A method for detecting out-of-roundness of train wheels based on three-dimensional information

Abstract

The present invention discloses a method for detecting the out-of-roundness of train wheels based on three-dimensional information, including image acquisition and preprocessing, laser centerline extraction, laser coordinate transformation and wheel out-of-roundness detection. The present invention only needs to use a small number of linear array cameras and a combination of 3D laser scanner modules to generate multiple two-dimensional image data and laser scanning data with depth information, extract the tread area of ​​the laser scan, and reconstruct the wheel tread profile with depth information based on an elliptical model. Whether it is the local out-of-roundness of the wheel or the overall out-of-roundness, the method of the present invention can accurately calculate it through the wheel tread profile data, and generate a test report in a timely and dynamic manner for the customer to review and repair.

Description

Technical Field

[0001] The invention belongs to the field of digital image processing, and in particular relates to a train wheel out-of-roundness detection method based on three-dimensional information, and is applied to a train wheel pair online detection product. Background Art

[0002] With the rapid development of my country's rail transit, the traditional manual method of wheel out-of-roundness detection can no longer meet the daily operation needs, and the emergence of high-precision intelligent non-contact detection equipment has ensured the safety of increasing train operations. In train operation, wheels are key components for driving safety. Friction between the wheels and the rails during operation will cause wheel wear; during turning, insufficient lubrication and other conditions can easily cause the overall wheel to be out of round; when braking is unreasonable, the wheels will have local radius changes such as scratches, peeling, and falling blocks, causing local wheel out-of-roundness. In severe cases, the out-of-roundness of wheels can lead to train accidents, so it is necessary to dynamically detect the out-of-roundness of train wheels.

[0003] At present, the out-of-roundness detection of train wheels mainly includes static detection and dynamic detection. Static detection is mainly carried out manually when the train is stopped, which can no longer meet the growing demand for trains. Dynamic detection is a non-contact online detection method performed by intelligent equipment without affecting the running of the train.

[0004] Chinese patent ZL201310556634 discloses a laser sensor-based urban rail vehicle wheel out-of-roundness detection device and method, the method uses multiple laser sensors, all of which are coplanar with the wheel circumference to be measured, and are installed under the wheel according to a certain geometric relationship. The sensor simultaneously detects the wheel to obtain the detection point, and the diameter is obtained by least square fitting, and then the diameter within the entire circumference of the wheel is averaged in sections, and the wheel out-of-roundness is obtained by subtracting the maximum value from the minimum value. Although the invention can measure the wheel out-of-roundness, it requires that all laser sensors be coplanar with the wheel circumference to be measured and are installed under the wheel according to a certain geometric relationship. This installation method has high requirements for preliminary work, requires a large number of laser sensors, and fits the circle through the wheel tread data, and the accuracy is relatively low. Summary of the invention

[0005] In view of the shortcomings and deficiencies in the above-mentioned prior art, the present invention aims to provide a method for detecting the out-of-roundness of train wheels based on three-dimensional information.

[0006] The present invention adopts the following technical solutions to solve the above technical problems:

[0007] A method for detecting out-of-roundness of train wheels based on three-dimensional information, comprising:

[0008] Image acquisition and preprocessing to obtain laser line images;

[0009] Laser centerline extraction: extract the centerline of the laser image, with two-dimensional coordinates (x, y);

[0010] Laser coordinate transformation: The image point set generated by the laser line on the wheel surface, i.e., the (x, y) point, is converted into the coordinates (X, Y) in the real world coordinate system. w , Y w , Z w );

[0011] Wheel out-of-roundness detection: Extract the 3D data of the tread area from the converted 3D laser data of the wheel; extract the OX w Z W Tread coordinates in the plane; according to the general equation of an ellipse A·X w 2 +B·X w ·Z W +C·Z w 2 +E·X w +F·Z W +1=0, the wheel tread is modeled according to the ellipse model, and the center and the lengths of the major and minor axes of the fitted ellipse are obtained. The lengths of the major and minor axes can determine the size of the wheel, and the center of the ellipse can determine the position of the wheel axle. The fitted tread profile data is convenient for subsequent wheel out-of-roundness detection.

[0012] Furthermore, the wheel out-of-roundness information can be calculated based on the tread profile data: first, the starting coordinates are marked according to the wheel tread profile, and all diameters on the wheel tread profile are counted according to the center of the ellipse; then the maximum and minimum values ​​of all diameters are calculated, and finally, the maximum value of the diameter is subtracted from the minimum value, and the difference is recorded as the out-of-roundness of the wheel tread.

[0013] Furthermore, the laser center line extraction: the pre-processed laser line image is adjusted and enlarged several times to achieve a sub-pixel precision image; an adaptive edge detection algorithm is used to detect edge points of the enlarged laser line image, and then the edge detection image is morphologically expanded; an adaptive threshold is calculated by statistically analyzing the histogram of the enlarged laser line image; the enlarged laser line image is segmented using the adaptive threshold, and the segmented image is refined; the expanded points are merged through region growing processing, and finally the center line of the laser image is extracted, and the coordinates of the center line are recorded as (x, y).

[0014] Furthermore, for the image acquisition: a plurality of acquisition devices are distributed beside the track to form an acquisition device array, and the acquisition devices include linear array cameras and laser scanners.

[0015] Furthermore, the laser coordinate transformation: the (x, y) and the coordinates in the world coordinate system (x w , Y w , Z w ) is as follows:

[0016]

[0017]

[0018]

[0019] Where, f is the focal length of the linear array camera, (X c , Y c , Z c ) is the coordinate in the line array camera coordinate system, H is the parameter obtained by calibration of the line array camera; the corresponding relationship between the coordinates of the two-dimensional image point and the three-dimensional coordinates of the corresponding target can be known through the above formula.

[0020] Furthermore, in the world coordinate system OX w Y w Z w , the three-dimensional wheel information is corrected and processed to facilitate the stitching and reconstruction of the wheel data: the point cloud data of the non-wheel part is segmented through the point cloud clustering algorithm DNSCN, and the point cloud data with only the wheel information is extracted. The curved wheel is corrected into a regular rectangular wheel through the coordinate mapping method, and the laser line data after coordinate transformation and correction is obtained.

[0021] Furthermore, the preprocessing: using Gaussian filtering to perform noise reduction processing on the original data to obtain a relatively smooth laser line image, and then improving the clarity of the laser line image by contrast stretching, and finally deleting interference data unrelated to the laser line by contour extraction and judgment. The train wheel out-of-roundness detection method based on three-dimensional information of the present invention only needs to use a small number of linear array cameras and 3D laser scanner modules in combination to generate multiple two-dimensional image data and laser scanning data with depth information, extract the tread area of ​​the laser scan, and reconstruct the wheel tread contour with depth information based on the elliptical model. Whether it is the local out-of-roundness of the wheel or the overall out-of-roundness, the method of the present invention can accurately calculate it through the wheel tread contour data, and generate a detection report in a timely and dynamic manner for the customer to review and repair. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is the overall flow chart of the present invention;

[0023] FIG2 is a schematic diagram of the acquisition device of the present invention: (a) is a schematic diagram of data acquisition of the acquisition device; (b) is a schematic diagram of the position layout of multiple acquisition devices;

[0024] Figure 3 It is the original wheel laser line drawing;

[0025] Figure 4 It is a laser line processing diagram;

[0026] Figure 5 It is a three-dimensional coordinate transformation diagram;

[0027] Figure 6 It is the wheel tread profile. DETAILED DESCRIPTION

[0028] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0029] from Figure 1 It can be seen that the train wheel out-of-roundness detection method based on three-dimensional information of the present invention has four main implementation steps, namely data acquisition, laser data processing, laser coordinate transformation and wheel out-of-roundness detection. The specific situation of each step is as follows:

[0030] 1. Data Collection

[0031] The acquisition equipment used in the method of the present invention is installed on both sides of the track at the throat section of the train in and out of the depot, and the number of acquisition equipment to be installed will be selected according to the size of the wheel. The acquisition equipment includes components such as laser scanners, linear array cameras, light sources, transmitters, receivers, and temperature control. In this embodiment, a total of 10 acquisition equipment are installed in the EMU inspection shed. Only 5 equipment are required to be installed on one side of the track to complete the data acquisition of a single wheel. The equipment layout is shown in Figure 2 (b), which shows the equipment layout on the left side of the track. The layout on the right side of the track is the same as the left side of the track. In addition, the acquisition equipment is installed on the side of the track and keeps a safe distance from it, cannot be higher than the height of the rail surface, and the angle between it and the track is about 5 degrees. The distance between each acquisition equipment is about 600 mm. Since the diameter of the EMU wheel is 910 mm, it can be calculated that the wheel circumference is about 2857 mm, and the length collected by a single acquisition equipment is 600 mm, then the total continuous collection length of the 5 acquisition equipment is 3000 mm, and the total length of the data collected by the acquisition equipment is greater than the wheel circumference, indicating that the equipment installation layout meets the wheel collection requirements.

[0032] Due to the fixed imaging characteristics of the wheel image acquisition device, when the wheel passes, each wheel is scanned thousands of times to capture the detailed status of each wheel. Figure 2(a) shows a schematic diagram of the device collecting wheel data. The collected wheel images are numbered in turn according to the layout of the acquisition device and the shooting order, and transmitted to the server via TCP for storage, which is convenient for subsequent data processing and analysis.

[0033] 2. Laser Data Processing

[0034] like Figure 3The figure shows the original laser images of some wheels collected by a certain collection device. Since the obtained data are interfered by light source, noise, etc., which affects the subsequent data processing and measurement accuracy, this embodiment uses Gaussian filtering to reduce the noise of the original data to obtain a smoother laser line image, and then improves the clarity of the laser line image by contrast stretching. Finally, the interference data unrelated to the laser line is deleted by contour extraction and judgment. The preprocessed image is recorded as Img.

[0035] Further extract the center coordinates of the laser line, the specific steps are:

[0036] Adjust Img to 2 times to achieve sub-pixel precision image, denoted as ImgResize;

[0037] Adaptive edge detection algorithm is used to detect edge points of the enlarged laser line image ImgResize, and then morphological dilation processing is performed on the edge detection image. The image after dilation processing is recorded as ImgDilate.

[0038] Count the histogram of ImgResize and calculate the adaptive threshold based on the histogram information;

[0039] Using the adaptive threshold segmentation image ImgResize obtained in step (3), the segmented image is thinned and recorded as ImgThin;

[0040] Perform region growth on ImgThin, process and merge the points of ImgDilate, and finally extract the center line of the laser image, such as Figure 4 As shown, the center line coordinates are (x, y).

[0041] 3. Laser coordinate transformation

[0042] Since the installation positions of the line array camera and the laser scanner are relatively fixed, in this embodiment, the line array camera is first calibrated, and then the laser scanner is calibrated to obtain the position of the laser scanner relative to the line array camera. The data of the laser scanner is reflected through camera imaging, so that the two-dimensional laser image data can be converted into a three-dimensional world coordinate system.

[0043] Specifically, the image point set (x, y) generated by the wheel surface laser line extracted in step 2 is converted into the coordinates (X w , Y w , Z w ), according to the principle of photogrammetry, the relationship between them is as follows:

[0044]

[0045]

[0046] Where f is the focal length of the linear array camera, H is the variable parameter obtained through calibration, (X c , Y c , Z c ) is the coordinate in the camera coordinate system. Since the installation positions of the laser scanner and camera in the acquisition equipment are relatively fixed, the camera is calibrated in advance through the black and white chessboard, and the transformation parameter H of the calibrated camera can be obtained.

[0047] Since the coordinates of the laser line image and the position of the calibration object are known, the laser scanner is calibrated and the laser plane equation p 1 ·X w +p 2 ·Y W +p 3 ·Z w +t=0 is written as P·X=T, where p 1 2 +p 2 2 +p 3 2 =1, P = [p 1 , p 2 , p 3 ],X=[X w Y w Z w ] T , T = -t, and the SVD (Singular Value Decomposition) decomposition method is used to calculate and obtain the parameter P of the calibrated laser scanner.

[0048] Since any laser point on the wheel tread image is the result of the intersection of a laser plane and an object, the position of the wheel tread 3D information in the world coordinate system can be obtained through the above coordinate transformation. That is, once the laser scanner and the linear array camera are calibrated, any laser point on the wheel image can be converted to obtain the position of the wheel tread and wheel rim 3D information in the world coordinate system.

[0049] During the data collection process, the distance between the wheel and the laser scanner changes, that is, the laser line data collected at a close distance is wider, and the laser line data collected at a distance is narrower, so the collected laser line is in an arc shape in the world coordinate system, so it is necessary to w Y w Z w, the three-dimensional wheel tread information is corrected and processed to facilitate the subsequent fitting and reconstruction of the wheel tread data. This embodiment uses the point cloud clustering algorithm DNSCN (Density-Based Spatial Clustering of Applications with Noise) to segment the point cloud data of the non-wheel part, extract the point cloud data with only wheel information, and finally correct the curved wheel into a regular wheel through the coordinate mapping method, such as Figure 5 Shown is the laser line data after coordinate transformation correction.

[0050] 4. Wheel out-of-roundness detection

[0051] The three-dimensional coordinates of the wheel tread laser line in the world coordinate system are obtained through step three. In order to obtain the true information of the tread profile, the following steps are adopted. First, in the world coordinate system, the three-dimensional data of the tread area is extracted from the transformed three-dimensional laser data of the wheel; then, the three-dimensional data of the tread area at 70±5 mm from the side of the wheel rim is extracted in sequence. w Z W Tread coordinates in the plane; finally, the wheel tread is modeled according to the ellipse model according to the general equation of the ellipse. Specifically, the general equation of the ellipse is

[0052] A.X w 2 +B·X w ·Z W +C·Z w 2 +D·X w +E·Z W +1=0

[0053] Write the above formula as

[0054]

[0055] make Then there is

[0056] X w 2 +a·X w ·Z W +b·Z w 2 +c·X w +d·Z W +e=0

[0057] Written in matrix form:

[0058]

[0059] By using the SVD (Singular Value Decomposition) decomposition method to calculate, we can get the values ​​of parameters a, b, c, d, and e, and further get A, B, C, D, and E.

[0060]

[0061]

[0062]

[0063]

[0064] Through the above fitting calculation, the center of the ellipse (x c , z c ), major axis length R 1 and the minor axis length R 2 , where the length of the major and minor axes determines the size of the wheel, and the center of the ellipse determines the position of the wheel axle, such as Figure 6 As shown, it is a comparison between the real data and the fitting data. The tread profile data after fitting is convenient for the subsequent wheel out-of-roundness detection.

[0065] Based on the obtained tread profile data, the wheel out-of-roundness information can be calculated. Specifically:

[0066] (1) Mark the starting coordinates (x) according to the wheel tread contour 1 , z 1 ), according to the ellipse center (x c , z c ), we can get the straight line segment L;

[0067] (2) Calculate the intersection point of the straight line segment L and the other end of the tread profile, denoted as (x 11 , z 11 ), through (x 1 , z 1 ) and (x 11 , z 11 ) to obtain the diameter of the tread profile;

[0068] (3) according to steps (1) and (2), all diameters on the wheel tread profile are counted, and the maximum and minimum values ​​of all diameters are calculated;

[0069] (4) Subtract the minimum diameter from the maximum diameter, and the difference is recorded as the out-of-roundness of the wheel tread.

[0070] The above embodiment discloses a method for detecting the out-of-roundness of a train wheel based on three-dimensional depth information. The method uses a combination of multiple linear array cameras and a 3D laser scanner to generate multiple two-dimensional image data and laser scanning data with depth information, extract the tread area scanned by the laser, and reconstruct the wheel tread profile with depth information based on an elliptical model. Whether it is the local out-of-roundness of the wheel or the overall out-of-roundness, this embodiment can accurately calculate it through the wheel tread profile data, and generate a test report in a timely and dynamic manner for the customer to review and repair.

[0071] The present invention has been disclosed as above in terms of preferred embodiments, but they are not intended to limit the present invention. Anyone familiar with this technology can make various changes or modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims of the present invention.

Claims

1. A train wheel out-of-roundness detection method based on three-dimensional information, characterized in that: The method comprises: Image acquisition and preprocessing to obtain laser line images: multiple acquisition devices are distributed on the trackside to form an acquisition device array, including line array cameras and laser scanners; Laser centerline extraction: extract the centerline of the laser image, with two-dimensional coordinates (x, y): adjust and magnify the preprocessed laser line image several times to achieve sub-pixel precision image; use adaptive edge detection algorithm to detect edge points of the amplified laser line image, and then perform morphological expansion processing on the edge detection image; calculate the adaptive threshold by statistically analyzing the histogram of the amplified laser line image; use the adaptive threshold to segment the amplified laser line image, and perform refinement processing on the segmented image; merge the expanded points through region growing processing, and finally extract the centerline of the laser image, with the centerline coordinates recorded as (x, y); Laser coordinate transformation: The image point set generated by the laser line on the wheel surface, i.e. (x, y), is converted into coordinates in the real world coordinate system. ; In the world coordinate system , correct the three-dimensional wheel information to facilitate the splicing and reconstruction of the wheel data: segment the point cloud data of the non-wheel part through the point cloud clustering algorithm DNSCN, extract the point cloud data with only the wheel information, correct the curved wheel into a regular rectangular wheel through the coordinate mapping method, and obtain the laser line data after coordinate transformation and correction; Wheel out-of-roundness detection: Extract the 3D data of the tread area from the converted 3D laser data of the wheel; extract the 3D data of the tread area at 70±5 mm from the side of the wheel rim in sequence. Tread coordinates in the plane; according to the general equation of an ellipse , the wheel tread is modeled according to the ellipse model, and the center and the length of the major and minor axes of the fitted ellipse are obtained. The length of the major and minor axes can determine the size of the wheel, and the center of the ellipse can determine the position of the wheel axle. The tread profile data after fitting is convenient for subsequent wheel out-of-roundness detection; The wheel out-of-roundness information can be calculated based on the tread profile data: first, the starting coordinates are marked according to the wheel tread profile, and all diameters on the wheel tread profile are counted according to the center of the ellipse; then the maximum and minimum values ​​of all diameters are calculated, and finally the maximum value of the diameter is subtracted from the minimum value, and the difference is recorded as the out-of-roundness of the wheel tread.

2. A train wheel defect detection method based on three-dimensional information as claimed in claim 1, characterized in that: The laser coordinate transformation: the (x, y) coordinates in the world coordinate system The relationship is as follows: ， ， ， Where f is the focal length of the linear array camera, is the coordinate in the line array camera coordinate system, and H is the parameter obtained by calibration of the line array camera. The above formula can be used to know the corresponding relationship between the coordinates of the two-dimensional image point and the three-dimensional coordinates of the corresponding target.

3. The method for detecting out-of-roundness of train wheels based on three-dimensional information as claimed in claim 1, characterized in that: The preprocessing includes: using Gaussian filtering to reduce noise on the original data to obtain a relatively smooth laser line image, then improving the clarity of the laser line image by contrast stretching, and finally deleting interference data unrelated to the laser line by contour extraction and judgment.

Images