CN104655043A - Contour registration method based on machine-vision contour measurement for train rails - Google Patents

Abstract

The invention discloses a contour registration method based on the machine-vision contour measurement for train rails. The contour measurement comprises the following steps: S1, acquiring a steel rail image; S2, obtaining a single-pixel curve of the cross section of the steel rail through the processing of a computer; S3, performing calibration restoration; S4, performing contour registration, wherein the contour registration comprises the following substeps: S41, uniformly placing an actual contour line and a standard contour line on a coordinate system, and obtaining an included angle formed between the actual contour line and the standard contour line by utilizing the asymptotes of the two contour lines; S42, enabling the actual contour line to rotate, so as to enable the asymptotes of the two contour lines to be in parallel; S53, horizontally moving the actual contour line, so that the endpoints of the actual contour line overlap with the endpoints of the standard contour line. According to the contour registration method disclosed by the invention, the actual contour line and the standard contour line are unified on the coordinate system, and the actual contour line is processed, so that the asymptotes of the actual contour line and the standard contour line are in parallel, and the endpoints of the actual contour line and the standard contour line are overlapped; besides, the registration result is accurate, the observation and the calculation are convenient, and reliable abrasion data can be obtained.

Description

A kind of based on profile method for registering in the measurement of machine vision train guide rail profile

Technical field

The invention belongs to the contactless automatic fields of measurement based on image, relate to image procossing, laser and optics, computing machine, automation field, specifically refer to a kind of based on profile method for registering in the measurement of machine vision train guide rail profile.

Background technology

Rail, as the important carrier in transportation by railroad, is one of most base unit in transportation by railroad, and its state quality directly affects the security of operation of train.Especially, when train runs under high speed, heavy duty, high density, rail wear strengthens, and the lost of life, causes high risks to train safe.The Ministry of Railways is in rail track " dimension rule ", and, severe injury light for rail defines the limit value of rail head abrasion, and requires will carry out in time processing or changing, to ensure the security of operation of train after rail head of rail abrasion exceed limit value when detecting.

Along with the lifting of transportation by railroad rate of traffic flow, being increased sharply of travelling speed, the abrasion of rail are more and more serious, especially the sharp radius curve Innerside wear at rail head place is particularly serious, have impact on the even running of train, the geometrical deviation of small track all likely causes train (especially high ferro) rocks and various vibration, impact is had, even because path wear increases the weight of to cause the skew laterally run on the comfort level of train driving and traveller's.When taking turns transverse force and increasing, just likely can there is the situations such as derail, bring great harm to passenger and traffic safety.The degree of wear of rail determines the serviceable life of rail, when attrition value arrives certain limit, needs to change in time.The measurement data of rail wear is that work business system determines that rail is the need of the important evidence changed, and therefore, detects the abrasion value size of track exactly, and to analyze each section abrasion situation be a very important job.

Mechanical Contact mensuration is that China makes a kind of method with the most use at present, slide calliper rule are mainly adopted to carry out hand dipping, this method is rely on manually completely, the shortcoming of contact type measurement is that speed is too slow, error is large, human factor impact is large, efficiency is low, and is not easy to ex-post analysis statistics and files.Another branch of contact type measurement is exactly the rail wear detector of machinery and electronics, the technological synthesis application and development such as able to programme.Also have photoelectric coding method, displacement sensing method and the non-contact type measuring method based on image procossing in addition, this method obtains the contour curve of rail profile indirectly by sensor, and the curve of drafting and the curve comparison of standard are obtained wearing valve.Photoelectric encoder and displacement transducer can be divided into again according to the difference of sensor.The MINIPROF system of greenwood company of Denmark development, it is made up of a notebook, special expansion box and wheel track measurement mechanism.Contact tested Rail Surface by the little magnet-wheel on measurement mechanism, the little magnet-wheel of manual control, obtains the polar data of sensor by computing machine, calculate profile after being converted to Cartesian coordinates.The track of magnet-wheel core can go on record, and computing machine just can calculate the actual profile of section.Domestic photoelectric coding checkout equipment has the WRS2000 shaped steel rail rail head appearance instrument for measuring and recording by railway institute of Tongji University and Jingjiang Hua Xing company cooperation research and development, and it is realized by high-precision photoelectric encoder.Tongji University still further developed a kind of wave abrasion measuring device of low-arranged sliding type railway steel rail in addition, this device adopts simple mechanical component to form longitudinal measuring basis and guide rail, and adopt USB interface and control technology and computer radio transmission measurement result, can immediately result be presented on screen.

A kind of noncontact measurement based on machine vision profilometry, have speed fast, the advantages such as profile overall picture, data management be convenient can be obtained.In recent years due to the high speed development of Digital image technology, laser technology and computer technology, the train guide rail profile measuring technique based on machine vision is greatly developed.A kind of noncontact measurement based on machine vision profilometry, have speed fast, the advantages such as profile overall picture, data management be convenient can be obtained.

Measure based on machine vision train guide rail profile, wherein the link of most critical is by obtained guide rail profile and real profile comparison, and could calculate and carry out abrasion calculating, the process of this comparison is called profile registration.Profile registration not pair time, the abrasion size calculated has very large, full of prunes deviation, and only have profile registration correct, the abrasion size calculated is only trust data within the scope of certain error.

Summary of the invention

The object of the invention is to overcome the deficiencies in the prior art, profile method for registering in providing a kind of contactless train guide rail profile to measure, by treated guide rail real contour line and the unification of standard wheels profile in same coordinate system, and real contour line is rotated, the process such as translation, make real contour line parallel with the asymptotic line of standard wheels profile and end points coincidence, the relative position of two outline lines is determined with this, registration result is accurate, clear and easy to understand, observe and convenience of calculation, can obtain reliable abrasion data.

The object of the invention is to be achieved through the following technical solutions: a kind of measure based on machine vision train guide rail profile in profile method for registering, wherein, what profile measurement adopted is non-contact measurement, and it can improve the work efficiency that on-the-spot railway detects department greatly.The cardinal principle measured obtains rail profile contour curve by laser light cutting method, simultaneously, CCD camera gathers laser image, and image information is transferred in the process chip of the computing machine be loaded on dolly, through measurement of rail wear based program, a series of process is carried out to the image collected, obtain single pixel curve of rail profile.By the matching and correlation to curve image, detect with railway the rail wear point that department defines and compare, draw the value of corresponding side wear and vertical abrasion.

Profile measurement specifically comprises the following steps: S1, gathers rail image; S2, processes the rail image transmitting collected to computing machine, and obtains single pixel curve of rail profile; S3, carries out demarcation reduction to single pixel curve, obtains real contour line; S4, carries out profile registration;

Profile registration in described step S4 comprises following sub-step:

S41: using the most end end points on guide rail inoperative limit as the scaling point of outline line, i.e. initial point, set up cartesian coordinate system, by the real contour line of guide rail and the unification of standard wheels profile in cartesian coordinate system, get the asymptotic line of two outline lines respectively, and utilize the asymptotic line of two outline lines to obtain angle theta between the two;

S42: rotate real contour line with twiddle factor R (θ), makes the asymptotic line of two outline lines be parallel to each other, obtains postrotational real contour line;

S43: postrotational real contour line is carried out translation, makes this real profile line endpoints overlap with the end points of standard wheels profile, completes the registration of outline line, and now, the real contour line relative position after standard wheels profile and registration is determined.

Further, the method gathering rail image in described step S1 is: obtain rail profile contour curve by laser light cutting method, and adopt CCD camera to gather laser image, CCD camera collection be the scattered light of profiled outline, the scattered light of this profile is two dimensional image model.

Further, demarcation in step S3 reduction refers to the two dimensional image model for collecting, and by calculating the information of actual three dimensions epigraph of reducing, is reconstructed the process of the corresponding relation between two dimensional image and three dimensional practicality by geometric model.Described step S3 specifically comprises following sub-step:

S31: by the change of camera imaging and pattern distortion, if its hidden parameter relational expression is:

$\left\{\begin{array}{c}x=a{u}^2+\mathrm{buv}+c{v}^2+\mathrm{du}+\mathrm{ev}+f\\ y=g{u}^2+\mathrm{huv}+i{v}^2+\mathrm{ju}+\mathrm{kv}+l\end{array}\right.---\left(1\right)$

Wherein, if P (x, y) represents is world coordinate system, P (u, what v) represent is image space coordinate system, parameter a, b, c, d, e, f, g, h, i, j, k, l is undetermined coefficient, and the object of demarcation is exactly will obtain stealthy undetermined parameter by the relation of fixing world coordinate system and image space coordinate, then obtains corresponding world coordinates by the parameter of stealth and image space coordinate points in actual applications and fastens object coordinates point;

S32: in order to eliminate the impact of indivedual points on result, adopt least square method to ask for undetermined parameter: the optimum matching being asked for function parameter by the quadratic sum asking for least error, adopts 20 groups of corresponding coordinates to ask for stealthy parameter;

S33: adopt semi-automation to determine calibration method, by presetting the coordinate points of the point in calibration plate, i.e. object coordinates, gathering image space coordinate figure corresponding in image space plane by the method for image procossing, formula (1) being separated into the form of matrix:

B＝[x ₁x ₂… x ₂₀y ₁y ₂… y ₂₀] ^T(3)

C＝[a b c d e f g h i j k l] ^T(4)

C＝A ^-1B (5)

Wherein, formula (2) is object coordinates system matrix, and formula (3) is corresponding image space coordinate points matrix, and formula (4) is matrix of coefficients, and through type (5) solves matrix of coefficients totally 12 parameters;

S34: in measuring system, is transformed into formula (5):

A＝BC ^-1(6)

Test matrix of coefficients C is known, then carrys out by the coordinate B matrix of image mid point the object coordinates that reverse goes out correspondence, the namely real contour line of rail profile section;

S35: in the process of image calibration reduction, because the image gathered there will be geometry deformation, so need to carry out geometric transformation to the image obtained, geometric transformation is in another image by the pixel-map in image, comprise the mapping of horizontal direction and the mapping of vertical direction, geometric transformation comprises the content of two aspects: pixel coordinate conversion and brightness interpolating:

S351: pixel coordinate conversion is mapping between output image and input picture, the plane of delineation (x, y) is be through geometric transformation to converted relational expression to the plane of delineation (X, Y):

$\left\{\begin{array}{c}X=\underset{r=0}{\overset{m}{\mathrm{\Σ}}}\underset{k=0}{\overset{m-r}{\mathrm{\Σ}}}{a}_{\mathrm{rk}}{x}^r{y}^k\\ Y=\underset{r=0}{\overset{m}{\mathrm{\Σ}}}\underset{k=0}{\overset{m-r}{\mathrm{\Σ}}}{b}_{\mathrm{rk}}{x}^r{y}^k\end{array}\right.---\left(7\right)$

Like this, by the coefficient a of conversion _rkand b _rkjust can determine the relation of geometric transformation, whole coordinate system may be changed when image procossing is carried out in geometric transformation, be determined the information of coordinate system transformation by Jacobi coefficient:

$J=\left|\frac{\∂(X,Y)}{\∂(x,y)}\right|=\left|\begin{array}{cc}\frac{\∂X}{\∂x}& \frac{\∂X}{\∂y}\\ \frac{\∂Y}{\∂x}& \frac{\∂Y}{\∂y}\end{array}\right|---\left(8\right)$

As J=0, conversion is irreversible, work as J=1, represent that image area has unchangeability when converting, when measuring system gathers image, the most incident image geometry distortion is exactly the rotation of image, in the geometric transformation of image, rotating R is the most frequently used one conversion, and transformation relation is such as formula (9):

$R:\left\{\begin{array}{c}X=x\sin \mathrm{\θ}+y\cos \mathrm{\θ}\\ Y=-x\sin \mathrm{\θ}+y\cos \mathrm{\θ}\\ J=1\end{array}\right.---\left(9\right)$

So, as long as obtain the angle θ rotated, just by formula (9), image is carried out rotational transform;

S352: the object of brightness interpolating is the gray-scale value of the pixel determined after conversion, the new picture point obtained after conversion may be a non-integral coordinate, need to process to the pixel value of image the gray scale obtaining this point, it has direct impact to the quality of image after geometric transformation, and the method for interpolation comprises arest neighbors, linear, bicubic three kinds.

Particularly, described arest neighbors Outlet approach concrete operations are: establish the result of gray-level interpolation to be expressed as Convolution Formula:

$f_n(X,Y)=\underset{l=-\∞}{\overset{\∞}{\mathrm{\Σ}}}\underset{k=-\∞}{\overset{\∞}{\mathrm{\Σ}}}{g}_s(\mathrm{l\ΔX},\mathrm{k\ΔY})h_n(X-\mathrm{l\ΔX},Y-\mathrm{k\ΔY})---\left(10\right)$

Wherein, h _nwhat represent is interpolation kernel, represents very little field;

Arest neighbors interpolation is that its interpolation formula is expressed as by point (X, Y) with the gray-scale value put to replace this from the gray-scale value of its nearest point:

f ₁(X,Y)＝g _s[round(x),round(y)] (11)。

Described linear interpolation considers four consecutive point of point (X, Y), and the gray-scale value of these four points is represented the gray-scale value of this point, interpolation formula is:

f ₂(X,Y)＝(1-a)(1-b)g _s(l,k)+a(1-b)g _s(l+1,k)

(12)。

+b(1-a)g _s(l,k+1)+abg _s(l+1,k+1)

The invention has the beneficial effects as follows:

1, the profile method for registering in providing a kind of contactless train guide rail profile to measure, by treated guide rail real contour line and the unification of standard wheels profile in same coordinate system, and real contour line is rotated, the process such as translation, make real contour line parallel with the asymptotic line of standard wheels profile and end points coincidence, the relative position of two outline lines is determined with this, registration result is accurate, clear and easy to understand, observe and convenience of calculation, then the calculating of relevant abrasion is carried out in the abrasion definition specified according to the Ministry of Railways, can obtain reliable abrasion data;

2, adopt contactless measurement, adopt laser light cutting method to obtain rail profile contour curve, and adopt CCD camera to gather laser image, and the image transmitting collected is processed to computing machine; There is image acquisition speed fast, the advantage such as profile overall picture and data management convenience can be obtained;

3, by rail profile line that laser line generator and CCD camera collect, outline line relatively is reliably obtained through the image processing means such as noise processed, connection of broken lines, and then suitable demarcation reduction treatment is carried out to outline line, two dimensional image model CCD camera collected is by calculating the information of actual three dimensions epigraph of reducing, can by elimination of the distortion, rediscover outline line, makes result of calculation more accurately and reliably.

Accompanying drawing explanation

Fig. 1 is profile measurement schematic diagram of the present invention;

Fig. 2 is rail profile skeleton pattern figure of the present invention;

Fig. 3 is outline line registration process schematic diagram of the present invention;

Fig. 4 is the square trrellis diagram of standard;

Fig. 5 is the image formed after two kinds typically distortion;

Fig. 6 is the image rotation schematic diagram due to guide rail, the generation of CCD attitudes vibration of the present invention;

Fig. 7 is linear gradation interpolation schematic diagram.

Embodiment

Further illustrate technical scheme of the present invention below in conjunction with accompanying drawing, but the content that the present invention protects is not limited to the following stated.

Profile measurement principle as shown in Figure 1, employing be non-contact measurement, it greatly can improve on-the-spot railway and detect the work efficiency of department.The cardinal principle measured obtains rail profile contour curve by laser light cutting method, adopt semiconductor county laser instrument Emission Lasers, simultaneously, CCD camera gathers laser image, and image information is transferred in the process chip of the computing machine be loaded on dolly, through measurement of rail wear based program, a series of process is carried out to the image collected, obtain single pixel curve of rail profile.By the matching and correlation to curve image, detect with railway the rail wear point that department defines and compare, draw the value of corresponding side wear and vertical abrasion.

In measurement of rail wear based process, detect department and fix, have corresponding definition to rail head of rail Detection of content according to rail, main measuring content comprises side wear, vertical abrasion and always wears away, and Figure 2 shows that rail wear section and calibration schematic diagram.In figure, what W1 represented is vertical abrasion, and its definition is the thickness of 2/3 place's vertical direction abrasion of the rail head width on distance rail inoperative limit; What W2 represented is side wear, and its definition is that the definition of total abrasion is that vertical wearing valve adds 1/2 times of horizontal wearing valve from the rail head top down width that wears away in the horizontal direction of 16mm.

The rail profile line collected by laser line generator and ccd array, is obtained outline line relatively reliably through image processing meanses such as noise processed (filtering etc.), connection of broken lines, and then carries out suitable calibration process to outline line.But the plane formed due to laser line generator is not orthogonal with contoured surface, therefore the outline line obtained not is real outline line, that is there is distortion, so need elimination of the distortion, with rediscover outline line.

A kind of based on profile method for registering in the measurement of machine vision train guide rail profile, wherein, profile measurement specifically comprises the following steps: S1, gathers rail image; S2, processes the rail image transmitting collected to computing machine, and obtains single pixel curve of rail profile; S3, carries out demarcation reduction to single pixel curve, obtains real contour line; S4, carries out profile registration;

Profile registration in described step S4 comprises following sub-step:

S41: as shown in Fig. 3 (a), using the most end end points P on guide rail inoperative limit as the scaling point of outline line, i.e. initial point, set up cartesian coordinate system, by the real contour line of guide rail and the unification of standard wheels profile in cartesian coordinate system, get the asymptotic line of two outline lines respectively, and utilize the asymptotic line of two outline lines to obtain angle theta between the two;

S42: rotate the real contour line in Fig. 3 (a) with twiddle factor R (θ), makes the asymptotic line of two outline lines be parallel to each other, obtains postrotational real contour line;

S43: the postrotational real contour line in Fig. 3 (b) is carried out translation, this real profile line endpoints is overlapped with the end points of standard wheels profile, as shown in Fig. 3 (c), complete the registration of outline line, now, the real contour line relative position after standard wheels profile and registration is determined.

Further, the method gathering rail image in described step S1 is: obtain rail profile contour curve by laser light cutting method, and adopt CCD camera to gather laser image, CCD camera collection be the scattered light of profiled outline, the scattered light of this profile is two dimensional image model.

When obtaining rail profile image based on light cross-section method, due to CCD camera collection is the scattered light of profiled outline, CCD camera imaging plane, laser plane are at an angle, based in the non-contact measurement system of machine vision, all can be designed into the demarcation reduction of image.This is because, for the two dimensional image model that CCD camera collects, need, by calculating the information of actual three dimensions epigraph of reducing, to reconstruct the corresponding relation between two dimensional image and three dimensional practicality by the geometric model of CCD imaging.The process obtaining the parameter of this geometric model is exactly the process that CCD camera is demarcated.

Ideally, video camera imaging is a kind of linear pin-hole imaging model.But in practice, the camera lens that video camera is installed or lens group, due to the reason of processing or other some reasons, there will be the phenomenon of distortion in the place away from optical axis center during imaging.Be illustrated in figure 4 the square trrellis diagram of standard, shown in Fig. 5 (a) He Fig. 5 (b) is the image that Fig. 4 is formed after two kinds typically distortion.After CCD imaging, the distortion type of appearance is primarily of two classes: the barrel distortion shown in the pincushion type distortion shown in Fig. 5 (a) and Fig. 5 (b).Now, camera imaging model is a kind of nonlinear, so the process of image rectification also should comprise the impact of distortion correction on image.

Camera plane is relative with the position of laser plane is fixing, so, only need surely show once for the mechanical hook-up fixed.But when actual measurement, because dolly shake and other unknown causes are on the impact of machinery, in order to ensure the accuracy measured, still need regularly to demarcate camera.Demarcation reduction in described step S3 specifically comprises following sub-step:

S31: the model adopting hidden parameter to calibrate reacts the corresponding relation between two dimensional image coordinate and three dimensional practicality, these parameters calculated do not have concrete physical significance, just characterize corresponding variation relation; By the change of camera imaging and pattern distortion, if its hidden parameter relational expression is:

$\left\{\begin{array}{c}x=a{u}^2+\mathrm{buv}+c{v}^2+\mathrm{du}+\mathrm{ev}+f\\ y=g{u}^2+\mathrm{huv}+i{v}^2+\mathrm{ju}+\mathrm{kv}+l\end{array}\right.---\left(1\right)$

Wherein, if P (x, y) represents is world coordinate system, P (u, what v) represent is image space coordinate system, parameter a, b, c, d, e, f, g, h, i, j, k, l is undetermined coefficient, and the object of demarcation is exactly will obtain stealthy undetermined parameter by the relation of fixing world coordinate system and image space coordinate, then obtains corresponding world coordinates by the parameter of stealth and image space coordinate points in actual applications and fastens object coordinates point;

S32: the stealthy parameter asked in theory in formula (1) needs 6 groups of corresponding image space points and object space point, but in order to eliminate the impact of indivedual points on result, adopt least square method to ask for undetermined parameter: the optimum matching being asked for function parameter by the quadratic sum asking for least error, adopt 20 groups of corresponding coordinates to ask for stealthy parameter;

S33: adopt semi-automation to determine calibration method, by presetting the coordinate points of the point in calibration plate, i.e. object coordinates, gathering image space coordinate figure corresponding in image space plane by the method for image procossing, formula (1) being separated into the form of matrix:

B＝[x ₁x ₂… x ₂₀y ₁y ₂… y ₂₀] ^T(3)

C＝[a b c d e f g h i j k l] ^T(4)

C＝A ^-1B (5)

Wherein, formula (2) is object coordinates system matrix, and formula (3) is corresponding image space coordinate points matrix, and formula (4) is matrix of coefficients, and through type (5) solves matrix of coefficients totally 12 parameters;

S34: in measuring system, is transformed into formula (5):

A＝BC ^-1(6)

Test matrix of coefficients C is known, then carrys out by the coordinate B matrix of image mid point the object coordinates that reverse goes out correspondence, the namely real contour line of rail profile section;

S35: in the process of image calibration reduction, the visual position of laser plane and camera is fixing, and the scaling parameter obtained by formula (5) records in the lab, and during calibration, dolly lies on the rail level of rail head of rail.In actual measurement, the position of dolly can not lie on rail level completely, inevitably has inclination, shake in the process of promotion.Illustrate for the some sections on rail, when dolly tilts, just in the section surface thereof perpendicular to Rail Surface, now the position of laser plane still matches with the section of this point, do not change, but the position due to camera is the bottom being fixed on dolly, the rotation of dolly has just driven the rotation of camera position, and the position be equivalent to as plane and object plane there occurs small change.Now, the stealthy parameter that through type (1) obtains is to when having the rail profile curve of certain inclination to calibrate, and the image obtained just has certain inclination.When graphical analysis, because the image gathered there will be geometry deformation, so need to carry out geometric transformation to the image obtained, the impact brought to graphical analysis could be out of shape by removal of images like this.Geometric transformation be by the pixel-map in image in another image, comprise the mapping of horizontal direction and the mapping of vertical direction.When carrying out geometric transformation, mainly comprise two aspects, i.e. pixel coordinate conversion and brightness interpolating:

S351: pixel coordinate conversion is mapping between output image and input picture, the plane of delineation (x, y) is be through geometric transformation to converted relational expression to the plane of delineation (X, Y):

$\left\{\begin{array}{c}X=\underset{r=0}{\overset{m}{\mathrm{\Σ}}}\underset{k=0}{\overset{m-r}{\mathrm{\Σ}}}{a}_{\mathrm{rk}}{x}^r{y}^k\\ Y=\underset{r=0}{\overset{m}{\mathrm{\Σ}}}\underset{k=0}{\overset{m-r}{\mathrm{\Σ}}}{b}_{\mathrm{rk}}{x}^r{y}^k\end{array}\right.---\left(7\right)$

Like this, by the coefficient a of conversion _rkand b _rkjust can determine the relation of geometric transformation, whole coordinate system may be changed when image procossing is carried out in geometric transformation, be determined the information of coordinate system transformation by Jacobi coefficient:

$J=\left|\frac{\∂(X,Y)}{\∂(x,y)}\right|=\left|\begin{array}{cc}\frac{\∂X}{\∂x}& \frac{\∂X}{\∂y}\\ \frac{\∂Y}{\∂x}& \frac{\∂Y}{\∂y}\end{array}\right|---\left(8\right)$

As J=0, conversion is irreversible, works as J=1, represent that image area has unchangeability when converting, when measuring system gathers image, the most incident image geometry distortion is exactly the rotation of image, and Fig. 6 is the image rotation schematic diagram for producing due to guide rail, CCD attitudes vibration; Solid line shown in figure is the actual curve obtained, the rail image collected when dotted line is and keeps flat, and the degree of tilt of two curves is θ.In the geometric transformation of image, rotating R is the conversion of the most frequently used one, transformation relation such as formula for:

$R:\left\{\begin{array}{c}X=x\sin \mathrm{\θ}+y\cos \mathrm{\θ}\\ Y=-x\sin \mathrm{\θ}+y\cos \mathrm{\θ}\\ J=1\end{array}\right.---\left(9\right)$

So, as long as obtain the angle θ rotated, just by formula (9), image is carried out rotational transform;

S352: the object of brightness interpolating is the gray-scale value of the pixel determined after conversion, the new picture point obtained after conversion may be a non-integral coordinate, need to process to the pixel value of image the gray scale obtaining this point, it has direct impact to the quality of image after geometric transformation, and the method for interpolation comprises arest neighbors, linear, bicubic three kinds.

Particularly, described arest neighbors Outlet approach concrete operations are: establish the result of gray-level interpolation to be expressed as Convolution Formula:

$f_n(X,Y)=\underset{l=-\∞}{\overset{\∞}{\mathrm{\Σ}}}\underset{k=-\∞}{\overset{\∞}{\mathrm{\Σ}}}{g}_s(\mathrm{l\ΔX},\mathrm{k\ΔY})h_n(X-\mathrm{l\ΔX},Y-\mathrm{k\ΔY})---\left(10\right)$

Wherein, h _nwhat represent is interpolation kernel, represents very little field;

Arest neighbors interpolation is that its interpolation formula is expressed as by point (X, Y) with the gray-scale value put to replace this from the gray-scale value of its nearest point:

f ₁(X,Y)＝g _s[round(x),round(y)] (11)。

Particularly, described linear interpolation considers four consecutive point of point (X, Y), and the gray-scale value of these four points is represented the gray-scale value of this point, interpolation formula is:

f ₂(X,Y)＝(1-a)(1-b)g _s(l,k)+a(1-b)g _s(l+1,k)

(12)。

+b(1-a)g _s(l,k+1)+abg _s(l+1,k+1)

Fig. 7 is linear interpolation schematic diagram, and wherein solid line represents input picture, and dotted line represents the image grid that output image obtains through inverse transformation.

The geometric transformation problem of image when solving measurement by pixel transform and brightness interpolating.But, above rotation is only applicable to the low-angle variation that dolly rotates, here set the maximum anglec of rotation is by experiment 5 °, when the angle rotated is excessive, owing to also needing the factor considered to be the aberration problems of image during image restoring, between three dimensional practicality image to two-dimensional image plane, be transformed to nonlinear relation.At this moment, rotate the attitudes vibration caused just there will be larger error by reacting dolly as the change of plane.So when designing dolly, having a scroll wheel be fixed and clamped to the position of dolly on rail, allowing little truck position lie in rail rail level as far as possible.In the process promoted, we also will allow dolly keep equilibrium position.Like this, the impact of nonlinear transformation on result can be reduced.

Technique effect of the present invention is further illustrated below in conjunction with experiment.

When certain rail welding plant is tested, the rail type of experiment is 60 rails.The flow process of whole measuring system according to setting is gathered rail profile curve and preserve data.By automatic and manual mode image data respectively, system can normally be run.Following table 1 is depicted as the contrast of system for field measurement data and kind of calliper data, reliability and stability when data carry out evaluating system field survey by experiment.

The reliability and stability of table 1 field survey verification msg

Table 1 (a) field survey different measuring point experimental data

The same measurement point experimental data of table 1 (b) field survey

Standard deviation in table is the dispersion degree of data deviation average, and above two list datas can be found out, system still has good Stability and veracity for in-site measurement.

When experienced Chongqing line is tested, it is very serious that guide rail has all worn away, and the rail head of rail has back edge phenomenon, and it is very serious that rail limit to be measured all wears away.Following table 2 is depicted as the one piece of data wherein tested, and by obtaining the information of rail wear after data analysis software process, and is printed by printing reports.

Table 2 field survey wears away comparatively heavy rail experimental data

In form, what first print is the essential information measured.First form is the abrasion statistics exceeded standard, the measurement data of a certain section that what table 2 represented is.Wherein, the font that have passed different colours for the point beyond slight wound or severely injured standard marks.In in-site measurement, the sampling interval of setting is 3m, and the spacing that dolly is passed by is 250m.System is for manually and automatically sampling and can normally work.A contrast has roughly been passed through with the carrying out of kind of calliper (in measuring at the scene for the data collected, kind of calliper also has some enchancement factors and occurs), wherein vertically wear away error roughly at ± 0.5mm, level abrasion error is ± 1.0mm.

Experiment determines rationality and the correctness of matching process of the present invention, and this measuring method can stable operation obtain good measurement effect, meets the measurement requirement that the Ministry of Railways specifies.

Those of ordinary skill in the art will appreciate that, embodiment described here is to help reader understanding's principle of the present invention, should be understood to that protection scope of the present invention is not limited to so special statement and embodiment.Those of ordinary skill in the art can make various other various concrete distortion and combination of not departing from essence of the present invention according to these technology enlightenment disclosed by the invention, and these distortion and combination are still in protection scope of the present invention.

Claims (6)

1. a profile method for registering in measuring based on machine vision train guide rail profile, it is characterized in that, profile measurement comprises the following steps: S1, gathers rail image; S2, processes the rail image transmitting collected to computing machine, and obtains single pixel curve of rail profile; S3, carries out demarcation reduction to single pixel curve, obtains real contour line; S4, carries out profile registration;

Profile registration in described step S4 comprises following sub-step:

S41: using the most end end points on guide rail inoperative limit as the scaling point of outline line, i.e. initial point, set up cartesian coordinate system, by the real contour line of guide rail and the unification of standard wheels profile in cartesian coordinate system, get the asymptotic line of two outline lines respectively, and utilize the asymptotic line of two outline lines to obtain angle theta between the two;

S42: rotate real contour line with twiddle factor R (θ), makes the asymptotic line of two outline lines be parallel to each other, obtains postrotational real contour line;

S43: postrotational real contour line is carried out translation, makes this real profile line endpoints overlap with the end points of standard wheels profile, completes the registration of outline line, and now, the real contour line relative position after standard wheels profile and registration is determined.

2. profile method for registering according to claim 1, it is characterized in that, the method gathering rail image in described step S1 is: obtain rail profile contour curve by laser light cutting method, and adopt CCD camera to gather laser image, CCD camera collection be the scattered light of profiled outline, the scattered light of this profile is two dimensional image model.

3. profile method for registering according to claim 1, it is characterized in that, demarcation reduction in described step S3 refers to the two dimensional image model for collecting, by calculating the information of actual three dimensions epigraph of reducing, reconstructed the process of the corresponding relation between two dimensional image and three dimensional practicality by geometric model.

4. profile method for registering according to claim 3, is characterized in that, described step S3 specifically comprises following sub-step:

S31: by the change of camera imaging and pattern distortion, if its hidden parameter relational expression is:

Wherein, if P (x, y) represents is world coordinate system, P (u, what v) represent is image space coordinate system, parameter a, b, c, d, e, f, g, h, i, j, k, l is undetermined coefficient, and the object of demarcation is exactly will obtain stealthy undetermined parameter by the relation of fixing world coordinate system and image space coordinate, then obtains corresponding world coordinates by the parameter of stealth and image space coordinate points in actual applications and fastens object coordinates point;

S32: in order to eliminate the impact of indivedual points on result, adopt least square method to ask for undetermined parameter: the optimum matching being asked for function parameter by the quadratic sum asking for least error, adopts 20 groups of corresponding coordinates to ask for stealthy parameter;

S33: adopt semi-automation to determine calibration method, by presetting the coordinate points of the point in calibration plate, i.e. object coordinates, gathering image space coordinate figure corresponding in image space plane by the method for image procossing, formula (1) being separated into the form of matrix:

B＝[x ₁x ₂… x ₂₀y ₁y ₂… y ₂₀] ^T(3)

C＝[a b c d e f g h i j k l] ^T(4)

C＝A ^-1B (5)

Wherein, formula (2) is object coordinates system matrix, and formula (3) is corresponding image space coordinate points matrix, and formula (4) is matrix of coefficients, and through type (5) solves matrix of coefficients totally 12 parameters;

S34: in measuring system, is transformed into formula (5):

A＝BC ^-1(6)

Test matrix of coefficients C is known, then carrys out by the coordinate B matrix of image mid point the object coordinates that reverse goes out correspondence, the namely real contour line of rail profile section;

S35: in the process of image calibration reduction, because the image gathered there will be geometry deformation, so need to carry out geometric transformation to the image obtained, geometric transformation is in another image by the pixel-map in image, comprise the mapping of horizontal direction and the mapping of vertical direction, geometric transformation comprises the content of two aspects: pixel coordinate conversion and brightness interpolating:

S351: pixel coordinate conversion is mapping between output image and input picture, the plane of delineation (x, y) is be through geometric transformation to converted relational expression to the plane of delineation (X, Y):

Like this, by the coefficient a of conversion _rkand b _rkjust can determine the relation of geometric transformation, whole coordinate system may be changed when image procossing is carried out in geometric transformation, be determined the information of coordinate system transformation by Jacobi coefficient:

As J=0, conversion is irreversible, work as J=1, represent that image area has unchangeability when converting, when measuring system gathers image, the most incident image geometry distortion is exactly the rotation of image, in the geometric transformation of image, rotating R is the most frequently used one conversion, and transformation relation is such as formula (9):

So, as long as obtain the angle θ rotated, just by formula (9), image is carried out rotational transform;

S352: the object of brightness interpolating is the gray-scale value of the pixel determined after conversion, the new picture point obtained after conversion may be a non-integral coordinate, need to process to the pixel value of image the gray scale obtaining this point, it has direct impact to the quality of image after geometric transformation, and the method for interpolation comprises arest neighbors, linear, bicubic three kinds.

5. profile method for registering according to claim 4, is characterized in that, described arest neighbors Outlet approach concrete operations are: establish the result of gray-level interpolation to be expressed as Convolution Formula:

Wherein, h _nwhat represent is interpolation kernel, represents very little field;

Arest neighbors interpolation is that its interpolation formula is expressed as by point (X, Y) with the gray-scale value put to replace this from the gray-scale value of its nearest point:

f ₁(X,Y)＝g _s[round(x),round(y)] (11)。

6. profile method for registering according to claim 4, is characterized in that, described linear interpolation considers four consecutive point of point (X, Y), and the gray-scale value of these four points is represented the gray-scale value of this point, interpolation formula is:

f ₂(X,Y)＝(1-a)(1-b)g _s(l,k)+a(1-b)g _s(l+1,k) (12)。

+b(1-a)g _s(l,k+1)+abg _s(l+1,k+1)