CN104236498A - Method for measuring end size of switch rail - Google Patents

Abstract

The invention relates to a method for measuring end size of a switch rail, overcomes the defect that indication error of a tape measure manually used for measurement, reading errors caused by tension changes during measuring and errors caused by temperature changes finally cause a final measurement error, aims to directly measure the size of a related part of each end face of a rail member in rail workshops, to achieve noncontact measurement and to finish measurement on the site. The method has the advantages that the method is flexible and convenient, measurement errors are reduced, measurement results are accurate, measuring precision and measuring speed are greatly increased, and production efficiency is greatly improved accordingly.

Description

A kind of measuring method of railroad turnout steel rail part face size

Technical field

The invention belongs to measuring method technical field, relate to a kind of measuring method of railroad turnout steel rail part face size.

Background technology

Track switch is the crucial track equipment that railway realizes that train turns line operation, its technical merit has embodied a concentrated reflection of the manufacture level of a state-owned railroads track, the high precision of switch manufacture and assembling and high ride are one of important technical indicators, and the relative dimensions manufacturing accuracy of railroad turnout steel rail part is one of key factor affecting track switch high precision and high ride.It is all the mode manually adopting steel tape to measure that the face size of current railroad turnout steel rail part is measured usually, but the error that the reading error that when error of indication of steel tape itself, measurement, tension variations causes, temperature variation are brought all likely causes final measuring error, precision is low, measures efficiency low.

Summary of the invention

The present invention is in order to overcome the defect of prior art, devise a kind of measuring method of railroad turnout steel rail part face size, achieve non-contact measurement, overcome that artificial measuring error is large when adopting steel tape to measure, precision is low, inefficient defect, reduce measuring error, improve measuring accuracy and measure efficiency.

The concrete technical scheme that the present invention takes is:

A measuring method for railroad turnout steel rail part face size, key is: described method comprises the following steps:

A, preparation rail part and calibrating block, calibrating block be centre dig one with the scaling board in the size of rail part end face, hole that shape is identical, the middle position of scaling board is positioned at by cutouts, web of the rail position, both sides is with chequered with black and white chessboard figure, in addition, also respectively have one piece of chessboard figure in this scaling board left and right sides, two pieces of chessboard figure become fixed angle with scaling board, and three is an entirety;

B, carry out industrial camera demarcation: adopt Zhang Zhengyou plane reference method to demarcate, the image of shooting calibrating block, the operation of extraction angle point is carried out to the image of calibrating block, extract all angle points, set up the corresponding relation between image coordinate system, industrial camera coordinate system and world coordinate system three coordinate systems, obtain the inside and outside parameter of industrial camera, the relation can set up between pixel coordinate with corresponding spatial point coordinate is as follows:

$\begin{array}{c}z\left[\begin{array}{c}u\\ v\\ 1\end{array}\right]=\left[\begin{array}{ccc}1/\mathrm{dX}& \tan \mathrm{\α}/\mathrm{dY}& u_0\\ 0& 1/\mathrm{dY}& v_0\\ 0& 0& 1\end{array}\right]\left[\begin{array}{cccc}f& 0& 0& 0\\ 0& f& 0& 0\\ 0& 0& 1& 0\end{array}\right]\left[\begin{array}{c}x\\ y\\ z\\ 1\end{array}\right]\\ =\begin{array}{c}\left[\begin{array}{cccc}{f}_x& \mathrm{\γ}& u_0& 0\\ 0& f_y& v_0& 0\\ 0& 0& 1& 0\end{array}\right]\left[\begin{array}{cc}R& T\\ 0^T& 1\end{array}\right]\left[\begin{array}{c}{X}_w\\ Y_w\\ Z_w\\ 1\end{array}\right]\end{array}\end{array}$

Wherein, the coordinate of (x, y, z) representation space point P in industrial camera coordinate system, α denotation coordination axle out of plumb error, f _x=f/dX, f _y=f/dY is the equivalent focal length represented with pixel wide and length, and γ represents the nonorthogonality of image coordinate axle;

C, rail part is positioned over ground or is placed on platform, the flange of rail kiss the earth of rail part or platform are placed, remove the foreign material around rail part, calibrating block is close to rail part end face, the height of adjustment two industrial cameras and angle, make two industrial cameras and rail part end face be in sustained height, and allow two industrial cameras just to rail part end face, require that the optical axis of two industrial cameras is parallel, the optical axis of two industrial cameras is perpendicular to rail part end face;

D, open lamphouse light source, rail part end face is illuminated in side, makes the complete jag facial contour of light, to remove shade, adjust two industrial cameras, make rail part end face fall into the center of screen picture, and take two each screen views of industrial camera, gather the image of rail part end face now;

E, control device first carry out medium filtering process to the image of steps d gained, then carry out binaryzation, adopt the edge detection algorithm based on Canny to obtain the edge of image;

F, according to the Gestalt rule in topological consciousness theory, the profile of rail part end face should be continuously, closed, complete, if the parameterized form at an edge is B=v (t)={ v ₁(t ₁), v ₂(t ₂) ..., v _n(t _n), 0≤t _i≤ 1, if 1≤i≤n. v (0)=v (1), then B is a closed contour; If this closed contour just in time contains the end face of rail part, then B is exactly required for us; But often contour of object is made up of a series of edge, these edges are discontinuous often, propose a kind of method getting up to form continuous boundary by two edge conjunction here: suppose that there are two end points p at an edge of the image that step e obtains ₁(x, y) and p ₂(x, y), and have both direction, the direction gradient angle of regulation end points is the direction of these two end points,

In formula, R (p ₁) _y, R (p ₁) _xrepresent some p respectively ₁in the projection in y direction and x direction; R (p ₂) _y, R (p ₂) _xrepresent some p respectively ₂in the projection in y direction and x direction, if one in the direction of an end points at edge and its surrounding edge, the angular separation of an end points at a certain bar edge is minimum, then these two end points straight lines are coupled together and are merged into a new edge, finally detect one meet continuity, closed, integrality edge be exactly the closed contour of rail part end face;

G, the data fitting of the closed contour obtained in step f is become geometric primitive, least square fitting principle is utilized to carry out fitting a straight line, extract the N number of discrete point on graph line edge feature, ask for the fitting a straight line y=kx+b of these points, this N number of discrete point coordinate (x obtained by step f _i, y _i) (i=1,2 ..., N) meet straight-line equation or as far as possible from straight line very close to, ask the minimum value of expression:

$S=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{[y_i-({\mathrm{kx}}_i+b)]}^2$

The matching of circle the N number of point on profile is carried out company to the squared-distance of fitting circle add summation, minimizes it:

$S=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(\sqrt{{(x_i-a)}^2+{(y_i-b)}^2}-r)}^2$

Wherein, (a, b) is central coordinate of circle, r is radius of a circle, image outline can be divided into straight line and curve like this, by the detection and indentification to these shapes, according to the relation between the pixel coordinate set up in step b with corresponding spatial point coordinate, obtain the size that we want;

The standard size of the size of step g gained with the same model rail part preserved in a database is compared by h, control device, if measured size is in the margin tolerance of permission, end surface shape and the size that then can conclude this rail part are qualified, otherwise point out underproof position, for processing provides foundation further.

The margin tolerance allowed in step h is: rail head width ± 0.5mm; Web of the rail thickness+0.75mm ,-0.5mm; Flange of rail width+1.0mm ,-2.0mm; High+the 0.8mm of rail.

The invention has the beneficial effects as follows: the measuring method providing a kind of railroad turnout steel rail part face size, overcome the error of indication due to steel tape itself existed when artificial employing steel tape is measured, the reading error that during measurement, tension variations causes, the error that temperature variation is brought causes the defect of final measuring error, the measurement of rail part end face region of interest size directly can be realized at rail part workshop, achieve non-contact measurement, surveying work just can be completed at the scene, flexibly and easily, reduce measuring error, measurement result is accurate, to greatly improve precision and the speed of measurement, thus greatly enhance productivity.

Accompanying drawing explanation

Fig. 1 is the structural representation of measurement mechanism in the present invention.

Fig. 2 is the position relationship schematic diagram of the present invention's calibrating block, industrial camera and rail part when measuring.

In accompanying drawing, 1 represents base, and 2 represent industrial camera, 3 represent bracing frame, and 4 represent leading screw, and 5 represent motor, 6 represent reductor, and 7 represent slide block, 8 representative location cross bars, 9 representative location vertical poles, 10 represent all-directional tripod head, and 11 represent shaft coupling, 12 represent pulley, and 13 represent control desk, and 14 represent support bar, 15 represent Turning-over plate chair, and 16 represent rail part, and 17 represent calibrating block.

Embodiment

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

As shown in Figure 1, a kind of railroad turnout steel rail part face size measurement mechanism, comprise base 1, two industrial cameras 2, control device, arrangement for adjusting height, two industrial cameras 2 are all connected with control device with arrangement for adjusting height, arrangement for adjusting height is vertically arranged on base 1, two industrial cameras 2 are symmetricly set on the both sides of arrangement for adjusting height in the horizontal direction and are formed to be elevated with arrangement for adjusting height and coordinate, and two industrial cameras 2 are formed to be elevated with base 1 by arrangement for adjusting height and coordinate.

Arrangement for adjusting height comprises the bracing frame 3 of door type structure, leading screw 4, motor 5, reductor 6, fixed support, the output shaft of motor 5 is connected with reductor 6 by shaft coupling 11, the bottom of bracing frame 3 is fixedly connected with base 1, leading screw 4 is vertically arranged on the inside of bracing frame 3 door type structure, motor 5 is arranged on the top of bracing frame 3 and is connected with control device, the top perforating branch support 3 of leading screw 4 is fixedly connected with the output shaft of reductor 6, the bottom of leading screw 4 and base 1 are formed and are rotatably assorted, leading screw 4 is set with slide block 7, fixed support comprises location cross bar 8, the two ends of location cross bar 8 are respectively arranged with a location vertical pole 9, two location vertical poles 9 are symmetrical arranged and all cross bar 8 is vertical arranges with location, the end of two location vertical poles 9 is all fixed with all-directional tripod head 10, industrial camera 2 is hinged with all-directional tripod head 10, location cross bar 8 is fixedly connected with slide block 7, industrial camera 2 coordinates to be formed to be elevated with base 1 by slide block 7 and leading screw 4 and coordinates.During measurement, motor 5 drives reductor 6 to rotate by shaft coupling 11, reductor 6 drives leading screw 4 to rotate, slide block 7 is made to rise on leading screw 4 or decline, slide block 7 drives two industrial cameras 2 to rise or decline by fixed support, and when two industrial cameras 2 arrive required height, motor 5 stops operating, and then utilizes all-directional tripod head 10 to adjust the angle of industrial camera 2, structure is simple, easy to operate.

As to further improvement of the present invention, multiple pulley 12 being set in the bottom of base 1, making this device convenient laborsaving when moving horizontally.

As to further improvement of the present invention, set up the control desk 13 for placing control device, the back side of control desk 13 is fixedly connected with bracing frame 3, the bottom of control desk 13 is fixed with support bar 14, the bottom of support bar 14 is fixedly connected with the upper end of base 1, be arranged on by control device on control desk 13, integrally, during measurement, handled easily personal observations controls.

As to further improvement of the present invention, set up the Turning-over plate chair 15 matched with control desk 13, the bottom of Turning-over plate chair 15 is fixedly connected with base 1, and operating personnel can be sitting in the enterprising line operate of Turning-over plate chair 15, convenient comfortable.

Use a measuring method for railroad turnout steel rail part face size measurement mechanism, key is: described method comprises the following steps:

A, preparation rail part 16 and calibrating block 17, calibrating block 17 be centre dig one with the scaling board in the size of rail part 16 end face, hole that shape is identical, the middle position of scaling board is positioned at by cutouts, web of the rail position, both sides is with chequered with black and white chessboard figure, in addition, also respectively have one piece of chessboard figure in this scaling board left and right sides, two pieces of chessboard figure become fixed angle with scaling board, and three is an entirety;

B, carry out industrial camera demarcation: it is exactly set up the relation between industrial camera 2 image pixel coordinates and corresponding spatial point coordinate that industrial camera 2 is demarcated, the operation of extraction angle point is carried out to the calibrating block image of the different angles collected, extract all angle points, the image coordinate system set up, corresponding relation between industrial camera 2 coordinate system and world coordinate system three coordinate systems, try to achieve the inside and outside parameter of industrial camera 2, the accuracy of the spatial point coordinate that industrial camera 2 parameter of demarcating is determined, finally have influence on the accuracy of dimensional measurement, what the present invention adopted is the scaling method based on two-dimensional planar template that Zhang Zhengyou proposes, the method does not need concrete orientation and the displacement information of knowing plane template movement, and plane template makes simple, more flexible compared to additive method, due to the singularity of the calibrating block 17 that the present invention adopts, timing signal only needs collection chessboard figure, and industrial camera 2 is fixed, so only need to demarcate once, (Zhengyou Zhang.A Flexible New Technique for Camera Calibration.IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE after system being carried out to demarcation, VOL.22, NO.11, NOVEMBER 2000), the relation can set up between pixel coordinate with corresponding spatial point coordinate is as follows:

$\begin{array}{c}z\left[\begin{array}{c}u\\ v\\ 1\end{array}\right]=\left[\begin{array}{ccc}1/\mathrm{dX}& \tan \mathrm{\α}/\mathrm{dY}& u_0\\ 0& 1/\mathrm{dY}& v_0\\ 0& 0& 1\end{array}\right]\left[\begin{array}{cccc}f& 0& 0& 0\\ 0& f& 0& 0\\ 0& 0& 1& 0\end{array}\right]\left[\begin{array}{c}x\\ y\\ z\\ 1\end{array}\right]\\ =\begin{array}{c}\left[\begin{array}{cccc}{f}_x& \mathrm{\γ}& u_0& 0\\ 0& f_y& v_0& 0\\ 0& 0& 1& 0\end{array}\right]\left[\begin{array}{cc}R& T\\ 0^T& 1\end{array}\right]\left[\begin{array}{c}{X}_w\\ Y_w\\ Z_w\\ 1\end{array}\right]\end{array}\end{array}$

Wherein, the coordinate of (x, y, z) representation space point P in industrial camera 2 coordinate system, α denotation coordination axle out of plumb error, f _x=f/dX, f _y=f/dY is the equivalent focal length represented with pixel wide and length, and γ represents the nonorthogonality of image coordinate axle;

C, rail part 16 is positioned over ground or is placed on platform, the flange of rail kiss the earth of rail part 16 or platform are placed, remove the foreign material around rail part 16, calibrating block 17 is close to rail part 16 end face, the height of adjustment two industrial cameras 2 and angle, two industrial cameras 2 and rail part 16 end face is made to be in sustained height, and allow two industrial cameras 2 just to rail part 16 end face, require that the optical axis of two industrial cameras 2 is parallel, the optical axis of two industrial cameras 2 perpendicular to rail part 16 end face, as shown in Figure 2;

D, open lamphouse light source, rail part 16 end face is illuminated in side, make the complete jag facial contour of light, to remove shade, adjust two industrial cameras 2, rail part 16 end face is made to fall into the center of screen picture, and take two each screen views of industrial camera 2, gather the image of rail part 16 end face now, due to the singularity of the calibrating block 17 of employing, the foreign material of surrounding can be masked, outstanding rail part 16 end face, we only need two industrial cameras 2 respectively according to a sheet photo, but the impact of environmental factor is inevitable, a few width of many collections when gathering image, general three width, select the work that the good width of effect carries out the later stage,

E, control device first carry out medium filtering process to the image of steps d gained, then carry out binaryzation, adopt the edge detection algorithm based on Canny to obtain the edge of image, are specially:

Medium filtering is a kind of nonlinear digital filter technology first proposed by Tukey, through being usually used in removing the noise in image or other signal, traditional medium filtering generally adopts the moving window containing odd number point, replace the gray-scale value of specified point by the intermediate value of each point gray-scale value in window, the mathematical description of medium filtering is: set S as pixel (x ₀, y ₀) Neighbourhood set, the element that (x, y) is S, the gray-scale value of point (x, y) h (x, y) represents, | S| is the number of element in S set, to point (x ₀, y ₀) be smoothly expressed as

$h^{\′}(x_0,y_0)={\left[\underset{(x,y)\∈S}{\mathrm{sort}}h(x,y)\right]}_{\frac{\left|S\right|+1}{2}}$

Wherein, sort is expressed as h (x, y) sort, this is the principle of standard medium filtering, standard medium filtering needs to carry out a large amount of sequence work, calculated amount is very large, the performance simultaneously removing impulsive noise is larger by the impact of filter window size, certain contradiction is there is in suppression picture noise and protection details, therefore present invention uses a kind of median filtering algorithm of improvement, namely based on the filtering algorithm in extreme value and direction, noise image is divided into suspicious noise spot and signaling point set by the extreme point that first whether this algorithm be its neighborhood according to certain point, then 8 direction gradient templates in suspicious noise spot structure neighborhood are determined whether this point is real noise spot, finally medium filtering is carried out to noise spot, result shows that this algorithm denoising effect is obvious, Edge preservation (draws from Dou Yan very well simultaneously, Kong Lingfu. a kind of filtering algorithm [J] based on extreme value and direction. computer engineering and application .2007, 43 (28): 41-43.),

Then binaryzation is carried out to image, edge detection algorithm is adopted to obtain the edge of image, Canny edge detection operator is that J.Canny proposed in 1986, has the advantage that signal to noise ratio (S/N ratio) is large, accuracy of detection is high and calculated amount is little, be widely used compared with other edge detection operator; According to the definition of Canny operator, the intensity of center edge is the maximal value on gradient direction, and judge whether have maximal value in each neighborhood on its gradient direction respectively, if so, then this point is marginal point; Canny edge detection algorithm is the maximal value finding shade of gray in topography, test problems is converted into the process of the maximum value found a function, and detect strong and weak two edges respectively by two threshold values, therefore the method noise resisting ability is strong, and Canny edge detection algorithm specifically comprises: Gaussian smoothing, gradient calculation, non-maxima suppression and double-threshold comparison four is most of; What traditional Canny edge detection algorithm adopted is Gaussian smoothing, and Gaussian filter also can be smoothing as high fdrequency component using edge while filtering noise, marginal information may be lost when non-maxima suppression afterwards; Fixing high-low threshold value is utilized to carry out edge extracting, lack the adaptivity to different images, and arranging of high-low threshold relies on artificial acquisition completely, automaticity is low, for smooth, we adopt the median filtering algorithm of aforementioned improvement, the marginal information of needs can be retained like this, then calculate the amplitude and direction that have passed through level and smooth rear gradient of image and gray scale, carry out the finite difference formulations of single order local derviation, amplitude is M (x, y), gradient direction is H (x, y), and expression formula is:

$M(x,y)=\sqrt{f_x{(x,y)}^2+f_y{(x,y)}^2}$ H(x,y)＝arctan(f _y(x,y)/f _x(x,y))

Wherein, f _x(x, y) and f _y(x, y)

The neighborhood be respectively in the x and y direction is poor; Then non-maxima suppression is carried out, finally select the method that Ostu chooses as threshold value, its maximum feature is its adaptivity is based on the grey level histogram of image, get the variance maximal value of target and background as segmentation threshold, because one dimension Ostu algorithm can only find single threshold value, we using the threshold value that obtains as high threshold, it is 2/5 as Low threshold, these two values are realized double-threshold comparison as the high-low threshold value in Canny algorithm, and the sub-pixel edge based on least square fitting detects and measurement result can be made to reach more high precision;

F, utilize Gestalt Rule Extraction to go out the complete outline of rail part 16 end face, Gestalt psychology conclusion is numerous, and we mainly utilize its conclusion relevant to image procossing,

The first, continuity: refer to the special consciousness of the one of the mankind to lines, very short for some distance or incorporating aspects close to each other are easily become as a whole by people;

The second, integrality: refer to that the mankind can tend to incoherent figure jaggy is made it conjunction as much as possible thus becomes as a whole;

3rd, similarity: refer to that similar part (such as color is close) is easily combined into as a whole by people;

Also have the characteristics such as propinquity, (straight/bent) is linear altogether, closed, collimation and Curvature varying are minimum in addition, think that human visual system tends to them integrally to carry out perception, according to above rule gestalt in topological consciousness theory, use multiscale analysis technology, profile detected is above organized into groups, thus obtain objects in images closed contour, specific as follows:

According to the Gestalt rule in topological consciousness theory, the profile of rail part 16 end face should be continuous, closed, complete, if the parameterized form at an edge is B=v (t)={ v ₁(t ₁), v ₂(t ₂) ..., v _n(t _n), 0≤t _i≤ 1, if 1≤i≤n. v (0)=v (1), then B is a closed contour; If this closed contour just in time contains the end face of rail part 16, then B is exactly required for us; But often contour of object is made up of a series of edge, these edges are discontinuous often, propose a kind of method getting up to form continuous boundary by two edge conjunction here: suppose that there are two end points p at an edge of the image that step e obtains ₁(x, y) and p ₂(x, y), and have both direction, the direction gradient angle of regulation end points is the direction of these two end points,

In formula, R (p ₁) _y, R (p ₁) _xrepresent some p respectively ₁in the projection in y direction and x direction; R (p ₂) _y, R (p ₂) _xrepresent some p respectively ₂in the projection in y direction and x direction, if one in the direction of an end points at edge and its surrounding edge, the angular separation of an end points at a certain bar edge is minimum, then these two end points straight lines are coupled together and are merged into a new edge, finally detect one meet continuity, closed, integrality edge be exactly the closed contour of rail part 16 end face;

G, the data fitting of the closed contour obtained in step f is become geometric primitive, least square fitting principle is utilized to carry out fitting a straight line, extract the N number of discrete point on graph line edge feature, ask for the fitting a straight line y=kx+b of these points, this N number of discrete point coordinate (x obtained by step f _i, y _i) (i=1,2 ..., N) meet straight-line equation or as far as possible from straight line very close to, ask the minimum value of expression:

$S=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{[y_i-({\mathrm{kx}}_i+b)]}^2$

The matching of circle the N number of point on profile is carried out company to the squared-distance of fitting circle add summation, minimizes it:

$S=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(\sqrt{{(x_i-a)}^2+{(y_i-b)}^2}-r)}^2$

Wherein, (a, b) is central coordinate of circle, r is radius of a circle, image outline can be divided into straight line and curve like this, by the detection and indentification to these shapes, according to the relation between the pixel coordinate set up in step b with corresponding spatial point coordinate, obtain the size that we want;

The standard size of the size of step g gained with the same model rail part 16 preserved in a database is compared by h, control device, if measured size is in the margin tolerance of permission, end surface shape and the size that then can conclude this rail part 16 are qualified, otherwise point out underproof position, for processing provides foundation further, the margin tolerance of permission refers to: rail head width ± 0.5mm; Web of the rail thickness+0.75mm ,-0.5mm; Flange of rail width+1.0mm ,-2.0mm; High+the 0.8mm of rail.

Measuring method in the present invention overcomes the error of indication due to steel tape itself existed when artificial employing steel tape is measured, the reading error that during measurement, tension variations causes, the error that temperature variation is brought causes the defect of final measuring error, the measurement of rail part end face region of interest size directly can be realized at rail part 16 workshop, achieve non-contact measurement, surveying work just can be completed at the scene, flexibly and easily, reduce measuring error, measurement result is accurate, greatly can improve precision and the speed of measurement, thus greatly enhance productivity.

Claims (2)

1. a measuring method for railroad turnout steel rail part face size, is characterized in that: described method comprises the following steps:

A, preparation rail part (16) and calibrating block (17), calibrating block (17) be centre dig one with the scaling board in the size of rail part (16) end face, hole that shape is identical, the middle position of scaling board is positioned at by cutouts, web of the rail position, both sides is with chequered with black and white chessboard figure, in addition, also respectively have one piece of chessboard figure in this scaling board left and right sides, two pieces of chessboard figure become fixed angle with scaling board, and three is an entirety;

B, carry out industrial camera demarcation: adopt Zhang Zhengyou plane reference method to demarcate, the image of shooting calibrating block (17), the operation of extraction angle point is carried out to the image of calibrating block (17), extract all angle points, set up the corresponding relation between image coordinate system, industrial camera (2) coordinate system and world coordinate system three coordinate systems, obtain the inside and outside parameter of industrial camera (2), the relation can set up between pixel coordinate with corresponding spatial point coordinate is as follows:

$\begin{array}{c}z\left[\begin{array}{c}u\\ v\\ 1\end{array}\right]=\left[\begin{array}{ccc}1/\mathrm{dX}& \tan \mathrm{\α}/\mathrm{dY}& u_0\\ 0& 1/\mathrm{dY}& v_0\\ 0& 0& 1\end{array}\right]\left[\begin{array}{cccc}f& 0& 0& 0\\ 0& f& 0& 0\\ 0& 0& 1& 0\end{array}\right]\left[\begin{array}{c}x\\ y\\ z\\ 1\end{array}\right]\\ =\left[\begin{array}{cccc}{f}_x& \mathrm{\γ}& u_0& 0\\ 0& f_y& v_0& 0\\ 0& 0& 1& 0\end{array}\right]\left[\begin{array}{cc}R& T\\ 0^T& 1\end{array}\right]\left[\begin{array}{c}{X}_w\\ Y_w\\ Z_w\\ 1\end{array}\right]\end{array}$

Wherein, the coordinate of (x, y, z) representation space point P in industrial camera (2) coordinate system, α denotation coordination axle out of plumb error, f _x=f/dX, f _y=f/dY is the equivalent focal length represented with pixel wide and length, and γ represents the nonorthogonality of image coordinate axle;

C, rail part (16) is positioned over ground or is placed on platform, the flange of rail kiss the earth of rail part (16) or platform are placed, remove rail part (16) foreign material around, calibrating block (17) is close to rail part (16) end face, adjust height and the angle of two industrial cameras (2), two industrial cameras (2) and rail part (16) end face is made to be in sustained height, and allow two industrial cameras (2) just to rail part (16) end face, require that the optical axis of two industrial cameras (2) is parallel, the optical axis of two industrial cameras (2) is perpendicular to rail part (16) end face,

D, open lamphouse light source, rail part (16) end face is illuminated in side, make the complete jag facial contour of light, to remove shade, adjust two industrial cameras (2), make rail part (16) end face fall into the center of screen picture, and take two each screen views of industrial camera (2), gather the image of rail part (16) end face now;

E, control device first carry out medium filtering process to the image of steps d gained, then carry out binaryzation, adopt the edge detection algorithm based on Canny to obtain the edge of image;

F, according to the Gestalt rule in topological consciousness theory, the profile of rail part (16) end face should be continuously, closed, complete, if the parameterized form at an edge is B=v (t)={ v ₁(t ₁), v ₂(t ₂) ..., v _n(t _n), 0≤t _i≤ 1, if 1≤i≤n. v (0)=v (1), then B is a closed contour; If this closed contour just in time contains the end face of rail part (16), then B is exactly required for us; But often contour of object is made up of a series of edge, these edges are discontinuous often, propose a kind of method getting up to form continuous boundary by two edge conjunction here: suppose that there are two end points p at an edge of the image that step e obtains ₁(x, y) and p ₂(x, y), and have both direction, the direction gradient angle of regulation end points is the direction of these two end points,

In formula, R (p ₁) _y, R (p ₁) _xrepresent some p respectively ₁in the projection in y direction and x direction; R (p ₂) _y, R (p ₂) _xrepresent some p respectively ₂in the projection in y direction and x direction, if one in the direction of an end points at edge and its surrounding edge, the angular separation of an end points at a certain bar edge is minimum, then these two end points straight lines are coupled together and are merged into a new edge, finally detect one meet continuity, closed, integrality edge be exactly the closed contour of rail part (16) end face;

G, the data fitting of the closed contour obtained in step f is become geometric primitive, least square fitting principle is utilized to carry out fitting a straight line, extract the N number of discrete point on graph line edge feature, ask for the fitting a straight line y=kx+b of these points, this N number of discrete point coordinate (x obtained by step f _i, y _i) (i=1,2 ..., N) meet straight-line equation or as far as possible from straight line very close to, ask the minimum value of expression:

$S=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{[y_i-({\mathrm{kx}}_i+b)]}^2$

The matching of circle the N number of point on profile is carried out company to the squared-distance of fitting circle add summation, minimizes it:

$S=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(\sqrt{{(x_i-a)}^2+{(y_i-b)}^2}-r)}^2$

Wherein, (a, b) is central coordinate of circle, r is radius of a circle, image outline can be divided into straight line and curve like this, by the detection and indentification to these shapes, according to the relation between the pixel coordinate set up in step b with corresponding spatial point coordinate, obtain the size that we want;

The standard size of the size of step g gained with the same model rail part (16) preserved in a database is compared by h, control device, if measured size is in the margin tolerance of permission, end surface shape and the size that then can conclude this rail part (16) are qualified, otherwise point out underproof position, for processing provides foundation further.

2. the measuring method of a kind of railroad turnout steel rail part face size according to claim 1, is characterized in that: the margin tolerance allowed in step h is: rail head width ± 0.5mm; Web of the rail thickness+0.75mm ,-0.5mm; Flange of rail width+1.0mm ,-2.0mm; High+the 0.8mm of rail.