CN105414206A - Online rapid positioning method for cold-rolled strip steel edge - Google Patents

Abstract

An online rapid positioning method for the edge of cold-rolled strip steel comprises the steps of processing a strip steel real-time image collected by a camera to obtain a strip steel synchronous gray image; carrying out longitudinal median filtering on the gray image, wherein the gray value of each pixel point on the image is equal to the gray intermediate value of the longitudinally adjacent point; layering images, and respectively determining left and right fluctuation maximum blocks of each layer from the left and right directions; performing image longitudinal compression to respectively form left and right compressed image sequences; respectively generating left and right edge fluctuation sequences; then, determining an edge fluctuation threshold value of each fluctuation sequence, respectively searching a fluctuation sequence value of which the first one is larger than the corresponding edge fluctuation threshold value in the fluctuation sequence from the left direction and the right direction, and determining the left edge position and the right edge position of the layer and a step value; and finally, determining the statistical boundaries of the left and right sides. The invention has the advantages of reliable work, simple algorithm, high running speed, good processing effect, no influence of strip steel texture, color difference, defects and illumination and suitability for industrial production and various computing equipment applications.

Description

The online method for rapidly positioning of a kind of cold-strip steel edge

Technical field

The invention belongs to industrial machine vision technique field, particularly the online method for rapidly positioning of a kind of cold-strip steel edge.

Background technology

It is a key issue in belt steel rolling process that edge with steel detects, and the processing links such as centering process, trimming process, strip width detection, steel strip surface defect Information locating for band steel all plays vital effect.

Traditional steel edge portion detection method has mechanical probe method, photodetection method etc.Mechanical detection method by contact zones steel direct-detection, but very easily causes damage to the edge of band steel and forms crimping.Photodetection method, by photoelectric combination formula equipment, non-contact detecting can go out steel edge portion, but needs to rely on basic automatization means, equipment relative complex and costliness, and because system element is more, easily breaks down, difficult maintenance.

Along with the development of machine vision technique, obtain industry based on image analysis method determination steel edge portion position and more and more paid close attention to.Image Edge-Detection is one of the most basic problem of image procossing and computer vision field, and how to extract image edge information is fast, accurately study hotspot both domestic and external always, and the detection at edge simultaneously is also a difficult problem in image procossing.

" a kind of method of Image Edge-Detection " that Authorization Notice No. CN1O2156996B provides is the improvement to canny edge detection method, can the edge of positioning image more accurately, and the edge blurry avoiding image again can not be too responsive to noise.Application publication number CN1O2521836A discloses " a kind of edge detection method based on certain kinds gray level image ", and the method can improve rim detection precision and the degree of accuracy, effectively realizes the detection at edge.Above two kinds of methods are all improve the one of conventional method, but method amount of calculation is comparatively large, the quick running environment that uncomfortable crossed belt steel is produced.

Application publication number CN1O3226829A discloses one " method for detecting image edge based on edge enhancement operator ", and the method can realize the precision location at edge, and the edge of different size can respond preferably, and can reduce undetected as far as possible.But the method the cold-strip steel edge not being suitable for substrate complexity (be such as with zinc flower galvanized sheet or aluminium plating zinc plate etc.) detect.Application publication number is " a kind of method for detecting image edge average based on gray scale difference value " of CN1O3150735A, propose a kind ofly utilize the method identification object edge of the average combining adaptive threshold value of gray scale difference value and extract profile, although the method processing speed comparatively fast, is not suitable for substrate complexity and may there is the cold-strip steel edge detection of edge defect equally.

Summary of the invention

Object of the present invention aim to provide a kind of can when cold-strip steel high-speed runs, determine non-contacting, online, instant steel edge portion position, and the online method for rapidly positioning of all kinds of cold-strip steel edge of the not situation such as tape steel superficial makings, aberration, uneven illumination, edge defect impact.

For this reason, the solution that the present invention takes is:

The online method for rapidly positioning of a kind of cold-strip steel edge, it is characterized in that, position fixing process is divided into the location of left lateral and the location in the right portion, and concrete grammar and step are:

1, edge Image Acquisition: by collected by camera band steel real-time synchronization image, then turn gray proces to the image collected, obtains the synchronous gray level image of band steel.

2, Image semantic classification: first carry out longitudinal medium filtering to the synchronous gray level image of band steel obtained, namely on image, the gray value of each pixel equals the gray scale median of its longitudinal point of proximity.

3, search fluctuation largest block: for the search of the left-hand side undulated layer largest block, from image upper left quarter, horizontal M row pixel, the capable pixel of longitudinal N, with the size of M × N, divide block check from left to right, according to gray scale initial survey situation determination the left-hand side undulated layer largest block in block; For right undulated layer largest block then from image upper right quarter, with the size of M × N, divide block check from right to left, make to use the same method and determine right undulated layer largest block; Every one deck largest block longitudinal extent below the first floor is still the capable pixel of N, and lateral extent is then equal to the lateral extent of last layer largest block;

4, image longitudinal compression is carried out: left lateral is located, the left undulated layer largest block determined in step (3) expands to 3M × n-quadrant to the left and right, left and right extension length is identical, calculates the longitudinal mean value often arranged, and forms the left compressed image sequence FY of 3M × 1 size _l(x, y), wherein x=0,1,2 ..., y is the number of plies; For location, the right portion, method is identical, forms right compressed image sequence FY _r(x, y);

5, volatility series is generated:

For left lateral location, volatility series FB _lthe computational methods of (x, y) are:

FB _L(i,y)＝|FY _L(i,y)-FY _L(i-1,y)|(i>0)

For location, the right portion, volatility series FB _rx the computational methods of () are:

FB _r(i, y)=| FY _r(i, y)-FY _r(i+1, y) | (i< compressed image sequence length)

6, a layer marginal position is determined:

For left lateral location, first calculate volatility series FB _lruffling threshold k B in (x, y) _l(y):

KB _L(y)＝AVG _L(y)+SDV _L(y)

Wherein, AVG _ly () is volatility series FB _lthe mean value of (x, y), SDV _ly () is volatility series FB _lthe standard deviation of (x, y);

Then, volatility series FB is searched for from left to right _lin (x, y), first is greater than KB _ly the volatility series value of (), original image horizontal pixel point position (x) corresponding to this value is this layer of left hand edge position LB _l(Y) the volatility series value, corresponding to this position is designated as this layer of left hand edge and jumps rank value FS _l(y);

For location, the right portion, first calculate volatility series FB _rruffling threshold k B in (x, y) _r(y):

KB _R(y)＝AVG _R(y)+SDV _R(y)

Wherein, AVG _ry () is volatility series FB _rthe mean value of (x, y), SDV _ry () is volatility series FB _rthe standard deviation of (x, y);

Then, volatility series FB is searched for from right to left _rin (x, y), first is greater than KB _ry the volatility series value of (), original image horizontal pixel point position (x) corresponding to this value is this layer of right hand edge position LB _r(Y) the volatility series value, corresponding to this position is designated as this layer of right hand edge and jumps rank value FS _r(y);

7, statistical boundary is determined: for the situation calculating multi-layer image statistical boundary, left lateral statistical boundary TGB _lcomputational methods be:

${\mathrm{TGB}}_L=\underset{i=0}{\overset{Y}{\mathrm{\&Sigma;}}}{\mathrm{LB}}_L\left(i\right)\&times;\frac{{\mathrm{FS}}_L\left(i\right)}{{\mathrm{TS}}_L}$

${\mathrm{TS}}_L=\underset{i=0}{\overset{Y}{\mathrm{\&Sigma;}}}{\mathrm{FS}}_L\left(i\right)$

Wherein, y is the number of plies.

The right portion statistical boundary TGB _rcomputational methods be:

${\mathrm{TGB}}_R=\underset{i=0}{\overset{Y}{\mathrm{\&Sigma;}}}{\mathrm{LB}}_R\left(i\right)\&times;\frac{{\mathrm{FS}}_R\left(i\right)}{{\mathrm{TS}}_R}$

${\mathrm{TS}}_R=\underset{i=0}{\overset{Y}{\mathrm{\&Sigma;}}}{\mathrm{FS}}_R\left(i\right)$

In described step 3, be if the following test condition of this block first fit according to gray scale initial survey situation decision method in gray scale initial survey situation determination the left-hand side undulated layer largest block in block, be then defined as the left-hand side undulated layer largest block: test condition:

N _M>T _K

$N_M=\underset{\left(\right|F(x,y)-F(x-N,y)|>C_K)}{\mathrm{\&Sigma;}}1$

Wherein, C _kgray scale for strip substrate and non-band steel part fluctuates predetermined minimum, T _kfor judging constant.

Beneficial effect of the present invention is:

1, be applicable to industrial production running environment, reliable operation, algorithm is simple, is easy to realize.

2, for all kinds of camera design, treatment effect is good, and the speed of service is fast.

3, the various computing equipment such as computer and single-chip microcomputer is applicable to.

4, the impact of the present invention's not situation such as tape steel superficial makings, aberration, uneven illumination, edge defect.

5, steel edge portion is detected roughly and meticulous detection all applicable.

Accompanying drawing explanation

Fig. 1 is the online quick positioning system pie graph of cold-strip steel edge;

Fig. 2 is the online method for rapidly positioning flow chart of cold-strip steel edge;

Fig. 3 is cold-strip steel edge online method for rapidly positioning effect figure.

In figure: band steel 1, image capture device 2, lighting device 3, light belt 4.

Detailed description of the invention

Below in conjunction with accompanying drawing, the invention will be further described.

The online quick positioning system of cold-strip steel edge of the present invention is formed as shown in Figure 1: system carries out Real-time Collection by the band steel 1 of image capture device 2 to motion, is thrown light on by lighting device 3.Image capture device 2 can be single or multiple camera (camera) or other image capture devices.Lighting device 3 can be the various light sources such as general light modulation, LED integrated lamp, laser scanning.Lighting device 3 projects belt steel surface and forms light belt 4, and IMAQ point is completely wrapped in light belt 4.The flip flop equipment of camera can be equipped with sychronisation, also can asynchronously trigger.System is captured in line image by image acquisition units, and communication modes can be digital or analog.System carries out pretreatment by graphics processing unit to image, and graphics processing unit can be separate unit (as singlechip equipment etc.) or integrated unit (program module as in computer).System is confirmed steel edge portion by edge detecting unit, and edge detecting unit also can be separate unit (as singlechip equipment etc.) or integrated unit (program module as in computer).

The concrete steps following (flow process is shown in Fig. 2) of the online method for rapidly positioning of cold-strip steel edge of the present invention:

Step one: edge Image Acquisition.Be with steel 1 synchronous images in real time by collected by camera, then gray proces turned to the image collected, obtain band, 1 synchronous gray level image.

Step 2: Image semantic classification.First carry out longitudinal medium filtering to the synchronous gray level image of the band steel 1 obtained, namely on image, the gray value of each pixel equals the gray scale median of its longitudinal point of proximity.Now the selection range of longitudinal point of proximity can be upper two pixels of analysis site and lower two pixels.

Step 3: search fluctuation largest block.For the search of the left-hand side undulated layer largest block, from image upper left quarter, size (laterally 50 row with 50 × 10, longitudinally 10 row), divide block check from left to right, if a certain piece of following test condition of first fit, be then defined as the left-hand side undulated layer largest block.

N _M>T _K

$N_M=\underset{\left(\right|F(x,y)-F(x-N,y)|>C_K)}{\mathrm{\&Sigma;}}1$

Wherein, C _kgray scale for strip substrate and non-band steel part fluctuates predetermined minimum, is set as 5; T _kfor judging constant, T _kget 150.

For right undulated layer largest block then from image upper right quarter, with the size of 50 × 10, divide block check from right to left, make to use the same method and determine right undulated layer largest block.Every one deck largest block height below the first floor is still 10, and lateral extent is then directly equal to last layer lateral extent.

Step 4: carry out image longitudinal compression.For left lateral location, in 150 × 10 regions that the left undulated layer largest block determined in previous step is expanded to the left and right, calculate the longitudinal mean value often arranged, form the left compressed image sequence FY of 150 × 1 sizes _l(x, y).For location, the right portion, method is identical, forms right compressed image sequence FY _r(x, y).

Step 5: generate volatility series.

For left lateral location, volatility series FB _lthe computational methods of (x, y) are as follows:

FB _L(i,y)＝FY _L(i,y)-FY _L(i-1,y)(i>0)

For location, the right portion, volatility series FB _rx the computational methods of () are as follows:

FB _r(i, y)=FY _r(i, y)-FY _r(i+1, y) (i< compressed image sequence length)

Step 6: determine a layer marginal position.

For left lateral location, first calculate volatility series FB _lruffling threshold k B in (x, y) _l(y).

KB _L(y)＝AVG _L(y)+SDV _L(y)

Wherein, AVG _ly () is volatility series FB _lthe mean value of (x, y), SDV _ly () is volatility series FB _lthe standard deviation of (x, y).Then, volatility series FB is searched for from left to right _lin (x, y), first is greater than KB _ly the volatility series value of (), original image horizontal pixel point position (x) corresponding to this value is this layer of left hand edge position LB _l(Y) the volatility series value, corresponding to this position is this layer of left hand edge and jumps rank value FS _l(y).

For location, the right portion, first calculate volatility series FB _rruffling threshold k B in (x, y) _r(y).

KB _R(y)＝AVG _R(y)+SDV _R(y)

Wherein, AVG _ry () is volatility series FB _rthe mean value of (x, y), SDV _ry () is volatility series FB _rthe standard deviation of (x, y); Then, volatility series FB is searched for from right to left _rin (x, y), first is greater than KB _ry the volatility series value of (), original image horizontal pixel point position (x) corresponding to this value is this layer of right hand edge position LB _r(Y) the volatility series value, corresponding to this position is designated as this layer of right hand edge and jumps rank value FS _r(y).

Step 7: repeat step 3 to step 6, successively calculate, until calculated the left and right edges position of bottom layer image and the rank value that jumps.

Step 8: determine statistical boundary.For the situation calculating multi-layer image statistical boundary, left lateral statistical boundary TGB _lcomputational methods as follows:

${\mathrm{TGB}}_L=\underset{i=0}{\overset{Y}{\mathrm{\&Sigma;}}}{\mathrm{LB}}_L\left(i\right)\&times;\frac{{\mathrm{FS}}_L\left(i\right)}{{\mathrm{TS}}_L}$

${\mathrm{TS}}_L=\underset{i=0}{\overset{Y}{\mathrm{\&Sigma;}}}{\mathrm{FS}}_L\left(i\right)$

Wherein, y is the number of plies.

The right portion statistical boundary TGB _rcomputational methods as follows:

${\mathrm{TGB}}_R=\underset{i=0}{\overset{Y}{\mathrm{\&Sigma;}}}{\mathrm{LB}}_R\left(i\right)\&times;\frac{{\mathrm{FS}}_R\left(i\right)}{{\mathrm{TS}}_R}$

${\mathrm{TS}}_R=\underset{i=0}{\overset{Y}{\mathrm{\&Sigma;}}}{\mathrm{FS}}_R\left(i\right)$

Fig. 3 display be edge Detection results for lvanized cold rolled strip.Inconsistent from image the right and left portion band steel illumination condition, left lateral is comparatively dark, and the right portion is brighter, and substrate steel sheet has more obvious texture, and part edge band steel gradation of image is near or below non-band steel district gradation of image.The method can determine the edge position of galvanized sheet preferably under such condition.

Claims (1)

1. the online method for rapidly positioning of cold-strip steel edge, is characterized in that, position fixing process is divided into the location of left lateral and the location in the right portion, and concrete grammar and step are:

(1) edge Image Acquisition: by collected by camera band steel real-time synchronization image, then turn gray proces to the image collected, obtains the synchronous gray level image of band steel;

(2) Image semantic classification: first carry out longitudinal medium filtering to the synchronous gray level image of band steel obtained, namely on image, the gray value of each pixel equals the gray scale median of its longitudinal point of proximity;

(3) search fluctuation largest block: for the search of the left-hand side undulated layer largest block, from image upper left quarter, horizontal M row pixel, the capable pixel of longitudinal N, with the size of M × N, divide block check from left to right, according to gray scale initial survey situation determination the left-hand side undulated layer largest block in block; For right undulated layer largest block then from image upper right quarter, with the size of M × N, divide block check from right to left, make to use the same method and determine right undulated layer largest block; Every one deck largest block longitudinal extent below the first floor is still the capable pixel of N, and lateral extent is then equal to the lateral extent of last layer largest block;

(4) image longitudinal compression is carried out: left lateral is located, the left undulated layer largest block determined in step (3) expands to 3M × n-quadrant to the left and right, left and right extension length is identical, calculates the longitudinal mean value often arranged, and forms the left compressed image sequence FY of 3M × 1 size _l(x, y), wherein x=0,1,2 ..., y is the number of plies; For location, the right portion, method is identical, forms right compressed image sequence FY _r(x, y);

(5) volatility series is generated:

For left lateral location, volatility series FB _lthe computational methods of (x, y) are:

FB _L(i,y)＝|FY _L(i,y)-FY _L(i-1,y)|(i>0)

For location, the right portion, volatility series FB _rx the computational methods of () are:

FB _r(i, y)=| FY _r(i, y)-FY _r(i+1, y) | (i< compressed image sequence length)

(6) a layer marginal position is determined:

For left lateral location, first calculate volatility series FB _lruffling threshold k B in (x, y) _l(y):

KB _L(y)＝AVG _L(y)+SDV _L(y)

Wherein, AVG _ly () is volatility series FB _lthe mean value of (x, y), SDV _ly () is volatility series FB _lthe standard deviation of (x, y);

Then, volatility series FB is searched for from left to right _lin (x, y), first is greater than KB _ly the volatility series value of (), original image horizontal pixel point position (x) corresponding to this value is this layer of left hand edge position LB _l(Y) the volatility series value, corresponding to this position is designated as this layer of left hand edge and jumps rank value FS _l(y);

For location, the right portion, first calculate volatility series FB _rruffling threshold k B in (x, y) _r(y):

KB _R(y)＝AVG _R(y)+SDV _R(y)

Wherein, AVG _ry () is volatility series FB _rthe mean value of (x, y), SDV _ry () is volatility series FB _rthe standard deviation of (x, y);

Then, volatility series FB is searched for from right to left _rin (x, y), first is greater than KB _ry the volatility series value of (), original image horizontal pixel point position (x) corresponding to this value is this layer of right hand edge position LB _r(Y) the volatility series value, corresponding to this position is designated as this layer of right hand edge and jumps rank value FS _r(y);

(7) statistical boundary is determined: for the situation calculating multi-layer image statistical boundary, left lateral statistical boundary TGB _lcomputational methods be:

${\mathrm{TGB}}_L=\underset{i=0}{\overset{Y}{\mathrm{\&Sigma;}}}{\mathrm{LB}}_L\left(i\right)\&times;\frac{{\mathrm{FS}}_L\left(i\right)}{{\mathrm{TS}}_L}$

${\mathrm{TS}}_L=\underset{i=0}{\overset{Y}{\mathrm{\&Sigma;}}}{\mathrm{FS}}_L\left(i\right)$

Wherein, y is the number of plies;

The right portion statistical boundary TGB _rcomputational methods be:

${\mathrm{TGB}}_R=\underset{i=0}{\overset{Y}{\mathrm{\&Sigma;}}}{\mathrm{LB}}_R\left(i\right)\&times;\frac{{\mathrm{FS}}_R\left(i\right)}{{\mathrm{TS}}_R}$

${\mathrm{TS}}_R=\underset{i=0}{\overset{Y}{\mathrm{\&Sigma;}}}{\mathrm{FS}}_R\left(i\right)$

In described step (3), be if the following test condition of this block first fit according to gray scale initial survey situation decision method in gray scale initial survey situation determination the left-hand side undulated layer largest block in block, be then defined as the left-hand side undulated layer largest block: test condition:

N _M>T _K

$N_M=\underset{\left(\right|F(x,y)-F(x-N,y)|>C_K)}{\mathrm{\&Sigma;}}1$

Wherein, C _kgray scale for strip substrate and non-band steel part fluctuates predetermined minimum, T _kfor judging constant.