A construction quality safety assessment system and method

Abstract

The present invention discloses a construction quality and safety assessment system and method. Through multi-angle shooting, key parts of the construction site are fully covered, and missed detection due to incomplete angles is avoided, thereby improving the accuracy of subsequent construction quality and safety assessment. In addition, it is very critical to cover the detection areas of cracks and settlements. By edge recognition of construction images, image processing and analysis, building crack images and ground settlement images are identified. Edge recognition technology is used to extract cracks in building images and settlement features in ground images, which can effectively distinguish between cracks and settlement areas, and efficiently identify problematic areas, providing clear data support for subsequent quality assessments. Finally, the building crack images and the ground settlement images are processed and analyzed by a pre-trained construction quality and safety assessment model, and the safety assessment score of the current construction site is obtained as an output.

Description

Technical Field

[0001] The present invention relates to the field of building construction quality and safety assessment, and in particular to a building construction quality and safety assessment system and method. Background Art

[0002] Construction quality and safety assessment is crucial to construction safety. High-quality construction can improve the safety performance of buildings.

[0003] Traditional construction quality and safety assessment methods usually use manual assessment or 3D laser point cloud data recognition, and then analyze and evaluate the recognition results of 3D laser point cloud data based on artificial intelligence algorithms to obtain the construction quality and safety assessment results;

[0004] However, while manual assessments are simple, they rely heavily on experience and subjective judgment, making them susceptible to factors such as operator emotion and expertise. This results in instability and low accuracy. Especially in large-scale or complex construction projects, manual inspections are time-consuming and prone to omissions, making it difficult to provide efficient and reliable assessments.

[0005] Laser point cloud data, as a relatively advanced assessment method, uses laser scanning technology to obtain three-dimensional point cloud data of buildings, accurately capturing the building's geometry and details. Although laser point cloud data has relatively good assessment results, its practical application faces the problem of huge deployment and data processing costs. The installation, commissioning, and data collection processes of laser equipment are cumbersome and require high technical requirements and manual intervention. Furthermore, the subsequent steps of laser point cloud data processing are very complex, involving extensive data conversion and analysis, which is time-consuming and requires a high-performance computing platform, limiting its application in most construction projects.

[0006] With the development of image processing technology, high-definition cameras and multi-angle acquisition techniques are widely used on construction sites. Researchers have further discovered that by acquiring multiple construction images and combining them with edge recognition technology, they can effectively identify safety hazards such as building cracks and ground subsidence, thereby providing data support for construction quality and safety assessments. However, due to the complexity of building cracks, traditional edge recognition technology cannot fully detect and analyze detailed building cracks, resulting in poor assessment results. Summary of the Invention

[0007] The purpose of the present invention is to provide a construction quality safety assessment system and method to solve the above technical problems pointed out in the prior art.

[0008] The present invention provides a construction quality and safety assessment system, comprising an image acquisition module, a recognition module and an assessment module;

[0009] The image acquisition module is used to acquire multiple construction images of the current construction site through multi-angle acquisition using a high-definition camera;

[0010] The building construction image includes a building image and a building ground image;

[0011] The recognition module is used to analyze and process the building construction image through edge recognition to obtain a building crack image and a ground subsidence image of the current construction site;

[0012] The evaluation module is used to process and analyze the building crack image and the ground subsidence image through a pre-trained construction quality safety evaluation model, and output a safety evaluation score for the current construction site.

[0013] Preferably, the identification module includes a pre-processing module, a depth image generation and alignment module, a crack identification module and a settlement identification module;

[0014] The preprocessing module is configured to obtain a first building construction image through a preprocessing operation based on the building construction image;

[0015] The first building construction image includes a first building image and a first building ground image;

[0016] The depth image generation and alignment module is used to obtain the camera internal parameters of the high-definition camera; generate the building construction image based on the camera internal parameters to obtain a depth building construction image; align the depth building construction image with the first building ground image to obtain a construction image group;

[0017] The crack recognition module is configured to obtain a building crack image through edge detection and recognition analysis based on the first building image;

[0018] The settlement recognition module is used to obtain a ground settlement image through edge detection, recognition and analysis based on the construction image group.

[0019] Specifically, the crack identification module includes a first extraction module, a clustering module, a second extraction module and an iterative analysis module;

[0020] The first extraction module is configured to extract multi-dimensional feature data of each building image pixel point in the first building image;

[0021] The multi-dimensional feature data includes grayscale value feature data, gradient value feature data and gradient direction feature data;

[0022] The clustering module is configured to cluster the building image pixels based on the multi-dimensional feature data to obtain a plurality of clusters; determine the image area of ​​the cluster based on the grayscale value mean of each building image pixel in the cluster; obtain an image area to be processed based on the image area of ​​the cluster; and construct an image area label matrix L based on the cluster labels of the image area to be processed;

[0023] The image area includes a background area image, a crack area image and an edge area image;

[0024] The image area to be processed includes a crack area image and an edge area image;

[0025] Each element in the image region label matrix L represents the image region to which each building image pixel point in the first building image belongs;

[0026] The number of clusters is 3;

[0027] The second extraction module is used to extract the position feature data of each building image pixel in the image area label matrix L; construct a spatial distribution feature matrix S based on the position feature data; extract the shape features of the spatial distribution feature matrix S to construct a shape feature matrix M;

[0028] The iterative analysis module is used to perform iterative analysis on the image of the area to be processed based on the spatial distribution feature matrix S and the shape feature matrix M in combination with probability calculation and morphological processing to identify and obtain a building crack image.

[0029] Specifically, the iterative analysis module includes a first construction submodule, a second construction submodule, a first segmentation submodule, an iterative analysis submodule and a third construction submodule;

[0030] The first construction submodule is used to construct a comprehensive feature matrix C based on the spatial distribution feature matrix S and the shape feature matrix M;

[0031] The second construction submodule is used to calculate the probability r of each pixel point of the building image belonging to the crack area image based on the comprehensive feature matrix C; and construct a scoring matrix R based on the probability r;

[0032] The first segmentation submodule is configured to mark, based on a preset initial segmentation threshold T0, pixel points of the building image corresponding to a probability r greater than or equal to the initial segmentation threshold T0 in the scoring matrix R as crack regions, thereby obtaining a first crack region image; and perform a morphological processing operation on the first crack region image to output a second crack region image;

[0033] The iterative analysis submodule is used to perform iterative analysis and screening based on the second crack region image in combination with the spatial distribution feature matrix S and the shape feature matrix M to obtain target crack pixel points;

[0034] The third construction submodule is used to construct a building crack image based on all the target crack pixel points.

[0035] Specifically, the iterative analysis submodule includes a second segmentation submodule, a seed point acquisition module, an iterative optimization growth module and a fourth construction submodule;

[0036] The second segmentation submodule is configured to segment the second crack region image based on a preset second segmentation threshold T1 to obtain a first target crack pixel point and a target crack pixel point to be determined;

[0037] The seed point acquisition module is used to acquire all first target crack pixel points at the outermost periphery of the first target crack pixel point bounding box and determine them as seed points;

[0038] The iterative optimization growth module is used to optimize the spatial distribution characteristics of the seed points. and shape characteristics Performing iterative optimization growth processing analysis on the target crack pixel point to be determined to obtain a second target crack pixel point;

[0039] The fourth construction submodule is configured to obtain a target crack pixel point based on the first target crack pixel point and the second target crack pixel point.

[0040] Specifically, the iterative optimization growth module is further used to traverse each of the seed points, obtain the target crack pixel points to be determined in the neighborhood pixels of the seed point, and determine them as the pixel points to be grown;

[0041] Traverse each of the pixels to be grown, based on the spatial distribution characteristics of the pixels to be grown Shape characteristics of the growth And the spatial distribution characteristics of the seed points and shape characteristics Calculate the growth similarity measure between the pixel point to be grown and the seed point ;

[0042] Based on the growth similarity metric Perform judgment iteration and output to obtain a second target crack pixel point among the target crack pixel points to be determined.

[0043] Preferably, the iterative optimization growth module is specifically used to determine the growth similarity metric Is it greater than or equal to the preset growth similarity measurement threshold? ; If so, determine that the pixel point to be grown is the second target crack pixel point; determine whether all the pixel points to be grown have been traversed. If so, obtain a new seed point based on the second target crack pixel point corresponding to the seed point, and return the new seed point to the above operation to continue the traversal operation until all the target crack pixel points to be determined have been traversed, and output all the second target crack pixel points.

[0044] During a specific implementation, the iterative optimization growth module is further configured to obtain a third crack region image based on the first target crack pixel point and the current second target crack pixel point; divide the third crack region image into a plurality of grid regions of equal size; and obtain the second target crack pixel point in each grid region as a first candidate seed point.

[0045] Selecting the first candidate pixel point with the highest growth similarity measure in each of the grid areas as the second candidate seed point;

[0046] Calculate the local density of each of the second candidate seed points based on the preset window radius w ;

[0047] Select the local density of the second candidate seed points Greater than or equal to the preset local density threshold of the second candidate seed point The second candidate seed point is used as the new seed point.

[0048] Accordingly, the present invention also provides a construction quality safety assessment method, comprising the following steps:

[0049] Use high-definition cameras to acquire multiple construction images of the current construction site through multi-angle acquisition;

[0050] Analyzing and processing the building construction image through edge recognition to obtain a building crack image and a ground subsidence image of the current construction site;

[0051] The building crack image and the ground subsidence image are processed and analyzed by a pre-trained construction quality safety assessment model, and a safety assessment score of the current construction site is output.

[0052] Compared with the prior art, the embodiments of the present invention have at least the following technical advantages:

[0053] From the analysis of the above-mentioned construction quality and safety assessment system and method provided by the present invention, it can be known that in specific applications, through multi-angle shooting methods, the key parts of the construction site can be fully covered to ensure that each area where cracks or settlements may occur is carefully observed, providing sufficient data support for subsequent crack and settlement detection, avoiding missed detections due to incomplete angles, and improving the accuracy of subsequent construction quality and safety assessments. Covering the detection area for cracks and settlements is very critical, because cracks and settlements are the two most important indicators in construction quality and safety, affecting the stability and safety of the building; further, by performing edge recognition on the construction image, processing and analyzing the image, identifying the building crack image and the ground settlement image, the edge recognition technology is used to extract the crack features in the building image and the settlement features in the ground image, which can effectively distinguish between cracks and settlement areas, and efficiently identify problematic areas, providing clear data support for subsequent quality assessments. ; Finally, the identified building crack images and ground subsidence images are input into the pre-trained construction quality and safety assessment model for processing and analysis, and finally the safety assessment score of the construction site is output; by converting the crack and settlement images into score form, the construction quality and safety assessment model can quantify the safety status of the construction site, concretize the severity of each crack and settlement, form a scoring system, simplify the complex construction quality assessment process, and accurately reflect the potential risks and hidden dangers of the construction site according to the assessment score, helping technical personnel to promptly discover safety problems in construction, so as to carry out targeted remediation; the above-mentioned embodiment of the present application ensures comprehensive coverage of key parts of the construction site through multi-angle high-definition image acquisition, and the edge recognition technology effectively extracts the image features of building cracks and ground subsidence. Finally, through the pre-trained construction quality and safety assessment model, the severity of cracks and settlement is quantified and accurately assessed, providing data support for subsequent remedial measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a schematic diagram of the overall architecture of a construction quality and safety assessment system;

[0055] Figure 2 This is a schematic diagram of the main process of a construction quality and safety assessment method;

[0056] Figure 3 A schematic diagram of the operation steps for identifying building crack images and ground subsidence images at a current construction site in a construction quality safety assessment method;

[0057] Figure 4 A schematic diagram of the operation steps for analyzing and obtaining building crack images in a construction quality safety assessment method;

[0058] Figure 5A schematic diagram of cluster label simulation in a construction quality and safety assessment method;

[0059] Figure 6 A schematic diagram of the operation steps for identifying and obtaining a building crack image in a building construction quality safety assessment method;

[0060] Figure 7 2. A simulation diagram of a construction quality safety assessment method for distinguishing a first target crack pixel point from a target crack pixel point to be determined by a second segmentation threshold T1;

[0061] Figure 8 A schematic diagram of the operation steps for analyzing and obtaining a second target crack pixel point in a construction quality safety assessment method;

[0062] Figure 9 Schematic diagram of the operational steps for screening and obtaining new seed points in a construction quality and safety assessment method.

[0063] Figure numerals: image acquisition module 10, recognition module 20, evaluation module 30, preprocessing module 21, depth image generation and alignment module 22, crack recognition module 23, settlement recognition module 24, first extraction module 231, clustering module 232, second extraction module 233, iterative analysis module 234; first construction submodule 2341, second construction submodule 2342, first segmentation submodule 2343, iterative analysis submodule 2344, third construction submodule 2345, second segmentation submodule 23441, seed point acquisition module 23442, iterative optimization growth module 23443, fourth construction submodule 23444. DETAILED DESCRIPTION

[0064] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0065] The present invention will be further described in detail below through specific embodiments in conjunction with the accompanying drawings.

[0066] Example 1

[0067] like Figure 1 As shown, the first embodiment of the present invention provides a construction quality safety assessment system, including an image acquisition module 10, a recognition module 20 and an assessment module 30;

[0068] The image acquisition module 10 is used to acquire multiple construction images of the current construction site through multi-angle acquisition using a high-definition camera;

[0069] The building construction image includes a building image and a building ground image;

[0070] The recognition module 20 is configured to analyze and process the construction image through edge recognition to obtain a building crack image and a ground subsidence image of the current construction site;

[0071] The evaluation module 30 is used to process and analyze the building crack image and the ground subsidence image using a pre-trained construction quality safety evaluation model, and output a safety evaluation score for the current construction site.

[0072] Preferably, the identification module 20 includes a pre-processing module 21, a depth image generation and alignment module 22, a crack identification module 23 and a settlement identification module 24;

[0073] The pre-processing module 21 is configured to obtain a first building construction image through a pre-processing operation based on the building construction image;

[0074] The first building construction image includes a first building image and a first building ground image;

[0075] The depth image generation and alignment module 22 is used to obtain the camera internal parameters of the high-definition camera; generate the building construction image based on the camera internal parameters to obtain a depth building construction image; align the depth building construction image with the first building ground image to obtain a construction image group;

[0076] The crack recognition module 23 is configured to obtain a building crack image through edge detection and recognition analysis based on the first building image;

[0077] The settlement identification module 24 is configured to obtain a ground settlement image through edge detection, identification and analysis based on the construction image group.

[0078] Specifically, the crack identification module 23 includes a first extraction module 231, a clustering module 232, a second extraction module 233 and an iterative analysis module 234;

[0079] The first extraction module 231 is configured to extract multi-dimensional feature data of each building image pixel in the first building image;

[0080] The multi-dimensional feature data includes grayscale value feature data, gradient value feature data and gradient direction feature data;

[0081] The clustering module 232 is configured to cluster the building image pixels based on the multi-dimensional feature data to obtain a plurality of clusters; determine the image region of the cluster based on the mean grayscale value of each building image pixel in the cluster; obtain an image region to be processed based on the image region of the cluster; and construct an image region label matrix L based on the cluster labels of the image region to be processed;

[0082] The image area includes a background area image, a crack area image and an edge area image;

[0083] The image area to be processed includes a crack area image and an edge area image;

[0084] Each element in the image region label matrix L represents the image region to which each building image pixel point in the first building image belongs;

[0085] The number of clusters is 3;

[0086] The second extraction module 233 is used to extract the position feature data of each building image pixel in the image region label matrix L; construct a spatial distribution feature matrix S based on the position feature data; extract the shape features of the spatial distribution feature matrix S to construct a shape feature matrix M;

[0087] The iterative analysis module 234 is used to perform iterative analysis on the image of the area to be processed based on the spatial distribution feature matrix S and the shape feature matrix M in combination with probability calculation and morphological processing to identify and obtain a building crack image.

[0088] Specifically, the iterative analysis module 234 includes a first construction submodule 2341, a second construction submodule 2342, a first segmentation submodule 2343, an iterative analysis submodule 2344 and a third construction submodule 2345;

[0089] The first construction submodule 2341 is configured to construct a comprehensive feature matrix C based on the spatial distribution feature matrix S and the shape feature matrix M;

[0090] The second construction submodule 2342 is configured to calculate, based on the comprehensive feature matrix C, the probability r that each pixel point in the building image belongs to the crack region image; and construct a scoring matrix R based on the probability r;

[0091] The first segmentation submodule 2343 is configured to mark, based on a preset initial segmentation threshold T0, pixel points of the building image corresponding to a probability r greater than or equal to the initial segmentation threshold T0 in the scoring matrix R as crack regions, thereby obtaining a first crack region image; and perform morphological processing on the first crack region image to output a second crack region image.

[0092] The iterative analysis submodule 2344 is configured to perform iterative analysis and screening based on the second crack region image in combination with the spatial distribution feature matrix S and the shape feature matrix M to obtain target crack pixel points;

[0093] The third construction submodule 2345 is used to construct a building crack image based on all the target crack pixel points.

[0094] Specifically, the iterative analysis submodule 2344 includes a second segmentation submodule 23441, a seed point acquisition module 23442, an iterative optimization growth module 23443 and a fourth construction submodule 23444;

[0095] The second segmentation submodule 23441 is configured to segment the second crack region image based on a preset second segmentation threshold T1 to obtain first target crack pixel points and target crack pixel points to be determined;

[0096] The seed point acquisition module 23442 is configured to acquire all first target crack pixel points at the outermost periphery of the first target crack pixel point bounding box and determine them as seed points;

[0097] The iterative optimization growth module 23443 is used to optimize the spatial distribution characteristics of the seed points based on the and shape characteristics Performing iterative optimization growth processing analysis on the target crack pixel point to be determined to obtain a second target crack pixel point;

[0098] The fourth construction submodule 23444 is configured to obtain a target crack pixel point based on the first target crack pixel point and the second target crack pixel point.

[0099] Specifically, the iterative optimization growth module 23443 is further configured to traverse each of the seed points, obtain target crack pixel points to be determined from the neighboring pixel points of the seed point, and determine them as pixel points to be grown;

[0100] Traverse each of the pixels to be grown, based on the spatial distribution characteristics of the pixels to be grown Shape characteristics of the growth And the spatial distribution characteristics of the seed points and shape characteristics Calculate the growth similarity measure between the pixel point to be grown and the seed point ;

[0101] The growth similarity metric The calculation method is: ;

[0102] Where, 、 is the weight coefficient, and satisfies α+β=1; and is the norm parameter;

[0103] Determine the growth similarity metric Is it greater than or equal to the preset growth similarity measurement threshold? ; If so, determine that the pixel point to be grown is the second target crack pixel point; determine whether all the pixel points to be grown have been traversed. If so, obtain a new seed point based on the second target crack pixel point corresponding to the seed point, and return the new seed point to the above operation to continue the traversal operation until all the target crack pixel points to be determined have been traversed, and output all the second target crack pixel points.

[0104] During specific implementation, the iterative optimization growth module 23443 is further configured to obtain a third crack region image based on the first target crack pixel point and the current second target crack pixel point; divide the third crack region image into a plurality of grid regions of equal size; and obtain the second target crack pixel point in each grid region as a first candidate seed point.

[0105] Selecting the first candidate pixel point with the highest growth similarity measure in each of the grid areas as the second candidate seed point;

[0106] Calculate the local density of each of the second candidate seed points based on the preset window radius w ;

[0107] The local density of the second candidate seed point The calculation method is: ;

[0108] Where, is the window radius; is the offset in the vertical direction; is the offset in the horizontal direction;

[0109] Select the local density of the second candidate seed points Greater than or equal to the preset local density threshold of the second candidate seed point The second candidate seed point is used as the new seed point.

[0110] Example 2

[0111] like Figure 2 As shown, the second embodiment of the present invention provides a construction quality safety assessment method, including the following steps:

[0112] Step S10: using a high-definition camera to acquire multiple construction images of the current construction site through multi-angle acquisition;

[0113] The building construction image includes a building image and a building ground image;

[0114] It should be noted that the above-mentioned embodiment of the present application uses a high-definition camera to capture the current construction site from multiple angles to obtain multiple construction images; the above-mentioned construction images refer to building parts (such as wall joints, window frames, door frames, etc.) that are prone to cracks; the above-mentioned building ground images refer to ground areas that are prone to building settlement; the above-mentioned building cracks and building ground settlement are of vital importance to the quality and safety of construction;

[0115] Shooting from multiple angles ensures full coverage of key parts of the building, ensures comprehensive shooting of the construction site, and improves the accuracy of subsequent construction quality and safety assessments.

[0116] Step S20: analyzing and processing the building construction image through edge recognition to obtain a building crack image and a ground subsidence image of the current construction site;

[0117] Step S30: Processing and analyzing the building crack image and the ground subsidence image using a pre-trained construction quality safety assessment model to output a safety assessment score for the current construction site;

[0118] It should be noted that the above embodiment of the present application is to input the building crack data and the ground subsidence data into the pre-trained construction quality and safety assessment model, and then process and analyze the construction quality and safety assessment model to output a safety assessment score; after obtaining the safety assessment score, the technician will generate a construction quality and safety assessment report based on the safety assessment score and send it to the management party. Further, the management party will take corresponding remedial measures based on the construction quality and safety assessment report; specifically, the pre-trained construction quality and safety assessment model is based on preset rules, and the building crack image and ground subsidence data of the current construction site are input into the pre-trained construction quality and safety assessment model. The settlement image is converted into a score, and each crack and settlement severity corresponds to a certain deduction point, forming the building's construction safety assessment score. For example, a slight crack or settlement deducts 5 points, a moderate deduction 10 points, and a severe deduction 20 points. The building construction safety assessment score is calculated based on the total severity score of all cracks in the building image and all settlements in the ground image. Furthermore, remedial measures mean that slight cracks or settlements can be repaired and reinforced, such as crack grouting and ground reinforcement; moderate cracks or settlements require structural reinforcement or foundation adjustment; and severe cracks or settlements may require demolition and reconstruction or large-scale repairs.

[0119] The above-mentioned embodiment of the present application can fully cover the key parts of the construction site through multi-angle shooting, ensuring that every area where cracks or settlements may occur is carefully observed, providing sufficient data support for subsequent crack and settlement detection, avoiding missed detections due to incomplete angles, and improving the accuracy of subsequent construction quality and safety assessments. Multi-angle acquisition is to cover the detection areas of cracks and settlements, which is very critical because cracks and settlements are the two most important indicators in construction quality and safety, affecting the stability and safety of buildings; further, by performing edge recognition on the construction images, processing and analyzing the images, identifying building crack images and ground settlement images, edge recognition technology is used to extract cracks in building images and settlement features in ground images, which can effectively distinguish crack and settlement areas, and efficiently identify problematic areas, providing clear data support for subsequent quality assessments;

[0120] Finally, the identified building crack images and ground subsidence images are input into a pre-trained construction quality and safety assessment model for processing and analysis, and finally a safety assessment score for the construction site is output; by converting the crack and settlement images into score form, the construction quality and safety assessment model can quantify the safety status of the construction site, concretize the severity of each crack and settlement, form a scoring system, and simplify the complex construction quality assessment process. According to the assessment score, it can accurately reflect the potential risks and hidden dangers of the construction site, help technical personnel to promptly discover safety problems in construction, and thus carry out targeted remediation; the above-mentioned embodiment of the present application ensures comprehensive coverage of key parts of the construction site through multi-angle high-definition image acquisition, and edge recognition technology effectively extracts image features of building cracks and ground subsidence. Finally, through the pre-trained construction quality and safety assessment model, the severity of cracks and settlement is quantified and accurately assessed, providing data support for subsequent remedial measures.

[0121] Specifically, if Figure 3 As shown, in step S20, the building crack image and the ground subsidence image of the current construction site are obtained by analyzing and processing the building construction image through edge recognition, including the following steps:

[0122] Step S21: obtaining a first building construction image through a preprocessing operation based on the building construction image;

[0123] The first building construction image includes a first building image and a first building ground image;

[0124] It should be noted that the preprocessing operation in the above-mentioned embodiment of the present application includes a grayscale processing operation, a denoising processing operation and a contrast enhancement processing operation; the above-mentioned image grayscale operation is to convert a color building construction image into a grayscale image to simplify the processing process; the denoising processing operation is to use Gaussian filtering, median filtering and other methods to denoise the grayscale image to eliminate random noise and salt and pepper noise in the image; the contrast enhancement processing operation is to use histogram equalization or adaptive contrast enhancement technology to enhance the contrast of cracks and settlement areas in the image to make them easier to identify; the above-mentioned building construction image includes a building image and a building ground image; accordingly, the first construction image obtained after the preprocessing operation is performed on the building construction image will correspondingly include the first building image obtained after the building image is preprocessed and the first building ground image obtained after the building ground image is preprocessed;

[0125] Step S22: Obtaining the camera internal parameters of the high-definition camera; generating a depth building construction image based on the camera internal parameters of the building construction image; aligning the depth building construction image with the first building ground image to obtain a construction image group;

[0126] It should be noted that the above embodiment of the present application aligns the depth image with the first building construction image to ensure that the spatial correspondence between the 2D image and the depth data is accurate, thereby ensuring that the subsequent analysis of the building crack image and the ground subsidence image is accurate;

[0127] The generation of a depth building construction image based on the in-camera building construction image in the above-mentioned embodiments of the present application is common knowledge to those skilled in the art and will not be elaborated in this application.

[0128] Step S23: obtaining a building crack image through edge detection and recognition analysis based on the first building image;

[0129] Step S24: obtaining a ground subsidence image through edge detection, recognition and analysis based on the construction image group.

[0130] It should be noted that ground subsidence often manifests as depressions or height changes in local areas, which are manifested as sudden changes in depth data in depth images. Generally, depth images of flat ground images often show uniform depth changes. Based on this, the embodiments of the present application combine depth images (i.e., the above-mentioned construction image group) with edge detection to not only detect the shape of the subsidence area, but also accurately assess the depth and range of the subsidence.

[0131] The above-mentioned embodiment of the present application first performs pre-processing operations such as grayscale conversion, denoising and contrast enhancement on the construction image to obtain a "first construction image". Removing noise from the image can significantly improve the accuracy of subsequent analysis, avoid noise affecting the detection of cracks and settlement areas, and enhance the contrast of cracks and settlement areas, making these areas more conspicuous, facilitating subsequent edge detection and analysis, and improving the accuracy of image recognition; further, the camera intrinsic parameters of the high-definition camera are obtained, and the camera intrinsic parameters are used to generate a depth construction image together with the construction image. Through accurate camera intrinsic parameters, it can ensure accurate spatial alignment of 2D images and depth data, so that subsequent analysis can be based on a consistent The spatial reference of the building image is established, and the spatial correspondence between the depth data and the actual image is accurate, so that the spatial information of the crack image and the ground subsidence image is synchronized, thereby improving the accuracy of the analysis; further, through edge detection technology, the significant edges in the image are extracted, and the boundaries of the crack area are clearly identified, so that the position and shape of the crack can be accurately located, which is convenient for subsequent processing. The characteristics of the cracks are accurately extracted from the building image, and the potential safety hazard areas are marked. Based on the changes in the depth image (such as the sudden change in the depth data), the area of ​​ground subsidence is accurately identified and its severity is evaluated. Combined with edge detection, the boundaries of the settlement area can be accurately delineated, thereby improving the accuracy of settlement analysis.

[0132] During the specific implementation of the above-mentioned embodiment of the present application, technical personnel found that the detection and analysis of building crack images may result in incomplete or insufficiently accurate images of the identified building cracks due to reasons such as the complexity of the cracks, thereby causing inaccurate evaluation results in the subsequent construction quality and safety assessment. Therefore, the recognition of building crack images requires further refined analysis to identify a complete building crack image.

[0133] Specifically, if Figure 4 As shown, in step S23, a building crack image is obtained by edge detection and recognition analysis based on the first building image, including the following operation steps:

[0134] Step S231: extracting multi-dimensional feature data of each building image pixel in the first building image;

[0135] The multi-dimensional feature data includes grayscale value feature data, gradient value feature data and gradient direction feature data;

[0136] It should be noted that the above-mentioned embodiments of the present application can effectively extract edge information of cracks by calculating gradient values ​​and gradient directions (i.e., the above-mentioned gradient value feature data and gradient direction feature data). Local changes in cracks are usually accompanied by large gradient changes, especially in edge areas. Therefore, the gradient features provide significant edge information for subsequent crack detection, helping to separate cracks from other patterns or backgrounds; by considering grayscale values ​​(i.e., the above-mentioned grayscale value feature data), the features not only rely on edge changes, but can also capture grayscale differences in crack areas, which helps to distinguish background textures from actual crack areas and reduce the possibility of misidentification.

[0137] Step S232: clustering the building image pixels based on the multi-dimensional feature data to obtain a plurality of clusters; determining the image regions of the clusters based on the grayscale value mean of each building image pixel in the clusters; filtering the image regions of the clusters to obtain the image regions to be processed; and constructing the image region label matrix L based on the cluster labels of the image regions to be processed.

[0138] The image area includes a background area image, a crack area image and an edge area image;

[0139] The image area to be processed includes a crack area image and an edge area image;

[0140] Each element in the image region label matrix L represents the image region to which each building image pixel point in the first building image belongs;

[0141] The number of clusters is 3; it should be noted that, Figure 5As shown in the figure, the cluster label refers to the label of the image area where each pixel is located; the cluster label includes the background area image label (label 1 in the figure), the crack area image label (label 3 in the figure) and the edge area image label (label 2 in the figure); clusters are of three types, namely, clusters composed of background area pixels, clusters composed of crack area pixels and clusters composed of edge area pixels;

[0142] It should be noted that the above clustering generally uses the K-means clustering method. Since the first building image in the embodiment of the present application mainly contains three regions, namely the background region (the image region of the building), the crack region (the region where cracks appear on the building), and the edge region (the edge region of the cracks on the building); therefore, the embodiment of the present application clusters the building image pixels through K-means clustering based on the multi-dimensional feature data to obtain three clusters, which correspond to the background region image cluster cluster, the crack region image cluster cluster, and the edge region image cluster cluster respectively;

[0143] Furthermore, under normal circumstances, the pixels of the background area image appear in a uniform grayscale form, the pixels of the crack area image appear in a deepened grayscale form, and the pixels of the edge area image appear in a gradient grayscale form. Therefore, the embodiment of the present application determines the image area of ​​each cluster by the grayscale value mean of the pixel points of each building image in the cluster, that is, the cluster with the smallest grayscale value mean is determined as the background area image, the cluster with a medium grayscale value mean is determined as the edge area image, and the cluster with the largest grayscale value mean is determined as the crack area image; further, since in the clustering process of this step, only a rough clustering process of the pixels is performed, the cracks and the edges of the cracks (that is, the crack area image and the edge area image) cannot be completely and finely distinguished to obtain a complete crack area image (the complete crack area image is the above-mentioned building crack image). Therefore, the crack area image and the edge area image need to be further processed and analyzed. The above-mentioned embodiment of the present application screens the image area to be processed (that is, the crack area image and the edge area image) to screen out the background area image that accounts for a large proportion in the first building image, thereby reducing the subsequent calculation pressure and improving the efficiency of analysis and calculation;

[0144] The above-mentioned embodiment of the present application groups images according to pixel features (such as grayscale value, gradient, direction, etc.) through K-means clustering, which can automatically distinguish crack areas from other areas (such as background, edge). By setting the initial cluster number k=3, the crack area, edge area and background area can be effectively divided; the label matrix in the clustering algorithm can clearly mark the category to which each pixel belongs. Through these labels, subsequent steps can further process different areas (cracks, edges, background), effectively reducing the confusion caused by complex crack structures.

[0145] Step S233: extracting position feature data of each building image pixel in the image region label matrix L; constructing a spatial distribution feature matrix S based on the position feature data; extracting shape features of the spatial distribution feature matrix S to construct a shape feature matrix M;

[0146] It should be noted that the above-mentioned embodiments of the present application use the spatial distribution feature matrix S and the shape feature matrix M to deeply analyze the geometric morphology of each cluster, including the connectivity and shape (such as aspect ratio, contour, area, etc.) of the cluster, to improve the effect of identifying broken crack areas or cracks with complex morphology, and avoid missed detection or false detection.

[0147] Step S234: performing iterative analysis on the image of the area to be processed based on the spatial distribution feature matrix S and the shape feature matrix M in combination with probability calculation and morphological processing to identify and obtain a building crack image.

[0148] It should be noted that step S231 to step S232 is a coarse segmentation processing operation, which aims to filter out the background area image in the first building image, and step S233-step S234 is a fine segmentation processing operation, which aims to perform fine segmentation processing on the image of the area to be processed obtained after filtering out the background area image, that is, to perform fine segmentation on the crack area image and the edge of the crack area image (edge ​​area image), filter out the edge, and obtain a complete crack area image, that is, the building crack image.

[0149] Specifically, if Figure 6 As shown, in step S234, based on the spatial distribution feature matrix S and the shape feature matrix M and in combination with probability calculation and morphological processing, the image of the area to be processed is iteratively analyzed and processed to identify and obtain a building crack image, including the following steps:

[0150] Step S2341: constructing a comprehensive feature matrix C based on the spatial distribution feature matrix S and the shape feature matrix M;

[0151] It should be noted that the above embodiment of the present application first normalizes the spatial distribution feature matrix S and the shape feature matrix M to ensure that different features have the same scale; further, the normalized spatial distribution feature matrix S and the shape feature matrix M are fused by feature weighting to obtain a comprehensive feature matrix C;

[0152] Step S2342: Calculating based on the comprehensive feature matrix C, obtaining the probability r that each pixel point in the building image belongs to the crack region image; constructing a scoring matrix R based on the probability r;

[0153] It should be noted that the above embodiment of the present application calculates the probability value of each pixel belonging to a crack based on the comprehensive feature matrix C (which can be calculated using a Sigmoid function or a Softmax function), and generates a scoring matrix R. Each element in the scoring matrix R represents the probability that each pixel in the building image belongs to a crack;

[0154] Step S2343: Based on a preset initial segmentation threshold T0, pixel points of the building image corresponding to the probability r greater than or equal to the initial segmentation threshold T0 in the scoring matrix R are marked as crack regions, thereby obtaining a first crack region image; and the first crack region image is subjected to morphological processing to output a second crack region image.

[0155] It should be noted that, in the probability calculation of the above-mentioned embodiment of the present application and the segmentation processing operation of the initial segmentation threshold T0, the first crack area image obtained will filter out most of the edge area images (that is, most of the pixels at the edges of the crack area images that do not belong to the building crack image are filtered out). However, due to the complexity of the building cracks, a small part of the edges will still remain. In order to ensure the accuracy of the subsequent construction quality and safety assessment, it is also necessary to screen out and optimize the small part of the remaining edges to obtain a more refined building crack image. For specific operations, please refer to the subsequent steps.

[0156] Step S2344: performing iterative analysis and screening based on the second crack region image in combination with the spatial distribution feature matrix S and the shape feature matrix M to obtain target crack pixel points;

[0157] Step S2345: constructing a building crack image based on all the target crack pixel points.

[0158] It should be noted that the above-mentioned embodiment of the present application first normalizes the spatial distribution feature matrix S and the shape feature matrix M to ensure that they are fused at the same scale, and fuses the normalized matrices by feature weighting to obtain a comprehensive feature matrix C, so that in the complex building crack image, the contribution of each feature is fully utilized to enhance the accurate identification of the crack area; further, by calculating the crack probability of each pixel, the crack and non-crack areas are accurately distinguished, and accurate probability information is provided for subsequent segmentation, thereby preventing misjudgment or omission in complex crack scenes and improving the recognition accuracy of the crack image; further, using a preset initial segmentation threshold T0, the pixels in the scoring matrix R are classified according to the probability value, and those greater than or equal to T0 are marked as crack areas to obtain a first crack area image; the initial threshold T0 is used to effectively remove pixels that do not meet the crack characteristics, screen out most of the non-crack areas, and ensure the initial accuracy of the crack area image; and further optimize the edge processing through morphological processing to eliminate the problem of unclear edges or missing small parts of cracks due to the complex morphology of building cracks, thereby improving the fineness of the image and providing a better basis for subsequent analysis;

[0159] Furthermore, based on the second crack area image, an iterative analysis is performed in combination with the spatial distribution feature matrix S and the shape feature matrix M to further optimize the identification of the crack area, screen out the target crack pixel points, further explore the characteristics of complex cracks in the image, and combine the spatial distribution and shape characteristics to identify crack areas that may be missed, especially in cracks with complex structures or fuzzy shapes, thereby improving the comprehensiveness and accuracy of the image; finally, through the aggregation of target crack pixel points, the integrity and accuracy of the crack area are ensured, avoiding missed detection or misjudgment, ensuring the high-quality output of the building crack image, and providing reliable image support for subsequent building quality and safety assessments.

[0160] Specifically, in step S2344, based on the second crack region image and in combination with the spatial distribution feature matrix S and the shape feature matrix M, target crack pixel points are screened and obtained, including the following operation steps:

[0161] Step S23441: Segmenting the second crack region image based on a preset second segmentation threshold T1 to obtain a first target crack pixel point and a target crack pixel point to be determined;

[0162] It should be noted that if Figure 7As shown, the above embodiment of the present application distinguishes the pixel points in the second crack area image by the second segmentation threshold T1, and obtains the first target crack pixel point (the pure black part of the crack in the figure) and the target crack pixel point to be determined (the gray-white part of the crack in the figure); wherein the first target pixel point is determined to be a pixel point in the building crack image. Generally, the building crack image appears as a deepened grayscale, while the edge area image pixel point (the edge area image pixel point here refers to the above-mentioned small part of the edge residue) appears as a grayscale gradient. The embodiment of the present application determines the building image pixel point (that is, the second crack area image pixel point) with a grayscale value greater than or equal to the second segmentation threshold T1 as the first target crack pixel point, and sets it aside for use and processes it in subsequent steps, mainly for distinguishing the target crack pixel point to be determined (that is, the building image pixel point with a grayscale value less than the second segmentation threshold T1) and identifying the pixel point belonging to the building crack image;

[0163] The second segmentation threshold T1 in the embodiment of the present application is different from the first segmentation threshold T0 described above. The difference is that the first segmentation threshold T0 is a probability threshold, which represents the threshold for determining the probability that a pixel point in the building image belongs to a crack; the second segmentation threshold T1 is a gray value threshold, which represents the threshold for determining that a pixel point in the building image belongs to a pixel point in the building crack image;

[0164] Step S23442: obtaining all first target crack pixel points on the outermost periphery of the first target crack pixel point bounding box and determining them as seed points;

[0165] Step S23443: Based on the spatial distribution characteristics of the seed points and shape characteristics Performing iterative optimization growth processing analysis on the target crack pixel point to be determined to obtain a second target crack pixel point;

[0166] Step S23444: obtaining a target crack pixel point based on the first target crack pixel point and the second target crack pixel point;

[0167] It should be noted that the first target crack pixel point and the second target crack pixel point together constitute the target crack pixel point.

[0168] Specifically, if Figure 8 As shown, in step S23443, based on the spatial distribution characteristics of the seed points and shape characteristics Performing iterative optimization growth processing analysis on the target crack pixel point to be determined to obtain a second target crack pixel point includes the following steps:

[0169] Step S234431: traverse each seed point, obtain the target crack pixel point to be determined in the neighborhood pixel points of the seed point and determine it as the pixel point to be grown;

[0170] Step S234432: traverse each of the pixels to be grown, and based on the spatial distribution characteristics of the pixels to be grown Shape characteristics of the growth And the spatial distribution characteristics of the seed points and shape characteristics Calculate the growth similarity measure between the pixel point to be grown and the seed point ;

[0171] The growth similarity metric The calculation method is: ;

[0172] Where, 、 is the weight coefficient, and satisfies α+β=1; and is the norm parameter (controls the similarity calculation, usually can be taken as 1 (Manhattan distance) or 2 (Euclidean distance));

[0173] It should be noted that the spatial distribution features to be grown and the shape features to be grown of the pixel points to be grown in the above-mentioned embodiment of the present application and the spatial distribution features and shape features of the seed points are all extracted from the above-mentioned spatial distribution feature matrix and shape feature matrix according to the corresponding pixel points;

[0174] Step S234433: Determine the growth similarity metric Is it greater than or equal to the preset growth similarity measurement threshold? ; If yes (if no, filter out), then determine that the pixel point to be grown is the second target crack pixel point; judge whether all the pixel points to be grown have been traversed, if yes (if no, return to the above step S234432 to continue traversing until all the pixel points to be grown have been traversed, and the second target crack pixel point corresponding to the current seed point is output), then filter and obtain a new seed point based on the second target crack pixel point corresponding to the seed point, and return the new seed point to the above step S234431 to continue the traversal operation until all the target crack pixel points to be determined have been traversed, and all the second target crack pixel points are output.

[0175] It should be noted that the above-mentioned embodiment of the present application first initializes the growth range of the crack by identifying the seed point and its neighborhood, laying the foundation for subsequent expansion and screening; further, based on the spatial distribution characteristics and shape characteristics of the pixel point to be grown and the seed point, the growth similarity measurement between the pixel point to be grown and the seed point is calculated, and the pixel points similar to the seed point are screened out to ensure that the crack expansion direction is consistent with the initial seed point. By calculating the similarity, it is possible to effectively determine which pixels belong to the same crack area; further, according to the calculated growth similarity measurement, by setting the growth similarity threshold, the pixel points similar to the seed point are effectively screened out to avoid irrelevant or noise pixels being misjudged as part of the crack; by continuously updating the seed point and continuing to traverse, the crack area is gradually expanded to ensure that cracks are identified in different parts of the image. Through the iterative process, all crack pixel points are finally identified to ensure the comprehensiveness of crack detection.

[0176] During the specific implementation of the above-mentioned embodiment of the present application, the technicians also found that after traversing the target crack pixel points to be determined of all seed points, multiple second target crack pixel points are obtained. At this time, the number of second target crack pixel points is very large. If the second target crack pixel points are directly used as new seed points to re-traverse the growth operation, the amount of calculation is very large, which will cause a huge waste of computing power. Therefore, it is necessary to screen the obtained second target crack pixel points to obtain new seed points, thereby reducing the number of iterations, realizing high-speed calculation of the building model and reducing the waste of computing power. For details, please refer to the subsequent steps and Figure 8 .

[0177] Specifically, if Figure 9 As shown, in step S234433, a new seed point is obtained by screening based on the second target crack pixel point corresponding to the seed point, including the following steps:

[0178] Step S2344331: obtaining a third crack region image based on the first target crack pixel point and the current second target crack pixel point; dividing the third crack region image into a plurality of grid regions of equal size; obtaining the second target crack pixel point in each grid region as a first candidate seed point;

[0179] It should be noted that, in the above embodiment of the present application, the second target crack pixel point refers to the pixel point in the above step S234433 that is iteratively calculated through the initial seed point to measure the growth similarity. The second target crack pixel point is obtained by the subsequent judgment; the third crack area image formed based on the first target crack pixel point and the current second target crack pixel point is different from the second crack area image in that all pixels in the third crack area image belong to the pixel points of the building crack image, and the third crack area image is different from the building crack image in that the third crack area image is not as complete as the pixel points of the building crack image, and some of the second target crack pixel points still exist in the second crack area image;

[0180] The above-mentioned embodiment of the present application divides the third crack area image into grid areas of equal size, and obtains the first candidate seed points in each grid area, aiming to divide the currently obtained second target crack pixel points according to their spatial distribution, thereby obtaining the first candidate seed points of each grid area. In subsequent operations, new seed points are screened based on the first candidate seed points to avoid excessive concentration of the selected seed points, and are evenly distributed among all the currently obtained second target crack pixel points.

[0181] Step S2344332: selecting the first candidate pixel point with the highest growth similarity metric in each of the grid areas as the second candidate seed point;

[0182] It should be noted that the above-mentioned embodiment of the present application combines the growth similarity measurement of the first candidate pixel point in each grid area with the non-maximum suppression algorithm to ensure that only one most significant seed point (i.e., the above-mentioned second candidate seed point) is retained in each grid area.

[0183] Step S2344333: Calculate the local density of each of the second candidate seed points based on the preset window radius w ;

[0184] The local density of the second candidate seed point The calculation method is: ;

[0185] Where, is the window radius; is the vertical offset (value ranges from -w to w); l is the horizontal offset (value ranges from -w to w); Step S2344334: Select the local density of the second candidate seed point Greater than or equal to the preset local density threshold of the second candidate seed point The second candidate seed point is used as the new seed point;

[0186] It should be noted that the above embodiment of the present application further optimizes the selection of new seed points through density analysis to ensure uniform distribution of new seed points in space and improve the coverage of crack identification;

[0187] The above-mentioned embodiment of the present application first combines the first target crack pixel point and the current second target crack pixel point to obtain a third crack region image containing more crack information. Further, the third crack region image is divided into multiple grid regions of equal size, thereby optimizing the selection of seed points according to the spatial distribution. Then, the second target crack pixel point is selected from each grid region as the first candidate seed point, and the second target crack pixel point is distributed to different regions according to the spatial distribution, effectively avoiding excessive concentration of seed points and ensuring a more uniform distribution of cracks, thereby improving the coverage and accuracy of crack detection.

[0188] Furthermore, in each grid area, the first candidate pixel point with the highest growth similarity metric is selected as the second candidate seed point to ensure that only one most significant seed point is retained in each grid area, thereby avoiding repeated calculation of multiple similar seed points for the same crack area. The non-maximum suppression algorithm can effectively screen out the most significant seed points, reduce redundant calculations, and thus improve the efficiency of the algorithm; further, through density calculation, seed points with relatively uniform and dense distribution are screened out to ensure the uniform distribution of new seed points in space, avoiding excessive concentration of seed points in certain areas, affecting the integrity and accuracy of crack detection; finally, the second candidate seed point with a local density greater than or equal to the preset threshold is selected as the new seed point to ensure that the selected seed points are well representative in space and reasonably distributed; the seed points screened out by density analysis have a higher spatial distribution uniformity, avoiding excessive concentration of seed points, thereby improving the global coverage of crack identification, and through the limitation of the local density threshold, seed points with reasonable distribution and representativeness can be further screened out to ensure the stability and robustness of the crack detection algorithm.

[0189] In summary, the present invention proposes a construction quality and safety assessment system and method. By performing edge recognition on construction images, processing and analyzing the images, and identifying building crack images and ground subsidence images, the system and method can effectively distinguish crack and subsidence areas, and efficiently identify problematic areas, providing clear data support for subsequent quality assessment. Finally, the identified building crack images and ground subsidence images are input into a pre-trained construction quality and safety assessment model for processing and analysis, and finally output a safety assessment score for the construction site. By converting the crack and subsidence images into a score form, the construction quality and safety assessment model can quantify the safety status of the construction site, specify the severity of each crack and subsidence, form a scoring system, simplify the complex construction quality assessment process, and accurately reflect the potential risks and hidden dangers of the construction site based on the assessment scores, helping technical personnel to promptly discover safety problems in construction and thus carry out targeted remediation.

[0190] Specifically, when identifying ground subsidence images, the camera parameters are combined to generate a depth image, thereby accurately identifying and analyzing the location and subsidence of the ground;

[0191] Furthermore, in the process of recognizing the building crack image, a background area image in the first building image is firstly screened out through a coarse segmentation processing operation, and then a fine segmentation processing operation is performed on the image of the area to be processed obtained after the background area image is screened out, that is, the crack area image and the edge of the crack area image (edge ​​area image) are finely segmented, and the edge is screened out to obtain a complete crack area image, that is, a building crack image;

[0192] Furthermore, during the fine segmentation operation, the spatial distribution features and shape features are fused to obtain a comprehensive feature matrix C. The scoring matrix R is calculated based on the comprehensive feature matrix C. Then, based on the scoring matrix R, the spatial distribution feature matrix S and the shape feature matrix M, combined with the preset initial segmentation threshold T0 and the preset second segmentation threshold T1, an analysis is performed to obtain the building crack image.

[0193] In the further operation process, first, by identifying the seed point and its neighborhood, the growth range of the crack is initialized, laying the foundation for subsequent expansion and screening; further, based on the spatial distribution characteristics and shape characteristics of the pixel point to be grown and the seed point, the growth similarity measurement between the pixel point to be grown and the seed point is calculated, and the pixel points similar to the seed point are screened out to ensure that the crack expansion direction is consistent with the initial seed point. By calculating the similarity, it is possible to effectively determine which pixels belong to the same crack area; further, based on the calculated growth similarity measurement, by setting the growth similarity threshold, the pixel points similar to the seed point are effectively screened out to avoid irrelevant or noise pixels being misjudged as part of the crack; by continuously updating the seed point and continuing to traverse, the crack area is gradually expanded to ensure that cracks are identified in different parts of the image. Through the iterative process, all crack pixels are finally identified to ensure the comprehensiveness of crack detection;

[0194] In the further iterative process, the local density threshold is limited to further screen out seed points with reasonable distribution and representativeness to ensure the stability and robustness of the crack detection algorithm.

[0195] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. A person skilled in the art may modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A construction quality safety assessment system, characterized in that: It includes image acquisition module, recognition module and evaluation module; The image acquisition module is used to acquire multiple construction images of the current construction site through multi-angle acquisition using a high-definition camera; The building construction image includes a building image and a building ground image; The recognition module is used to analyze and process the building construction image through edge recognition to obtain a building crack image and a ground subsidence image of the current construction site; The assessment module is configured to process and analyze the building crack image and the ground subsidence image using a pre-trained construction quality safety assessment model, and output a safety assessment score for the current construction site; The recognition module includes a pre-processing module, a depth image generation and alignment module, a crack recognition module and a settlement recognition module; The preprocessing module is configured to obtain a first building construction image through a preprocessing operation based on the building construction image; The first building construction image includes a first building image and a first building ground image; The depth image generation and alignment module is used to obtain the camera internal parameters of the high-definition camera; generate the building construction image based on the camera internal parameters to obtain a depth building construction image; align the depth building construction image with the first building ground image to obtain a construction image group; The crack recognition module is configured to obtain a building crack image through edge detection and recognition analysis based on the first building image; The settlement recognition module is configured to obtain a ground settlement image through edge detection and recognition analysis based on the construction image group; The crack identification module includes a first extraction module, a clustering module, a second extraction module and an iterative analysis module; The first extraction module is configured to extract multi-dimensional feature data of each building image pixel point in the first building image; The multi-dimensional feature data includes grayscale value feature data, gradient value feature data and gradient direction feature data; The clustering module is configured to cluster the building image pixels based on the multi-dimensional feature data to obtain a plurality of clusters; determine the image area of ​​the cluster based on the grayscale value mean of each building image pixel in the cluster; obtain an image area to be processed based on the image area of ​​the cluster; and construct an image area label matrix L based on the cluster labels of the image area to be processed; The second extraction module is used to extract the position feature data of each building image pixel in the image area label matrix L; construct a spatial distribution feature matrix S based on the position feature data; extract the shape features of the spatial distribution feature matrix S to construct a shape feature matrix M; The iterative analysis module is used to perform iterative analysis on the image area to be processed based on the spatial distribution feature matrix S and the shape feature matrix M in combination with probability calculation and morphological processing to identify and obtain a building crack image.

2. A construction quality safety assessment system according to claim 1, characterized in that: The multi-dimensional feature data includes grayscale value feature data, gradient value feature data and gradient direction feature data; The image area includes a background area image, a crack area image and an edge area image; The image area to be processed includes a crack area image and an edge area image; Each element in the image region label matrix L represents the image region to which each building image pixel point in the first building image belongs; The number of clusters is 3.

3. A construction quality safety assessment system according to claim 2, characterized in that: The iterative analysis module includes a first construction submodule, a second construction submodule, a first segmentation submodule, an iterative analysis submodule and a third construction submodule; The first construction submodule is used to construct a comprehensive feature matrix C based on the spatial distribution feature matrix S and the shape feature matrix M; The second construction submodule is used to calculate the probability r of each pixel point of the building image belonging to the crack area image based on the comprehensive feature matrix C; and construct a scoring matrix R based on the probability r; The first segmentation submodule is configured to mark, based on a preset initial segmentation threshold T0, pixel points of the building image corresponding to a probability r greater than or equal to the initial segmentation threshold T0 in the scoring matrix R as crack regions, thereby obtaining a first crack region image; and perform a morphological processing operation on the first crack region image to output a second crack region image; The iterative analysis submodule is used to perform iterative analysis and screening based on the second crack region image in combination with the spatial distribution feature matrix S and the shape feature matrix M to obtain target crack pixel points; The third construction submodule is used to construct a building crack image based on all the target crack pixel points.

4. A construction quality safety assessment system according to claim 3, characterized in that: The iterative analysis submodule includes a second segmentation submodule, a seed point acquisition module, an iterative optimization growth module and a fourth construction submodule; The second segmentation submodule is configured to segment the second crack region image based on a preset second segmentation threshold T1 to obtain a first target crack pixel point and a target crack pixel point to be determined; The seed point acquisition module is used to acquire all first target crack pixel points at the outermost periphery of the first target crack pixel point bounding box and determine them as seed points; The iterative optimization growth module is used to optimize the spatial distribution characteristics of the seed points. and shape characteristics Performing iterative optimization growth processing analysis on the target crack pixel point to be determined to obtain a second target crack pixel point; The fourth construction submodule is configured to obtain a target crack pixel point based on the first target crack pixel point and the second target crack pixel point.

5. A construction quality safety assessment system according to claim 4, characterized in that: The iterative optimization growth module is further configured to traverse each of the seed points, obtain target crack pixel points to be determined from the neighboring pixel points of the seed point, and determine them as pixel points to be grown; Traverse each of the pixels to be grown, based on the spatial distribution characteristics of the pixels to be grown Shape characteristics of the growth And the spatial distribution characteristics of the seed points and shape characteristics Calculate the growth similarity measure between the pixel point to be grown and the seed point ; Based on the growth similarity metric Perform judgment iteration and output to obtain a second target crack pixel point among the target crack pixel points to be determined.

6. A construction quality safety assessment system according to claim 5, characterized in that: The iterative optimization growth module is specifically used to determine the growth similarity metric Is it greater than or equal to a preset growth similarity measurement threshold K'; if so, determine that the pixel point to be grown is the second target crack pixel point; determine whether all the pixel points to be grown have been traversed, and if so, obtain a new seed point based on the second target crack pixel point corresponding to the seed point, and return the new seed point to the above operation to continue the traversal operation until all the target crack pixel points to be determined have been traversed, and all the second target crack pixel points are output.

7. A construction quality safety assessment system according to claim 6, characterized in that: During a specific implementation, the iterative optimization growth module is further configured to obtain a third crack region image based on the first target crack pixel point and the current second target crack pixel point; divide the third crack region image into a plurality of grid regions of equal size; and obtain the second target crack pixel point in each grid region as a first candidate seed point. Selecting the first candidate pixel point with the highest growth similarity measure in each of the grid areas as the second candidate seed point; Calculate the local density of each of the second candidate seed points based on the preset window radius w ; Select the local density of the second candidate seed point Greater than or equal to the preset local density threshold of the second candidate seed point The second candidate seed point is used as the new seed point.

Images