Bridge pavement crack intelligent detection and analysis method based on machine vision

Abstract

The present application belongs to the technical field of intelligent crack detection, and relates to a bridge pavement crack intelligent detection and analysis method based on machine vision. The present application identifies candidate crack regions through image acquisition and processing, determines real crack regions through division and calculation, further divides spatial districts, classifies adjacent crack relationships, and finally generates a crack propagation rate distribution map. The method identifies real cracks by confirming crack convergence points and combining neighborhood geometric morphological features, effectively solves the false edge misdetection problem caused by ignoring crack convergence characteristics in the prior art, significantly reduces the misdetection rate, greatly improves the accuracy and robustness of crack detection, and generates an expansion rate distribution map by dividing spatial districts, classifying adjacent crack relationships, and combining a mechanical model, overcoming the limitations of the prior art in lacking adjacent crack spatial relationship identification and being difficult to analyze the expansion driving mechanism, and can accurately judge crack expansion behavior, providing a solid basis for engineering structure operation and maintenance decision-making.

Description

Technical Field

[0001] This invention belongs to the field of intelligent crack detection technology, and relates to a machine vision-based intelligent detection and analysis method for bridge pavement cracks. Background Technology

[0002] As a core component of transportation infrastructure, the structural health of bridges directly affects their durability and operational safety. Under the combined effects of long-term loads, environmental erosion, and material aging, cracks inevitably develop in bridge pavements. These cracks are key indicators for assessing the integrity and safety of bridge structures. Therefore, efficient and accurate detection and analysis of bridge pavement cracks are crucial.

[0003] During crack detection, non-structural texture features such as stains, scratches, and road markings often create suspected crack areas with shapes highly similar to actual cracks, leading to false positives and false negatives, affecting the accuracy and completeness of the detection. Therefore, effective identification of suspected crack areas is crucial, as it can effectively classify and suppress non-crack interference areas, improving the anti-interference capability and coverage of the detection system.

[0004] In recent years, various technical methods have been developed for identifying suspected crack areas. For example, the concrete crack detection method for bridge engineering proposed in Chinese invention patent publication number CN120070360A analyzes the length and width characteristics of suspected crack areas to screen key areas, then combines the degree of extension regularity and density to determine credibility indicators, and adjusts the significance value to screen out real cracks, thereby improving detection accuracy.

[0005] However, the aforementioned method for screening key regions based on length and width features does not consider the convergence characteristics of cracks. This feature, as an important morphological attribute characterizing the gradual decrease in local width at the crack's start or end, plays a crucial role in distinguishing between real and false cracks. If the identification and analysis of convergence points are ignored during crack detection, non-convergent pseudo-edges may be misjudged as valid cracks, leading to an increased false detection rate. Furthermore, it is difficult to accurately define the start and end positions of cracks, affecting the quantitative assessment of key parameters such as crack length and development trend, thereby reducing the accuracy of crack detection.

[0006] Crack detection requires not only accurate identification of current geometric features, but also precise judgment of its propagation behavior, which is crucial to the comprehensiveness and effectiveness of the detection. Crack propagation behavior is typically driven by multiple factors. Besides conventional external and intrinsic factors such as load stress and material aging, the spatial relationship between adjacent cracks also influences the crack propagation path and rate. Therefore, a lack of effective identification of the spatial relationship between adjacent cracks will significantly limit the analysis of crack propagation driving mechanisms, weaken the forward-looking diagnostic capability of detection methods, and, moreover, misjudgment of the spatial relationship between adjacent cracks can easily lead to misjudgments of crack propagation behavior, creating potential risks for the operation and maintenance decisions of engineering structures. Summary of the Invention

[0007] In view of this, in order to solve the problems mentioned in the background technology, a machine vision-based intelligent detection and analysis method for bridge pavement cracks is proposed.

[0008] The objective of this invention can be achieved through the following technical solution: a machine vision-based intelligent detection and analysis method for bridge pavement cracks, comprising: acquiring a surface image of the crack area to be detected through an image acquisition device, and processing and identifying the surface image through edge detection and morphological analysis to obtain several candidate crack areas.

[0009] By traversing the sampling points along the crack's central axis point by point and calculating its normal width, the maximum width point of several candidate crack regions is identified. Based on the maximum width point, each candidate crack region is divided into a first sub-region and a second sub-region.

[0010] The width attenuation index of the first sub-region and the second sub-region is calculated based on each sampling point, and the first convergence point and the second convergence point are determined.

[0011] Several target regions with a first convergence point and a second convergence point are selected from several candidate crack regions, and several real crack regions are identified from each target region through geometric analysis and morphological discrimination.

[0012] Density clustering is used to divide several real crack regions into several spatial areas, and geometric similarity comparison is used to classify the spatial relationship between adjacent cracks.

[0013] Based on the spatial relationship classification of adjacent cracks and the geometric parameters of the actual cracks in each spatial region, a crack propagation rate distribution map is generated by analyzing the inverted mechanical parameters and propagation criteria.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0015] (1) The present invention identifies the crack convergence point by the crack width change rate and identifies the real crack based on the geometric features in the neighborhood of the convergence point. This solves the technical defect of the prior art that non-convergent pseudo-edges are mistakenly detected as valid cracks because the crack convergence characteristics are ignored. This reduces the false detection rate, improves the accuracy and robustness of crack detection, and provides reliable structured data support for the subsequent quantitative analysis and behavior assessment of crack propagation trends.

[0016] (2) This invention divides real cracks into spatial regions, classifies the spatial relationships between adjacent cracks, and generates a crack propagation rate distribution map by combining a mechanical inversion model. This solves the limitation of existing technologies that lack the identification of spatial relationships between adjacent cracks and make it difficult to analyze the crack propagation driving mechanism. It enhances the forward-looking diagnostic capability of the detection method, accurately judges crack propagation behavior, provides a reliable basis for engineering structure operation and maintenance decisions, and reduces potential operation and maintenance risks. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the method steps of the present invention.

[0019] Figure 2 This is a flowchart illustrating the process of obtaining the crack convergence point in this invention.

[0020] Figure 3 This is a flowchart illustrating the classification of spatial relationships between adjacent cracks in this invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1 As shown, the present invention provides a machine vision-based intelligent detection and analysis method for bridge pavement cracks, including: acquiring surface images of the crack area to be detected through an image acquisition device, and processing and identifying several candidate crack areas by edge detection and morphological analysis of the surface image.

[0023] Furthermore, the method for obtaining the candidate crack regions is as follows: using an image acquisition device installed on a mobile device, surface images of the crack regions to be detected on the bridge pavement are acquired under controlled lighting conditions.

[0024] It should be noted that the mobile device is a detection vehicle or inspection robot, and the image acquisition device is a high-resolution industrial camera, a line scan camera, or a multispectral imaging device. Under controlled lighting conditions, the relative height, angle, and lighting conditions between the device and the road surface are kept stable during the image acquisition process to ensure consistent image quality. The acquired surface images usually contain the effects of uneven ambient lighting, dust, water stains, marking interference, and sensor noise, so preprocessing is required.

[0025] The surface image is preprocessed to generate preprocessed image data.

[0026] It should be noted that the method for preprocessing the surface image is as follows: the color image is converted into a grayscale image to reduce data redundancy and improve computational efficiency. Then, an adaptive histogram equalization method is used to enhance the local contrast of the image, making weak details such as cracks more clearly visible. To address the problem of uneven illumination, background modeling or homomorphic filtering techniques are used to correct the illumination of the image, eliminating brightness abrupt changes caused by shadows or reflections. Median filtering or nonlocal mean denoising algorithms are used to suppress random noise in the image while preserving the structural information of the crack edges.

[0027] Multi-scale edge detection is performed on the preprocessed image data to generate a binary edge image and extract the edge pixel coordinate set.

[0028] Morphological closing operations and connected component analysis are performed on the edge pixel coordinate set to obtain several candidate crack regions at the pixel level.

[0029] By traversing the sampling points along the crack's central axis point by point and calculating its normal width, the maximum width point of several candidate crack regions is identified. Based on the maximum width point, each candidate crack region is divided into a first sub-region and a second sub-region.

[0030] It should be noted that the sampling points were obtained by uniformly sampling along the central axis of the crack.

[0031] Furthermore, the method for dividing the first sub-region and the second sub-region is as follows: the skeleton of the candidate crack region is extracted to obtain the crack centerline.

[0032] The boundary contours of candidate crack regions are obtained through edge detection and contour extraction algorithms.

[0033] In a specific embodiment, the method for obtaining the boundary contour of the candidate crack region is as follows: using edge detection algorithms such as the Canny algorithm and the Sobel algorithm, gradient calculation and threshold segmentation are performed on the pixel gray-level changes in the candidate crack region to extract edge pixels with continuous gray-level jump characteristics. Using contour extraction algorithms such as the chain code-based contour tracking algorithm and the eight-neighbor contour search algorithm, based on the edge pixel coordinate set, the edge pixels are traversed according to the pixel connectivity rules, and adjacent edge pixels that meet the connectivity conditions are connected in sequence to form a closed or continuous curve, thereby obtaining the complete boundary contour of the candidate crack region.

[0034] Calculate the normal direction of each sampling point on the central axis of the crack, identify the two intersection points of the normal direction and the boundary profile, and use the Euclidean distance between the two points as the width value of the sampling point.

[0035] By traversing all sampling points, the maximum width point corresponding to the maximum width value is identified. Using this maximum width point as the segmentation benchmark, the candidate crack region is divided into two sub-regions: the region pointing from the maximum width point to the crack start end is defined as the first sub-region, and the region pointing to the crack end is defined as the second sub-region.

[0036] It should be noted that if there are multiple maximum width points, the largest one should be selected as the main dividing point, or the global shape of the crack should be used as an auxiliary factor for judgment.

[0037] It should be explained that the crack width is a core geometric parameter characterizing the crack morphology. Real cracks have significant convergence characteristics, while pseudo cracks usually have irregular width variations or no convergence characteristics. By calculating the normal width of the sampling point on the central axis of the crack, the width distribution pattern of the crack along the length direction can be obtained, which provides key data support for subsequent identification of convergence points and differentiation between real cracks and pseudo cracks, avoiding false detections caused by relying solely on edge morphology.

[0038] The width distribution of real cracks typically exhibits a widening pattern in the middle and narrowing at both ends. The point of maximum width is a natural dividing point for crack morphological changes. Using this point as a benchmark for partitioning allows for independent analysis of the width changes in the two sub-regions. This ensures accurate location of the convergence point where the width decay in each sub-region tends to stabilize when calculating the width decay index of each sub-region. Simultaneously possessing the convergence point of both sub-regions is the core criterion for determining real cracks and provides accurate morphological boundary data for subsequent quantification of crack propagation behavior.

[0039] The width attenuation index of the first sub-region and the second sub-region is calculated based on each sampling point, and the first convergence point and the second convergence point are determined.

[0040] For further details, please refer to Figure 2As shown, the steps for obtaining the first convergence point and the second convergence point are as follows: S1, calculate the width attenuation index of the first sub-region based on each sampling point on the central axis of the candidate crack region, using the following formula: ,in For the first Width value of each sampling point For the first Width value of each sampling point For the first Width value of each sampling point The sampling point number, ,in This represents the total number of sampling points.

[0041] It should be explained that the formula for calculating the width attenuation index quantifies the trend of width attenuation along the central axis of the crack by the ratio of the width change difference between three adjacent sampling points. This formula, by using the ratio of three differences rather than a single difference, can accurately capture the uniformity of the actual crack attenuation and avoid the defect that a single difference cannot distinguish between regular attenuation and random fluctuations.

[0042] S2. When the width attenuation index approaches 1 for the first time, and the corresponding width value is less than the crack width limit specified in the engineering specification, the sampling point corresponding to the width attenuation index is taken as the first convergence point.

[0043] It should be explained that the condition that the width attenuation index approaches 1 for the first time indicates that the width change of three consecutive sampling points tends to be consistent, which means that the crack width attenuation has entered a stable stage. By replacing subjective visual judgment, it provides an objective basis for the identification of subsequent convergence points and avoids the error of human judgment.

[0044] It should be noted that the crack width limit specified in the engineering specifications refers to the maximum allowable width threshold that is clearly set based on national / industry technical standards related to bridge construction and operation and maintenance, combined with the structural type and design service life of the bridge, to determine whether the crack needs intervention.

[0045] This limit is a core quantitative indicator for measuring the impact of cracks on the safety and durability of bridge structures. It is obtained by directly consulting the national / industry authoritative technical specifications in the field of bridge construction, combining the bridge-specific design documents, and referring to the local or industry supplementary standards corresponding to the special environment of the project location.

[0046] S3. Replace the first sub-region in step S1 with the second sub-region, and repeat steps S1-S2 to obtain the second convergence point.

[0047] It needs to be explained that crack width is essentially the cracking deformation of a material caused by stress. That is, the greater the stress, the more obvious the crack opening; the greater the stress gradient, the more drastic the change in crack width. Therefore, based on the physical characteristics of stress release at the crack tip, the crack width usually exhibits the characteristic of rapid decay followed by gradual convergence near the endpoint. Thus, the convergence point is determined by analyzing the width decay index.

[0048] Several target regions with a first convergence point and a second convergence point are selected from several candidate crack regions, and several real crack regions are identified from each target region through geometric analysis and morphological discrimination.

[0049] Furthermore, the method for identifying the real crack region is as follows: selecting several target regions from several candidate crack regions that simultaneously contain a first convergence point and a second convergence point.

[0050] For each target region, the width variation rate is calculated based on the width values ​​of each sampling point along the crack's central axis.

[0051] Historical crack width variation data is obtained, and crack width gradient benchmark value is calculated. If the width variation rate is less than the crack width gradient benchmark value, the area is determined to be the first real crack area. If the width variation rate is greater than or equal to the crack width gradient benchmark value, the area is determined to be the candidate real crack area.

[0052] It should be noted that the crack width gradient benchmark value is a quantitative threshold reflecting the normal range of the width change rate of a real crack along the central axis. It is obtained by collecting sample data of bridge pavement that have been clearly identified as real cracks in historical inspections, extracting the width values ​​of each sampling point along the central axis of these real cracks, calculating the width change rate of each real crack sample, and then performing statistical analysis on the width change rate of all samples. The crack width gradient benchmark value is obtained by the mean or median of the width change rate of all samples calculated above, or by combining engineering experience to determine a reasonable range.

[0053] It should be explained that the width change rate being less than the crack width gradient benchmark value indicates that the width change of the target area along the crack central axis is gentle and regular, which conforms to the normal rate range of historical real crack width changes. Its morphological characteristics are highly matched with the convergence characteristics of real cracks, so it can be directly identified as a real crack area.

[0054] The fact that the width change rate is greater than or equal to the crack width gradient benchmark value indicates that the width change of the target area is drastic and irregular, exceeding the normal range of real crack width changes. Its morphological characteristics are closer to the disordered changes of pseudo cracks. Therefore, it cannot be directly determined as a real crack and needs to be classified as a candidate real crack area.

[0055] Structural features of the neighborhood of the convergence point in the candidate real crack region are extracted, and the second real crack region is identified and determined by the feature weighted scoring mechanism. The real crack region is obtained by judging region by region according to the above judgment method.

[0056] Furthermore, the method for identifying the second real crack region is as follows: taking the convergence point in the candidate real crack region as the center, the neighborhood window is cropped according to the preset pixel size, and a comprehensive score is calculated by weighted summation based on the index. When the comprehensive score is greater than or equal to the preset score threshold, the region is determined to be the second real crack region.

[0057] It should be noted that the preset pixel size needs to be determined by combining the common morphological scale of cracks, the resolution of the image acquisition device and the sampling point density. It is usually based on the principle of being able to completely cover the smallest area around the convergence point that reflects the structural characteristics of the crack.

[0058] It should be explained that the convergence point is the key location for distinguishing between real cracks and pseudo cracks, but its determination depends on the surrounding local structural features rather than information from a single pixel. By extracting a neighborhood window, we can focus on the core area around the convergence point, eliminate other irrelevant interference from the road surface, and accurately extract the local structural features used to determine real cracks. This provides a targeted analysis range for calculating the comprehensive score of subsequent indicators such as crack development degree, dispersion characteristics, and tortuosity index.

[0059] It should be noted that the specific weighting of the comprehensive score calculated by weighted summation is based on the physical characteristics of bridge cracks, and the importance of different indicators varies. Crack development degree reflects the crack propagation trend and is highly important, generally assigned a high weight. Dispersion characteristics are used to determine whether cracks are concentrated or disordered, to distinguish noise, and are usually assigned a medium weight. The tortuosity index reflects the standard deviation of the rate of curvature change; real cracks are usually smoother, while pseudo-cracks such as scratches or shadows are more tortuous, therefore this indicator has strong discriminative ability and is also assigned a high weight. The bifurcation index counts the number of topological bifurcations; concrete cracks often have bifurcation characteristics and can be used as a discriminative criterion, assigned a medium or high weight depending on the situation. In a preferred embodiment, the initial weights of each feature can be set as follows: crack development degree weight 0.3, dispersion characteristics weight 0.2, tortuosity index weight 0.3, and bifurcation index weight 0.2. Subsequently, supervised learning, such as using a logistic regression model, can be used for training and optimization on a labeled dataset.

[0060] For example, by combining machine learning, supervised learning can be used to extract four features as inputs from labeled real and fake crack samples to train a classifier. The model automatically learns the optimal weights and fixes them into the system as fixed parameters. Before determining the weights, the scores of each feature usually need to be normalized, such as by min-max or z-score transformation, to eliminate differences in units and ensure the rationality and comparability of the weighted summation.

[0061] The preset score threshold is a critical score value set based on historical detection data to determine whether a candidate real crack region is a second real crack region. It is determined by collecting a large number of real crack neighborhood windows and pseudo crack neighborhood windows that have been confirmed through manual review or authoritative detection methods, calculating the comprehensive score of each type of sample, and using statistical analysis methods such as plotting ROC curves to find the optimal critical value. This threshold provides an objective and quantitative standard for the discrimination of candidate real crack regions, replacing subjective visual judgment and avoiding discrimination errors caused by differences in the experience of detection personnel. In this invention, the preset score threshold is typically set between 0.65 and 0.75.

[0062] The method for obtaining the indicators is as follows: (a) Perform multi-scale edge detection on the neighborhood window to extract fine edge structures, and calculate the density and spatial distribution entropy of short edge segments in the region in front of the main crack tip as crack development degree and diffusion characteristics, respectively.

[0063] In a specific embodiment, the method for obtaining the degree of crack development and diffusion characteristics is as follows: performing multi-scale edge detection or enhanced filtering such as Gabor filtering or LoG operator on the neighborhood window, extracting fine edge structures, and statistically analyzing the density and spatial distribution entropy of short edge segments in the region in front of the tip of the main crack.

[0064] It should be explained that the degree of crack development indicates that more tiny pre-cracks or branch buds have appeared around the tip of the main crack, reflecting that the crack is in an active state of dynamic expansion.

[0065] The spatial distribution entropy is an indicator describing the degree of dispersion of short edge segments in the region in front of the tip of the main crack: if the short edge segments are concentrated in a small area, the distribution entropy value is low, indicating that the crack propagation direction is relatively singular; if the short edge segments are dispersed in multiple directions, the distribution entropy value is high, indicating that the crack propagation trend is more complex and may branch out in multiple directions.

[0066] (b) Calculate the local curvature at each sampling point along the central axis of the crack, and calculate its rate of change of curvature. Use the standard deviation of the rate of change of curvature as the tortuosity index. Analyze the number of edge topological bifurcations in the curvature direction using morphological skeleton and graph theory, and use it as the bifurcation index.

[0067] In a specific embodiment, the method for obtaining the tortuosity index and the bifurcation index is as follows: sampling at equal intervals along the central axis of the crack within a neighborhood window, calculating the local curvature at each point; calculating the standard deviation of the rate of change of the curvature sequence as the tortuosity index, and detecting the number of significant reversals of curvature direction or edge topological bifurcations as the bifurcation index.

[0068] It should be explained that the tortuosity index is used to quantify the complexity of the tortuosity of the crack direction. The larger the standard deviation, the more obvious the difference in the degree of tortuosity of each segment of the crack, and the more irregular and tortuous the overall direction of the crack. The smaller the standard deviation, the more gentle and straight the crack direction.

[0069] The bifurcation index is used to measure whether a crack branches and the number of branches. The larger the bifurcation index value, the more branches the crack branches. When the value is zero, it means that the crack is a single main crack without branches.

[0070] It should be explained that the second method for identifying real crack regions is a multi-dimensional morphological feature quantitative evaluation. The four indicators can accurately distinguish between real cracks and false cracks, such as crack-like images formed by surface scratches and stains, from two key dimensions: dynamic expansion potential and morphological structural features.

[0071] On the one hand, real cracks, as a manifestation of bridge structural damage, will dynamically expand with changes in load and temperature. The precursor to their expansion is the appearance of tiny pre-cracked segments at the tip of the main crack, and the expansion trend may show a multi-directional dispersion. On the other hand, false cracks are static surface defects, and such short edge structures will not be produced at the tip. The two can be directly distinguished by two indicators: the degree of crack development and the spatial distribution entropy.

[0072] On the other hand, real cracks are affected by the distribution of structural stress, and their direction often has a natural curvature. They may also branch out due to stress concentration. False cracks are mostly regular straight lines or simple shapes, with almost no obvious bends or bifurcations. The two indicators of bend index and bifurcation index can further enhance the differentiation effect.

[0073] These four indicators complement each other, capturing the dynamic nature of real cracks while covering their typical morphological characteristics. They effectively compensate for the shortcomings of relying solely on the width change rate as a single indicator, which is susceptible to image noise interference and has a high misjudgment rate.

[0074] It should be understood that the aforementioned indicators essentially focus on the two most common core features of bridge pavement cracks: dynamic expansion related features and morphological structure related features. These two types of features are sufficient to handle the identification of false cracks in most conventional scenarios. However, in special scenarios such as weathering cracks in old bridges and low-temperature shrinkage cracks in asphalt pavements, it is necessary to make a comprehensive judgment based on the specific circumstances such as the material of the actual object being tested, the usage environment, and the type of damage.

[0075] This invention identifies crack convergence points by measuring crack width variation rate and identifies real cracks based on geometric features in the neighborhood of the convergence point. This solves the technical defect of existing technologies that ignore crack convergence characteristics, leading to the misdetection of non-convergent pseudo-edges as valid cracks. This reduces the false detection rate, improves the accuracy and robustness of crack detection, and provides reliable structured data support for subsequent quantitative analysis and behavioral assessment of crack propagation trends.

[0076] Furthermore, the method for extracting the boundary contour curves on both sides of the candidate real crack region includes: generating a normal line perpendicular to the tangent direction of the crack's central axis at each sampling point along the direction of the crack's central axis.

[0077] Extract the edge contour data of candidate real crack regions.

[0078] In a specific embodiment, the method for obtaining the edge contour data is as follows: the image data of the candidate real crack region is processed by edge detection and contour extraction algorithms such as multi-scale edge detection and Canny edge detection to obtain initial contour information containing crack edge pixels.

[0079] The two intersection points located on both sides of the crack's central axis are obtained by intersecting the normal of each sampling point with the edge contour of the candidate real crack region, and are taken as the contour points corresponding to that sampling point.

[0080] It should be noted that the basis for taking the two intersection points on both sides of the crack's central axis as the contour points corresponding to the sampling point is that the edge contour of the candidate real crack region is essentially a discrete set of edge pixels. If these pixels are directly connected, problems such as burrs and breaks may occur due to noise interference, making it impossible to form a smooth and continuous contour curve.

[0081] However, selecting contour points by intersecting normals is equivalent to using the central axis as a skeleton, selecting key nodes from discrete edge pixels according to uniform geometric rules, and then connecting the contour points on the same side in sequence. This can naturally form a continuous curve that fits the true boundary of the crack, and the curve direction is highly matched with the extension trend of the crack's central axis, which is more in line with the requirements of subsequent morphological analysis for regular contour data.

[0082] Connect all sampling points to the contour points on the same side to form the boundary contour curves on both sides of the candidate real crack region.

[0083] Density clustering is used to divide several real crack regions into several spatial areas, and geometric similarity comparison is used to classify the spatial relationship between adjacent cracks.

[0084] Furthermore, the method for dividing several real crack regions into several spatial regions is as follows: extract the set of sampling points of the crack centerline of all real crack regions, perform density clustering on the set of sampling points of the crack centerline to obtain several clusters, and merge the candidate crack regions corresponding to all sampling points belonging to the same cluster into one spatial region.

[0085] It should be noted that the density clustering algorithms used, such as the DBSCAN algorithm and other typical density clustering methods, are highly adaptable to the actual distribution and detection needs of bridge pavement cracks. On the one hand, real cracks are mostly caused by stress concentration or material aging, and their spatial distribution is non-uniform and irregular. Density clustering does not require a preset number of partitions. It automatically clusters based on the spatial density and connectivity of the sampling points along the crack's central axis, which can accurately identify the natural boundaries between dense and sparse areas, avoiding the crack fragmentation or forced grouping problems caused by traditional grid partitioning.

[0086] On the other hand, sampling points within the same cluster are spatially adjacent and have a high density, usually originating from the same structural damage mechanism and have strong structural correlation. Classifying them into the same area can ensure that subsequent spatial relationship analysis focuses on potentially interacting crack groups, eliminating interference from isolated cracks, and is more in line with actual damage logic.

[0087] In addition, the number and extent of cracks in each zone can be controlled after zoning, which helps to reduce the complexity of mechanical parameter inversion and propagation rate analysis; assessment based on strongly correlated zones can more accurately identify local propagation risks and provide regional and targeted decision support for bridge maintenance.

[0088] For further details, please refer to Figure 3 As shown, the classification method for the spatial relationship of adjacent cracks is as follows: for each spatial region, extract the principal direction vector of all real cracks in the region.

[0089] Calculate the Euclidean distance similarity among all sampled points of the centerline of the real cracks within the region. And calculate their mean, denoted as . .

[0090] In a specific embodiment, the Euclidean distance similarity calculation formula is as follows: in, and Sampling points and The coordinate value in the i-th dimension.

[0091] Calculate the angle similarity between the principal directions of each pair of cracks. And calculate the mean of the similarity of all included angles, denoted as . .

[0092] It should be noted that the similarity of the angles between the main directions of the cracks is defined as the absolute value of the cosine of the angle between the two direction vectors.

[0093] In a specific embodiment, the formula for calculating the cosine similarity of the included angle is: in, and These represent the two direction vectors corresponding to the main direction of the crack.

[0094] like and If so, the spatial relationship between adjacent cracks is determined to be a parallel cooperative configuration.

[0095] like and If so, the spatial relationship between adjacent cracks is determined to be an intersecting configuration.

[0096] like If the two cracks are not significantly related, then it is determined that there is no significant spatial relationship between adjacent cracks.

[0097] It needs to be explained that when At that time, regardless of whether the similarity of the included angle is higher than its mean, it is determined that there is no significant spatial relationship. This is because the primary premise of the spatial correlation of cracks is that the spatial distance is close. Only cracks that are close to each other can generate mechanical interaction due to stress transmission, load distribution, etc. Cracks that are too far apart, even if they have similar directions, will not form an actual structural relationship. Therefore, they do not need to be classified as parallel cooperative or intersecting spatial configurations with mechanical significance.

[0098] Based on the spatial relationship classification of adjacent cracks and the geometric parameters of the actual cracks in each spatial region, a crack propagation rate distribution map is generated by analyzing the inverted mechanical parameters and propagation criteria.

[0099] This invention divides real cracks into spatial regions, classifies the spatial relationships between adjacent cracks, and generates a crack propagation rate distribution map by combining a mechanical inversion model. This solves the limitations of existing technologies that lack identification of spatial relationships between adjacent cracks and make it difficult to analyze the crack propagation driving mechanism. It enhances the forward-looking diagnostic capability of the detection method, accurately judges crack propagation behavior, provides a reliable basis for engineering structure operation and maintenance decisions, and reduces potential operation and maintenance risks.

[0100] Furthermore, the method for obtaining the crack propagation rate distribution map is as follows: for each spatial region, extract the width value and spatial coordinates of all real crack sampling points along the crack centerline, and calculate the width change rate of each sampling point per unit time.

[0101] Based on the width and spatial coordinates of the crack at each sampling point, combined with the fracture mechanics model and the classification results of the spatial relationship between adjacent cracks, the stress intensity factor of each crack is obtained through inversion calculation.

[0102] In a specific embodiment, the fracture mechanics model needs to be selected according to the characteristics of bridge pavement materials and crack structure, such as the Mode I model in linear elastic fracture mechanics, and the model parameters need to be specifically modified according to the classification results of the spatial relationship between adjacent cracks: for parallel cooperative configurations, the cooperative stress coefficient needs to be increased to amplify the stress contribution; for cross configurations, the cross stress concentration factor needs to be introduced to correct the stress distribution at the cross position; and for no significant spatial relationship, the basic parameters of the model are used.

[0103] The extracted crack sampling point width value and spatial location coordinates are used as known inputs and substituted into the modified fracture mechanics model. The equations about the stress intensity factor are solved by numerical iterative algorithms such as least squares method and Newton's iteration method. This minimizes the error between the theoretical crack geometric parameters corresponding to the calculated stress intensity factor and the actual detection parameters, and finally obtains the stress intensity factor of each crack at each sampling point.

[0104] The crack width change rate is converted into the corresponding stress intensity factor change rate using a quantitative correlation formula between the crack width change rate and the stress intensity factor change rate.

[0105] In a specific embodiment, the width value, spatial coordinates, and width change rate per unit time of each sampling point along the central axis of each crack are extracted. Then, based on material properties, crack type, and engineering specifications, a correlation formula is selected, for example, by combining the elastic modulus and geometric parameters to establish a linear or nonlinear functional relationship, such as based on material mechanics derivation. The width change rate of each sampling point is directly substituted into this formula, and without iterative inversion, the corresponding stress intensity factor change rate can be output.

[0106] By combining the fracture mechanics propagation criterion of the cracked material under test, the rate of change of stress intensity factor is transformed into the input parameter required by the criterion, and then substituted into the criterion formula to calculate the crack propagation rate per unit time.

[0107] In a specific embodiment, based on the mechanical properties of the material containing the cracks in the bridge pavement to be detected and the type of crack propagation, a suitable fracture mechanics propagation criterion is selected, and the core input parameter requirements of the criterion are clarified.

[0108] For example, the fracture mechanics extension criterion is the Paris criterion, whose input parameters include the stress intensity factor amplitude. The material fatigue crack propagation coefficient C and exponent m are used to convert the stress intensity factor change rate into the values ​​of parameters C and m obtained from material tests and crack geometric parameters, as well as the material fatigue crack propagation coefficient C and exponent m, into the values ​​in the criterion. Finally, the transformed input parameters are substituted into the Paris criterion formula: The crack propagation rate per unit time at each sampling point was calculated.

[0109] Based on the crack propagation rate and spatial location coordinates, discrete crack propagation rates are transformed into crack propagation rate distribution maps through spatial interpolation.

[0110] It should be noted that the spatial interpolation specifically uses the Kriging interpolation method.

[0111] The parameters involved in the above formula are all dimensionless and calculated numerically. The formula is a formula obtained from the most recent real situation by collecting a large amount of data and simulating it with software. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0112] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0113] Those skilled in the art will recognize that the algorithmic steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0114] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0115] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0116] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A machine vision-based intelligent detection and analysis method for bridge pavement cracks, characterized in that, include: The surface image of the crack area to be detected is acquired by an image acquisition device, and several candidate crack areas are obtained by processing and identifying the surface image through edge detection and morphological analysis. By traversing the sampling points along the crack axis point by point and calculating its normal width, the maximum width point of several candidate crack regions is identified, and each candidate crack region is divided into a first sub-region and a second sub-region based on the maximum width point. The width attenuation index of the first sub-region and the second sub-region is calculated based on each sampling point, and the first convergence point and the second convergence point are determined accordingly. The steps for obtaining the first convergence point and the second convergence point are as follows: S1. Calculate the width attenuation index of the first sub-region based on the sampling points along the central axis of the candidate crack region. The formula is as follows: ,in For the first Width value of each sampling point For the first Width value of each sampling point For the first Width value of each sampling point The sampling point number, ,in This represents the total number of sampling points; S2. When the width attenuation index approaches 1 for the first time, and the corresponding width value is less than the crack width limit specified in the engineering specification, the sampling point corresponding to the width attenuation index is taken as the first convergence point. S3. Replace the first sub-region in step S1 with the second sub-region, and repeat steps S1-S2 to obtain the second convergence point. Several target regions with a first convergence point and a second convergence point are selected from several candidate crack regions, and several real crack regions are identified from each target region through geometric analysis and morphological discrimination. Density clustering is used to divide several real crack regions into several spatial areas, and geometric similarity comparison is used to classify the spatial relationship between adjacent cracks. Based on the spatial relationship classification of adjacent cracks and the geometric parameters of the actual cracks in each spatial region, a crack propagation rate distribution map is generated by analyzing the inverted mechanical parameters and propagation criteria.

2. The intelligent detection and analysis method for bridge pavement cracks based on machine vision as described in claim 1, characterized in that: The method for obtaining the candidate crack regions is as follows: Using an image acquisition device mounted on a mobile device, surface images of the crack area to be detected on the bridge pavement are acquired under controlled lighting conditions. The surface image is preprocessed to generate preprocessed image data; Multi-scale edge detection is performed on the preprocessed image data to generate a binary edge image and extract the edge pixel coordinate set; Morphological closing operations and connected component analysis are performed on the edge pixel coordinate set to obtain several candidate crack regions at the pixel level.

3. The intelligent detection and analysis method for bridge pavement cracks based on machine vision as described in claim 1, characterized in that: The method for dividing the first sub-region and the second sub-region is as follows: The skeleton of the candidate crack region is extracted to obtain the crack centerline; The boundary contours of candidate crack regions are obtained through edge detection and contour extraction algorithms; Calculate the normal direction of each sampling point on the central axis of the crack, identify the two intersection points of the normal direction and the boundary profile, and use the Euclidean distance between the two points as the width value of the sampling point. By traversing all sampling points, the maximum width point corresponding to the maximum width value is identified. Using this maximum width point as the segmentation benchmark, the candidate crack region is divided into two sub-regions: the region pointing from the maximum width point to the crack start end is defined as the first sub-region, and the region pointing to the crack end is defined as the second sub-region.

4. The intelligent detection and analysis method for bridge pavement cracks based on machine vision as described in claim 1, characterized in that: The method for identifying the actual crack region is as follows: Select several target regions from several candidate crack regions that simultaneously contain the first convergence point and the second convergence point; For each target region, the width change rate is calculated based on the width values ​​of each sampling point along the crack axis. Historical crack width variation data were obtained, and the crack width gradient benchmark value was statistically calculated. If the width change rate is less than the crack width gradient benchmark value, the region is determined to be the first real crack region; if the width change rate is greater than or equal to the crack width gradient benchmark value, the region is determined to be a candidate real crack region. Structural features of the neighborhood of convergence point in the candidate real crack region are extracted, and the second real crack region is identified and determined through a feature weighted scoring mechanism. The actual crack area is obtained by judging each area according to the above judgment method.

5. The intelligent detection and analysis method for bridge pavement cracks based on machine vision as described in claim 4, characterized in that: The method for identifying the second real crack region is as follows: Centered on the convergence point in the candidate real crack region, a neighborhood window is cropped according to a preset pixel size, and a comprehensive score is calculated by weighted summation based on the following indicators: (a) Perform multi-scale edge detection on the neighborhood window to extract fine edge structures, and calculate the density and spatial distribution entropy of short edge segments in the region in front of the main crack tip as crack development degree and diffusion characteristics, respectively. (b) Calculate the local curvature at each sampling point along the central axis of the crack, and calculate the rate of change of curvature. Use the standard deviation of the rate of change of curvature as the tortuosity index. The number of edge topological bifurcations along the curvature direction is used as a bifurcation index through morphological skeleton and graph theory analysis. When the overall score is greater than or equal to a preset score threshold, the region is determined to be the second real crack region.

6. The intelligent detection and analysis method for bridge pavement cracks based on machine vision as described in claim 1, characterized in that: The method for dividing several real crack regions into several spatial regions is as follows: Extract the set of sampling points along the crack centerline of all real crack regions, perform density clustering on the set of sampling points along the crack centerline to obtain several clusters, and merge the candidate crack regions corresponding to all sampling points belonging to the same cluster into a spatial region.

7. The intelligent detection and analysis method for bridge pavement cracks based on machine vision as described in claim 1, characterized in that: The classification method for the spatial relationship of adjacent cracks is as follows: For each spatial region, extract the principal direction vectors of all real cracks within the region; Calculate the Euclidean distance similarity among all sampled points of the centerline of the real cracks within the region. And calculate their mean, denoted as . ; Calculate the angle similarity between the principal directions of each pair of cracks. And calculate the mean of the similarity of all included angles, denoted as . ; like and If so, the spatial relationship between adjacent cracks is determined to be a parallel cooperative configuration; like and If so, the spatial relationship between adjacent cracks is determined to be an intersecting configuration; like If the two cracks are not significantly related, then it is determined that there is no significant spatial relationship between adjacent cracks.

8. The intelligent detection and analysis method for bridge pavement cracks based on machine vision as described in claim 1, characterized in that: The method for obtaining the crack propagation rate distribution map is as follows: For each spatial region, extract the width value and spatial coordinates of all real crack sampling points along the crack centerline, and calculate the width change rate of each sampling point per unit time. Based on the width and spatial coordinates of the crack at each sampling point, combined with the fracture mechanics model and the classification results of the spatial relationship between adjacent cracks, the stress intensity factor of each crack is obtained through inversion calculation. The crack width change rate is converted into the corresponding stress intensity factor change rate through a quantitative correlation formula between the crack width change rate and the stress intensity factor change rate. Combining the fracture mechanics propagation criterion of the cracked material to be tested, the rate of change of stress intensity factor is transformed into the input parameter required by the criterion, and substituted into the criterion formula to calculate the crack propagation rate per unit time. Based on the crack propagation rate and spatial location coordinates, discrete crack propagation rates are transformed into crack propagation rate distribution maps through spatial interpolation.

Images