Method for quantitatively detecting concrete cracks based on machine vision

Abstract

The present application relates to the technical field of machine vision detection, and particularly relates to a concrete crack quantitative detection method based on machine vision, which comprises the following steps: firstly, acquiring multiple images and local calibration information of a region to be detected, registering the multiple images, and performing scale mapping to construct a local measurement field; secondly, performing multi-scale local phase analysis and bilateral phase symmetry analysis based on the local measurement field to generate a crack structure response field; thirdly, extracting a center line and matching bilateral boundaries, and correcting the center line by using boundary matching residual error; and finally, performing topological stability analysis and geometric consistency analysis to output crack length, width, area and orientation. The present application can reduce scale conversion error, improve fine crack recognition ability and quantitative result consistency.

Description

Technical Field

[0001] This invention relates to the field of machine vision inspection technology, and more specifically to a method for quantitative detection of concrete cracks based on machine vision. Background Technology

[0002] Concrete components are susceptible to cracking during service due to loads, temperature, shrinkage, and environmental erosion. Crack development affects structural durability, safety assessments, and maintenance decisions; therefore, crack size information is crucial for inspection, maintenance, and lifespan management. Current practices still rely heavily on manual observation and measurement for some detection methods, which are highly subjective and lack consistency in repeated measurements. Some image detection methods depend on single-image interpretation, standardized scaling, or simple edge extraction, making them susceptible to variations in shooting angle, local scale changes, surface texture interference, and weak responses from minute cracks. This results in unstable crack connectivity, significant deviations in width calculations, and difficulty in simultaneously achieving a unified quantitative representation of length, width, area, and orientation. Summary of the Invention

[0003] This invention provides a machine vision-based quantitative detection method for concrete cracks, which addresses at least the problem of how to achieve unified quantitative detection of the length, width, area, and orientation of concrete cracks under conditions of local scale variations, easy omission of fine cracks, and unstable boundary matching.

[0004] A machine vision-based quantitative detection method for concrete cracks, the method comprising: Acquire multiple frames of images and local calibration information of the region to be tested, register the multiple frames of images, and perform scale mapping on the registered multiple frames of images based on the local calibration information to obtain the local metric field; Based on the local metric field, multi-scale local phase analysis and bilateral phase symmetry analysis are performed on the registered multi-frame images to generate the crack structure response field; Centerline extraction and bilateral boundary matching are performed based on the crack structure response field and local metric field to obtain boundary matching residuals. The centerline extraction results are then corrected based on the boundary matching residuals, and the crack topology and width distribution are determined based on the corrected centerline extraction results. Topological stability analysis and geometric consistency analysis were performed based on crack topology, width distribution and boundary matching residuals to obtain the length, width, area and direction of concrete cracks.

[0005] In one possible implementation, scale mapping is performed on the registered multi-frame images to obtain a local metric field, including: extracting calibration feature points from local calibration information; determining the local mapping relationship of the region to be measured based on the calibration feature points; determining the entity scale corresponding to each position in the registered multi-frame images according to the local mapping relationship, and generating a local metric field.

[0006] In one possible implementation, based on a local metric field, multi-scale local phase analysis and bilateral phase symmetry analysis are performed on the registered multi-frame images to generate a crack structure response field. This includes: performing uniform-scale resampling on the registered multi-frame images based on the local metric field; performing multi-scale filtering on the uniform-scale resampled multi-frame images to determine local phase information and local principal direction; performing symmetrical sampling on both sides of the normal direction based on the local principal direction, and generating a crack structure response field based on the symmetrical sampling results.

[0007] In one possible implementation, multi-scale local phase analysis includes: filtering multiple frames of images resampled at a unified scale at multiple scales; determining the local phase information at each location based on the filtering results corresponding to each scale; and determining the local principal direction at each location based on the local phase information.

[0008] In one possible implementation, the two-sided phase symmetry analysis includes: selecting at least one set of symmetrical sampling points on both sides of the normal along the local principal direction; determining the phase change relationship of the symmetrical sampling points respectively; and determining the central response, left boundary response and right boundary response in the crack structure response field based on the phase change relationship and local phase information.

[0009] In one possible implementation, centerline extraction and bilateral boundary matching are performed based on the crack structure response field and the local metric field, including: determining seed points based on the crack structure response field; performing path expansion based on the seed points and the local metric field to obtain candidate centerlines; establishing a cross-sectional search region based on the candidate centerlines, and performing bilateral boundary matching within the cross-sectional search region to obtain boundary matching residuals; correcting the candidate centerlines based on the boundary matching residuals, and determining the crack topology and width distribution based on the corrected candidate centerlines.

[0010] In one possible implementation, path extension is performed based on seed points and local metric fields to obtain candidate centerlines, including: determining the path extension cost at each location based on the crack structure response field; performing path extension along at least two directions based on the path extension cost; retaining a preset number of candidate extension paths, and using interconnected candidate extension paths as candidate centerlines.

[0011] In one possible implementation, bilateral boundary matching is performed within the cross-sectional search area to obtain boundary matching residuals. This includes: extracting the boundary candidate point sets on both sides of the cross-sectional search area; performing minimum cost matching on the boundary candidate point sets based on the spacing relationship, width continuity, and response correspondence of the boundary candidate points in the crack structure response field in the local metric field to obtain the width distribution and boundary matching residuals; and adjusting the path extension cost corresponding to the candidate centerline based on the boundary matching residuals.

[0012] In one possible implementation, topological stability analysis and geometric consistency analysis are performed based on crack topology, width distribution, and boundary matching residuals, including: determining topological stability based on the changes in crack topology corresponding to different response thresholds in the crack structure response field; and determining geometric consistency based on the deviation between the area value obtained by integrating the width distribution along the centerline extraction results and the crack area corresponding to the crack topology.

[0013] In one possible implementation, obtaining the length, width, area, and orientation of concrete cracks includes: removing crack segments from the crack topology if the topological stability or geometric consistency of the crack segments in the crack topology does not meet preset conditions; and determining the length, width, area, and orientation of concrete cracks based on the centerline length, width distribution, crack area, and orientation of the centerline extraction results corresponding to the remaining crack segments.

[0014] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows: By employing multi-frame image registration and local calibration mapping techniques, a unified expression of location-related scales was achieved; by employing multi-scale local phase analysis and bilateral phase symmetry analysis techniques, collaborative identification of crack centers and boundaries was achieved; by employing boundary matching residual backpropagation correction techniques, consistent constraints on centerline and boundary results were achieved; and by employing topological stability analysis and geometric consistency analysis techniques, the screening of abnormal crack segments and stable output of quantitative results were achieved. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the execution flow of the method of the present invention; Figure 2 This is a schematic diagram of the crack topology and bilateral boundaries in the test area in a specific embodiment of the present invention; Figure 3 This is a local phase and bilateral boundary response curve of a certain cross section in a specific embodiment of the present invention; Figure 4 This is a diagram showing the distribution of the main crack width along the path in a specific embodiment of the present invention; Figure 5 This is a statistical graph showing the stability and geometric consistency under different response thresholds in a specific embodiment of the present invention. Detailed Implementation

[0016] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0017] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0018] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0019] Machine vision is a technology that uses imaging devices to acquire image information of a target surface and achieves automatic perception and quantitative analysis through image processing, feature extraction, spatial mapping, and result output. Its core lies not merely in "capturing images," but in transforming the grayscale, texture, edges, shape, and spatial relationships in an image into calculable and comparable structured information, thereby enabling the localization, identification, measurement, and state characterization of the target object. Compared to conventional imaging methods that only record images, machine vision emphasizes the correspondence between imaging results and subsequent calculations, providing a unified data foundation for size quantification, morphological analysis, and result verification. Based on this, this invention applies machine vision to the structural extraction and size calculation of concrete cracks. By establishing a correspondence between image information and entity scale, it achieves quantitative detection of crack length, width, area, and orientation.

[0020] like Figure 1 As shown, a machine vision-based quantitative detection method for concrete cracks includes: Acquire multiple frames of images and local calibration information of the region to be tested, register the multiple frames of images, and perform scale mapping on the registered multiple frames of images based on the local calibration information to obtain the local metric field; In this embodiment of the invention, the area to be tested is first continuously imaged to acquire multiple frames of images under the same field of view, and local calibration information is recorded simultaneously. Preferably, multiple frames are acquired continuously within a short period to reduce the impact of crack morphology changes and environmental disturbances. Subsequently, image registration is performed on the multiple frames to ensure that the same surface position in different frames falls into a unified image coordinate system. After registration, the correspondence between the image position and the entity size of the area to be tested is established using the local calibration information, and this correspondence is written to each image position to form a local metric field. The local metric field provides the basis for the entity scale at different locations within the area to be tested, providing a unified spatial reference for subsequent crack structure analysis, boundary matching, and size calculation.

[0021] Scale mapping is performed on the registered multi-frame images to obtain a local metric field, including: extracting calibration feature points from local calibration information; determining the local mapping relationship of the region to be measured based on the calibration feature points; determining the entity scale corresponding to each position in the registered multi-frame images according to the local mapping relationship, and generating a local metric field.

[0022] In one embodiment, scale mapping can be performed separately after multi-frame image registration is completed. Local calibration information can be a calibration patch attached near the region to be measured, or an array of calibration dots projected onto the surface of the region to be measured. Regardless of the method used, the local calibration information includes at least a plurality of calibration feature points with known entity spacing.

[0023] First, calibration feature points are located in the registered multi-frame images. Localization can be achieved using corner detection, bright spot center extraction, or edge intersection extraction, combined with grayscale thresholding and neighborhood connectivity to remove false points. To ensure mapping stability, it is preferable to retain calibration feature points distributed around the periphery of the test area and spatially dispersed. When the number of calibration feature points is insufficient or concentrated in local areas, valid calibration feature points in the current frame can be reselected, or corresponding calibration feature points from adjacent frames can be used to supplement the points. After completing the extraction of calibration feature points, the local mapping relationship of the test area is determined based on the coordinates of the calibration feature points in the image and the corresponding entity coordinates. For relatively smooth concrete areas, a single-plane mapping relationship can be directly established; for areas with undulating surfaces, honeycomb textures, or edge transitions, the test area can be divided into multiple local sub-regions, and segmented mapping relationships can be established for each sub-region. After establishing the segmented mapping relationship, smooth transitions can be applied between adjacent local sub-regions to avoid abrupt changes in local scale at boundary positions.

[0024] Next, the entity scale corresponding to each position in the registered multi-frame images is calculated based on the local mapping relationship. Specifically, this can be calculated point-by-point according to the image grid, or calculated pixel by pixel and then interpolated and expanded. For each image position, the entity length per unit pixel corresponding to the current position is obtained.

[0025] Finally, the unit pixel entity length corresponding to each location is written into a unified scale matrix to generate a local metric field. The local metric field can be understood as a spatial scale distribution map of the area to be measured, with each location in the local metric field corresponding to a specific entity scale value. Subsequently, in the extraction of the crack centerline, boundary location matching, and the calculation of crack length, width, and area, the corresponding scale value in the local metric field is directly called to complete the entity size conversion, thereby avoiding local measurement distortion caused by using a single scale.

[0026] Based on the local metric field, multi-scale local phase analysis and bilateral phase symmetry analysis are performed on the registered multi-frame images to generate the crack structure response field; In this embodiment of the invention, after establishing the local metric field, crack structure analysis is performed on the registered multi-frame images. During processing, the local metric field is first used to unify the image scale at different locations. Then, under the unified scale condition, multi-scale local phase analysis is performed on the multi-frame images to extract phase and orientation information corresponding to the crack's elongated, continuous, and gradually changing grayscale characteristics. Subsequently, bilateral phase symmetry analysis is performed around the local principal direction at each location, on both sides of the normal, to determine whether the current location simultaneously possesses the structural features of a central trough and bilateral boundaries. Finally, the consistent responses from the multi-frame images are converged to generate a crack structure response field. The crack structure response field is used to characterize the spatial distribution of the crack's central and boundary regions within the test area, providing direct input for subsequent centerline extraction and boundary matching.

[0027] Based on the local metric field, multi-scale local phase analysis and bilateral phase symmetry analysis are performed on the registered multi-frame images to generate the crack structure response field. This includes: performing unified-scale resampling on the registered multi-frame images based on the local metric field; performing multi-scale filtering on the unified-scale resampled multi-frame images to determine local phase information and local principal direction; performing symmetrical sampling on both sides of the normal direction based on the local principal direction, and generating the crack structure response field based on the symmetrical sampling results.

[0028] In one embodiment, when performing multi-scale local phase analysis and bilateral phase symmetry analysis on the registered multi-frame images based on the local metric field, uniform-scale resampling is first performed on the registered multi-frame images, followed by multi-scale filtering and symmetry sampling. The purpose of uniform-scale resampling is to eliminate the influence of different entity sizes corresponding to pixels at different locations within the test area, so that the crack width and orientation features calculated subsequently at different locations are under the same scale reference.

[0029] In practice, the unit pixel entity length at each location in the local metric field can be read first. Then, a target entity resolution can be selected, and the registered multi-frame images can be remapped onto a unified grid according to the target entity resolution. The target entity resolution can be set based on the lower limit of the expected crack width in the area to be measured, so that small cracks remain distinguishable after resampling. During resampling, bilinear interpolation or local neighborhood interpolation can be used to maintain grayscale continuity and avoid introducing obvious jagged edges. After completing the unified-scale resampling, each frame image is filtered at multiple scales. These multiple scales correspond to crack structures with different width ranges, used to accommodate both fine and wider cracks.

[0030] After filtering, a set of response results is generated at each location. Based on the response results at each scale, the local phase information and local principal direction of that location in the current frame are extracted. The local phase information reflects the structural characteristics of grayscale changes, and the local principal direction reflects the extension direction of the linear structure. After obtaining the local principal direction, symmetrical sampling paths are established on both sides of the normal direction, and the response changes at the sampling positions on both sides are read and compared to see if the changes on both sides satisfy the basic structure of the crack cross-section. If a continuous trough appears near the center location, and the two sides show relatively stable boundary changes, then that location is marked as a candidate for a valid crack structure.

[0031] Finally, the analysis results at the same location in multiple frames are merged. During merging, responses that appear stably in multiple frames are prioritized, while isolated anomalous responses appearing only in a single frame are suppressed. After the above processing, the crack structure response field is obtained. The crack structure response field can be stored in the form of a response map with the same size as the image, and the strength of the crack structure at each location is recorded. After this processing, subsequent steps do not need to repeatedly extract crack clues from the original grayscale image; instead, the crack structure response field can be directly used for structure localization and boundary constraints.

[0032] Multi-scale local phase analysis includes: filtering multiple frames of images resampled at a unified scale at multiple scales; determining the local phase information at each location based on the filtering results at each scale; and determining the local principal direction at each location based on the local phase information.

[0033] In one embodiment, multi-scale local phase analysis is used to extract local structural features consistent with crack morphology from multiple frames of images after uniform-scale resampling, and to determine the orientation basis required for subsequent bilateral phase symmetry analysis.

[0034] In practice, the first step is to filter multiple frames of images resampled at a unified scale across multiple scales. These multiple scales can be understood as multiple different analysis windows or different feature width ranges, each scale corresponding to a different type of crack width. Smaller scales are primarily used to capture the narrow line features of fine cracks; larger scales are primarily used to capture the continuous trough features of wider cracks. In actual execution, the images can be processed step-by-step from smallest to largest scale, with directional correlation filtering applied at each scale to obtain the structural response at each location. After filtering across multiple scales, the filtering results for each scale need to be summarized. During summarization, the results from a single scale are not directly used; instead, the strength and stability of the responses at the same location are compared across different scales. If a location shows a weak response only at a few scales, that location is more likely to belong to surface texture or noise; if a location consistently shows a stable response within adjacent scales, that location is more likely to belong to a true crack structure.

[0035] Based on this principle, responses at each scale can be filtered, and scale results with more stable responses can be retained. Then, the local phase information at each location is determined based on the retained scale results. This local phase information primarily characterizes whether the grayscale change at the current location is closer to a linear trough, an edge transition, or no obvious structure. For the crack center, the local phase typically exhibits a more pronounced trough-type response; for the crack edge, the local phase typically exhibits a boundary-type response transitioning from the background to the trough. After determining the local phase information, the local principal direction at each location is determined based on the direction with the strongest response or the direction with the best continuity among adjacent directions. The local principal direction represents the main extension direction of the current linear structure in the plane. To improve stability, the direction results can be smoothed within a small neighborhood to avoid frequent jumps in direction between adjacent pixels due to surface roughness.

[0036] Through the steps described above, multi-scale local phase analysis can provide two fundamental results for each location: whether the current location possesses crack structure characteristics, and the preferred extension direction of the crack structure at the current location. The former is used to determine whether to proceed to subsequent analysis, while the latter is used to define the normal sampling direction in the bilateral phase symmetry analysis, thereby ensuring that subsequent boundary identification unfolds around the correct structural orientation.

[0037] The two-sided phase symmetry analysis includes: selecting at least one set of symmetrical sampling points on both sides of the normal along the local principal direction; determining the phase change relationship of the symmetrical sampling points respectively; and determining the central response, left boundary response and right boundary response in the crack structure response field based on the phase change relationship and local phase information.

[0038] In one embodiment, bilateral phase symmetry analysis is used to further determine whether the local structure conforms to the cross-sectional morphology of the crack, and the linear candidate region obtained from the local phase analysis is refined into the center response, the left boundary response, and the right boundary response.

[0039] In the specific processing, first, using the local principal direction as a reference, determine the normal direction perpendicular to the local principal direction. Then, select at least one set of symmetrical sampling points on both sides of the normal direction. The symmetrical sampling points can be set at fixed intervals or dynamically set according to the expected crack width corresponding to the current scale. The reason for using at least one set of symmetrical sampling points is that the crack boundary may have slight undulations at different locations, and relying solely on sampling at a single interval can easily lead to misjudgment. After setting multiple sets of symmetrical sampling points, both near-side and far-side changes can be observed simultaneously, thereby improving the reliability of structural judgment. After completing the symmetrical sampling, the phase change of each set of symmetrical sampling points is read.

[0040] If there is a clear trough feature near the center location, and boundary changes from the background to the trough appear at similar distances on both sides of the normal direction, then the current location simultaneously meets the basic conditions of a crack center and bilateral boundaries. If there is a clear boundary change on one side while the other side is missing, or if the changes on both sides are significantly asymmetrical, then the current location is more likely to belong to ordinary surface texture, unilateral shadow edge, or local contamination boundary. Based on this judgment approach, each location is classified. For locations that satisfy the bilateral phase symmetry relationship, the center location is recorded in the center response, the transition position on the left side of the normal direction is recorded in the left boundary response, and the transition position on the right side of the normal direction is recorded in the right boundary response.

[0041] For locations that do not satisfy the bilateral phase symmetry relationship, the response value is reduced or not included in the crack structure response field. To avoid local isolation and misjudgment, after completing the single-point analysis, the spatial continuity between the center response and the left and right boundary responses can be checked within the neighborhood. If the center response of adjacent locations can extend continuously along the local principal direction, and the left and right boundary responses remain basically stable in terms of normal distance, then the region is retained as a valid crack structure; if the response shows discrete jump points or the left and right boundaries cannot form a continuous correspondence within the neighborhood, then the region is treated as a low-confidence region. After the bilateral phase symmetry analysis, the crack structure response field is no longer just a simple strength-weakness map, but simultaneously contains structural information of the crack center and both sides of the boundary. In this way, subsequent steps can directly rely on the center response when extracting the centerline, and can directly rely on the left and right boundary responses when performing boundary matching, thereby reducing redundant calculations and improving the coherence of crack geometry reconstruction.

[0042] Centerline extraction and bilateral boundary matching are performed based on the crack structure response field and local metric field to obtain boundary matching residuals. The centerline extraction results are then corrected based on the boundary matching residuals, and the crack topology and width distribution are determined based on the corrected centerline extraction results. In this embodiment of the invention, after the crack structural response field is formed, the crack geometry restoration stage begins. This stage is based on the center and boundary responses provided by the crack structural response field, and combined with the entity scale constraints provided by the local metric field. First, a centerline characterizing the crack's extension trend is extracted, and then bilateral boundary matching is performed around the centerline. After bilateral boundary matching is completed, the boundary matching residual is calculated, and the centerline extraction result is corrected using this residual to ensure the centerline position is consistent with the bilateral boundary positions. After correction, the crack topology is determined based on the stable centerline connectivity, and the width distribution is determined based on the bilateral boundary spacing corresponding to each center position. This process allows the crack's connectivity and width variations to be obtained synchronously within the same processing chain.

[0043] Centerline extraction and bilateral boundary matching based on crack structural response field and local metric field include: determining seed points based on crack structural response field; expanding the path based on seed points and local metric field to obtain candidate centerlines; establishing a cross-sectional search region based on candidate centerlines and performing bilateral boundary matching within the cross-sectional search region to obtain boundary matching residuals; correcting candidate centerlines based on boundary matching residuals and determining crack topology and width distribution based on corrected candidate centerlines.

[0044] In one embodiment, centerline extraction and bilateral boundary matching are not performed directly on the original grayscale image, but rather under the joint constraints of the crack structure response field and the local metric field. The added constraint is that seed points are first determined using the crack structure response field, then candidate centerlines are formed using the seed points, and finally bilateral boundary matching is performed around the candidate centerlines. The boundary matching results are then used in reverse to correct the centerline. The purpose of this setup is to process the crack center position and crack boundary position within the same computational path, avoiding the cumulative errors caused by fixing the centerline first and then independently finding the boundaries.

[0045] In the specific processing, pixels with strong central responses and relatively stable spatial positions are first extracted from the crack structure response field as candidate seed points. These candidate seed points are then merged to remove duplicate points that are too close together, retaining representative central positions within the local area. Subsequently, starting from each seed point, and combining the entity scale information in the local metric field, a path is extended in the direction the crack may extend, forming a set of candidate center lines. These candidate center lines are not directly used as the final result but rather as the basis for subsequent boundary matching.

[0046] After obtaining the candidate centerline, a cross-sectional search region is established segment by segment along the extension direction of the candidate centerline. The cross-sectional search region is typically approximately perpendicular to the candidate centerline and covers the possible boundary locations on both sides of the candidate centerline. Then, in each cross-sectional search region, candidate boundary locations on the left and right sides are searched, and bilateral boundary matching is performed according to the solid scale given by the local metric field, the width variation of adjacent locations, and the strength of the boundary response in the crack structure response field. After bilateral boundary matching is completed, the boundary pairing result at the current location and the difference between the current pairing result and the candidate centerline can be obtained. This difference is the boundary matching residual.

[0047] Boundary matching residuals reflect whether the current candidate centerline is located in a reasonable middle region between the two boundaries. A small boundary matching residual indicates that the candidate centerline and boundary pairing is consistent; a consistently large boundary matching residual indicates that the current candidate centerline is offset. In this case, the candidate centerline is corrected based on the boundary matching residuals, gradually moving it closer to the middle region of the boundary pairing. After correction, the crack topology is formed based on the connectivity and branching relationships between adjacent center positions. Simultaneously, the width distribution, varying along the crack extension direction, is formed based on the entity distance between the two boundaries of each center position. In this way, the centerline, boundaries, topology, and width mutually constrain each other in the same processing, resulting in a more coherent crack geometry restoration.

[0048] Path extension based on seed points and local metric fields to obtain candidate centerlines includes: determining the path extension cost at each location based on the crack structure response field; extending the path along at least two directions based on the path extension cost; retaining a preset number of candidate extension paths, and using interconnected candidate extension paths as candidate centerlines.

[0049] In one embodiment, the path extension stage further defines how candidate centerlines are formed. The new limitation is that instead of simply connecting points one by one according to the local center response, a path extension cost is first constructed, and then multi-directional extension is performed based on the path extension cost, retaining multiple candidate extension paths. This approach avoids the crack being directly truncated when it bifurcates, is occluded, or the local response weakens, and also facilitates the subsequent selection of more reasonable centerlines based on boundary matching results.

[0050] In practice, the effective locations in the crack structure response field are traversed first, and the path extension cost is calculated for each location. The path extension cost can be calculated by considering the strength of the central response at the current location, the degree of directional consistency between the current location and its adjacent locations, the change in the entity scale corresponding to the current location, and the spatial continuity between the current location and the generated paths. The stronger the central response, the smoother the directional change, and the more stable the scale change, the smaller the path extension cost; conversely, the greater the path extension cost.

[0051] After calculating the path extension cost, the path is extended from each seed point in at least two directions. These at least two directions generally correspond to the positive and negative directions of the main crack direction, allowing the seed point to grow in both directions. If there is a significant bifurcation trend within the area to be tested, additional extension directions can be retained if the local costs are similar. During path extension, a unique extension result is not determined directly at once. Instead, the path extension costs of adjacent candidate positions are compared segment by segment, prioritizing the connection of positions with lower path extension costs and continuous directions. To avoid the path being influenced by local noise, a minimum continuous length and a maximum allowable turning angle can be set during the path extension process. If a segment is too short or the turning angle between adjacent segments is too large, that segment can be downgraded to a low-confidence path.

[0052] As the path extends forward, multiple candidate extension paths are formed. When screening these candidate extension paths, a preset number are retained. This preset number can be pre-set based on the complexity of the cracks within the test area. Subsequently, spatially adjacent, directionally continuous candidate extension paths that can form a complete linear structure are merged based on connectivity, forming candidate centerlines. Isolated paths that cannot form stable connections with other paths and are significantly shorter can be directly eliminated or temporarily retained as branches to be verified. The candidate centerlines obtained in this way not only preserve the continuity of the crack backbone but also provide sufficient adjustment space for subsequent boundary matching and centerline correction. Especially when cracks have local occlusion, blurred edges, or small branches, retaining multiple possible extension results during the path expansion stage, followed by further confirmation in subsequent steps, can reduce early accidental deletion.

[0053] Perform bilateral boundary matching within the cross-sectional search area to obtain boundary matching residuals, including: extracting the boundary candidate point sets on both sides of the cross-sectional search area respectively; performing minimum cost matching on the boundary candidate point sets based on the spacing relationship, width continuity and response correspondence of the boundary candidate points in the crack structure response field in the local metric field to obtain the width distribution and boundary matching residuals; and adjusting the path extension cost corresponding to the candidate centerline according to the boundary matching residuals.

[0054] In one embodiment, the bilateral boundary matching stage further defines the formation method of the width distribution and boundary matching residuals. The new limitation is that the boundaries are not determined independently through a unilateral search, but rather candidate point sets for both sides are extracted separately within the cross-sectional search area. Then, left and right boundary pairs are formed through minimum cost matching, and the path extension cost corresponding to the candidate centerline is adjusted in reverse using the matching results. The reason for this design is that the crack width is essentially determined by the correspondence between the left and right boundaries within the same cross-section. If the left and right boundaries are determined independently, it is easy for one side to fall on the true boundary and the other side to fall on the texture edge.

[0055] In the specific processing, samples are first taken along the candidate centerline at fixed intervals. A cross-sectional search region is established at each sampling location. The cross-sectional search region is usually centered on the sampling location, expanding in a direction approximately perpendicular to the candidate centerline, and covering a preset range on both sides of the candidate centerline. Then, the left boundary candidate point set is extracted on the left side of the cross-sectional search region, and the right boundary candidate point set is extracted on the right side of the cross-sectional search region. The extraction of boundary candidate points is mainly based on the left and right boundary responses in the crack structure response field, while also incorporating grayscale transition continuity to remove isolated noise points. After obtaining the candidate point sets for both sides, the candidate points on the left and right sides are combined and evaluated.

[0056] During the evaluation, first, it is checked whether the solid spacing between the left and right candidate points falls within a reasonable range; then, it is checked whether the width change between the current cross-section and the adjacent cross-section is continuous; simultaneously, it is checked whether the boundary responses of the left and right candidate points in the crack structure response field can form corresponding boundary structures. Based on the above three factors, the matching cost is calculated for each pair of candidate boundary points. The smaller the matching cost, the more the current boundary pair conforms to the true crack boundary. Subsequently, the boundary pair with the smallest matching cost within the current cross-section is selected as the final boundary matching result for that location. The solid distance between the left and right boundaries is the local width at the current location, and the local widths at each location are arranged along the candidate centerline to form a width distribution. At the same time, it is also necessary to determine the boundary matching residual based on the deviation between the middle position of the final boundary pair and the current candidate centerline position.

[0057] If the midpoint of the boundary pair deviates significantly from the candidate centerline, it indicates that the current candidate centerline does not fall within the central region of the actual crack. In this case, the boundary matching residual is fed back to the path extension stage to adjust the path extension cost of the corresponding position of the current candidate centerline and its neighborhood. A larger boundary matching residual results in a more significant increase in the corresponding path extension cost; a smaller residual maintains the original path extension cost. After cost adjustment, the local position of the candidate centerline can be re-examined, and the candidate centerline can be moved closer to the midpoint of the boundary pair. In this way, bilateral boundary matching not only outputs the width distribution but also directly participates in centerline correction. By using the boundary information in reverse for centerline updating, the final centerline can be closer to the actual crack axis, thus ensuring consistency between the crack topology and width distribution.

[0058] Topological stability analysis and geometric consistency analysis were performed based on crack topology, width distribution and boundary matching residuals to obtain the length, width, area and direction of concrete cracks.

[0059] In this embodiment of the invention, after constructing the crack topology, width distribution, and boundary matching residuals, the quantitative result confirmation stage begins. This stage first performs topological stability analysis on the crack topology, and then performs geometric consistency analysis on the correspondence between the width distribution and the crack area. Through these two analyses, crack segments with unstable connectivity or unbalanced size relationships are eliminated. Subsequently, only crack segments that meet the conditions are quantitatively calculated. The length is obtained by accumulating the centerline, the width by statistically analyzing the width distribution, the area by accumulating the crack regions, and the orientation is determined by the direction of the centerline extension. This ensures that the final output crack size results are consistent with the crack geometry and reduces the impact of local mismatches on the results.

[0060] Topological stability analysis and geometric consistency analysis are performed based on crack topology, width distribution, and boundary matching residuals. This includes: determining topological stability based on the changes in crack topology corresponding to different response thresholds in the crack structure response field; and determining geometric consistency based on the deviation between the area value obtained by integrating the width distribution along the centerline extraction results and the crack area corresponding to the crack topology.

[0061] In one embodiment, the added constraint in this stage is that the crack topology and width distribution obtained in the previous stage are not directly used as the final result. Instead, topological stability analysis and geometric consistency analysis are introduced first to screen the crack segments for validity. The purpose of this setting is to check the structural reliability of the crack segments again before quantitative output, so as to avoid unstable crack segments caused by local noise, transient fractures, or boundary mismatches being directly included in the final result.

[0062] In the specific processing, multiple response thresholds are first set based on the crack structure response field, and candidate crack regions are generated sequentially under multiple threshold levels in descending order. Each response threshold corresponds to a set of crack topologies. Subsequently, the changes in the crack trunk, branches, endpoints, and junctions are compared under different response thresholds. If the same crack segment can remain connected under multiple adjacent response thresholds and the centerline position changes little, the crack segment is determined to be topologically stable; if a crack segment only appears briefly under individual response thresholds, or breaks, merges, or shifts significantly when the threshold changes slightly, the crack segment is determined to be topologically unstable. To improve the executability of the judgment, the number of times each crack segment is retained, the change in connectivity length, and the endpoint displacement under different response thresholds can be further recorded, and these three types of results can be combined for stability determination. After completing the topological stability analysis, geometric consistency analysis is then performed. The core of geometric consistency analysis is to check whether the area value calculated from the width distribution is consistent with the actual crack area corresponding to the crack topology.

[0063] Specifically, the width distribution can be read segment by segment along the centerline and accumulated along the length of the centerline to obtain the integrated area; simultaneously, the crack area can be directly calculated based on the crack region enclosed by the crack topology. The geometric consistency deviation can be calculated using the following formula: ;

[0064] in, For geometric consistency deviation, For the integral area, The crack area is denoted as .

[0065] When the geometric consistency deviation is small, it indicates that the width distribution is consistent with the crack region morphology. When the geometric consistency deviation is consistently large, it indicates that the current crack segment has boundary offset, centerline deviation, or local area incorrect connection. In this case, boundary matching residuals can also be used for cross-judgment. If both the boundary matching residual and the geometric consistency deviation are large, the current crack segment is directly marked as a low-confidence crack segment. If the boundary matching residual is small but the geometric consistency deviation is slightly large, only the confidence level of the current crack segment can be reduced, and it can be retained for further screening. Through the above process, topological stability analysis is responsible for constraining the connectivity reliability of the crack segment, and geometric consistency analysis is responsible for constraining the dimensional self-consistency of the crack segment. Together, they form the screening basis before the final quantitative output.

[0066] The process involves obtaining the length, width, area, and orientation of concrete cracks, including: removing crack segments from the crack topology if the topological stability or geometric consistency of the crack segments does not meet preset conditions; and determining the length, width, area, and orientation of concrete cracks based on the centerline length, width distribution, crack area, and orientation of the centerline extraction results corresponding to the remaining crack segments.

[0067] In one embodiment, this stage further defines the generation method for length, width, area, and orientation. The new limitation is that instead of uniformly counting all crack segments, crack segments that do not meet topological stability or geometric consistency requirements are first removed from the crack topology, and then the remaining crack segments are quantitatively calculated. This is because the crack topology may simultaneously contain main cracks, short branches, and pseudo-crack segments introduced by noise. If these are not removed first, local abnormal segments will directly affect the length accumulation, width statistics, and orientation determination.

[0068] In the specific processing, each crack segment in the crack topology is first read segment by segment, and the topological stability and geometric consistency results corresponding to that crack segment are checked. If the topological stability of a crack segment is lower than a preset lower limit, it indicates that the connectivity of the crack segment is unstable under different response thresholds, and the crack segment is removed from the crack topology. If the topological stability of a crack segment meets the requirements, but the geometric consistency deviation exceeds a preset upper limit, it indicates that the width distribution of the crack segment is inconsistent with the crack area, and the crack segment is also removed. After the removal is completed, the remaining crack segments constitute the final effective crack structure. Subsequently, quantitative calculations are performed on each crack segment in the final effective crack structure.

[0069] When calculating length, the solid distance between adjacent sampling points is accumulated segment by segment along the centerline to obtain the centerline length of a single crack segment. If multiple crack segments are connected end-to-end and belong to the same main trunk, they can be accumulated to form the total length. When calculating width, the width distribution of the corresponding crack segment is directly read, and the average width, maximum width, or width variation along the crack is output as needed. When calculating area, the crack region area of ​​the current crack segment in the crack topology is calculated, and the areas of multiple adjacent crack segments corresponding to the same main trunk are merged when necessary. When calculating orientation, it is determined based on the extension direction of the centerline in the plane. Specifically, the local direction of each segment of the centerline can be extracted first, and then the main orientation can be determined based on the continuous extension trend of the main trunk of the centerline.

[0070] For crack structures with branches, the main orientation is preferentially adopted from the direction of the main crack segment, while the branch directions are retained as supplementary directional information. If there is only one remaining crack segment, the orientation is directly taken from the centerline direction of that segment; if there are multiple remaining crack segments, the dominant orientation can be determined according to the proportion of crack segment length. After the above processing, the final length, width, area, and orientation are all based on the selected effective crack segments. This preserves the main geometric information of the real cracks while avoiding interference from unstable crack segments on the final quantitative results.

[0071] In one specific embodiment, taking a concrete component with a main crack and a branch crack on its surface as an example, the area to be measured is 240 mm by 135 mm. The component material is common bridge deck repair concrete, with a slightly rough texture and local color difference on the surface. An industrial camera with a resolution of 4032×3024 and a lens focal length of 25 mm was used for data acquisition, continuously acquiring three frames. A slight horizontal shift between adjacent frames creates a parallax displacement of approximately 6 mm. Four local calibration points were arranged at the four corners of the area to be measured, with the corresponding entity coordinates being the upper left, upper right, lower right, and lower left corners, respectively. After registration of the three frames, the average reprojection error of the four calibration points was 0.78 pixels. After establishing a local mapping relationship based on the four calibration points, a local metric field was obtained, where the entity scale at the center of the area to be measured is 0.0411 mm per pixel, and the entity scale at the four corners ranges from 0.0396 mm per pixel to 0.0432 mm per pixel. The results indicate that slight scale variations exist within the same field of view, and directly using a single scale will introduce local measurement errors. However, a local measurement field can provide a unified basis for subsequent size conversions at various locations.

[0072] After establishing the local metric field, the three frames of images were resampled at a uniform scale, and multi-scale local phase analysis and bilateral phase symmetry analysis were performed under the uniform scale condition. The multi-scale analysis window covers the common range of fine and medium-width cracks. After processing, a continuous crack structure response field was obtained in the test area. This crack structure response field simultaneously provides the central structure response and the bilateral boundary responses. Subsequently, seed points were selected from the region with strong central structure response and stable position in the three frames, resulting in a total of 9 effective seed points, including 7 corresponding to the main crack trunk and 2 corresponding to the branch cracks. Path expansion was performed starting from these seed points, and after a residual correction of the centerline based on the bilateral boundary matching results, the cumulative length of the main crack centerline was 220.35 mm, and the cumulative length of the branch crack centerline was 59.25 mm. The main crack extends from the upper left to the lower right with an equivalent strike angle of 339.9°; the branch crack extends from the bifurcation point to the upper right with an equivalent strike angle of 37.5°.

[0073] like Figure 2As shown in the figure, this diagram is created by overlaying the corrected centerline, bilateral boundaries, and calibration points after local metric field conversion. The four square points in the figure are calibration points, located near the four corners of the area to be measured; the solid blue line represents the main crack centerline, and the dashed blue line represents the branch crack centerlines; the red and orange banded boundaries outside the two centerlines represent the crack boundary range obtained through bilateral boundary matching. The main crack extends continuously from the upper left to the lower right, branching at a cumulative length of approximately 120 mm to 125 mm. The "cross-sectional measurement points" in the figure are the sampling locations for subsequent cross-sectional analysis, corresponding to a representative measurement section in the middle of the main crack trunk. The figure clearly shows that the centerline is located in the middle region of the bilateral boundaries, and the trunk and branches near the branching points maintain a clear connection, indicating that the centerline correction based on the boundary matching residuals can stably restore the crack topology.

[0074] against Figure 2 A cross-sectional search area was established at the measurement points on the cross section, approximately perpendicular to the centerline, and the local phase and boundary responses on both sides of the cross section were read. In this cross section, the left matching boundary is located 0.34 mm to the left of the center position, and the right matching boundary is located 0.38 mm to the right of the center position, corresponding to a local crack width of 0.72 mm. Without boundary matching residual backpropagation, the initial centerline is offset by approximately 0.07 mm relative to the midpoints of the left and right boundaries; after one residual correction, this offset is reduced to 0.01 mm. This demonstrates that the boundary matching residual not only evaluates the reliability of the current pairing result but also directly participates in the centerline position correction, making the centerline closer to the true crack axis.

[0075] like Figure 3 As shown in the figure. Figure 2 The cross-sectional search area corresponding to the mid-section measurement point is extracted. The horizontal axis represents the cross-sectional position, and the vertical axis represents the normalized response value. The blue curve represents the central structural response, the red curve represents the left boundary response, and the green curve represents the right boundary response. The blue curve forms a peak at the center position, indicating that there is a clear crack center structure at the current location; the red and green curves form boundary peaks on the left and right sides of the center, respectively, and the peak positions are consistent with the final matching boundary positions. The three vertical reference lines in the figure show the left matching boundary, the center position, and the right matching boundary, respectively. This figure reflects that the current cross-section simultaneously satisfies the structural conditions of a central trough and paired bilateral boundaries, therefore, this cross-section can be used as an effective width measurement section.

[0076] After matching the bilateral boundaries of all cross sections, the width distribution of the main crack was accumulated along the centerline of the main crack. The width of the main crack ranged from 0.22 mm to 0.72 mm, with an average width of 0.41 mm. The maximum width occurred at a cumulative length of approximately 135.82 mm, precisely near the branch influence area. The width of the branch cracks ranged from 0.18 mm to 0.40 mm, with an average width of 0.30 mm. The integrated area of ​​the widths of the main crack and branch cracks was 109.4 square millimeters. The crack area directly obtained from the crack topology region was 113.1 square millimeters. The deviation between the two is small, indicating that the width distribution is basically consistent with the crack region range.

[0077] like Figure 4 As shown in the figure, this graph is plotted by expanding the total width results of all cross-sections along the cumulative length of the centerline. The blue curve represents the actual width of each cross-section, the orange curve represents the smoothed width variation trend, and the green dashed line indicates an average width of 0.41 mm. The light-shaded area represents the branch influence zone, where the main crack width increases significantly, reaching a maximum width of 0.72 mm at approximately 135.82 mm. Before entering the branch influence zone, the width gradually increases from approximately 0.22 mm to over 0.40 mm; after crossing the branch influence zone, the width gradually decreases again. This result is consistent with... Figure 2 The consistent topological morphology indicates that the width distribution is not a random fluctuation, but corresponds to the actual geometric structure of the crack.

[0078] After obtaining the crack topology and width distribution, the crack retention under different response thresholds was compared. Five response thresholds were specifically set: 0.35, 0.40, 0.45, 0.50, and 0.55. When the response threshold was 0.35, the effective connectivity length ratio was 100%, but there were many local false connections, resulting in a geometric consistency deviation of 5.6%. When the response threshold was 0.40, the effective connectivity length ratio was 99.1%, and the geometric consistency deviation decreased to 3.9%. When the response threshold was 0.45, the effective connectivity length ratio was 97.6%, and the geometric consistency deviation further decreased to 3.3%. When the response threshold was further increased to 0.50 and 0.55, local thin segments of the crack began to be truncated, and the effective connectivity length ratio decreased to 92.4% and 84.8%, respectively, while the geometric consistency deviation increased to 4.8% and 8.1%, respectively. Therefore, in this embodiment, a response threshold of 0.45 is preferred.

[0079] Geometric consistency deviation is calculated using the following formula: ;

[0080] in, For geometric consistency deviation, The integral area is obtained by integrating the width distribution along the centerline. The crack area is obtained from the statistics of the crack topological region.

[0081] In this embodiment, a preferred response threshold of 0.45 is selected. =109.4 square millimeters, =113.1 square millimeters, and after substituting, the geometric consistency deviation is 3.27%. This deviation is at a low level, indicating that there is good self-consistency between the width calculation result and the crack area.

[0082] like Figure 5 As shown in the figure, this graph was obtained by repeatedly performing topology filtering and area consistency calculations on five sets of response thresholds. The blue bars represent the percentage of effective connected length, and the red line represents the geometric consistency deviation. As the response threshold increases from 0.35 to 0.45, the geometric consistency deviation gradually decreases, indicating that false connections caused by excessively low thresholds are gradually suppressed. When the response threshold continues to rise to 0.50 and 0.55, the blue bars decrease significantly, and the red line rises again, indicating that true fine fracture segments begin to be over-removed. The "optimal threshold" in the figure corresponds to 0.45, at which point the percentage of effective connected length is relatively high and the geometric consistency deviation is minimal, making it more suitable for the final quantitative output.

[0083] Further comparison with the same set of data revealed that, with boundary matching residual backpropagation correction disabled, the position of the maximum width point of the main crack shifted by approximately 0.12 mm relative to the final result, increasing the geometric consistency deviation to 8.4%. With boundary matching residual backpropagation correction enabled, the shift in the maximum width point position decreased to 0.01 mm, and the geometric consistency deviation dropped to 3.27%. This demonstrates that boundary matching residual backpropagation not only improves the local centerline position but also significantly enhances the self-consistency of the overall dimensional results. Ultimately, this embodiment outputs a main crack length of 220.35 mm, a branch crack length of 59.25 mm, an average main crack width of 0.41 mm, a maximum main crack width of 0.72 mm, and a total crack area of ​​113.1 square millimeters. The main crack direction is from the upper left to the lower right, and the branch direction is from the bifurcation point to the upper right. This set of implementation data shows that the present invention can simultaneously provide the crack connectivity, width variation, and area results in the same processing step, with clear geometric correspondences between the results, making it suitable for quantitative detection of concrete cracks.

[0084] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0085] The above are merely embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for quantitative detection of concrete cracks based on machine vision, characterized in that, The method includes: Acquire multiple frames of images and local calibration information of the region to be tested, register the multiple frames of images, and perform scale mapping on the registered multiple frames of images based on the local calibration information to obtain the local metric field; Based on the local metric field, multi-scale local phase analysis and bilateral phase symmetry analysis are performed on the registered multi-frame images to generate the crack structure response field; Based on the crack structure response field and the local metric field, centerline extraction and bilateral boundary matching are performed to obtain boundary matching residuals. The centerline extraction results are then corrected based on the boundary matching residuals. Based on the corrected centerline extraction results, the crack topology and width distribution are determined. Based on the crack topology, the width distribution, and the boundary matching residual, topological stability analysis and geometric consistency analysis are performed to obtain the length, width, area, and orientation of the concrete cracks.

2. The method according to claim 1, characterized in that, The process of scaling the registered multi-frame images to obtain the local metric field includes: Extract calibration feature points from the local calibration information; The local mapping relationship of the region to be tested is determined based on the calibration feature points; Based on the local mapping relationship, the entity scale corresponding to each position in the registered multi-frame image is determined, and the local metric field is generated.

3. The method according to claim 1, characterized in that, The step of performing multi-scale local phase analysis and bilateral phase symmetry analysis on the registered multi-frame images based on the local metric field to generate the crack structure response field includes: Based on the local metric field, the registered multi-frame images are resampled at a uniform scale. Multi-scale filtering is performed on multiple frames of images after uniform scale resampling to determine local phase information and local principal direction; Symmetrical sampling is performed on both sides of the normal direction based on the local principal direction, and the crack structure response field is generated based on the symmetric sampling results.

4. The method according to claim 3, characterized in that, The multi-scale local phase analysis includes: Filter multiple frames of images resampled at a uniform scale at multiple scales; The local phase information at each location is determined based on the filtering results corresponding to each scale; The local principal direction at each position is determined based on the local phase information.

5. The method according to claim 3, characterized in that, The two-sided phase symmetry analysis includes: At least one set of symmetrical sampling points is selected on both sides of the normal of the local main direction; Determine the phase change relationship of the symmetrical sampling points respectively; The center response, left boundary response, and right boundary response in the crack structure response field are determined based on the phase change relationship and the local phase information.

6. The method according to claim 1, characterized in that, The process of extracting the centerline and matching the bilateral boundaries based on the crack structure response field and the local metric field includes: Seed points are determined based on the crack structure response field; Based on the seed points and the local metric field, path expansion is performed to obtain candidate centerlines; A cross-sectional search region is established based on the candidate centerline, and bilateral boundary matching is performed within the cross-sectional search region to obtain the boundary matching residual. The candidate centerline is corrected based on the boundary matching residual, and the crack topology and width distribution are determined based on the corrected candidate centerline.

7. The method according to claim 6, characterized in that, The process of extending the path based on the seed point and the local metric field to obtain the candidate centerline includes: The path extension cost at each location is determined based on the crack structure response field. Based on the path extension cost, the path is extended in at least two directions; A preset number of candidate extension paths are retained, and interconnected candidate extension paths are used as candidate centerlines.

8. The method according to claim 6, characterized in that, The step of performing bilateral boundary matching within the cross-sectional search area to obtain the boundary matching residual includes: Extract the boundary candidate point sets on both sides of the cross-sectional search region respectively; Based on the spacing relationship and width continuity in the local metric field, and the response correspondence of the boundary candidate points in the crack structure response field, minimum cost matching is performed on the boundary candidate point set to obtain the width distribution and the boundary matching residual. Adjust the path extension cost corresponding to the candidate centerline based on the boundary matching residual.

9. The method according to claim 1, characterized in that, The topological stability analysis and geometric consistency analysis based on the crack topology, the width distribution, and the boundary matching residuals include: The topological stability is determined based on the changes in the crack topology corresponding to different response thresholds in the crack structure response field. Geometric consistency is determined by the deviation between the area value obtained by integrating the width distribution along the centerline and the crack area corresponding to the crack topology.

10. The method according to claim 9, characterized in that, The obtained length, width, area, and orientation of the concrete cracks include: If the topological stability or geometric consistency of a crack segment in the crack topology does not meet the preset conditions, the crack segment is removed from the crack topology. Based on the centerline length of the remaining crack segment, the width distribution, the crack area, and the direction of the centerline extraction result, the length, width, area, and direction of the concrete crack are determined.

Images