CN111397528B - Portable train wheel regular section contour structure optical vision measurement system and method - Google Patents
Portable train wheel regular section contour structure optical vision measurement system and method Download PDFInfo
- Publication number
- CN111397528B CN111397528B CN202010222623.9A CN202010222623A CN111397528B CN 111397528 B CN111397528 B CN 111397528B CN 202010222623 A CN202010222623 A CN 202010222623A CN 111397528 B CN111397528 B CN 111397528B
- Authority
- CN
- China
- Prior art keywords
- point
- axis
- general
- contour
- normal section
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/24—Measuring arrangements characterised by the use of optical techniques for measuring contours or curvatures
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Image Analysis (AREA)
- Length Measuring Devices By Optical Means (AREA)
Abstract
The invention discloses a portable train wheel normal section contour structured light vision measuring system and a method thereof. The invention provides a method for measuring the profile of the right section of a train wheel without being constrained by the position and the posture of a sensor, which has the characteristics of high measurement precision, high operation flexibility, high automation degree, strong robustness and the like.
Description
Technical Field
The invention relates to the field of railway detection and measurement, in particular to a method for measuring a contour line of a regular section of a train wheel.
Background
The train wheels are important components in a railway transportation system, and the wheel wear condition and key geometric parameters of the wheels can be determined by measuring the contour lines of the regular sections of the train wheels, so that the wheels reaching the wear limit can be replaced and overhauled in time, and important guarantee is provided for the safety of railway traffic.
At present, the measurement methods for the contour line of the regular section of a train wheel are mainly divided into contact type and non-contact type. The contact type measuring method adopts a sensor sensitive element to be in direct contact with the surface of the wheel so as to obtain three-dimensional information of the profile of the right section of the wheel. LIJ-4D and MiniProf are typical contact measurement devices. The former measures wheel parameters with a specifically designed vernier caliper, and the latter uses a small magnetic wheel to roll along the profile to complete the measurement. The contact type measuring equipment needs to be operated by professional personnel, the measuring precision is greatly influenced by the surface contact condition, the measuring time is long, the efficiency is low, and the MiniProf needs to take 5min to complete one-time complete profile measurement. In contrast, the non-contact measurement method has high automation degree and high measurement efficiency, and is currently the mainstream method. The line structured light sensor is widely applied to a non-contact type measuring system due to the characteristics of high point cloud obtaining speed and high precision. The problems faced by using a structured light sensor to measure the profile of the right section of a train wheel are as follows: the structured light sensor can only acquire a point cloud at the intersection of the light plane and the wheel surface, and when the laser plane is not aligned with the wheel axis, the contour acquired by the structured light sensor is a general contour rather than a normal section wheel surface. The conventional method for measuring the profile of the right section of the train wheel by adopting the structured light sensor can be divided into on-line measurement and off-line measurement according to application scenes. In an on-line measuring system, the structured light vision sensors are usually installed on two sides of a rail, and when a train passes through the on-line measuring system, the structured light vision sensors can continuously scan the surface of the wheel according to the installed trigger control device, so that the automatic measurement of the wheel profile and the geometric parameters of the wheel profile is completed. The off-line measurement is usually applied to train stopping maintenance or wheel spinning maintenance in a workshop, at the moment, the wheel is in a static state, and in order to measure the contour line of the right section of the wheel under the condition, two schemes of adding constraint to the wheel or adding constraint to a sensor are usually adopted. The wheel drop type laser wheel set detection system is used for pushing wheels into a measurement system in a specific pose after the wheels are disassembled so as to finish the accurate measurement of the size of the wheel set. The portable profile measuring instrument mostly adopts a measuring method of adding constraint to the sensor, for example, the auxiliary alignment device is used for ensuring that the laser plane is aligned with the wheel axis, so that the measurement of the profile of the right section is completed. In comparison, the method has the advantages of low cost, simplicity and convenience in operation and the like, but under the influence of the auxiliary alignment device, the methods have a single measurement visual angle and cannot guarantee that the complete wheel profile is obtained through measurement. The method for measuring the positive section profile of the complete wheel comprises the following steps: marking points are pasted on the surface of the wheel, the periphery of the wheel is scanned by laser, the point cloud is spliced by the marking points to obtain the complete surface of the wheel, and then the profile of the right section is calculated. However, the method for assisting in positioning the pose of the sensor by means of the mark points is affected by the pasting limitation of the mark points, is complex in operation and low in efficiency, and is not suitable for large-scale application scenes.
In summary, in the existing method for measuring the contour of the cross section of the train wheel based on the structured light, the pose of the wheel or the sensor needs to be restricted to achieve the purpose of measuring the contour line of the cross section, and the method has the main disadvantages that the measurement precision is affected by the installation precision of the sensor, the measurement visual angle is limited, the measured contour is incomplete, the operation flexibility is poor, and the like.
Disclosure of Invention
The invention solves the problems: the defects of the prior art are overcome, the portable train wheel normal section profile structure optical vision measuring system and method are provided, the train wheel can be measured in a multi-view normal section profile without constraining the position of a sensor and the position of a wheel, and the complete normal section profile is obtained by a profile splicing method.
In order to achieve the purpose, the technical scheme of the invention is realized as follows:
the invention discloses a portable optical vision measuring system for a train wheel normal section profile structure, which comprises: the system comprises a multi-line structured optical vision sensor, a portable computer and a measurement control device;
the multi-line structured light vision sensor is formed by fixedly mounting a plurality of line lasers and industrial cameras and is used for reconstructing a plurality of general contour lines on the surface of the train wheel; during measurement, a light bar is formed by the intersection line of the laser plane of the multi-line structured light vision sensor and the wheel surface of the train, an industrial camera shoots the light bar to obtain a light bar picture, and the three-dimensional coordinate of a point at the light bar, namely the general contour of the surface of the train, is reconstructed according to the small hole imaging model of the industrial camera and the constraint of the laser plane; the general contour line is the intersection line of the laser plane and the surface of the train wheel at a general position, and is different from the normal section contour of the train wheel, and the normal section contour is the intersection line of the laser plane and the surface of the train wheel when passing through the axis of the train wheel;
the portable computer is connected with the multi-line structured light vision sensor, is responsible for processing light bar images shot by the industrial camera and provides a computing platform for recovering the normal section contour from a general contour, and specifically comprises the following steps: 1) the system is responsible for processing tasks of light bar pictures, including light bar identification in images, light bar center extraction and general wheel surface contour reconstruction; 2) providing a computing platform for recovering the normal section contour from the general contour, wherein the specific computing process comprises the tasks of wheel axis estimation, normal contour rotation projection to a trans-axial plane and normal section contour splicing under multiple visual angles;
the measurement control device is integrated with the multi-line structured light vision sensor and is arranged on the movable carrier, connected with the multi-line structured light vision sensor, used for controlling the triggering of the industrial camera and the multi-line lasers, connected with the portable computer and used for controlling the acquisition and storage of the light bar images;
under the control of the measurement control device, the light strip images on the surfaces of the train wheels at different visual angles are obtained by shooting through the multi-line structured light vision sensor; then inputting the light bar image into a portable computer, reconstructing general outlines of train wheels at different visual angles through a structured light visual model, and projecting the general outlines to an over-axle plane in a rotating manner around wheel axes so as to obtain a normal section outline; and finally, splicing the measurement results under multiple viewing angles to obtain the complete normal section profile of the train wheel.
The method for reconstructing a plurality of general contour lines on the surface of the train wheel by the multi-line structured light vision sensor comprises the following steps:
(1) calibrating the multi-line structured light vision sensor by using the target to obtain internal parameters, distortion coefficients and a plane equation of a laser plane of the industrial camera in a coordinate system of the industrial camera;
(2) moving the multi-line structured light sensor to a certain position during measurement, enabling a laser plane to hit the surface of a train wheel to form a light bar, shooting by an industrial camera to obtain a light bar image, extracting the light bar center of a sub-pixel in the light bar image by using a Hessian matrix-based method, correcting the light bar center coordinate by using the distortion coefficient of the industrial camera obtained by calibration in the step (1), and obtaining the corrected light bar sub-pixel center coordinate;
(3) according to the principle of a structured light visual model: determining a ray passing through the optical center by using the internal reference of the industrial camera obtained by calibration in the step (1) and the corrected light bar sub-pixel central point obtained in the step (2), wherein the intersection point of the ray and the laser plane obtained by calibration in the step (1) is a three-dimensional coordinate corresponding to the light bar center, namely recovering the three-dimensional coordinates of the plurality of laser light bars in the image, and reconstructing a plurality of general contours on the surface of the train wheel.
The invention discloses a portable optical vision measuring method for a train wheel normal section profile structure, which comprises the following steps:
(1) reconstructing a plurality of general contours of the surface of the train wheel by using the multi-line structured light vision sensor;
(2) matching corresponding points among a plurality of general contours, estimating the position of a train wheel axis, and projecting the rotation of the general contours around the train wheel axis to an over-axle plane so as to recover the normal section contour from the general contours; the corresponding point, namely the matching pair, is any point on the general profile, a rotating circle on the surface of the train wheel is obtained by rotating around the wheel axis, and the intersection point of the rotating circle and other general profiles and the point form a pair of corresponding points, namely the matching pair;
(3) and (3) moving the structured light vision sensor to different visual angles of the inner end surface, the outer end surface and the wheel tread surface of the train wheel, repeating the step (1) and the step (2), measuring to obtain the normal section profiles at different visual angles, and splicing the normal section profiles at different visual angles to obtain the complete normal section profile of the train wheel.
In the step (2), the method for matching corresponding points among the general contours and estimating the positions of the wheel axes of the train comprises the following steps:
a. establishing a profile conversion model, and converting the general profile into a normal section profile on an over-axis plane by rotating around an axis under the condition of giving the position of the wheel axis;
b. b, establishing an axis estimation model to estimate the position of the wheel axis required by the step a, wherein the axis estimation model takes a matching pair among the general outlines as input to estimate the position of the wheel axis, and if the axis estimation is accurate, the matching pair on the general outlines is overlapped into a normal section outline on an axis passing plane after passing through the outline conversion model in the step a;
c. establishing an iterative contour optimization algorithm framework, wherein the framework comprises the following two sub-steps: 1) firstly, extracting and matching feature points on a general outline to obtain an initial value of a matching pair; 2) then the frame iteratively utilizes the axis estimation model in the step b, the axis estimation model is substituted with the matching pairs to update the train wheel axis position, then the outline conversion model in the step a is utilized, the general outline is converted into an over-axle plane according to the updated wheel axis position, and the closest point pair is searched to update the matching pair;
d. repeatedly executing substep 2) in step c) until the algorithm converges, namely: the estimated axis position is not changed, or the error of the corresponding point after a certain iteration is smaller than a given threshold value, wherein the error of the corresponding point refers to the distance between the point pairs after the corresponding point rotates to the over-axis plane; and obtaining a matching pair between the general profiles after iterative convergence as a final matching pair, wherein the obtained axis position is the train wheel axis position obtained by final estimation.
In the step (2), the method for recovering the normal section profile from the general profile of the train wheel surface comprises the following steps: matching corresponding point pairs among a plurality of general profiles and estimating the position of the wheel axis, then rotationally projecting the corresponding points to an over-axis plane to obtain projection point pairs, and taking the average value of the projection point pairs as a point on the final normal section profile, thus recovering the normal section profile of the train wheel from the general profiles.
In the step (3), the method for splicing the normal section profiles at different viewing angles to obtain the complete normal section profile of the train wheel specifically comprises the following steps: extracting curvature extreme points as feature points for two frames of normal section profiles measured at adjacent positions, calculating the tangential direction and the normal direction of the feature points, establishing a feature point local coordinate system, intercepting points on equidistant concentric circles on the profiles by taking the feature points as the circle center, converting the obtained intersection points into the feature point local coordinate system to obtain a feature description point row, matching the feature points on different normal section profiles according to the feature description point row, estimating initial Euclidean transformation, then performing nearest point iteration by adopting an ICP (inductively coupled plasma) algorithm to obtain accurate Euclidean transformation of the normal section profiles measured at the adjacent two frames, finally converting all the profiles into a global coordinate system by taking the coordinate system of the first profile as the global coordinate system according to the Euclidean transformation between the adjacent positions to finish multi-view profile splicing, and obtaining a complete normal section profile.
The contour transformation model in the step a is as follows: given the position of the axis of the train wheel, taking a point M on the axis as an origin O, taking the direction along the axis as a Y axis and taking the direction vertical to the axis as an X axis, establishing an axis plane coordinate system O-XY, and for a point s on a general profile, corresponding to a point p on a positive section profile, rotating the point s around an axis to a passing axis plane to obtain the coordinate of the point p in the axis plane coordinate system:
the general contour transformation model is a contour transformation model, and transformation from a general contour to a normal section contour is described, namely, an intersection point obtained by rotating a point on the general contour to a passing axis plane around a wheel is a point on the normal section contour.
The axis estimation model in the step b is as follows: inputting a set S of matching pairs on the general profile, and estimating the position of an axis according to the least square principle, wherein the axis (P, M) is estimated by the following mathematical model:
wherein (P, M) is the position parameter of the axle line of the train wheel, P is a unit vector and represents the direction of the axle line of the wheel, M is a point on the axle, (P, M) is the estimated value of the position parameter of the axle line of the train wheel under the condition that the objective function is minimum, S is a matching pair set, and (S, d) is one of the matching pairs.
In the contour optimization algorithm framework of the step c: the specific steps of extracting and matching the feature points on the general outline to obtain the initial values of the matching pairs are as follows:
(1) calculating curvature values of all points on a general contour point by point to obtain a curvature sequence, defining a sign of the curvature values according to the bending direction in order to facilitate subsequent matching, and defining that clockwise bending is a positive value and anticlockwise bending is a negative value;
(2) extracting extreme points of the curvature sequence, firstly performing Gaussian smoothing on the curvature sequence, calculating a derivative, performing Taylor expansion on points with curvatures larger than a set threshold value in the sequence, and taking the points with the derivative being zero as local extreme points in order to overcome the influence of noise and improve the stability of feature extraction; in order to avoid extracting excessive local mechanism points at close positions, a maximum value suppression measure is adopted to ensure that only the maximum value point with the maximum curvature is reserved as a final characteristic point within a certain range;
(3) and (3) matching the feature points, respectively calculating the relative sizes of curvature differences of the feature points between the general contours, taking the points with the curvature differences smaller than a threshold value as candidate matching pairs, and clustering the candidate matching pairs by using a moving average clustering algorithm according to the characteristic that the connecting lines of the correct matching pairs have similar spatial orientation so as to eliminate wrong matching pairs, thus obtaining the initial values of the matching pairs between the general contours.
In the contour optimization algorithm framework of the step c: the specific steps of searching the nearest point pair to update the matching pair are as follows: searching the nearest point on the converted contour to obtain a nearest point pair, and updating the existing matching pair according to the point on the general contour corresponding to the nearest point pair; in order to improve the matching precision, a method of performing linear interpolation near the closest point is adopted, namely, a point of the intersection of the normal direction of one point of the matching pair and the tangential direction of the other point of the matching pair is taken as the final closest point.
Compared with the prior art, the invention has the advantages that:
(1) the method adopts the multi-line structured light sensor to simultaneously acquire a plurality of general profiles on the surface of the train wheel, and estimates the position of the wheel axis by matching corresponding points on the general profiles, thereby converting the general profiles into the normal section profiles;
(2) analysis is carried out on the principle, and the invention can be used for measuring the profile of the normal section as long as the extreme point of the curvature of the profile can be extracted, so the invention has robustness to the shape of the measured profile;
(3) the invention can measure the high-precision complete right section contour, after the right section contour is measured at multiple visual angles, the ICP algorithm is adopted to splice the right section contours at different visual angles to obtain the complete right section contour, the essence is the splicing problem of a two-dimensional curve, mark points do not need to be pasted around a measured piece, and the high-precision measurement requirement can be met, and experiments show that the precision of the invention for measuring the right section contour of the train wheel can reach 0.068 mm.
Drawings
FIG. 1 is a schematic diagram of a portable optical vision measuring system for a train wheel normal cross-section profile structure according to the present invention;
FIG. 2 is a flow chart of the general implementation of a portable optical vision measurement method for the cross-section profile structure of a train wheel according to the present invention;
FIG. 3 is a diagram of a train wheel to be measured and a multi-line structured optical vision sensor;
FIG. 4 is a process of acquiring a general contour of a wheel surface by the multi-line structured light sensor;
FIG. 5 is a diagram of feature extraction and matching on a general outline;
FIG. 6 is a graph of the exact match of points on a generic contour after iterative optimization;
fig. 7 is a normal cross-sectional profile measured at multiple viewing angles and a complete normal cross-sectional profile obtained by stitching the normal cross-sectional profiles.
Detailed Description
The present invention will be described in further detail with reference to the following drawings and specific embodiments.
Fig. 1 and fig. 2 are general implementation flow charts of a portable train wheel normal section profile structured light vision measurement system and method of the invention, respectively. The train wheel normal section contour structure optical vision measuring system consists of a multi-line structure optical vision sensor, a portable computer and measurement control software, can be freely moved to different visual angles, and reconstructs the general contour of a train wheel under multiple visual angles. The method for measuring the wheel normal section contour structure optical vision recovers the normal section contour of the train wheel from the general contour of the train wheel surface, and obtains a complete normal section contour by splicing the normal section contours at different viewing angles.
The general contour reconstruction steps are: (1) calibrating internal parameters and distortion coefficients of the industrial camera and an equation of a light plane of the multi-line structured light under the industrial camera; (2) the handheld multi-line structured light sensor surrounds the wheel tread of the train and collects light bar images of the laser plane on the surface of the wheel from multiple angles; (3) extracting the sub-pixel center of the optical strip in the image obtained in the step (2), and removing the lens distortion of the industrial camera by using the distortion removal coefficient calibrated in the step (1) to obtain the image coordinate of the optical strip center after distortion removal; (4) according to the structured light vision measurement model, by utilizing the internal reference of the industrial camera obtained by calibration in the step (1), a ray passing through the optical center can be determined according to the distorted light strip sub-pixel central point obtained in the step (3), and the intersection point of the ray and the laser plane obtained by calibration in the step (1) is the three-dimensional coordinate corresponding to the light strip center, so that the three-dimensional coordinates of the middle points of all light strips in the image can be reconstructed, and a plurality of general contours on the surface of the train wheel can be reconstructed.
A portable train wheel normal section contour structure optical vision measuring method. The method takes a plurality of general profiles on the surface of the train wheel as data input, estimates the position of the wheel axis through the general profiles, and rotationally projects the general profiles onto an over-axle plane according to the axis position to restore the normal section profile. And moving the sensors, respectively measuring the normal section profiles at different viewing angles, and finally splicing the normal section profiles at multiple viewing angles to recover the complete normal section profile of the train wheel.
In order to recover the normal section contour from the general contour, the method of the present invention mainly includes two models, a contour transformation model and an axis estimation model, and a contour optimization framework, and the following describes the models and algorithms in detail.
The effect of the contour transformation model is to project a general rotation onto the transaxial plane. Given the position of the wheel axis, taking a point M on the axis as an origin O, taking the direction along the axis as a Y axis, taking the direction vertical to the axis as an X axis, establishing an axis plane coordinate system O-XY, and for a point s on the general profile, corresponding to a point p on the positive cross-sectional profile, rotating s around the axis to a point p on the cross-axis plane to obtain the coordinates of the point p in the axis plane coordinate system:
where (P, M) represents a parameter describing the position of the wheel axis, P is a unit vector representing the direction of the wheel axis, and M is a point on the axis. The contour transformation model T (| P, M) describes the transformation from the general contour to the normal section contour, namely, the point on the general contour rotates around the wheel, and the intersection point obtained by intersecting the trans-axis plane is the point on the normal section contour.
To introduce the axis estimation model, the definition of the corresponding points (matching pairs) is given first. The tread of the train wheel is a rotating curved surface, one point on the tread rotates around a shaft to form a rotating circle, and points on the same rotating circle on different general contours are called corresponding points (matching pairs). Assuming (s, d) is a pair of matching pairs, the following equation should be satisfied:
T(s|P,M)＝T(d|P,M)
wherein T (| P, M) is a contour transformation model, P, M is a train wheel axis parameter, P is an axis direction, and M is a point on the axis.
The axis estimation model is used for estimating train wheel axis position parameters (P, M) required by the profile conversion model. The axis estimation model is built based on the fact that: the corresponding points on the general profile are rotated around the axis to the transaxial plane, and the projection of the two points is superposed on the normal section profile. The position of the axes is thus estimated according to the principle of least squares, the axes (P, M) being estimated by the following mathematical model:
wherein, S is a corresponding point set on the general outline, (S, d) is a pair of matching pairs, and (P, M) is an estimated value of the train wheel axis position parameter under the condition that the objective function is minimum. In order to quickly solve the nonlinear least square problem in the axis estimation model, an objective function is approximated by two linear least square problems, and P and M are respectively solved, specifically:
p is solved with the following linear least squares problem:
st.||P||＝1
where N is the number of matching pairs,(s)i,di) (i-1, 2 … N) is a pair of matching pairs.
M is solved with the following least squares problem:
wherein
Is an anti-symmetric matrix of the axis direction P, N is the number of matching pairs,(s)i,di) Is a matched pair of one pair.
The axis estimation model needs to know the matching pairs among the general contours, but the matching pairs are not known, so the invention constructs the following contour optimization algorithm framework by using an iterative optimization strategy, and the recovery of the normal section contour from the general contour comprises the following steps:
a. and extracting curvature extreme points on the general contour as feature points, and matching the feature points to obtain an initial value of a matching pair. Firstly, calculating curvature values of all points on a general contour point by point to obtain a curvature sequence, defining the curvature values as symbols according to the bending direction in order to be beneficial to subsequent matching, and defining that clockwise bending is a positive value and anticlockwise bending is a negative value. Then, extracting an extreme point of the curvature sequence, firstly performing Gaussian smoothing on the curvature sequence, calculating a derivative, performing Taylor expansion on a point with the curvature being larger than a set threshold value in the sequence, and taking a point with the derivative being zero as a local extreme point in order to overcome the influence of noise and improve the stability of feature extraction; in order to avoid extracting excessive local mechanism points at close positions, a maximum value suppression measure is adopted, and only the maximum value point with the maximum curvature is guaranteed to be reserved as a final characteristic point within a certain range. And finally, matching the feature points to obtain initial values of the matched pairs, respectively calculating the relative sizes of the curvature difference of the feature points among the general contours, taking the points with the curvature difference smaller than a threshold value as candidate matched pairs, and clustering the candidate matched pairs by utilizing a moving average clustering algorithm according to the characteristic that the connecting lines of the correct matched pairs have similar spatial orientation so as to eliminate wrong matched pairs, thus obtaining the initial values of the matched pairs among the general contours.
b. According to the following steps, the initial value of the matching pair obtained in the step a is used as input, the matching pair is optimized in an iterative mode, and the normal section profile is obtained finally: 1) substituting the matching pairs into an axis estimation model to estimate the axis position; 2) the estimated axis position is brought into a contour transformation model, so that a general contour can be rotationally projected to an axial plane to obtain a transformed contour, and the distance between matched pairs on the transformed contour is calculated to serve as a corresponding error; 3) searching a nearest point pair on the converted contour and updating the existing matching pair, and in order to improve the searching precision of the nearest point pair, taking the intersection point of a normal line of one point in the nearest point pair and a tangent line of another point as the nearest point after interpolation, and updating the existing matching pair by using the point pair on the general contour corresponding to each group of nearest points; 4) repeatedly executing the steps 1), 2) and 3) until the position of the wheel axis obtained by estimation is not changed any more or the error of the corresponding point is smaller than a given threshold value; and rotating the general profile to the shaft passing plane according to the axis position of the train wheel obtained by the last iteration, wherein the general profile is the profile of the right section of the train wheel.
c. And measuring and splicing the multi-view normal section profile. Moving the sensor to other visual angles and positions, repeating the steps a and b, completing the steps of extracting and matching characteristic points on the general profile at other visual angles, recovering the profile of the normal section and the like, obtaining the profile of the normal section at multiple visual angles, and splicing according to the following steps to obtain the complete profile of the normal section: firstly, calculating a transformation matrix between profiles of a measured normal section under two adjacent visual angles, extracting curvature extreme points on the profiles as feature points, calculating description point rows on equidistant concentric circles of the feature points, matching the feature points on different profiles to calculate initial Euclidean transformation between the two profiles, and calculating accurate Euclidean transformation between the two profiles by adopting a closest point iterative algorithm (ICP); and then, taking the measurement result of the first frame as a reference coordinate system, transforming all the normal section profiles to the reference coordinate system according to Euclidean transformation between two adjacent frames, and obtaining the complete normal section profiles by adopting the direction of nearest point-to-mean fusion for the overlapped parts of the profiles.
The following description will be given with reference to specific examples.
The method specifically comprises the following steps:
step 11: referring to fig. 3, an industrial camera and a plurality of line lasers are mounted on a freely movable carrier to form a multi-line structured optical vision sensor, and parameters of the sensor are calibrated. The installation distance of the laser is about 20mm, the included angle between the light plane and the main shaft of the industrial camera is about 60 degrees, the resolution of the industrial camera is 2448 x 2048 pixels, the power of the laser is 30mW, the diameter of the wheel to be measured is 1040mm, and the working distance of the multi-line structured light sensor is 400 mm.
Calibrating the internal reference and distortion coefficient of the industrial camera by adopting a Zhang Zhengyou camera calibration method, wherein the calibration result is as follows:
the equation of the light plane under the coordinate system of the industrial camera can be calibrated by using the structured light plane calibration technology proposed by Sunsylvania et al [ J.Sun, G.Zhang, Q.Liu, and Z.Yang, "Universal method for calibrating structured-light vision sensor on the spot," J.Mech.Eng., vol.45, No.03, pp.174-177,2009 ], and the calibration result is:
step 12: as shown in fig. 4, the sensor is moved to a certain viewing angle, the light strip of the light plane on the wheel surface is photographed to obtain a light strip image, the center of the light strip is extracted, then the three-dimensional coordinates of the light strip are restored according to the structured light vision model, and the general contour of the wheel surface is reconstructed.
And extracting the sub-pixel coordinates of the center of the light bar by using a Hessian-based method and carrying out distortion correction. Because camera lenses generally have different degrees of distortion (mainly radial distortion and tangential distortion), the ideal pinhole imaging model is not provided. Therefore, the extracted light bar center sub-pixel image coordinates are substituted into the Brown distortion model, and more accurate light bar center image coordinates are obtained. The Brown distortion model is shown below:
wherein:
normalizing image coordinates for extracting distortion characteristic points; (x, y) are the undistorted normalized image coordinates; (k)1,k2,p1,p2) Is a distortion parameter.
And substituting the corrected light bar sub-pixel coordinates into the structured light vision model to reconstruct the general contour of the wheel surface. Utilizing the internal reference K of the industrial camera calibrated in the step 11 and the laser light plane equation { W }1,W2,W3The three-dimensional points on the general outline can be reconstructed by the following model:
wherein X is the three-dimensional point coordinate on the general outline, X is the image coordinate of the light bar center, and s is the scale factor of the industrial camera imaging model. Using the above model for each light plane, a plurality of general profiles { l } can be reconstructed1,l2,l3}。
Step 13: as shown in fig. 5, extracting and matching feature points on the general contour to obtain an initial value S of the matching pair0. Calculating the curvature of each point on the general contour point by point to obtain a curvature sequence, and according to the curve bending directionThe points that curve clockwise take a positive curvature and the points that curve counterclockwise take a negative curvature. And then, smoothing the curvature sequence by using a Gaussian core, and taking a point with the curvature larger than a threshold value and the curvature gradient of zero as a characteristic point. And calculating the curvature difference among the characteristic points on different contours, wherein the points with the curvature difference smaller than a threshold value are used as candidate matching pairs. Connecting lines between correct matching pairs have similar spatial directions in space, so that the wrong matching pairs are eliminated according to the moving mean clustering to obtain initial matching pairs S0。
Step 14: initial matching pair S0Substituting the contour optimization algorithm into a contour optimization algorithm frame, iteratively optimizing and matching the positions of the pair and the wheel axis according to the following steps, and finally obtaining the normal section contour:
step1 inputs the initial matching pair S0Let the iteration count t equal to 0 and set the corresponding point error threshold to epsilonmax；
Step2 brings the current matching pair StEstimating the position of the axis (P, M) from the axis estimation modelt；
Step3 is based on the current axis position (P, M)tGeneral profile, { l }1,l2,l3Rotating to an over-axis plane by using a contour conversion model to obtain a converted contour
And calculating the distance between corresponding points on the converted contour, namely corresponding point errors. Judging whether the iteration reaches a termination condition, namely the axis position estimated in the Step2 is not changed or the error of the corresponding point is less than a given threshold value epsilonmaxIf yes, go to step5, otherwise go to step 4;
step4, searching the nearest points on the converted contour pairwise to obtain a nearest point set
According to the nearest point set CtSelecting corresponding points on the general contour to update the corresponding point set
Let iteration count t be t +1, go to Step 2;
step5, the iteration process is ended, and the transformed contour { L ] obtained in the last iteration is used1,L2,L3And f, fusing to obtain the final normal section outline L.
Fig. 6 is a matching pair on the resulting general profile and thus a right cross-sectional profile at that viewing angle can be obtained.
Step 15: as shown in fig. 7, the sensor is moved to other viewing angles and positions, steps 12, 13 and 14 are repeated, the steps of obtaining general profiles at other viewing angles, matching initial feature points, recovering normal section profiles and the like are completed, normal section profiles at a plurality of different viewing angles are obtained, and the normal section profiles at multiple viewing angles are spliced to obtain a complete normal section profile according to the following steps:
a. calculating the curvature of each point by point to obtain a curvature sequence for each normal section contour, and extracting an extreme point of the curvature sequence as a characteristic point p0By the feature point p0The position coordinate of the point is used as an original point o, a characteristic point local coordinate system o-xy is established by taking the normal direction and the tangential direction as an x axis and a y axis, k concentric circles with equal intervals are made by taking the o as the circle center and are intersected with the profile of the right cross section to respectively obtain concentric circle point rows, the coordinate of the intersection point is converted to be under the characteristic point local coordinate system o-xy and is sequentially arranged along the x axis direction
The point sequence is the descriptor point sequence of the feature point.
b. And matching feature points on the profiles of the normal sections of two adjacent frames according to the descriptor point column, calculating an initial Euclidean transformation matrix, completing rough splicing of the profiles, and then adopting an ICP (inductively coupled plasma) algorithm to iterate a nearest point, and calculating accurate Euclidean transformation between adjacent frames. And then, taking the measurement result of the first frame as a reference coordinate system, transforming all the normal section profiles to the reference coordinate system according to Euclidean transformation between two adjacent frames, and obtaining the complete normal section profiles by adopting the direction of nearest point-to-mean fusion for the overlapped parts of the profiles.
In summary, the present invention provides a portable train wheel normal section profile measuring system and method, including: the method comprises the following contents of general contour acquisition, a contour transformation model, an axis estimation model, an iterative contour optimization algorithm, splicing of normal section contours under multiple viewing angles and the like. Experimental results show that the precision of the method for measuring the profile of the front section of the train wheel is 0.068mm, and the method has the advantages of strong flexibility, high precision, strong robustness and the like. Compared with the method for measuring the profile of the positive section of the train wheel by using the MiniProf, the method can greatly shorten the measurement time, wherein the single measurement time of the MiniProf is 5min, and the method can finish the measurement within 1 min. Compared with the traditional caliper LIJ-4D (with the precision of 0.1mm), the invention has the advantages of no need of manual alignment and higher precision.
While the invention has been described with respect to specific preferred embodiments, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention.
Claims (8)
1. A portable train wheel normal section profile structure optical vision measuring system is characterized by comprising: the system comprises a multi-line structured optical vision sensor, a portable computer and a measurement control device;
the multi-line structured light vision sensor is formed by fixedly mounting a plurality of line lasers and industrial cameras and is used for reconstructing a plurality of general contour lines on the surface of the train wheel; during measurement, a light bar is formed by the intersection line of the laser plane of the multi-line structured light vision sensor and the wheel surface of the train, an industrial camera shoots the light bar to obtain a light bar picture, and the three-dimensional coordinate of a point at the light bar, namely the general contour of the surface of the train, is reconstructed according to the small hole imaging model of the industrial camera and the constraint of the laser plane; the general contour line is the intersection line of the laser plane and the surface of the train wheel at a general position, and is different from the normal section contour of the train wheel, and the normal section contour is the intersection line of the laser plane and the surface of the train wheel when passing through the axis of the train wheel;
the portable computer is connected with the multi-line structured light vision sensor, is responsible for processing light bar images shot by the industrial camera and provides a computing platform for recovering the normal section contour from a general contour, and specifically comprises the following steps: 1) the system is responsible for processing tasks of light bar pictures, including light bar identification in images, light bar center extraction and general wheel surface contour reconstruction; 2) providing a computing platform for recovering the normal section contour from the general contour, wherein the computing process comprises the tasks of wheel axis estimation, normal contour rotation projection to a trans-axial plane and normal section contour splicing under multiple visual angles;
the measurement control device is integrated with the multi-line structured light vision sensor and is arranged on the movable carrier, connected with the multi-line structured light vision sensor, used for controlling the triggering of the industrial camera and the multi-line lasers, connected with the portable computer and used for controlling the acquisition and storage of the light bar images;
under the control of the measurement control device, the light strip images on the surfaces of the train wheels at different visual angles are obtained by shooting through the multi-line structured light vision sensor; then inputting the light bar image into a portable computer, reconstructing general outlines of train wheels at different visual angles through a structured light visual model, and projecting the general outlines to an over-axle plane in a rotating manner around wheel axes so as to obtain a normal section outline; finally, splicing measurement results under multiple viewing angles to obtain a complete normal section profile of the train wheel;
in the multi-line structured light vision sensor, the reconstruction of the three-dimensional coordinates of the points at the light bars, namely the general contour of the train surface, comprises the following processes:
(1) calibrating the multi-line structured light vision sensor by using the target to obtain internal parameters, distortion coefficients and a plane equation of a laser plane of the industrial camera in a coordinate system of the industrial camera;
(2) moving the multi-line structured light sensor to a certain position during measurement, enabling a laser plane to hit the surface of a train wheel to form a light bar, shooting by an industrial camera to obtain a light bar image, extracting the light bar center of a sub-pixel in the light bar image by using a Hessian matrix-based method, correcting the light bar center coordinate by using the distortion coefficient of the industrial camera obtained by calibration in the step (1), and obtaining the corrected light bar sub-pixel center coordinate;
(3) according to the principle of a structured light visual model: determining a ray passing through the optical center by using the internal reference of the industrial camera obtained by calibration in the step (1) and the corrected light bar sub-pixel central point obtained in the step (2), wherein the intersection point of the ray and the laser plane obtained by calibration in the step (1) is a three-dimensional coordinate corresponding to the light bar center, namely recovering the three-dimensional coordinate of the light bars of the lasers in the image, and reconstructing a plurality of general contours on the surface of the train wheel.
2. A portable train wheel normal section contour structure optical vision measuring method is characterized by comprising the following steps:
(1) reconstructing a plurality of general contours of the surface of the train wheel by using the multi-line structured light vision sensor;
(2) matching corresponding points among a plurality of general contours, estimating the position of a train wheel axis, and projecting the rotation of the general contours around the train wheel axis to an over-axle plane so as to recover the normal section contour from the general contours; the corresponding point, namely the matching pair, is any point on the general profile, a rotating circle on the surface of the train wheel is obtained by rotating around the wheel axis, and the intersection point of the rotating circle and other general profiles and the point form a pair of corresponding points, namely the matching pair;
(3) moving the structured light vision sensor to different visual angles of the inner end surface, the outer end surface and the wheel tread surface of the train wheel, repeating the step (1) and the step (2), measuring to obtain the normal section profiles at different visual angles, and splicing the normal section profiles at different visual angles to obtain the complete normal section profile of the train wheel;
in the step (2), the method for matching corresponding points among the general contours and estimating the positions of the wheel axes of the train comprises the following steps:
a. establishing a profile conversion model, and converting the general profile into a normal section profile on an over-axis plane by rotating around an axis under the condition of giving the position of the wheel axis;
b. b, establishing an axis estimation model to estimate the position of the wheel axis required by the step a, wherein the axis estimation model takes a matching pair among the general outlines as input to estimate the position of the wheel axis, and if the axis estimation is accurate, the matching pair on the general outlines is overlapped into a normal section outline on an axis passing plane after passing through the outline conversion model in the step a;
c. establishing an iterative contour optimization algorithm framework, wherein the framework comprises the following two sub-steps: 1) firstly, extracting and matching feature points on a general outline to obtain an initial value of a matching pair; 2) then the frame iteratively utilizes the axis estimation model in the step b, the axis estimation model is substituted with the matching pairs to update the train wheel axis position, then the outline conversion model in the step a is utilized, the general outline is converted into an over-axle plane according to the updated wheel axis position, and the closest point pair is searched to update the matching pair;
d. and (4) repeatedly executing substep 2) in the step c until the algorithm converges, namely: the estimated axis position is not changed, or the error of the corresponding point after a certain iteration is smaller than a given threshold value, wherein the error of the corresponding point refers to the distance between the point pairs after the corresponding point rotates to the over-axis plane; and obtaining a matching pair between the general profiles after iterative convergence as a final matching pair, wherein the obtained axis position is the train wheel axis position obtained by final estimation.
3. The portable optical vision measurement method for the train wheel normal section profile structure according to claim 2, wherein the method comprises the following steps: in the step (2), the method for recovering the normal section profile from the general profile of the train wheel surface comprises the following steps: matching corresponding point pairs among a plurality of general profiles and estimating the position of the wheel axis, then rotationally projecting the corresponding points to an over-axis plane to obtain projection point pairs, and taking the average value of the projection point pairs as a point on the final normal section profile, thus recovering the normal section profile of the train wheel from the general profiles.
4. The portable optical vision measurement method for the train wheel normal section profile structure according to claim 2, wherein the method comprises the following steps: in the step (3), the method for splicing the normal section profiles at different viewing angles to obtain the complete normal section profile of the train wheel specifically comprises the following steps: extracting curvature extreme points as feature points for two frames of normal section profiles measured at adjacent positions, calculating the tangential direction and the normal direction of the feature points, establishing a feature point local coordinate system, intercepting points on equidistant concentric circles on the profiles by taking the feature points as the circle center, converting the obtained intersection points into the feature point local coordinate system to obtain a feature description point row, matching the feature points on different normal section profiles according to the feature description point row, estimating initial Euclidean transformation, then performing nearest point iteration by adopting an ICP (inductively coupled plasma) algorithm to obtain accurate Euclidean transformation of the normal section profiles measured at the adjacent two frames, finally converting all the profiles into a global coordinate system by taking the coordinate system of the first profile as the global coordinate system according to the Euclidean transformation between the adjacent positions to finish multi-view profile splicing, and obtaining a complete normal section profile.
5. The portable optical vision measurement method for the train wheel normal section profile structure according to claim 2, wherein the method comprises the following steps: the contour transformation model in the step a is as follows: given the position of the axis of the train wheel, taking a point M on the axis as an origin O, taking the direction along the axis as a Y axis and taking the direction vertical to the axis as an X axis, establishing an axis plane coordinate system O-XY, and for a point s on a general profile, corresponding to a point p on a positive section profile, rotating the point s around an axis to a passing axis plane to obtain the coordinate of the point p in the axis plane coordinate system:
the general contour transformation model is a contour transformation model, and transformation from a general contour to a normal section contour is described, namely, an intersection point obtained by rotating a point on the general contour to a passing axis plane around a wheel is a point on the normal section contour.
6. The portable optical vision measurement method for the train wheel normal section profile structure according to claim 2, wherein the method comprises the following steps: the axis estimation model in the step b is as follows: inputting a set S of matching pairs on the general profile, and estimating the position of an axis according to the least square principle, wherein the axis (P, M) is estimated by the following mathematical model:
wherein (P, M) is the position parameter of the axle line of the train wheel, P is a unit vector and represents the direction of the axle line of the wheel, M is a point on the axle, (P, M) is the estimated value of the position parameter of the axle line of the train wheel under the condition that the objective function is minimum, S is a matching pair set, and (S, d) is one of the matching pairs.
7. The portable optical vision measurement method for the train wheel normal section profile structure according to claim 2, wherein the method comprises the following steps: in the contour optimization algorithm framework of the step c: the specific steps of extracting and matching the feature points on the general outline to obtain the initial values of the matching pairs are as follows:
(1) calculating curvature values of all points on a general contour point by point to obtain a curvature sequence, defining a sign of the curvature values according to the bending direction in order to facilitate subsequent matching, and defining that clockwise bending is a positive value and anticlockwise bending is a negative value;
(2) extracting extreme points of the curvature sequence, firstly performing Gaussian smoothing on the curvature sequence, calculating a derivative, performing Taylor expansion on points with curvatures larger than a set threshold value in the sequence, and taking the points with the derivative being zero as local extreme points in order to overcome the influence of noise and improve the stability of feature extraction; in order to avoid extracting excessive local mechanism points at close positions, a maximum value suppression measure is adopted to ensure that only the maximum value point with the maximum curvature is reserved as a final characteristic point within a certain range;
(3) and (3) matching the feature points, respectively calculating the relative sizes of curvature differences of the feature points between the general contours, taking the points with the curvature differences smaller than a threshold value as candidate matching pairs, and clustering the candidate matching pairs by using a moving average clustering algorithm according to the characteristic that the connecting lines of the correct matching pairs have similar spatial orientation so as to eliminate wrong matching pairs, thus obtaining the initial values of the matching pairs between the general contours.
8. The portable optical vision measurement method for the train wheel normal section profile structure according to claim 2, wherein the method comprises the following steps: in the contour optimization algorithm framework of the step c: the specific steps of searching the nearest point pair to update the matching pair are as follows: searching the nearest point on the converted contour to obtain a nearest point pair, and updating the existing matching pair according to the point on the general contour corresponding to the nearest point pair; in order to improve the matching precision, a method of performing linear interpolation near the closest point is adopted, namely, a point of intersection of the normal direction of one point in the closest point pair and the tangential direction of the other point in the closest point pair is used as the final closest point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010222623.9A CN111397528B (en) | 2020-03-26 | 2020-03-26 | Portable train wheel regular section contour structure optical vision measurement system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010222623.9A CN111397528B (en) | 2020-03-26 | 2020-03-26 | Portable train wheel regular section contour structure optical vision measurement system and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111397528A CN111397528A (en) | 2020-07-10 |
| CN111397528B true CN111397528B (en) | 2021-03-09 |
Family
ID=71436632
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010222623.9A Active CN111397528B (en) | 2020-03-26 | 2020-03-26 | Portable train wheel regular section contour structure optical vision measurement system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111397528B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114234808B (en) * | 2021-12-17 | 2022-10-28 | 湖南大学 | Size measuring method and device for deformation area of rotary magnetic pulse crimping part |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2004309344A (en) * | 2003-04-08 | 2004-11-04 | Act Denshi Kk | Cross-sectional shape measuring method, reference tool used therefor, cross-sectional shape measuring device with supporter, and supporter for cross-sectional shape measuring device |
| CN101571373A (en) * | 2009-06-11 | 2009-11-04 | 天津大学 | Method for measuring geometric parameters of profile steel based on multi-linear structured light vision sensors |
| CN102901457A (en) * | 2012-10-18 | 2013-01-30 | 北京航空航天大学 | Dynamic measurement method and system for train wheel diameter |
| CN104424630A (en) * | 2013-08-20 | 2015-03-18 | 华为技术有限公司 | Three-dimension reconstruction method and device, and mobile terminal |
| CN105241397A (en) * | 2015-06-29 | 2016-01-13 | 北航温州研究院 | Real-time measuring splicing method and device based on structured light |
| CN108955576A (en) * | 2018-10-31 | 2018-12-07 | 湖南东映碳材料科技有限公司 | Multi-line structured light self-calibrating method and system in profile of steel rail dynamic detection |
-
2020
- 2020-03-26 CN CN202010222623.9A patent/CN111397528B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2004309344A (en) * | 2003-04-08 | 2004-11-04 | Act Denshi Kk | Cross-sectional shape measuring method, reference tool used therefor, cross-sectional shape measuring device with supporter, and supporter for cross-sectional shape measuring device |
| CN101571373A (en) * | 2009-06-11 | 2009-11-04 | 天津大学 | Method for measuring geometric parameters of profile steel based on multi-linear structured light vision sensors |
| CN102901457A (en) * | 2012-10-18 | 2013-01-30 | 北京航空航天大学 | Dynamic measurement method and system for train wheel diameter |
| CN104424630A (en) * | 2013-08-20 | 2015-03-18 | 华为技术有限公司 | Three-dimension reconstruction method and device, and mobile terminal |
| CN105241397A (en) * | 2015-06-29 | 2016-01-13 | 北航温州研究院 | Real-time measuring splicing method and device based on structured light |
| CN108955576A (en) * | 2018-10-31 | 2018-12-07 | 湖南东映碳材料科技有限公司 | Multi-line structured light self-calibrating method and system in profile of steel rail dynamic detection |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111397528A (en) | 2020-07-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106558080B (en) | Monocular camera external parameter online calibration method | |
| CN103499302B (en) | The camshaft diameter dimension On-line Measuring Method of structure based light Vision imaging system | |
| CN109827502B (en) | High-precision calibration method for line-structured light vision sensor for calibration point image compensation | |
| CN112985293B (en) | Binocular vision measurement system and measurement method for single-camera double-spherical mirror image | |
| CN112734841B (en) | Method for realizing positioning by using wheel type odometer-IMU and monocular camera | |
| CN105698699A (en) | A binocular visual sense measurement method based on time rotating shaft constraint | |
| CN108955576A (en) | Multi-line structured light self-calibrating method and system in profile of steel rail dynamic detection | |
| JP2011509398A (en) | Method for processing a three-dimensional image of a tire surface so that it can be used to inspect the tire surface | |
| CN102750697A (en) | Parameter calibration method and device | |
| JPH10253322A (en) | Method and apparatus for designating position of object in space | |
| CN106996748A (en) | Wheel diameter measuring method based on binocular vision | |
| CN102810205A (en) | Method for calibrating camera shooting or photographing device | |
| CN113129430B (en) | Underwater three-dimensional reconstruction method based on binocular structured light | |
| JP2014500484A (en) | Method for pre-processing 3D image of tire surface using continuous B-spline deformation | |
| CN112815843B (en) | On-line monitoring method for printing deviation of workpiece surface in 3D printing process | |
| CN114705122A (en) | Large-field stereoscopic vision calibration method | |
| CN111189416A (en) | Structural light 360-degree three-dimensional surface shape measuring method based on characteristic phase constraint | |
| CN111397528B (en) | Portable train wheel regular section contour structure optical vision measurement system and method | |
| CN114396894B (en) | Pipeline straightness detection device and method based on laser ranging and monocular vision | |
| CN114170284B (en) | Multi-view point cloud registration method based on active landmark point projection assistance | |
| CN116625258A (en) | Chain spacing measuring system and chain spacing measuring method | |
| CN116563377A (en) | Mars rock measurement method based on hemispherical projection model | |
| CN111998823B (en) | Target ranging method based on binocular different-light-source ranging device | |
| CN113920201A (en) | Polar line geometric constraint fisheye camera calibration method | |
| JP2003532061A (en) | Method for measuring a three-dimensional object or a set of objects |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |