High-rise building deformation monitoring method and system based on unmanned aerial vehicle oblique photography

Abstract

The application relates to the technical field of three-dimensional modeling, in particular to a high-rise building deformation monitoring method and system based on unmanned aerial vehicle oblique photography. The method comprises the following steps: determining a building attribution coefficient; determining an accessory coefficient based on the color difference between each data point and all data points in the cluster cluster where the data point is located and the distance from all data points to the gap midpoint, and combining the building attribution coefficient to determine a building possibility coefficient, so as to screen out building points from all data points of unlabeled point clouds; and constructing a three-dimensional building model based on all building points and the point cloud of each building, so as to be used for building deformation monitoring. The application solves the problem that point cloud data is mixed and classification is difficult due to narrow floor distance and obstruction of sundries in unmanned aerial vehicle oblique photography, thereby affecting the three-dimensional modeling and deformation monitoring precision, and improves the precision of high-rise building three-dimensional modeling and the accuracy of high-rise building deformation monitoring.

Description

Technical Field

[0001] This application relates to the field of 3D modeling technology, specifically to a method and system for monitoring the deformation of high-rise buildings based on UAV oblique photography. Background Technology

[0002] High-rise buildings are prone to deformation due to variations in load, geology, and structure. This deformation directly impacts their structural safety and service life; therefore, accurate monitoring is a crucial task throughout the entire building lifecycle. However, traditional geometric and physical analysis methods are inefficient and lack sufficient accuracy, failing to meet the demands of modern large-scale monitoring. Unmanned aerial vehicle (UAV) oblique photogrammetry technology efficiently acquires multi-angle information about buildings, providing reliable data for constructing high-precision 3D models and significantly improving the accuracy and efficiency of building deformation monitoring.

[0003] However, with urban development, many old high-rise buildings have become "handshake buildings" due to their close proximity. These buildings are filled with air conditioner units, security grilles, electrical cables, and clothes drying, posing a serious challenge to drone-based oblique photography. These obstacles not only severely obscure the key features of the buildings themselves but also introduce a large amount of interfering point clouds, causing a sharp decline in point cloud data quality. Ultimately, this reduces the accuracy of 3D modeling of high-rise buildings and the accuracy of deformation monitoring. Summary of the Invention

[0004] To address the aforementioned technical problems, the purpose of this application is to provide a method and system for monitoring the deformation of high-rise buildings based on UAV oblique photography. The specific technical solution adopted is as follows: In a first aspect, embodiments of this application provide a method for monitoring the deformation of high-rise buildings based on UAV oblique photography, the method comprising the following steps: Obtain the 3D point cloud of the target building complex and the normal vector of each data point in it. Pre-label the point cloud of each building in the 3D point cloud, and record the remaining point cloud as the unlabeled point cloud. For unlabeled point clouds, the unlabeled point clouds are divided into multiple point clouds based on the angle between the normal vectors of any two data points. All data points in each point cloud are clustered. Based on the dispersion of the distance between all data points in the cluster where each data point is located, and the dispersion of the angle between the normal vectors of all data points in the neighborhood of each data point, the distribution characteristic value of each data point is determined. Combined with the total number of data points in the cluster where each data point is located, the building affiliation coefficient of each data point is determined. For unlabeled point clouds, the midpoints of the gaps between each data point are obtained. Based on the color difference between each data point and all data points in its cluster, the color feature value of each data point is determined. Combined with the distances from each data point and all data points in its cluster to the midpoints of the gaps, the coefficients of the attached objects of each data point are determined. Based on the building affiliation coefficient and the attached object coefficient, the building probability coefficient of each data point is determined to filter out building points from all data points in the unlabeled point cloud; a three-dimensional building model is constructed based on all building points and the point cloud of each building for building deformation monitoring.

[0005] Preferably, the step of classifying the unlabeled point cloud into multiple point cloud categories includes: In an unlabeled point cloud, if the angle between the normal vectors of data point m and data point n is less than a preset angle threshold, then data point m and data point n are classified into one type of point cloud. All data points in the unlabeled point cloud are traversed to obtain all types of point clouds.

[0006] Preferably, the expression for the distribution characteristic value of each data point is: In the formula, This represents the distribution characteristic value of data point i in the unlabeled point cloud; This indicates the degree of dispersion of distances between all data points in the cluster to which data point i belongs in the unlabeled point cloud; This indicates the degree of dispersion of the angle between the normal vectors of all data points in the neighborhood of data point i in the unlabeled point cloud.

[0007] Preferably, the method for determining the building affiliation coefficient of each data point is as follows: Calculate the sum of the distribution characteristic value of each data point and the preset first factor, and record it as the distribution sum. The ratio of the total number of data points in the cluster to which each data point in the unlabeled point cloud belongs to the distribution sum is used as the building affiliation coefficient of each data point in the unlabeled point cloud, where the preset first factor is a constant greater than 0.

[0008] Preferably, the method for determining the color feature value of each data point is as follows: Calculate the difference between the pixel value of each data point and the mean pixel value of all data points in its neighborhood, and record it as the pixel value difference of each data point. Calculate the product of the pixel value of each data point and the corresponding pixel value difference, and use it as the color feature value of each data point.

[0009] Preferably, the expression for the accessory coefficient of each data point is: In the formula, This represents the coefficient of the attached objects of data point i in the unlabeled point cloud; This represents the color feature value of data point i in the unlabeled point cloud; This indicates the degree of dispersion of the distances from all data points in the cluster to the midpoint of the corresponding gap in the unlabeled point cloud; This represents the distance from data point i in the unlabeled point cloud to the midpoint of its gap; norm() represents the normalization function; This indicates a constant that is pre-defined as being greater than 0.

[0010] Preferably, the method for determining the building probability coefficient of each data point is as follows: Calculate the sum of the attached object coefficients of each data point and the preset second factor, and record it as the attached sum. Divide the building attribution coefficient of each data point in the unlabeled point cloud by the normalized value of the attached sum, and use it as the building probability coefficient of each data point in the unlabeled point cloud. The preset second factor is a constant greater than 0.

[0011] Preferably, the step of filtering building points from all data points in the unlabeled point cloud includes: The building probability coefficient of all data points in the unlabeled point cloud is used as the input of the threshold segmentation algorithm, and the output is the segmentation threshold. Data points in the unlabeled point cloud that are greater than or equal to the segmentation threshold are regarded as building points.

[0012] Preferably, the construction of the 3D building model based on all building points and the point cloud of each building includes: All building points and their point clouds are used as input to the 3D modeling software, and the resulting 3D model is used as the 3D architectural model of the target building complex.

[0013] Secondly, embodiments of this application also provide a high-rise building deformation monitoring system based on UAV oblique photography, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any of the above-described high-rise building deformation monitoring methods based on UAV oblique photography.

[0014] This application has at least the following beneficial effects: This application addresses the problem of numerous non-building debris mixed in with the point cloud data of gaps between buildings in "handshake buildings" scenarios. It employs normal vector angle classification and clustering analysis, combining the spatial distribution dispersion of the point cloud with the consistency of the normal vectors to construct a building attribution coefficient. This effectively distinguishes regular building structures from chaotic non-building points, significantly improving the classification accuracy of the point cloud data and providing a reliable data foundation for high-precision 3D modeling and building deformation monitoring. Furthermore, this application addresses the issue of metal attachments in "handshake buildings" scenarios being easily misidentified as the building itself. By introducing color features and gap distance information to construct an attachment coefficient, it effectively distinguishes highly reflective metal attachments such as security grilles and air conditioner units from the main building structure, significantly improving the accuracy of point cloud classification and providing strong support for the accuracy and reliability of subsequent 3D modeling. Finally, this application integrates the building attribution coefficient and the attachment coefficient to construct a building probability coefficient, achieving accurate differentiation between the building itself, metal attachments, and non-building debris. This effectively improves the quality of point cloud data and enhances the accuracy of 3D modeling and deformation monitoring of high-rise buildings. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating the steps of a high-rise building deformation monitoring method based on UAV oblique photography, provided as an embodiment of this application; Figure 2 A schematic diagram illustrating the building probability coefficient extraction process provided in one embodiment of this application; Figure 3 A flowchart illustrating the building point screening process provided in one embodiment of this application. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by this application to achieve the intended inventive objective, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the high-rise building deformation monitoring method and system based on UAV oblique photography proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0019] The following description, in conjunction with the accompanying drawings, details the specific scheme of the high-rise building deformation monitoring method and system based on UAV oblique photography provided in this application.

[0020] Please see Figure 1 The diagram illustrates a flowchart of a method for monitoring the deformation of high-rise buildings based on UAV oblique photography, according to an embodiment of this application. The method includes the following steps: Step S1: Obtain the 3D point cloud of the target building complex and the normal vector of each data point in it. Pre-label the point cloud of each building in the 3D point cloud, and record the remaining point cloud as unlabeled point cloud.

[0021] A drone is used to fly over the target building complex along a preset route and perform oblique photography to acquire image data of the target building complex. The image data is then input into point cloud extraction software to output a 3D point cloud of the target building complex. The 3D point cloud contains the position coordinates, color information (RGB value), normal vector, and timestamp of each data point.

[0022] Furthermore, since the buildings within the visible range have a high degree of visualization, the building point cloud data can be pre-annotated based on data fusion in advance. This allows for the pre-acquisition of some point cloud data belonging to the buildings themselves in the 3D point cloud of the target building complex, i.e., point cloud data with high visualization and clearly belonging to the main structure of the buildings. Point cloud processing software is then used to pre-annotate the point clouds of each building, and the remaining point clouds are recorded as unannotated point clouds.

[0023] In this embodiment, the point cloud processing software uses DJI Terra to pre-annotate the point clouds corresponding to each building in the target building complex. This is a well-known technology, and the specific annotation process will not be described in detail.

[0024] Step S2: For the unlabeled point cloud, divide the unlabeled point cloud into multiple point clouds based on the angle between the normal vectors of any two data points; cluster all data points in each point cloud, and determine the distribution characteristic value of each data point based on the dispersion of the distance between all data points in the cluster where each data point is located, and the dispersion of the angle between the normal vectors of all data points in the neighborhood of each data point. Combined with the total number of data points in the cluster where each data point is located, determine the building affiliation coefficient of each data point.

[0025] In the scenario of "handshake buildings," due to the extremely short distances between one or both sides of a building and other buildings, drones cannot perform gap-based photography. They can only obtain information about the relationship between two buildings through narrow gaps. Furthermore, due to residents' daily needs, these gaps often contain a large amount of non-building debris, such as electrical wires and cables, clothes drying, and protruding structural elements of the buildings themselves, such as balconies and window sills. These factors result in a large amount of non-building point cloud data mixed in with the final generated 3D point cloud, severely affecting the accuracy of subsequent modeling. Therefore, it is crucial to effectively distinguish this point cloud data to ensure the accuracy and reliability of the 3D building model for building deformation monitoring.

[0026] Specifically, point cloud data of buildings themselves typically exhibits strong spatial regularity. For example, structures such as balconies and window sills extending from buildings often share similarities in location, size, and appearance, resulting in point clouds that are locally concentrated and have similar overall distribution intervals. Furthermore, the normal vectors of building point clouds are relatively regular; for instance, the normal heights of the point cloud on one side of a square balcony are consistent, and the point cloud density is high, with a large number of points. In contrast, point clouds of non-building elements in the gaps between buildings, such as clothes drying in the sun, show significant disorder. Due to the diverse styles or wrinkles of clothing, the normal directions of their point clouds are highly chaotic, and the point cloud aggregation is low.

[0027] Therefore, based on the above analysis, this embodiment divides unlabeled point clouds into multiple point cloud classes based on the angle between the normal vectors of any two data points. All data points in each class are clustered, and the distribution characteristic value of each data point is determined based on the dispersion of the distance between all data points in each cluster and the dispersion of the angle between the normal vectors of all data points in the neighborhood of each data point. Combined with the total number of data points in each cluster, the building affiliation coefficient of each data point is determined. The specific process is as follows: First, this embodiment classifies unlabeled point clouds into multiple point cloud types based on the angle between the normal vectors of any two data points. Specifically: In an unlabeled point cloud, if the angle between the normal vectors of data point m and data point n is less than a preset angle threshold, then data point m and data point n are classified into one type of point cloud. All data points in the unlabeled point cloud are traversed to obtain all types of point clouds. Each type of point cloud is used to represent point clouds that are relatively consistent in local geometry, that is, to represent a set of data points with similar normal vector orientations.

[0028] It should be noted that the preset angle threshold is set manually. In this embodiment, the preset angle threshold is 2°. In actual applications, as other implementation methods, implementers can also set it according to specific circumstances. This embodiment does not impose any special restrictions.

[0029] The method for calculating the angle between the normal vectors is a well-known technique, and the specific calculation process will not be elaborated here.

[0030] Furthermore, in this embodiment, all data points in each type of point cloud are clustered. Based on the dispersion of the distance between all data points in the cluster to which each data point belongs, and the dispersion of the angle between the normal vectors of all data points in the neighborhood of each data point, the distribution characteristic value of each data point is determined, specifically as follows: In this embodiment, all data points in each type of point cloud are first used as input to the clustering algorithm. The metric distance is set to the Euclidean distance between data points, the stage distance is set to 3, the elbow rule is used to determine the number of clusters, and all clusters are output. Each cluster represents an independent object or part of an object that is similar in geometric direction and continuous in spatial location.

[0031] It should be noted that there are many commonly used clustering algorithms. In this embodiment, the density peak clustering (DPC) algorithm is used to cluster the data points. In practical applications, as other implementation methods, implementers may also use other clustering methods such as density clustering (DBSCAN) depending on the specific circumstances. This embodiment does not impose any special restrictions on the selection of clustering algorithms.

[0032] The calculation method of Euclidean distance, the determination of the number of clusters using the elbow rule, and the process of clustering all data points in each type of point cloud using the density peak clustering (DPC) algorithm are all well-known techniques. The specific process of calculating Euclidean distance, determining the number of clusters using the elbow rule, and clustering data points using the density peak clustering (DPC) algorithm will not be described in detail here.

[0033] Furthermore, this embodiment determines the distribution characteristic value of each data point based on the dispersion of the distance between all data points in the cluster to which each data point belongs, and the dispersion of the angle between the normal vectors of all data points in the neighborhood of each data point. Specifically: As one implementation method, in this embodiment, the distribution characteristic value of data point i in the unlabeled point cloud is... The expression is: In the formula, This indicates the degree of dispersion of distances between all data points in the cluster to which data point i belongs in the unlabeled point cloud; This indicates the degree of dispersion of the angle between the normal vectors of all data points in the neighborhood of data point i in the unlabeled point cloud.

[0034] It should be noted that there are many methods to measure the dispersion of data. In this embodiment, the variance of the distance between all data points in the cluster where data point i is located in the unlabeled point cloud is taken as the dispersion of the distance between all data points in the cluster where data point i is located in the unlabeled point cloud. The variance of the angle between the normal vectors of all data points in the neighborhood of data point i in the unlabeled point cloud is taken as the dispersion of the angle between the normal vectors of all data points in the neighborhood of data point i in the unlabeled point cloud. In practical applications, as other implementation methods, implementers may also choose other methods to measure the dispersion of data, such as standard deviation or coefficient of variation, depending on the specific circumstances. This embodiment does not impose any special restrictions on the selection of methods to measure the dispersion of data.

[0035] It should be noted that the process of constructing the neighborhood of a data point is as follows: a circular area with each data point as the center and r as the radius is the neighborhood of each data point. In practical applications, as other implementation methods, implementers can also set their own neighborhood construction methods according to specific circumstances. This embodiment does not impose any special restrictions. In this embodiment, the value of r is 0.5m.

[0036] It should be noted that, unless otherwise specified, all content involving the calculation of dispersion in this embodiment uses the variance calculation method.

[0037] Based on the distribution characteristic values ​​of each data point in the unlabeled point cloud, we can understand that the distribution characteristic values ​​reflect the geometric regularity and directional consistency of the local area where the data point is located. The dispersion of the distance between all data points in the cluster where the current data point is located reflects the dispersion of the data point distribution. The dispersion of the angle between the normal vectors of all data points in the neighborhood of the current data point reflects the consistency of the point cloud direction in the local area where the data point is located. The larger the product of the two, the more dispersed and chaotic the point cloud distribution in the local area where the current data point is located, and the more likely the current data point belongs to a non-building point, such as clothing, wires, or other messy objects. Conversely, the smaller the product of the two, the more concentrated and consistent the point cloud distribution in the local area where the current data point is located, and the more likely the current data point belongs to a building body, such as a regular building structure like a balcony or window sill.

[0038] Furthermore, this embodiment is based on the distribution characteristic values ​​of each data point, combined with the total number of data points in the cluster to which each data point belongs, specifically as follows: In this embodiment, the sum of the distribution characteristic value of each data point and the preset first factor is calculated and recorded as the distribution sum. The ratio of the total number of data points in the cluster to which each data point in the unlabeled point cloud belongs to the distribution sum is used as the building affiliation coefficient of each data point in the unlabeled point cloud. The preset first factor is a constant greater than 0 to prevent the denominator from being 0. Its value is set manually. In this embodiment, the preset first factor is 0.01. In actual application, as other implementation methods, implementers can also set it according to specific circumstances. This embodiment does not impose any special restrictions.

[0039] Based on the building affiliation coefficient of each data point in the unlabeled point cloud, we can understand that the building affiliation coefficient reflects the probability that the data point in the unlabeled point cloud belongs to a building. If there are more data points in the cluster where the current data point is located, it means that the point cloud distribution in the local area where the current data point is located is more dense and the building structure is more prominent. Therefore, the larger the corresponding building affiliation coefficient, the greater the probability that the current data point belongs to a building. At the same time, if the distribution characteristic value of the current data point is larger, it means that the geometric features of the local area where the current data point is located are more regular. The corresponding building affiliation coefficient is larger, indicating that the probability that the current data point belongs to a building is greater. Conversely, if there are fewer data points in the cluster where the current data point is located, it indicates that the point cloud distribution in the local area where the current data point is located is more sparse, and it is more likely to belong to discrete noise or cluttered objects. Therefore, the smaller the corresponding building affiliation coefficient, the less likely the current data point belongs to a building. At the same time, if the distribution characteristic value of the current data point is smaller, it indicates that the geometric features of the local area where the current data point is located are more irregular and the orientation is more chaotic, and it is more likely to belong to unstructured objects such as clothing. The corresponding building affiliation coefficient is also smaller, further indicating that the current data point is less likely to belong to a building.

[0040] Thus, this embodiment addresses the problem of numerous non-building debris mixed in with the point cloud between buildings in a "handshake building" scenario. By classifying and clustering the normal vector angles, and combining the spatial distribution dispersion of the point cloud with the consistency of the normal vectors to construct building attribution coefficients, it effectively distinguishes between regular building structures and chaotic non-building points, significantly improving the classification accuracy of point cloud data and providing a reliable data foundation for high-precision 3D modeling and building deformation monitoring.

[0041] Step S3: For the unlabeled point cloud, obtain the midpoint of the gap between each data point. Based on the color difference between each data point and all data points in its cluster, determine the color feature value of each data point. Combine the distance from each data point and all data points in its cluster to the midpoint of the gap, determine the attachment object coefficient of each data point.

[0042] For residential buildings, in addition to the building's own prominent ancillary structures, such as balconies and windowsills, there may also be ancillary objects that are closely related to the building but are not part of the building itself. For example, burglar bars added by residents for security, and air conditioner outdoor units attached when air conditioners are installed. These objects are often located between building gaps. Since burglar bars are highly related to windowsills and balconies in terms of position and structure, and the overall structure of air conditioner outdoor units is also relatively regular, if only the geometric features of step S2, such as point cloud distribution and normal consistency, are used for judgment, they are easily misjudged as point clouds of the building itself, leading to structural errors or decreased accuracy in subsequent modeling. Therefore, it is necessary to further introduce other features or constraints to distinguish them.

[0043] Specifically, for accessories such as security grilles or air conditioner outdoor units, which are usually made of metal, in the poorly lit gaps between buildings, if there is sufficient light, their reflectivity is stronger than that of the building's main structures, such as balconies and window sills, resulting in a more prominent overall brightness. In addition, as the degree of obstruction in the gaps gradually increases from top to bottom and from outside to inside, the area closer to the center of the gap is less affected by obstruction and has stronger reflectivity. Therefore, the point cloud data of such accessories often exhibits high brightness and complex surface texture. Specifically, if a data point belongs to an accessory, the closer it is to the center of the gap, the higher its brightness; within its cluster, its brightness is also more prominent. At the same time, compared with building accessories, the flatness of such point clouds fluctuates more greatly with respect to the distance from the center point of the gap.

[0044] Therefore, based on the above analysis, this embodiment, for unlabeled point clouds, obtains the midpoints of the gaps between each data point, determines the color feature value of each data point based on the color difference between each data point and all data points in its cluster, and determines the attachment object coefficient of each data point by combining the distances from each data point and all data points in its cluster to the midpoints of the gaps, specifically: First, in this embodiment, for the unlabeled point cloud, the midpoints of the gaps between each data point are obtained, specifically: In the unlabeled point cloud, the center point of the gap between the nearest handshake buildings is taken as the midpoint of the gap for each data point. This process is repeated for all data points to obtain the midpoint of the gap for each data point.

[0045] It should be noted that the specific process for determining the midpoint of the gap is as follows: Based on the point cloud data of each building obtained from the pre-annotation in step S1, the relative positional relationship between any two adjacent buildings is determined. For any two adjacent buildings, the centroids of the point clouds of the two buildings are calculated, and the midpoint of the line connecting the two centroids is initially determined as the initial gap center point between the two buildings. Considering the irregularity of the building plan shape, in order to improve the positioning accuracy, a sphere with a preset distance as the radius is used as the reference, and the closest point pair between the two buildings is found within the search window. The midpoint of the closest point pair is calculated as the final gap center point. The calculation logic of the preset distance is as follows: Assuming the average width of the buildings is 20 meters and the distance between the centroids of the two buildings is 30 meters, then the distance from the gap center to each building is approximately (30-20) / 2=5 meters. Therefore, the preset distance of the search window is set to 4~5 meters. In this embodiment, as one implementation method, the preset distance is set to 5 meters. In actual application, the implementer can also set it according to the specific situation. This embodiment does not impose any special restrictions.

[0046] Furthermore, in this embodiment, the color feature value of each data point is determined based on the color difference between each data point and all data points in its cluster, specifically as follows: In this embodiment, the difference between the pixel value of each data point and the average pixel value of all data points in its neighborhood is calculated and denoted as the pixel value difference of each data point. The product of the pixel value of each data point and the corresponding pixel value difference is calculated as the color feature value of each data point. It should be noted that there are many methods to measure the difference between data points. In this embodiment, the absolute value of the difference between the pixel value of each data point in the unlabeled point cloud and the average pixel value of all data points in its neighborhood is taken as the difference between the pixel value of each data point in the unlabeled point cloud and the average pixel value of all data points in its neighborhood is used. In practical applications, as other implementation methods, the implementer may also use other methods to measure the difference between data points, such as the square or ratio of the difference, depending on the specific circumstances. This embodiment does not impose any special restrictions on the selection of methods for measuring the difference between data points.

[0047] It should be noted that, in this embodiment, since the three-dimensional point cloud contains color information, the three-dimensional point cloud is converted into a grayscale image to obtain the grayscale value of each data point. The pixel value of each data point is the corresponding grayscale value in this embodiment. The specific process of obtaining the grayscale image of the three-dimensional point cloud is a well-known technology and will not be described in detail here.

[0048] Based on the color feature values ​​of each data point in the unlabeled point cloud, it can be understood that the color feature values ​​reflect the prominence of the current data point in terms of brightness. If the difference between the pixel value of the current data point and the average pixel value of all data points in its neighborhood is greater, and the pixel value of the current data point is larger, it indicates that the brightness of the current data point is significantly higher than that of the surrounding point cloud, showing obvious brightness anomalies. This is usually related to the high reflectivity of metal materials, such as security grilles and air conditioner outdoor units. Therefore, the current data point is more likely to belong to an accessory object, and the corresponding color feature value is larger. Conversely, if the difference between the pixel value of the current data point and the average pixel value of all data points in its neighborhood is smaller, and the pixel value of the current data point is smaller, it means that the brightness of the current data point is closer to that of the surrounding point cloud and does not show obvious brightness anomalies. It is more likely to belong to a large area of ​​uniform material surface such as the main body of a building or the ground. Therefore, the current data point is less likely to belong to an accessory object, and the corresponding color feature value is also smaller.

[0049] Furthermore, this embodiment determines the accessory coefficient of each data point based on the color feature value of each data point, and in combination with the distance from each data point and all data points in its cluster to the midpoint of the gap, specifically as follows: As one implementation method, in this embodiment, the coefficients of the attached objects of data point i in the point cloud are not labeled. The expression is: In the formula, This represents the color feature value of data point i in the unlabeled point cloud; This indicates the degree of dispersion of the distances from all data points in the cluster to the midpoint of the corresponding gap in the unlabeled point cloud; This represents the distance from data point i to the midpoint of its gap; norm() represents the normalization function; This represents a preset constant greater than 0, used to prevent the denominator from being 0. Under the premise of ensuring that the denominator is not 0 and does not excessively affect the calculation result, the implementer can also set it according to the specific situation. This embodiment does not impose any special restrictions.

[0050] It should be noted that, in this embodiment, the color feature value... and Normalization was performed using the maximum-minimum normalization method, and the color feature values ​​were then normalized. and Mapped to the range of [0,1], in practical applications, as other implementation methods, implementers may also choose other normalization methods according to specific circumstances. This embodiment does not impose any special restrictions.

[0051] The method of normalizing data using the maximum and minimum value normalization method is a well-known technique and will not be elaborated further.

[0052] Based on the accessory coefficients of each data point, we can understand that the accessory coefficients reflect the probability that a data point belongs to a building accessory. If the color feature value of data point i in the unlabeled point cloud is larger, it means that the brightness of data point i is higher, indicating that the data point i is more likely to be made of metal materials such as burglar bars or air conditioner outdoor units. The corresponding accessory coefficient is larger, indicating that the data point i is more likely to be an accessory and less likely to be part of the building. At the same time, if the dispersion of the distances from all data points in the cluster to the midpoint of the corresponding gap is greater, and the distance from data point i to its midpoint is smaller, it means that the data point i is more likely to be an accessory, and the corresponding accessory coefficient is larger. Conversely, if the color feature value of data point i in the unlabeled point cloud is smaller, it indicates that the brightness of data point i is lower and there is no significant difference from the surrounding environment. It is more likely to belong to a non-metallic, low-reflectivity surface such as the main body of the building, walls, or the ground. Therefore, the possibility that data point i belongs to an accessory is smaller, and the corresponding accessory coefficient is also smaller. At the same time, if the dispersion of the distances from all data points in the cluster to the midpoint of the corresponding gap in the unlabeled point cloud is smaller, and the distance from data point i to its midpoint is larger, it indicates that the overall position of the cluster is more regular and farther away from the edge of the main body of the building. It is more likely to belong to the main structure of the building. Therefore, the possibility that data point i belongs to an accessory is also smaller, and the corresponding accessory coefficient is also smaller.

[0053] Thus, this embodiment addresses the problem of metal attachments in "handshake building" scenarios being easily misidentified as the building itself. By introducing color features and gap distance information to construct attachment coefficients, it effectively distinguishes highly reflective metal attachments such as security grilles and air conditioner outdoor units from the main building structure, significantly improving the accuracy of point cloud classification and providing strong support for the accuracy and reliability of subsequent 3D modeling.

[0054] Step S4: Based on the building affiliation coefficient and the attached object coefficient, determine the building probability coefficient of each data point to filter out building points from all data points in the unlabeled point cloud; construct a 3D building model based on all building points and the point cloud of each building for building deformation monitoring.

[0055] Based on the building attribution coefficient obtained in step S2 and the accessory object coefficient obtained in step S3, the building probability coefficient of each data point in the unlabeled point cloud is further determined as follows: In this embodiment, the sum of the attached object coefficient of each data point and the preset second factor is calculated and recorded as the attached sum. The normalized value of the building attribution coefficient of each data point in the unlabeled point cloud divided by the attached sum is used as the building probability coefficient of each data point in the unlabeled point cloud. The preset second factor is a constant greater than 0 to prevent the denominator from being 0. Its value is set manually. In this embodiment, the value of the preset second factor is 0.01. In actual application, as other implementation methods, implementers can also set it according to specific circumstances. This embodiment does not impose any special restrictions.

[0056] It should be noted that in this embodiment, the maximum-minimum normalization method is used to normalize the result of the ratio of the building affiliation coefficient to the attached sum of values ​​of each data point in the unlabeled point cloud, mapping it to the range of [0,1]. In actual application, as other implementation methods, implementers may also use other normalization methods according to specific circumstances. This embodiment does not impose any special restrictions on the selection of normalization methods.

[0057] Preferably, the schematic diagram of the building possibility coefficient extraction process provided in this embodiment is as follows: Figure 2 As shown.

[0058] Based on the building probability coefficients of each data point in the unlabeled point cloud, we can understand that the building probability coefficients reflect the comprehensive probability that a data point belongs to a building. If the building belonging coefficient of the current data point in the unlabeled point cloud is larger, it means that the geometric features of the current data point are more consistent with the building, and the corresponding building probability coefficient is higher. At the same time, if the accessory object coefficient of the current data point in the unlabeled point cloud is smaller, it means that the current data point is more likely to belong to the building, and therefore, the corresponding building probability coefficient is higher. Conversely, if the building affiliation coefficient of the current data point in the unlabeled point cloud is smaller, it indicates that the geometric characteristics of the current data point are less consistent with the building itself. For example, if the distribution is sparse or the shape is irregular, it is more likely to belong to vegetation, debris or noise. Therefore, the corresponding building probability coefficient is lower. At the same time, if the accessory object coefficient of the current data point in the unlabeled point cloud is larger, it indicates that the current data point is more likely to belong to accessory structures such as security grilles or air conditioner outdoor units, rather than the main building. Therefore, the corresponding building probability coefficient is also lower.

[0059] Furthermore, in this embodiment, building points are selected from all data points in the unlabeled point cloud based on the building probability coefficient. Specifically: In this embodiment, the building probability coefficients of all data points in the unlabeled point cloud are used as the input of the threshold segmentation algorithm, and the output is the segmentation threshold. Data points in the unlabeled point cloud that are greater than or equal to the segmentation threshold are regarded as building points.

[0060] Preferably, the flowchart of the building point screening process provided in this embodiment is as follows: Figure 3 As shown.

[0061] It should be noted that there are many commonly used threshold segmentation algorithms. In this embodiment, the maximum inter-class variance algorithm is used to obtain the segmentation threshold and divide the data points. In practical applications, as other implementation methods, implementers may also use other threshold segmentation algorithms such as cross-validation according to specific circumstances. This embodiment does not impose any special restrictions on the selection of threshold segmentation algorithms.

[0062] The process of obtaining the segmentation threshold and dividing the data points using the Otsu's inter-class variance algorithm is a well-known technique and will not be elaborated further.

[0063] Furthermore, all building points in the unlabeled point cloud and the point clouds of each building obtained from the pre-labeling in step S1 are output to the 3D modeling software to model the building point cloud. The output 3D model is used as the 3D model of the target building complex.

[0064] It should be noted that there are many commonly used 3D modeling software programs. In this embodiment, ContextCapture modeling software is used to model each building in the target building complex. In actual applications, as other implementation methods, implementers may also use other software such as Pix4Dmapper software depending on the specific circumstances. This embodiment does not impose any special restrictions.

[0065] The process of using ContextCapture modeling software to create 3D models of building point clouds and building points is a well-known technique and will not be elaborated further.

[0066] Furthermore, in this embodiment, the 3D model of the target building complex is used as input to the GOM Inspect software, and the deformation results of each building in the 3D model are output. The deformation results are presented as a building visualization model with deformation heatmap, displacement arrows, and maximum settlement index parameters.

[0067] The process of using GOM Inspect software to calculate building deformation results is a well-known technique and will not be described in detail here.

[0068] Thus, this embodiment addresses the problem of difficulty in point cloud classification and decreased modeling accuracy caused by narrow building spacing and obstruction by debris in drone oblique photography of "handshake buildings" scenarios. By integrating multi-dimensional features such as geometric distribution, normal vector consistency, color brightness, and gap position, a building attribution coefficient and ancillary object coefficient are constructed, and a building probability coefficient is comprehensively determined. This enables accurate differentiation of the building body, metal accessories, and non-building debris, effectively improving the quality of point cloud data and enhancing the accuracy of 3D modeling of high-rise buildings and the accuracy of deformation monitoring of high-rise buildings.

[0069] Based on the same inventive concept as the above methods, this application also provides a high-rise building deformation monitoring system based on UAV oblique photography, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described high-rise building deformation monitoring methods based on UAV oblique photography.

[0070] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments of this specification have been described above. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0071] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0072] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A method for monitoring the deformation of high-rise buildings based on UAV oblique photography, characterized in that, The method includes the following steps: Obtain the 3D point cloud of the target building complex and the normal vector of each data point in it. Pre-label the point cloud of each building in the 3D point cloud, and record the remaining point cloud as the unlabeled point cloud. For unlabeled point clouds, the unlabeled point clouds are divided into multiple point clouds based on the angle between the normal vectors of any two data points. All data points in each point cloud are clustered. Based on the dispersion of the distance between all data points in the cluster where each data point is located, and the dispersion of the angle between the normal vectors of all data points in the neighborhood of each data point, the distribution characteristic value of each data point is determined. Combined with the total number of data points in the cluster where each data point is located, the building affiliation coefficient of each data point is determined. For unlabeled point clouds, the midpoints of the gaps between each data point are obtained. Based on the color difference between each data point and all data points in its cluster, the color feature value of each data point is determined. Combined with the distances from each data point and all data points in its cluster to the midpoints of the gaps, the coefficients of the attached objects of each data point are determined. Based on the building affiliation coefficient and the attached object coefficient, the building probability coefficient of each data point is determined to filter out building points from all data points in the unlabeled point cloud; a three-dimensional building model is constructed based on all building points and the point cloud of each building for building deformation monitoring.

2. The method for monitoring the deformation of high-rise buildings based on UAV oblique photography as described in claim 1, characterized in that, The method of classifying unlabeled point clouds into multiple point cloud categories includes: In an unlabeled point cloud, if the angle between the normal vectors of data point m and data point n is less than a preset angle threshold, then data point m and data point n are classified into one type of point cloud. All data points in the unlabeled point cloud are traversed to obtain all types of point clouds.

3. The method for monitoring the deformation of high-rise buildings based on UAV oblique photography as described in claim 1, characterized in that, The expression for the distribution characteristic value of each data point is: In the formula, This represents the distribution characteristic value of data point i in the unlabeled point cloud; This indicates the degree of dispersion of distances between all data points in the cluster to which data point i belongs in the unlabeled point cloud; This indicates the degree of dispersion of the angle between the normal vectors of all data points in the neighborhood of data point i in the unlabeled point cloud.

4. The method for monitoring the deformation of high-rise buildings based on UAV oblique photography as described in claim 1, characterized in that, The method for determining the building affiliation coefficient for each data point is as follows: The ratio of the total number of data points in the cluster to the distribution characteristic value.

5. The method for monitoring the deformation of high-rise buildings based on UAV oblique photography as described in claim 1, characterized in that, The method for determining the color feature value of each data point is as follows: Calculate the sum of the distribution characteristic value of each data point and the preset first factor, and record it as the distribution sum. The ratio of the total number of data points in the cluster to which each data point in the unlabeled point cloud belongs to the distribution sum is used as the building affiliation coefficient of each data point in the unlabeled point cloud, where the preset first factor is a constant greater than 0.

6. The method for monitoring the deformation of high-rise buildings based on UAV oblique photography as described in claim 1, characterized in that, The expression for the coefficient of the attached object for each data point is as follows: In the formula, This represents the coefficient of the attached objects of data point i in the unlabeled point cloud; This represents the color feature value of data point i in the unlabeled point cloud; This indicates the degree of dispersion of the distances from all data points in the cluster to the midpoint of the corresponding gap in the unlabeled point cloud; This represents the distance from data point i in the unlabeled point cloud to the midpoint of its gap; norm() represents the normalization function; This indicates a constant that is pre-defined as being greater than 0.

7. The method for monitoring the deformation of high-rise buildings based on UAV oblique photography as described in claim 1, characterized in that, The method for determining the building probability coefficient for each data point is as follows: Calculate the sum of the attached object coefficients of each data point and the preset second factor, and record it as the attached sum. Divide the building attribution coefficient of each data point in the unlabeled point cloud by the normalized value of the attached sum, and use it as the building probability coefficient of each data point in the unlabeled point cloud. The preset second factor is a constant greater than 0.

8. The method for monitoring the deformation of high-rise buildings based on UAV oblique photography as described in claim 1, characterized in that, The process of filtering building points from all data points in the unlabeled point cloud includes: The building probability coefficient of all data points in the unlabeled point cloud is used as the input of the threshold segmentation algorithm, and the output is the segmentation threshold. Data points in the unlabeled point cloud that are greater than or equal to the segmentation threshold are regarded as building points.

9. The method for monitoring the deformation of high-rise buildings based on UAV oblique photography as described in claim 1, characterized in that, The construction of a 3D building model based on all building points and the point cloud of each building includes: All building points and their point clouds are used as input to the 3D modeling software, and the resulting 3D model is used as the 3D architectural model of the target building complex.

10. A high-rise building deformation monitoring system based on UAV oblique photography, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the high-rise building deformation monitoring method based on UAV oblique photography as described in any one of claims 1-9.

Images