Steel rail curvature detection method and system based on machine vision

Abstract

The invention provides a steel rail curvature detection method and system based on machine vision, and the method comprises the steps: synchronously collecting three-dimensional point cloud data and a two-dimensional image of a steel rail; voxel filtering and statistical filtering are carried out on the three-dimensional point cloud data to remove noise, and distortion correction and gray level equalization processing are carried out on a two-dimensional image; processing the two-dimensional image by using a deep learning semantic segmentation algorithm, extracting the edge contour of the steel rail, and generating a three-dimensional contour model of the steel rail in combination with the three-dimensional point cloud data; fitting the three-dimensional contour point cloud of the steel rail by adopting an RANSAC algorithm to obtain a central axis of the steel rail; bending angles of the steel rail in the vertical direction and the tangential direction are calculated, included angles of normal vectors of adjacent sections are calculated, and the torsion degree is determined; bending and torsion angle threshold values are set, when a detection value exceeds the threshold values, sound-light alarm is triggered, real-time communication with a production line PLC system is carried out through an industrial Ethernet, a bending degree value and an abnormal state signal are transmitted to a production control system, and automatic rolling mill parameter adjustment or shutdown maintenance triggering is supported.

Description

Technical Field

[0001] This invention relates to the field of steel rolling production quality inspection technology, and more particularly to a method and system for detecting rail curvature based on machine vision. Background Technology

[0002] In railway construction, the quality of rails directly affects the safe operation and service life of the railway. The curvature of the rails is one of the core indicators for measuring their quality. During the production process, due to the influence of various factors such as rolling technology, equipment precision, and environmental factors, defects such as bending and torsion may occur. These defects not only cause the rails to rub against guides and side plates, resulting in surface quality defects, but also significantly affect the control of end specifications, and may even lead to railway operation safety accidents. Therefore, accurate detection of rail curvature is a crucial link in ensuring the quality of rail production.

[0003] Currently, rail curvature detection technologies mainly include manual visual inspection, offline inspection, and some online inspection methods based on single sensors. Manual visual inspection relies on the experience of the inspectors; although it is low-cost, its efficiency is low, and it is greatly affected by human factors, making it difficult to guarantee accuracy. Offline inspection is usually performed after rail production is completed. While it can provide relatively accurate results, it cannot provide real-time feedback on problems during the production process, making it difficult to dynamically adjust the production process. Furthermore, existing online inspection technologies mostly employ single 2D visual inspection or contact measurement. 2D visual inspection can only acquire planar image information and cannot identify the three-dimensional curvature and torsion state of the rail; contact measurement is greatly affected by environmental factors such as high temperature and vibration, has high equipment maintenance costs, and its detection accuracy is easily affected by interference.

[0004] Existing rail curvature detection technologies have several shortcomings. First, their efficiency is low, unable to adapt to the high-speed rail movement (4-5 m / s) during production, leading to a large number of defective products. Second, they lack real-time capability, only allowing for post-production inspection after rail tapping, failing to detect curvature anomalies during production and hindering online monitoring and real-time adjustments to the tapping status. Furthermore, their accuracy is insufficient; manual visual inspection relies on experience, and offline inspection lacks real-time data feedback, making it impossible to accurately quantify the vertical and tangential curvature angles of the rail. Finally, their automation level is low; inspection results cannot be linked with the production control system in real time, making automatic adjustments to mill parameters difficult. Therefore, there is an urgent need for an automated inspection technology capable of real-time, online, and accurate rail curvature detection, and integrated with the production system. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a machine vision-based method and system for detecting rail curvature. By integrating high-frame-rate point cloud reconstruction with deep learning algorithms, this invention enables online measurement of multidimensional bending / torsional defects in rails during the rolling process and adaptive control of rolling mill parameters.

[0006] The technical means employed in this invention are as follows:

[0007] A machine vision-based method for detecting rail curvature includes:

[0008] S1. Install 3D intelligent sensing terminals and high-definition industrial cameras symmetrically on both sides of the roller conveyor to cover the detection area of ​​0-5m long at the end of the rail.

[0009] S2. When the rail enters the detection area, the sensor is triggered to start the equipment and simultaneously collect the three-dimensional point cloud data and two-dimensional image of the rail.

[0010] S3. Perform voxel filtering and statistical filtering on the acquired 3D point cloud data to remove noise, and perform distortion correction and grayscale equalization on the 2D image.

[0011] S4. Use deep learning semantic segmentation algorithm to process the preprocessed two-dimensional image, extract the rail edge contour, and combine it with three-dimensional point cloud data to generate a three-dimensional contour model of the rail.

[0012] S5. The RANSAC algorithm is used to fit the three-dimensional contour point cloud of the rail to obtain the center axis of the rail, which is used as the benchmark for the curvature calculation.

[0013] S6. Based on geometric principles, the bending angles of the rail in the vertical and tangential directions are calculated using trigonometric functions, and the degree of torsion is determined by calculating the angle between the normal vectors of adjacent sections.

[0014] S7. Set the bending angle threshold and torsion angle threshold. When the detected value exceeds the threshold, trigger an audible and visual alarm and communicate with the production line PLC system in real time via industrial Ethernet to transmit the bending value and abnormal status signal to the production control system. This supports automatic adjustment of mill parameters or triggering shutdown for maintenance.

[0015] Furthermore, in step S1, the 3D intelligent sensing terminal has a ranging accuracy of <±1mm, a field of view of 80°×80°, a measurement frequency of 10Hz, and a protection level of IP67; the high-definition industrial camera has a resolution of 3840×2160, 8 million pixels, and supports PoE power supply and H.265 encoding.

[0016] Furthermore, in step S2, the acquisition of the three-dimensional point cloud data and the two-dimensional image of the rail are kept in spatiotemporal synchronization, with a synchronization error of ≤1ms.

[0017] Furthermore, in step S4, the deep learning semantic segmentation algorithm adopts a fusion architecture of CNN and Transformer, and the training data includes multi-angle samples of rails under high temperature and vibration environments.

[0018] Further, step S6 specifically includes:

[0019] The formula for calculating the vertical bending angle is as follows:

[0020] θ1 = arctan(Δh / L)

[0021] Where θ1 represents the vertical bending angle, Δh represents the vertical height difference, and L represents the detection length;

[0022] The formula for calculating the tangential bending angle is as follows:

[0023] θ2=arctan(Δw / L)

[0024] Where θ2 represents the tangential bending angle and Δw represents the lateral deviation;

[0025] Calculate the angle between the normal vectors of adjacent sections to determine the degree of torsion. The calculation formula is as follows:

[0026]

[0027] Where θ represents the angle between the normal vectors of two adjacent cross sections. and

[0028] These represent the normal vectors of two adjacent cross sections.

[0029] Furthermore, in step S7, the set threshold for vertical bending angle is ≥3°, the threshold for tangential bending angle is ≥2°, and the threshold for torsion angle is ±1.5°.

[0030] Furthermore, step S7 also includes data storage, storing raw data, detection results and adjustment records, with a storage period of ≥3 months.

[0031] Furthermore, the rail curvature detection method is applicable to the detection of rails used in high-speed railways with a speed of ≥350km / h, with a curvature control accuracy of ≤1mm / m. It can be adapted to the detection of flange inclination or web curvature of H-beams and I-beams through parameterized configuration.

[0032] This invention also provides a machine vision-based rail curvature detection system based on the aforementioned machine vision-based rail curvature detection method, comprising: a multimodal data acquisition module, a data processing module, a feature extraction and analysis module, a curvature calculation module, and a closed-loop control module, wherein:

[0033] The multimodal data acquisition module includes 3D intelligent sensing terminals and high-definition industrial cameras symmetrically installed on both sides of the roller conveyor, covering a detection area of ​​0-5m in length at the end of the rail; when the rail enters the detection area, the sensor is triggered to start the device, and the three-dimensional point cloud data and two-dimensional image of the rail are collected simultaneously.

[0034] The data processing module performs voxel filtering and statistical filtering on the acquired 3D point cloud data to remove noise, and performs distortion correction and grayscale equalization on the 2D image.

[0035] The feature extraction and analysis module uses a deep learning semantic segmentation algorithm to process the preprocessed two-dimensional image, extract the rail edge contour, and combine it with three-dimensional point cloud data to generate a three-dimensional contour model of the rail; and uses the RANSAC algorithm to fit the three-dimensional contour point cloud of the rail to obtain the central axis of the rail as the benchmark for curvature calculation.

[0036] The curvature calculation module, based on geometric principles, calculates the bending angle of the rail in the vertical and tangential directions using trigonometric functions, and determines the degree of torsion by calculating the angle between the normal vectors of adjacent sections.

[0037] The closed-loop control module sets bending angle thresholds and torsion angle thresholds. When the detected value exceeds the threshold, it triggers an audible and visual alarm and communicates with the production line PLC system in real time via industrial Ethernet to transmit the bending value and abnormal status signal to the production control system. This supports automatic adjustment of mill parameters or triggering shutdown for maintenance.

[0038] Furthermore, the communication protocol between the closed-loop control module and the production line PLC system is Industrial Ethernet, which supports the real-time transmission of curvature values, abnormal signals, and adjustment commands.

[0039] Compared with the prior art, the present invention has the following advantages:

[0040] 1. This invention achieves high-precision, real-time online detection and automated quality control of rail curvature by deeply integrating multimodal sensing devices with deep learning algorithms and real-time linkage with production control systems, resulting in a significant improvement in detection efficiency, accuracy, and automation level.

[0041] 2. This invention supports dynamic detection of high-speed moving rails at speeds of 4-5 m / s. The detection frequency is synchronized with the production rhythm, realizing real-time operation of the entire process from detection to feedback adjustment. This avoids the lag of offline detection and reduces the rate of missed detection of defective products.

[0042] 3. This invention achieves simultaneous acquisition and processing of three-dimensional point cloud data and two-dimensional images of high-speed moving rails through multi-modal data acquisition and fusion technology. It can simultaneously detect the vertical and tangential bending angles and torsional state of the rails, achieving a multi-dimensional and precise detection effect with a detection accuracy of up to 0.5 mm. It quantifies the bending degree value, provides a precise basis for adjusting rolling mill parameters, effectively controls the specifications of the rail ends, and improves the consistency of product quality.

[0043] 3. This invention uses deep learning algorithms to preprocess data, extract features, and calculate curvature, achieving accurate extraction of the rail profile and precise fitting of the center axis, thereby accurately quantifying the curvature value and providing a precise basis for adjusting rolling mill parameters. This achieves the technical effect of improving quality stability and reduces the rate of excessive curvature within 5m of the end of the heavy rail by 60%.

[0044] 4. This invention achieves automatic analysis of test results and intelligent adjustment of production parameters through real-time linkage control technology with the production line PLC system, reducing errors from manual visual inspection and experience-based adjustments, and improving the automation level and standardization of the steel rolling process.

[0045] 5. Through high-protection-level equipment design and optimized algorithm robustness, this invention ensures the long-term stable operation of the detection system in the harsh environment of high temperature, dust, and strong vibration in the hot rolling production line. The equipment protection level reaches IP67, and the operating temperature range is -40℃ to 80℃, achieving a strong environmental adaptability and ensuring the reliability of the system.

[0046] 6. This invention enables the long-term preservation of historical test data and images through data storage and traceability functions, supporting rapid traceability of quality issues and continuous optimization of production processes. The storage period is no less than 3 months, achieving the technical effect of data traceability, helping enterprises improve the accuracy and robustness of the testing system and optimize production processes.

[0047] In summary, this invention, through the deep integration of machine vision technology and deep learning algorithms, constructs an efficient and accurate rail curvature detection system, which significantly improves the automation level and control accuracy of rail production quality inspection. It has significant industrial application value and broad prospects for promotion. Based on the above reasons, this invention can be widely promoted in fields such as steel rolling production quality inspection. Attached Figure Description

[0048] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a flowchart of the method of the present invention.

[0050] Figure 2 This is a schematic diagram of the detection rail bending in the horizontal direction provided in an embodiment of the present invention.

[0051] Figure 3 This is a schematic diagram of the detection rail bending in the vertical direction provided in an embodiment of the present invention.

[0052] Figure 4 This is a schematic diagram of the detection rail bending in the torsional direction provided in an embodiment of the present invention. Detailed Implementation

[0053] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0054] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products or devices.

[0055] like Figure 1 As shown, this invention provides a machine vision-based method for detecting rail curvature, comprising:

[0056] S1. Install 3D intelligent sensing terminals and high-definition industrial cameras symmetrically on both sides of the roller conveyor to cover the detection area of ​​0-5m long at the end of the rail.

[0057] S2. When the rail enters the detection area, the sensor is triggered to start the equipment, and the three-dimensional point cloud data and two-dimensional image of the rail are collected simultaneously to ensure the spatiotemporal consistency of the data.

[0058] S3. Perform voxel filtering and statistical filtering on the acquired 3D point cloud data to remove noise, and perform distortion correction and grayscale equalization on the 2D image to improve data quality.

[0059] S4. Use deep learning semantic segmentation algorithm to process the preprocessed two-dimensional image, extract the rail edge contour, and combine it with three-dimensional point cloud data to generate a three-dimensional contour model of the rail.

[0060] S5. The RANSAC algorithm is used to fit the three-dimensional contour point cloud of the rail to obtain the center axis of the rail, which is used as the benchmark for the curvature calculation.

[0061] S6. Based on geometric principles, the bending angles of the rail in the vertical and tangential directions are calculated using trigonometric functions, and the degree of torsion is determined by calculating the angle between the normal vectors of adjacent sections.

[0062] S7. Set the bending angle threshold and torsion angle threshold. When the detected value exceeds the threshold, trigger an audible and visual alarm and communicate with the production line PLC system in real time via industrial Ethernet to transmit the bending value and abnormal status signal to the production control system. This supports automatic adjustment of mill parameters or triggering shutdown for maintenance.

[0063] In this embodiment, multimodal sensing devices are deployed on both sides of the roller conveyor to collect 3D point cloud data and 2D images of the rail in real time. Deep learning algorithms are used to process the data, extract the rail profile, fit the central axis, calculate the vertical and tangential bending angles, and detect the torsional state. The detection results are transmitted to the production control system in real time, enabling automatic alarms for abnormal states and coordinated adjustments to production parameters.

[0064] In a specific implementation, as a preferred embodiment of the present invention, in step S1, two 3D intelligent sensing terminals and two high-definition industrial cameras are symmetrically installed on both sides of the roller conveyor, wherein:

[0065] The 3D intelligent sensing terminal has a ranging accuracy of <±1mm, a field of view of 80°×80°, a measurement frequency of 10Hz, and a protection rating of IP67.

[0066] The high-definition industrial camera has a resolution of 3840×2160, 8 megapixels, and supports PoE power supply and H.265 encoding.

[0067] In a specific implementation, as a preferred embodiment of the present invention, in step S2, the acquisition of the three-dimensional point cloud data and the two-dimensional image of the rail is kept in spatiotemporal synchronization, with a synchronization error of ≤1ms.

[0068] In a specific implementation, as a preferred embodiment of the present invention, in step S4, the deep learning semantic segmentation algorithm adopts a fusion architecture of CNN and Transformer, and the training data includes multi-angle samples of rails under high temperature and vibration environments.

[0069] In a specific implementation, as a preferred embodiment of the present invention, step S6 specifically includes:

[0070] The formula for calculating the vertical bending angle is as follows:

[0071] θ1 = arctan(Δh / L)

[0072] Where θ1 represents the vertical bending angle, Δh represents the vertical height difference, and L represents the detection length, typically a distance from the end of the rail, such as 0-5m. In this embodiment, the vertical bending angle reflects the degree to which the rail deviates from an ideal straight line in the vertical direction. Based on geometric principles, the bending of the rail is simplified to the relationship between the hypotenuse and the base of a right triangle, and the angle is calculated using trigonometric functions.

[0073] The formula for calculating the tangential bending angle is as follows:

[0074] θ2=arctan(Δw / L)

[0075] Where θ2 represents the tangential bending angle, Δw represents the lateral deviation of the rail end point from the central axis in the horizontal direction within the detection interval, and L represents the detection length, which is generally a distance from the end of the rail, such as 0-5m. In this embodiment, the horizontal bending angle is used to measure the degree of bending of the rail in the horizontal direction. Based on geometric relationships, the horizontal deviation and the detection length are constructed into a right triangle, and the angle is calculated using the arctangent function.

[0076] Calculate the angle between the normal vectors of adjacent sections to determine the degree of torsion. The calculation formula is as follows:

[0077]

[0078] Where θ represents the angle between the normal vectors of two adjacent cross sections. and These represent the normal vectors of two adjacent cross-sections. In this embodiment, the torsional bending angle is used to determine whether the rail has twisted around its own axis. The degree of twist is determined by calculating the angle between the normal vectors of adjacent cross-sections. The normal vector describes the direction of the cross-section; when the rail twists, the direction of the normal vectors of adjacent cross-sections changes, and the magnitude of the angle reflects the degree of twist. By analyzing the difference in normal vectors of the point clouds of adjacent cross-sections, when the change in the angle between the normal vectors exceeds a preset threshold, it is determined to be a twisting state. Figure 2-4 The figures shown are schematic diagrams for detecting the bending of the rail in the horizontal, vertical, and torsional directions.

[0079] In a specific implementation, as a preferred embodiment of the present invention, in step S7, the set vertical bending angle threshold is ≥3°, the tangential bending angle threshold is ≥2°, and the torsion angle threshold is ±1.5°.

[0080] In a specific implementation, as a preferred embodiment of the present invention, step S7 further includes data storage, storing raw data, detection results and adjustment records, with a storage period of ≥3 months.

[0081] In specific implementation, as a preferred embodiment of the present invention, the rail curvature detection method is applicable to the detection of rails used in high-speed railways with a speed of ≥350km / h, with a curvature control accuracy of ≤1mm / m. It is adapted to the detection of flange inclination or web curvature of H-beams and I-beams through parameterized configuration.

[0082] This invention also provides a machine vision-based rail curvature detection system based on the aforementioned machine vision-based rail curvature detection method, comprising: a multimodal data acquisition module, a data processing module, a feature extraction and analysis module, a curvature calculation module, and a closed-loop control module, wherein:

[0083] The multimodal data acquisition module includes 3D intelligent sensing terminals and high-definition industrial cameras symmetrically installed on both sides of the roller conveyor, covering a detection area of ​​0-5m in length at the end of the rail; when the rail enters the detection area, the sensor is triggered to start the device, and the three-dimensional point cloud data and two-dimensional image of the rail are collected simultaneously.

[0084] The data processing module performs voxel filtering and statistical filtering on the acquired 3D point cloud data to remove noise, and performs distortion correction and grayscale equalization on the 2D image.

[0085] The feature extraction and analysis module uses a deep learning semantic segmentation algorithm to process the preprocessed two-dimensional image, extract the rail edge contour, and combine it with three-dimensional point cloud data to generate a three-dimensional contour model of the rail; and uses the RANSAC algorithm to fit the three-dimensional contour point cloud of the rail to obtain the central axis of the rail as the benchmark for curvature calculation.

[0086] The curvature calculation module, based on geometric principles, calculates the bending angle of the rail in the vertical and tangential directions using trigonometric functions, and determines the degree of torsion by calculating the angle between the normal vectors of adjacent sections.

[0087] The closed-loop control module sets bending angle thresholds and torsion angle thresholds. When the detected value exceeds the threshold, it triggers an audible and visual alarm and communicates with the production line PLC system in real time via industrial Ethernet to transmit the bending value and abnormal status signal to the production control system. This supports automatic adjustment of mill parameters or triggering shutdown for maintenance.

[0088] The embodiments of the present invention are described simply because they correspond to those in the embodiments above. For any similarities, please refer to the descriptions in the embodiments above, which will not be elaborated here.

[0089] In a specific implementation, as a preferred embodiment of the present invention, the communication protocol between the closed-loop control module and the production line PLC system is industrial Ethernet, which supports real-time transmission of curvature values, abnormal signals and adjustment commands.

[0090] Example

[0091] Taking the testing of 60kg / m rails as an example:

[0092] Equipment Installation and Initialization: 3D intelligent sensing terminals and high-definition industrial cameras are symmetrically installed on both sides of the mill exit roller conveyor at a height of 1.5m above the rail surface, ensuring an unobstructed detection area. The equipment is connected to the industrial control computer via a gigabit switch. Initialization and calibration are performed, importing the rail standard model parameters and setting the curvature detection threshold.

[0093] Data Acquisition and Processing: When the rail enters the detection area at a speed of 4.5 m / s, the sensor triggers the equipment to acquire data. The 3D terminal acquires point cloud data at a frequency of 10 Hz, and the camera captures images at a frame rate of 20 fps. Voxel filtering (5mm voxel size) is performed on the acquired 3D point cloud data. 3 Remove discrete noise; perform distortion correction on 2D images; and use histogram equalization to enhance edge features.

[0094] Curvature Calculation and State Recognition: A deep learning model is used to perform semantic segmentation on 2D images to extract the rail contour, and a three-dimensional centerline is fitted using 3D point cloud data. The vertical height difference Δh = 3mm between the endpoint and the axis within the detection interval (5m) and the tangential lateral deviation Δw = 2mm are calculated. The vertical curvature angle θ1 = arctan(3 / 5000) ≈ 0.034° and the tangential curvature angle θ2 = arctan(2 / 5000) ≈ 0.023° are obtained, both of which do not exceed the threshold and are judged to be in a normal state.

[0095] Anomaly Feedback and System Linkage: If a vertical bending angle θ1 = 3.5° is detected (exceeding the threshold of 3°), the system immediately triggers an audible and visual alarm and sends an adjustment signal (signal code 0x01) to the PLC. Upon receiving the signal, the PLC automatically reduces the vertical roller pressure by 0.5mm. Operators can view the real-time detection results and the change in bending angle after adjustment through the monitoring interface. After confirming the adjustment effect, the system stores the detection data and adjustment records in the database.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting rail curvature based on machine vision, characterized in that, include: S1. Install 3D intelligent sensing terminals and high-definition industrial cameras symmetrically on both sides of the roller conveyor to cover the detection area of ​​0-5m long at the end of the rail. S2. When the rail enters the detection area, the sensor is triggered to start the equipment and simultaneously collect the three-dimensional point cloud data and two-dimensional image of the rail. S3. Perform voxel filtering and statistical filtering on the acquired 3D point cloud data to remove noise, and perform distortion correction and grayscale equalization on the 2D image. S4. Use deep learning semantic segmentation algorithm to process the preprocessed two-dimensional image, extract the rail edge contour, and combine it with three-dimensional point cloud data to generate a three-dimensional contour model of the rail. S5. The RANSAC algorithm is used to fit the three-dimensional contour point cloud of the rail to obtain the center axis of the rail, which is used as the benchmark for the curvature calculation. S6. Based on geometric principles, the bending angles of the rail in the vertical and tangential directions are calculated using trigonometric functions, and the degree of torsion is determined by calculating the angle between the normal vectors of adjacent sections. S7. Set the bending angle threshold and torsion angle threshold. When the detected value exceeds the threshold, trigger an audible and visual alarm and communicate with the production line PLC system in real time via industrial Ethernet to transmit the bending value and abnormal status signal to the production control system. This supports automatic adjustment of mill parameters or triggering shutdown for maintenance.

2. The method for detecting rail curvature based on machine vision according to claim 1, characterized in that, In step S1, the 3D intelligent sensing terminal has a ranging accuracy of <±1mm, a field of view of 80°×80°, a measurement frequency of 10Hz, and a protection level of IP67; the high-definition industrial camera has a resolution of 3840×2160, 8 million pixels, and supports PoE power supply and H.265 encoding.

3. The method for detecting rail curvature based on machine vision according to claim 1, characterized in that, In step S2, the acquisition of the three-dimensional point cloud data and the two-dimensional image of the rail are kept in spatiotemporal synchronization, with a synchronization error of ≤1ms.

4. The method for detecting rail curvature based on machine vision according to claim 1, characterized in that, In step S4, the deep learning semantic segmentation algorithm adopts a fusion architecture of CNN and Transformer, and the training data includes multi-angle samples of rails under high temperature and vibration environments.

5. The method for detecting rail curvature based on machine vision according to claim 1, characterized in that, Step S6 specifically includes: The formula for calculating the vertical bending angle is as follows: θ1 = arctan(Δh / L) Where θ1 represents the vertical bending angle, Δh represents the vertical height difference, and L represents the detection length; The formula for calculating the tangential bending angle is as follows: θ2=arctan(Δw / L) Where θ2 represents the tangential bending angle and Δw represents the lateral deviation; Calculate the angle between the normal vectors of adjacent sections to determine the degree of torsion. The calculation formula is as follows: Where θ represents the angle between the normal vectors of two adjacent cross sections. and These represent the normal vectors of two adjacent cross sections.

6. The method for detecting rail curvature based on machine vision according to claim 1, characterized in that, In step S7, the set threshold for vertical bending angle is ≥3°, the threshold for tangential bending angle is ≥2°, and the threshold for torsion angle is ±1.5°.

7. The method for detecting rail curvature based on machine vision according to claim 1, characterized in that, Step S7 also includes data storage, storing raw data, test results and adjustment records, with a storage period of ≥3 months.

8. The method for detecting rail curvature based on machine vision according to claim 1, characterized in that, The rail curvature detection method is applicable to the detection of rails used in high-speed railways with a speed of ≥350km / h, with a curvature control accuracy of ≤1mm / m. It can be adapted to the detection of flange inclination or web curvature of H-beams and I-beams through parameterized configuration.

9. A rail curvature detection system based on machine vision, implemented according to the rail curvature detection method based on any one of claims 1-8, characterized in that, include: The system comprises a multimodal data acquisition module, a data processing module, a feature extraction and analysis module, a curvature calculation module, and a closed-loop control module, among which: The multimodal data acquisition module includes 3D intelligent sensing terminals and high-definition industrial cameras symmetrically installed on both sides of the roller conveyor, covering a detection area of ​​0-5m in length at the end of the rail; when the rail enters the detection area, the sensor is triggered to start the device, and the three-dimensional point cloud data and two-dimensional image of the rail are collected simultaneously. The data processing module performs voxel filtering and statistical filtering on the acquired 3D point cloud data to remove noise, and performs distortion correction and grayscale equalization on the 2D image. The feature extraction and analysis module uses a deep learning semantic segmentation algorithm to process the preprocessed two-dimensional image, extract the rail edge contour, and combine it with three-dimensional point cloud data to generate a three-dimensional contour model of the rail; and uses the RANSAC algorithm to fit the three-dimensional contour point cloud of the rail to obtain the central axis of the rail as the benchmark for curvature calculation. The curvature calculation module, based on geometric principles, calculates the bending angle of the rail in the vertical and tangential directions using trigonometric functions, and determines the degree of torsion by calculating the angle between the normal vectors of adjacent sections. The closed-loop control module sets bending angle thresholds and torsion angle thresholds. When the detected value exceeds the threshold, it triggers an audible and visual alarm and communicates with the production line PLC system in real time via industrial Ethernet to transmit the bending value and abnormal status signal to the production control system. This supports automatic adjustment of mill parameters or triggering shutdown for maintenance.

10. The rail curvature detection system based on machine vision according to claim 9, characterized in that, The closed-loop control module communicates with the production line PLC system via an industrial Ethernet protocol, supporting real-time transmission of curvature values, abnormal signals, and adjustment commands.

Images