Crack identification and building structure safety analysis method
Through the improved YOLOV5 algorithm and BIM technology, combined with the building structure images collected by the drone, a three-dimensional database was established, which solved the problems of low efficiency and poor accuracy of traditional crack identification methods, and achieved high-precision crack identification and safety evaluation of building structures , providing important data support.
Patent Information
- Application Number
- CN202410946791.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-16
- Publication Date
- 2024-11-08
AI Technical Summary
Traditional crack identification methods are inefficient and susceptible to human factors, and cannot accurately reflect the safety of the building structure. Simple crack identification is not enough to conduct safety assessment of the building structure.
The improved YOLOV5 algorithm is combined with BIM technology, and the building structure images are collected through drones and a three-dimensional database is established to achieve high-precision identification of cracks and safety analysis of building structures, and comprehensive evaluation is performed using the BIM model.
It improves the accuracy and efficiency of crack identification, provides strong data support for the maintenance and repair of building structures, and realizes accurate assessment of building structure safety.
Smart Images

Figure CN118918498A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of building engineering, and in particular to a method for crack identification and building structure safety analysis. Background Art
[0002] With the rapid development of the construction industry, the safety of building structures has attracted more and more attention. As a common form of structural damage, accurate identification and evaluation of cracks are of great significance to ensure the stability and safety of building structures. Traditional crack identification methods mainly rely on manual visual inspection, which is not only inefficient but also easily affected by human factors, resulting in inaccurate identification results.
[0003] In recent years, deep learning technology has made significant progress in the field of image processing and provided a new solution for crack identification. The deep learning model can automatically extract the characteristics of cracks by learning the characteristic information in a large amount of image data, and realize accurate crack identification. However, simple crack identification cannot directly reflect the safety of building structures, and comprehensive evaluation is also required in combination with detailed information of the building structure.
[0004] As a digital building design and management method, Building Information Modeling (BIM) technology can integrate detailed information such as geometry, materials, and construction of building structures, providing strong data support for the analysis and evaluation of building structures. By combining BIM technology with deep learning, seamless connection between crack identification and building structure information can be achieved, providing the possibility for accurate analysis of building structure safety.
[0005] Therefore, the present invention proposes a crack identification and building structure safety analysis method based on the improved YOLOV5 algorithm and BIM technology, aiming to use the improved YOLOV5 model to perform high-precision identification of cracks in building structure images, and to build a digital model of the building structure in combination with BIM technology, and to achieve accurate assessment of the safety of the building structure through comparative analysis. This method can not only improve the accuracy and efficiency of crack identification, but also provide strong data support for the maintenance and repair of building structures, and has important practical application value.
[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention
[0007] The purpose of the present invention is to provide a method for crack identification and building structure safety analysis to solve the problems in the above-mentioned background technology.
[0008] In order to achieve the above object, the present invention provides the following technical solution: a method for crack identification and building structure safety analysis, comprising the following steps:
[0009] S1, using the labeled crack dataset to train the crack detection network model;
[0010] S2, using the labeled crack dataset to train the crack segmentation network model;
[0011] S3, using a camera and a laser rangefinder mounted on a drone to collect the image of the building structure to be detected and the distance between the image and the building;
[0012] S4, dividing the obtained high-resolution image into a plurality of image blocks to be processed and inputting them into the crack detection network model trained in step S1 to obtain a high-resolution image of the structure to be detected with crack information;
[0013] S5, obtaining a crack segmentation map of the image block to be processed using the crack segmentation network model trained in step S2;
[0014] S6, automatically splicing the crack segmentation maps of each block to be processed to obtain the crack segmentation map of the structure to be detected;
[0015] S7, determining the actual physical size of the crack according to the crack segmentation map;
[0016] S8. Establish a BIM model of the building structure according to the building design information, and form a three-dimensional database of the BIM model;
[0017] S9. Import the crack information into the three-dimensional database, and determine whether the building structure is safe based on the information in the three-dimensional database.
[0018] Preferably, the crack detection network model is an improved YOLOV5 model, specifically comprising:
[0019] (1) Input layer, which includes transferring each batch of training data to the neural network through the data loader, normalizing the data size, and then using data augmentation to expand the data set;
[0020] (2) The backbone network Backbone part includes the first CBS module, the first ADM module, the first C3DCN module, the second ADM module, the second C3DCN module, the third ADM module, the third C3DCN module, the fourth ADM module, the fourth C3DCN module, and the first SPPFCSPC module, wherein the input of each module comes from the output of the previous module;
[0021] (3) The Neck part of the neck network includes the first CBS module, the second CBS module, the first Concat module, the first CSPStage module, the first nn.Upsample module, the third CBS module, the second Concat module, the second CSPStage module, the second nn.Upsample module, the third Concat module, the third CSPStage module, the fourth CBS module, the fourth Concat module, the fourth CSPStage module, the fifth CBS module, the sixth CBS module, the fifth Concat module, and the fifth CSPStage module, wherein the input of the second CBS module comes from the output of the third C3DCN module of the Backbone network, the input of the first Concat module comes from the output of the first and second CBS modules of the Neck network, and the input of the third CBS module comes from The output of the second C3DCN module of the Backbone network, the input of the second Concat module comes from the output of the third C3DCN of the Backbone network, the third CBS module of the Neck network, and the first nn.Upsample module, the input of the third Concat module comes from the output of the second C3DCN of the Backbone network and the second nn.Upsample module of the Neck network, the input of the fourth Concat module comes from the output of the fourth CBS module of the Neck network and the second CSPStage module, the input of the fifth CBS module comes from the output of the second CSPStage module of the Neck network, the input of the sixth CBS module comes from the output of the fourth CSPStage module of the Neck network, and the input of the fifth Concat module comes from the output of the first CSPStage module of the Neck network, the fifth and sixth CBS modules;
[0022] (4) Final detection of the head part: This includes analyzing the shape and features of pavement cracks, clustering the training set using a clustering algorithm to obtain nine initial anchor boxes, and applying anchor boxes to the feature map to grid-predict the feature map using the set NMS confidence parameters. The detected targets with confidence levels higher than the set threshold are retained.
[0023] Preferably, the ADM module is an improved part, and the specific operation is as follows:
[0024] The input feature map is first downsampled by the SPD layer, and then the CA attention layer is used to determine the importance of different channels and spatial positions to emphasize important information. Finally, a 1x1 CBS layer is used to adjust the number of channels using learnable parameters.
[0025] Preferably, the C3DCN module is an improved part, and the specific operations are as follows:
[0026] The features are divided into two parts through a 1x1 CBS layer. The first part does not undergo additional processing. The second part processes the feature map by introducing a deformable convolutional Bottleneck layer. Finally, the channel correlation between the two parts is decoupled through a concatenation operation and a 1x1 CBS layer.
[0027] Preferably, the SPPFCSPC module is an improved part, and the specific operation is as follows: the input feature map is divided into two parts through a 1x1 CBS layer, the first part is directly transmitted to the output end without any additional operation, and the second part first extracts the detail information of the basic feature map through a 3×3 CBS layer, and then divides the extracted detail information into four parts through a 1×1 CBS layer, and uses three 5×5 Maxpool2d layers to process features of different scales, and then uses a 1×1 CBS layer and a 3×3 CBS layer to highlight the importance of foreground features and improve insensitivity to small space changes. Finally, it is spliced with the first part without additional operation, and the channel correlation of the two parts is decoupled through a 1x1 CBS layer to obtain the final output.
[0028] Preferably, the Neck network is an improved part, and its core module CSPStage specifically operates as follows: the input feature map is spliced, and then the input feature map is divided into two parts through a 1x1 CBS layer. The first part does not undergo additional processing, and the second part processes the feature map through several residual modules consisting of a 3×3 RepConv layer and a 3×3 CBS layer, and then splices it with the first part without additional operation, and decouples the channel correlation of the two parts through a 1x1 CBS layer to obtain the final output.
[0029] Preferably, the crack segmentation network model is a UNet network.
[0030] Preferably, the UAV flies according to a preset moving path, the camera collects high-resolution images of the surface of the structure to be detected during the flight of the UAV, and the laser rangefinder records the actual physical distance between the camera and the structure to be detected.
[0031] Preferably, the actual physical size of the crack is determined according to the crack segmentation map, and the specific steps include:
[0032] (1) extracting the center line and contour of the crack in the crack segmentation image to obtain the center line and contour of the crack;
[0033] (2) obtaining a crack length characteristic value of the crack in pixels according to a Euclidean distance calculation method;
[0034] (3) obtaining a crack width characteristic value of the crack in pixels according to the distance between the perpendicular line of the center line of the test point and the edge line;
[0035] (4) The actual physical size of the crack is calculated based on the crack characteristic value, the distance information between the image and the surface of the structure to be detected, and the camera parameters.
[0036] Preferably, the conclusion of whether the building structure is safe is obtained by importing the three-dimensional crack database into structural safety analysis software, inputting load information into the structural safety analysis software, and analyzing the software.
[0037] In the above technical solution, the technical effects and advantages provided by the present invention are:
[0038] The present invention improves the crack detection network. Compared with the existing method, the method provided by the embodiment of the present invention has higher crack detection accuracy.
[0039] The present invention uses a drone to replace manual inspection, which saves a lot of manpower and material resources while increasing the speed of inspection.
[0040] The present invention performs three-dimensional digital modeling based on BIM technology, effectively improving the digital level of crack detection, facilitating the subsequent use of technical software to accurately evaluate the safety of the building structure, and facilitating the subsequent monitoring and maintenance of the building structure to be detected. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.
[0042] Figure 1 A schematic diagram of a method flow of a crack identification and building structure safety analysis method provided by an embodiment of the present invention;
[0043] Figure 2 A crack detection network model diagram provided by an embodiment of the present invention;
[0044] Figure 3 ADM module diagram in the crack detection network model provided by an embodiment of the present invention;
[0045] Figure 4 A C3DCN module diagram in a crack detection network model provided by an embodiment of the present invention;
[0046] Figure 5 This is a diagram of the SPPFCSPC module in the crack detection network model provided in an embodiment of the present invention.
[0047] Figure 6 A diagram of the CSPStage module in the crack detection network model provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0048] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.
[0049] The present invention provides Figures 1 to 6 A crack identification and building structure safety analysis method shown includes the following steps:
[0050] S1, using the labeled crack dataset to train the crack detection network model;
[0051] S2, using the labeled crack dataset to train the crack segmentation network model;
[0052] S3, using a camera and a laser rangefinder mounted on a drone to collect the image of the building structure to be detected and the distance between the image and the building;
[0053] S4, dividing the obtained high-resolution image into a plurality of image blocks to be processed and inputting them into the crack detection network model trained in step S1 to obtain a high-resolution image of the structure to be detected with crack information;
[0054] S5, obtaining a crack segmentation map of the image block to be processed using the crack segmentation network model trained in step S2;
[0055] S6, automatically splicing the crack segmentation maps of each block to be processed to obtain the crack segmentation map of the structure to be detected;
[0056] S7, determining the actual physical size of the crack according to the crack segmentation map;
[0057] S8. Establish a BIM model of the building structure according to the building design information, and form a three-dimensional database of the BIM model;
[0058] S9, importing the crack information into the three-dimensional database, and judging whether the building structure is safe according to the information in the three-dimensional database;
[0059] Among them, the specific steps of S1 include:
[0060] S1-1: Use drone cameras to take pictures of cracks on building structures to obtain high-resolution pictures;
[0061] S1-2: Divide the image into several image blocks of the same size to be processed;
[0062] S1-3: Use Labelimg software to manually annotate the image blocks to be processed, mark the locations of the cracks with rectangular boxes, and generate XML format files;
[0063] S1-4: Divide the training set and validation set proportionally to train the improved YOLOV5 crack detection network.
[0064] The improved YOLOV5 network model is as follows Figure 2 As shown, including:
[0065] (1) Input layer: This includes transferring each batch of training data to the neural network through the data loader, normalizing the data size, and augmenting the data set using data enhancement methods such as mosaic, random inversion, and color conversion.
[0066] (2) The backbone network includes the first CBS module, the first ADM module, the first C3DCN module, the second ADM module, the second C3DCN module, the third ADM module, the third C3DCN module, the fourth ADM module, the fourth C3DCN module, and the first SPPFCSPC module. The input of each module comes from the output of the previous module.
[0067] (3) The Neck part of the neck network includes the first CBS module, the second CBS module, the first Concat module, the first CSPStage module, the first nn.Upsample module, the third CBS module, the second Concat module, the second CSPStage module, the second nn.Upsample module, the third Concat module, the third CSPStage module, the fourth CBS module, the fourth Concat module, the fourth CSPStage module, the fifth CBS module, the sixth CBS module, the fifth Concat module, and the fifth CSPStage module. The input of the second CBS module comes from the output of the third C3DCN module of the Backbone network. The input of the first Concat module comes from the output of the first and second CBS modules of the Neck network. The input of the third CBS module comes from the output of the second C3DCN module of the Backbone network. The input of the second Concat module comes from the output of the third C3DCN of the Backbone network, the third CBS module of the Neck network, and the output of the first nn.Upsample module. The input of the third Concat module comes from the output of the second C3DCN of the Backbone network and the second nn.Upsample module of the Neck network. The input of the fourth Concat module comes from the output of the fourth CBS module and the second CSPStage module of the Neck network. The input of the fifth CBS module comes from the output of the second CSPStage module of the Neck network. The input of the sixth CBS module comes from the output of the fourth CSPStage module of the Neck network. The input of the fifth Concat module comes from the output of the first CSPStage module, the fifth and sixth CBS modules of the Neck network. The remaining inputs that are not explained are all from the output of the previous module.
[0068] (4) Final detection of the head part: First, by analyzing the shape and characteristics of the pavement cracks, the training set is clustered using a clustering algorithm to obtain 9 anchor boxes. Then, the NMS confidence parameter is set, and the anchor box is applied to the feature map to grid the feature map prediction, and the detected targets with confidence higher than the set threshold are retained.
[0069] Furthermore, the ADM module is an improved module, such as Figure 3 As shown in the figure, the input feature map is first downsampled by the SPD layer, and then the CA attention layer determines the importance of different channels and spatial positions to emphasize important information. Finally, a 1x1 CBS layer is used to adjust the number of channels using learnable parameters.
[0070] Furthermore, the C3DCN module is an improved module, such as Figure 4 As shown in the figure, a 1x1 CBS layer is used to divide the features into two parts. The first part does not undergo additional processing, and the second part processes the feature map by introducing a deformable convolutional Bottleneck layer. Finally, the channel correlation between the two parts is decoupled through a concatenation operation and a 1x1 CBS layer.
[0071] Furthermore, the SPPFCSPC module is an improved part, such as Figure 5 As shown in the figure, the input feature map is divided into two parts through a 1x1 CBS layer. The first part is directly transmitted to the output without any additional operation. The second part first extracts the detail information of the basic feature map through a 3x3 CBS layer. Then, the extracted detail information is divided into four parts through a 1x1 CBS layer, and three 5x5 Maxpool2d layers are used to process features of different scales. Then, a 1x1 CBS layer and a 3x3 CBS layer are used to highlight the importance of the area close to the center, and improve the insensitivity to small spatial changes. Finally, it is spliced with the first part without any additional operation, and the channel correlation of the two parts is decoupled through a 1x1 CBS layer to obtain the final output.
[0072] Furthermore, Neck network is the improved part, and its core module CSPStage, such as Figure 6 As shown in the figure, the input feature map is concatenated, and then divided into two parts through a 1x1 CBS layer. The first part does not undergo additional processing, and the second part processes the feature map through several residual modules consisting of a 3x3 RepConv layer and a 3x3 CBS layer, and then concatenates it with the first part without additional operation, and uses a 1x1 CBS layer to decouple the channel correlation of the two parts to obtain the final output.
[0073] In this embodiment, the specific steps of S2 include:
[0074] S2-1: Use Labelme software to annotate the image blocks to be processed obtained in step S1, set the pixel label of the crack area to 255, set the background pixel to 0, and construct a segmentation data set.
[0075] S2-2: Divide the dataset into training set and validation set and train the UNet segmentation network.
[0076] Specifically, in this step, the UNet segmentation network is constructed by an encoder constructed by double convolution and a decoder constructed by deconvolution.
[0077] In an embodiment, step S3 uses a camera and a laser rangefinder mounted on a drone to collect the image of the building structure and the distance between the image and the building, including the drone flying according to a preset moving path, and the camera collecting a high-resolution image of the surface of the structure to be detected during the flight of the drone. The laser rangefinder records the actual physical distance between the structure to be detected captured by the camera and the captured image.
[0078] In this embodiment, determining the actual physical size of the crack according to the crack segmentation map includes:
[0079] (1) extracting the center line and contour of the crack in the crack segmentation image to obtain the center line and contour of the crack;
[0080] (2) obtaining a crack length characteristic value of the crack in pixels according to a Euclidean distance calculation method;
[0081] (3) obtaining a crack width characteristic value of the crack in pixels according to the distance between the perpendicular line of the center line of the test point and the edge line;
[0082] (4) The actual physical size of the crack is calculated based on the crack characteristic value, the distance information between the image and the surface of the structure or component to be detected, and the camera parameters.
[0083] Furthermore, the center line and contour extraction of the crack segmentation map is performed by converting the crack segmentation map into a binary matrix and completing it through the "findcontours" function in opencv.
[0084] Furthermore, the crack length characteristic value of the crack in pixels is obtained according to the Euclidean distance calculation method. The specific calculation formula is: In the formula, x k+1 ,y k+1 and x k ,y k are the pixel coordinates of the k+1th and kth pixel points on the center line respectively. The total length of the crack L d For: L d =∑d k .
[0085] Furthermore, the width of the crack is determined based on the center line of the crack by drawing a perpendicular line to the center line of the crack to be measured, intersecting the contour at two points, and calculating the distance between the two points.
[0086] Preferably, a three-dimensional database containing crack information is established using BIM technology based on the actual physical information of the cracks and design information of the building structure.
[0087] Preferably, the information of the three-dimensional database is imported into the structural safety analysis software, the load information is input into the structural safety analysis software, and a judgment conclusion of whether the structure to be inspected is safe is obtained from the output of the structural safety analysis software.
[0088] Preferably, the staff performs maintenance and renovation on the building structure based on the judgment conclusion of whether the building structure is safe and the three-dimensional visualization result.
[0089] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.
Claims
1. A method for crack identification and building structure safety analysis, characterized in that: The following steps are involved: S1, using the labeled crack dataset to train the crack detection network model; S2, using the labeled crack dataset to train the crack segmentation network model; S3, using a camera and a laser rangefinder mounted on a drone to collect the image of the building structure to be detected and the distance between the image and the building; S4, dividing the obtained high-resolution image into a plurality of image blocks to be processed and inputting them into the crack detection network model trained in step S1 to obtain a high-resolution image of the structure to be detected with crack information; S5, obtaining a crack segmentation map of the image block to be processed using the crack segmentation network model trained in step S2; S6, automatically splicing the crack segmentation maps of each block to be processed to obtain the crack segmentation map of the structure to be detected; S7, determining the actual physical size of the crack according to the crack segmentation map; S8. Establish a BIM model of the building structure according to the building design information, and form a three-dimensional database of the BIM model; S9. Import the crack information into the three-dimensional database, and determine whether the building structure is safe based on the information in the three-dimensional database.
2. A crack identification and building structure safety analysis method according to claim 1, characterized in that: The crack detection network model is an improved YOLOV5 model, which specifically includes: (1) Input layer, which includes transferring each batch of training data to the neural network through the data loader, normalizing the data size, and then using data augmentation to amplify the data set; (2) The backbone network Backbone part includes the first CBS module, the first ADM module, the first C3DCN module, the second ADM module, the second C3DCN module, the third ADM module, the third C3DCN module, the fourth ADM module, the fourth C3DCN module, and the first SPPFCSPC module, wherein the input of each module comes from the output of the previous module; (3) The Neck part of the neck network includes the first CBS module, the second CBS module, the first Concat module, the first CSPStage module, the first nn.Upsample module, the third CBS module, the second Concat module, the second CSPStage module, the second nn.Upsample module, the third Concat module, the third CSPStage module, the fourth CBS module, the fourth Concat module, the fourth CSPStage module, the fifth CBS module, the sixth CBS module, the fifth Concat module, and the fifth CSPStage module, wherein the input of the second CBS module comes from the output of the third C3DCN module of the Backbone network, the input of the first Concat module comes from the output of the first and second CBS modules of the Neck network, and the input of the third CBS module comes from The output of the second C3DCN module of the Backbone network, the input of the second Concat module comes from the output of the third C3DCN of the Backbone network, the third CBS module of the Neck network, and the first nn.Upsample module, the input of the third Concat module comes from the output of the second C3DCN of the Backbone network and the second nn.Upsample module of the Neck network, the input of the fourth Concat module comes from the output of the fourth CBS module of the Neck network and the second CSPStage module, the input of the fifth CBS module comes from the output of the second CSPStage module of the Neck network, the input of the sixth CBS module comes from the output of the fourth CSPStage module of the Neck network, and the input of the fifth Concat module comes from the output of the first CSPStage module of the Neck network, the fifth and sixth CBS modules; (4) Final detection of the head part: This includes analyzing the shape and features of pavement cracks, clustering the training set using a clustering algorithm to obtain nine initial anchor boxes, and applying anchor boxes to the feature map to grid-predict the feature map using the set NMS confidence parameters. The detected targets with confidence levels higher than the set threshold are retained.
3. A crack identification and building structure safety analysis method according to claim 2, characterized in that: The ADM module is an improved part, and the specific operation is as follows: The input feature map is first downsampled by the SPD layer, and then the CA attention layer is used to determine the importance of different channels and spatial positions to emphasize important information. Finally, a 1x1 CBS layer is used to adjust the number of channels using learnable parameters.
4. A crack identification and building structure safety analysis method according to claim 2, characterized in that: The C3DCN module is an improved part, and the specific operation is as follows: The features are divided into two parts through a 1x1 CBS layer. The first part does not undergo additional processing. The second part processes the feature map by introducing a deformable convolutional Bottleneck layer. Finally, the channel correlation between the two parts is decoupled through a concatenation operation and a 1x1 CBS layer.
5. A crack identification and building structure safety analysis method according to claim 2, characterized in that: The SPPFCSPC module is an improved part, and its specific operation is as follows: the input feature map is divided into two parts through a 1x1 CBS layer. The first part is directly transmitted to the output end without any additional operation. The second part first extracts the detail information of the basic feature map through a 3×3 CBS layer, and then divides the extracted detail information into four parts through a 1×1 CBS layer, and uses three 5×5 Maxpool2d layers to process features of different scales. Then, a 1×1 CBS layer and a 3×3 CBS layer are used to highlight the importance of foreground features and improve insensitivity to small spatial changes. Finally, it is spliced with the first part without additional operation, and the channel correlation of the two parts is decoupled through a 1x1 CBS layer to obtain the final output.
6. A crack identification and building structure safety analysis method according to claim 2, characterized in that: The Neck network is an improved part, and its core module CSPStage operates as follows: the input feature map is spliced, and then the input feature map is divided into two parts through a 1x1 CBS layer. The first part does not undergo additional processing, and the second part processes the feature map through several residual modules consisting of a 3×3 RepConv layer and a 3×3 CBS layer, and then splices it with the first part without additional operation, and decouples the channel correlation of the two parts through a 1x1 CBS layer to obtain the final output.
7. A crack identification and building structure safety analysis method according to claim 1, characterized in that: The crack segmentation network model is a UNet network.
8. A crack identification and building structure safety analysis method according to claim 1, characterized in that: The UAV flies according to a preset moving path, the camera collects high-resolution images of the surface of the structure to be detected during the flight of the UAV, and the laser rangefinder records the actual physical distance between the camera and the structure to be detected.
9. A crack identification and building structure safety analysis method according to claim 1, characterized in that: The actual physical size of the crack is determined according to the crack segmentation map, and the specific steps include: (1) extracting the center line and contour of the crack in the crack segmentation image to obtain the center line and contour of the crack; (2) obtaining a crack length characteristic value of the crack in pixels according to a Euclidean distance calculation method; (3) obtaining a crack width characteristic value of the crack in pixels according to the distance between the perpendicular line of the center line of the test point and the edge line; (4) The actual physical size of the crack is calculated based on the crack characteristic value, the distance information between the image and the surface of the structure to be detected, and the camera parameters.
10. A crack identification and building structure safety analysis method according to claim 1, characterized in that: The conclusion of whether the building structure is safe is obtained by importing the three-dimensional crack database into the structural safety analysis software, inputting the load information into the structural safety analysis software, and analyzing by the structural safety analysis software.
Citation Information
Patent Citations
Crack identification method and building structure safety monitoring method
CN111651816A
Structural apparent crack detection method, device and system
CN114049356A
Lightweight large-scale building high-altitude steel structure weld defect identification model and weld quality detection device and method
CN116342531A
Unmanned aerial vehicle visual angle image small object target real-time identification method based on improved YOLOv5
CN118115899A
Cited By
Unmanned aerial vehicle image crack detection method, system and device and storage medium
CN120220000A
Building thermal insulation outer wall defect identification method and system based on image identification
CN122313291A
