can by calculate bianry image in the horizontal direction with the one dimension projection matrix of vertical direction, the number of judging one dimension projection matrix nonzero value is to obtain the initial normal direction of striation.

Here, at first adopt adaptive threshold method binaryzation optical strip image, calculate respectively then bianry image in the horizontal direction with the one dimension projection matrix of vertical direction, the number that compares nonzero value in two one dimension projection matrixes, have the trend of the corresponding striation of one dimension projection matrix of more nonzero value, its corresponding vertical direction is the initial normal direction (n of striation _x, n _y), promptly have the initial normal direction of the corresponding striation of one dimension projection matrix of less nonzero value.Above-mentioned employing adaptive threshold method binaryzation optical strip image (Otsu N.A threshold selectionmethod from gray level histograms[J] .Transaction on System Man and Cybernetic, 1979,9 (1): be prior art 62-66), do not describe in detail at this.

Step 12: according to the initial striation central point row of the initial normal direction computation structure light striation of the determined structural light strip of step 11.

In the ideal case, the one dimension optical strip image can be approximated to be the linear structure that has the Gaussian intensity profile along the light stripe centric line normal direction, as shown in Figure 2.With one dimension striation central point be initial point, the One dimensional Mathematical Model that distributes along this striation normal direction gradation of image can be expressed as $I\left(x\right)=h\&CenterDot;e^{-\mathrm{\&lambda;}{x}^2}$ , wherein, h, λ are non-vanishing arithmetic number, and the striation center is exactly the peak of function, satisfies formula (1):

$2{\&Integral;}_a^{x_0}I\left(x\right)\mathrm{dx}={\&Integral;}_a^{b}I\left(x\right)\mathrm{dx}---\left(1\right)$

Wherein, x ₀Be the coordinate at striation center, a, b are two borders of one dimension striation.Because I (x) is at x ₀The integration of the right and left equate, so x ₀Also can be called center of energy.In the practical application, because the restriction of camera gray scale sampling depth, striation intensity profile function will be as shown in Figure 3, and it can obtain striation central point x by formula (1) equally ₀Obviously, two kinds of intensity profile functions are identical by the center of energy point that formula (1) obtains.

Here, consider the discreteness of image, for point (x, y) and the initial normal direction (n of normalization _x, n _y), can obtain according to step-length l a plurality of, as 2m+1 sub-pix coordinate: (x+k * l * n _x, y+k * l * n _y), (k ∈ m ,-m+1 ..., m}), wherein, the physical significance of step-length l is the distance of discrete point, generally between 0.2～0.4, m represents that the sub-pix point is at (n _x, n _y) depart from the direction (and x, distance y), these points with interval l be positioned at striation cross point (x, on cross section y), formula (1) becomes formula (2):

$\underset{k=-m}{\overset{k=m}{\mathrm{\&Sigma;}}}I(x+k\&times;l\&times;n_x,y+k\&times;l\&times;n_y)=2\&times;[\underset{k=-m}{\overset{k=x_0}{\mathrm{\&Sigma;}}}I(x+k\&times;l\&times;n_x,y+k\&times;l\&times;n_y)+I(x+(x_0+\mathrm{\&Delta;})\&times;l\&times;n_x,y+(x_0+\mathrm{\&Delta;})\&times;l\&times;n_y)]$

(2)

Wherein, k, x ₀∈ m ..., m}, I (x+k * l * n _x, y+k * l * n _y) gray scale of the discrete pixel of expression, obtain Δ ∈ (0,1), then (x+ (x by utilizing known rounded coordinate gray-level interpolation ₀+ Δ) * l * n _x, y+ (x ₀+ Δ) * l * n _y) be exactly the striation center that the striation cross section is asked.

According to step 11, initial normal direction only may be a level or vertical, so by pressing line scanning or pressing the column scan image, using formula (2) just can obtain n initial striation central point, promptly initial striation central point is listed as.

Step 13:, obtain the light strip center of structured light point range of degree of precision by the repeatedly iteration between striation center point range and the striation central spot normal direction.

The specific implementation flow process of the described repeatedly iteration of step 13 may further comprise the steps as shown in Figure 4:

Step 131: a plurality of striation central points in the initial striation central point row that extraction step 12 obtains.

Here, described extraction is meant: for (x more arbitrarily in the striation center point range of obtaining _i, y _i), in the point range of described striation center, get its most contiguous 2q striation central point, herein, q is a positive integer.

Step 132: the normal direction of the calculation procedure 131 striation central point correspondence of obtaining.

Here, according to formula (3)～(5), the normalization normal direction of each striation central point that calculation procedure 131 is got obtains: (n _X1, n _Y1) ..., (n _Xn, n _Yn).Wherein, (n _Xn, n _Yn) expression striation central point (x _n, y _n) normal direction located.

$t_x=\underset{i=1}{\overset{i=2q}{\mathrm{\&Sigma;}}}(x_{i+1}-x_i)---\left(3\right)$

$t_y=\underset{i=1}{\overset{i=2q}{\mathrm{\&Sigma;}}}(y_{i+1}-y_i)---\left(4\right)$

$(n_x,n_y)=(t_y-t_x)/\sqrt{t_x^2+t_y^2}---\left(5\right)$

Step 133: utilize the normal direction of each striation central point of step 132 calculating, calculate new striation center point range by formula (2): $(x_1^t,y_1^t),......,(x_n^t,y_n^t),$ Wherein, t represents iterations.

Step 134: judge whether to stop iteration, if step 13 treatment scheme finishes; Otherwise, execution in step 135.

Here, judge whether that the foundation that stops iteration can having multiple choices.For example, can be by judging $\sqrt{\underset{k=0}{\overset{k=n}{\mathrm{\&Sigma;}}}({(x_k^t-x_k^{t-1})}^2+{(y_k^t-y_k^{t-1})}^2)/n}<\mathrm{\&epsiv;}$ Whether set up, judge whether to stop iteration, here, n, ε is little positive number, generally gets 0.05 if being the number that described step 131 is extracted the striation central point, $\sqrt{\underset{k=0}{\overset{k=n}{\mathrm{\&Sigma;}}}({(x_k^t-x_k^{t-1})}^2+{(y_k^t-y_k^{t-1})}^2)/n}<\mathrm{\&epsiv;}$ Set up, mean the result who uses alternative manner can not change gained substantially, proceeding interative computation has not had too big meaning, so stop iteration, step 13 treatment scheme finishes; Otherwise, execution in step 135.Certainly, the upper limit T of iterations can be set also, if iterations t=T then stops iteration, step 13 treatment scheme finishes; Otherwise, t=t+1, execution in step 135 continues iteration.In the practical application, T gets 5～10 usually.

Step 135: according to the striation center point range of step 133 calculating

By formula (3)～(5), calculate the normal direction of new striation center point range After, return step 133.

Step 14: the striation center point range of the degree of precision that step 13 is obtained is carried out smooth operation, extracts final light strip center of structured light.

The striation center of energy point range that step 13 iteration is obtained is designated as: (x ₁, y ₁), (x ₂, y ₂) ... (x _n, y _n).

Based on formula (6):

$({\stackrel{\&OverBar;}{x}}_m^1,{\stackrel{\&OverBar;}{y}}_m^1)=({\stackrel{\&OverBar;}{x}}_m,{\stackrel{\&OverBar;}{y}}_m)+({\stackrel{\&OverBar;}{x}}_{m+1},{\stackrel{\&OverBar;}{y}}_{m+1})/2,(m=1,\&CenterDot;\&CenterDot;\&CenterDot;,n-1)---\left(6\right)$

To above-mentioned point range carry out once level and smooth after, obtain point range $({\stackrel{\&OverBar;}{x}}_1^1,{\stackrel{\&OverBar;}{y}}_1^1),({\stackrel{\&OverBar;}{x}}_2^1,{\stackrel{\&OverBar;}{y}}_2^1).......,({\stackrel{\&OverBar;}{x}}_{n-1}^1,{\stackrel{\&OverBar;}{y}}_{n-1}^1).$ To the point range that newly obtains, carry out above-mentioned smooth operation repeatedly, carry out p time level and smooth after, obtain the net result point range: $({\stackrel{\&OverBar;}{x}}_1^p,{\stackrel{\&OverBar;}{y}}_1^p),({\stackrel{\&OverBar;}{x}}_2^p,{\stackrel{\&OverBar;}{y}}_2^p),......,({\stackrel{\&OverBar;}{x}}_{n-p}^p,{\stackrel{\&OverBar;}{y}}_{n-p}^p).$ Usually, p gets 5～10 and gets final product.

Below be one about simulation example of the present invention:

At first, has the desirable optical strip image of intensity profile as shown in Figure 3 by the computing machine generation, as shown in Figure 5, its resolution is 768 * 576, be used for simulating the ideal space distribution of striation gray scale, then, adding average is 0, standard deviation is the Gaussian distribution noise of σ, is used for the real image of virtual space striation.The detailed process that optical strip image shown in Figure 5 is extracted is as follows:

According to above-mentioned steps 11, adopt the adaptive threshold method to obtain the binaryzation optical strip image, as shown in Figure 6, it is as follows at the one dimension projection matrix of level and vertical direction to calculate bianry image then respectively:

The one dimension projection matrix of horizontal direction: [0 ..., 0,2,3,4,6,7,7 ..., 127,109,89,63,1,0,0 ..., 0], its size is 1 * 768, the nonzero value number is 104.

The one dimension projection matrix of vertical direction: [5,5,4,5,5,5,4,4 ..., 5,5,4,5,5,5,4] ', its size is 576 * 1, the nonzero value number is 576.

Because the one dimension projection matrix of horizontal direction has less nonzero value, so the initial normal direction of striation is a horizontal direction.

According to above-mentioned steps 12,, think that each row pixel is exactly an one dimension cross section of striation because the initial normal direction of striation is a horizontal direction, thereby by the line scanning image, utilize formula (2), select parameter l=0.3 and m=15, calculate 576 initial striation central points: (400.068,0), (400.497,1), (401.248,2), ..., (416.95,574), (416.094,575).

According to step 13, to choose all initial striation central points that calculate and carry out iteration, in the iterative process, related parameter choosing is: q=5, l=0.3, m=15, T=10, ε=0.05, final iteration obtains comprising the striation center point range of 575 points: (400.146 ,-0.0557085), (400.619,0.912298), (401.13,2.06592) ..., (417.975,571.868), (416.959,572.947), (416.581,573.811).

According to step 14, use formula (6) to above-mentioned point range carry out 10 times level and smooth after, obtain comprising the net result point range of 565 points: (407.07,10.1059), (407.763,11.0993), (408.454,12.0798) ..., (424.24,561.87), (423.619,562.859), (423.01,563.86).

Fig. 7 is the extracting section effect synoptic diagram of emulation striation shown in Figure 5, and table 1 extracts accuracy data for emulating image adds Gauss's noise light bar center:

Table 1

In this simulation example, computer configuration is: CPU is Intel Core 2.0G, internal memory 1G, operating system is Windows XP, and compilation tool is VC 6.0, and the size of pending image is 768 * 576, adopt light strip center of structured light extracting method of the present invention, elapsed time is 6.5 milliseconds.

As seen from the above, the present invention does not relate to adding in a large number of image filtering fully and takes advantage of operation, image for 768 * 576 pixels, only consume 6.5 milliseconds, and use through improving the striation center sub-pix extraction algorithm based on the Hessian matrix of filtering method, adopt the processor of 2.4G dominant frequency, need to consume milliseconds up to a hundred, if be used in combination the image preprocessing means, extracting the partial image region at striation place handles, utilize the Hessian matrix method equally, elapsed time also needs about 40 milliseconds, so processing speed of the present invention is very fast; From extracting on the precision, at image shown in Figure 4, extraction precision of the present invention is better than 0.12 pixel, can satisfy the requirement of structure light vision measuring systems fully.On versatility, because the present invention has adopted iterative idea, can calculate the normal direction of striation each point preferably, so the shape of optical strip image is very little to extraction precision influence of the present invention.In addition, from experimental result as can be seen, the present invention has noise antijamming capability preferably for image.Therefore, the present invention can reduce operand, improve processing speed, and versatility be good, antijamming capability is strong when guarantee accuracy of detection.

The above is preferred embodiment of the present invention only, is not to be used to limit protection scope of the present invention.

Claims (7)

1, a kind of extracting method of light strip center of structured light is characterized in that, this method comprises:

Obtain the initial striation central point row of structural light strip;

By striation central point in the described initial striation central point row and the repeatedly iteration between the described striation central spot normal direction, obtain light strip center of structured light;

The described light strip center of structured light of obtaining is carried out smooth operation, extract final light strip center of structured light.

2, extracting method according to claim 1 is characterized in that, describedly obtains initial striation central point and classifies as: after determining the initial normal direction of described structural light strip, obtain initial striation central point row according to the initial normal direction of determined striation.

3, extracting method according to claim 2 is characterized in that, the initial normal direction of described definite striation is:

The binaryzation optical strip image;

Calculate the one dimension projection matrix of bianry image respectively in level and vertical direction;

The direction of determining to have the one dimension projection matrix correspondence of less nonzero value is the initial normal direction of striation.

4, extracting method according to claim 3 is characterized in that, the initial normal direction of the striation that described basis is determined is obtained initial striation central point and classified as:

Initial normal direction is a horizontal direction, obtains initial striation central point row by line scanning; Perhaps, initial normal direction is a vertical direction, obtains initial striation central point row by column scan.

5, extracting method according to claim 1 is characterized in that, and is described by the repeatedly iteration between striation central point and the described striation central spot normal direction, extracts light strip center of structured light and comprises:

A, extract a plurality of striation central points in the initial striation central point row obtained, calculate the normal direction of described each striation central point;

B, calculate new striation central point according to the normal direction of described striation central point;

C, judge whether to stop iteration, if then carry out described smooth operation; Otherwise the normal direction of the described new striation central point of calculation procedure b is returned step b.

6, extracting method according to claim 5 is characterized in that, describedly judges whether to stop iteration and is: judge $\sqrt{\underset{k=0}{\overset{k=n}{\mathrm{\&Sigma;}}}({(x_k^t-x_k^{t-1})}^2+{(y_k^t-y_k^{t-1})}^2)/n}<\mathrm{\&epsiv;}$ Whether set up, wherein, n, ε is little positive number if being the number that described step a extracts the striation central point.

7, extracting method according to claim 5 is characterized in that, this method further comprises: the upper limit T that iterations is set;

Accordingly, describedly judge whether to stop iteration and be: judge whether iterations t=T sets up.

Images