Crack quantitative identification method and system fusing photogrammetry and semantic key points

Abstract

The application discloses a kind of crack quantitative identification method and system of fusing photogrammetry and semantic key point, belong to crack identification and measurement technical field;By introducing photogrammetry space correction in crack identification stage, the mapping relationship between pixel coordinates and object coordinates is established, so that the crack detection result has consistent physical scale under different imaging conditions, reliable measurement of crack width and length is realized;At the same time, by combining multiscale edge enhancement and semantic key point guided geometric consistency constraint, the adaptability of crack identification process to complex road texture, illumination change and environmental noise is improved, and the misidentification of non-crack structure is reduced;In addition, without relying on high-density three-dimensional reconstruction or complex point cloud processing, through constraint path reconstruction and skeleton measurement mode, integrated processing of crack identification and physical measurement is realized, the measurement accuracy is guaranteed while the computational complexity is reduced, the applicability and practicality in mobile inspection and engineering application scenarios are improved.

Description

Technical Field

[0001] This invention relates to a method and system for quantitative crack identification that integrates photogrammetry and semantic key points. In particular, it relates to a method and system for achieving millimeter-level quantitative identification and measurement of cracks in pavement and subgrade structures by introducing photogrammetric geometric correction, semantic key point constraints and topological reconstruction mechanisms under monocular or multi-view image conditions. This invention belongs to the interdisciplinary technical field of computer vision, photogrammetry and civil engineering structural health monitoring. Background Technology

[0002] With the increasing demand for refined maintenance and structural safety monitoring of highway infrastructure, the automatic identification and quantitative measurement of cracks in pavement and subgrade structures (such as concrete pavement, asphalt pavement, retaining walls, culverts, slope protection, etc.) has become an important research direction in the fields of traffic engineering and civil engineering.

[0003] The existing technologies mainly include the following categories: (1) Crack detection methods based on traditional computer vision. These methods usually use image edge detection and line feature extraction algorithms to detect gray-scale change areas in road surface images, thereby identifying potential crack areas. These methods can achieve a certain degree of crack detection effect in experimental environments with relatively simple textures and stable lighting conditions. (2) Image semantic segmentation methods based on deep learning. These methods use convolutional neural networks or Transformer structures to perform target detection or semantic segmentation of road surface cracks. Although they have strong feature expression capabilities in complex scenes, the output results remain at the pixel scale and require additional calibration to achieve physical size measurement. (3) Crack measurement methods based on photogrammetry or 3D reconstruction. Although these methods have high spatial accuracy, they require high-overlapping images or high-density point clouds, are computationally complex, and are difficult to meet the real-time requirements of mobile inspection or embedded deployment. (4) The comprehensive application of existing technologies mostly remains in a separate process of "detect first, then measure", lacking a unified spatial constraint mechanism.

[0004] Based on the aforementioned existing technologies, in practical engineering applications, crack identification results generally suffer from insufficient consistency in physical scale. Whether using traditional computer vision methods or deep learning-based image segmentation methods, the processing object is primarily two-dimensional pixel information, and the detection results only reflect the relative position and morphological features of the crack in the image. When the imaging height, shooting angle, or camera parameters change, it is difficult to maintain a stable correspondence between pixel size and actual physical size, making it impossible to directly and reliably quantify crack width and length. Furthermore, in real road environments, road surfaces are typically accompanied by complex textures such as aggregate, water stains, tire tracks, repair marks, and shadow variations. Existing methods often rely on local grayscale changes or gradient information for feature extraction, making them sensitive to environmental noise and prone to misidentifying non-crack structures as cracks, thus affecting the accuracy and stability of the identification results. The main reason for these problems is that existing technologies do not introduce a unified object-space coordinate system or ground sampling distance constraints in the crack identification stage, and also lack high-level semantic constraints on the geometric continuity and topological features of cracks.

[0005] On the other hand, while crack measurement methods based on photogrammetry or 3D reconstruction can achieve high spatial measurement accuracy, they typically require highly overlapping imagery or high-density point cloud data, involve complex feature matching and 3D reconstruction processes, consume significant computational resources, and have stringent data acquisition requirements, making them difficult to meet the real-time operational needs of mobile inspections or embedded devices. Furthermore, some existing fusion solutions often separate crack detection and dimensional measurement into independent processing stages, lacking a unified spatial constraint mechanism. This causes errors to accumulate across multiple processing stages, affecting the reliability of the final measurement results. The fundamental reason is that current technologies have not yet formed a holistic technical solution that simultaneously introduces photogrammetric geometric correction and semantic structural constraints during the crack identification stage, resulting in a fragmented identification and measurement process and limiting its engineering applicability. Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] The technical problem to be solved by this invention is to address the common problems in existing pavement and subgrade crack identification technologies, such as the lack of unified physical scale constraints in crack identification results, insufficient adaptability to complex pavement textures and environmental noise, high computational complexity of photogrammetry methods, and the disconnect between crack identification and dimensional measurement processes.

[0008] (II) Technical Solution

[0009] To address the aforementioned technical problems, this invention provides a method for quantitative crack identification that integrates photogrammetry and semantic key points, comprising the following steps:

[0010] S1. Data Acquisition: Acquire the original image of the structure surface, perform geometric correction and orthophoto transformation on the original image based on camera pose information, establish a mapping benchmark between pixels and physical dimensions, and determine the real physical dimension corresponding to a single pixel.

[0011] S2, Edge Enhancement Processing: Edge enhancement is performed on the orthophoto image obtained in S1 to form a candidate edge set;

[0012] S3. Semantic Key Point Extraction: Extract semantic key points with crack topological structure features from the candidate edge set and construct a local geometric description. Use geometric consistency constraints to filter false edges and divide candidate edges into structural crack edges and non-crack texture noise edges. The semantic key points include crack endpoints, intersections and bifurcation points. The geometric consistency constraints are implemented based on the random sampling consensus algorithm.

[0013] S4. Crack Path Reconstruction: Guided by semantic key points, perform constrained line feature search and fracture compensation to reconstruct a crack path network with complete topological connections.

[0014] S5. Quantitative Measurement: Based on the mapping benchmark established in S1, the crack path network is refined by skeletonization and normal projection search is performed to realize the real physical measurement of crack width and length.

[0015] Furthermore, in S1, the camera pose information includes the camera mounting height, pitch angle, and extrinsic parameter information. The geometric correction is a perspective correction process, and then the perspective image is converted into an orthographic projection image by inverse perspective transformation.

[0016] Furthermore, the S1 data acquisition includes the following steps:

[0017] S1.1 Image Acquisition: Using monocular or multi-view imaging equipment, images are acquired of the surface of the road surface or roadbed structure to obtain the original image or image sequence to be detected;

[0018] S1.2 Camera parameter calibration: Perform intrinsic parameter calibration on the camera of the imaging device to obtain focal length, principal point position and distortion parameters; under multi-view imaging conditions, simultaneously obtain the relative extrinsic parameter relationship between each camera;

[0019] S1.3 Photogrammetric geometric correction: Based on the camera installation height, pitch angle and external parameter information, establish the mapping relationship between the pixel coordinate system and the object coordinate system, and perform perspective correction processing on the original image;

[0020] S1.4 Orthographic Projection and Scale Unification: Inverse perspective transformation is used to convert the perspective image into an orthographic projection image, and the ground sampling distance is calculated and used as the real physical size corresponding to a single pixel.

[0021] Specifically, in S2, the edge enhancement processing includes noise suppression using anisotropic diffusion filtering, different scale perception by constructing an image pyramid, and crack guidance enhancement using directional gradient operators.

[0022] More specifically, the S2 edge enhancement process includes the following steps:

[0023] S2.1 Noise Suppression Filtering: Anisotropic diffusion filtering is applied to the orthophoto image to suppress background noise while preserving crack edge features;

[0024] S2.2, Image Construction at Different Scales: Construct image pyramids at different scales and process images at different resolution levels;

[0025] S2.3 Directional Edge Enhancement: Different directional gradient operators are introduced on images at various scales to perform directional enhancement on edge structures with linear extension characteristics, forming candidate edge sets at different scales.

[0026] More preferably, the S3 semantic key point extraction includes the following steps:

[0027] S3.1 Semantic Key Point Extraction: Extract semantic key points with crack topological significance from the candidate edge set, including crack endpoints, intersections, and bifurcation points;

[0028] S3.2 Local geometric description construction: Construct local geometric description information for each semantic key point to represent the directional consistency and extension trend in its neighborhood;

[0029] S3.3, Geometric Consistency Constraint Filtering: Based on the random sampling consistency algorithm, geometric consistency verification is performed on semantic key points and their associated edges to remove isolated edge points that do not conform to the continuous topological characteristics of the crack;

[0030] S3.4 Edge Semantic Classification: Combining the fuzzy C-means clustering method, candidate edges are divided into structural crack edges and non-crack texture noise edges based on the connectivity, contrast and geometric consistency characteristics of the edges.

[0031] More specifically, in S4, fracture compensation is to dynamically compensate and connect fractured areas caused by changes in illumination, shading, or texture interference, based on the continuity of the fracture extension direction.

[0032] More specifically, the S5 quantitative measurement includes the following steps:

[0033] S5.1 Crack skeleton extraction: Morphological refinement of the crack path network is performed to extract the sub-pixel level central skeleton line of the crack.

[0034] S5.2 Crack width calculation: At each sampling node on the center skeleton line of the crack, search for the crack edge projection point along the normal direction, and combine it with the real physical size corresponding to a single pixel obtained in S1 to convert the pixel width into the real physical width.

[0035] S5.3 Crack Length Calculation: Integrate the distance between each node along the crack skeleton path to calculate the true physical length of the crack;

[0036] S5.4 Output Results: Output the physical width, length, and spatial distribution information of the cracks.

[0037] On the other hand, the present invention also provides a crack quantitative identification system that integrates photogrammetry and semantic key points, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the crack quantitative identification method that integrates photogrammetry and semantic key points described above.

[0038] (III) Beneficial Effects

[0039] The above-described technical solution of the present invention has the following advantages:

[0040] First, this invention achieves physical scale consistency in crack identification results, improving measurement accuracy to the millimeter level. This is because, unlike existing 2D image processing methods, it employs photogrammetric geometric constraints to establish a stable mapping relationship between pixels and actual physical dimensions during the identification stage, avoiding scale distortion caused by changes in imaging parameters. Second, it enhances adaptability to complex road environments and reduces the false recognition rate. This advantage stems from the introduction of semantic keypoint extraction and geometric consistency screening mechanisms. These differ from low-level feature processing in traditional edge detection; they capture the crack topology for high-level semantic constraints, enhancing the algorithm's robustness. Finally, it achieves integrated processing of identification and measurement, reducing computational complexity. This is thanks to constrained path reconstruction and skeleton-based refinement measurement methods. Unlike existing separate processes, it directly integrates semantic constraints and scale uniformity without relying on high-density 3D reconstruction, achieving efficient end-to-end processing.

[0041] The key point of this invention is:

[0042] 1. In the initial stage of crack identification, photogrammetric spatial correction and ground sampling distance constraints are introduced to ensure that crack identification and physical measurement are completed under a unified object-space coordinate system. Furthermore, semantic key point constraints enable reliable screening and reconstruction of crack structures. Compared to existing technologies, traditional crack detection methods typically perform edge or semantic segmentation within a two-dimensional image plane. Their identification results only possess relative geometric meaning at the pixel level, and physical dimensions often rely on post-processing conversion or empirical estimation, with inconsistent scales under different acquisition conditions. This invention, through inverse perspective transformation and scale unification processing, directly places the crack detection process within an orthophoto image with real physical scale constraints, avoiding scale drift problems in subsequent measurement stages.

[0043] 2. This invention uses semantic key points and geometric consistency constraints to screen candidate cracks structurally, rather than simply relying on local grayscale or gradient features. Existing technologies, whether traditional edge detection methods or some deep learning methods, primarily rely on pixel intensity changes or local feature responses for judgment, easily misidentifying non-crack structures such as gravel textures and tire tracks as cracks. This invention extracts semantic key points with topological significance, such as crack endpoints and bifurcation points, and combines them with geometric consistency constraints for screening, giving the crack identification process a clear structural semantic basis, thereby effectively suppressing non-crack noise in complex road surface environments.

[0044] In addition to the technical problems solved by the present invention, the technical features of the technical solutions constituted by the present invention, and the advantages brought about by the technical features of these technical solutions as described above, other technical features of the present invention and the advantages brought about by these technical features will be further explained in conjunction with the accompanying drawings. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are 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.

[0046] Figure 1 This is a schematic flowchart of the method of the present invention.

[0047] Figure 2 This is a schematic diagram of the photogrammetric spatial correction and orthophoto projection of the present invention.

[0048] Figure 3 This is a schematic diagram illustrating the extraction of semantic key points in this invention.

[0049] Figure 4 This is a schematic diagram of the crack path reconstruction of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0051] Example 1

[0052] like Figure 1-4 As shown (attached) Figure 2 middle, The effective width of the camera's single-frame field of view projected onto the road surface. Sensor width

[0053] A method for quantitative identification of pavement and subgrade cracks that integrates photogrammetric constraints and semantic key points includes the following steps:

[0054] Step 1: Image acquisition and photogrammetric spatial calibration (the goal of this step is to establish the metric relationship between image pixels and the real physical world).

[0055] Step 1.1: Image Acquisition;

[0056] Images of the road surface or roadbed structure are acquired by using monocular or multi-view imaging equipment (or imaging sensors) installed on inspection vehicles, robots, or fixed monitoring platforms to obtain the original images or image sequences to be inspected.

[0057] Step 1.2: Camera parameter calibration;

[0058] The camera intrinsic parameters of the imaging device are calibrated to obtain the intrinsic parameter matrix. Including focal length Principal point coordinates In addition to radial and tangential distortion parameters; under multi-view imaging conditions, the relative external parameter relationships between each camera are obtained simultaneously.

[0059] Step 1.3: Photogrammetric geometric correction;

[0060] Based on the camera's installation height, pitch angle, and external parameters, a mapping relationship between the pixel coordinate system and the object coordinate system is established, and perspective correction processing is performed on the original image.

[0061] Step 1.4: Orthographic projection and scale unification;

[0062] Inverse perspective transformation (IPM) is used to convert the perspective image into an orthographic projection image, and the ground sample distance (GSD) is calculated to determine the true physical size corresponding to a single pixel, providing a unified dimensional benchmark for subsequent crack size measurement. Specifically, to eliminate the perspective reduction effect, the vertical distance from the camera lens center to the road surface (which can be calculated from the fixed installation height and pitch angle) is used. and pitch angle Construct the homography matrix This transforms the original perspective view into an orthographic projection image. The coordinate transformation formula is as follows:

[0063]

[0064] in, These are the pixel coordinates of the original image. These are the coordinates of the ground object.

[0065] Calculate the actual physical length (unit: mm / pixel) represented by a single pixel in an orthophoto image, i.e., calculate the Ground Sample Distance (GSD):

[0066]

[0067] in The vertical distance from the center of the camera lens to the road surface. The effective focal length of the camera. This refers to the pixel size of the camera's image sensor. After obtaining the orthographic image through inverse perspective transformation, the image is located within the area of ​​interest on the road surface. Approximately constant, thus ensuring the consistency of physical scale in crack measurement results at different imaging distances. Through It can directly convert the width of subsequently detected pixels into the physical width.

[0068] Step 2: Multi-scale anisotropic edge enhancement processing (for the texture and graininess of road surface gravel, this step uses nonlinear filtering technology).

[0069] Step 2.1: Noise suppression filtering;

[0070] Anisotropic diffusion filtering is applied to the orthophoto image to suppress road surface debris and texture noise while maintaining the clarity and continuity of crack edges. Specifically, the Perona-Malik (PM) model is used to smooth noise while controlling the diffusion coefficient to prevent crack edges from being blurred. The diffusion equation is expressed as:

[0071]

[0072] in, For image gradient, For divergence operators. Diffusion guidance function. Defined as:

[0073]

[0074] Here This is the edge sensitivity parameter. When... When the crack is large (likely at the edge of a crack), It approaches zero, stopping diffusion to protect the boundary.

[0075] Step 2.2: Multi-scale image construction;

[0076] A multi-scale image pyramid is constructed to process images at different resolution levels, taking into account both the coarse-scale main crack structure and the fine-scale crack branch features. Specifically, a multi-scale image pyramid is constructed. The Pyramid of Gauss. In Layer identification of main cracks, in Layer identification of subtle branches.

[0077] Step 2.3: Directional edge enhancement;

[0078] Multi-directional gradient operators are introduced on images at various scales to enhance the directionality of edge structures with linear extension characteristics, forming a multi-scale candidate edge set.

[0079] Step 3: Semantic key point extraction and geometric consistency screening (introducing semantic constraints to eliminate false edges);

[0080] Step 3.1: Semantic Key Point Extraction;

[0081] Semantic key points with crack topological significance are extracted from the candidate edge set, including crack endpoints, intersections, and bifurcation points. These key points serve as core anchors for semantic constraints, capturing the structural semantics of cracks (such as branching and connectivity). Their unique role lies in elevating low-level edge features to a high-level topological representation, avoiding the fragmentation problem of traditional edge detection, thereby significantly improving the overall coherence and accuracy of crack identification. Specifically, topological nodes are extracted from the enhanced edge graph, defined as a set. Key points include: endpoints: points within a connected component that have only one adjacent pixel; intersections (branching points): bifurcation points with three or more adjacent pixels.

[0082] Step 3.2: Construction of local geometric description;

[0083] For each semantic keypoint, a local geometric description is constructed to characterize the directional consistency and extension trend within its neighborhood. This description further strengthens the role of semantic constraints, ensuring that keypoints are not disturbed by isolated noise and providing a reliable geometric semantic foundation.

[0084] Step 3.3: Filtering based on geometric consistency constraints;

[0085] Based on the Random Sample Consensus (RANSAC) algorithm, geometric consistency verification is performed on semantic keypoints and their associated edges, eliminating isolated edge points that do not conform to the continuous topological characteristics of the crack. Specifically, curve fitting is performed on the edge points in the neighborhood of each keypoint. The RANSAC algorithm is used to find the maximum set of interior points. If a certain set of edge points does not conform to a linear or second-order curve model... If an interior point is not found, it is considered noise (such as road surface spots or light and shadow interference) and is removed. The interior point discrimination criterion is: the perpendicular distance from the point to the fitted curve. ( This is a configurable parameter. Its specific value can be obtained through experimental calibration to achieve an optimal balance between noise suppression and feature preservation in specific application scenarios. For example, in one specific embodiment, its value is 3 pixels. Or, the... It can be set according to the ground sampling distance (GSD). Preferably, Set to 1 to 3 times the pixel distance corresponding to GSD, for example = 2 GSD).

[0086] Step 3.4: Edge semantic classification;

[0087] Combining the fuzzy C-means clustering method, candidate edges are divided into structural crack edges and non-crack texture noise edges based on their connectivity, contrast, and geometric consistency characteristics. Specifically, the FCM algorithm is used to perform attribute clustering on edge pixels, with the objective function being:

[0088]

[0089] in The feature vector contains gradient direction, local curvature, and contrast. Through clustering, edges are automatically labeled as structural cracks or non-crack textures, achieving preliminary semantic-level classification. It should be noted that FCM, as a mature algorithm, has a consensus in academia regarding the form of its objective function and the meaning of its symbols. Therefore, those skilled in the art can understand the meaning of each symbol based on general knowledge. For example: For pixels Category The membership degree of has a value range of [0,1]. The fuzzy index has a value of 2.0. For the first Cluster centers of classes This represents the total number of edge pixels. The preset number of clusters is set to 3 in this embodiment, corresponding to the main crack, micro-crack, and pseudo-edge, respectively.

[0090] Step 4: Constrained crack path reconstruction;

[0091] Step 4.1: Guided Line Feature Search and Improved Hough Transform;

[0092] Guided by the selected semantic key points, a constrained line feature search is performed within their local neighborhoods to reconstruct the linear segment structure of the cracks, avoiding the generation of random paths. Specifically, in the Hough parameter space, instead of blind voting across the entire graph, the direction of the semantic key points determined in step 3 is used. Centered on, within a limited angular range Perform local accumulation:

[0093]

[0094] This step can effectively reduce false positives of straight lines caused by random noise.

[0095] Step 4.2: Fracture compensation and connection;

[0096] Based on the continuity of the crack's extension direction, dynamic compensation and connection are performed on crack fracture areas caused by changes in lighting, shading, or texture interference. Specifically, for intermittent cracks caused by lighting shadows, the connection is based on the tangent slope of adjacent endpoints. , Calculate the connection cost using the Euclidean distance and the connection cost:

[0097]

[0098] like If the distance is below a preset threshold (which can be determined based on historical data or experimental verification. In one embodiment, it can be set as an empirical value, such as 5.0 (when the distance unit is pixels and the slope difference is dimensionless). Alternatively, it can be dynamically set based on the statistical distribution (such as the median or mean) of the calculated Cost for all endpoints to be connected), then spline interpolation is used to automatically complete the path. For the two endpoints and In an orthographically projected image, the Euclidean distance (in pixels) is given by k1 and k2, which are the slopes of the tangents at the two endpoints. and These are weighting coefficients, used to balance the contributions of spatial distance and directional consistency to the total connectivity cost. In one specific implementation, an optimized set of weight values ​​can be obtained through experimental calibration, for example, by setting... = 0.7, = 0.3, and The sum is usually 1 to achieve the best connection effect of fracture cracks in most road surface scenarios.

[0099] Step 4.3: Generation of the crack topology network;

[0100] The reconstructed linear segments are topologically connected to form a complete crack path network.

[0101] Step 5: Quantitative measurement of cracks based on skeleton refinement;

[0102] Step 5.1: Extraction of the fracture skeleton;

[0103] Morphological refinement is performed on the crack path network to extract the subpixel-level central skeleton line of the crack (or extract the center line with a single pixel width).

[0104] Step 5.2: Crack width calculation;

[0105] At each sampling node along the skeleton line at the center of the crack, the projection points of the crack edge are searched along the normal direction. Combined with the ground sampling distance obtained in step 1, the pixel width is converted into the actual physical width. Specifically, the width is measured at the skeleton points. Calculate the normal direction Search for edge points on both sides along the normal. , Actual physical width The calculation formula is:

[0106]

[0107] Step 5.3: Crack length calculation;

[0108] The true physical length of the crack is calculated by integrating the distances between nodes along the crack skeleton path. Specifically, the length is measured by integrating the entire skeleton segment; the total length is:

[0109] , This represents the spatial coordinates of the i-th node (pixel) on the skeleton line segment. The meaning is the total number of nodes (or pixels) sampled or extracted on the skeleton line segment;

[0110] Step 5.4: Output the results;

[0111] The system outputs information on the physical width, length, and spatial distribution of cracks for subsequent highway maintenance assessment and decision support.

[0112] Example 2

[0113] This embodiment provides a crack quantitative identification system (or electronic device) that integrates photogrammetry and semantic key points, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the crack quantitative identification method that integrates photogrammetry and semantic key points as described in the above embodiment.

[0114] Alternatively, the crack quantitative identification system integrating photogrammetry and semantic key points may further include: a data acquisition module for acquiring a target image of the road surface or subgrade structure to be detected and its corresponding imaging geometric parameters (or camera pose parameters); a geometric correction module for performing geometric correction on the target image based on the imaging geometric parameters using a photogrammetric constraint mechanism, establishing a mapping relationship between the image pixel coordinate system and the object-side physical coordinate system, and determining the ground sampling distance; a key point detection module for inputting the geometrically corrected target image into a pre-trained semantic key point detection model to extract the semantic feature key points of the crack; a topology reconstruction module for constructing the geometric topology of the crack based on the extracted semantic feature key points and combining a topology reconstruction mechanism to generate crack skeleton lines; and a quantitative measurement module for calculating the physical size parameters of the crack based on the ground sampling distance and the geometric topology of the crack.

[0115] This embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the crack quantitative identification method fused with photogrammetry and semantic key points as described in the above embodiment.

[0116] In the description of the invention, unless otherwise stated, the terms "multiple" or "multiple groups" mean two or more.

[0117] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for quantitative crack identification integrating photogrammetry and semantic key points, characterized in that, Includes the following steps: S1. Data Acquisition: Acquire the original image of the structure surface, perform geometric correction and orthophoto transformation on the original image based on camera pose information, establish a mapping benchmark between pixels and physical dimensions, and determine the real physical dimension corresponding to a single pixel. S2, Edge Enhancement Processing: Edge enhancement is performed on the orthophoto image obtained in S1 to form a candidate edge set; S3. Semantic Key Point Extraction: Extract semantic key points with crack topological structure features from the candidate edge set and construct a local geometric description. Use geometric consistency constraints to filter false edges and divide candidate edges into structural crack edges and non-crack texture noise edges. The semantic key points include crack endpoints, intersections and bifurcation points. The geometric consistency constraints are implemented based on the random sampling consensus algorithm. S4. Crack Path Reconstruction: Guided by semantic key points, perform constrained line feature search and fracture compensation to reconstruct a crack path network with complete topological connections. S5. Quantitative Measurement: Based on the mapping benchmark established in S1, the crack path network is refined by skeletonization and normal projection search is performed to realize the real physical measurement of crack width and length. The acquisition of S1 data includes the following steps: S1.1 Image Acquisition: Using monocular or multi-view imaging equipment, images are acquired of the surface of the road surface or roadbed structure to obtain the original image or image sequence to be detected; S1.2 Camera parameter calibration: Perform intrinsic parameter calibration on the camera of the imaging device to obtain focal length, principal point position and distortion parameters; under multi-view imaging conditions, simultaneously obtain the relative extrinsic parameter relationship between each camera; S1.3 Photogrammetric geometric correction: Based on the camera installation height, pitch angle and external parameter information, establish the mapping relationship between the pixel coordinate system and the object coordinate system, and perform perspective correction processing on the original image; S1.4 Orthographic Projection and Scale Unification: Inverse perspective transformation is used to convert the perspective image into an orthographic projection image, and the ground sampling distance is calculated and used as the real physical size corresponding to a single pixel; The S3 semantic key point extraction includes the following steps: S3.1 Semantic Key Point Extraction: Extract semantic key points with crack topological significance from the candidate edge set, including crack endpoints, intersections, and bifurcation points; S3.2 Local geometric description construction: Construct local geometric description information for each semantic key point to represent the directional consistency and extension trend in its neighborhood; S3.3, Geometric Consistency Constraint Filtering: Based on the random sampling consistency algorithm, geometric consistency verification is performed on semantic key points and their associated edges to remove isolated edge points that do not conform to the continuous topological characteristics of the crack; S3.4 Edge Semantic Classification: Combining the fuzzy C-means clustering method, candidate edges are divided into structural crack edges and non-crack texture noise edges based on the connectivity, contrast and geometric consistency characteristics of the edges.

2. The quantitative crack identification method according to claim 1, characterized in that: In step S1, the camera pose information includes the camera mounting height, pitch angle, and external parameter information. The geometric correction is a perspective correction process, and then the perspective image is converted into an orthographic projection image by inverse perspective transformation.

3. The quantitative crack identification method according to claim 1, characterized in that: In S2, the edge enhancement processing includes noise suppression using anisotropic diffusion filtering, perception of different scales by constructing an image pyramid, and crack guidance enhancement using directional gradient operators.

4. The quantitative crack identification method according to claim 3, characterized in that: The S2 edge enhancement process includes the following steps: S2.1 Noise Suppression Filtering: Anisotropic diffusion filtering is applied to the orthophoto image to suppress background noise while preserving crack edge features; S2.2, Image Construction at Different Scales: Construct image pyramids at different scales and process images at different resolution levels; S2.3 Directional Edge Enhancement: Different directional gradient operators are introduced on images at various scales to perform directional enhancement on edge structures with linear extension characteristics, forming candidate edge sets at different scales.

5. The quantitative crack identification method according to claim 1, characterized in that: In S4, fracture compensation is to dynamically compensate and connect fractured areas caused by changes in illumination, shading, or texture interference, based on the continuity of the crack extension direction.

6. The quantitative crack identification method according to claim 1, characterized in that, The S5 quantitative measurement includes the following steps: S5.1 Crack skeleton extraction: Morphological refinement of the crack path network is performed to extract the sub-pixel level central skeleton line of the crack. S5.2 Crack width calculation: At each sampling node on the center skeleton line of the crack, search for the crack edge projection point along the normal direction, and combine it with the real physical size corresponding to a single pixel obtained in S1 to convert the pixel width into the real physical width. S5.3 Crack Length Calculation: Integrate the distance between each node along the crack skeleton path to calculate the true physical length of the crack; S5.4 Output Results: Output the physical width, length, and spatial distribution information of the cracks.

7. A crack quantitative identification system integrating photogrammetry and semantic key points, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, it implements the crack quantitative identification method that combines photogrammetry and semantic key points as described in any one of claims 1-6.

Images