A building inspection method and system based on existing building structure

Abstract

The present invention discloses a building inspection method and system based on an existing building structure. The point cloud model obtained by using a laser radar has high-precision and high-density three-dimensional spatial information, and can accurately reflect the geometric shape and detail features of the building structure. The image data obtained by the image acquisition device helps to understand the building status more intuitively. The point cloud model and the image data are aligned and fused, so that the inspection package contains rich geometric information and visual information at the same time. By integrating the data acquisition, alignment fusion and evaluation processes into one inspection package, the automation and standardization of the inspection process are realized. Compared with the traditional manual inspection method, the inspection method based on the laser radar and image acquisition technology can complete the data acquisition and processing work more quickly, and significantly improve the inspection efficiency. At the same time, since the automated inspection reduces the labor cost and time cost, it also has high cost-effectiveness.

Description

Technical Field

[0001] The present invention relates to the technical field of building complex monitoring, and in particular to a building inspection method and system based on existing building structures. Background Art

[0002] In the process of modern urbanization, with the rapid development of the construction industry, a large number of existing buildings have gradually entered the aging period, and their structural safety has become an important issue that needs urgent attention. These buildings not only carry the lives and work of residents, but are also directly related to public safety and social stability. Therefore, regular inspections of existing building structures and timely discovery and handling of potential safety hazards are of great significance to protecting people’s lives and property.

[0003] However, traditional building inspection methods often rely on manual on-site inspections, which have problems such as low efficiency, strong subjectivity, and difficulty in comprehensive coverage. Especially in high-rise buildings, complex structures or inaccessible areas, the difficulty and cost of manual inspections increase significantly. Summary of the invention

[0004] In view of this, the present invention proposes a building inspection method and system based on existing building structures, which can solve the defects of low efficiency, strong subjectivity and difficulty in comprehensive coverage in the prior art.

[0005] The technical solution of the present invention is achieved in this way:

[0006] A building inspection method based on an existing building structure, specifically comprising:

[0007] Obtain point cloud models of existing building structures based on LiDAR;

[0008] Acquire image data of existing building structures based on an image acquisition device;

[0009] The point cloud model and image data are registered and fused to obtain the inspection package;

[0010] The safety of the building structure is assessed based on the inspection package, thereby realizing the inspection of the building.

[0011] As a further optional solution of the building inspection method based on the existing building structure, the point cloud model of the existing building structure obtained based on the laser radar specifically includes:

[0012] Initializing the posture of the inspection instrument, wherein the initial posture of the inspection instrument is defined as the origin posture of the world coordinate system;

[0013] Based on the initial attitude of the inspection meter, the attitude of the inspection meter is predicted by combining the velocity and accelerometer difference readings;

[0014] Based on the predicted posture of the inspection instrument and the existing building structure, predict the point cloud data that the inspection instrument can collect;

[0015] The model is constructed based on the point cloud data collected by the predictive inspection instrument to obtain an initial point cloud model;

[0016] Collect the current point cloud data of the inspection instrument in real time based on laser radar;

[0017] Compare the point cloud data actually collected with the point cloud data that can be collected by the predictive inspection instrument to obtain error information;

[0018] The initial point cloud model is adjusted according to the error information to obtain the final point cloud model.

[0019] As a further optional solution of the building inspection method based on the existing building structure, the point cloud data actually collected is compared with the point cloud data that can be collected by the predictive inspection instrument to obtain error information, which specifically includes:

[0020] Extract the actual straight line edge features in the building interior from the actually collected point cloud data;

[0021] Obtain predicted straight line edge features inside the building from the initial point cloud model;

[0022] The geometric properties of the actual straight line edge features are compared with the predicted straight line edge features to obtain error information.

[0023] As a further optional solution of the building inspection method based on the existing building structure, the image data of the existing building structure is obtained based on the image acquisition device, specifically including:

[0024] Constructing a building structure unsafe factor identification model, wherein the input of the building structure unsafe factor identification model is an existing building structure image, and the output is a building structure image marked with building structure unsafe factors;

[0025] Collecting images of existing building structures according to an image acquisition device;

[0026] Input the collected existing building structure images into the building structure unsafe factor identification model for processing;

[0027] The building structure unsafe factor identification model outputs a building structure image that marks the building structure unsafe factors.

[0028] As a further optional solution of the building inspection method based on the existing building structure, the construction of the building structure unsafe factor identification model specifically includes:

[0029] Collect and manually annotate sample sets, where sample set annotations are composed of bounding boxes and masks;

[0030] Use a pre-trained convolutional neural network to extract feature maps of the sample set;

[0031] Generate target areas containing unsafe factors in feature maps based on region proposal network;

[0032] Classify the target area containing unsafe factors and regress the bounding box to obtain the classification result and bounding box regression result;

[0033] The feature map that is the same as the classification result and the bounding box regression result is processed according to the feature pyramid network to generate a pixel mask map containing unsafe factors;

[0034] Based on a combination of multiple loss functions, the classification results, bounding box regression results and pixel mask images containing unsafe factors are jointly trained to generate a building structure unsafe factor identification model; wherein the multiple loss functions include a classification loss function, a bounding box regression loss function and a mask loss function.

[0035] As a further optional solution of the building inspection method based on the existing building structure, the point cloud model and the image data are registered and fused to obtain the inspection package, which specifically includes:

[0036] Extract feature points from the point cloud model and image data respectively;

[0037] Based on the feature matching algorithm and the extracted feature points, corresponding feature point pairs are established between the point cloud model and the image data;

[0038] Using the matched feature point pairs, the transformation matrix from the point cloud model to the image data is solved;

[0039] According to the transformation matrix, the color information in the image data is mapped to the point cloud model to achieve the registration and fusion of the point cloud model and image data.

[0040] As a further optional solution of the building inspection method based on the existing building structure, the building structure safety assessment based on the inspection package specifically includes:

[0041] Transmit the inspection package to the inspector's terminal via the wireless network;

[0042] The inspector checks the inspection package in the current inspection record through the inspector terminal, and evaluates the safety of the building structure based on the inspection package.

[0043] A building inspection system based on an existing building structure, comprising:

[0044] The first acquisition module is used to acquire a point cloud model of an existing building structure based on a laser radar;

[0045] A second acquisition module is used to acquire image data of the existing building structure based on the image acquisition device;

[0046] The registration and fusion module is used to register and fuse the point cloud model and image data to obtain the inspection package;

[0047] The assessment module is used to assess the safety of building structures based on the inspection package, thereby realizing building inspections.

[0048] A computing device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements any one of the steps of the above-mentioned building inspection method based on an existing building structure when executing the computer program.

[0049] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the computer program implements any one of the steps of the above-mentioned building inspection method based on an existing building structure.

[0050] The beneficial effects of the present invention are as follows: the point cloud model obtained by using the laser radar has high-precision and high-density three-dimensional spatial information, which can accurately reflect the geometric form and detailed features of the building structure; the image data obtained by the image acquisition device provides visual information such as the appearance, texture and color of the building structure, which helps to understand the building status more intuitively; the point cloud model and the image data are registered and fused, which not only realizes the precise alignment of spatial information, but also gives the color information in the image data to the point cloud model through color mapping, so that the inspection package contains rich geometric information and visual information at the same time; by integrating the data acquisition, registration fusion and evaluation processes into one inspection package, the automation and Standardization has greatly reduced the labor intensity of manual inspections and improved inspection efficiency and accuracy. Automated inspections can also reduce the impact of human factors on inspection results, making the evaluation results more objective and reliable. The comprehensive data in the inspection package provides a rich source of information for building structure safety assessments. Based on these data, assessors can conduct a comprehensive and in-depth analysis of the stability, integrity, and damage of building structures. Compared with traditional manual inspections, inspection methods based on lidar and image acquisition technology can complete data collection and processing faster, significantly improving inspection efficiency. At the same time, since automated inspections reduce labor and time costs, they are also cost-effective. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0052] Figure 1 A schematic diagram of a flow chart of a building inspection method based on an existing building structure according to the present invention;

[0053] Figure 2 The present invention is a schematic diagram of the composition of a building inspection system based on an existing building structure. DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] refer to Figure 1-2 , a building inspection method based on an existing building structure, specifically comprising:

[0056] Obtain point cloud models of existing building structures based on LiDAR;

[0057] Acquire image data of existing building structures based on an image acquisition device;

[0058] The point cloud model and image data are registered and fused to obtain the inspection package;

[0059] The safety of the building structure is assessed based on the inspection package, thereby realizing the inspection of the building.

[0060] In this embodiment, the point cloud model obtained by the laser radar has high-precision and high-density three-dimensional spatial information, which can accurately reflect the geometric form and detailed features of the building structure. The image data obtained by the image acquisition device provides visual information such as the appearance, texture and color of the building structure, which helps to understand the building status more intuitively. The point cloud model and the image data are registered and fused, which not only realizes the precise alignment of spatial information, but also gives the color information in the image data to the point cloud model through color mapping, so that the inspection package contains rich geometric information and visual information at the same time. By integrating the data acquisition, registration fusion and evaluation processes into one inspection package, the automation and standardization of the inspection process are realized. Standardization greatly reduces the labor intensity of manual inspections and improves inspection efficiency and accuracy. Automated inspections can also reduce the impact of human factors on inspection results, making the evaluation results more objective and reliable. The comprehensive data in the inspection package provides a rich source of information for building structure safety assessments. Based on these data, assessors can conduct a comprehensive and in-depth analysis of the stability, integrity, and damage of building structures. Compared with traditional manual inspections, inspection methods based on lidar and image acquisition technology can complete data collection and processing faster, significantly improving inspection efficiency. At the same time, since automated inspections reduce labor and time costs, they are also cost-effective.

[0061] Preferably, the step of acquiring a point cloud model of an existing building structure based on a laser radar specifically includes:

[0062] Initializing the posture of the inspection instrument, wherein the initial posture of the inspection instrument is defined as the origin posture of the world coordinate system;

[0063] Based on the initial attitude of the inspection meter, the attitude of the inspection meter is predicted by combining the velocity and accelerometer difference readings;

[0064] Based on the predicted posture of the inspection instrument and the existing building structure, predict the point cloud data that the inspection instrument can collect;

[0065] The model is constructed based on the point cloud data collected by the predictive inspection instrument to obtain an initial point cloud model;

[0066] Collect the current point cloud data of the inspection instrument in real time based on laser radar;

[0067] Compare the point cloud data actually collected with the point cloud data that can be collected by the predictive inspection instrument to obtain error information;

[0068] The initial point cloud model is adjusted according to the error information to obtain the final point cloud model.

[0069] In this embodiment, the posture of the inspection meter is initialized and combined with the speed and accelerometer difference readings to predict the posture of the inspection meter. This method can plan the data collection path in advance, reduce unnecessary scanning and repetitive work, and thus improve the overall efficiency of data collection; based on the predicted posture of the inspection meter and the point cloud data that can be collected based on the existing building structure, it is helpful to estimate the coverage and quality of the data before data collection, and ensure that key areas are fully scanned; through the real-time collection of current point cloud data by lidar, and compared with the predicted point cloud data, errors and deviations in the data collection process can be discovered in time, and the initial point cloud model can be adjusted according to the error information, realizing dynamic optimization of the model. This implementation The real-time feedback and adjustment mechanism ensures the high precision and quality of the final point cloud model. Taking into account the various uncertainties that may be encountered during the inspection process (such as environmental interference, equipment failure, etc.), this technical solution enhances the robustness and fault tolerance of the system through the prediction and comparison mechanism. Even if deviations are encountered during the actual collection process, they can be corrected by adjusting the model to ensure the smooth progress of the inspection task. The entire data collection and model building process is highly automated, reducing human intervention and subjective judgment, and improving the objectivity and accuracy of the inspection work. The final point cloud model has high precision and high integrity, providing a solid data foundation for subsequent building structure safety assessment, damage detection, 3D visualization and other applications.

[0070] Preferably, the point cloud data actually collected is compared with the point cloud data that can be collected by the predictive inspection instrument to obtain error information, which specifically includes:

[0071] Extract the actual straight line edge features in the building interior from the actually collected point cloud data;

[0072] Obtain predicted straight line edge features inside the building from the initial point cloud model;

[0073] The geometric properties of the actual straight line edge features are compared with the predicted straight line edge features to obtain error information.

[0074] In this embodiment, the straight edge feature is a significant feature in the building structure, which has high geometric stability and recognition. By extracting these features for comparison, the difference between the actual collected data and the predicted data can be more accurately reflected. The geometric attribute comparison (such as position, direction, length, etc.) further ensures the accuracy and reliability of the comparison, which helps to obtain more accurate error information; the extraction and comparison of straight edge features help to eliminate local data inconsistency caused by factors such as point cloud density and noise. By focusing on the overall structural features, the overall consistency and comparability of the data can be improved; the error information directly reflects the deviation between the actual collected data and the predicted data, providing a clear direction and basis for model adjustment. Based on these error information, the initial point cloud model can be adjusted in a targeted manner to reduce the error and improve the accuracy of the model; the automatic extraction and comparison process of straight edge features is highly automated, reducing the need for manual intervention and subjective judgment, which not only improves work efficiency, but also reduces the risk of human error; accurate error information provides an important reference for subsequent building structure safety assessment, damage detection and other applications. By understanding the difference between the model and the actual structure, the safety performance of the building can be more accurately evaluated, and corresponding maintenance and management strategies can be formulated.

[0075] Preferably, the method of acquiring image data of the existing building structure based on the image acquisition device specifically includes:

[0076] Constructing a building structure unsafe factor identification model, wherein the input of the building structure unsafe factor identification model is an existing building structure image, and the output is a building structure image marked with building structure unsafe factors;

[0077] Collecting images of existing building structures according to an image acquisition device;

[0078] Input the collected existing building structure images into the building structure unsafe factor identification model for processing;

[0079] The building structure unsafe factor identification model outputs a building structure image that marks the building structure unsafe factors.

[0080] In this embodiment, the building structure unsafe factor identification model is trained to accurately identify potential problems in the building structure, such as cracks, corrosion, deformation and other unsafe factors. This automated identification method greatly improves the accuracy and efficiency of identification and reduces the subjectivity and omissions of manual inspection. By marking images of unsafe factors in the building structure, the safety status of the building structure can be intuitively displayed, which provides an important reference for subsequent safety assessment, damage detection, maintenance and reinforcement, etc., and helps to timely discover and deal with potential safety hazards. The image acquisition device can quickly obtain a large amount of image data of the building structure, and the building structure unsafe factor identification model can quickly process these data and output results. This efficient data processing process shortens the entire inspection and evaluation cycle and improves work efficiency. Traditional building structure inspections often require a lot of manpower input and have problems such as subjectivity and fatigue. By adopting this technical solution, the demand for manual inspection can be greatly reduced and the labor cost can be reduced. The image acquisition device and the building structure unsafe factor identification model can be integrated into the remote monitoring system to realize real-time monitoring and data analysis of the building structure. Once an unsafe factor is found, the system can immediately issue an alarm and generate a corresponding report to provide timely feedback and decision support for managers.

[0081] Preferably, the construction of the building structure unsafe factor identification model specifically includes:

[0082] Collect and manually annotate sample sets, where sample set annotations are composed of bounding boxes and masks;

[0083] Use a pre-trained convolutional neural network to extract feature maps of the sample set;

[0084] Generate target areas containing unsafe factors in feature maps based on region proposal network;

[0085] Classify the target area containing unsafe factors and regress the bounding box to obtain the classification result and bounding box regression result;

[0086] The feature map that is the same as the classification result and the bounding box regression result is processed according to the feature pyramid network to generate a pixel mask map containing unsafe factors;

[0087] Based on a combination of multiple loss functions, the classification results, bounding box regression results and pixel mask images containing unsafe factors are jointly trained to generate a building structure unsafe factor identification model; wherein the multiple loss functions include a classification loss function, a bounding box regression loss function and a mask loss function.

[0088] In this embodiment, by collecting and manually annotating sample sets, the accuracy and representativeness of the data are ensured. The dual annotation method of bounding boxes and masks not only identifies the location of unsafe factors, but also finely depicts their shape and range, providing a solid foundation for subsequent feature extraction and recognition; using pre-trained convolutional neural networks for feature extraction, fully utilizing the powerful capabilities of deep learning in the field of image processing, this method can quickly and effectively extract high-level, discriminative features from original images, providing strong support for subsequent target detection and segmentation; generating target regions containing unsafe factors based on region proposal networks (RPNs), this step significantly reduces the amount of computation for subsequent processing and improves the accuracy of target detection, RPN can automatically screen out candidate regions that may contain unsafe factors based on feature maps, providing effective input for subsequent classification and regression tasks; classifying target regions containing unsafe factors The classification task is used to determine the category of the target area (i.e., the type of unsafe factor), while the bounding box regression is used to accurately adjust the position and size of the target area so that it can more accurately reflect the actual situation of the unsafe factor. The feature map is processed by the feature pyramid network (FPN) to generate a pixel mask map containing the unsafe factors. This step realizes the fine depiction of the unsafe factors at the pixel level, providing more detailed information for subsequent damage assessment and repair plan design. The classification results, bounding box regression results and pixel mask map are jointly trained based on a combination of multiple loss functions to ensure the optimization and convergence of the entire recognition model. The classification loss function, bounding box regression loss function and mask loss function are optimized for different tasks respectively. The joint training achieves the coordination and balance between multiple tasks, thereby improving the recognition performance and generalization ability of the entire model.

[0089] Preferably, the step of registering and fusing the point cloud model and the image data to obtain the inspection package specifically includes:

[0090] Extract feature points from the point cloud model and image data respectively;

[0091] Based on the feature matching algorithm and the extracted feature points, corresponding feature point pairs are established between the point cloud model and the image data;

[0092] Using the matched feature point pairs, the transformation matrix from the point cloud model to the image data is solved;

[0093] According to the transformation matrix, the color information in the image data is mapped to the point cloud model to achieve the registration and fusion of the point cloud model and image data.

[0094] In this embodiment, feature points are extracted from the point cloud model and the image data respectively, and corresponding feature point pairs are established using a feature matching algorithm. This process ensures the accuracy of data fusion. Feature points are usually significant points or key points in the data. They are highly recognizable and consistent between different data types, so they can be used as reference points for data fusion. The transformation matrix from the point cloud model to the image data is solved by matching the feature point pairs. This step establishes the spatial position correspondence between the two data types. The transformation matrix contains transformation parameters such as rotation and translation, which can accurately map the information in the image data to the point cloud model, and realize the precise alignment of the two data in spatial position. The color information in the image data is mapped to the point cloud model, and the registration and fusion of the point cloud model and the image data is realized. This process not only retains the three-dimensional geometric information of the point cloud model, but also increases its color information, making the point cloud model more intuitive and vivid. Color information is of great significance for the identification, analysis and evaluation of building structures, and can provide richer visual clues and judgment basis. The inspection package after registration and fusion contains the comprehensive information of the point cloud model and the image data, providing more comprehensive and accurate data support for building inspection. Inspection personnel can use the inspection package to quickly understand the actual situation of the building structure, discover potential safety hazards and damage problems, and take corresponding measures to deal with them in a timely manner. This efficient and accurate inspection method greatly improves the efficiency and accuracy of the inspection work; the registered and fused data can be used for a variety of subsequent analyses and applications, such as damage detection, safety assessment, and three-dimensional reconstruction of building structures. These data not only provide an important reference basis for building management and maintenance, but also provide strong support for building design and renovation.

[0095] Preferably, the assessment of building structure safety based on the inspection package specifically includes:

[0096] Transmit the inspection package to the inspector's terminal via the wireless network;

[0097] The inspector checks the inspection package in the current inspection record through the inspector terminal, and evaluates the safety of the building structure based on the inspection package.

[0098] In this embodiment, the inspection package is transmitted to the inspector's terminal in real time through the wireless network, ensuring the timeliness and sharing of the inspection data. The inspector can directly obtain the latest inspection records and data at the inspection site without carrying a large amount of physical media or waiting for data to be transmitted back, thereby improving the response speed and efficiency of the inspection work; the inspector can easily view the inspection package in the current inspection record through the inspector's terminal, including the point cloud model, image data and the comprehensive information after registration and fusion. This intuitive and convenient data viewing method helps the inspector to quickly understand the actual situation of the building structure and provides strong support for subsequent evaluation work; the inspector can conduct a comprehensive safety assessment of the building structure based on the detailed data and comprehensive information in the inspection package. The point cloud model in the inspection package provides the three-dimensional geometric information of the building structure, and the image data provides visual information such as color and texture. The combination of the two makes the evaluation results more accurate and reliable. At the same time, inspectors can quickly identify potential safety hazards and damage problems based on the data in the inspection package and the actual situation, thereby improving the efficiency and accuracy of the evaluation work. Wireless network transmission also makes remote monitoring and guidance possible. Senior managers or experts can remotely view the inspection package and evaluation results through the network, and provide real-time guidance and supervision to the inspectors' work. This remote collaboration method not only improves the professional level of the inspection work, but also enhances the flexibility and efficiency of teamwork. The data and evaluation results in the inspection package can also be used to optimize the inspection process and make decision support. Through the analysis and mining of the inspection data, the weak links and potential problems in the inspection work can be discovered, and then improvement measures and optimization plans can be proposed. At the same time, the evaluation results can also provide a scientific basis for decisions such as the maintenance, reinforcement, renovation and upgrading of building structures.

[0099] A building inspection system based on an existing building structure, comprising:

[0100] The first acquisition module is used to acquire a point cloud model of an existing building structure based on a laser radar;

[0101] A second acquisition module is used to acquire image data of the existing building structure based on the image acquisition device;

[0102] The registration and fusion module is used to register and fuse the point cloud model and image data to obtain the inspection package;

[0103] The assessment module is used to assess the safety of building structures based on the inspection package, thereby realizing building inspections.

[0104] A computing device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements any one of the steps of the above-mentioned building inspection method based on an existing building structure when executing the computer program.

[0105] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the computer program implements any one of the steps of the above-mentioned building inspection method based on an existing building structure.

[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A building inspection method based on an existing building structure, characterized in that: Specifically include: Obtain point cloud models of existing building structures based on LiDAR; Acquire image data of existing building structures based on an image acquisition device; The point cloud model and image data are registered and fused to obtain the inspection package; Evaluate the safety of building structures according to the inspection package, thereby realizing building inspections; The step of obtaining a point cloud model of an existing building structure based on a laser radar specifically includes: Initializing the posture of the inspection instrument, wherein the initial posture of the inspection instrument is defined as the origin posture of the world coordinate system; Based on the initial attitude of the inspection meter, the attitude of the inspection meter is predicted by combining the velocity and accelerometer difference readings; Based on the predicted posture of the inspection instrument and the existing building structure, predict the point cloud data that the inspection instrument can collect; The model is constructed based on the point cloud data collected by the predictive inspection instrument to obtain an initial point cloud model; Collect the current point cloud data of the inspection instrument in real time based on laser radar; Compare the point cloud data actually collected with the point cloud data that can be collected by the predictive inspection instrument to obtain error information; Adjust the initial point cloud model according to the error information to obtain the final point cloud model; The point cloud data actually collected is compared with the point cloud data that can be collected by the predictive inspection instrument to obtain error information, which specifically includes: Extract the actual straight line edge features in the building interior from the actually collected point cloud data; Obtain predicted straight line edge features inside the building from the initial point cloud model; The geometric properties of the actual straight line edge features are compared with the predicted straight line edge features to obtain error information.

2. A building inspection method based on an existing building structure according to claim 1, characterized in that: The method of acquiring image data of the existing building structure based on the image acquisition device specifically includes: Constructing a building structure unsafe factor identification model, wherein the input of the building structure unsafe factor identification model is an existing building structure image, and the output is a building structure image marked with building structure unsafe factors; Collecting images of existing building structures according to an image acquisition device; Input the collected existing building structure images into the building structure unsafe factor identification model for processing; The building structure unsafe factor identification model outputs a building structure image that marks the building structure unsafe factors.

3. A building inspection method based on an existing building structure according to claim 2, characterized in that: The construction of the building structure unsafe factor identification model specifically includes: Collect and manually annotate sample sets, where sample set annotations are composed of bounding boxes and masks; Use a pre-trained convolutional neural network to extract feature maps of the sample set; Generate target areas containing unsafe factors in feature maps based on region proposal network; Classify the target area containing unsafe factors and regress the bounding box to obtain the classification result and bounding box regression result; The feature map that is the same as the classification result and the bounding box regression result is processed according to the feature pyramid network to generate a pixel mask map containing unsafe factors; Based on a combination of multiple loss functions, the classification results, bounding box regression results and pixel mask images containing unsafe factors are jointly trained to generate a building structure unsafe factor identification model; wherein the multiple loss functions include a classification loss function, a bounding box regression loss function and a mask loss function.

4. A building inspection method based on an existing building structure according to claim 3, characterized in that: The point cloud model and the image data are registered and fused to obtain the inspection package, which specifically includes: Extract feature points from the point cloud model and image data respectively; Based on the feature matching algorithm and the extracted feature points, corresponding feature point pairs are established between the point cloud model and the image data; Using the matched feature point pairs, the transformation matrix from the point cloud model to the image data is solved; According to the transformation matrix, the color information in the image data is mapped to the point cloud model to achieve the registration and fusion of the point cloud model and image data.

5. A building inspection method based on an existing building structure according to claim 4, characterized in that: The assessment of building structure safety based on the inspection package specifically includes: Transmit the inspection package to the inspector's terminal via the wireless network; The inspector checks the inspection package in the current inspection record through the inspector terminal, and evaluates the safety of the building structure based on the inspection package.

6. A building inspection system based on an existing building structure, characterized in that: include: The first acquisition module is used to acquire a point cloud model of an existing building structure based on a laser radar; A second acquisition module is used to acquire image data of the existing building structure based on the image acquisition device; The registration and fusion module is used to register and fuse the point cloud model and image data to obtain the inspection package; The assessment module is used to assess the safety of building structures according to the inspection package, thereby realizing building inspections; The step of obtaining a point cloud model of an existing building structure based on a laser radar specifically includes: Initializing the posture of the inspection instrument, wherein the initial posture of the inspection instrument is defined as the origin posture of the world coordinate system; Based on the initial attitude of the inspection meter, the attitude of the inspection meter is predicted by combining the velocity and accelerometer difference readings; Based on the predicted posture of the inspection instrument and the existing building structure, predict the point cloud data that the inspection instrument can collect; The model is constructed based on the point cloud data collected by the predictive inspection instrument to obtain an initial point cloud model; Collect the current point cloud data of the inspection instrument in real time based on laser radar; Compare the point cloud data actually collected with the point cloud data that can be collected by the predictive inspection instrument to obtain error information; Adjust the initial point cloud model according to the error information to obtain the final point cloud model; The point cloud data actually collected is compared with the point cloud data that can be collected by the predictive inspection instrument to obtain error information, which specifically includes: Extract the actual straight line edge features in the building interior from the actually collected point cloud data; Obtain predicted straight line edge features inside the building from the initial point cloud model; The geometric properties of the actual straight line edge features are compared with the predicted straight line edge features to obtain error information.

7. A computing device, characterized in that The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of any one of claims 1 to 5 of the building inspection method based on an existing building structure are implemented.

8. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the steps of any one of claims 1 to 5 of the building inspection method based on an existing building structure are implemented.

Images