Bridge structure crack image detection and width measurement method and system

Abstract

This invention relates to the field of image detection, specifically to a method and system for detecting and measuring the width of cracks in bridge structures. The method includes the following steps: based on a set of attitude perturbation parameters and a set of crack edge points extracted from a crack image sequence, a measurement plane for each frame of the image is constructed in a stable reference frame through multi-plane fitting, generating a primary reference plane sequence. The reference plane is used to simultaneously calibrate and correlate crack edge points across frames, obtaining a set of crack edge trajectories. A crack skeleton point set is constructed by combining crack skeleton data. A primary normal measurement path sequence is reconstructed based on this set. Spatial coordinate fitting is performed on the measurement path segments to construct a secondary reference plane sequence and update the normal path, obtaining a secondary normal measurement path sequence. Normal sampling is performed along the updated normal path. The crack width is calculated by combining attitude perturbation parameters and the trajectory set, and the width measurement result is output. This invention improves the accuracy and stability of crack measurement in dynamic detection scenarios.

Description

Technical Field

[0001] This invention relates to the field of image detection, specifically to a method and system for detecting and measuring the width of cracks in bridge structures. Background Technology

[0002] In the daily operation and maintenance and periodic safety inspection of bridge structures, crack detection, as a key indicator for evaluating the health status of concrete components, is directly related to the durability and operational safety of the bridge structure. In recent years, with the development of high-resolution cameras and image processing technology, image analysis-based crack identification and measurement methods have gradually emerged, improving the automation and data processing capabilities of crack detection to a certain extent. To further improve detection efficiency, some systems also combine mobile platforms such as drones and inspection vehicles to achieve batch image acquisition of cracks on the bridge surface, supplemented by crack identification algorithms to extract crack areas.

[0003] In existing technologies, under dynamic detection scenarios (such as bridge vibration, wind load, or detection platform shaking), the image acquisition process is easily affected by attitude disturbances, leading to inter-frame drift in crack location and consequently inaccurate crack width measurement. Existing methods often neglect the impact of attitude changes on the measurement reference plane, failing to construct a stable and unified reference measurement plane, which is a problem we need to solve. Summary of the Invention

[0004] The purpose of this application is to provide a method for image detection and width measurement of cracks in bridge structures. This method effectively reduces the interference of UAV attitude disturbance and bridge structure vibration on the stability of image acquisition, improves the accuracy of crack edge positioning and width measurement, and thus enhances the reliability of automated bridge crack detection results.

[0005] The objective of this application can be achieved through the following technical solution: Firstly, a method for image detection and width measurement of cracks in bridge structures, comprising the following steps:

[0006] Acquire image sequence data of cracks in bridge structures and their image drift data, along with attitude angle data of UAVs. Perform time-series correlation calculations on the attitude angle data and the image drift data to obtain a set of attitude disturbance parameters.

[0007] Based on the set of attitude perturbation parameters and the set of crack edge points extracted from the crack image sequence data, a measurement plane for each frame of image in a stable reference system is constructed by multi-plane fitting, generating a reference plane sequence.

[0008] Based on the primary reference surface sequence and crack image sequence data, the crack edge points in the crack image sequence data are synchronously calibrated and correlated with each frame to generate a crack edge trajectory set.

[0009] Based on the crack edge trajectory set and the crack skeleton data extracted from the crack image sequence data, a crack skeleton point set is constructed; for the measurement points in the crack skeleton point set, based on the first-order reference surface sequence, the normal measurement path corresponding to the measurement point is reconstructed to obtain the first-order normal measurement path sequence.

[0010] Based on the first normal measurement path sequence, a measurement path segment representing the degree of change in the concentrated edge position of the crack edge trajectory is obtained. The spatial distribution coordinates of the measurement path segment are refitted and updated to generate a secondary reference surface sequence. Based on the updated secondary reference surface sequence, the direction and structure of the measurement path segment in the first normal measurement path sequence are updated to obtain the secondary normal measurement path sequence.

[0011] Along the updated secondary normal measurement path sequence, normal sampling is performed on the crack edge points, and the width measurement data of the measurement points is calculated based on the attitude perturbation parameter set and the crack edge trajectory set to generate the crack width measurement result.

[0012] Secondly, the bridge structure crack image detection and width measurement system includes the following modules:

[0013] The image acquisition module is used to acquire image sequence data of cracks in the bridge structure and its image drift data, as well as attitude angle data of the UAV. It performs time-series correlation calculation on the attitude angle data and the image drift data to obtain a set of attitude disturbance parameters.

[0014] The reference surface construction module is used to construct the measurement plane of each frame of image in a stable reference system by means of multi-plane fitting based on the set of attitude perturbation parameters and the set of crack edge points extracted from the crack image sequence data, and generate a reference surface sequence.

[0015] The crack edge module is used to synchronously calibrate and perform inter-frame association on crack edge points in the crack image sequence data based on the primary reference surface sequence and crack image sequence data, and generate a crack edge trajectory set.

[0016] The normal path module is used to construct a set of crack skeleton points based on the crack edge trajectory set and the crack skeleton data extracted from the crack image sequence data; for the measurement points in the crack skeleton point set, the normal measurement path corresponding to the measurement point is reconstructed based on the first-order reference surface sequence to obtain a first-order normal measurement path sequence.

[0017] The image optimization module is used to obtain measurement path segments that characterize the degree of change in the concentrated edge position of the crack edge trajectory based on the primary normal measurement path sequence, and generate a secondary reference surface sequence by refitting and updating the spatial distribution coordinates of the measurement path segments, and update the direction and structure of the measurement path segments in the primary normal measurement path sequence based on the updated secondary reference surface sequence to obtain the secondary normal measurement path sequence.

[0018] The crack width measurement module is used to sample the normal vector of the crack edge points along the updated secondary normal measurement path sequence, and calculate the width measurement data of the measurement points based on the attitude perturbation parameter set and the crack edge trajectory set to generate the crack width measurement result.

[0019] Compared with the prior art, the beneficial effects of this application are:

[0020] 1. This application constructs a primary reference plane sequence and a secondary reference plane sequence to progressively update and spatially adaptively correct the crack measurement plane, so that the crack width measurement is always based on the local geometric structure under a stable reference system. This helps to improve the instability of the measurement plane caused by attitude changes in the dynamic detection scenario of the prior art, and improves the spatial consistency and reliability of the crack width measurement results.

[0021] 2. This application introduces attitude angle variation to correct the normal direction projection of the crack width measurement value during the crack width calculation stage. Furthermore, it constructs a crack width error range based on the change amplitude of attitude angle within a preset time window. This ensures that the crack width measurement result not only has a definite value but also provides the measurement uncertainty range caused by dynamic attitude disturbances, thereby improving the reliability and application security of crack detection results in engineering evaluation. In dynamic detection scenarios, it realizes explicit modeling and quantitative constraints on the sources of crack width measurement error, which helps to reduce the problem of measurement result fluctuation and insufficient reliability that may be caused by relying solely on static images or single-frame measurements in existing technologies. It is suitable for high-precision crack detection and long-term health monitoring in complex environments such as bridges. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the steps of the bridge structure crack image detection and width measurement method of this application;

[0023] Figure 2 This is a schematic diagram of the modules of the bridge structure crack image detection and width measurement system of this application. Detailed Implementation

[0024] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0025] Application Overview:

[0026] In traditional UAV inspection and image measurement technologies for bridge structural cracks, when faced with complex on-site environments such as continuous wind loads, bridge vibrations, or unstable UAV flight, the acquired image sequences will experience non-rigid deformation and drift due to multi-source disturbances. This significantly interferes with the stable tracking of crack edges and the accurate measurement of crack width. Existing technologies typically rely on single-frame image analysis or simple image registration, failing to effectively integrate UAV attitude angle data with the vibration and displacement data of the bridge structure itself. They also lack dynamic modeling and compensation for the local three-dimensional geometry of the cracks. This results in the measurement reference plane (or line) failing to maintain spatial consistency in the image sequence, causing significant fluctuations in the width measurement results due to changes in viewpoint and structural micro-movements, thus reducing the reliability of the assessment conclusions.

[0027] For example, when conducting drone-based crack inspections on a highway bridge in a lightly windy environment with light vehicle traffic, the system simultaneously acquired crack image sequences, drone attitude angles, and bridge acceleration data. Traditional methods rely solely on image features for edge detection. When analyzing a sequence of images showing periodic lateral displacement of crack edges due to slight torsional vibration of the bridge, the system mistakenly identified this pixel displacement caused by structural vibration as the actual positional change of the crack edges. Because the measurement path direction was not dynamically corrected based on the actual local surface normal of the crack, in frames with a specific vibration phase, the "edge points" sampled along the original image plane normal actually deviated from the true crack boundary. This resulted in periodic abnormal peaks and troughs in the calculated crack width value, severely distorting the true crack width and its changing trend, leading to biases in the structural safety assessment.

[0028] Without addressing the aforementioned issues, automated crack detection and quantitative assessment based on UAV imagery will be difficult to reliably apply in engineering practice. Inaccurate crack width data could mislead maintenance decisions, leading to insufficient maintenance of actually dangerous cracks or excessive intervention in harmless cracks, resulting in safety hazards or resource waste. In the long term, the uncertainty and low repeatability of measurement results will hinder the large-scale adoption of this technology, forcing engineering inspections to remain highly reliant on time-consuming, labor-intensive, and subjective close-range manual surveys, thus failing to leverage the efficient and comprehensive technological advantages of UAV inspections.

[0029] Therefore, this application provides a method for image detection and width measurement of cracks in bridge structures, such as... Figure 1 As shown, it includes the following steps:

[0030] Acquire image sequence data of cracks in bridge structures and their image drift data, along with attitude angle data of UAVs. Perform time-series correlation calculations on the attitude angle data and the image drift data to obtain a set of attitude disturbance parameters.

[0031] Based on the set of attitude perturbation parameters and the set of crack edge points extracted from the crack image sequence data, a measurement plane for each frame of image in a stable reference system is constructed by multi-plane fitting, generating a reference plane sequence.

[0032] Based on the primary reference surface sequence and crack image sequence data, the crack edge points in the crack image sequence data are synchronously calibrated and correlated with each frame to generate a crack edge trajectory set.

[0033] Based on the crack edge trajectory set and the crack skeleton data extracted from the crack image sequence data, a crack skeleton point set is constructed; for the measurement points in the crack skeleton point set, based on the first-order reference surface sequence, the normal measurement path corresponding to the measurement point is reconstructed to obtain the first-order normal measurement path sequence.

[0034] Based on the first normal measurement path sequence, a measurement path segment representing the degree of change in the concentrated edge position of the crack edge trajectory is obtained. The spatial distribution coordinates of the measurement path segment are refitted and updated to generate a secondary reference surface sequence. Based on the updated secondary reference surface sequence, the direction and structure of the measurement path segment in the first normal measurement path sequence are updated to obtain the secondary normal measurement path sequence.

[0035] Along the updated secondary normal measurement path sequence, normal sampling is performed on the crack edge points, and the width measurement data of the measurement points is calculated based on the attitude perturbation parameter set and the crack edge trajectory set to generate the crack width measurement result.

[0036] In some of the embodiments described above in this application, an attitude disturbance parameter set is proposed to compensate for attitude disturbance. However, in this process, since the crack image sequence data, image drift data and attitude angle data come from different sensors and each has an independent sampling frequency and timestamp, the data time is not synchronized, and the attitude disturbance information cannot be accurately aligned, which affects the accuracy of the attitude disturbance parameter set and causes errors in subsequent measurement plane construction and crack width measurement.

[0037] To address this, this application further proposes a process for obtaining image sequence data of cracks in a bridge structure and its image drift data, along with attitude angle data of a UAV. The process involves performing time-series correlation calculations on the attitude angle data and the image drift data to obtain an attitude disturbance parameter set. This process includes: the crack image sequence data comprising bridge structure image frames and their corresponding image acquisition timestamps; the image drift data being structural acceleration data acquired by accelerometers deployed on the bridge structure, including x-direction acceleration, y-direction acceleration, z-direction acceleration, and their corresponding time series; the attitude angle data comprising measured values ​​of pitch angle, roll angle, and yaw angle, and their angle acquisition timestamps; establishing a time correspondence based on the image acquisition timestamps and the angle acquisition timestamps; converting attitude angle data and image drift data with different sampling frequencies to a unified time resolution using an interpolation algorithm; and performing time-axis pairing of the attitude angle data and image drift data to obtain a dataset corresponding to the time-stamp-aligned attitude angle data and image drift data, denoted as the attitude disturbance parameter set.

[0038] In one preferred embodiment, the UAV image acquisition module acquires image frames at a fixed frame rate and records precise timestamps; its inertial measurement unit (IMU) acquires pitch, roll, and yaw angles at a higher frequency, along with angle acquisition timestamps; simultaneously, image drift data can be acquired by onboard accelerometers or MEMS sensors deployed on the bridge, recording triaxial acceleration and its time series. The system unifies the time reference of each sensor (e.g., GPS clock), using the image acquisition time point as a reference, and resamples the high-frequency attitude angle data and acceleration data through linear interpolation or cubic spline interpolation to form a data stream with unified time resolution. Finally, the system performs one-to-one matching of the three types of data based on the time index, constructing a comprehensive parameter set including image frames, attitude angles, and structural acceleration, which serves as the basic input for subsequent measurement plane fitting and attitude compensation.

[0039] Through the above technical solution, this application can effectively solve the problem of attitude information distortion caused by the time asynchrony between image data and sensor data, ensuring the consistency between crack image frames and corresponding attitude perturbation states. The attitude perturbation parameter set established thereby can significantly improve the spatial stability of subsequent measurement plane construction, further enhancing the accuracy and robustness of crack edge calibration and width measurement.

[0040] In some of the solutions described above in this application, without interpolation resampling and time alignment processing, the time drift between different sensors will directly affect the attitude compensation accuracy, causing the crack edges in the image to shift or distort during the spatial mapping process, thereby affecting the reliability of the measurement path direction and crack width results. Therefore, constructing a highly consistent and time-aligned attitude perturbation parameter set is a key step in achieving accurate crack detection in dynamic scenes.

[0041] In some embodiments described above in this application, a primary reference plane sequence is proposed to be constructed based on the attitude perturbation parameter set and the crack edge point set extracted from the crack image sequence data to provide a measurement plane under a stable reference frame. However, in its implementation, due to the influence of attitude perturbation, the extraction and compensation of crack edge points may be inaccurate, resulting in an unstable measurement plane and an inability to effectively eliminate inter-frame drift.

[0042] To address this, this application further proposes a method for generating a reference plane sequence by constructing a measurement plane for each frame of image in a stable reference frame through multi-plane fitting, based on the aforementioned attitude perturbation parameter set and crack image sequence data. This process includes: performing grayscale and noise filtering on each frame of image in the crack image sequence data to obtain a preprocessed crack image sequence; calculating the grayscale gradient components of pixels along the horizontal and vertical directions within the processed crack image sequence; and calculating the gradient magnitude of pixels using the Euclidean norm based on the grayscale gradient components in the horizontal and vertical directions; performing principal component analysis on the pixel set of the preprocessed crack image sequence to obtain the crack principal axis direction vector; and calculating the direction projection of the grayscale gradient components of pixels based on the crack principal axis direction vector to obtain the direction projection of pixels perpendicular to the crack principal axis. The upward gradient component is selected, and pixels with gradient magnitudes greater than a first preset threshold are selected as candidate crack edge pixels. Candidate crack edge pixels with an absolute value of the difference between the mean grayscale values ​​of the two preset neighborhoods along the crack's main axis direction greater than a second preset threshold are selected as crack edge points, thus constructing a crack edge point set. Based on the attitude angle data and image drift data corresponding to the timestamp of the current image frame in the attitude perturbation parameter set, attitude compensation processing is performed on the crack edge point set, transforming the crack edge points from the image acquisition coordinate system to a unified reference coordinate system, obtaining the attitude perturbation corrected crack edge point set. Based on the crack edge point set, a weighted least squares fitting method is used to construct the measurement plane of the corresponding image frame in a stable reference system, and the measurement planes of each image frame are arranged in chronological order to generate a primary reference plane sequence.

[0043] In a preferred embodiment, after attitude compensation, this application constructs a measurement plane using a weighted least squares fitting method for the crack edge point set corresponding to each image frame. The weight allocation can be set based on the gradient magnitude or spatial density of the edge points to enhance the fitting contribution of high-confidence points. Finally, the measurement planes of all image frames are arranged in the order of image acquisition time to form a reference plane sequence, providing a stable geometric benchmark for subsequent crack edge calibration, skeleton extraction, and width measurement.

[0044] Through the above technical solutions, this application solves the problems of inconsistent spatial positions of crack edge points, measurement plane drift, and unstable geometric reference caused by attitude perturbation. Compared with the prior art, this method introduces a dual edge extraction mechanism of gradient direction projection and grayscale domain contrast, which improves the crack edge discrimination capability; coordinate compensation achieved by combining attitude perturbation parameter set ensures the consistency of spatial reference between different image frames; and the introduction of weighted least squares fitting method further enhances the robustness and accuracy of the measurement plane, ultimately realizing the continuous construction of measurement reference in crack image sequence.

[0045] In some of the solutions described above in this application, if a gradient direction extraction mechanism guided by the principal axis direction is not introduced, the crack edge points are easily affected by image texture or false edges, increasing the risk of false detection. If attitude perturbation compensation is ignored, the same crack will be spatially misaligned in different image frames due to changes in the UAV's attitude, severely affecting the accuracy of subsequent crack trajectory construction and width measurement. Therefore, constructing a crack edge point set based on gradient intensity + principal axis direction + attitude perturbation compensation, and fitting a stable measurement plane sequence accordingly, is a key technical means to improve the robustness of the system in dynamic detection environments.

[0046] In some embodiments described above in this application, a method is proposed to synchronously calibrate and correlate crack edge points with frames based on a single reference surface sequence to generate a set of crack edge trajectories, thereby establishing continuous trajectories for crack edges. However, during its implementation, due to image drift caused by attitude perturbations, the positions of crack edge points may change significantly between frames, leading to difficulties in inter-frame matching, discontinuous or inaccurate trajectory generation, and consequently affecting the accuracy and reliability of subsequent crack width measurements.

[0047] To address this, this application further proposes a process for synchronously calibrating and inter-frame associating crack edge points in crack image sequence data based on a primary reference surface sequence and crack image sequence data, generating a crack edge trajectory set as follows: Based on the primary reference surface sequence, crack edge points are mapped to a unified spatial reference system to generate an intra-frame calibration point set; intra-frame calibration point sets of adjacent frame images are obtained, the difference between Euclidean distance and grayscale mean is calculated, and the intra-frame calibration point sets whose Euclidean distance is less than a preset spatial threshold and whose absolute value of the grayscale mean difference is less than a preset grayscale similarity threshold are selected to establish an inter-frame matching relationship; the corresponding intra-frame calibration point sets are connected according to the time sequence and the inter-frame matching relationship to obtain a crack edge trajectory sequence; trajectory segments in the crack edge trajectory sequence that are continuously missing for more than a preset number of frames are removed, and linear interpolation is performed to complete and merge them based on the positions of the intra-frame calibration points and the grayscale mean to construct a crack edge trajectory set.

[0048] In the crack image sequence processing stage, a unified spatial reference coordinate system is first constructed based on the primary reference surface sequence. The crack edge point set extracted from each frame image after preprocessing (such as Canny edge detection and morphological processing) is then projected onto this unified spatial reference system to form an intra-frame calibration point set. The projection transformation process is based on the camera intrinsic parameters, attitude perturbation parameter set, and the geometric equation of the primary reference surface to ensure that the crack edge point coordinates of each frame image are uniformly expressed.

[0049] Subsequently, for the intra-frame calibration point sets of two adjacent frames, the 3D Euclidean distance and the difference in grayscale mean between any pair of points are calculated sequentially. A spatial threshold (e.g., 2cm) and a grayscale similarity threshold (e.g., 20 / 255) are set. When the Euclidean distance between a pair of points is less than the former, and the absolute value of the difference in the grayscale mean of their 5×5 neighborhood is less than the latter, an inter-frame matching relationship is established. This "spatial + grayscale" dual-threshold screening mechanism effectively suppresses the risk of trajectory drift caused by mismatches.

[0050] Based on the established inter-frame matching relationship, matching points are connected in the order of image frame acquisition time to form a preliminary crack edge trajectory sequence. For missing frame segments in the trajectory, if the number of missing consecutive frames does not exceed a preset tolerance limit (e.g., 5 frames), a linear interpolation method is used to estimate the corresponding points of the missing frames based on the point positions and grayscale averages of adjacent valid frames, thus restoring the continuity of the trajectory. If the number of missing frames exceeds the preset limit, the trajectory segment is discarded.

[0051] Through the above technical solution, this application can effectively solve the problem of discontinuous and unstable crack edge trajectories caused by UAV attitude disturbances and image frame differences in dynamic detection scenarios. On the one hand, the intra-frame calibration mechanism under a unified reference coordinate system ensures the consistency of spatial coordinates of each frame image, avoiding projection drift of crack edge points at different times. On the other hand, the inter-frame matching method combining spatial distance and grayscale features improves matching accuracy, and the interpolation completion mechanism enhances trajectory continuity, significantly improving the spatiotemporal tracking capability of crack edge points.

[0052] In some of the solutions described above, direct inter-frame matching based on image pixel positions is easily affected by changes in viewing angle and platform micro-vibrations, leading to a high matching error rate. This can cause interruptions or deviations in crack trajectories, ultimately affecting the accuracy of width measurement results. This application significantly improves the stability and robustness of crack trajectory recognition by introducing unified spatial mapping under attitude perturbation compensation and similarity screening based on image intensity features. This provides continuous and reliable data support for crack skeleton generation and width measurement.

[0053] In some of the embodiments described above in this application, a set of crack skeleton points is proposed to support crack width measurement. However, during its implementation, due to attitude perturbation and image noise, the initial skeleton points may contain invalid points, resulting in an inaccurate skeleton point set and affecting the accuracy of subsequent measurements. To address this, this application proposes a method for constructing a crack skeleton point set. The specific process includes: analyzing the crack region image using the Zhang-Suen skeletonization algorithm based on the crack principal axis direction vector to obtain an initial skeleton point set; constructing a neighborhood window centered on the initial skeleton point, with a preset pixel 1 along the principal axis direction and a preset pixel 2 along the normal direction; searching for crack edge trajectory points in the corresponding frame within the neighborhood window; selecting initial skeleton points whose pixel distance to the crack edge trajectory points is less than a preset pixel 3 as valid skeleton points and including them in the skeleton point set; sorting the skeleton point set according to frame order and their projection distance along the principal axis direction; constructing a sliding window with a preset pixel 4 step size; generating a skeleton fitting curve using a local fifth-order polynomial fitting method limited by the principal axis direction; and generating a crack skeleton sampling point set by sampling at equal intervals; wherein each skeleton sampling point includes its pixel coordinates, the index of its corresponding image frame, and information on the principal axis tangent direction and normal direction.

[0054] As a preferred embodiment, the solution of this application is implemented as follows: After acquiring the image sequence data of bridge structure cracks, the crack area image is first binarized, and the Zhang-Suen skeletonization algorithm is used to extract an initial skeleton point set. Then, a neighborhood window is constructed based on the crack principal axis direction, with each skeleton point as the center. The window size is jointly determined by a preset pixel 1 along the principal axis direction and a preset pixel 2 along the normal direction. The crack edge trajectory points of the frame to which the skeleton point belongs are searched within the window. If the pixel distance from the skeleton point to any edge trajectory point is less than a preset pixel 3, it is determined to be a valid skeleton point and retained.

[0055] After obtaining the effective skeleton point set, the skeleton points are sorted according to the image frame order and the projection distance of the skeleton points along the principal axis. A sliding window is constructed with a preset step size of four pixels. A local fifth-order polynomial is used to fit the skeleton points within the window along the principal axis, generating a skeleton fitting curve. Subsequently, sampling is performed on this fitting curve at fixed pixel intervals to obtain the crack skeleton point set. Each skeleton point includes not only two-dimensional image coordinates but also its corresponding image frame index, principal axis tangent direction, and normal direction, providing directional basis for subsequent width measurement.

[0056] Through the above technical solutions, this application can effectively eliminate invalid skeleton points caused by image noise or pose disturbances, improving the accuracy and stability of the skeleton point set. Simultaneously, local fifth-order fitting can smooth the crack path and accurately represent its geometric orientation, while equidistant sampling ensures the uniformity of the measurement point distribution. The final crack skeleton point set not only possesses good geometric continuity but also includes complete directional information, significantly improving the robustness and accuracy of crack width measurement.

[0057] In some of the aforementioned solutions in this application, the initial skeleton points are directly extracted from binary images, which easily leads to misidentified false points. If these points are used directly for width measurement without verification, they can easily cause misjudgment of direction or sampling offset, resulting in measurement errors. Therefore, this solution introduces crack edge trajectory points as a basis for validity judgment. By constructing a neighborhood window and performing spatial pairing, the spatial reliability of the skeleton points is effectively improved. Subsequently, through sliding window fitting and spacing sampling mechanisms, the skeleton orientation is modeled and standardized, addressing the technical shortcomings of traditional skeleton extraction, such as easy breakage and lack of directionality. This provides solid support for the directional stability and sampling consistency of crack width.

[0058] In some of the embodiments described above in this application, a method for reconstructing the normal measurement direction path is proposed to obtain a stable measurement path. However, in this process, due to attitude disturbances and inter-frame drift, the normal direction may be inconsistent in adjacent frames, resulting in discontinuous path trajectories and measurement errors.

[0059] To address this, this application proposes a method for reconstructing the normal measurement direction path corresponding to a measurement point in a crack skeleton point set, based on a first-order reference surface sequence, to obtain a first-order normal measurement path sequence. The method includes: obtaining the crack principal axis direction vector in the frame image where the measurement point is located; constructing a normal direction vector based on the crack principal axis direction vector; and generating a line segment of a preset pixel region length along the normal direction vector in the image, starting from the pixel coordinates of the measurement point, as the normal measurement path of the measurement point; spatially aligning skeleton points in adjacent image frames that are within a preset measurement pixel range of Euclidean distance from the measurement point; calculating the angle between the corresponding normal direction vectors; selecting angles smaller than a preset threshold angle and adding them to the continuous path trajectory of the measurement point; sequentially connecting these angles to obtain the normal measurement path of the measurement point in multiple frames of images; and summarizing the normal measurement paths of all measurement points to generate a first-order normal measurement path sequence.

[0060] For a given measurement point in the crack skeleton point set, the system first extracts the crack principal axis direction vector from the image frame in which it is located. For example, by using principal component analysis (PCA) or local linear fitting methods, the system identifies the extension trend of the surrounding skeleton points in the local region of the image and determines the principal axis direction accordingly. Subsequently, based on this principal axis direction vector, the system constructs the corresponding normal direction vector through orthogonal transformation in two-dimensional image coordinates, ensuring that the normal measurement path is perpendicular to the crack orientation.

[0061] Starting from the pixel coordinates of the measurement point, the system generates a line segment of fixed length (e.g., 20 pixels) along the normal direction vector, which serves as the normal measurement path for the current image frame. This path is used for subsequent sampling operations of the crack width and constitutes the spatial measurement reference for the measurement point in that frame.

[0062] To enhance the continuity of the measurement path over time, the system further searches for skeleton points in adjacent image frames that are close to the measurement point in terms of Euclidean distance (e.g., less than 5 pixels away). The local normal direction vector of each candidate point is calculated, and the angle between this vector and the normal direction of the measurement point in the current frame is calculated. If the angle is less than a preset threshold (e.g., 10°), the two directions are considered to be consistent, and an inter-frame trajectory connection can be constructed. This process can be iteratively executed to construct a continuous and consistent normal measurement path trajectory across multiple subsequent image frames.

[0063] Finally, the system summarizes the normal measurement paths established by each measurement point in multiple frames of images to form a sequence of normal measurement paths. This sequence can be used as the data sampling path for crack width measurement, supporting subsequent extraction of normal pixel pairs and calculation of width values.

[0064] Through the above technical solution, this application effectively solves the problem of inconsistent normal directions in adjacent frames caused by attitude disturbances and inter-frame drift in traditional methods during dynamic detection scenarios, leading to discontinuous path trajectories and measurement errors. Specifically, based on the stable reference system provided by a primary reference surface sequence, this solution provides a unified and stable benchmark for calculating the normal direction, thereby reducing the impact of dynamic disturbances on the accuracy of the normal direction. By obtaining the crack principal axis direction vector and constructing the normal direction vector, the perpendicularity of the measurement direction to the crack orientation is ensured. More importantly, by introducing an inter-frame alignment mechanism, using Euclidean distance to filter spatially close skeleton points, and combining the angle threshold of the normal direction vectors for direction consistency judgment, this solution can effectively filter out frames with drastic changes in the normal direction caused by attitude disturbances, thereby ensuring the continuity and stability of the constructed normal measurement path in multi-frame images. This continuous and stable normal measurement path provides a reliable benchmark for subsequent accurate crack width sampling, significantly improving the accuracy and robustness of crack width measurement.

[0065] In some of the solutions described above in this application, traditional methods often assume that the pose is stable during the acquisition of each frame of image, and then directly construct the normal path for measurement after the crack skeleton points are extracted. This lacks a cross-frame continuity control mechanism and is prone to sudden changes in path direction due to viewpoint disturbances, which affects the stability of crack width estimation.

[0066] In some embodiments described above in this application, a primary reference surface sequence and a primary normal measurement path sequence are generated based on an attitude perturbation parameter set and crack image sequence data to compensate for the influence of attitude perturbation and measure crack width. However, in its implementation, the primary normal measurement path sequence may not accurately capture the degree of change in crack edge position under dynamic scenes. This is because attitude perturbation and image drift cause inconsistencies in edge point positions between frames, making the measurement path based on the primary reference surface unstable and affecting the accuracy of subsequent width measurements.

[0067] In response, this application proposes an improved scheme, which obtains measurement path segments that characterize the degree of change in the concentrated edge position of crack edge trajectory based on the first normal measurement path sequence, and generates a secondary reference surface sequence by refitting and updating the spatial distribution coordinates of the measurement path segments. Specifically, the scheme includes the following steps: extracting the coordinates of the crack skeleton points corresponding to each normal path in the primary normal measurement path sequence and the crack edge points on both sides in the crack image sequence data to obtain the edge point side coordinate set; aligning the edge point side coordinate set on the same path by a preset sliding frame number, calculating the standard deviation of the lateral and longitudinal displacements of the edge points within the preset sliding frame number, and selecting the path segments traversed by the time period corresponding to the standard deviation of the lateral displacement being greater than or equal to a preset lateral pixel or the standard deviation of the longitudinal displacement being greater than or equal to a preset longitudinal pixel as the measurement path segment; extracting the coordinate data of the skeleton points and edge points within the time period of the measurement path segment respectively, and constructing a crack cross-section edge curve group based on the least squares fitting method; generating a single-frame reference surface by using a surface interpolation fitting algorithm on the crack cross-section edge curve group, and performing spatial fitting on the single-frame reference surface with the skeleton point as the center and the edge points as the boundary; combining the reference surfaces into a spatially continuous surface sequence according to the time frame order to obtain a secondary reference surface sequence.

[0068] As a preferred embodiment, the solution of this application is implemented as follows: First, for each primary normal measurement path, its corresponding crack skeleton point is extracted, and the coordinates of the intersection points of the two edge points of the skeleton point in each frame image are found along the normal direction in the image sequence, forming an edge point side coordinate set. This coordinate set is used to reflect the edge dynamic response of the normal path on the time axis.

[0069] Then, a sliding window process is applied to the edge point coordinate set. The sliding window width is set to 5 frames, and the inter-frame standard deviations of the horizontal (x-direction) and vertical (y-direction) coordinates are calculated separately along the same path. If the standard deviation in either direction exceeds a preset threshold (e.g., 0.5 pixels) within a certain time window, it is considered that there is significant spatial perturbation in that path segment, and this path segment is marked as a measurement path segment, requiring further fitting calculation.

[0070] Next, the system extracts the coordinate data of the crack skeleton points and edge points within the corresponding time periods of these measurement path segments. For each frame of data, least-squares polynomial fitting is performed on the edge points on both sides of the crack to construct a set of crack cross-section edge curves. This fitting process uses the skeleton points as the central axis reference to ensure the geometric continuity and boundary integrity of the fitted curves.

[0071] Subsequently, based on the crack cross-section edge curves of each frame, a single-frame reference surface is generated using a surface interpolation fitting method (such as thin-plate splines or B-splines). This surface is centered on the skeleton points and bounded by the edge curves, ensuring that its spatial morphology is consistent with the actual crack structure. All single-frame reference surfaces are then combined according to the image frame time sequence to form a spatially continuous sequence of secondary reference surfaces, achieving stable surface connections across the time dimension.

[0072] Through the above technical solution, this application can effectively identify and handle the measurement path instability problem caused by attitude disturbances and inconsistent motion at the crack edge. Compared with the strategy of fitting only skeleton points for the primary reference surface, the secondary reference surface models the edge changes of key path segments, which better reflects the actual dynamic changes of the crack; it identifies high-fluctuation regions through a sliding window and performs fine fitting only on unstable path segments, balancing efficiency and accuracy; the final constructed secondary reference surface sequence is continuous in time and smooth in space, providing a highly robust reference plane for subsequent normal measurement of crack width, significantly improving the consistency and reliability of the measurement results.

[0073] In some of the solutions described above in this application, the primary reference surface is mainly constructed based on static skeleton point fitting, which is insufficient to handle the micro-drifts or noise fluctuations in the crack edge trajectory between image frames caused by factors such as illumination, attitude perturbation, and structural complexity. This solution dynamically identifies highly variable regions by evaluating the standard deviation of the horizontal and vertical displacements of crack edge points on the time axis, and performs cross-sectional curve fitting and surface interpolation reconstruction on these key path segments, thus maintaining measurement accuracy while avoiding redundant computational overhead.

[0074] In some embodiments described above in this application, a sequence of primary reference surfaces and a sequence of primary normal measurement paths are generated based on an attitude perturbation parameter set to construct a stable measurement plane and path. However, in this process, due to the influence of attitude perturbation, some path segments in the primary normal measurement path sequence may have directional deviations, resulting in inaccurate crack width measurements because the path direction is inconsistent with the actual crack normal direction, and cannot effectively reflect the dynamic changes of the crack edge.

[0075] To address this issue, this application proposes a method for updating the orientation and structure of measurement path segments in a primary normal measurement path sequence. The method first obtains the local reference surface of the path starting point in the secondary reference surface sequence at the corresponding time frame, based on the location of the crack skeleton point corresponding to the measurement path segment within the path starting point position in the primary normal measurement path sequence. Then, based on the spatial fitting result of the local reference surface, the local normal direction at the path starting point is calculated. Next, the original normal direction of the measurement path segment in the primary normal measurement path sequence is compared with the local normal direction corresponding to the local reference surface. When the direction deviation between the original normal direction and the local normal direction exceeds a preset direction deviation threshold, the measurement path segment needs to be updated. Subsequently, using the crack skeleton point corresponding to the measurement path segment requiring updating as the path starting point, a new normal measurement path is reconstructed according to the local normal direction of the local reference surface. The reconstructed path is then truncated based on a preset path length constraint to obtain the updated measurement path segment. Finally, the updated measurement path segments replace the corresponding original measurement path segments in the first normal measurement path sequence, and all measurement path segments are summarized to obtain the updated second normal measurement path sequence.

[0076] Specifically, this application further proposes a method for generating a secondary normal measurement path sequence, which can effectively correct the deviation of the original normal direction and improve the fit between the measurement path and the crack morphology. The method first obtains the local reference surface at the corresponding time frame from the secondary reference surface sequence based on the path starting position of the crack skeleton point corresponding to the measurement path segment in the primary normal measurement path sequence. Then, based on the spatial fitting result of the local reference surface, the local normal direction at the path starting point is calculated, which more realistically reflects the actual geometric orientation of the crack boundary in that region.

[0077] Next, the system compares the local normal direction with the original normal direction, for example, by calculating the angle between the vectors or by determining the angular deviation between them. If the result shows that the direction deviation exceeds a preset direction deviation threshold (e.g., 5°), it is considered that there is a significant direction error in the measured path segment, and a direction update operation needs to be performed.

[0078] For the path segment that needs to be updated, the system uses its crack skeleton point as the starting point of the path, reconstructs a new normal measurement path according to the local normal direction of the local reference surface, and truncates the path based on the preset path length constraint (e.g., set to the maximum width of the crack or the measurement range allowed by the image) to obtain the updated measurement path segment.

[0079] Finally, the system replaces the corresponding path segments in the original primary normal measurement path sequence with all the updated measurement path segments, and summarizes them together with the path segments that have not been updated, thereby generating a structurally complete and directionally accurate secondary normal measurement path sequence, providing a more stable and reliable directional reference for subsequent accurate crack width measurement.

[0080] As a preferred embodiment, the solution of this application is implemented as follows: First, for each measurement path segment in the primary normal measurement path sequence, the corresponding crack skeleton point and path starting position are extracted. In the secondary reference surface sequence, the local region point set closest to the path starting point in terms of time frame and spatial position is selected, and a local reference surface is constructed using weighted polynomial fitting or radial basis function interpolation. Subsequently, the local normal direction is obtained by calculating the gradient direction of the surface at the path starting point. If the angle between this direction and the original normal direction is greater than 5°, the system determines that the path segment needs to be reconstructed. At this time, a new path that meets the preset length constraint is generated along the local normal direction, starting from the skeleton point, and this new path replaces the original path segment. After all path segments are processed, they are recombined to form a new secondary normal measurement path sequence.

[0081] Through the above technical solution, this application can effectively identify and correct the directional deviation caused by attitude disturbance in a single normal measurement path sequence, making the direction of the measurement path closer to the actual geometric normal direction of the crack. This significantly improves the accuracy and stability of crack width measurement, especially in dynamic acquisition environments, and can more realistically reflect the dynamic changes of the crack edge, thus providing more reliable data support for bridge structural health monitoring.

[0082] In some embodiments described above in this application, normal sampling and width calculation are proposed along a sequence of secondary normal measurement paths. However, in this process, due to the influence of attitude disturbances, the sampling points may be unstable, resulting in inaccurate width measurement. Furthermore, the measurement error lacks quantification, making it impossible to provide reliable crack width assessment results.

[0083] To address this, this application further proposes a method for sampling the crack edge points along an updated quadratic normal measurement path sequence, and calculating the width measurement data of the measurement points based on the attitude perturbation parameter set and the crack edge trajectory set to generate crack width measurement results. The specific steps include: using the crack skeleton point at the starting point of the measurement path segment in the updated quadratic normal measurement path sequence as the sampling center, performing pixel-level equidistant sampling along the normal direction of the measurement path segment within a preset sampling length range to obtain candidate crack edge sampling points located on both sides of the normal direction; calculating the inter-frame displacement of the candidate crack edge sampling points within consecutive image frames in the crack image sequence data, and calculating the change in the angle of the direction vector of the quadratic normal measurement path corresponding to the candidate crack edge sampling points whose inter-frame displacement is less than a preset pixel displacement threshold in adjacent frames; selecting candidate crack edge sampling points whose angle change is less than a preset direction change threshold to obtain the first crack edge sampling point and the second crack edge sampling point located on both sides of the normal direction of the crack skeleton point. The process involves several steps: First, calculating the pixel distance between the first and second crack edge sampling points in the image coordinate system to obtain the original crack width measurement value for the corresponding measurement path segment. Second, based on the timestamp corresponding to the original crack width measurement value, obtaining a matching attitude angle parameter set from the attitude disturbance parameter set, which includes pitch, roll, and yaw angles. Third, based on the angle change of the attitude angle parameter set, calculating the normal direction projection of the line vector connecting the first and second crack edge sampling points to convert the original crack width measurement value into an attitude-corrected crack width measurement value. Fourth, calculating the upper and lower limits of the corrected crack width measurement value based on the change amplitude of the attitude angle parameter set within the corresponding preset time window to obtain the crack width error interval corresponding to the measurement path segment. Finally, summarizing the crack width measurement values ​​obtained along the secondary normal measurement path sequence and their corresponding crack width error intervals to form a crack width measurement value set and a crack width error interval set, which is output as the crack width measurement result.

[0084] Specifically, using the starting crack skeleton point of the measurement path segment in the updated secondary normal measurement path sequence as the sampling center, pixel-level equidistant sampling is performed along the normal direction of the path within a set sampling length range to obtain candidate crack edge sampling points located on both sides of the normal. In the crack image sequence data, the position change of each candidate sampling point is tracked in consecutive frames, the inter-frame pixel displacement is calculated, and the angle change between the corresponding normal direction vectors in adjacent frames is further calculated. By setting pixel displacement thresholds and angle change thresholds, edge point pairs with high temporal stability in multiple frames are selected as the first and second crack edge sampling points, respectively. The pixel distance between these two points in the image coordinate system is calculated to obtain the original crack width measurement value.

[0085] Subsequently, based on the image acquisition timestamp corresponding to the original measurement value, a matching set of attitude angle parameters is obtained from the attitude disturbance parameter set, and the measured values ​​of pitch, roll, and yaw angles are extracted. Based on these attitude angles, the projection of the vector connecting the crack edge points onto the normal direction of the measurement path in three-dimensional space is calculated, and the image pixel width value is corrected to a spatial measurement value. Then, based on the angular change amplitude of the attitude angle parameter set within a set time window, an error estimation model is constructed to obtain the upper and lower limits of the crack width measurement value for that path segment, forming an error interval. Finally, the corrected crack width measurements of all measurement path segments and their corresponding error intervals are summarized, and the output is a set of crack width measurements and a set of error intervals, constituting the final crack width measurement result.

[0086] In a preferred embodiment, when the UAV inspects the bridge structure, the image acquisition module acquires image frames at a frequency of 10Hz and records timestamps. The secondary normal measurement path sequence is generated by the image optimization module. The crack width measurement module selects one path segment, centers on its starting skeleton point, and samples at equal intervals along the normal direction, for example, 20 pixels on each side of the skeleton point, with a 1-pixel interval. Each sampling point is tracked by optical flow or feature point matching algorithms over 5 consecutive frames, recording its inter-frame pixel offset and calculating the angle change with its corresponding normal path. If the inter-frame displacement is less than 1.0 pixel and the angle change is less than 2°, it is considered a stable point pair. At this point, the pixel distance between the point pair is extracted as the original crack width measurement value, and the attitude angle information under the matching timestamp is obtained from the attitude perturbation parameter set. Using the camera extrinsic attitude angle, combined with a 3D projection correction model, the crack width is geometrically corrected to obtain the attitude-corrected width value. Next, calculate the standard deviation of the attitude angle variation within a ±0.5 second window, and estimate its error range based on a preset confidence factor, for example, by multiplying the standard deviation by an empirical weighting coefficient to obtain the upper and lower limits. Summarize the results of all measurement path segments sequentially and output them as image annotations, tables, or data interfaces.

[0087] Through the above technical solutions, this application can achieve high-confidence crack width calculation in a dynamic sampling environment. Analysis of the inter-frame displacement and normal direction angle variation of edge points significantly improves the stability screening accuracy of edge point pairs and suppresses measurement errors caused by image jitter or slight noise. The introduction of the attitude angle parameter set realizes the conversion from the original pixel width to the spatial scale, avoiding projection deviations caused by UAV attitude tilt. The error interval construction mechanism provides a quantifiable description of uncertainty for each measurement result, enhancing the credibility of the results in structural assessment or risk warning.

[0088] In some of the solutions described above in this application, without inter-frame stability analysis and attitude compensation, the crack width measurement results are easily affected by sampling position offset and perspective distortion, causing them to deviate from the true value. Furthermore, without error interval quantification, the system struggles to determine the reliability of the measurement, limiting its engineering application in bridge health monitoring. By constructing a complete width measurement quality assurance system, both detection accuracy and measurement reliability are improved, providing engineering-approved detection capabilities for actual inspection tasks.

[0089] In another embodiment, this application also provides a system for detecting and measuring the width of cracks in bridge structures, such as... Figure 2 As shown, it includes the following modules:

[0090] The image acquisition module is used to acquire image sequence data of cracks in the bridge structure and its image drift data, as well as attitude angle data of the UAV. It performs time-series correlation calculation on the attitude angle data and the image drift data to obtain a set of attitude disturbance parameters.

[0091] The reference surface construction module is used to construct the measurement plane of each frame of image in a stable reference system by means of multi-plane fitting based on the set of attitude perturbation parameters and the set of crack edge points extracted from the crack image sequence data, and generate a reference surface sequence.

[0092] The crack edge module is used to synchronously calibrate and perform inter-frame association on crack edge points in the crack image sequence data based on the primary reference surface sequence and crack image sequence data, and generate a crack edge trajectory set.

[0093] The normal path module is used to construct a set of crack skeleton points based on the crack edge trajectory set and the crack skeleton data extracted from the crack image sequence data; for the measurement points in the crack skeleton point set, the normal measurement path corresponding to the measurement point is reconstructed based on the first-order reference surface sequence to obtain a first-order normal measurement path sequence.

[0094] The image optimization module is used to obtain measurement path segments that characterize the degree of change in the concentrated edge position of the crack edge trajectory based on the primary normal measurement path sequence, and generate a secondary reference surface sequence by refitting and updating the spatial distribution coordinates of the measurement path segments, and update the direction and structure of the measurement path segments in the primary normal measurement path sequence based on the updated secondary reference surface sequence to obtain the secondary normal measurement path sequence.

[0095] The crack width measurement module is used to sample the normal vector of the crack edge points along the updated secondary normal measurement path sequence, and calculate the width measurement data of the measurement points based on the attitude perturbation parameter set and the crack edge trajectory set to generate the crack width measurement result.

[0096] Specifically, the image acquisition module is responsible for acquiring image sequences of bridge cracks, image drift data, and attitude angle data, and generating an attitude perturbation parameter set based on timestamp alignment, providing a basis for subsequent compensation. The reference surface construction module combines the edge point set and attitude perturbation parameters, and uses multi-plane fitting to generate a primary reference surface sequence to establish a stable spatial benchmark. The crack edge module uses the primary reference surface sequence to calibrate crack edge points and perform inter-frame correlation, generating a crack edge trajectory set to ensure the consistency and continuity of crack tracking. The normal path module constructs a primary normal measurement path sequence based on crack trajectory and skeleton point data, combined with the reference surface direction. The image optimization module identifies offset path segments, refits to generate a secondary reference surface sequence, and updates the measurement path direction to obtain a secondary normal measurement path sequence. The crack width measurement module performs normal sampling along this path, and combines attitude perturbation parameters and trajectory information to output the crack width measurement result after attitude compensation. Through the coordinated operation of these modules, the system achieves a high-precision crack measurement closed loop in dynamic environments.

[0097] As a preferred embodiment, the specific implementation of this application is as follows: The image acquisition module consists of a high-resolution industrial camera, a three-axis IMU, and a GPS module. The camera acquires a crack image sequence at 30 frames per second, the IMU outputs three-axis acceleration data, and the GPS provides attitude angle data. The management center processes the multi-source data uniformly based on timestamp alignment and interpolation algorithms to generate an attitude perturbation parameter set. The reference surface construction module preprocesses the images, extracts the crack edge point set, combines it with the attitude perturbation parameters, and uses RANSAC to fit the bridge measurement plane to generate a primary reference surface sequence. The crack edge module maps the edge points to a unified reference system, matches adjacent frame calibration points based on the Kd-tree algorithm to form a crack edge trajectory set, and fills in missing segments through interpolation. The normal path module combines the skeleton point set and the reference surface to generate a primary normal measurement path sequence based on the local tangent direction. The image optimization module identifies path segments with significant crack edge changes, fits the edge curves using a fifth-order polynomial, uses RBF surface interpolation to generate a secondary reference surface sequence, and updates the normal direction accordingly to generate a secondary normal measurement path sequence. The crack width measurement module extracts edge point pairs along the secondary normal direction, calculates the pixel width, and performs projection correction in combination with attitude angle data to output the actual crack width and error range.

[0098] Through the above system design, this application effectively overcomes the impact of attitude disturbances on the accuracy of crack image detection and width measurement. By constructing a unified reference surface and optimizing the measurement path, stable identification of crack edges and accurate orientation modeling are achieved. Based on the disturbance compensation and path update mechanism, the system improves its adaptability to structural deformation and detection errors in dynamic environments, ensuring the reliability and practicality of crack width measurement results in complex detection scenarios.

[0099] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for image detection and width measurement of cracks in bridge structures, characterized in that, Includes the following steps: Acquire image sequence data of cracks in bridge structures and their image drift data, along with attitude angle data of UAVs. Perform time-series correlation calculations on the attitude angle data and the image drift data to obtain a set of attitude disturbance parameters. Based on the set of attitude perturbation parameters and the set of crack edge points extracted from the crack image sequence data, a measurement plane for each frame of image in a stable reference system is constructed by multi-plane fitting, generating a reference plane sequence. Based on the primary reference surface sequence and crack image sequence data, the crack edge points in the crack image sequence data are synchronously calibrated and correlated with each frame to generate a crack edge trajectory set. Based on the crack edge trajectory set and the crack skeleton data extracted from the crack image sequence data, a crack skeleton point set is constructed; for the measurement points in the crack skeleton point set, based on the first-order reference surface sequence, the normal measurement path corresponding to the measurement point is reconstructed to obtain the first-order normal measurement path sequence. Based on the first normal measurement path sequence, a measurement path segment representing the degree of change in the concentrated edge position of the crack edge trajectory is obtained. The spatial distribution coordinates of the measurement path segment are refitted and updated to generate a secondary reference surface sequence. Based on the updated secondary reference surface sequence, the direction and structure of the measurement path segment in the first normal measurement path sequence are updated to obtain the secondary normal measurement path sequence. Along the updated secondary normal measurement path sequence, normal sampling is performed on the crack edge points, and the width measurement data of the measurement points is calculated based on the attitude perturbation parameter set and the crack edge trajectory set to generate the crack width measurement result.

2. The method for image detection and width measurement of bridge structural cracks according to claim 1, characterized in that, The process of obtaining the attitude perturbation parameter set includes: The crack image sequence data includes bridge structure image frames and their corresponding image acquisition timestamps. The attitude angle data includes the measured values ​​of pitch angle, roll angle and yaw angle and their angle acquisition timestamps; Based on the image acquisition timestamp and the angle acquisition timestamp, a time correspondence is established. The attitude angle data and image drift data with different sampling frequencies are converted to a unified time resolution by an interpolation algorithm, and corresponding pairing is performed on the time axis to obtain the attitude disturbance parameter set.

3. The method for image detection and width measurement of bridge structural cracks according to claim 1, characterized in that, The process of generating a reference surface sequence includes: Each frame of the crack image sequence data is subjected to grayscale processing and noise filtering to obtain a preprocessed crack image sequence. The gradient magnitude of the pixel is calculated based on the grayscale gradient components of the pixel in the crack image sequence along the horizontal and vertical directions. Principal component analysis is performed on the pixel set of the preprocessed crack image sequence to obtain the crack principal axis direction vector. Based on the crack principal axis direction vector, the gray-level gradient components of the pixels are calculated by directional projection to obtain the gradient components of the pixels in the direction perpendicular to the crack principal axis. Pixels with gradient magnitudes greater than a first preset threshold are selected as candidate crack edge pixels. Candidate crack edge pixels whose absolute value of the difference between the mean gray values ​​of two preset neighborhoods along the main axis of the crack are greater than a second preset threshold are selected as crack edge pixels, and a crack edge point set is constructed. Based on the attitude angle data and image drift data corresponding to the timestamp of the current image frame in the attitude perturbation parameter set, attitude compensation processing is performed on the crack edge point set to transform the crack edge points from the image acquisition coordinate system to a unified reference coordinate system, thereby obtaining the crack edge point set after attitude perturbation correction. Based on the crack edge point set, a weighted least squares fitting method is used to construct the measurement plane of the corresponding image frame in a stable reference system, and the measurement planes of each image frame are arranged in chronological order to generate a primary reference plane sequence.

4. The method for image detection and width measurement of bridge structural cracks according to claim 1, characterized in that, The process of generating the crack edge trajectory set is as follows: Based on a single reference surface sequence, crack edge points are mapped to a unified spatial reference system to generate an intra-frame calibration point set; Obtain the intra-frame calibration point set of adjacent frame images, calculate the difference between Euclidean distance and grayscale mean, select the intra-frame calibration point set whose Euclidean distance is less than a preset spatial threshold and whose absolute value of the difference between grayscale mean is less than a preset grayscale similarity threshold, and establish inter-frame matching relationship. By connecting the corresponding intra-frame calibration point sets according to the time sequence and inter-frame matching relationship, the crack edge trajectory sequence is obtained; Trajectory segments with consecutive missing values ​​exceeding a preset number of frames in the crack edge trajectory sequence are removed, and linear interpolation is performed to complete and merge the segments based on the calibration point positions and grayscale average values ​​within the preceding and following frames to construct a crack edge trajectory set.

5. The method for image detection and width measurement of bridge structural cracks according to claim 3, characterized in that, The process of constructing the crack skeleton point set includes: Based on the crack principal axis direction vector, the Zhang-Suen skeletonization algorithm is used to analyze the crack region image and obtain the initial skeleton point set. Construct a neighborhood window centered on the initial skeleton point, with a preset pixel 1 along the main axis and a preset pixel 2 along the normal direction. Search for crack edge trajectory points in the corresponding frame within the neighborhood window. Select the initial skeleton point whose pixel distance from the crack edge trajectory point is less than a preset pixel 3 as the effective skeleton point and include it in the crack skeleton data. Based on the crack skeleton data, the data is sorted according to frame order and its projection distance in the principal axis direction. A sliding window is constructed with a preset step size of four pixels. A local fifth-order polynomial fitting method limited by the principal axis direction is used to generate a skeleton fitting curve. Crack skeleton points are sampled at equal intervals to generate a set of crack skeleton points. The skeleton points include their pixel coordinates, the index of the image frame to which they belong, and the information of the principal axis tangent direction and normal direction.

6. The method for image detection and width measurement of bridge structural cracks according to claim 3, characterized in that, The process of obtaining a sequence of normal measurement paths includes: Obtain the crack principal axis direction vector in the frame image where the measurement point is located, construct a normal direction vector based on the crack principal axis direction vector, and generate a line segment with a preset pixel area length in the image along the normal direction vector, starting from the pixel coordinates of the measurement point, as the normal measurement path of the measurement point; Spatial alignment is performed on skeleton points in adjacent image frames that are within a preset range of Euclidean distances from the measurement point. The angle between the corresponding normal direction vectors is calculated, and angles smaller than a preset threshold angle are selected and added to the continuous path trajectory of the measurement point. These angles are then sequentially connected to obtain the normal measurement path of the measurement point in multiple image frames. Finally, the normal measurement paths of all measurement points are summarized to generate a normal measurement path sequence.

7. The method for image detection and width measurement of bridge structural cracks according to claim 1, characterized in that, The process of generating a secondary reference surface sequence includes: Extract the coordinates of the crack skeleton points corresponding to each normal path in the single normal measurement path sequence and the crack edge points on both sides in the crack image sequence data to obtain the edge point side coordinate set; The edge point side coordinate set is aligned on the same path by a preset sliding frame number. The standard deviation of the horizontal and vertical displacement of the edge point within the preset sliding frame number is calculated. The path segment traversed in the corresponding time period with the standard deviation of the horizontal displacement greater than or equal to a preset horizontal pixel or the standard deviation of the vertical displacement greater than or equal to a preset vertical pixel is marked as the measurement path segment. The coordinate data of the skeleton points and edge points within the time period of the measurement path segment are extracted respectively, and the crack section edge curve group is constructed based on the least squares fitting method; A surface interpolation fitting algorithm is used to generate a single-frame reference surface for the crack cross-section edge curve group. The single-frame reference surface is spatially fitted with the skeleton point as the center and the edge point as the boundary. The reference surface is combined into a spatially continuous surface sequence according to the time frame order to obtain a secondary reference surface sequence.

8. The method for image detection and width measurement of bridge structural cracks according to claim 1, characterized in that, The process of obtaining the sequence of secondary normal measurement paths includes: Based on the path starting position of the crack skeleton point corresponding to the measurement path segment in the first normal measurement path sequence, the local reference surface of the path starting point in the corresponding time frame in the second reference surface sequence is obtained, and the local normal direction at the path starting point is calculated based on the spatial fitting result of the local reference surface. The original normal direction of the measurement path segment in the first normal measurement path sequence is compared with the local normal direction corresponding to the local reference surface. When the direction deviation between the original normal direction and the local normal direction exceeds a preset direction deviation threshold, the measurement path segment needs to be updated in direction. Using the crack skeleton point corresponding to the measurement path segment that needs to be updated as the starting point of the path, a new normal measurement path is reconstructed according to the local normal direction of the local reference surface, and the reconstructed path is truncated based on the preset path length constraint to obtain the updated measurement path segment. Replace the original measurement path segment in the first normal measurement path sequence with the updated measurement path segment, and summarize all measurement path segments to obtain the updated second normal measurement path sequence.

9. The method for image detection and width measurement of bridge structural cracks according to claim 1, characterized in that, The process of generating crack width measurement results includes: Using the starting crack skeleton point of the measurement path segment in the updated secondary normal measurement path sequence as the sampling center, pixel-level equidistant sampling is performed along the normal direction of the measurement path segment within a preset sampling length range to obtain candidate crack edge sampling points located on both sides of the normal direction. For the candidate crack edge sampling points in the crack image sequence data, calculate their inter-frame displacement between adjacent frames in consecutive image frames. Calculate the change in the angle of the direction vector of the secondary normal measurement path corresponding to the candidate crack edge sampling points whose inter-frame displacement is less than a preset pixel displacement threshold. Select candidate crack edge sampling points whose angle change is less than a preset direction change threshold to obtain the first crack edge sampling point and the second crack edge sampling point located on both sides of the crack skeleton point normal direction. Calculate the pixel distance between the first crack edge sampling point and the second crack edge sampling point in the image coordinate system to obtain the original crack width measurement value of the corresponding measurement path segment. Based on the timestamp corresponding to the original crack width measurement value, a matching attitude angle parameter set is obtained from the attitude disturbance parameter set, the attitude angle parameter set including pitch angle, roll angle and yaw angle; Based on the angle change of the attitude angle parameter set, the normal direction projection calculation is performed on the line vector connecting the first crack edge sampling point and the second crack edge sampling point to convert the original crack width measurement value into the attitude-corrected crack width measurement value. Based on the change of the attitude angle parameter set within the corresponding preset time window, the upper and lower limits of the corrected crack width measurement value are calculated to obtain the crack width error range corresponding to the measurement path segment. The crack width measurements obtained along the secondary normal measurement path sequence and their corresponding crack width error intervals are summarized to form a crack width measurement set and a crack width error interval set, which are then output as the crack width measurement results.

10. A system for detecting and measuring the width of cracks in bridge structures, specifically applied to the method for detecting and measuring the width of cracks in bridge structures as described in any one of claims 1 to 9, comprising a management center, characterized in that, The management center communication connection includes an image acquisition module, a reference surface construction module, a crack edge module, a normal path module, an image optimization module, and a crack width measurement module. The image acquisition module is used to acquire image sequence data of cracks in the bridge structure and its image drift data, as well as attitude angle data of the UAV. It performs time-series correlation calculation on the attitude angle data and the image drift data to obtain a set of attitude disturbance parameters. The reference surface construction module is used to construct the measurement plane of each frame of image in a stable reference system by means of multi-plane fitting based on the set of attitude perturbation parameters and the set of crack edge points extracted from the crack image sequence data, and generate a reference surface sequence. The crack edge module is used to synchronously calibrate and perform inter-frame association on crack edge points in the crack image sequence data based on the primary reference surface sequence and crack image sequence data, and generate a crack edge trajectory set. The normal path module is used to construct a set of crack skeleton points based on the crack edge trajectory set and the crack skeleton data extracted from the crack image sequence data; for the measurement points in the crack skeleton point set, the normal measurement path corresponding to the measurement point is reconstructed based on the first-order reference surface sequence to obtain a first-order normal measurement path sequence. The image optimization module is used to obtain measurement path segments that characterize the degree of change in the concentrated edge position of the crack edge trajectory based on the primary normal measurement path sequence, and generate a secondary reference surface sequence by refitting and updating the spatial distribution coordinates of the measurement path segments, and update the direction and structure of the measurement path segments in the primary normal measurement path sequence based on the updated secondary reference surface sequence to obtain the secondary normal measurement path sequence. The crack width measurement module is used to sample the normal vector of the crack edge points along the updated secondary normal measurement path sequence, and calculate the width measurement data of the measurement points based on the attitude perturbation parameter set and the crack edge trajectory set to generate the crack width measurement result.

Images