Sub-pixel level displacement monitoring method and monitoring system for rail expansion joint

Abstract

The application discloses a kind of steel rail expansion regulator subpixel level displacement monitoring method and monitoring system, the method includes the following steps: S1, to steel rail expansion regulator carries out multimodal data acquisition;S2, based on the gradient direction consistency constraint of target edge carries out subpixel level edge positioning;S3, using adaptive weight fusion algorithm carries out multispectral collaborative imaging;S4, temperature data is collected in real time by temperature sensor, and systematic deviation is eliminated by using thermal expansion displacement compensation model;S5, the data in S2, S3 and S4 are fused with multimodal data fusion.The application constructs high-precision steel rail displacement monitoring system by innovative fusion multispectral imaging, subpixel analysis and physical compensation mechanism, forms anti-interference, high-reliability monitoring output;Significantly improve the safety guarantee capability of high-speed railway seamless line, effectively improve the monitoring precision, reduce the false alarm rate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a rail expansion joint sub-pixel level displacement monitoring method and a monitoring system, which is especially suitable for real-time diagnosis of the safety state of the seamless track in the high-speed railway bridge section, and can effectively cope with the coupling effect of train impact load and temperature stress. BACKGROUND

[0002] The relative displacement amount of the point rail-base rail of the rail expansion joint is the core parameter for evaluating the safety state of the seamless track, and the monitoring accuracy is directly related to the safety of high-speed railway operation. Industry specifications clearly require that when the displacement amount exceeds the safety threshold, an emergency warning must be triggered. However, due to the coupling of environmental interference and physical effects, the false positive rate is high. The traditional monitoring method has three defects: (1) the mechanical displacement sensor is affected by train vibration, and it is difficult to meet the sub-millimeter level precision requirement; (2) the signal-to-noise ratio of optical equipment deteriorates in bad weather such as rain and fog, and the feature recognition rate decreases significantly; (3) the existing method ignores the physical mechanism of rail thermal expansion, resulting in systematic deviation of the data. The current machine vision monitoring scheme still faces fundamental limitations: the algorithm based on pixel-level edge detection is limited by the theoretical limit of spatial resolution, and the displacement measurement error is ≥0.3mm; at the same time, the visible light imaging has a feature recognition rate of less than 60% in low light, strong backlight or rain and snow interference. Engineering practice shows that the above defects lead to an unacceptably high false negative rate in extreme weather. SUMMARY

[0003] The present application discloses a rail expansion joint sub-pixel level displacement monitoring method and a monitoring system to solve the problems in the prior art. The present application first deploys a composite target group containing geometric coding and near-infrared light sources, synchronously collects images using a visible light-near-infrared dual-spectrum camera, and overcomes rain, fog and strong light interference through a signal-to-noise ratio adaptive weight fusion algorithm; then realizes 0.1 pixel level edge positioning based on gradient direction consistency constraint and bicubic spline interpolation, breaking through the traditional visual measurement precision limit; at the same time, real-time data is collected through a temperature sensor, and the thermal expansion equation is applied to eliminate systematic deviation; finally, an adaptive Kalman filter is used to fuse multi-source displacement data to form an anti-interference and high-reliability monitoring output, effectively improving the monitoring accuracy and reducing the false positive rate.

[0004] The present application is realized by the following technical solutions:

[0005] The present application first provides a rail expansion joint sub-pixel level displacement monitoring method, including the following steps:

[0006] S1, multi-modal data acquisition of the rail expansion joint is performed;

[0007] S2, sub-pixel level edge positioning based on gradient direction consistency constraint of the target edge;

[0008] S3, using an adaptive weight fusion algorithm for multispectral collaborative imaging;

[0009] S4, collecting temperature data in real time through a temperature sensor, and eliminating systematic deviation by applying a thermal expansion displacement compensation model;

[0010] S5, multi-modal data fusion of the data in S2, S3 and S4.

[0011] As a further solution, the specific method of S1 is:

[0012] A composite target group is arranged on the side of the switch rail and the basic rail, and a high-contrast geometric code and a near-infrared light unit are arranged on the surface of the switch rail and the basic rail; the images of the switch rail and the basic rail are synchronously collected by a visible light-near infrared dual-band camera, and the rail temperature data of the switch rail and the basic rail are collected in real time by a temperature sensor.

[0013] As a further solution, the target in S2 is a chessboard target, the chessboard target is a PVC coated with a nano anti-fouling coating on the surface, the coating can prevent glare diffuse reflection, thereby reducing the interference on gradient calculation, and the chessboard grid is 4 rows x 9 columns, wherein the white area reflectivity is 92%, and the black area reflectivity is 8%.

[0014] As a further solution, the specific method of S2 is:

[0015] S21, calculating the visible light image gradient field :

[0016] Firstly, the visible light image is converted into a gray-scale image, which effectively simplifies the calculation and retains the brightness information of the image;

[0017] Secondly, the intensity change rate of each pixel in the image is calculated as a gradient vector using the Sobel operator, including the horizontal gradient formed by vertical edge convolution and the vertical gradient formed by horizontal edge convolution, representing the gradient component of each pixel, and the gradient amplitude is calculated as follows by substituting the gray-scale image coordinates

[0018] ;

[0019] wherein the gray-scale image pixel coordinates are ;

[0020] S22, calculating the visible light image direction angle :

[0021] The direction angle represents the direction of gradient change, and is calculated as follows:

[0022] ;

[0023] where the atan2 function is an extension of the arctangent function that outputs angles between -π and π radians.

[0024] S23, the gradient direction standard deviation minimum point is solved in the edge neighborhood to accurately locate the target edge:

[0025] ;

[0026] where, denotes the coordinates of the target edge point, i.e., the coordinates of the gradient direction standard deviation minimum point to be solved; denotes the gradient direction at the point , usually measured in radians or degrees; denotes the average gradient direction within the edge neighborhood Ω;

[0027] S24, precision cycle verification is performed through bicubic spline interpolation to improve the pixel positioning accuracy after visible light image processing, with a precision of 0.1 pixel level;

[0028] ;

[0029] where, denotes the position increment after improving the precision, denotes the original position estimate, denotes the gray value function of the visible light image, denotes the horizontal and vertical pixel coordinates of the visible light image, respectively.

[0030] As a further scheme, the specific method of S3 is:

[0031] For the visible light and near-infrared images collected by the visible light-near-infrared dual-band camera described in S1, an adaptive weight fusion algorithm is used:

[0032] ;

[0033] where, denotes the fused image intensity value,

[0034] the weight and the weight are dynamically adjusted according to the real-time signal-to-noise ratio:

[0035] ,

[0036] denotes the intensity value of the visible light, denotes the intensity value of the near-infrared image.

[0037] then ; wherein, represents the near-infrared compensation quantity after weight fusion.

[0038] As a further scheme, the specific method of S4 is: obtaining the rail temperature obtained by the temperature sensor of S1 and the visible light displacement value obtained by S2 A temperature-displacement coupling correction model is established to eliminate the influence of ambient temperature:

[0039] ;

[0040] In the formula, represents the temperature displacement change amount, represents the actual measured displacement value, is the rail linear expansion coefficient, is the reference length of the expansion device, which is obtained by field calibration by a total station, Real-time acquisition by a temperature sensor.

[0041] As a further scheme, the specific method of S5 is:

[0042] The visible light displacement value obtained by S2, the near-infrared compensation quantity obtained by S3, and the temperature correction quantity obtained by S4 are input into an adaptive Kalman filter for multi-modal data fusion:

[0043] ;

[0044] wherein, represents the visible light displacement value with high precision, represents the near-infrared compensation quantity after weight fusion, represents the temperature compensation quantity, is the Kalman gain, which determines the weight of the displacement value and the compensation quantity value in state updating. The Kalman gain considers a certain size of sliding window in the fusion process, and the Kalman gain Error variance matrix is calculated based on the historical residual data of each sensor in the last 30 groups Update, the period is 10 seconds, and the formula is as follows:

[0045] ;

[0046] wherein, H represents the state transition matrix of the Kalman filter (the state components are fixed and unchanged), ; respectively corresponding to the fusion weights of the visible light displacement value, the near-infrared compensation quantity, and the temperature correction quantity.

[0047] The application also provides a rail expansion regulator sub-pixel level displacement monitoring system, which applies the rail expansion regulator sub-pixel level displacement monitoring method, and comprises:

[0048] Composite target group: titanium alloy substrate surface etching ring code and embedded 850nm near-infrared LED, installed on the side of the frog and the basic rail;

[0049] Multispectral imaging unit: visible light + near-infrared dual-sensor camera, used for real-time image acquisition of the frog and the basic rail and data transmission to the edge computing terminal;

[0050] Temperature compensation module: used for real-time acquisition of the temperature of the frog and the basic rail and data transmission to the edge computing terminal;

[0051] Edge computing terminal: data processing through the built-in sub-pixel solving algorithm, and output of the calculation result;

[0052] Safety warning platform: transmission of the calculation result of the edge computing terminal to the BIM model through the 5G private network, used for realizing three-level alarm.

[0053] As a further scheme, the three-level alarm includes:

[0054] First-level alarm: triggering of the audible and light warning when the displacement exceeds the first-level warning threshold;

[0055] Second-level alarm: sending of the short message to the maintenance personnel when the displacement exceeds the second-level warning threshold;

[0056] Third-level alarm: automatic triggering of the line speed reduction instruction when the displacement exceeds the third-level warning threshold.

[0057] As a further scheme, the edge computing terminal integrates the NVIDIA Jetson module, and the CUDA-accelerated sub-pixel solving optimization library is built-in.

[0058] The characteristics and beneficial effects of the present application are:

[0059] (1) The present application creatively integrates multispectral imaging, sub-pixel analysis and physical compensation mechanism, and constructs a high-precision rail displacement monitoring system: first, the composite target group containing geometric coding and near-infrared light source is deployed, the visible light-near-infrared dual-spectrum camera is used to synchronously acquire images, the signal-to-noise ratio adaptive weight fusion algorithm is used to overcome rain, fog and strong light interference; then, based on the gradient direction consistency constraint and the bicubic spline interpolation, the 0.1 pixel level edge positioning is realized, the traditional visual measurement precision limit is broken through; at the same time, the temperature sensor is used to acquire data in real time, and the thermal expansion equation is used to eliminate systematic deviation; finally, the adaptive Kalman filter is used to fuse multi-source displacement data, and the anti-interference and high-reliability monitoring output is formed; the present application creatively integrates optical perception, physical modeling and edge computing, and significantly improves the safety guarantee capability of the high-speed railway seamless line. BRIEF DESCRIPTION OF DRAWINGS

[0060] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0061] Figure 1 The system deployment schematic diagram described in the embodiments of the present application;

[0062] Figure 2 The signal-to-noise ratio weight dynamic adjustment image described in the embodiments of the present application;

[0063] Figure 3 The sub-pixel positioning algorithm flow chart described in the embodiments of the present application;

[0064] Figure 4 The temperature compensation effect verification diagram described in the embodiments of the present application;

[0065] Figure 5 The chessboard target schematic diagram described in the embodiments of the present application. DETAILED DESCRIPTION

[0066] In order to facilitate the understanding of the present application, the following will give a more comprehensive description of the present application, and the embodiments of the present application are given, but the scope of the present application is not limited thereto.

[0067] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the purpose of facilitating the description of the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the device or element indicated must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second" and the like are only for the purpose of description and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features limited by "first", "second" and the like can explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0068] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connection", "linking" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integrally connected; it can be mechanical connection, or electrical connection; it can be directly connected, or indirectly connected through intermediate medium, or internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0069] The sub-pixel level displacement monitoring method of the rail expansion regulator, as shown in Figures 1 to 5 The method comprises the following steps:

[0070] S1, multi-modal data acquisition of the rail expansion regulator,

[0071] The composite target group is arranged on the side of the switch rail and the basic rail, and the high-contrast geometric code and the near-infrared light-emitting unit are arranged on the surface of the switch rail and the basic rail. The images of the switch rail and the basic rail are synchronously collected by the visible light-near-infrared dual-band camera, and the track temperature data are collected by the temperature sensor in real time, so that the multi-modal data coverage is effectively ensured.

[0072] The composite target group comprises: a substrate surface etched with a ring-shaped code and an infrared LED;

[0073] Preferably, the infrared LED is embedded in the surface of the substrate, and the material of the substrate is titanium alloy. The near-infrared light-emitting unit is an infrared LED. The wavelength of the infrared LED is not less than 850 nm. The frame rate of the visible light-near-infrared dual-band camera is greater than or equal to 120 fps, and the protection level is IP68. The temperature sensor is a Pt100 platinum resistance temperature sensor with an accuracy of ±0.1℃. The geometric code is a concentric ring structure with a line width of 0.5 mm and an inter-ring spacing of 1 mm.

[0074] The present method collects track temperature and image data in multiple aspects, and fuses multiple data, thereby increasing the data coverage, providing more comprehensive data for subsequent data processing, and ensuring the accuracy and comprehensiveness of the later data processing.

[0075] S2, sub-pixel level edge positioning based on gradient direction consistency constraint of target edge:

[0076] Based on the gradient direction consistency constraint of the target edge, the visible light image collected by the visible light-near-infrared dual-band camera in S1 is solved in sub-pixel coordinates:

[0077] The target is a chessboard target, which is PVC coated with a nano anti-fouling coating on the surface, which can prevent diffuse reflection of glare and thus reduce interference with gradient calculation. The chessboard is 4 rows x 9 columns, with a white area reflectivity of 92% and a black area reflectivity of 8%. Such contrast mainly affects the signal-to-noise ratio; Since the accuracy of the target directly affects the accuracy of the sub-pixel coordinate solution, the accuracy verification experiment is accurate and reliable, so the errors easily caused in the generation and processing of the target are calculated:

[0078] Firstly: 20 times of laser tracker repeated measurement are made on the experimental chessboard target, and the theoretical position and actual position of the target feature point reference coordinates are obtained, as shown in Table 1:

[0079] Table 1

[0080]

[0081] S21, calculate the gradient field of the visible light image :

[0082] Firstly, the visible light image is converted into a gray scale image, which effectively simplifies the calculation and retains the brightness information of the image;

[0083] Secondly, the intensity change rate of each pixel in the image is calculated as a gradient vector using Sobel operator, including horizontal gradient formed by vertical edge convolution and vertical gradient formed by horizontal edge convolution , representing the gradient component of each pixel. The gradient amplitude is calculated as follows, and the higher the value, the stronger the edge.

[0084] ;

[0085] Wherein, the gray scale image pixel coordinates are .

[0086] S22, calculate the direction angle of the visible light image :

[0087] The direction angle represents the direction of gradient change, and is calculated as follows.

[0088] ;

[0089] Wherein, the atan2 function is an extension of the inverse tangent function, and the output angle is between -π and π radians.

[0090] S23, use the gradient field and the direction angle of the visible light image to solve the gradient direction standard deviation minimum point in the edge neighborhood , and accurately locate the target edge:

[0091] ;

[0092] wherein, denotes the coordinate of the target edge point, i.e. the coordinate of the gradient direction standard deviation minimum point to be solved; denotes the gradient direction at the point ; denotes the average gradient direction within the edge neighborhood Ω.

[0093] S24, the gradient calculation result is verified by bicubic spline interpolation precision cycle, the pixel positioning precision after visible light image processing is improved, and the precision reaches 0.1 pixel level;

[0094] ;

[0095] wherein, denotes the position increment after improving the precision, denotes the original position estimation, denotes the gray value function of the visible light image, denotes the horizontal and vertical pixel coordinates of the visible light image respectively.

[0096] After improving the precision of the new position, the position increment is detected for convergence, if the precision meets the requirements, it is ended, otherwise the gradient direction constraint and bicubic spline interpolation calculation of the new position are performed again. This step breaks through the limit of pixel level resolution, and effectively improves the theoretical recognition precision.

[0097] In order to compare the performance of the present displacement monitoring method, the pure gradient direction method and the pure bicubic spline method are selected to verify the experiment, and the experimental results are shown in Table 2.

[0098] Table 2

[0099]

[0100] From the above Table 2, the error precision of the present combined method is controlled below 0.1 pixel level, and the average error is obviously better than the other two methods, thereby improving the monitoring precision of the track and reducing the false positive rate.

[0101] S3, using adaptive weight fusion algorithm for multispectral collaborative imaging:

[0102] The visible light and near-infrared images collected by the visible light-near-infrared dual-band camera described in S1 are processed by using adaptive weight fusion algorithm:

[0103] ;

[0104] wherein, denotes the intensity value of the fused image,

[0105] weight and weight Dynamic adjustment according to real-time signal-to-noise ratio:

[0106] ,

[0107] wherein, and respectively visible light and near-infrared image signal-to-noise ratio, represent the intensity value of visible light, represent the intensity value of near-infrared image.

[0108] then wherein, represent the weight of the near-infrared compensation after fusion; this step effectively fuses the visible light detail features by penetrating rain and fog through near-infrared active illumination.

[0109] S4, real-time acquisition of temperature data by temperature sensor, application of thermal expansion displacement compensation model to eliminate systematic deviation:

[0110] The temperature of the steel rail obtained by the temperature sensor of S1 and the visible light displacement obtained by S2 Establish a temperature-displacement coupling correction model to eliminate the influence of environmental temperature:

[0111] ;

[0112] In the formula, represent the temperature displacement change, represent the actual measured displacement, is the expansion coefficient of the steel rail, is the reference length of the stretcher, which is obtained by field calibration by a total station, Real-time acquisition by platinum resistance temperature sensor.

[0113] S5, multi-modal data fusion of data in S2, S3 and S4;

[0114] Input the visible light displacement value obtained by S2, the near-infrared compensation amount obtained by S3, and the temperature correction amount obtained by S4 into the adaptive Kalman filter for multi-modal data fusion:

[0115] ;

[0116] wherein, represent the precision of the visible light displacement value, represent the near-infrared compensation amount after weight fusion, represent the temperature compensation amount, is the Kalman gain, which determines the weight of displacement value and compensation value in state update. The Kalman gain considers a certain size of sliding window in the fusion process, and the Kalman gain Error variance matrix is calculated based on the historical residual data of each sensor in the last 30 groups Update, the period is 10 seconds, and the formula is as follows:

[0117]

[0118] H represents the state transition matrix of the Kalman filter (the state components are fixed and unchanged), respectively corresponding to the fusion weights of visible light displacement value, near-infrared compensation value and temperature correction value.

[0119] Comparative Example 1

[0120] The comparative example scheme only arranges a normal black and white chessboard target without functional coating at the track monitoring point, uses a single visible light band camera to implement image acquisition, and completely lacks a temperature sensing device. The theoretical accuracy limit value achieved by the Canny edge detection and Harris corner matching algorithm is less than 0.5 pixels. Due to the lack of multispectral anti-interference mechanism, the feature point loss lock probability increases by 300% under rain and fog weather. Meanwhile, the lack of temperature compensation mechanism leads to a systematic error of 2.0-2.3 mm caused by the thermal expansion effect of the steel rail under a temperature difference of 15°C, which exceeds the safety threshold of high-speed rail. The field verification data show that the average error of displacement monitoring under standard light is 0.45 pixels, but under night low-illumination conditions, it deteriorates to 0.68 pixels, and the feature recognition rate is less than 55%. The false positive rate breaks through the 60% technical red line under rain and fog weather, and more seriously, the uncompensated displacement directly triggers false alarm when the rail temperature changes by 15°C, which makes the false positive rate increase to 45-48%. Under the conventional working condition of ±10°C temperature change, the monitoring confidence is less than 55%, which does not have the engineering applicability of all-weather high-precision monitoring of high-speed rail.

[0121] Example 1

[0122] ​In the side of the frog and the basic rail, a special composite target group is deployed, which cooperates with a high-frame-rate visible and near-infrared dual-band camera and a high-precision Pt100 temperature sensor to synchronously collect multi-modal data. Through step S2, the image processing is performed by using a chessboard target coated with a nano anti-fouling coating, combined with Sobel gradient field calculation and gradient direction consistency constraint, the sub-pixel level edge features are accurately located, and the bicubic spline interpolation technology is used for cyclic verification, finally realizing a stable positioning accuracy of up to 0.1 pixel level. From step S3, the system dynamically calculates the fusion weight of visible and near-infrared images based on real-time signal-to-noise ratio, and adaptively generates high-quality fusion images, which significantly improves the feature recognition ability in adverse weather conditions such as rain, fog and backlight. From step S4, a steel rail thermal expansion compensation model is established, and the displacement deviation caused by temperature is accurately calculated using real-time rail temperature data. For example, when the day-night temperature difference reaches 15°C, about 2.7mm of systematic error can be automatically deducted. From step S5, the visual displacement value, near-infrared compensation amount and temperature correction amount are input into the adaptive Kalman filter, which dynamically updates the gain coefficient based on the latest 30 groups of historical residual data, with a period of 10 seconds, and finally outputs the fused high-reliability displacement sequence. The test data shows that in the 30-day line monitoring, the average error of the feature point displacement is stably controlled within 0.07 pixels (about 0.1mm), the maximum error is not more than 0.097 pixels, and the overall false alarm rate is less than 5%.

[0123] The present application creatively integrates multi-modal imaging and physical compensation mechanism, realizes high-precision and high-robustness monitoring of the rail expansion regulator displacement without the description of quantitative indicators: through multi-spectral collaborative imaging, environmental interference is effectively overcome, sub-pixel positioning technology breaks through the limitation of traditional precision, embedded thermal expansion compensation eliminates systematic deviation, and finally a real-time monitoring system suitable for the harsh requirements of high-speed railway is established, which significantly improves the safety guarantee ability of the seamless line.

[0124] The rail expansion regulator sub-pixel level displacement monitoring method comprises:

[0125] The composite target group: a titanium alloy substrate surface etched with a ring code and embedded with an 850nm near-infrared LED, is installed on the side of the frog and the basic rail.

[0126] The multi-spectral imaging unit: a visible light + near-infrared dual-sensor camera is used to collect images of the frog and the basic rail in real time and transmit the data to the edge computing terminal.

[0127] The temperature compensation module: is used to collect the temperature of the frog and the basic rail in real time and transmit the data to the edge computing terminal.

[0128] The edge computing terminal: processes the collected data through the built-in sub-pixel solving algorithm and outputs the calculation results.

[0129] Safety warning platform: the calculation result of the edge computing terminal is transmitted to the BIM model through the 5G private network, and is used to realize three-level alarm.

[0130] The application creatively integrates optical imaging technology, physical law modeling and edge intelligent computing, solves the problems of large environmental interference, insufficient precision and lack of physical effect compensation in traditional monitoring, and provides reliable guarantee for safe operation of high-speed railway seamless line.

[0131] Preferably, the edge computing terminal integrates an NVIDIA Jetson module, and a CUDA-accelerated sub-pixel solution optimization library is built-in.

[0132] Preferably, the three-level alarm mechanism of the safety warning platform comprises:

[0133] First-level alarm: sound and light warning is triggered when the displacement exceeds the first-level warning threshold;

[0134] Second-level alarm: send a short message to the maintenance personnel when the displacement exceeds the second-level warning threshold;

[0135] Third-level alarm: automatically trigger the line speed reduction instruction when the displacement exceeds the third-level warning threshold.

[0136] The application also provides a non-transient computer readable storage medium (industrial grade SD card) storing the following computer instructions:

[0137] Adaptive spectrum fusion algorithm code;

[0138] Sub-pixel edge positioning optimization program;

[0139] Kalman filter dynamic weight update logic, when the processor executes the instructions, realizes the displacement monitoring method.

[0140] The application also provides a computer device comprising a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the monitoring method as described above, comprising the following steps:

[0141] Synchronously collecting visible light and near-infrared images of the switch rail and the basic rail through the dual-spectrum camera;

[0142] Enhancing the target feature by using an adaptive weight fusion algorithm;

[0143] Realizing sub-pixel edge positioning based on gradient direction consistency constraint;

[0144] Eliminating the influence of thermal expansion through a temperature-displacement coupling model;

[0145] Fusing multi-source displacement data by using an adaptive Kalman filter;

[0146] The early warning mechanism is triggered when the fusion displacement value exceeds a safety threshold.

[0147] The application also provides an electronic device, comprising a processor, a memory and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes instructions for realizing the monitoring method as described above, which comprises:

[0148] The multispectral imaging unit is controlled to synchronously collect visible light and near-infrared images at a frame rate of 120 fps;

[0149] A sub-pixel positioning algorithm is executed to solve the edge coordinates of the target;

[0150] Temperature sensor data are read in real time, and a thermal expansion compensation amount is calculated;

[0151] An adaptive Kalman filtering program is run to fuse multi-source displacement data;

[0152] The displacement monitoring result is transmitted to a BIM safety platform through a 5G communication module;

[0153] An audible and light alarm device is activated when the displacement exceeds a limit;

[0154] The electronic device is integrated into an edge computing terminal with an IP68 protection level, and is deployed on a railway site to realize unattended monitoring.

[0155] Those skilled in the art will understand that embodiments of the application can be provided as methods, systems, or computer program products. Therefore, the application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0156] The application is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the flowcharts and / or block diagrams. Figure 1 one flow or multiple flows and / or blocks Figure 1means for performing the function specified in the block or blocks.

[0157] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 one or more flowcharts and / or blocks Figure 1 one or more blocks or means for performing the function specified in the block or blocks.

[0158] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 one or more flowcharts and / or blocks Figure 1 one or more blocks or steps for performing the function specified in the block or blocks.

[0159] The above description is only a specific implementation of the present application, and is not intended to limit the protection scope of the present application. It should be understood by those skilled in the art that various modifications or changes can be made to the disclosed technical solutions without inventive labor, and all these modifications or changes should be covered within the protection scope of the present application.

Claims

1. A subpixel-level displacement monitoring method for rail expansion joints, characterized in that: Includes the following steps: S1. Perform multimodal data acquisition on the rail expansion joint; S2. Sub-pixel-level edge localization based on gradient direction consistency constraints of the target edge; S3. Adaptive weight fusion algorithm is used for multispectral collaborative imaging; S4. Collect temperature data in real time through temperature sensors and apply a thermal expansion displacement compensation model to eliminate systematic deviations; S5. Perform multimodal data fusion on the data from S2, S3 and S4; The specific method for S1 is as follows: Composite target groups are deployed on the sides of the switch rail and the base rail, and high-contrast geometric coding and near-infrared emitting units are set on the surface of the switch rail and the base rail. Images of the switch rail and the base rail are simultaneously acquired by a visible light-near-infrared dual-band camera, and temperature sensors acquire the rail temperature data of the switch rail and the base rail in real time. The specific method for S2 is as follows: S21. Calculate the gradient field of a visible light image. : First, convert the visible light image into a grayscale image; Secondly, the Sobel operator is used to calculate the intensity change rate of each pixel in the image as a gradient vector, including the horizontal gradient formed by vertical edge convolution. and the vertical gradient formed by horizontal edge convolution. , representing the gradient component of each pixel, is used to calculate the gradient magnitude by substituting the grayscale coordinates as follows; ； Wherein, the pixel coordinates of the grayscale image are ; S22. Calculate the direction angle of the visible light image. : The direction angle represents the direction of gradient change and is calculated as follows: ； The atan2 function is an extension of the arctangent function, and its output angle is between -π and π radians. S23, in the edge neighborhood Minimum point of the standard deviation of the gradient direction in the inner solution: ； in, This represents the coordinates of the target edge point, i.e., the coordinates of the point where the standard deviation of the gradient direction is minimized. Indicates at point The gradient direction at a given point is measured in radians or angles. This indicates the average gradient direction within the edge neighborhood Ω; S24. Accuracy verification is performed through bicubic spline interpolation. ； in, This represents the position increment after improving accuracy. This represents the original location estimate. The grayscale value function representing a visible light image. , These represent the horizontal and vertical pixel coordinates of a visible light image, respectively. The specific method for S5 is as follows: The visible light shift value obtained in S2, the near-infrared compensation value obtained in S3, and the temperature correction value obtained in S4 are input into an adaptive Kalman filter for multimodal data fusion. ； in, This represents a visible light shift value with considerable accuracy. This represents the near-infrared compensation amount after weighted fusion. Indicates the temperature compensation amount. The Kalman gain determines the weights of displacement and compensation values ​​in state updates; the Kalman gain considers a sliding window during the fusion process. The error variance matrix was calculated based on the most recent 30 sets of historical residual data from each sensor. The update cycle is 10 seconds, and the formula is as follows: ； Where H represents the state transition matrix of the Kalman filter. ; These correspond to the fusion weights of visible light shift value, near-infrared compensation amount, and temperature correction amount, respectively.

2. The sub-pixel level displacement monitoring method for rail expansion joints according to claim 1, characterized in that: The target in S2 is a checkerboard target, which is a PVC surface coated with a nano anti-fouling coating. The checkerboard grid has 4 rows and 9 columns, with a white area reflectivity of 92% and a black area reflectivity of 8%.

3. The sub-pixel level displacement monitoring method for rail expansion joints according to claim 1, characterized in that: The specific method for S3 is as follows: For the visible light and near-infrared images acquired by the visible light-near-infrared dual-band camera described in S1, an adaptive weight fusion algorithm is used: ； in, The weights represent the intensity values ​​of the fused image. and weight Dynamically adjusted based on real-time signal-to-noise ratio: ， This represents the intensity value of visible light. Indicates the intensity value of a near-infrared image; but ;in, This represents the near-infrared compensation amount after weighted fusion.

4. The sub-pixel level displacement monitoring method for rail expansion joints according to claim 1, characterized in that: The specific method for S4 is as follows: The rail temperature obtained from the temperature sensor in S1 and the visible light displacement obtained in S2 are compared... Establish a temperature-displacement coupled correction model: ； In the formula, It represents the amount of temperature displacement change. This represents the actual measured displacement. This is the coefficient of linear expansion of the rail. The reference length for the expansion joint was obtained through on-site calibration using a total station. Data is collected in real time by a temperature sensor.

5. A subpixel-level displacement monitoring system for rail expansion joints, employing the subpixel-level displacement monitoring method for rail expansion joints as described in any one of claims 1 to 4, characterized in that: include: Composite target assembly: A ring code is etched on the surface of a titanium alloy substrate and an 850nm near-infrared LED is embedded therein, which is then mounted on the sides of the switch rail and the base rail. Multispectral imaging unit: Visible light + near-infrared dual sensor camera, used to acquire images of the switch rail and base rail in real time and transmit the data to the edge computing terminal; Temperature compensation module: used to collect the temperature of the switch rail and the base rail in real time and transmit the data to the edge computing terminal; Edge computing terminal: Processes the collected data using a built-in sub-pixel resolution algorithm and outputs the calculation results; Safety early warning platform: Transmits the calculation results of the edge computing terminal to the BIM model through a 5G private network to realize three-level alarm.

6. The sub-pixel displacement monitoring system for rail expansion joints according to claim 5, characterized in that: Level 3 alarms include: Level 1 alarm: An audible and visual alarm is triggered when the displacement exceeds the Level 1 warning threshold; Level 2 alarm: When the displacement exceeds the level 2 warning threshold, a text message is sent to maintenance personnel; Level 3 alarm: When the displacement exceeds the level 3 warning threshold, a line speed reduction command will be automatically triggered.

7. The sub-pixel displacement monitoring system for rail expansion joints according to claim 5, characterized in that: The edge computing terminal integrates the NVIDIA Jetson module and has a built-in CUDA-accelerated subpixel resolution optimization library.

Images