CN102519400B - Large slenderness ratio shaft part straightness error detection method based on machine vision - Google Patents

Abstract

The invention, which belongs to a shaft part straightness error detection field, discloses a large slenderness ratio shaft part straightness error detection method based on machine vision. The method comprises the following steps: (1) installing standard components; (2) installing a camera; (3) installing workpiece; (4) collecting an image; (5) processing the image; (6) establishing a coordinate system; (7) calculating and acquiring a position coordinate of a center of a circle; (8) unifying the coordinate; (9) according to the coordinate of the center of the circle of a cross section, using a current straightness evaluation method to calculate the straightness. According to the invention, a position relation between the standard components is taken as a link and a coordinate position relation between the images is established so as to avoid using an image splicing technology with long time consuming. Simultaneously, a distance from the camera to the part does not need to be the same with the distance from the camera to the standard components. Measuring difficulty can be reduced and measuring precision can be increased.

Description

Big L/D ratio axial workpiece straightness error detection method based on machine vision

Technical field

The present invention relates to a kind of axial workpiece straightness error detection method, more particularly, relate to a kind of big L/D ratio axial workpiece straightness error detection method based on machine vision.

Background technology

Vision is human perception and the world of understanding periphery, obtains the topmost means of knowledge from the Nature.In the information of obtaining the mankind, 80% from vision, and remaining is 20% from the sense of hearing, the sense of taste, sense of touch etc.Machine vision, refers to and utilizes computing machine and some utility appliance to realize people's visual performance, thereby realizes by the extraneous things of two dimensional image perception and objective three-dimensional world, is an emerging subject, is one of interesting research frontier.

Along with the development of machine vision technique, there is a kind of means or the new detection method that carrier is used, i.e. image detecting technique based on machine vision of image being used as to detection and transmission of information.It is to take contemporary optics as basis, melts the modern detecting that the science and technology such as optoelectronics, computer graphics, information processing, machine vision are integrated.The instrument and equipment of the image detecting technique based on machine vision can be realized intellectuality, digitizing, miniaturization, networking and multifunction, possess online detection, real-time analysis, the real-time ability of controlling, in fields such as military affairs, industry, medical science, obtain extensive concern and application.

Image detecting method based on machine vision has had application in mechanical component physical dimension detects, and mainly concentrates on dimensional measurement and the dimensional measurement based on Image Mosaics based on single image.Single image detection method is with piece image, to show the overall picture of whole part, obtains the physical dimension of part by image operation.But because the field range of camera suffers restraints, because the method is only applicable to the measurement of smaller parts.For heavy parts, image detecting technique can be taken the different parts of part, obtain several local overlapping high precision images, then utilize the information redundancy between image to carry out the complete image that Image Mosaics obtains part, to spliced image analysis, can obtain the complete structure size of part.A large amount of documents and materials point out, this detection method need to be carried out a large amount of data storage and computing, certainly will affect the real-time of detection method.

Existing mechanical component verticality measuring method, mainly concentrates on the straight line degree measurement based on single image.Verticality measuring method based on single image is with piece image, to show the overall picture of measured axis, by the calculation process of this piece image being obtained to the straightness error of part.But because field range and the precision of camera is inversely prroportional relationship, in order to obtain higher accuracy of detection, the physical dimension of tested part must retrain within the specific limits.The post-doctoral research work report < < vision criteria line Measurement Technique research > > of University Of Tianjin (Lu Rongsheng) is 140mm to diameter, length is that the Linearity surveying method of the weldless steel tube of 1.5m is studied, and is the domestic big L/D ratio axial workpiece Linearity surveying method research scholar early that carries out.The method adopting in its report is a kind of laser vision method, and checkout equipment precision is higher, but comparison in equipment is expensive.In addition, need to carry out global calibration to imageing sensor, algorithm real-time is subject to certain restrictions.

1. prior art one related to the present invention

The technical scheme of prior art one: certain rail vehicle door is about 2m with change lead screw, and diameter is about 2cm, and length-diameter ratio is 100, belongs to big L/D ratio axial workpiece.Its linearity is the important indicator of quality control, is also its its functional attributes, from client's feedack, even detect at production link, uses after a period of time, and its linearity still there will be larger error.Research is found, is mainly because the precision of detection means is not high enough.

At present, production line is that leading screw is placed on work top in the detection method of use, goes to observe the situation that contacts of leading screw and work top with eyes, and the local feeler gauge by standard of printing opacity goes detection, and somewhere can be put feeler gauge into, the straightness error at this place has been described

The shortcoming of prior art one: above-mentioned detection method exists following shortcoming:

(1) can not accurately provide the data of straightness error;

(2) can not formulate rational correcting scheme according to data;

(3) this post is larger to workman's skills involved in the labour dependence;

(4) workman's labour intensity is larger;

(5) detection speed is slow, cannot realize production line automation.

2. prior art two related to the present invention

The technical scheme of prior art two: Fig. 1 selects from the Lu Rong of University Of Tianjin victory post-doctoral performance report < < vision criteria line Measurement Technique research > >, this test macro sends light beam plane by semiconductor laser LD, after it cuts mutually with weldless steel tube, outer round surface at steel pipe forms an elliptic arc, by CCD gamma camera, received the image of each elliptic arc, machine realtime graphic is processed and is tried to achieve each coordinate of elliptic arc center in world coordinate system as calculated, use corresponding algorithm, obtain the straightness error of steel pipe.

The shortcoming of prior art two: adopted the semiconductor laser projector in Fig. 1, this element belongs to high-precision product, price comparison is expensive.In method, adopt a plurality of ccd video cameras to gather image, need the image unification gathering can carry out the calculating of linearity under same coordinate system, therefore need comparatively complicated calibration algorithm.Demarcation is a difficult point in machine vision, and existing various calibration techniques have certain limitation, available in certain accuracy rating.As low in linear calibration's computational accuracy, nonlinear calibration needs repeatedly nonlinear iteration, and computing velocity is low, thus calibration technique become a bottleneck.Therefore, said method again by other high precision apparatus carried out global calibration, this has brought difficulty just to the popularization of method.

3. prior art three related to the present invention

The technical scheme of prior art three: the part planar dimension of the academic dissertation < < of Shandong University based on machine vision automatically measured and write the linearity that detects part by machine vision method in > > (Marvin's is beautiful), detailed process is as follows: image capture device obtains the overall picture of part, use image processing method to obtain gray-scale map, gray-scale map is carried out to binary conversion treatment, use again edge detection operator, obtain edge pixel level point set, if need further to improve precision, re-use sub-pixel edge location technology, obtain sub-pixel edge point set, edge point set carries out least square fitting and obtains ideal line, marginal point is concentrated and apart from maximal value, is considered to straightness error apart from ideal line.

The shortcoming of prior art three: the verticality measuring method based on single image is to show the overall picture of tested part with piece image, by the calculation process of this piece image being obtained to the straightness error of part.But because field range and the precision of camera is inversely prroportional relationship, in order to obtain the overall picture of part, must sacrifice certain precision; Or in order to obtain higher accuracy of detection, the physical dimension of tested part must retrain within the specific limits.

Summary of the invention

1. the technical matters that invention will solve

The invention reside in and overcome deficiency of the prior art, a kind of big L/D ratio axial workpiece straightness error detection method based on machine vision is provided, by the high precision image of quick each regional area of acquisition measured axis, adopt simple image processing algorithm to obtain the centre coordinate of a plurality of positions, the evaluation of straightness error will be carried out again under separate central point unification to coordinate system, to obtain whole axle straightness error value, and overproof part is carried out to record, for follow-up correction provides data.

2. technical scheme

For achieving the above object, technical scheme provided by the invention is:

A kind of big L/D ratio axial workpiece straightness error detection method based on machine vision of the present invention, the steps include:

(1) standard component is installed:

Many groups of the contour installations of standard component, the corresponding sides of all standard components are parallel to each other, and the physical dimension of two standard components and the position relationship between standard component adopt conventional high precision testing tool to measure and obtain;

(2) camera is installed:

Guarantee that imaging plane is parallel with the upper surface of standard component, and phase function is along the axis direction of measured axis with move perpendicular to work top direction;

(3) workpiece is installed:

Measured axis is placed in visual field, adjusts position, make the axis of measured axis parallel or vertical with standard component limit;

(4) image acquisition:

When camera is in i sample point, adjust camera position, making camera is D to the distance of measured axis ₁, gather 1 width image, be designated as image _1i, adjust camera position, making camera is D to measured axis distance ₂, then gather 1 width image, be designated as image _2i; Along measured axis axis direction, move camera to next sampled point i+1 place, adjust camera position, making camera is D to the distance of measured axis ₁, gather 1 width image, be designated as image _1i+1, adjust camera position, making camera is D to measured axis distance ₂, then gather 1 width image, be designated as image _2i+1, so along axis direction, move camera, until to last sampled point, complete the collection of whole axis direction epigraph sequence, wherein, D ₁and D ₂two unknown variablees, but D ₁and D ₂between difference be known quantity;

(5) image is processed:

Image ₁in sequence image, i width image is processed, and adopts Harris Robust Algorithm of Image Corner Extraction to extract 8 angle points of two standard components, uses Canny Boundary extracting algorithm to extract the contour edge of axle; Image ₂in sequence image, i width image is processed, and adopts Harris Robust Algorithm of Image Corner Extraction to extract 8 angle points of two standard components, uses Canny Boundary extracting algorithm to extract the contour edge of axle;

(6) set up coordinate system:

Image ₁on image sequence Plays part, the angle point in the lower left corner is true origin, and x axle parallels to the axis, and y axle is perpendicular to axis, and z axle is perpendicular to work top;

(7) calculate home position coordinate:

At sequence image image _1imeasured axis profile on get 2 points, guarantee that 2 points are identical perpendicular to axis direction epigraph coordinate, at sequence image image _2imeasured axis profile on get 2 and image _1iin the corresponding point of point; According to pinhole imaging system principle, above-mentioned got 4 points are concyclic, photocentre is vertical with the determined straight line in point and the center of circle with the determined straight line of point, the distance of mobile camera moving (| D ₁-D ₂|) 4 conditions, calculate home position coordinate;

Wherein, calculate home position coordinate time and set up system mathematic model, when camera is during in 1# position, the symmetric points p on measured axis profile ₉and p ₁₁imaging; When camera is during in 2# position, the symmetric points p on measured axis profile ₈and p ₁₀imaging, 8 angle points (true origin and p on standard component _i(i=1,2,3 ..., 7)) between distance be known parameter, then according to pinhole imaging system principle, p _i(i=8,9,10,11) 4 concyclic, O _cp ₈⊥ p ₈o, O _cp ₉⊥ p ₉it is as follows that known these conditions of the distance of o, mobile camera moving (1# position is to 2# positional distance) are set up system mathematic model:

O _w-X _wy _wz _wfor world coordinate system, O _c-X _cy _cz _cfor camera coordinates system, 1#, 2# are camera two positions of living in, and o is the center of circle of cross section, axle somewhere circle, p _i(i=1,2,3 ..., 11) and coordinate is respectively: p ₁(x ₁, 0,0), p ₂(x ₁, y ₂, 0), p ₃(0, y ₂, 0); p ₄(x ₄, y ₄, 0), p ₅(x ₅, y ₄, 0), p ₆(x ₄, y ₆, 0), p ₇(x ₅, y ₆, 0); p ₈(x ₈, y, z ₈), p ₉(x ₉, y, z ₉), p ₁₀(x ₁₀, y, z ₁₀), p ₁₁(x ₁₁, y, z ₁₁);

When camera is during in 1# position, camera photocentre O _cto center of circle o distance, be D ₁, when camera is during in 2# position, camera photocentre O _cto center of circle o distance, be D ₂, and suppose that center of circle coordinate under world coordinate system is o (x, y, z), have:

x ₉+x ₁₁＝2x (1.1)

z ₉＝z ₁₁ (1.2)

${\stackrel{\&RightArrow;}{a}}_1(x_9-x,0,z_9-z+D_1)\&CenterDot;{\stackrel{\&RightArrow;}{b}}_1(x_9-x,0,z_9-z)=0---\left(1.3\right)$

$\frac{f}{\mathrm{dx}}\&CenterDot;(x_9-x)/(z_9-z+D_1)-\frac{f}{\mathrm{dx}}\&CenterDot;(x_{11}-x)/(z_{11}-z+D_1)=C_1---\left(1.4\right)$

${\stackrel{\&OverBar;}{\mathrm{\&beta;}}}_x^1=\frac{f}{\mathrm{dx}}\&CenterDot;\frac{1}{-z+D_1}---\left(1.5\right)$

${\stackrel{\&OverBar;}{\mathrm{\&beta;}}}_{\mathrm{<figure-callout\; id="1"\; label="y"\; filenames="111215062404.png,111215062407.png"\; state="\{\{state\}\}">y</figure-callout>}}^1=\frac{f}{\mathrm{dy}}\&CenterDot;\frac{1}{-z+D_1}---\left(1.6\right)$

x ₈+x ₁₀＝2x (1.7)

z ₈＝z ₁₀ (1.8)

${\stackrel{\&RightArrow;}{a}}_2(x_8-x,0,z_8-z+D_2)\&CenterDot;{\stackrel{\&RightArrow;}{b}}_2(x_8-x,0,z_8-z)=0---\left(1.9\right)$

$\frac{f}{\mathrm{dx}}\&CenterDot;(x_8-x)/(z_8-z+D_2)-\frac{f}{\mathrm{dx}}\&CenterDot;(x_{10}-x)/(z_{10}-z+D_2)=C_2---\left(1.10\right)$

(x ₈-x) ²+(z ₈-z) ²＝(x ₉-x) ²+(z ₉-z) ² (1.11)

D ₁-D ₂＝C (1.12)

${\stackrel{\&OverBar;}{\mathrm{\&beta;}}}_x^2=\frac{f}{\mathrm{dx}}\&CenterDot;\frac{1}{-z+D_2}---\left(1.13\right)$

${\stackrel{\&OverBar;}{\mathrm{\&beta;}}}_{\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="111215062407.png"\; state="\{\{state\}\}">y</figure-callout>}}^2=\frac{f}{\mathrm{dy}}\&CenterDot;\frac{1}{-z+D_2}---\left(1.14\right)$

Wherein, for straight line O _cp ₉direction,

for straight line p ₉o direction,

for straight line O _cp ₈direction, for straight line p ₈o direction; C ₁for p ₉point look like p ₁₁distance between some picture; C ₂for p ₈point look like p ₁₀distance between some picture;

for when camera is during in 1# position, along X _cdirection of principal axis is measured ratio, by two standard components, is parallel to X _cthe length on 4 limits of axle and the ratio of the length of its corresponding picture determines,

for when camera is during in 1# position, along Y _cdirection of principal axis is measured ratio, by two standard components, is parallel to Y _cthe length on 4 limits of axle and the ratio of the length of its corresponding picture determines,

for when camera is during in 2# position, along X _cdirection of principal axis is measured ratio, by two standard components, is parallel to X _cthe length on 4 limits of axle and the ratio of the length of its corresponding picture determines,

for when camera is during in 2# position, along Y _cdirection of principal axis is measured ratio, by two standard components, is parallel to Y _cthe length on 4 limits of axle and the ratio of the length of its corresponding picture are determined; C is the distance that camera was moved to 2# position from 1# position; F is the focal length of camera; Dx is the size of pixel on directions X, and dy is pixel size in the Y direction;

Simultaneous formula (1.1)～formula (1.14), is the mathematical model of this measuring system, according to the mathematical model of this measuring system, calculates home position coordinate;

(8) unify coordinate:

According to the installation dimension relation between standard component, the above-mentioned coordinate calculating is unified under same coordinate system;

(9), according to the central coordinate of circle in cross section, utilize linearity assessment method, calculated line degree.

3. beneficial effect

Adopt technical scheme provided by the invention, compare with existing known technology, there is following remarkable result:

(1) a kind of big L/D ratio axial workpiece straightness error detection method based on machine vision of the present invention, against existing technologies one, the present invention is the carrier as detection and transmission of information with image, be stored in computing machine, can obtain straightness error data accurately, according to the straightness error obtaining, formulate rational correcting scheme, the long-time steady operation of machine vision, the mankind are difficult to for a long time same target be observed, and machine vision can be measured for a long time, analysis and identification mission;

(2) a kind of big L/D ratio axial workpiece straightness error detection method based on machine vision of the present invention, against existing technologies two, the invention belongs to common vision, do not need to adopt the expensive high-precision laser projector, the scaling method of machine vision Size Measuring System mostly adopts standard component method, the accurate dimension of standard component is passed to image, the major advantage of this method is not need the inside and outside parameter of calibrating camera, only need to calibrate the object plane resolution of video camera, can simplify calibration process, the method is only applicable to measurement plane dimension of object, as long as while guaranteeing each collection image, in visual field, there are two standard components simultaneously, can whole coordinates be united according to mutual alignment relation between two standard components, this calibration algorithm does not need to carry out repeatedly nonlinear iteration and calculates, computing velocity is higher, precision mainly depends on the imaging precision of standard component precision and camera, can require configuration standard part and video camera according to accuracy class,

(3) the present invention be take image as detecting and transmission of information carrier, can obtain the straightness error data with data accurate description, data are stored in to be calculated, and can link up with a plurality of departments such as design, production and production quality control, with the quality of improving product in good time;

(4) a kind of big L/D ratio axial workpiece straightness error detection method based on machine vision of the present invention, do not need longer Image Mosaics technology consuming time, obtain part overall picture, do not carry out very complicated and need the world coordinates that high precision apparatus is supported to demarcate yet, obtain axis each point absolute coordinates, carry out the calculating of linearity;

(5) accuracy of detection of the present invention is only relevant with standard component precision with the imaging precision of camera, and camera moves and do not need to do special processing with locating along measured axis axis direction;

(6) the present invention can realize any length-diameter ratio axial workpiece Linearity surveying task, only needs the size of proper extension worktable, increases the standard component of some.

Accompanying drawing explanation

Test macro in the Lu Rong of Tu1Wei University Of Tianjin victory post-doctoral performance report < < vision criteria line Measurement Technique research > >;

Fig. 2 is measuring system scheme of installation of the present invention;

Fig. 3 is Measuring System Models figure of the present invention.

Label declaration in schematic diagram:

1,3-standard component; 2-measured axis.

Embodiment

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

In conjunction with Fig. 2 and Fig. 3, technical solution of the present invention is set forth." 1 " in Fig. 2 and " 3 " are 2 standard components, and its size is not identical, and " 2 " in Fig. 2 are certain measured axis.Before measurement, to carry out the installation of camera, measured axis 2 and standard component 1,3.Standard component 1 and the contour installation of standard component 3, and parallel (or vertical) on assurance limit, conventional high precision testing tool measurement acquisition for the position relationship between the physical dimension of two standard components 1,3 and standard component 1,3.Camera is installed and need to be guaranteed that imaging plane is parallel with the upper surface of standard component 1,3, and phase function is along the axis direction of measured axis 2 with move perpendicular to work top direction.Measured axis 2 is installed in visual field, adjusts its position, guarantees the axis of measured axis 2 and the limit parallel (or vertical) of standard component 1,3.When camera is in i sample point, adjust camera position, making camera is D to the distance of measured axis 2 ₁, gather 1 width image, be designated as image _1i; Adjust camera position, making camera is D to measured axis 2 distances ₂, then gather 1 width image, be designated as image _2i.Along measured axis 2 axis directions, move camera to next sampled point i+1 place, adjust camera position, making camera is D to the distance of measured axis 2 ₁, gather 1 width image, be designated as image _1i+1; Adjust camera position, making camera is D to measured axis 2 distances ₂, then gather 1 width image, be designated as image _2i+1.So along axis direction, move camera, until to last sampled point, complete the collection of whole axis direction epigraph sequence.D ₁and D ₂two unknown variablees, but D ₁and D ₂between difference be known quantity.To image ₁in sequence image, i width image is processed, and extracts 8 angle points of two standard components 1,3 with Harris angle point (C.Harris, etc.1998) extraction algorithm, extracts the contour edge of axle with Canny (John Harris, 1986) Boundary extracting algorithm.To image ₂in sequence image, i width image is processed, and extracts 8 angle points of two standard components 1,3 with Harris Robust Algorithm of Image Corner Extraction, extracts the contour edge of axle with Canny Boundary extracting algorithm.Coordinate system is: image ₁on image sequence Plays part 1,3, the angle point in the lower left corner is true origin, and x axle is perpendicular to axis, and y axle parallels to the axis, and z axle is perpendicular to work top.At sequence image image _1imeasured axis 2 profiles on get 2 points, guarantee that 2 points are identical perpendicular to axis direction epigraph coordinate.At sequence image image _2imeasured axis 2 profiles on get 2 and image _1ithe point that mid point is corresponding.

Fig. 3 is Measuring System Models figure, when camera is during in 1# position, and the some p on measured axis 2 profiles ₉and p ₁₁imaging.When camera is during in 2# position, the some p on measured axis 2 profiles ₈and p ₁₀imaging.8 angle points (true origin and p on standard component 1,3 _i(i=1,2,3 ..., 7)) between distance be known parameter.Again according to pinhole imaging system principle, p _i(i=8,9,10,11) 4 concyclic, straight line O _cp ₈with p ₈vertical and the straight line O of o _cp ₉with p ₉o also distance (1# position is to 2# positional distance) vertical, mobile camera moving is known, calculates home position coordinate.Utilize the installation dimension relation between standard component 1,3, the central coordinate of circle calculating is unified under same coordinate system.According to the central coordinate of circle in each cross section, use existing linearity assessment method, calculate the linearity of this measured axis 2.

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

Embodiment:

A kind of big L/D ratio axial workpiece straightness error detection method based on machine vision of the present embodiment, the steps include:

(1) standard component is installed:

Many groups of the contour installations of standard component, shown in Fig. 2, " 1 " and " 3 " is 2 standard components, its size is not identical, " 2 " in Fig. 2 are certain measured axis, the corresponding sides of all standard components are parallel to each other, and the physical dimension of two standard components and the position relationship between standard component adopt conventional high precision testing tool to measure and obtain;

(2) camera is installed:

Guarantee that imaging plane is parallel with the upper surface of standard component, and phase function is along the axis direction of measured axis with move perpendicular to work top direction;

(3) workpiece is installed:

Measured axis is placed in visual field, adjusts position, make the axis of measured axis parallel or vertical with standard component limit;

(4) image acquisition:

When camera is in i sample point, adjust camera position, making camera is D to the distance of measured axis ₁, gather 1 width image, be designated as image _1i, adjust camera position, making camera is D to measured axis distance ₂, then gather 1 width image, be designated as image _2i; Along measured axis axis direction, move camera to next sampled point i+1 place, adjust camera position, making camera is D to the distance of measured axis ₁, gather 1 width image, be designated as image _1i+1, adjust camera position, making camera is D to measured axis distance ₂, then gather 1 width image, be designated as image _2i+1, so along axis direction, move camera, until to last sampled point, complete the collection of whole axis direction epigraph sequence, wherein, D ₁and D ₂two unknown variablees, but D ₁and D ₂between difference be known quantity;

(5) image is processed:

Image ₁in sequence image, i width image is processed, and adopts Harris Robust Algorithm of Image Corner Extraction to extract 8 angle points of two standard components, uses Canny Boundary extracting algorithm to extract the contour edge of axle; Image ₂in sequence image, i width image is processed, and adopts Harris Robust Algorithm of Image Corner Extraction to extract 8 angle points of two standard components, uses Canny Boundary extracting algorithm to extract the contour edge of axle;

(6) set up coordinate system:

Image ₁on image sequence Plays part, the angle point in the lower left corner is true origin, and x axle parallels to the axis, and y axle is perpendicular to axis, and z axle is perpendicular to work top;

(7) calculate home position coordinate:

At sequence image image _1imeasured axis profile on get 2 points, guarantee that 2 points are identical perpendicular to axis direction epigraph coordinate, at sequence image image _2imeasured axis profile on get 2 and image _1iin the corresponding point of point; According to pinhole imaging system principle, above-mentioned got 4 points are concyclic, photocentre is vertical with the determined straight line in point and the center of circle with the determined straight line of point, the distance of mobile camera moving (| D ₁-D ₂|) 4 conditions, calculate home position coordinate;

Wherein, calculate home position coordinate time and set up system mathematic model as shown in Figure 3, when camera is during in 1# position, the symmetric points p on measured axis profile ₉and p ₁₁imaging; When camera is during in 2# position, the symmetric points p on measured axis profile ₈and p ₁₀imaging, 8 angle points (true origin and p on standard component _i(i=1,2,3 ..., 7)) between distance be known parameter, then according to pinhole imaging system principle, p _i(i=8,9,10,11) 4 concyclic, O _cp ₈⊥ p ₈o, O _cp ₉⊥ p ₉it is as follows that known these conditions of the distance of o, mobile camera moving (1# position is to 2# positional distance) are set up system mathematic model:

O _w-X _wy _wz _wfor world coordinate system, O _c-X _cy _cz _cfor camera coordinates system, 1#, 2# are camera two positions of living in, and o is the center of circle of cross section, axle somewhere circle, p _i(i=1,2,3 ..., 11) and coordinate is respectively: p ₁(x ₁, 0,0), p ₂(x ₁, y ₂, 0), p ₃(0, y ₂, 0); p ₄(x ₄, y ₄, 0), p ₅(x ₅, y ₄, 0), p ₆(x ₄, y ₆, 0), p ₇(x ₅, y ₆, 0); p ₈(x ₈, y, z ₈), p ₉(x ₉, y, z ₉), p ₁₀(x ₁₀, y, z ₁₀), p ₁₁(x ₁₁, y, z ₁₁);

When camera is during in 1# position, camera photocentre O _cto center of circle o distance, be D ₁, when camera is during in 2# position, camera photocentre O _cto center of circle o distance, be D ₂, and suppose that center of circle coordinate under world coordinate system is o (x, y, z), have:

x ₉+x ₁₁＝2x (1.1)

z ₉＝z ₁₁ (1.2)

${\stackrel{\&RightArrow;}{a}}_1(x_9-x,0,z_9-z+D_1)\&CenterDot;{\stackrel{\&RightArrow;}{b}}_1(x_9-x,0,z_9-z)=0---\left(1.3\right)$

$\frac{f}{\mathrm{dx}}\&CenterDot;(x_9-x)/(z_9-z+D_1)-\frac{f}{\mathrm{dx}}\&CenterDot;(x_{11}-x)/(z_{11}-z+D_1)=C_1---\left(1.4\right)$

${\stackrel{\&OverBar;}{\mathrm{\&beta;}}}_x^1=\frac{f}{\mathrm{dx}}\&CenterDot;\frac{1}{-z+D_1}---\left(1.5\right)$

${\stackrel{\&OverBar;}{\mathrm{\&beta;}}}_{\mathrm{<figure-callout\; id="1"\; label="y"\; filenames="111215062404.png,111215062407.png"\; state="\{\{state\}\}">y</figure-callout>}}^1=\frac{f}{\mathrm{dy}}\&CenterDot;\frac{1}{-z+D_1}---\left(1.6\right)$

x ₈+x ₁₀＝2x (1.7)

z ₈＝z ₁₀ (1.8)

${\stackrel{\&RightArrow;}{a}}_2(x_8-x,0,z_8-z+D_2)\&CenterDot;{\stackrel{\&RightArrow;}{b}}_2(x_8-x,0,z_8-z)=0---\left(1.9\right)$

$\frac{f}{\mathrm{dx}}\&CenterDot;(x_8-x)/(z_8-z+D_2)-\frac{f}{\mathrm{dx}}\&CenterDot;(x_{10}-x)/(z_{10}-z+D_2)=C_2---\left(1.10\right)$

(x ₈-x) ²+(z ₈-z) ²＝(x ₉-x) ²+(z ₉-z) ² (1.11)

D ₁-D ₂＝C (1.12)

${\stackrel{\&OverBar;}{\mathrm{\&beta;}}}_x^2=\frac{f}{\mathrm{dx}}\&CenterDot;\frac{1}{-z+D_2}---\left(1.13\right)$

${\stackrel{\&OverBar;}{\mathrm{\&beta;}}}_{\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="111215062407.png"\; state="\{\{state\}\}">y</figure-callout>}}^2=\frac{f}{\mathrm{dy}}\&CenterDot;\frac{1}{-z+D_2}---\left(1.14\right)$

Wherein,

for straight line O _cp ₉direction,

for straight line p ₉o direction,

for straight line O _cp ₈direction,

for straight line p ₈o direction; C ₁for p ₉point look like p ₁₁distance between some picture; C ₂for p ₈point look like p ₁₀distance between some picture;

for when camera is during in 1# position, along X _cdirection of principal axis is measured ratio, by two standard components, is parallel to X _cthe length on 4 limits of axle and the ratio of the length of its corresponding picture determines,

for when camera is during in 1# position, along Y _cdirection of principal axis is measured ratio, by two standard components, is parallel to Y _cthe length on 4 limits of axle and the ratio of the length of its corresponding picture determines,

for when camera is during in 2# position, along X _cdirection of principal axis is measured ratio, by two standard components, is parallel to X _cthe length on 4 limits of axle and the ratio of the length of its corresponding picture determines,

for when camera is during in 2# position, along Y _cdirection of principal axis is measured ratio, by two standard components, is parallel to Y _cthe length on 4 limits of axle and the ratio of the length of its corresponding picture are determined; C is the distance that camera was moved to 2# position from 1# position; F is the focal length of camera; Dx is the size of pixel on directions X, and dy is pixel size in the Y direction;

Simultaneous formula (1.1)～formula (1.14), is the mathematical model of this measuring system, according to the mathematical model of this measuring system, calculates home position coordinate;

(8) unify coordinate:

According to the installation dimension relation between standard component, the above-mentioned coordinate calculating is unified under same coordinate system;

(9) according to the central coordinate of circle in cross section, use existing linearity assessment method, in the present embodiment, adopt and use 3DLSA (Hu Zhongxun etc.) linearity evaluation algorithm, calculate the linearity of this measured axis.

Big L/D ratio axial workpiece straightness error detection method based on machine vision of the present invention, a kind of brand-new based on realizing of Robot Vision big L/D ratio axial workpiece straightness error On-line Measuring Method, its unique distinction is: take image as detecting and transmission of information carrier, can obtain the straightness error data with data accurate description, data are stored in calculating, can link up with a plurality of departments such as design, production and production quality control, with the quality of improving product in good time.In method of the present invention, do not use expensive high-precision laser projecting apparatus.The mutual position relationship of the standard component of take is tie, sets up the coordinate position relation between each image, thereby avoids using longer Image Mosaics technology consuming time.In the process of establishing of measuring system mathematical model, directly do not use and measure than method (or being referred to as equivalent pixel method), therefore do not need strict guarantee camera to equate to distance between standard component to distance and the camera of part, reduce and measure difficulty, improve measuring accuracy.

Claims (2)

1. the big L/D ratio axial workpiece straightness error detection method based on machine vision, the steps include:

(1) standard component is installed:

Many groups of the contour installations of standard component, every group is two standard components, and the corresponding sides of all standard components are parallel to each other, and the physical dimension of two standard components of every group and the position relationship between standard component adopt conventional high precision testing tool to measure and obtain;

(2) camera is installed:

Guarantee that imaging plane is parallel with the upper surface of standard component, and phase function is along the axis direction of measured axis with move perpendicular to work top direction;

(3) workpiece is installed:

Measured axis is placed in visual field, adjusts position, make the axis of measured axis parallel or vertical with standard component limit;

(4) image acquisition:

When camera is in i sample point, adjust camera position, making camera is D to the distance of measured axis ₁, gather 1 width image, be designated as image _1i, adjust camera position, making camera is D to measured axis distance ₂, then gather 1 width image, be designated as image _2i; Along measured axis axis direction, move camera to next sampled point i+1 place, adjust camera position, making camera is D to the distance of measured axis ₁, gather 1 width image, be designated as image _1i+1, adjust camera position, making camera is D to measured axis distance ₂, then gather 1 width image, be designated as image _2i+1, so along axis direction, move camera, until to last sampled point, complete the collection of whole axis direction epigraph sequence, wherein, D ₁and D ₂two unknown variablees, but D ₁and D ₂between difference be known quantity;

(5) image is processed:

Image ₁in sequence image, i width image is processed, and adopts Harris Robust Algorithm of Image Corner Extraction to extract 8 angle points of two standard components, uses Canny Boundary extracting algorithm to extract the contour edge of axle; Image ₂in sequence image, i width image is processed, and adopts Harris Robust Algorithm of Image Corner Extraction to extract 8 angle points of two standard components, uses Canny Boundary extracting algorithm to extract the contour edge of axle;

(6) set up coordinate system:

Image ₁on image sequence Plays part, the angle point in the lower left corner is true origin, and x axle parallels to the axis, and y axle is perpendicular to axis, and z axle is perpendicular to work top;

(7) the home position coordinate in cross section, reference axis somewhere:

At sequence image image _1imeasured axis profile on get 2 points, guarantee that 2 points are identical perpendicular to axis direction epigraph coordinate, at sequence image image _2imeasured axis profile on get 2 and image _1iin the corresponding point of point; According to pinhole imaging system principle, above-mentioned got 4 points are concyclic, photocentre is vertical with the determined straight line in point and the center of circle with the determined straight line of point, the distance of mobile camera moving | D ₁-D ₂| 4 conditions, calculate home position coordinate;

(8) unify coordinate:

According to the installation dimension relation between standard component, the above-mentioned coordinate calculating is unified under same coordinate system;

(9), according to the central coordinate of circle in cross section, utilize linearity assessment method, calculated line degree.

2. the big L/D ratio axial workpiece straightness error detection method based on machine vision according to claim 1, it is characterized in that: in step (7), calculate home position coordinate time and set up system mathematic model, when camera is during in 1# position, the symmetric points p on measured axis profile ₉and p ₁₁imaging; When camera is during in 2# position, the symmetric points p on measured axis profile ₈and p ₁₀imaging, true origin and p on standard component _i, i=1,2,3 ..., the distance between 8 angle points of 7 is known parameter, then according to pinhole imaging system principle, p _i, i=8,9,10,11 4 concyclic, O _cp ₈⊥ p ₈o, O _cp ₉⊥ p ₉the distance that o, camera move to 2# position from 1# position, it is as follows that known these conditions are set up system mathematic model:

O _w-X _wy _wz _wfor world coordinate system, O _c-X _cy _cz _cfor camera coordinates system, 1#, 2# are camera two positions of living in, and o is the center of circle of cross section, axle somewhere circle, p _i, i=1,2,3 ..., 11 coordinates are respectively: p ₁(x ₁, 0,0), p ₂(x ₁, y ₂, 0), p ₃(0, y ₂, 0); p ₄(x ₄, y ₄, 0), p ₅(x ₅, y ₄, 0), p ₆(x ₄, y ₆, 0), p ₇(x ₅, y ₆, 0); p ₈(x ₈, y, z ₈), p ₉(x ₉, y, z ₉), p ₁₀(x ₁₀, y, z ₁₀), p ₁₁(x ₁₁, y, z ₁₁);

When camera is during in 1# position, camera photocentre O _cto center of circle o distance, be D ₁, when camera is during in 2# position, camera photocentre O _cto center of circle o distance, be D ₂, and suppose that center of circle coordinate under world coordinate system is o (x, y, z), have:

x ₉+x ₁₁＝2x (1.1)

z ₉＝z ₁₁ (1.2)

${\stackrel{\&RightArrow;}{a}}_1(x_9-x,0,z_9-z+D_1)\&CenterDot;{\stackrel{\&RightArrow;}{b}}_1(x_9-x,0,z_9-z)=0---\left(1.3\right)$

$\frac{f}{\mathrm{dx}}\&CenterDot;(x_9-x)/(z_9-z+D_1)-\frac{f}{\mathrm{dx}}\&CenterDot;(x_{11}-x)/(z_{11}-z+D_1)=C_1---\left(1.4\right)$

${\stackrel{\&OverBar;}{\mathrm{\&beta;}}}_x^1=\frac{f}{\mathrm{dx}}\&CenterDot;\frac{1}{-z+D_1}---\left(1.5\right)$

${\stackrel{\&OverBar;}{\mathrm{\&beta;}}}_y^1=\frac{f}{\mathrm{dy}}\&CenterDot;\frac{1}{-z+D_1}---\left(1.6\right)$

x ₈+x ₁₀＝2x (1.7)

z ₈＝z ₁₀ (1.8)

${\stackrel{\&RightArrow;}{a}}_2(x_8-x,0,z_8-z+D_2)\&CenterDot;{\stackrel{\&RightArrow;}{b}}_2(x_8-x,0,z_8-z)=0---\left(1.9\right)$

$\frac{f}{\mathrm{dx}}\&CenterDot;(x_8-x)/(z_8-z+D_2)-\frac{f}{\mathrm{dx}}\&CenterDot;(x_{10}-x)/(z_{10}-z+D_2)=C_2---\left(1.10\right)$

(x ₈-x) ²+(z ₈-z) ²＝(x ₉-x) ₂+(z ₉-z) ² (1.11)

D ₁-D ₂＝C (1.12)

${\stackrel{\&OverBar;}{\mathrm{\&beta;}}}_x^2=\frac{f}{\mathrm{dx}}\&CenterDot;\frac{1}{-z+D_2}---\left(1.13\right)$

${\stackrel{\&OverBar;}{\mathrm{\&beta;}}}_y^2=\frac{f}{\mathrm{dy}}\&CenterDot;\frac{1}{-z+D_2}---\left(1.14\right)$

Wherein,

for straight line O _cp ₉direction,

for straight line p ₉o direction,

for straight line O _cp ₈direction, for straight line p ₈o direction; C ₁for p ₉point look like p ₁₁distance between some picture; C ₂for p ₈point look like p ₁₀distance between some picture;

for when camera is during in 1# position, along X _cdirection of principal axis is measured ratio, by two standard components, is parallel to X _cthe length on 4 limits of axle and the ratio of the length of its corresponding picture determines,

for when camera is during in 1# position, along Y _cdirection of principal axis is measured ratio, by two standard components, is parallel to Y _cthe length on 4 limits of axle and the ratio of the length of its corresponding picture determines, for when camera is during in 2# position, along X _cdirection of principal axis is measured ratio, by two standard components, is parallel to X _cthe length on 4 limits of axle and the ratio of the length of its corresponding picture determines,

for when camera is during in 2# position, along Y _cdirection of principal axis is measured ratio, by two standard components, is parallel to Y _cthe length on 4 limits of axle and the ratio of the length of its corresponding picture are determined; C is the distance that camera was moved to 2# position from 1# position; F is the focal length of camera; Dx is the size of pixel on directions X, and dy is pixel size in the Y direction;

Simultaneous formula (1.1)～formula (1.14), is the mathematical model of this measuring system, according to the mathematical model of this measuring system, calculates home position coordinate.

Images