A train wheel set geometry parameter on-line detection method and system

Abstract

The present application provides a train wheel set geometry parameter online detection method and system, which belongs to the technical field of train wheel set detection. The steps of the present application are as follows: based on the internal parameters and external parameters of the camera of the planar checkerboard-concentric circle complementary target static calibration measurement system, the light plane equation is determined, and the light plane calibration of the camera is completed; the laser transmission sensor acquires the incoming train signal, the upper computer issues an instruction to turn on the light supplementing device, and the camera calibrated by the light plane is used to collect the train wheel set image; the camera posture error is compensated in real time based on the IMU gyroscope sensor and the improved PID negative feedback algorithm; the collected train wheel set image is preprocessed, and the laser light strip center line is extracted online by using the Steger algorithm; and the geometry parameters of the inner diameter, the outer diameter, the rim width and the flange thickness of the train wheel set are calculated online by using the light plane equation. The present application has the advantages of high calibration precision, fast real-time detection speed and suitability for complex working environments.

Description

A method and system for online detection of train wheelset geometric parameters Technical Field

[0001] This invention relates to the technical field of train wheelset detection, and in particular to an online detection method and system for train wheelset geometric parameters, which integrates photoelectric detection technology, image processing technology, and automatic control technology to perform online measurement of the geometric parameters of moving train wheelsets. Background Technology

[0002] With the significant increase in the transport speed, density, and load capacity of my country's rail transit trains, the safe operation of trains faces greater challenges. Wheelsets are the most important moving and load-bearing components of a train, operating in a complex and harsh environment, facing issues such as working with existing damage, frequent operation, and long service life. Wheelset geometric parameters are crucial indicators for assessing wheelset health, but routine periodic inspections have many shortcomings: manual use of tools such as wheel gauges and fourth-order inspection devices is subject to subjective judgment, resulting in large errors and low efficiency; offline wheel inspections are cumbersome, consuming significant manpower and resources, and require even greater support during peak transport periods such as the Spring Festival travel rush; data processing lacks real-time capabilities. Therefore, there is an urgent need to improve the intelligent, rapid, and accurate detection of train wheelset geometric parameters.

[0003] Calibrating the camera's intrinsic and extrinsic parameters and establishing the light plane equation are the core parts of line structured light sensor calibration. Currently, camera calibration is usually performed using Zhang's calibration method with a handheld target and backlighting or coaxial illumination. However, this method suffers from problems such as noise interference caused by hand-held shaking and repeated poses in individual images affecting calibration efficiency. While circular target calibration offers higher accuracy, it suffers from eccentricity errors. When fitting the light plane, there are fewer spatial feature points, resulting in low light plane fitting accuracy. Therefore, high-precision calibration of the camera's intrinsic and extrinsic parameters and establishment of the light plane equation are crucial prerequisites for wheelset geometry parameter measurement.

[0004] Patent application number 202011567389.X discloses a railway wheelset size detection device and its calibration method. The relative positions of the camera and laser exhibit higher stability, enabling integrated calibration of the camera and light source, significantly reducing on-site calibration time. The calibration method includes the following steps: Step a: Using a planar target as the imaging target, extract spatial target points and calculate the homography matrix between the target plane and the image plane; Step b: Activate the light source, projecting a laser beam onto the planar target. Map the beam calibration points to three-dimensional space using the homography matrix determined by the spatial target points, obtaining the three-dimensional coordinates of the laser beam in the camera coordinate system; Step c: Move the planar target multiple times to obtain the constraint equations for the camera's intrinsic parameters and the constraint equations for the three-dimensional coordinates of the light plane. Construct an objective function using the constraint equations, calculate the camera's intrinsic parameters and light plane parameters using the objective function, and obtain the optimal solution for the light plane parameters through nonlinear optimization. However, this method is affected by the quality of the captured calibration images; different image quality can lead to errors in the calibration results. Summary of the Invention

[0005] To address the technical problems of poor real-time performance and low accuracy in existing train wheelset inspection methods, this invention proposes an online detection method and system for train wheelset geometric parameters. It utilizes a planar checkerboard-concentric circle complementary target with integrated self-backlighting and adjustable posture for camera calibration. This method is simple, practical, and improves calibration efficiency, reducing manpower requirements on-site. It also significantly improves calibration accuracy compared to traditional methods. Furthermore, this invention enables online detection of wheelset geometric parameters, with real-time uploading of results for preventative work such as wheelset maintenance, replacement, and scheduling.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: an online detection method for train wheelset geometric parameters, comprising the following steps:

[0007] (a) Determine the optical plane equation based on the intrinsic and extrinsic parameters of the camera using the planar checkerboard-concentric circle complementary target static calibration measurement system, and complete the optical plane calibration of the camera;

[0008] (b) The laser beam sensor acquires the signal of the approaching vehicle, the host computer issues a command to turn on the supplementary lighting device, and uses the camera after the light plane is calibrated to acquire the image of the train wheelset;

[0009] (c) Real-time compensation of camera attitude error based on IMU gyroscope sensor and improved PID negative feedback algorithm;

[0010] (d) The acquired train wheelset images are preprocessed, and the center line of the laser light stripe is extracted online using the Steger algorithm;

[0011] (e) Using the light plane equation obtained in step (a), calculate the geometric parameters of the train wheelset's inner diameter, outer diameter, rim width, and flange thickness online, and upload the detection results in real time.

[0012] Preferably, step (a) is implemented as follows:

[0013] (a1) Before the train passes, a planar checkerboard-concentric circle complementary target is placed at the position where the rail wheelset passes. The height of the planar checkerboard-concentric circle complementary target is adjusted to maintain the height of the train wheelset position. The target pose is changed multiple times, and the camera sequentially collects multiple calibration images.

[0014] (a2) Using the checkerboard corner points of the calibration image read into Matlab as input points, calculate the camera intrinsic parameters, extrinsic parameters, and homography matrix;

[0015] (a3) Using complementary target images acquired by the camera, the coordinates of the center compensation point are obtained based on the eccentricity error compensation algorithm of the nonlinear optimization optimal solution model;

[0016] (a4) Stop iterative optimization when the position of the center compensation point calculated in the previous two calculations is less than 0.01 pixels, update the intrinsic and extrinsic parameters of the camera, and obtain the homography matrix from the three-dimensional world coordinate system to the two-dimensional image coordinate system;

[0017] (a5) Turn on the grid laser light source and the supplementary lighting device to project the grid laser onto the planar checkerboard-concentric circle complementary target. Use the pose adjuster to change the target pose multiple times, and the camera will sequentially capture multiple images containing light stripes.

[0018] (a6) The Steger algorithm is used to extract the center line of the light stripe projected onto the object under test. Multiple three-dimensional coordinate points on the center line of the light stripe are extracted, and the least squares method is applied to fit and determine the three-dimensional camera coordinate system. The structured light plane equations are derived, and the light plane calibration is completed.

[0019] Preferably, the planar checkerboard-concentric circle complementary target includes a substrate, on which laser-written checkerboard-concentric circle complementary two-dimensional marker targets are provided. The substrate is fixedly mounted on the housing. The back of the substrate is provided with a rectangular arrangement of self-backlit high-brightness LED integrated light sources. The rear of the housing is provided with a pose adjuster, which is connected to a telescopic height linkage. When acquiring calibration images, the entire planar checkerboard-concentric circle complementary target plane occupies one-half to one-third of the area in the camera's field of view. A laser beam sensor is installed three meters in front of the calibration position and one meter away from the rail. The laser beam sensor acquires the oncoming vehicle signal and transmits it to the lower-level main controller. The lower-level main controller communicates with the upper-level computer via a serial port, thereby controlling the upper-level computer to issue commands to turn on the grid laser light source and the supplementary lighting device. The camera takes pictures at intervals to acquire images of the wheelsets of the oncoming vehicle.

[0020] Preferably, the method for calculating the camera's intrinsic and extrinsic parameters and homography matrix in step (a2) is as follows:

[0021] Based on the principle of pinhole camera models, the camera projection imaging model can be represented as follows:

[0022] ;

[0023] in, As a scaling factor, , These represent the camera's intrinsic and extrinsic parameter matrices, respectively. , These represent the rotation matrix and translation matrix, respectively, corresponding to the transformation from the three-dimensional world coordinate system to the two-dimensional image coordinate system. , The camera in the image coordinate system shaft and The physical focal length of the axis , Representing the principal points That is, the relative offset of the origin of the image coordinate system in the pixel coordinate system; Let be the homogeneous coordinates of the point in the world coordinate system. Let be the homogeneous coordinates of the corresponding image point in the two-dimensional pixel coordinate system; the origin of the world coordinate system is set at the upper left corner of the target, the origin of the pixel coordinate system is set at the upper left corner of the image, and the origin of the image coordinate system is set at the center of the image.

[0024] When calibrating a camera using a planar checkerboard-concentric circle complementary target, the world coordinate system is established on the two-dimensional target plane. The simplified representation of the camera projection imaging model is as follows:

[0025] ;

[0026] in, r1, r2, and r3 represent the homography matrix calculated by calibrating multiple images; r1, r2, and r3 represent the rotation matrices, respectively. The column vectors, where t represents the translation matrix. Column vectors;

[0027] Camera distortion model The expression for direction is:

[0028] ;

[0029] in, These are the coefficients in the mathematical expression for radial distortion. These are the coefficients of the mathematical expression for tangential distortion, where r is the image coordinate point. To the main point distance ;

[0030] Incorporating a distortion factor, the objective function is to find the minimum positional difference between the 3D projection keypoints and the 2D detection points in the least squares sense, and to establish the optimization objective function as follows: ;

[0031] in, For the first The first image taken The pixel coordinates of the key marker points 3D key markers The projected coordinates; n is the number of calibration images, m is the number of key landmarks in the calibration images, and r i t i Let r1, r2, and r3 represent the i-th translation vector and rotation vector, respectively. The parts that overlap with the preceding r1, r2, and r3 do not overlap, because they all refer to the same quantity. The initial camera intrinsic parameters, extrinsic parameters, and homography matrix are obtained by solving the objective function using Matlab.

[0032] Preferably, the implementation method of the eccentricity error compensation algorithm based on the nonlinear optimization optimal solution model is as follows: Three pixel units are constrained at the pixel-level edge, and half a pixel unit is constrained at the sub-pixel edge. Pixel-level and sub-pixel-level positioning of the ellipse edge are performed respectively. Pixel-level edge positioning uses the first-order image edge operator, i.e., the Sobel operator, while sub-pixel-level positioning uses Zernike moment pixel positioning. The equation of the eccentric ellipse is fitted using existing numerical fitting methods in numerical analysis. Based on the relationship between the true projection of the center of the imaging plane and the compensation eccentricity position, the coordinates of the center compensation point are obtained using trigonometric relationships and vector formulas, and the Zhang Zhengyou calibration method.

[0033] Preferably, the method for calculating the coordinates of the center compensation point is as follows:

[0034] The method for determining the center position of the projected ellipse is as follows: perform ellipse edge positioning, and use pixel-level edge positioning on the ellipse edge. Applying Zernike pixel-wise localization to subpixel-level edges We fit an ellipse equation by taking points on the edge pixel band;

[0035] Fitting the equations of inner and outer eccentric circles At that time, for pixel-level edges and subpixel-level edges Perform pixel constraints:

[0036] ;

[0037] in, , , , These represent the coefficients of the fitted inner and outer eccentric ellipse equations, respectively.

[0038] The equation of an eccentric ellipse was fitted using numerical fitting methods.

[0039] Based on the triangular relationship, we have:

[0040] ;

[0041] These are the inner and outer diameters of the concentric circles of the planar target, respectively. The lengths are respectively , and The included angles are respectively ;

[0042] Based on Zhang Zhengyou's calibration method, using the outer checkerboard edge corners of a planar complementary target as the target, the camera model parameters were calibrated using Matlab's calibration tools. As the initial value for iterative optimization, we can obtain the following from the vector formula and the transformation relationship between the world coordinate system and the two-dimensional pixel coordinate system:

[0043] ;

[0044] ;

[0045] Find the length Parameter values:

[0046] ;

[0047] in, The point is the center of the planar target circle; the length is determined by similarity. The size of the parameter value;

[0048] In the pixel coordinate system, the centers of the circles that fit the equations of the sub-pixel ellipses are respectively... Projection of the target's center onto the two-dimensional image plane Substituting the point into the equation of the line, we can determine the line: ;

[0049] Where A1, B1, C1, A2, B2, and C2 represent the coefficients of the equation of the straight line determined by the projection points of the inner and outer circle fitting ellipse centers and the true center of the circle center, respectively;

[0050] During the calibration process, when there is a pose tilt angle between the camera's imaging plane and the target object, the point... With point The eccentricity error between them exhibits a quadratic nonlinear relationship within the feasible tilt angle range, denoted as... Point and ,by Establish for the center Two-dimensional pixel coordinate system, according to weight Sure The angle ratio is used to determine the slope of the straight line where the compensation point is located. for The number of eccentric errors in the image, and the simultaneous points The coordinates can be used to obtain the compensation line. equation:

[0051] ;

[0052] in, , These represent the magnitudes of the eccentricity error between the center of each fitted ellipse and the projection point of the true center of the inner and outer circles, respectively.

[0053] With point Center of the circle A circle with diameter and a straight line Focus can be obtained The weight of the center deviation between the two eccentric ellipses is then used to further... Compensation is performed to obtain the compensated position of the true projection point that approximates the center of the concentric circles, which is then used as the coordinates of the key input point for 3D projection. The objective function for the optimal solution of eccentricity error is:

[0054] ;

[0055] in, This represents the magnitude of the i-th eccentricity error value; This represents the mean coefficient in the compensation calculation; , They represent the center points of the circle. The x and y coordinates;

[0056] After the camera model is recalibrated, the calibration parameters are updated to the initial values ​​of the iteration. The iteration is repeated until the change in the eccentricity error between the current and last two positioning points is less than the threshold condition of 0.01 pixels, at which point the position point coordinates are obtained. It is the optimal center compensation point.

[0057] Preferably, the method for obtaining the center line of the light stripe is as follows:

[0058] The Steger algorithm is used to extract the linear equations of the light stripes formed by the grid laser projected onto a planar checkerboard-concentric complementary target in the image in pixel coordinates. :

[0059] ;

[0060] in, It is the equation of a straight line. coefficient, These are coordinates in a two-dimensional pixel coordinate system;

[0061] By using the coordinate transformation relationship of the camera's extrinsic parameters, we can obtain the coordinates in the 3D camera coordinate system. The equation for the center line of the lower light stripe is:

[0062] ;

[0063] in, 3D camera coordinate system The coefficients of the plane equation of the lower two-dimensional marker target. Coefficients of the equation of a straight line in a 3D camera coordinate system 3D camera coordinate system The coordinates below, This refers to the camera's intrinsic focal length.

[0064] Preferably, based on the signal from the IMU gyroscope sensor, an improved PID negative feedback algorithm is used to control a two-degree-of-freedom servo motor to reset the camera, restore the initial calibration posture, and then capture an image of the object under test. In the improved PID negative feedback algorithm, a low-pass filter is added before the proportional, integral, and derivative operations to filter out interference information such as high-frequency jitter and spike signals. The angle between the IMU gyroscope sensor and the preset main axis direction is compared and calculated in real time, and the camera is adjusted to restore the preset angle, locking the camera's focal plane and maintaining the preset shooting angle. The preprocessing includes: image grayscale processing, binarization, contrast enhancement, and image denoising using a high-pass filter to eliminate stray light interference information in the image. The geometric parameters of the train wheelset, including the inner diameter, outer diameter, rim width, and flange thickness, are obtained through online calculation and transformation using a determined homography matrix. The average of multiple measurements from the images acquired by the two cameras is calculated, and then the maximum, minimum, and average values ​​of the geometric parameters of the train wheelset, including the inner diameter, outer diameter, rim width, and flange thickness, are output.

[0065] An online detection system for train wheelset geometric parameters includes a first laser beam sensor, a second laser beam sensor, and multiple sets of image acquisition devices. The first and second laser beam sensors are symmetrically installed on the outer side of the rail as a laser beam group to acquire oncoming train signals. The multiple sets of image acquisition devices are respectively located on the inner or outer side of the rail behind the first and second laser beam sensors to acquire images of the train wheelset. The first laser beam sensor, the second laser beam sensor, and the multiple sets of image acquisition devices are all communicatively connected to a lower-level main controller, which communicates with a host computer via a serial port.

[0066] Preferably, each image acquisition device includes two cameras and two supplementary lighting devices, with the two cameras set at a certain distance apart. The distance between the two cameras is to increase the measurement data of the object under test, enhance the persuasiveness of the system data, and verify the stability of the system. The supplementary lighting devices are set on the rear or outer side of the cameras. The cameras are high-speed cameras, and the grid laser light source emits grid lasers, which can be projected onto the calibration target.

[0067] The image acquisition device comprises four groups: a first group, a second group, a third group, and a fourth group. The first and fourth groups are respectively positioned on the outer sides of the two rails and are symmetrical about the center line between the two rails. The second and third groups are also positioned on the outer sides of the two rails and are symmetrical about the center line between the two rails. The first and fourth groups are housed within a protective enclosure for the first camera light source. A supplementary lighting device is located at the bottom of the protective enclosure, and the camera is located at the top of the protective enclosure. A partition is located in the middle of the protective enclosure. The second and third groups are housed within a protective enclosure for the second camera light source. A supplementary lighting device is located at the bottom of the protective enclosure, and the camera is located at the top of the protective enclosure.

[0068] The first group of image acquisition devices includes a first camera, a fifth camera, a first supplementary lighting device, and a fifth supplementary lighting device. The first camera and the fifth camera are positioned separately on the outside of a rail. The first supplementary lighting device is positioned outside the first camera, and the fifth supplementary lighting device is positioned outside the fifth camera. The fourth group of image acquisition devices includes a third camera, a third supplementary lighting device, a seventh camera, and a seventh supplementary lighting device. The third camera and the seventh camera are positioned separately on the outside of another rail. The third supplementary lighting device is positioned outside the third camera, and the seventh supplementary lighting device is positioned outside the seventh camera. The third camera and the first camera are symmetrical about the center line between the two rails, and the seventh camera and the fifth camera are symmetrical about the center line between the two rails.

[0069] The second group of image acquisition devices includes a second camera, a sixth camera, a second supplementary lighting device, and a sixth supplementary lighting device. The second camera and the sixth camera are positioned alternately on the inner side of one rail. The second supplementary lighting device is positioned diagonally behind the second camera, and the sixth supplementary lighting device is positioned diagonally behind the sixth camera. The third group of image acquisition devices includes a fourth camera, a fourth supplementary lighting device, an eighth camera, and an eighth supplementary lighting device. The fourth camera and the eighth camera are positioned alternately on the inner side of another rail. The fourth supplementary lighting device is positioned diagonally behind the fourth camera, and the eighth supplementary lighting device is positioned diagonally behind the eighth camera. The second camera and the fourth camera are symmetrical about the center line between the two rails, and the sixth camera and the eighth camera are symmetrical about the center line between the two rails.

[0070] The first camera, second camera, third camera, fourth camera, fifth camera, sixth camera, seventh camera, and eighth camera are all mounted on two-degree-of-freedom servos. Each of the first camera, second camera, third camera, fourth camera, fifth camera, sixth camera, seventh camera, and eighth camera is equipped with an IMU gyroscope sensor. The first camera, the first supplementary lighting device, the second camera, the second supplementary lighting device, the third camera, the third supplementary lighting device, the fourth camera, the fourth supplementary lighting device, the fifth camera, the fifth supplementary lighting device, the sixth camera, the sixth supplementary lighting device, the seventh camera, the seventh supplementary lighting device, the eighth camera, the eighth supplementary lighting device, the IMU gyroscope sensor, and the two-degree-of-freedom servos are all communicatively connected to the lower-level main controller.

[0071] The present invention has the following advantages and beneficial effects:

[0072] 1. This invention is based on a static high-precision calibration camera using a planar checkerboard-concentric circle complementary target with integrated self-backlighting and adjustable pose. It determines the optical plane equation, eliminating the need for additional lighting during calibration. Flexible pose adjustment eliminates the influence of human subjectivity, making it simple, practical, and freeing up manpower on-site. Compared to traditional calibration algorithms, this invention significantly improves calibration efficiency and accuracy. This invention offers advantages such as high calibration accuracy, fast real-time detection speed, and applicability to complex working environments.

[0073] 2. During online actual field measurement, the camera may experience free vibration due to factors such as mechanical vibration, serpentine wheel movement, and weather conditions. Based on the IMU gyroscope sensor and the improved PID negative feedback algorithm, the camera attitude error is compensated in real time to eliminate focal plane defocus and shooting angle shift caused by the camera's free vibration, thereby locking the camera's shooting focal plane and maintaining the preset shooting angle.

[0074] 3. The present invention installs an integrated protective box for the camera and light source on the outer side of the rail and an integrated protective box for the camera and light source on the inner side of the rail. During online actual field measurement, it avoids the influence of weather conditions such as dust, sandstorms, and rain on the normal operation of the equipment. Attached Figure Description

[0075] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0076] Figure 1 is a schematic diagram of the process of the present invention.

[0077] Figure 2 is a flowchart of the camera calibration process of the present invention.

[0078] Figure 3 is a schematic diagram of the eccentricity error compensation model of the geometric relationship of concentric circle projection. In it, (a) is the side view geometric relationship diagram of concentric circle transmission projection, and (b) is the relationship between the real projection of the center of the imaging plane and the compensated eccentricity position.

[0079] Figure 4 is a schematic diagram of the overall structure of the detection system of the present invention.

[0080] Figure 5 is a structural diagram of a single wheelset detection device in the detection system of the present invention.

[0081] Figure 6 is a schematic diagram of the interior of the integrated protective box for the first camera light source on the outer side of the rail according to the present invention.

[0082] Figure 7 is an internal schematic diagram of the integrated protective box for the second camera light source inside the rail of the present invention.

[0083] In the diagram, 1 is the first laser beam sensor, 2 is the second laser beam sensor, 3 is the first camera, 4 is the first supplementary lighting device, 5 is the second camera, 6 is the second supplementary lighting device, 7 is the third camera, 8 is the third supplementary lighting device, 9 is the fourth camera, 10 is the fourth supplementary lighting device, 11 is the fifth camera, 12 is the fifth supplementary lighting device, 13 is the sixth camera, 14 is the sixth supplementary lighting device, 15 is the seventh camera, 16 is the seventh supplementary lighting device, 17 is the eighth camera, 18 is the eighth supplementary lighting device, 19 is the integrated protection box for the first camera light source, 20 is the integrated protection box for the second camera light source, 21 is the rail, 22 is the camera, 23 is the two-degree-of-freedom servo motor, 24 is the IMU gyroscope sensor, and 25 is the supplementary lighting device. Detailed Implementation

[0084] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0085] Example 1

[0086] As shown in Figure 1, an online detection method for train wheelset geometric parameters includes the following steps:

[0087] (a) Based on the internal and external parameters of the camera in the static calibration measurement system of planar checkerboard-concentric circle complementary target, determine the light plane equation and complete the light plane calibration of the camera.

[0088] As shown in Figure 2, further, in step (a1), before the train passes, a planar checkerboard-concentric circle complementary target is placed at the position where the rail wheelset passes. The height of the planar checkerboard-concentric circle complementary target is adjusted to maintain the height of the train wheelset position, so that the entire target plane occupies one-half to one-third of the area in the camera's field of view. The target pose is changed multiple times, and the camera sequentially acquires fifteen images with an image size of 1024*1280 pixels, which are saved to the Calibration1 folder for reading into Matlab for image calibration processing.

[0089] The planar checkerboard-concentric circle complementary target consists of a ceramic substrate on which a laser-written checkerboard-concentric circle complementary two-dimensional marker target (7×9 coprime array checkerboard 45 mm, inner circle diameter 30 mm, center distance between adjacent circles 90 mm, accuracy 0.001 mm) is mounted. The substrate is fixed on a photosensitive resin 3D printed shell. The back of the substrate has a rectangular arrangement of self-backlit high-brightness LED integrated light source (62 LEDs, 24 V power supply, light intensity 100,000±2,000 lux). A pose adjuster is located at the rear of the shell, connected to a telescopic height linkage (adjustment range 0~1,500 mm). The self-backlit high-brightness LED integrated light source avoids the uneven light intensity distribution and overexposure that can occur in complex environments with backlighting or coaxial illumination, leading to image distortion and preventing the captured calibration images from being properly used in calculations.

[0090] Because the terrain and topography vary for different train wheelsets and tracks, the telescopic height linkage needs to be adjusted to maintain the height of the planar checkerboard-concentric circle complementary target at the same height as the train wheelset, ensuring that the entire target plane occupies one-half to one-third of the camera's field of view. The pose adjuster is used to change the target's pose multiple times, as the images captured by the camera sequentially include images of the target facing the camera directly, as well as images of the target tilted. The telescopic height linkage adjusts the target's height.

[0091] Compared with independent visual calibration boards on the market, this integrated self-backlit, adjustable-pose planar checkerboard-concentric-circle complementary target has advantages such as high overall integration, flexible pose adjustment, high sharpness of diffuse reflection material targets, uniform backlighting in complex environments to effectively eliminate stray light interference, elimination of manual intervention to avoid actively introducing noise, and improved efficiency of multiple imaging sessions.

[0092] A camera model describes the process of mapping three-dimensional world coordinates to a two-dimensional image plane; it can be understood as a mathematical model of projecting three-dimensional spatial feature points onto image points in a two-dimensional image coordinate system. By calibrating the monocular camera model to obtain the homography matrix, the three-dimensional world coordinates of any point can be converted into two-dimensional image coordinates. Based on the principle of the pinhole camera model, the camera projection imaging model is expressed as:

[0093] ;

[0094] in, As a scaling factor, , These represent the camera's intrinsic and extrinsic parameter matrices, respectively. , These represent the rotation matrix and translation matrix, respectively, corresponding to the transformation from the three-dimensional world coordinate system to the two-dimensional image coordinate system. , The camera in a two-dimensional image coordinate system shaft and The physical focal length of the axis , Representing the principal points That is, the relative offset of the origin of the two-dimensional image coordinate system in the pixel coordinate system. The relative offset can be obtained through the camera calibration tool in Matlab. Let be the homogeneous coordinates of the point in the three-dimensional world coordinate system. Let be the homogeneous coordinates of the corresponding image point in the two-dimensional pixel coordinate system. The origin of the world coordinate system is set at the upper left corner of the target, the origin of the pixel coordinate system is set at the upper left corner of the image, and the origin of the image coordinate system is set at the center of the image.

[0095] When calibrating a camera using a planar checkerboard-concentric circle complementary target, the three-dimensional world coordinate system is established on the two-dimensional target plane. The camera projection imaging model can be simplified as follows:

[0096] ;

[0097] in, This is the homography matrix calculated by calibrating multiple images. r1, r2, r3, and t represent the rotation matrices, respectively. Translation matrix The column vectors and extrinsic parameter matrices can be obtained using Matlab's camera calibration tool.

[0098] Radial distortion of images can occur due to factors such as the camera sensor not being parallel to the lens and lens overlap. and tangential distortion Camera distortion model The expression for direction is:

[0099] ;

[0100] The mathematical polynomial expression for radial distortion is omitted within the parentheses, because it is usually used... , Five parameters are sufficient to eliminate distortion; These are the coefficients in the mathematical expression for radial distortion. These are the coefficients of the mathematical expression for tangential distortion, where r is the image coordinate. To the main point distance .

[0101] Incorporating a distortion factor, the objective function is to find the minimum positional difference between the 3D projection keypoints and the 2D detection points in the least squares sense, and to establish the optimization objective function as follows:

[0102] ;

[0103] Among them, image coordinate points To the main point The distance is , For two dimensions The first image taken The pixel coordinates of the key marker points 3D key markers The projected coordinates. n refers to the distance from the i=1th image to the nth image, m refers to the distance from the j=1th keypoint to the mth keypoint, and r... i t i These represent the i-th translation vector and rotation vector, respectively; the parts that overlap with the preceding r1, r2, and r3 do not overlap, because they all refer to the same quantity; the objective function is solved by calculation using Matlab, and the result is the initial intrinsic and extrinsic parameters of the camera.

[0104] Furthermore, in step (a2), the checkerboard corner points of the image imported by Matlab are used as input points to determine the side length of the checkerboard square to be 45 mm. The camera's intrinsic and extrinsic parameters and homography matrix are calculated using a high-performance computer. The camera calibration toolbox in Matlab is opened, and the calibration image is loaded. The side length of the calibration target checkerboard square is determined to be 45 mm. The checkerboard corner points of the calibration image are automatically determined. Matlab calculates the camera's intrinsic and extrinsic parameters and homography matrix, which serve as initial values ​​for iterative optimization. These initial values ​​are saved to the Matlab workspace.

[0105] Furthermore, in step (a3), the coordinates of the center compensation point are obtained by using the complementary target image acquired by the camera and the eccentricity error compensation algorithm based on the nonlinear optimization optimal solution model.

[0106] When taking calibration images, there is a planar angle between the camera and the planar checkerboard-concentric complementary target when the image is not taken directly. The concentric circular target is mapped as a distorted ellipse, and there is a deviation between the true projected position of the center of the concentric circle and the actual projected ellipse center position, which is defined as eccentricity error. The magnitude of eccentricity error is closely related to the diameter of the central target and the shooting angle.

[0107] The ellipse edges are constrained by three pixels at the pixel level and by half a pixel at the subpixel level. Then, pixel-level and subpixel-level localization are performed on the ellipse edges respectively. The pixel-level edge localization is performed using a first-order image edge operator, and the subpixel-level localization is performed using Zernike moment pixel localization. In this way, the equation of the eccentric ellipse is fitted using existing numerical fitting methods in numerical analysis.

[0108] The degree of eccentricity of the center of a planar checkerboard-concentric complementary target varies depending on the diameter of the center and the shooting angle of the camera. Based on the relationship between the true projection of the center of the target onto the imaging plane and the position of the compensated eccentricity, a nonlinear optimization optimal solution model is used to obtain the coordinates of the center compensation point using trigonometric relationships and vector formulas.

[0109] To determine the center position of the projected ellipse, the ellipse edge is first located, and the Sobel operator is used to locate the pixel-level edge of the ellipse. Then apply Zernike pixel-wise localization to subpixel-level edges. We fit an ellipse equation by taking points on the edge pixel band.

[0110] To improve calibration accuracy, the equations of the inner and outer eccentric circles are fitted. At that time, for pixel-level edges and subpixel-level edges Perform pixel constraints:

[0111] ;

[0112] in, , , , These represent the coefficients of the fitted inner and outer eccentric ellipse equations, respectively.

[0113] As shown in Figure 3. The point is the center of the planar target circle. The point is the projection of the true center of the target circle onto the two-dimensional image plane. Let be the inner and outer diameters of the concentric circles of the planar target, respectively. The centers of the circles, according to the sub-pixel fitted ellipse equation, are respectively... The coordinates of the 3D projection input points are after compensation by the eccentricity error algorithm. The lengths are respectively , and The included angles are respectively . To accurately locate the eccentricity compensation point, it is first necessary to determine the location of the actual projection point of the circle's center, that is, to find... and The relationship between them, according to the triangular relationship, is as follows:

[0114] ;

[0115] Based on the classic calibration method proposed by Zhang, the camera model parameters are calibrated using the outer checkerboard edge corners of a planar complementary target as the target and the calibration tools in Matlab. As the initial value for iterative optimization, based on the vector formula and the transformation relationship between the 3D camera coordinate system and the 2D pixel coordinate system, we can obtain:

[0116] ;

[0117] ;

[0118] The results can be obtained by combining the simplified methods. Parameter values:

[0119] ;

[0120] Similarity can be used to determine The parameter value is determined by the calibration distance and angle in the actual calibration experiment. The parameter values ​​are also different, and the actual projected position of the circle's center is located. Points will also produce significant neighborhood bias, which can then be determined. A point can only serve as a pseudo-projection point of the circle's center. As shown in Figure 3(b), there is a positional deviation between the true projection of the circle's center on the imaging plane and the center of the fitted ellipse. Furthermore, as the diameter of the target circle increases, the position of the fitted ellipse's center deviates further from the target circle's center. The larger the distance between the points, the better. In the pixel coordinate system, the fitted values ​​will be... Point and Substituting the point into the equation of the line determines the... :

[0121] ;

[0122] Where A1, B1, C1, A2, B2, and C2 represent the coefficients of the linear equation determined by the projection points of the inner and outer circle fitting ellipse centers and the true center of the ellipse center, respectively.

[0123] During the calibration process, when there is a pose tilt angle between the camera's imaging plane and the target object... Point and The eccentricity error between points exhibits a quadratic nonlinear relationship within the feasible tilt angle range, denoted as... Point and ,by Establish for the center A two-dimensional pixel coordinate system can be used according to weight Sure The angle ratio is used to determine the slope of the straight line where the compensation point is located. for The number of eccentricity errors in the image, combined with... The coordinates of the point can be used to obtain the compensation line. equation:

[0124] ;

[0125] in, , These represent the magnitude of the eccentricity error between the center of each fitted ellipse and the projection point of the true center of the inner and outer circles, respectively.

[0126] by The point is the center of the circle. A circle with diameter and a straight line Focus can be obtained The weight of the center deviation between the two eccentric ellipses is then used to further... Compensation is performed to obtain the compensated position of the true projection point that approximates the center of the concentric circles, which is then used as the coordinates of the key input point for 3D projection. The objective function for the optimal solution of eccentricity error is:

[0127] ;

[0128] in, This represents the magnitude of the i-th eccentricity error value; This represents the mean coefficient in the compensation calculation; , They represent the center points of the circle. The x and y coordinates.

[0129] After the camera model is recalibrated, the calibration parameters are updated to the initial values ​​of the iteration. The iteration is repeated until the change in the eccentricity error between the current and last two positioning points is less than the threshold condition of 0.01 pixels, at which point the position point coordinates are obtained. This is the optimal deviation error compensation point.

[0130] Furthermore, in step (a4), during the process of solving the optimal solution of the nonlinear optimization, the iteration optimization stops when the position of the center compensation point calculated in the previous two calculations is less than 0.01 pixels. The intrinsic and extrinsic parameters of the camera are updated, and the homography matrix from the three-dimensional world coordinate system to the two-dimensional image coordinate system is obtained. After determining the homography matrix, coordinate transformation can be performed, which can then be used to measure and calculate the size of the object.

[0131] Furthermore, in step (a5), the grid laser source and the supplementary lighting device are turned on, so that the grid laser is projected onto the planar checkerboard-concentric circle complementary target. The target pose is changed multiple times using the pose adjuster, and the camera sequentially acquires fifteen images with an image size of 1024*1280 pixels, which are then saved to the Calibration2 folder for import into Matlab for image processing.

[0132] The Steger algorithm is used to extract the linear equations of light stripes in an image in a two-dimensional pixel coordinate system. :

[0133] ;

[0134] in, It is the equation of a straight line. coefficient, These are coordinates in a two-dimensional pixel coordinate system. The light stripes are formed by projecting the grid laser light onto a planar checkerboard-concentric complementary target through a grid laser light source and a supplementary lighting device.

[0135] By using the coordinate transformation relationship of the camera's extrinsic parameters, we can obtain the coordinates in the 3D camera coordinate system. The equation for the center line of the lower light stripe is:

[0136] ;

[0137] in, 3D camera coordinate system The four coefficients are plane equation coefficients for the two-dimensional target object, and their values ​​can be positive or negative. Coefficients of the equation of a straight line in a 3D camera coordinate system 3D camera coordinate system The coordinates below, This refers to the camera's intrinsic focal length.

[0138] Furthermore, in step (a6), the Steger algorithm is used to extract the center line of the light stripe projected onto the object under test, and multiple three-dimensional coordinate points on the center line of the light stripe are extracted to increase the number of feature points. The least squares method is then applied to fit and determine the three-dimensional camera coordinate system. The optical plane equation of the structured light is used to complete the optical plane calibration. By calculating the centerline of the light stripe using the optical plane equation of the structured light, the measurement dimensions of the object under test can be obtained through coordinate transformation.

[0139] Furthermore, camera calibration and determination of the light plane equation are performed for the first camera 3, the second camera 5, the third camera 7, the fourth camera 9, the fifth camera 11, the sixth camera 13, the seventh camera 15, and the eighth camera 17. Eight cameras are set up, with the two cameras on one side of the wheelset forming a group. This allows each pair of cameras to measure the same wheelset parameters, calculating multiple sets of measurements. These two sets of data can then be used to determine the maximum, minimum, mean, and standard deviation, preventing the influence of a single data result on the analysis and judgment.

[0140] (b) The laser beam sensor acquires the signal of the approaching train, the host computer issues a command to turn on the supplementary lighting device, and uses the camera after the light plane is calibrated to acquire the image of the train wheelset.

[0141] Furthermore, in step (b), a laser beam sensor is installed three meters before the calibration position and one meter away from the rail. The laser beam sensor acquires the incoming vehicle signal and transmits it to the lower-level main controller. The lower-level main controller communicates with the upper-level computer via serial port. The upper-level computer is a high-performance computer. The upper-level controller then issues commands to activate the grid laser light source and supplementary lighting device. The camera periodically captures images of the wheelsets of the incoming vehicle. The image size is 1024*1280 pixels, and the images are saved to the corresponding camera image folder for later import into Matlab for image processing.

[0142] (c) Real-time compensation of camera attitude error based on IMU gyroscope sensor and improved PID negative feedback algorithm.

[0143] Because passing trains cause vibrations in both the camera light source integrated protection box on the outer and inner sides of the rail, a low-pass filter is added before the proportional, integral, and derivative operations in the improved PID negative feedback algorithm to filter out high-frequency jitter signals, spike signals, and other interference. A two-degree-of-freedom servo motor drives the camera body to adjust its attitude. The IMU gyroscope sensor compares and calculates the angle with the preset main axis direction in real time, adjusting the camera to restore the preset angle and locking the camera's focal plane to maintain the preset shooting angle. When a train passes or the environment causes the camera light source integrated protection box to shake or vibrate, the improved PID negative feedback algorithm, based on the IMU gyroscope sensor signal, controls the two-degree-of-freedom servo motor to reset the camera, restoring its initial calibration attitude and allowing it to capture images of the object under test. Without this algorithm, environmental vibrations, train vibrations, etc., would prevent the camera from directly facing the object under test, resulting in only part of the object being in the field of view.

[0144] (d) Preprocess the acquired train wheelset images and extract the center line of the laser stripe online;

[0145] Furthermore, in step (d), the acquired train wheelset image is preprocessed, including image grayscale processing, binarization, and contrast enhancement. Contrast enhancement can be achieved using grayscale mapping, setting pixels with grayscale values ​​less than 40 to 0 and pixels with grayscale values ​​greater than 160 to 1, uniformly mapping grayscale values ​​between 40 and 160 to 0-255, thereby enhancing contrast and making the image clearer with more obvious contrast. A high-pass filter is used for image denoising, with an initial threshold of 50, which eliminates stray light interference in the acquired image. Image processing retains the main information of the wheelset, and then the Steger algorithm is used to extract the center lines of multiple laser beams from the acquired image in step (c) online. The Steger algorithm can perform sub-pixel extraction, making the extracted laser beam center lines more accurate, thus resulting in more precise calculations of the object's size.

[0146] (e) Using the smooth plane equation obtained in step (a), calculate the geometric parameters of the train wheelset such as inner diameter, outer diameter, rim width, and flange thickness online, and upload the detection results in real time.

[0147] Further, in step (e), the centerline of the light stripe is calculated using the linear structured light plane equation, and the measured dimensions of the object under test can be obtained through coordinate transformation. Geometric parameters such as the inner diameter, outer diameter, rim width, and flange thickness of the train wheelset can be obtained through online calculation and transformation using the determined homography matrix. For example, the rim thickness is obtained by first extracting the centerline equation of the light stripe and then calculating it online using the determined homography matrix. After Matlab data processing, the first camera 3 and the fifth camera 11 are grouped together. Multiple measurements from the acquired images are averaged, and then the maximum, minimum, and average values ​​of the geometric parameters of the train wheelset—inner diameter, outer diameter, rim width, and flange thickness—are output. Because multiple measurements can be used for error analysis when a single camera has an error or the system is unstable, the measurements from the two groups of cameras are also compared to increase the data volume. Comparing the root mean square error of the two groups of data can determine the validity of the data, and then the maximum, minimum, and average measurement values ​​are output based on the results from the two groups.

[0148] Furthermore, the second camera 5 and the sixth camera 13 are grouped together, the third camera 7 and the seventh camera 15 are grouped together, the fourth camera 9 and the eighth camera 17 are grouped together, and the fifth camera 11 and the first camera 3 are grouped together. The detection results are processed by Matlab and uploaded in real time. The measurement results of the first camera are stored in one array, and the measurement results of the second camera are stored in a second array. Data processing is performed by calling Matlab.

[0149] The online detection method for train wheelset geometric parameters described in this embodiment can quickly and accurately detect geometric parameters such as inner diameter, outer diameter, rim width, and flange thickness of train wheelsets. By using the camera's internal and external parameters in a static high-precision calibration measurement system based on a planar checkerboard-concentric circle complementary target, the equation of the optical plane is determined, which significantly improves the calibration efficiency and accuracy compared to traditional calibration algorithms. During online actual field measurement, an IMU gyroscope sensor and an improved PID negative feedback algorithm are used to compensate for camera attitude errors in real time, eliminating focal plane defocus and shooting angle shift caused by camera free vibration, thus locking the camera's shooting focal plane and maintaining the preset shooting angle.

[0150] Example 2

[0151] As shown in Figure 4, an online detection system for train wheelset geometric parameters includes a first laser beam sensor 1, a second laser beam sensor 2, and multiple sets of image acquisition devices. The first laser beam sensor 1 and the second laser beam sensor 2 are symmetrically installed on the outer side of the rail 21 as a laser beam group to acquire oncoming train signals. Multiple sets of image acquisition devices are located on the inner or outer side of the rail 21 behind the first laser beam sensor 1 and the second laser beam sensor 2, for acquiring images of the train wheelset. The first laser beam sensor 1, the second laser beam sensor 2, and the multiple sets of image acquisition devices are all communicatively connected to a lower-level main controller, which communicates with a host computer via a serial port. When the host computer issues a control command, the lower-level main controller activates the laser grid light source supplementary lighting device of the image acquisition device, and the camera begins acquiring images.

[0152] Each image acquisition unit consists of two cameras and two supplementary lighting devices, with the two cameras positioned at a certain distance. This distance is to enhance the measurement data of the object under test, strengthen the persuasiveness of the system data, and verify the system's stability. If the cameras were positioned too close together, measurements would be taken at almost the same location, rendering the measurements meaningless. The supplementary lighting devices are positioned correspondingly to the rear or outer side of the cameras, without affecting image acquisition. The cameras are high-speed cameras, which have a fast frame rate, facilitating real-time online calculation of the object's parameters. The grid laser source emits grid laser light, which can be projected onto the calibration target, enabling the extraction of the light stripe centerline. The supplementary lighting devices provide ambient light to the high-speed cameras.

[0153] As shown in Figure 4, this embodiment includes four sets of image acquisition devices: a first set, a second set, a third set, and a fourth set. The first and fourth sets of image acquisition devices are respectively positioned on the outer sides of the two rails 21, and are symmetrical about the center line between the two rails 21. The second and third sets of image acquisition devices are also positioned on the outer sides of the two rails 21, and are symmetrical about the center line between the two rails 21. Because the position of a set of wheelsets is fixed, this arrangement ensures that a set of wheelsets can be simultaneously measured online in real time.

[0154] The first group of image acquisition devices includes a first camera 3, a fifth camera 11, a first supplementary lighting device 4, and a fifth supplementary lighting device 12. The first camera 3 and the fifth camera 11 are positioned spaced apart on the outside of a rail 21, with a typical interval of 5 meters, determined by the actual track layout and connections. The distance between the first camera 3 and the fifth camera 11 and the outside of the rail 21 is generally 2 meters. The first supplementary lighting device 4 is positioned outside the first camera 3, serving as its light source, and the fifth supplementary lighting device 12 is positioned outside the fifth camera 11, also serving as its light source. The fourth group of image acquisition devices includes a third camera 7, a third supplementary lighting device 8, a seventh camera 15, and a seventh supplementary lighting device 16. The third camera 7 and the seventh camera 15 are positioned spaced apart on the outside of another rail 21. The third supplementary lighting device 8 is positioned outside the third camera 7, and the seventh supplementary lighting device 16 is positioned outside the seventh camera 15. The third camera 7 and the first camera 3 are symmetrical about the centerline between the two rails 21, and the seventh camera 15 and the fifth camera 11 are symmetrical about the centerline between the two rails 21. As shown in Figures 5 and 6, the first and fourth image acquisition devices are housed inside the integrated protective box 19 for the first camera light source. The supplementary lighting device 25 is located at the lower part of the integrated protective box 19, and the camera 22 is located at the upper part of the integrated protective box 19. A partition is provided in the middle of the integrated protective box 19. The purpose of the partition is to prevent signal interference from generating noise, which would affect the noise of the images captured by the high-speed camera and thus the measurement results. At the same time, the partition prevents contact between the circuits, ensuring the stability of the system.

[0155] The second set of image acquisition devices includes a second camera 5, a sixth camera 13, a second supplementary lighting device 6, and a sixth supplementary lighting device 14. The second camera 5 and the sixth camera 13 are spaced apart on the inner side of a steel rail 21, with the spacing generally synchronized with the previous setting at 5m. The second camera 5 and the sixth camera 13 are 50cm away from the steel rail and placed at a 30-degree angle. The second supplementary lighting device 6 is positioned diagonally behind the second camera 5, and the sixth supplementary lighting device 14 is positioned diagonally behind the sixth camera 13. The third set of image acquisition devices includes a fourth camera 9, a fourth supplementary lighting device 10, an eighth camera 17, and an eighth supplementary lighting device 18. The fourth camera 9 and the eighth camera 17 are spaced apart on the inner side of another steel rail 21. The fourth supplementary lighting device 10 is positioned diagonally behind the fourth camera 9, and the eighth supplementary lighting device 18 is positioned diagonally behind the eighth camera 17. The second camera 5 and the fourth camera 9 are symmetrical about the center line between the two steel rails 21, and the sixth camera 13 and the eighth camera 17 are symmetrical about the center line between the two steel rails 21. As shown in Figure 6, the second and third sets of image acquisition devices are set in the integrated protective box 20 for the second camera light source. The supplementary lighting device 25 is set in the lower part of the integrated protective box 20 for the second camera light source, and the camera 22 is set in the upper part of the integrated protective box 20 for the second camera light source. The supplementary lighting device 25 and the camera 22 are relatively close to each other inside the integrated protective box 20 for the second camera light source. This arrangement of the supplementary lighting device makes the ambient light compensation more concentrated. If the distance is too far, the supplementary lighting effect will be poor.

[0156] As shown in Figures 6 and 7, the first camera 3, the second camera 5, the third camera 7, the fourth camera 9, the fifth camera 11, the sixth camera 13, the seventh camera 15, and the eighth camera 17 are all mounted on two-degree-of-freedom servo motors 23. Each of these cameras is equipped with an IMU gyroscope sensor 24. The two-degree-of-freedom servo motors 23 are housed within either the integrated protection box 19 for the first camera's light source or the integrated protection box 20 for the second camera's light source. Both the IMU gyroscope sensor 24 and the two-degree-of-freedom servo motors 23 are communicatively connected to the lower-level main controller. When a train passes by or the environment causes the integrated protection box for the camera's light source to vibrate or shake, the improved PID negative feedback algorithm, based on the signal from the IMU gyroscope sensor, can control the two-degree-of-freedom servo motors to reset the cameras, restore their initial calibration posture, and then capture images of the object under test. Without this algorithm, environmental vibrations, train vibrations, and other factors can prevent the cameras from directly facing the object under test, resulting in only a portion of the object being in the field of view.

[0157] The first camera 3, the first supplementary lighting device 4, the second camera 5, the second supplementary lighting device 6, the third camera 7, the third supplementary lighting device 8, the fourth camera 9, the fourth supplementary lighting device 10, the fifth camera 11, the fifth supplementary lighting device 12, the sixth camera 13, the sixth supplementary lighting device 14, the seventh camera 15, the seventh supplementary lighting device 16, the eighth camera 17, and the eighth supplementary lighting device 18 are all communicatively connected to the lower-level main controller. The lower-level main controller is a high-performance computer. While backing up and saving the acquired images, the high-performance computer processes the acquired images using the online detection method of Example 1 to extract the target light stripe centerline. Based on the static high-precision calibration measurement system, the camera's internal and external parameters and the light plane equation are calculated online in real time, and then the geometric parameters such as the train wheelset's inner diameter, outer diameter, rim width, and flange thickness are output. The detection results are uploaded in real time.

[0158] In this embodiment, an integrated protective box for the camera light source on the outside of the rail and an integrated protective box for the camera light source on the inside of the rail are installed on the outside of the camera and the supplementary lighting device. This avoids the influence of weather conditions such as dust, sandstorms, and rain on the normal operation of the equipment and is suitable for complex working environments.

[0159] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for online detection of geometric parameters of train wheelsets, characterized in that, The steps are as follows: (a) Based on the intrinsic and extrinsic parameters of the camera in the static calibration measurement system of the planar checkerboard-concentric circle complementary target, determine the light plane equation and complete the light plane calibration of the camera; the planar checkerboard-concentric circle complementary target includes a substrate, on which a laser-written checkerboard-concentric circle complementary two-dimensional marker target is provided, the substrate is fixedly set on the housing, and the back of the substrate is provided with a rectangular arrangement of self-backlit high-brightness LED integrated light source, and the rear of the housing is provided with a pose adjuster, which is connected to the telescopic height linkage; (b) Laser calibration of the target. (a) The sensor acquires the signal of the approaching vehicle, the host computer issues a command to turn on the supplementary lighting device, and the camera after light plane calibration is used to acquire images of the train wheelset; (c) The camera attitude error is compensated in real time based on the IMU gyroscope sensor and the improved PID negative feedback algorithm; (d) The acquired train wheelset images are preprocessed, and the center line of the laser light stripe is extracted online using the Steger algorithm; (e) The geometric parameters of the inner diameter, outer diameter, rim width and flange thickness of the train wheelset are calculated online using the light plane equation obtained in step (a), and the detection results are uploaded in real time.

2. The online detection method for train wheelset geometric parameters according to claim 1, characterized in that, The implementation method of step (a) is as follows: (a1) Before the train passes, a planar checkerboard-concentric circle complementary target is placed at the position where the rail wheelset passes. The height of the planar checkerboard-concentric circle complementary target is adjusted to maintain the height of the train wheelset position. The target pose is changed multiple times, and the camera sequentially acquires multiple calibration images; (a2) Using the checkerboard corner points of the calibration images read into Matlab as input points, the camera intrinsic parameters, extrinsic parameters, and homography matrix are calculated; (a3) ​​Using the complementary target images acquired by the camera, based on the nonlinear optimization optimal solution model... The eccentricity error compensation algorithm obtains the coordinates of the center compensation point; (a4) when the position of the center compensation point calculated in the previous two calculations is less than 0.01 pixels, the iteration optimization stops, the intrinsic and extrinsic parameters of the camera are updated, and the homography matrix from the three-dimensional world coordinate system to the two-dimensional image coordinate system is obtained; (a5) the grid laser light source and the supplementary lighting device are turned on, so that the grid laser is projected onto the planar checkerboard-concentric circle complementary target, and the target pose is changed multiple times using the pose adjuster, and the camera sequentially acquires multiple images containing light stripes; (a6) the Steger algorithm is used to extract the center line of the light stripe projected on the test object, extract the information of multiple three-dimensional coordinate points on the center line of the light stripe, and the least squares method is applied to fit and determine the three-dimensional camera coordinate system. The structured light plane equations are derived, and the light plane calibration is completed.

3. The online detection method for train wheelset geometric parameters according to claim 2, characterized in that, When acquiring calibration images, the entire planar checkerboard-concentric circle complementary target plane occupies one-half to one-third of the camera's field of view. A laser beam sensor is installed three meters in front of the calibration position and one meter away from the rail. The laser beam sensor acquires the signal of the approaching vehicle and transmits it to the lower-level main controller. The lower-level main controller communicates with the upper-level computer via serial port, and then controls the upper-level computer to issue instructions to turn on the grid laser light source and the supplementary lighting device. The camera then takes pictures at intervals to acquire images of the wheelsets of the approaching vehicle.

4. The online detection method for train wheelset geometric parameters according to claim 2 or 3, characterized in that, The method for calculating the camera's intrinsic and extrinsic parameters and homography matrix in step (a2) is as follows: Based on the principle of the pinhole camera model, the camera projection imaging model is represented as: ;in, As a scaling factor, 、 These represent the camera's intrinsic and extrinsic parameter matrices, respectively. 、 These represent the rotation matrix and translation matrix, respectively, corresponding to the transformation from the three-dimensional world coordinate system to the two-dimensional image coordinate system. 、 The camera in the image coordinate system shaft and The physical focal length of the axis 、 Representing the principal points That is, the relative offset of the origin of the image coordinate system in the pixel coordinate system; Let be the homogeneous coordinates of the point in the world coordinate system. Here, the coordinates are the homogeneous coordinates of the corresponding image point in the two-dimensional pixel coordinate system; the origin of the world coordinate system is set at the upper left corner of the target, the origin of the pixel coordinate system is set at the upper left corner of the image, and the origin of the image coordinate system is set at the center of the image; when calibrating the camera using a planar checkerboard-concentric circle complementary target, the world coordinate system is established on the two-dimensional target plane. The simplified representation of the camera projection imaging model is as follows: ;in, r1, r2, and r3 represent the homography matrix calculated by calibrating multiple images; r1, r2, and r3 represent the rotation matrices, respectively. The column vectors, where t represents the translation matrix. Column vectors; camera distortion model The expression for direction is: ;in, These are the coefficients in the mathematical expression for radial distortion. These are the coefficients of the mathematical expression for tangential distortion, where r is the image coordinate point. To the main point distance By incorporating a distortion factor, the objective is to find the minimum positional difference between the 3D projection keypoints and the 2D detection points in the least squares sense. The optimization objective function is then established as follows: ;in, For the first The first image taken The pixel coordinates of the key marker points 3D key markers The projected coordinates; n is the number of calibration images, m is the number of key landmarks in the calibration images, and r i t i Let r1, r2, and r3 represent the i-th translation vector and rotation vector, respectively. The parts that overlap with the preceding r1, r2, and r3 do not overlap, because they all refer to the same quantity. The initial camera intrinsic parameters, extrinsic parameters, and homography matrix are obtained by solving the objective function using Matlab.

5. The online detection method for train wheelset geometric parameters according to claim 4, characterized in that, The implementation method of the eccentricity error compensation algorithm based on the nonlinear optimization optimal solution model is as follows: A three-pixel constraint is applied to the pixel-level edge, and a half-pixel constraint is applied to the sub-pixel edge. Pixel-level and sub-pixel-level positioning of the ellipse edge are performed respectively. Pixel-level edge positioning uses the first-order image edge operator, i.e., the Sobel operator, while sub-pixel-level positioning uses Zernike moment pixel positioning. The equation of the eccentric ellipse is fitted using existing numerical fitting methods in numerical analysis. Based on the relationship between the true projection of the center of the imaging plane and the position of the compensated eccentricity, the coordinates of the center compensation point are obtained using trigonometric relationships and vector formulas, and the Zhang Zhengyou calibration method.

6. The online detection method for train wheelset geometric parameters according to claim 5, characterized in that, The method for calculating the coordinates of the center compensation point is as follows: The method for determining the position of the center of the projected ellipse is as follows: Perform ellipse edge positioning, and use pixel-level edge positioning for the ellipse edge. Applying Zernike pixel-wise localization to subpixel-level edges We fit an ellipse equation by taking points on the edge pixel band; Fitting the equations of inner and outer eccentric circles At that time, for pixel-level edges and subpixel-level edges Perform pixel constraints: ;in, 、 、 、 Let represent the coefficients of the fitted inner and outer eccentric ellipse equations, respectively; the eccentric ellipse equation is fitted using numerical fitting methods; based on trigonometric relationships, we have: ； These are the inner and outer diameters of the concentric circles of the planar target, respectively. The lengths are respectively ， and The included angles are respectively Based on Zhang Zhengyou's calibration method, using the outer checkerboard edge corners of a planar complementary target as the target, the camera model parameters were calibrated using Matlab's calibration tools. As the initial value for iterative optimization, we can obtain the following from the vector formula and the transformation relationship between the world coordinate system and the two-dimensional pixel coordinate system: ； ; Calculate the length Parameter values: ;in, The point is the center of the planar target circle; the length is determined by similarity. The magnitude of the parameter values; in the pixel coordinate system, the centers of the circles of the fitted ellipse equations for the sub-pixel points are respectively... Projection of the target's center onto the two-dimensional image plane Substituting the point into the equation of the line, we can determine the line: Where A1, B1, C1, A2, B2, and C2 represent the coefficients of the linear equation determined by the projection points of the inner and outer circle fitting ellipse centers and the true center of the ellipse center, respectively; during the calibration process, when there is a pose tilt angle between the camera imaging plane and the target object, the point... With point The eccentricity error between them exhibits a quadratic nonlinear relationship within the feasible tilt angle range, denoted as... Point and ,by Establish for the center Two-dimensional pixel coordinate system, according to weight Sure The angle ratio is used to determine the slope of the straight line where the compensation point is located. for The number of eccentric errors in the image, and the simultaneous points The coordinates can be used to obtain the compensation line. equation: ;in, 、 These represent the magnitudes of the eccentricity error between the center of each fitted ellipse and the projection point of the true center of the inner and outer circles, respectively; with points... Center of the circle A circle with diameter and a straight line Focus can be obtained The weight of the center deviation between the two eccentric ellipses is then used to... Compensation is performed to obtain the compensated position of the true projection point that approximates the center of the concentric circles, which is then used as the coordinates of the key marker input point for 3D projection. The objective function for the optimal solution of eccentricity error is: ;in, This represents the magnitude of the i-th eccentricity error value; This represents the mean coefficient in the compensation calculation; 、 They represent the center points of the circle. The x and y coordinates are obtained; after the camera model is recalibrated, the calibration parameters are updated to the initial values ​​of the iteration. The iteration is repeated until the change in the eccentricity error of the current two positioning points is less than the threshold condition of 0.01 pixels, at which point the position point coordinates are obtained. It is the optimal center compensation point.

7. The online detection method for train wheelset geometric parameters according to claim 5 or 6, characterized in that, The method for obtaining the center line of the light stripe is as follows: using the Steger algorithm to extract the linear equation of the light stripe formed by the grid laser projected onto the planar checkerboard-concentric complementary target in the pixel coordinate system image. ： ;in, It is the equation of a straight line. coefficient, These are coordinates in a two-dimensional pixel coordinate system; through coordinate transformation relationships of camera extrinsic parameters, the coordinates in a three-dimensional camera coordinate system are obtained. The equation for the center line of the lower light stripe is: ;in, 3D camera coordinate system The coefficients of the plane equation of the lower two-dimensional target marker. Coefficients of the equation of a straight line in a 3D camera coordinate system 3D camera coordinate system The coordinates below, This refers to the camera's intrinsic focal length.

8. The online detection method for train wheelset geometric parameters according to claim 7, characterized in that, Based on the signal from the IMU gyroscope sensor, an improved PID negative feedback algorithm is used to control a two-degree-of-freedom servo motor to reset the camera, restore the initial calibration posture, and then capture an image of the object under test. In the improved PID negative feedback algorithm, a low-pass filter is added before the proportional, integral, and derivative functions to filter out interference information such as high-frequency jitter signals and spike signals. The angle between the IMU gyroscope sensor and the preset main axis direction is compared and calculated in real time, and the camera is adjusted to restore the preset angle, so that the camera's focal plane is locked and the preset shooting angle is maintained. The preprocessing includes: image grayscale processing, binarization processing, contrast enhancement, and image denoising using a high-pass filter to eliminate stray light interference information in the image. The geometric parameters of the train wheelset, including inner diameter, outer diameter, rim width, and flange thickness, are obtained through online calculation and transformation using a defined homography matrix. The mean value of multiple measurements from images acquired by two cameras is calculated, and then the maximum, minimum, and average values ​​of the geometric parameters of the train wheelset, including inner diameter, outer diameter, rim width, and flange thickness, are output.

9. The detection system for the online detection method of train wheelset geometric parameters according to any one of claims 2-8, characterized in that, The system includes a first laser beam sensor (1), a second laser beam sensor (2), and multiple sets of image acquisition devices. The first laser beam sensor (1) and the second laser beam sensor (2) are symmetrically installed on the outside of the rail (21) as a laser beam group to acquire oncoming train signals. The multiple sets of image acquisition devices are respectively set on the inside or outside of the rail (21) behind the first laser beam sensor (1) and the second laser beam sensor (2) to acquire images of train wheelsets. The first laser beam sensor (1), the second laser beam sensor (2), and several sets of image acquisition devices are all connected to the lower-level main controller for communication. The lower-level main controller communicates with the upper-level computer via a serial port.

10. The detection system of the online detection method for train wheelset geometric parameters according to claim 9, characterized in that, Each image acquisition device includes two cameras and two supplementary lighting devices, with the two cameras spaced a certain distance apart. This distance is to increase the measurement data of the object under test, enhance the persuasiveness of the system data, and verify the stability of the system. The supplementary lighting devices are positioned correspondingly behind or outside the cameras. The cameras are high-speed cameras, and the grid laser source emits grid laser light, which can be projected onto the calibration target. There are four sets of image acquisition devices: a first set, a second set, a third set, and a fourth set. The first and fourth image acquisition devices are respectively set on the outside of the two rails (21), and the first and fourth image acquisition devices are symmetrical about the center line between the two rails (21); the second and third image acquisition devices are respectively set on the outside of the two rails (21), and the second and third image acquisition devices are symmetrical about the center line between the two rails (21); the first and fourth image acquisition devices are set inside the integrated protective box (19) of the first camera light source, the supplementary lighting device is set at the lower part of the integrated protective box (19) of the first camera light source, and the camera is set at the upper part of the integrated protective box (19) of the first camera light source; the first... A partition is provided in the middle of a camera light source integrated protective box (19); the second group of image acquisition devices and the third group of image acquisition devices are set in the second camera light source integrated protective box (20), the supplementary lighting device is set in the lower part of the second camera light source integrated protective box (20), and the camera is set in the upper part of the second camera light source integrated protective box (20); the first group of image acquisition devices includes a first camera (3), a fifth camera (11), a first supplementary lighting device (4), and a fifth supplementary lighting device (12), the first camera (3) and the fifth camera (11) are set apart on the outside of a steel rail (21); the first supplementary lighting device (4) is set on the outside of the first camera (3), the fifth supplementary lighting device (12) is set apart on the outside of the first camera (3), and the fifth supplementary lighting device (12) is set apart on the outside of the first camera (3). The supplementary lighting device (12) is set outside the fifth camera (11); the fourth group of image acquisition devices includes a third camera (7), a third supplementary lighting device (8), a seventh camera (15), and a seventh supplementary lighting device (16). The third camera (7) and the seventh camera (15) are set at intervals outside another rail (21). The third supplementary lighting device (8) is set outside the third camera (7). The seventh supplementary lighting device (16) is set outside the seventh camera (15). The third camera (7) and the first camera (3) are symmetrical about the center line between the two rails (21). The seventh camera (15) and the fifth camera (11) are symmetrical about the center line between the two rails (21).The second group of image acquisition devices includes a second camera (5), a sixth camera (13), a second supplementary lighting device (6), and a sixth supplementary lighting device (14). The second camera (5) and the sixth camera (13) are arranged side-by-side on the inner side of a steel rail (21). The second supplementary lighting device (6) is located diagonally behind the second camera (5), and the sixth supplementary lighting device (14) is located diagonally behind the sixth camera (13). The third group of image acquisition devices includes a fourth camera (9), a fourth supplementary lighting device (10), an eighth camera (17), and a sixth supplementary lighting device (14). Eight supplementary lighting devices (18), the fourth camera (9) and the eighth camera (17) are spaced apart on the inner side of another rail (21), the fourth supplementary lighting device (10) is set on the oblique rear side of the fourth camera (9), and the eighth supplementary lighting device (18) is set on the oblique rear side of the eighth camera (17); the second camera (5) and the fourth camera (9) are symmetrical about the center line between the two rails (21), and the sixth camera (13) and the eighth camera (17) are symmetrical about the center line between the two rails (21); the first camera (3), the fourth camera (9), the eighth camera (17), the eighth camera (9 ... The second camera (5), the third camera (7), the fourth camera (9), the fifth camera (11), the sixth camera (13), the seventh camera (15), and the eighth camera (17) are all mounted on a two-degree-of-freedom servo motor (23). The first camera (3), the second camera (5), the third camera (7), the fourth camera (9), the fifth camera (11), the sixth camera (13), the seventh camera (15), and the eighth camera (17) are all equipped with IMU gyroscope sensors (24). The first camera (3), the first supplementary lighting device ( 4) The second camera (5), the second supplementary lighting device (6), the third camera (7), the third supplementary lighting device (8), the fourth camera (9), the fourth supplementary lighting device (10), the fifth camera (11), the fifth supplementary lighting device (12), the sixth camera (13), the sixth supplementary lighting device (14), the seventh camera (15), the seventh supplementary lighting device (16), the eighth camera (17), the eighth supplementary lighting device (18), the IMU gyroscope sensor (24), and the two-degree-of-freedom servo motor (23) are all connected to the lower-level main controller for communication.

Images