Building model processing method and system and medium

Abstract

The invention relates to the technical field of building models, in particular to a building model processing method and system and a medium. The method comprises the following steps: acquiring initial image data of a high-rise structure through an aircraft, and synchronously recording motion state data; calculating a jitter characteristic parameter according to the initial image data and the motion state data; establishing a jitter compensation mechanism of the long-distance image based on the jitter characteristic parameters; processing the initial image data by using a jitter compensation mechanism to obtain stable image data; extracting structural feature point information in the stable image data; establishing a high-rise structure reference coordinate system according to the structural feature point information; and performing high-precision registration on pre-acquired multi-angle image data of the high-rise structure and the reference coordinate system of the high-rise structure to obtain registered image data. According to the invention, by analyzing the jitter characteristics of different frequency bands, a multi-level jitter compensation mechanism is established, and the stability problem during long-distance image acquisition is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of architectural models, and in particular to a method, system and medium for processing architectural models. Background Art

[0002] A tall structure is one with a relatively small cross-section and is typically designed primarily based on horizontal loads (particularly wind loads). An architectural model is a scaled-down version of the actual building, used to illustrate its exterior, interior structure, spatial layout, and material texture. It can be either physical or digital. As the core vehicle for project lifecycle management, architectural models play a vital role in the design, construction, and operation and maintenance phases.

[0003] Traditional methods for processing tall structural building models often have the following problems: in high-altitude strong wind environments, drones shake severely when shooting near wind towers, resulting in blurred images; during large bridge inspections, the images obtained are blurred and have ghosting due to the vibration of the inspection equipment itself and the slight movement of the bridge; when inspecting the tops of ultra-high buildings, the magnification effect of the long-range telephoto lens will amplify subtle jitters into obvious distortion. Summary of the Invention

[0004] Based on this, it is necessary for the present invention to provide a method, system and medium for processing a building model to solve at least one of the above technical problems.

[0005] To achieve the above object, a method for processing a building model includes the following steps:

[0006] Step S1: Acquire initial image data of a tall structure through an aircraft and simultaneously record motion state data; calculate jitter characteristic parameters based on the initial image data and the motion state data; and establish a jitter compensation mechanism for long-range images based on the jitter characteristic parameters;

[0007] Step S2: Processing the initial image data using a jitter compensation mechanism to obtain stabilized image data; extracting structural feature point information from the stabilized image data; and establishing a towering structure reference coordinate system based on the structural feature point information;

[0008] Step S3: High-precision registration of the pre-acquired multi-angle image data of the tall building structure with the tall building structure reference coordinate system to obtain registered image data; constructing a three-dimensional point cloud model of the tall building structure based on the registered image data; and extracting abnormal feature information of the building surface from the three-dimensional point cloud model according to the tall building structure reference coordinate system;

[0009] Step S4: Analyze the type of building surface damage based on the abnormal characteristic information of the building surface, and perform graded processing to obtain structural damage assessment data; generate a health status report based on the structural damage assessment data; compare and analyze the health status report with preset historical data to obtain the structural damage development trend; formulate a predictive maintenance plan based on the structural damage development trend to form inspection decision support data.

[0010] The present invention further provides a building model processing system for executing the above-mentioned building model processing method, wherein the building model processing system comprises:

[0011] The image acquisition and jitter compensation module is used to acquire initial image data of the tall structure through the aircraft and simultaneously record the motion status data; calculate the jitter characteristic parameters based on the initial image data and the motion status data; and establish a jitter compensation mechanism for long-range images based on the jitter characteristic parameters;

[0012] The image stabilization and coordinate system establishment module is used to process the initial image data using a jitter compensation mechanism to obtain stabilized image data; extract structural feature point information from the stabilized image data; and establish a towering structure reference coordinate system based on the structural feature point information;

[0013] The 3D modeling and anomaly detection module is used to perform high-precision registration of pre-acquired multi-angle image data of the towering structure with the towering structure reference coordinate system to obtain registered image data; construct a 3D point cloud model of the towering structure based on the registered image data; and extract abnormal feature information of the building surface from the 3D point cloud model according to the towering structure reference coordinate system;

[0014] The damage assessment and maintenance decision module is used to analyze the type of building surface damage based on abnormal characteristic information of the building surface, and perform graded processing to obtain structural damage assessment data; generate a health status report based on the structural damage assessment data; compare and analyze the health status report with preset historical data to obtain the development trend of structural damage; formulate a predictive maintenance plan based on the development trend of structural damage to form inspection decision support data.

[0015] The present invention also provides a computer medium storing a computer program, which implements the above-mentioned method for processing the building model when the computer program is executed.

[0016] The present invention takes the lateral displacement monitoring and prediction of tall structures as its core. By constructing a refined, multi-source fusion monitoring system and a highly reliable displacement prediction model, it comprehensively improves the state perception, dynamic response and risk prediction capabilities of tall structures under wind loads, earthquakes and environmental disturbances. First, in the data acquisition stage, the method introduces a variety of monitoring methods such as lidar, oblique photography, three-dimensional scanning and GNSS, and combines real-time collection with a periodic inspection mechanism. It not only ensures the high accuracy and full coverage of the monitoring data, but also effectively captures the deformation characteristics of tall structures at different time scales, especially the subtle displacement trends under extreme environmental conditions such as strong winds and temperature gradient changes. In terms of data preprocessing, the method integrates technical means such as coordinate adjustment, error filtering, time series reconstruction and outlier identification to standardize and robustly process the collected data. This not only improves the consistency and credibility of the monitoring data, but also provides an accurate and reliable basic data source for subsequent analysis and modeling, avoiding the influence of noise interference on the model prediction accuracy. In the process of feature extraction and dynamic change analysis, this method introduces algorithms such as wavelet analysis, multi-order difference and trend separation to perform time-frequency analysis and extract periodic patterns of structural displacement sequences. In particular, for the coexistence of low-frequency shaking and high-frequency disturbances common in tall structures, this method can accurately characterize their complex dynamic characteristics, and then identify different types of displacement behaviors caused by wind vibration, foundation settlement, and structural fatigue, providing hierarchical and multi-dimensional dynamic parameters for structural health assessment. For the prediction of lateral displacement, the method designs a multi-model fusion mechanism, integrating long short-term memory network (LSTM), support vector regression (SVR) and gray model (GM), respectively handling prediction tasks in short-term mutation, trend smoothing and data sparse scenarios. Through weighted fusion and time series calibration technology, the response speed and accuracy of the prediction model to future lateral displacement trends are improved. Especially in the process of strong wind mutation or earthquake-induced response, prediction output can be achieved at the minute level or even the second level, significantly improving the real-time and accuracy of monitoring and early warning. In the structural response identification and risk warning module, the method constructs a mapping relationship between lateral displacement, structural response, and safety level based on a multi-threshold judgment strategy and a multi-dimensional indicator system, achieving real-time identification and automatic alarm for phenomena such as excessive displacement, abnormal amplitude, and abnormal trend of tall structures. At the same time, the system also integrates structural design parameters and historical response models to customize threshold settings and risk level classification for different structural types (such as high towers, power towers, communication towers, super-high buildings, etc.), significantly enhancing adaptability and accuracy. In addition, the method also constructs a three-dimensional visualization interface and displacement evolution map, combining BIM or GIS platforms to display the real-time status, predicted trends, and risk levels of the structure in the form of dynamic layers, allowing monitoring personnel to intuitively grasp the current status of the structure and future change trends, realizing a management transition from "passive response" to "active intervention."This visualization strategy plays a key role in routine maintenance, pre-typhoon warning deployment, and emergency response to emergencies. In summary, this method not only bridges the entire technical chain from multi-source monitoring data acquisition, anomaly elimination, dynamic analysis, trend prediction, to intelligent early warning, but also enhances the operational safety assurance capabilities of tall structures. Its sensitivity, efficiency, and scalability make it widely applicable to displacement monitoring scenarios for a variety of tall structures, including urban high-rise buildings, large chimneys, transmission towers, communication towers, and wind turbine towers. It provides solid support for improving intelligent monitoring capabilities and safety management during the operational phase of structures. This method analyzes jitter characteristics across different frequency bands (low, medium, and high) and establishes a multi-level jitter compensation mechanism, effectively addressing stability issues during remote image acquisition. This technology significantly improves 3D model accuracy through high-precision registration (error controlled within 0.3 pixels) and the establishment of a structural reference coordinate system. It also establishes an objective and quantitative damage assessment system and risk index calculation method. By comparing historical data and developing an evolutionary model, it provides a scientific basis for predictive maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0018] Figure 1 Schematic diagram of the steps of the method for processing a building model of the present invention;

[0019] Figure 2 for Figure 1 Detailed step flow diagram of step S1;

[0020] Figure 3 for Figure 1 Detailed step flow chart of step S2 in FIG. DETAILED DESCRIPTION

[0021] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0022] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0023] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0024] To achieve this, please refer to Figures 1 to 3 The present invention provides a method for processing a building model, the method comprising the following steps:

[0025] Step S1: Acquire initial image data of a tall structure through an aircraft and simultaneously record motion state data; calculate jitter characteristic parameters based on the initial image data and the motion state data; and establish a jitter compensation mechanism for long-range images based on the jitter characteristic parameters;

[0026] Before conducting a tall structure health inspection, this embodiment of the present invention uses a quadrotor drone equipped with a high-resolution visible light camera and an inertial measurement unit (IMU) as the data acquisition platform. The drone ascends and descends vertically along the structure's facade at a speed of 5 meters per second and slowly orbits the structure's exterior with a radius of 3 meters. During flight, initial image data of the target building is acquired in real time. The IMU simultaneously records the drone's attitude angles (including pitch, roll, and yaw) and acceleration information. The system incorporates a timestamp mechanism into the image acquisition end to ensure synchronization between image frames and corresponding attitude data. After acquiring the image data, the system calculates the drone's jitter amplitude, jitter frequency, and main vibration direction by analyzing the displacement vector trends between key frames in the image sequence and combining the IMU data. This generates jitter characteristic parameters reflecting the stability of the flight image. Subsequently, based on these jitter characteristic parameters, a jitter compensation function is constructed between image frames to establish a long-range image jitter compensation mechanism. This mechanism corrects the image sequence using low-pass filtering to suppress high-frequency noise jitter and dynamically applies affine transformation to the jitter direction, taking into account flight trajectory and attitude changes.

[0027] Step S2: Processing the initial image data using a jitter compensation mechanism to obtain stabilized image data; extracting structural feature point information from the stabilized image data; and establishing a towering structure reference coordinate system based on the structural feature point information;

[0028] The embodiment of the present invention applies the jitter compensation mechanism obtained in step S1 to the initial image data, and performs distortion correction and repositioning on each frame of the image through a motion compensation model, thereby outputting stable image data and ensuring the continuity and consistency of the image sequence in the structural contour. Subsequently, a structural feature point extraction algorithm based on the fusion of weighted edge detection and deep learning is used to identify key nodes with geometric significance in the stable image, such as the connection nodes of the steel tower, the intersection of beams, the intersection of window frames, etc. Local contrast enhancement and illumination balance preprocessing strategies are introduced in the feature point extraction process to improve recognition robustness. After the extraction is completed, several key points located in symmetrical positions around the building are selected to construct a coordinate reference frame. Three of the points are used to construct a spatial triangle and one of the vertices is set as the coordinate origin. The other point defines the X-axis direction. The plane formed by the three points is used as the XY plane. The Z axis is defined by the vertical outward normal vector, thereby establishing a reference coordinate system for the towering structure.

[0029] Step S3: High-precision registration of the pre-acquired multi-angle image data of the tall building structure with the tall building structure reference coordinate system to obtain registered image data; constructing a three-dimensional point cloud model of the tall building structure based on the registered image data; and extracting abnormal feature information of the building surface from the three-dimensional point cloud model according to the tall building structure reference coordinate system;

[0030] After constructing a reference coordinate system for the towering structure, the present embodiment uses a dataset of images previously collected from multiple flight flights at different angles. Using an image registration algorithm based on sparse feature matching and affine projection consistency, the multi-angle images are precisely aligned to the reference coordinate system. Specifically, a modified SIFT (Scale-Invariant Feature Transform) algorithm combined with the RANSAC algorithm is used for feature matching. The registration process also incorporates spatial pose data recorded by the aircraft's IMU to improve the spatial accuracy of the registration. After image registration, the images are fed into a 3D reconstruction engine, which uses a combination of Structure from Motion (SBM) and Multi-View Stereo to generate a high-precision 3D point cloud model. The point cloud resolution is controlled to a minimum of 2,000 points per square meter to meet the requirements for building crack level analysis. Based on the established coordinate system, spatial slicing and projection analysis techniques are used to extract abnormal building surface features from the point cloud model. Different types of surface defects, such as cracks, expansion, and shedding, are identified and their specific spatial locations and dimensions are annotated.

[0031] Step S4: Analyze the type of building surface damage based on the abnormal characteristic information of the building surface, and perform graded processing to obtain structural damage assessment data; generate a health status report based on the structural damage assessment data; compare and analyze the health status report with preset historical data to obtain the structural damage development trend; formulate a predictive maintenance plan based on the structural damage development trend to form inspection decision support data.

[0032] Based on the surface anomaly feature information extracted in step S3, the present embodiment constructs a damage feature vector library. A trained convolutional neural network model is then used to classify and identify different damage types. This model uses input feature dimensions such as changes in the texture of the detached edge, the linear extension of the crack, and the trend of crack width variation to categorize abnormal areas into mild, moderate, and severe, and generates a structural damage level map. Based on this, a comprehensive assessment index system is constructed, combining the structure's design life, historical inspection data, and current damage information, to output structural damage assessment data. Furthermore, by comparing the current health status report with historical inspection data from the past three months, six months, and one year, a sliding window trend fitting algorithm is used to analyze the crack propagation rate and the frequency of new anomalies to determine the structural damage development trend. Based on these trend results and a maintenance cost model, early warning maintenance recommendations are formulated, such as recommending local reinforcement within three months or overall recoating after six months. Ultimately, inspection decision support data is generated to serve as a basis for subsequent operation and maintenance deployment.

[0033] Preferably, step S1 includes the following steps:

[0034] Step S11: acquiring initial image data of the tall structure through an image acquisition device carried by the aircraft;

[0035] In an embodiment of the present invention, an unmanned multi-rotor aircraft equipped with a 4K ultra-high-definition camera (resolution 3840×2160, frame rate 30 frames / second) is used to conduct remote inspections of urban high-rise buildings. A flight path is set before the flight mission begins. The aircraft starts from a position about 10 meters away from the building, rises from the bottom to the top along the vertical direction of the building, and completes a circle clockwise around the building, capturing image frames at an image acquisition interval of 0.5 meters. During the acquisition process, the camera is set to an exposure time of 1 / 1000 seconds, the ISO value is automatically adjusted to adapt to different lighting environments, and the image stabilization software and hardware joint working mode is enabled (the camera has built-in electronic stabilization plus the attitude data of the flight control). The image data is saved in real time to the local storage unit of the aircraft, and the frame data is set with a timestamp to ensure synchronization in subsequent processing to form an initial image data set.

[0036] Step S12: synchronously recording motion state data during the flight through the attitude sensor on the aircraft;

[0037] During the flight mission of step S11 of the embodiment of the present invention, the attitude sensor module on the aircraft works in real time. The module includes a nine-axis IMU system that integrates a three-axis gyroscope, a three-axis accelerometer, and a three-axis magnetometer. After the flight begins, the IMU records the attitude data of the aircraft at each moment at a frequency of 100 Hz, including the angular velocity and angular acceleration around the X, Y, and Z axes, the actual attitude angle (pitch angle, roll angle, and yaw angle), and the linear acceleration. Each piece of data is timestamped with the same time as the image acquisition time to ensure that the motion state data and the image data are strictly one-to-one corresponding, which is convenient for subsequent time domain fusion analysis. In addition, in order to improve data accuracy, a Kalman filter algorithm is provided in the system for dynamic filtering and correction of the attitude data, eliminating high-frequency error interference, and ensuring the stability and continuity of the motion state data.

[0038] Step S13: Compare consecutive frames in the initial image data, calculate the inter-frame displacement, and obtain image displacement data;

[0039] The embodiment of the present invention imports the initial image data obtained in step S11 into the edge computing module for image sequence processing, and uses the optical flow method (such as Lucas-Kanade sparse optical flow) to perform pixel-by-pixel motion estimation for adjacent frame images, and calculates the inter-frame displacement by detecting the displacement of feature points in the image between two frames. In the specific operation process, the image is first grayscaled and subjected to Gaussian filtering noise reduction processing, the edge texture area of the image is selected as the area of interest, the stable feature points in each frame are tracked, and their displacement vectors in the next frame are calculated, thereby constructing the displacement field between each frame. By averaging the displacement vectors of several feature points, the overall displacement amplitude and direction of the frame image are output to form image displacement data of continuous frames for subsequent dynamic jitter analysis.

[0040] Step S14: Calculating the jitter frequency and amplitude based on time domain analysis according to the image displacement data and the motion state data to obtain jitter characteristic parameters;

[0041] The embodiment of the present invention performs time alignment processing on the image displacement data extracted in step S13 and the posture sensor data obtained in step S12, performs data synchronization fusion on a unified time axis, and then conducts time domain analysis. The short-term stable interval in the image displacement sequence is extracted by sliding window technology, the periodic characteristics and amplitude changes of the image displacement changes in the interval are calculated, and the dominant jitter frequency and average jitter amplitude are extracted. The jitter frequency is estimated by analyzing the periodic fluctuations of the image displacement amplitude. The commonly used method is to extract the main frequency components after fast Fourier transforming the displacement time series. However, in order to improve the adaptability of time domain analysis, this embodiment adopts an improved autoregressive model (AR model) to directly extract the frequency distribution and corresponding amplitude changes in the time domain, thereby obtaining jitter characteristic parameters, including indicators such as main frequency band distribution, maximum jitter amplitude, average jitter amplitude and fluctuation period. These parameters will be used for the subsequent compensation mechanism construction.

[0042] Step S15: establishing jitter compensation strategies for different frequency bands according to jitter characteristic parameters;

[0043] According to the jitter characteristic parameters obtained in step S14, the embodiment of the present invention divides the image jitter effects into three frequency bands: low-frequency jitter (less than 0.5 Hz, mainly caused by slow swinging of the aircraft or ground wind interference), medium-frequency jitter (0.5 Hz to 3 Hz, commonly caused by oscillations during autonomous attitude adjustment of the aircraft), and high-frequency jitter (greater than 3 Hz, mainly caused by flight control response overshoot or body vibration). Corresponding compensation strategies are designed for each of the three frequency bands: for low-frequency jitter, an inter-frame position correction method based on motion estimation is used to reposition the image by smoothing the image frame displacement path; for medium-frequency jitter, a time-weighted average filtering technique is used to perform a weighted average of the positions of key feature points in consecutive frames to mitigate the impact of medium-amplitude mutations; and for high-frequency jitter, a dynamic rigid compensation model is established for the local area of the image, that is, the instantaneous displacement vector is estimated in each image sub-block, and the local image block is corrected by affine transformation to reduce the image blurring effect caused by high-frequency vibration, thereby achieving targeted compensation for various frequency bands.

[0044] Step S16: combining jitter compensation strategies of different frequency bands to form a jitter compensation mechanism for long-range images.

[0045] This embodiment of the present invention integrates the jitter compensation strategies designed for low, medium, and high frequency bands in step S15 to construct a composite jitter compensation mechanism for long-range images. This mechanism, based on multi-channel filtering, simultaneously feeds the input image stream into three compensation branches. Each branch independently processes the jitter characteristics of its corresponding frequency band. The processed image results are then fused and output based on frequency band priority and inter-frame stability assessment results, ultimately resulting in a stable, high-definition long-range image sequence. In application, this jitter compensation mechanism is embedded in the embedded image processing module at the image acquisition end, offering real-time processing capabilities and latency control within 200 milliseconds. It is suitable for long-range aerial photography scenarios such as urban high-rise buildings, transmission towers, and bridge pylons. The system supports automatic switching of compensation modes, dynamically adjusting filter weights based on the currently detected primary jitter frequency band, thereby ensuring high stability and availability of image data despite varying external interference.

[0046] It is particularly important that step S15 includes the following steps:

[0047] Perform frequency analysis on jitter characteristic parameters and divide them into three frequency bands based on jitter frequency: low frequency, medium frequency, and high frequency. The low frequency jitter characteristics, medium frequency jitter characteristics, and high frequency jitter characteristics are obtained. 0-2Hz is low frequency, 2-10Hz is medium frequency, and above 10Hz is high frequency.

[0048] After completing the initial extraction of jitter characteristic parameters, the embodiment of the present invention performs frequency domain conversion on the jitter displacement time series. Using an improved windowed Fourier transform method, the image displacement amplitude data is weighted with a Hanning window function, followed by a frequency domain transformation to reduce edge leakage effects, resulting in a spectrum. Based on the frequency division standard, the frequency range is divided into three segments: 0Hz to 2Hz is the low-frequency segment, representing slow disturbances such as slow aircraft tilt and wind direction changes; 2Hz to 10Hz is the mid-frequency segment, representing periodic jitter generated by routine attitude adjustments or feedback oscillations in the flight control system; and above 10Hz is the high-frequency segment, representing high-frequency noise such as propulsion device vibration and blade turbulence. Subsequently, a cluster analysis is performed on the energy peaks within different frequency bands to extract features such as the dominant frequency, average amplitude, and waveform stability within the three frequency bands. This data structure of low-frequency, mid-frequency, and high-frequency jitter features is then generated to drive the design of the frequency band compensation strategy.

[0049] Analyze the amplitude and direction characteristics of low-frequency jitter and design a low-frequency jitter compensation strategy based on overall displacement compensation;

[0050] The embodiment of the present invention focuses on the characteristics of low-frequency jitter, and mainly analyzes its amplitude change trend and displacement direction characteristics. A sequence of displacement vectors between consecutive image frames is selected, and the average displacement amplitude and direction are counted within the time window. If the direction is stable (such as the deviation is within ±15 degrees) and the displacement amplitude is significant (such as greater than the pixel displacement threshold of 2px), it is determined to be low-frequency jitter. This embodiment adopts a compensation method based on overall image translation, and the first N frames of images are translated in the opposite direction to the average direction as a whole. The translation amplitude is 95% of the average displacement amplitude to avoid image drift caused by over-compensation. During the inspection of high-rise buildings, if the image frame shows that the edge of the building presents a stable offset, the system will trigger the compensation strategy to correct the overall displacement, eliminate the problem of chronic image offset, and improve recognition accuracy.

[0051] Analyze the periodicity and regularity of IF jitter characteristics and design an IF jitter compensation strategy based on predictive compensation;

[0052] In the embodiment of the present invention, the medium-frequency jitter mainly manifests as a regular swing with obvious periodicity but small amplitude. This embodiment performs a periodic analysis on the medium-frequency jitter characteristics, and adopts the autocorrelation function method to analyze the image displacement data sequence, and extracts the jitter period length, regularity intensity and amplitude change trend therefrom. If the periodicity intensity is higher than the set threshold (such as the correlation peak is greater than 0.8), it means that the jitter is predictable. This embodiment introduces a compensation method based on a time series prediction model, adopts a first-order ARIMA model to predict the displacement trend of the next frame, and applies a reverse compensation offset to the image frame in advance. The compensation amplitude is dynamically adjusted according to the prediction error to improve the compensation accuracy. This strategy is applicable to the swaying phenomenon caused by wind interference on the aircraft, and can effectively maintain image stability in environments such as shooting around power towers.

[0053] Analyze the randomness and local characteristics of high-frequency jitter and design a high-frequency jitter compensation strategy based on local filtering;

[0054] The high-frequency jitter in the embodiment of the present invention has the characteristics of strong randomness and significant spatial locality, which often leads to local blurring of the image or unclear edges. This embodiment refines the high-frequency jitter features in the spatial dimension and divides the image into multiple small blocks (for example, each frame is divided into 8×8 areas). For each image block, its high-frequency disturbance index in several frames (such as the local pixel gradient change rate, instantaneous displacement standard deviation, etc.) is independently counted. If the high-frequency disturbance in a certain area is significantly higher than the average value of the whole frame (such as higher than twice the standard deviation), the local filtering compensation mechanism of the area is enabled. The compensation method uses bilateral filtering combined with an adaptive block alignment method to perform edge-preserving filtering on the high-frequency disturbance block, and at the same time locally calibrates the position of the block in space to align it with the adjacent area to avoid local image flickering. This strategy is suitable for remote monitoring or high-speed inspection scenarios, such as the processing of local vibration interference caused by high-speed attitude adjustment of the aircraft during wind turbine tower top structure inspection.

[0055] The compensation strength and threshold are determined according to the low-frequency jitter compensation strategy, the medium-frequency jitter compensation strategy and the high-frequency jitter compensation strategy, and the applicable conditions and triggering mechanism of the compensation strategy of each frequency band are established to obtain the jitter compensation strategy for different frequency bands.

[0056] This embodiment of the present invention establishes a multi-band compensation strategy management system, setting applicable conditions and triggering mechanisms for low-frequency, mid-frequency, and high-frequency compensation strategies to ensure that each strategy operates in the optimal scenario. Specifically, low-frequency compensation is enabled when the dominant frequency of image displacement is less than 2Hz and directional stability is greater than 85%; mid-frequency compensation is enabled when the dominant frequency is between 2Hz and 10Hz and exhibits significant periodicity; and high-frequency compensation is enabled when the high-band amplitude-energy ratio exceeds 30% of the total energy and the local disturbance area accounts for more than 10% of the image area. A decentralized fusion strategy is employed to avoid conflicts between different frequency-band strategies: for a single image frame, overall low-frequency compensation is first performed, followed by mid-frequency prediction adjustments, and finally, local high-frequency corrections. Each stage incorporates a compensation intensity threshold controller (e.g., the shift amount must not exceed 5% of the edge distance of the previous frame) to prevent image drift or distortion. The system also dynamically adjusts the trigger thresholds for each strategy based on actual compensation feedback, achieving adaptive control. This compensation mechanism is suitable for image jitter correction in a variety of complex environments, such as long-distance monitoring tasks of communication towers in mountainous areas or unmanned base stations at high altitudes, and can effectively enhance image clarity and structural stability.

[0057] Preferably, step S14 includes the following steps:

[0058] Step S141: extracting a displacement vector sequence from the image displacement data, thereby obtaining displacement change data in a two-dimensional plane;

[0059] In the embodiment of the present invention, the continuous image frames acquired by the image acquisition device are pre-processed and then subjected to inter-frame registration. The registration method adopts the pyramid-level optical flow method (Pyramidal Lucas-Kanade Optical Flow) to extract image feature points for key areas of the image (such as edges and corners) and track their position changes in adjacent frames. The displacement of each feature point between two adjacent frames is represented as a two-dimensional displacement vector, whose X-axis and Y-axis components correspond to the horizontal and vertical movements in the image coordinate system, respectively. The system extracts the image displacement vector corresponding to each frame within a time window (such as 15 frames within 1 second) to form time-serialized two-dimensional displacement data for subsequent speed and frequency analysis. In a building structure inspection scenario, if the aircraft flies around a high-rise building, the offset direction and value of the building edge in the image can clearly reflect the jitter behavior in the two-dimensional plane.

[0060] Step S142: extracting the attitude angle and acceleration information of the aircraft from the motion state data, thereby obtaining the flight attitude data of the aircraft;

[0061] The embodiment of the present invention utilizes the inertial measurement unit (IMU) carried by the aircraft to obtain attitude angle and acceleration information. The attitude angle is obtained in real time by a three-axis gyroscope to obtain the heading angle, pitch angle, and roll angle; the acceleration data is obtained by a three-axis accelerometer to obtain the acceleration components in the X, Y, and Z axes. The system begins recording IMU data after the aircraft takes off and aligns and synchronizes it with the image acquisition timestamp to ensure a one-to-one correspondence between image jitter and flight attitude. For example, when the aircraft performs lateral translation, small perturbations in the roll angle and changes in lateral acceleration will be recorded synchronously to analyze whether the image offset is caused by the aircraft's own motion. The data sampling frequency is set to 100Hz to ensure sufficient time resolution. This type of data plays a key role in analyzing the coupling relationship between attitude changes and image jitter.

[0062] Step S143: converting the displacement change data into a time series, calculating the time derivative of the displacement, and obtaining image velocity data;

[0063] The embodiment of the present invention organizes the acquired displacement vector sequence into a sequence in the order of time frames, that is, each time point corresponds to a two-dimensional displacement value. By performing a time derivative operation on the sequence, the rate of change of the displacement between adjacent frames is calculated, thereby obtaining the instantaneous velocity data of the image. The derivative calculation adopts the central difference method to calculate the average displacement difference between each frame and its previous and next frames to improve the stability and accuracy of the velocity estimation. The calculation result is a time-series two-dimensional image velocity value, which reflects the motion rate of the image in both the horizontal and vertical directions. In practice, if the displacement is an average of 2 pixels per frame and the frame rate is 15fps, the corresponding speed is about 30 pixels / second. This speed information directly reflects the image motion trend and the intensity of instantaneous changes.

[0064] Step S144: performing Fourier transform on the image velocity data and extracting its frequency components to obtain image spectrum data;

[0065] The embodiment of the present invention performs frequency domain analysis on the image velocity data sequence to identify periodic change characteristics. The processing method is fast Fourier transform (FFT), and the input is a velocity data sequence within a certain time window (for example, 1 second corresponds to 15 frames). The velocity components in the horizontal and vertical directions are independently FFTed to obtain the frequency amplitude spectrum of the velocity. In order to improve the frequency recognition accuracy, the input sequence is smoothed and windowed (such as Hamming window) before transformation to reduce spectrum leakage during the transformation process. The output spectrum contains energy intensity information at different frequency components, reflecting the dominant period and intensity of velocity changes in the image. For example, if a peak is found at a frequency of 6 Hz in the longitudinal velocity spectrum, it means that the aircraft or lens causes image jitter at a frequency of 6 times per second.

[0066] Step S145: identifying the main frequency peak according to the image spectrum data and determining the dominant jitter frequency;

[0067] After obtaining the image spectrum data, the embodiment of the present invention extracts the frequency component with the largest energy proportion in the spectrum through a frequency peak recognition algorithm. The specific method is to perform peak detection on the spectrum curve, extract the local maximum value therein, and set an energy threshold (such as a relative peak value higher than 20% of the total energy) to filter out background noise. The dominant frequency is selected from multiple peaks, that is, the frequency with the maximum energy value. The system also records the direction of the frequency (horizontal or vertical), the corresponding amplitude and its proportion in the entire spectrum for subsequent jitter amplitude estimation and compensation design. For example, if the dominant peak is found to be 7Hz in the mid-frequency band (such as 2-10Hz), the dominant jitter frequency is 7Hz, and the image corresponding to this direction will be given priority for mid-frequency compensation processing.

[0068] Step S146: Calculating the jitter amplitude data corresponding to each frequency component according to the flight attitude data and the dominant jitter frequency;

[0069] The embodiment of the present invention combines the flight attitude data of the aircraft with the dominant jitter frequency to estimate the actual image jitter amplitude at this frequency. The specific method is to use the frequency as an index to obtain the amplitude information of the frequency component in the image velocity spectrum, which represents the velocity intensity of the image at this frequency; then, combined with the acceleration value of the aircraft in the corresponding time period, the velocity integral and the attitude angle change are used to evaluate the image jitter amplitude corresponding to the frequency component. If the attitude angle frequency change trend of the aircraft within this frequency range has a high correlation with the image velocity frequency peak (such as a correlation coefficient greater than 0.7), it is further confirmed that the jitter is caused by the attitude, and its jitter amplitude in pixel units is extracted. For example, if the image velocity amplitude at a frequency of 6 Hz is 40 pixels / second, and the corresponding attitude angle swing angle of the aircraft is ±1 degree, the image jitter amplitude at this frequency is estimated to be approximately 2.5 pixels.

[0070] Step S147: establishing a jitter frequency-amplitude mapping table based on the jitter amplitude data and the dominant jitter frequency, and generating jitter characteristic parameters.

[0071] After obtaining the jitter amplitude under different frequency components, the embodiment of the present invention establishes a frequency-amplitude mapping relationship table to form a complete jitter characteristic parameter structure. The system forms a key-value pair with each identified frequency peak and its corresponding image jitter amplitude, constructs it into a table structure, and attaches a direction identifier (horizontal or vertical) and timestamp information to facilitate the rapid matching and triggering of the subsequent frequency band compensation mechanism. For example, a set of typical parameters can be expressed as: direction = vertical, frequency = 7Hz, amplitude = 2.8 pixels, time = 2.5s; this characteristic parameter table is then used to guide the design, implementation and effect evaluation of the jitter compensation strategy for each frequency band. The implementation of this step ensures the quantification, structuring and dynamic response capabilities of jitter identification, and is suitable for automated image stabilization systems.

[0072] Preferably, step S2 includes the following steps:

[0073] Step S21: applying a remote image jitter compensation mechanism to the initial image data to obtain a preliminary corrected image;

[0074] The jitter compensation mechanism for long-range images in the embodiment of the present invention is primarily implemented based on frequency domain filtering combined with a motion estimation model. First, a sequence of long-range images captured by an aircraft is extracted from the initial image data. The jitter frequency-amplitude mapping table established in the previous steps is used to identify the direction, frequency, and amplitude corresponding to the main jitter components in the image. Subsequently, a band-stop filter is used to process the image displacement time series to filter out the dominant jitter frequency components, for example, filtering the 7Hz and ±3Hz frequency bands to suppress the effects of periodic jitter. Simultaneously, a Kalman filter-based motion estimation algorithm is used to predict the true motion trajectory of the non-jitter components in the image, and the image content is realigned in the time series to generate a smooth image. The final output is a preliminary corrected image that has undergone frequency domain compensation and temporal reconstruction. In scenes captured on high-rise facades, jitter compensation can significantly straighten vertical edges and effectively remove periodic oscillation traces.

[0075] Step S22: performing low-frequency displacement compensation on the preliminary corrected image to correct the overall offset caused by aircraft drift and obtain stable image data;

[0076] The embodiment of the present invention continues to perform low-frequency displacement compensation on the basis of the preliminary corrected image to eliminate the global image offset caused by the drift of the aircraft. The method adopts the cumulative displacement trend fitting method within the time window to perform curve fitting on the average displacement trend of each key area (such as columns and building corners) in each frame of the image, and identify the low-frequency overall drift trajectory. Usually, the frequency of the trajectory change is less than 1Hz. The compensation method is to align the entire image in the X or Y direction by affine transformation, so that the image sequence remains stable and does not offset at the macro level. In actual applications, when the aircraft performs a smooth lateral movement to inspect high-rise curtain walls, if a slow left shift is observed in the image content, the system will perform low-frequency displacement smoothing and correction operations within 3-5 frames, thereby outputting stable image data, so that the subsequent structural feature extraction has spatial consistency.

[0077] Step S23: performing edge enhancement processing on the stabilized image data to improve the clarity of the structure outline and obtain a building enhanced image;

[0078] In order to improve the accuracy of subsequent structure recognition, the embodiment of the present invention performs edge enhancement processing on the stable image data and strengthens the geometric contour information of the building structure. The processing method adopts multi-scale Laplace operator enhancement combined with image brightness histogram equalization. First, the background illumination effect is suppressed by image grayscale normalization, and then the gradient mutation area in the image is enhanced by Laplace filtering to strengthen the linear structure contours such as the edges of columns, building frames, and window frames. In order to avoid the noise amplification effect caused by enhancement, the enhanced image is then subjected to edge-preserving noise reduction processing through bilateral filtering. This processing is suitable for front view images of high-rise buildings, especially in scenes where the structural contours are slightly blurred by fog or dust, the contour contrast can still be significantly improved, and the output enhanced image of the building is more conducive to the precise positioning of feature points.

[0079] Step S24: using the building enhanced image to perform high contrast area detection based on gradient analysis to obtain candidate locations of structural feature points;

[0080] Embodiments of the present invention This embodiment uses building enhanced images to detect high-contrast areas to obtain possible candidate locations of structural feature points. The method uses the Sobel operator to calculate the image gradient map, extracts the areas with large gradient amplitude values in the image, and then combines non-maximum suppression and threshold segmentation strategies to retain only the edge areas with significant gradient changes. Subsequently, these high-gradient areas are divided into image blocks (such as 8×8 pixels), and their local contrast levels and texture density indicators are evaluated, and the centers of image blocks with good texture changes are screened out as candidate locations of feature points. For example, in the junction area of high-rise building windows, obvious grayscale contrast and boundary line aggregation often appear. The system determines this area as a candidate structural feature area, laying the foundation for subsequent corner point screening.

[0081] Step S25: performing corner detection on the candidate positions of the structural feature points and screening the feature points with high stability, thereby obtaining the structural feature point information;

[0082] Embodiments of the Invention This embodiment uses the Harris corner detection algorithm to detect corners based on the aforementioned candidate positions, and screens them based on the corner response values and temporal stability. First, the local grayscale matrix changes of the image are calculated around the candidate positions, and the corner positions with large response intensity are extracted. Then, the position changes of these corner points in the image sequence are tracked through multiple frames of images; if a corner point is identified in consecutive frames and the position change is less than the set threshold (such as within 2 pixels), the corner point is considered stable. The filtered structural feature point information includes parameters such as its image coordinates, response intensity, and stability score. In the detection of high-rise curtain wall structures, window corners and beam-column intersections are often stable feature points. Extracting this information can be used to establish a structural contour model.

[0083] Step S26: analyzing the spatial distribution of structural feature point information and identifying key structural nodes;

[0084] After acquiring multiple structural feature points, the embodiment of the present invention identifies key structural nodes by analyzing their spatial distribution patterns in the image. The method adopts a strategy that combines cluster analysis with geometric structure fitting. First, spatially adjacent feature points are grouped based on a density clustering algorithm (such as DBSCAN), and then a straight line or rectangular structure is fitted to the feature points in each group. The intersections or endpoints with high fitting accuracy are identified as key structural nodes. This step is manifested in the front elevation image of a high-rise building as being able to accurately identify structural skeleton features such as window panes, column intersections, and floor boundaries. In a typical office building scene, the system can identify more than 20 key nodes distributed at regular intervals, providing a reference basis for the construction of a three-dimensional coordinate system.

[0085] Step S27: Using the key structural nodes as reference points, a three-dimensional rectangular coordinate system is established to form a reference coordinate system for the towering structure.

[0086] The embodiment of the present invention establishes a three-dimensional rectangular coordinate system for the building structure with reference to the identified key structural nodes. First, a stable node located at the center of the bottom of the building body in the image is selected as the origin node, the Z axis is defined by the direction of the vertical structural line approximately perpendicular to the Y axis of the image, the X axis is defined by the horizontal structural line of the floor, and the direction of the Y axis is determined by the right-hand rule to complete the setting of the coordinate axis direction. Subsequently, the positions of other key nodes in the image are calibrated relative to the coordinate system, and the pixel coordinates are mapped to the actual physical coordinates. The scale is normalized and corrected by the known building structure dimensions (such as floor height) to form a physically meaningful towering structure reference coordinate system. In urban high-rise inspection tasks, this coordinate system can be used as a unified reference framework between multi-source sensor data, image data and structural analysis models.

[0087] Preferably, performing high-precision registration of the pre-acquired multi-angle image data of the towering structure with the towering structure reference coordinate system in step S3 includes:

[0088] Acquire images of tall structures taken from different angles at intervals of 60-120 degrees to obtain multi-angle image data;

[0089] Perform feature point extraction based on Harris corner detection on multi-angle image data to obtain multi-angle image feature points;

[0090] Calculate the 128-dimensional descriptor vector of the multi-angle image feature points and set the matching threshold to 0.75 to obtain the feature point descriptor vector;

[0091] The structural feature points in the towering structure reference coordinate system are projected onto each viewing plane with an accuracy of ±0.5 pixels to form a standard corresponding point set;

[0092] The feature point descriptor vector is used to match the standard corresponding point mapping table, and the Euclidean distance threshold of 80 is used as the screening condition to establish the initial feature point correspondence relationship set;

[0093] The initial feature point correspondence set is iteratively optimized, the error tolerance threshold is set to 2.5 pixels, and abnormal matching points exceeding 15% of the total number of matches are eliminated to obtain a reliable feature point correspondence matrix;

[0094] Based on the reliable feature point correspondence matrix, a homography transformation matrix with an accuracy error of less than 0.8 pixels is calculated and applied to multi-angle image data for perspective correction, generating a spatially consistent image set with an edge error within 1.2 pixels.

[0095] The bicubic interpolation sub-pixel deformation compensation based on 16×16 pixel grid density is implemented on the spatially consistent image set to obtain the registered image data with a maximum error of less than 0.3 pixels.

[0096] In one embodiment of the present invention, a drone is used to fly around tall structures such as communication towers, television towers, or high-rise buildings from the ground. Image sequences are captured from at least three angles, ensuring that adjacent shooting directions form an angle of 60° to 120°. In specific operations, the flight mission planning system sets polygonal path points with a radius between 30 and 50 meters. Each path point controls the camera to focus on the center of the building structure to complete image acquisition. To ensure image clarity and angular accuracy, the flight altitude is controlled between 80% and 100% of the target structure's height, and the pan / tilt angle is controlled within 10° to ensure good image overlap across all viewing angles. This ultimately creates a multi-angle image dataset suitable for subsequent spatial registration and modeling. Feature point extraction is performed on the acquired multi-angle image data using the standard Harris corner detection method. First, the color image is converted to grayscale and pre-processed with Gaussian blur to reduce the influence of image noise. Next, the image gradient and local grayscale variation covariance matrix are calculated for each image. Based on the corner response function, several regions with the highest response intensity are selected as feature point locations. To avoid overcrowding or duplication of feature points, this method also performs non-maximum suppression and edge suppression on the extracted results. In typical applications, 300 to 500 highly responsive corner points can be extracted from each image, covering structurally significant areas such as window corners and frame intersections, thus laying the foundation for subsequent feature matching. After extracting multi-angle image feature points, the SIFT (Scale-Invariant Feature Transform) algorithm is used to construct a 128-dimensional feature descriptor for each corner point. In the specific implementation, with each Harris corner point as the center, the image block of its neighborhood area is extracted, and a multi-scale gradient direction histogram is constructed. The gradient directions are normalized and spliced to form a high-dimensional vector. In the descriptor matching stage, the matching ratio threshold is set to 0.75, that is, for any feature point, the Euclidean distance ratio between the best match and the second best match is less than 0.75 to be considered a valid match. This strategy can effectively filter out background interference features and improve matching accuracy. This step is suitable for feature preservation analysis when there are inconsistent elevation angles and viewing angles between images. After the establishment of the reference coordinate system of the towering structure is completed, this embodiment projects the position of the structural feature points in the coordinate system through the known camera external parameters and internal parameters, and projects the three-dimensional coordinate points to the image planes of each perspective. In order to control the geometric projection error, the structural feature points are mapped to the image coordinate system with an accuracy of ±0.5 pixels through the perspective transformation relationship based on the pinhole camera model as a standard corresponding point set. In this process, camera parameters such as focal length, principal point position and distortion parameters are set based on pre-flight calibration data to ensure mapping accuracy. In typical application scenarios, such as images taken from the north and south sides of a high-rise office building, the pixel coordinates of the corner points of the window frames on the same floor in each perspective image will be uniformly projected to provide a spatial reference for subsequent feature matching.The calculated image feature point descriptors are matched against the reference coordinate system mapping points, using Euclidean distance as the matching metric. The distance relationship between the descriptors and the standard point descriptors in each image is traversed. If the Euclidean distance is less than a set threshold of 80, the image feature point is considered to have a matching relationship with the reference point and is added to the initial feature point correspondence set. To improve efficiency, a KD-Tree is used to construct a feature space index and perform nearest neighbor search matching. In building facade images, consistent feature descriptors such as window corners and balcony edges often appear. This allows for the successful establishment of initial matching relationships across multiple angles, facilitating subsequent error removal. To improve matching quality, an iterative optimization of the initial feature point correspondences is performed based on the RANSAC (Random Sample Consensus) algorithm. The specific method involves fitting an initial homography model with a small number of randomly selected point pairs over multiple rounds. The geometric reprojection error of all matching point pairs is then calculated and determined to be less than a tolerance threshold of 2.5 pixels. If greater, the match is considered an outlier. The optimal result is the one with the largest number of inliers across all models. Matches with more than 15% outliers are then removed. This process forms an accurate and stable structural feature point correspondence matrix. In practice, for the front and side views of a five-story office building, over 80% of stable matching points can be effectively retained, minimizing mismatches. Based on the reliable feature point correspondence matrix, a least-squares fitting method is used to solve the homography transformation matrix, which is used to align the multi-angle images to a unified reference perspective. The transformation matrix is solved by reconstructing the projective geometric relationship between the corresponding points. A maximum fitting error limit of 0.8 pixels is set, meaning that the maximum deviation of the final transformation on the image plane should be within this range. The correction process involves projecting the source image onto the reference image coordinate system and performing edge padding to ensure overall structural alignment. Typically, after transforming the images from different angles, the alignment error of the building boundaries in the composite image does not exceed 1.2 pixels, meeting the requirements of subsequent high-precision applications such as stitching and depth inference. To further improve pixel-level registration accuracy between images, this embodiment uses a 16×16 pixel grid after perspective correction and performs nonlinear sub-pixel distortion compensation using bicubic interpolation. The method fits the image deformation vector field at each grid point, calculates the local deformation direction and magnitude, performs fine-grained image resampling, and uses a bicubic interpolation algorithm to achieve sub-pixel mapping, with particular optimization adjustments for edge regions and areas with complex textures. This process achieves sub-pixel alignment of the images, with a maximum residual error within 0.3 pixels, making it suitable for subsequent image fusion, 3D reconstruction, or high-precision damage identification.

[0097] Preferably, constructing a three-dimensional point cloud model of the towering structure based on the registered image data in step S3 includes:

[0098] Identify the same-name points in the overlapping areas between adjacent images on the registered image data to obtain the spatial coordinates of the feature points;

[0099] The spatial coordinates of the feature points are used to build a preliminary sparse point cloud, and the validity of the spatial points is verified by viewing angles greater than 15° to obtain a skeleton point cloud;

[0100] Determine the main plane and outline of the tall structure based on the skeleton point cloud, and divide the tall structure into different component areas to form a structural area index;

[0101] Dense matching is performed on each region in the structure region index to generate a gridded dense point cloud, where the search window for dense matching is set to 7×7 pixels and the matching step size is 2 pixels;

[0102] The gridded dense point cloud is merged according to the structural area index and local smoothing is performed based on radius query. The search radius is set to 0.05m, and points that deviate from the mean by 3 times the standard deviation are removed to obtain the filtered point cloud.

[0103] Based on the color information of the filtered point cloud and the corresponding registered image, an RGB value is assigned to each point to generate a three-dimensional point cloud model of the towering structure.

[0104] After obtaining multi-angle image data of a towering structure that has undergone perspective correction and sub-pixel registration processing, the embodiment of the present invention first analyzes the overlapping areas between adjacent image pairs of the registered image data. The image registration matrix is used to calculate the boundaries of the overlapping areas of view under different view angles. The scale-invariant feature transform (SIFT) algorithm is used to extract image feature points in the overlapping area. Combined with the pose estimation results, preliminary matching is performed using feature descriptors. Then, the initial matching points are eliminated using epipolar geometry constraints to screen out pairs of homonymous points that meet geometric consistency. These homonymous points are triangulated based on known camera intrinsic and extrinsic parameters, and the coordinates of each group of homonymous points in three-dimensional space are calculated to obtain a set of spatial feature points in the overlapping areas of adjacent images. The obtained set of spatial feature points is used as input, and a preliminary sparse point cloud model is constructed using the principle of structured beam triangulation. In this sparse point cloud, each spatial point comes from the matching results of homonymous points from two view angles. To improve the structural reliability of the point cloud, the angle between the image pairs associated with each spatial point is analyzed, and the angle between the optical axes of the two views is calculated. If the angle is greater than 15°, the point is considered a valid spatial point with a good triangulation basis; otherwise, the point is marked as a low-confidence point and discarded. The resulting point set, filtered in this way, becomes a skeleton point cloud, which has high 3D geometric accuracy and sparse coverage, making it suitable for subsequent structural analysis and detailed modeling. Within the constructed skeleton point cloud, a Random Sample Consensus Algorithm (RANSAC) algorithm is first used to identify large coplanar regions to identify the main planes of the tall structure, such as walls, towers, and platforms. Within the identified main plane boundaries, the structural outline is extracted using edge point clustering and projection analysis. Spatial connectivity is then used to partition the entire point cloud, dividing the structure into several subcomponents based on geometric properties, such as the base, midsection, platform, and top. Each subcomponent region is assigned a unique number to form a structural region index, which provides a logical framework for subsequent local dense reconstruction. Dense stereo matching is performed on the image regions within each component in the structural region index. A block-matching-based multi-view stereo reconstruction algorithm was employed, with a search window size of 7×7 pixels. Each pixel in the reference image corresponded to a 7×7 pixel window, and sliding matching was performed in the target image along the disparity direction. The matching step size was set to 2 pixels, balancing computational efficiency and density distribution. During the matching process, the normalized cross correlation (NCC) was used as a similarity metric. The matching positions with the highest similarity were selected and their 3D coordinates were recorded. After all image pairs were matched, a gridded dense point cloud corresponding to the structural region was accumulated. The gridded dense point clouds of all structural regions were merged. First, all local point clouds were spliced in a unified coordinate system based on the structural region index, and then local smoothing filtering was performed in 3D Euclidean space.A spherical search region with a radius of 0.05 meters is constructed around each point. The mean and standard deviation of the 3D coordinates of all neighboring points within this region are calculated. Outliers exceeding three standard deviations from the mean are removed to reduce noise. This processing effectively improves the overall continuity and surface smoothness of the point cloud, ultimately outputting filtered, high-precision point cloud data. After obtaining the filtered point cloud, the color information of the original image data is combined for 3D visualization and colorization. For each spatial point in the point cloud, its projected coordinates in the image plane are calculated based on its image position and the projection matrix of the corresponding viewpoint. The corresponding RGB values are extracted from the registered image using bilinear interpolation. If a point is visible from multiple viewpoints, its color values are weighted averaged, with the weight determined by the viewing angle and projection clarity. The resulting 3D point cloud exhibits high geometric accuracy and realistic surface texture, enabling complete 3D modeling of tall structures. It is suitable for a variety of applications, including structural monitoring, digital twins, and post-disaster assessments.

[0105] Preferably, extracting structural surface abnormal feature information from the three-dimensional point cloud model according to the towering structure reference coordinate system in step S3 includes:

[0106] The 3D point cloud model is transformed and aligned based on the towering structure reference coordinate system to obtain a standardized 3D point cloud in the reference system;

[0107] Based on the association between the towering structure reference coordinate system and the existing structural area index, the precise position mapping of each structural component in the reference coordinate system is established to form a reference system component mapping table;

[0108] Calculate the surface deviation distribution of each component in the reference system component mapping table in the towering structure reference coordinate system, identify areas with deviations exceeding 5mm, and mark them as potential abnormal areas in the reference system;

[0109] Perform spatial cluster analysis on potential abnormal areas to form abnormal characteristic patch data;

[0110] The geometric properties are calculated based on the abnormal feature patch data and the reference coordinate system of the towering structure to generate abnormal feature information on the building surface.

[0111] In order to make the three-dimensional point cloud model consistent with the position and direction of the actual high-rise structure, the embodiment of the present invention needs to perform a coordinate transformation operation on the point cloud model, with the goal of expressing the point cloud as standardized spatial data in the reference coordinate system of the high-rise structure. The specific operation is: select feature points with known positions in the actual high-rise structure (such as the center of the tower bottom, the corners of the platform, etc.) as anchor points, manually mark or automatically identify the positions corresponding to these feature points in the three-dimensional point cloud, use the quaternion method to calculate the rotation relationship between the point cloud coordinate system and the reference coordinate system, and establish a translation vector based on the absolute position of the anchor point, and finally form a complete rigid body transformation matrix. Apply this transformation to all point cloud data to complete the alignment of the point cloud model in the reference coordinate system. This process needs to ensure that the point cloud as a whole is consistent with the reference coordinate system after the transformation, for example, the axis of the tower is aligned with the Z axis, and the bottom center point is at the origin. Based on the three-dimensional point cloud that has completed the reference coordinate alignment, combined with the structural area index data established in the previous step, the point cloud part of each structural component is matched with the reference coordinate system. The specific method is: for each structural component point cloud subset, extract the spatial bounding box parameters of its boundary points in the reference coordinate system (i.e., the center coordinates, size, and orientation of the minimum circumscribed cuboid), and record its number in the index table. Then construct a component mapping table in the reference coordinate system. Each record in the table includes the component number, position, orientation, and circumscribed boundary body information in the reference system. This mapping table will serve as the structural skeleton for the digital twin modeling of high-rise structures, which will help with subsequent local analysis and deformation tracking. For each structural component area in the established reference system component mapping table, calculate the spatial deviation between its point cloud surface and the ideal structural surface. The ideal structural surface is imported from the design drawing or automatically constructed using a fitting algorithm (such as plane, cylinder, and cubic fitting). For each point cloud point, calculate its distance to the fitting surface and statistically analyze its deviation distribution. In actual engineering, 5 mm is used as the judgment threshold. If there are several points in a local area with an absolute value of deviation exceeding the threshold, it is marked as a potential abnormal area. In the specific implementation, a sliding window mechanism (such as dividing into 10×10 pixel equivalent grids) can be used to calculate the mean and variance of the deviation in the local area, and automatically mark the deviation clustering area. Spatial clustering analysis is performed on all marked potential anomalies to identify continuous areas of possible structural deformation or defects. Specifically, a density-based spatial clustering method (such as the DBSCAN algorithm) is used, the minimum point threshold is set to 30, and the neighborhood radius is set to 0.1 meters. The connectivity between the deviation points in space is converted into clustering conditions, and all anomaly points are aggregated to form multiple spatially discrete anomaly patches. Each anomaly patch contains a boundary range, a center position and its spatial morphological characteristics, and its component affiliation in the structural area index is marked, providing data support for subsequent fault assessment or structural operation and maintenance. For each anomaly patch, its geometric properties in the reference coordinate system of the towering structure are further calculated.Specifically, it includes the maximum linear size, area, volume (if the patch has a certain thickness), main direction angle (the angle between the main axis of the patch and the main axis of the structure), position height (Z-axis value), etc. of the abnormal patch. By combining these attributes, a data table of abnormal characteristic information on the building surface is formed. For example, in the inspection task of a certain transmission tower, the system detected a bulge abnormal area with a maximum deviation of 12 mm, a main direction angle of 9 degrees, and an area of approximately 0.15 square meters on the steel structure panel located 25 meters above the ground. The detection time can be combined with a timestamp to form a structural status database for long-term monitoring and structural health assessment.

[0112] Preferably, step S4 includes the following steps:

[0113] Step S41: determining the type of building surface damage based on the geometric form and distribution characteristics of the building abnormality feature information;

[0114] After obtaining the abnormal feature information of the building surface, the embodiment of the present invention needs to judge the damage type of the abnormal patch based on its geometric shape (such as aspect ratio, concave-convex trend, surface normal change) and spatial distribution characteristics (such as distribution at the edge, middle or node connection area). In the specific implementation, the system presets a set of typical building damage feature template libraries, including shedding (sharp boundaries and large missing depth), bulging (sudden change in normal direction in a continuous area but no obvious volume loss), cracks (linear defects with high aspect ratio and width less than 1 cm), corrosion (surface color change accompanied by slight depression), etc., and inputs the geometric and texture feature vectors of the abnormal patch through a classification model based on random forest or support vector machine for type identification. For example, if the aspect ratio of an abnormal area is greater than 5, the thickness is less than 3 mm, and it is distributed near the steel beam node, the model can determine it as crack-type damage.

[0115] Step S42: Calculate the number, area, and distribution density of damage based on the building surface damage type and building abnormality feature information to form quantitative index data;

[0116] After identifying each type of damage, the embodiment of the present invention performs quantitative statistics, area measurement, and density calculation on each type of damage. Among them, the area is estimated by projecting the area of the point cloud after triangulation, and the unit is square meters; the quantity refers to the number of independent patches of each type of damage; and the density is the proportion of the damage area per unit structural area (such as per square meter). Taking a 30-meter-high transmission tower as an example, a total of 12 bulge-type abnormal areas were detected, with an average area of 0.035 square meters, distributed in the area between 10 meters and 18 meters in the middle of the tower body. A density analysis of the total area of 20 square meters within this range was performed, and it was found that the bulge density was 2.1%. The final output forms a data table containing quantitative indicators such as the number, total area, and density of each type of damage.

[0117] Step S43: performing damage classification processing on the quantitative index data to obtain structural damage assessment data;

[0118] In order to accurately assess the severity of damage, the embodiment of the present invention classifies the above-mentioned quantitative indicator data into damage levels. This process is based on a preset grading standard library, and the level thresholds are set according to the area size, density value and damage type. For example, if the length of a crack exceeds 1 meter, it is classified as medium or above, and if the area exceeds 0.05 square meters, it is classified as a severe level; if the density of a bulge exceeds 3%, it is classified as a key monitoring level. A rule engine or decision tree is used to automatically match the indicator data of each type of damage, and divide it into three categories: mild, medium and severe, to form a structural damage grading result. This result is combined with the component index to generate a structural damage assessment data set, in which each structural component is labeled with the corresponding damage level.

[0119] Step S44: classifying the structural damage assessment data based on location, type, and severity to generate a structural damage distribution map;

[0120] After obtaining the structural damage assessment data, the embodiment of the present invention combines its location coordinates, damage type and level information to perform spatial visualization classification. The specific method is to construct a three-dimensional building structure model, assign corresponding color codes to each damaged area in the model (such as green for mild, yellow for moderate, and red for severe), and draw them separately according to component categories to generate a structural damage distribution map. The graph supports three-dimensional rotation browsing and layered viewing, which can be used for intuitive reference when making structural maintenance decisions. For example, in the inspection results of a certain tall tower, there are three severe corrosions at the base of the tower, which are displayed as red clustered patches, and the top of the tower is a medium crack. The distribution map reflects the different damage distribution trends at the top and bottom of the structure.

[0121] Step S45: Calculate the key risk index for the structural damage distribution map, and generate a risk heat map based on the structural importance and damage severity of the damage location;

[0122] The embodiment of the present invention calculates a key risk index for each damaged area based on the structural damage distribution map to evaluate its potential impact on the overall structural safety. The risk index comprehensively considers three dimensions: position weight (such as the main stress area is assigned a higher value), damage type risk coefficient (such as cracks are higher than bulges), and damage level coefficient (the most severe level). The product of the three is the risk value of the point. A risk heat map is then generated in the three-dimensional model, with high-risk areas highlighted in red and medium- and low-risk areas highlighted in orange, yellow, and green, respectively. Taking a certain communication tower as an example, the crack located at the intersection node in the middle of the tower body is assigned a high position weight and type coefficient, and the final risk value reaches 0.87 (the full score is 1), which appears as a red cluster area in the heat map.

[0123] Step S46: Based on the risk heat map and structural damage assessment data, damage statistics, risk level distribution and overall safety integration assessment are performed to generate a health status report;

[0124] The embodiment of the present invention combines the obtained risk heat map with the structural damage assessment data, and the system conducts multi-dimensional statistical analysis. First, the number and spatial distribution of each damage type and level are counted, and then the frequency of risk levels is summarized according to the affiliation of structural components to generate a structural risk level distribution table. At the same time, combined with the overall force model of the structure and the monitoring data, it is determined whether the risk patches are concentrated in the key load-bearing components or safety control areas, and the overall health status is comprehensively evaluated, and a report containing "damage type analysis", "high-risk area warning", "health level assessment" and other contents is output. For example, the report points out that the structure as a whole is in a sub-healthy state, mainly due to the severe corrosion of the tower base, which causes the risk level of this area to be "high", and repair actions need to be taken within the next 30 days.

[0125] Step S47: Compare and analyze the health status report with the preset historical data to obtain the structural damage development trend; formulate a predictive maintenance plan based on the structural damage development trend to form inspection decision support data.

[0126] In order to track and analyze the changing trends of structural health, the embodiment of the present invention compares the currently generated health status report with historical inspection data. The analysis objects include the increase or decrease in the number of damages, the changing trends of damage levels, and the changes in the scope of risk areas. For example, by comparing the data from three months ago with the current data, it is found that the cracks in a certain structural component have developed from a medium level to a severe level, and the length has increased by about 25%. The system then marks it as "developmental damage." The possibility of such damage developing in the next three months is then predicted based on the time series analysis model, and a predictive maintenance plan is formulated accordingly, such as "It is recommended to conduct ultrasonic crack depth detection and reinforcement of area X of the tower within the next 60 days." This ultimately forms an inspection decision support data set that includes future inspection priorities, maintenance recommendations, and key inspection areas.

[0127] It is particularly important that step S47 includes the following steps:

[0128] Step S471: extracting historical damage records related to the current tall structure from preset historical data, and performing comparative analysis with the data in the health status report to obtain damage comparison data;

[0129] Before performing structural damage trend analysis, the system first extracts historical records related to the current inspection target from a pre-set historical dataset. This historical data typically includes structural component number, damage type, inspection time, damage area, length, depth, and density. The extraction logic is based on the structure ID, component location code, and a time range matching strategy. For example, if a transmission tower numbered T-205 is currently inspected in 2025, the system automatically selects all inspection records for that number from the past five years and matches each of these records with the same component recorded in the current health status report. The system then compares damage changes for the same component or location in the historical and current data. For example, the 2023 inspection record indicates two cracks in the middle of the tower, with an average length of 0.75 meters. However, in the 2025 data, the cracks in this area have grown to 1.2 meters, and the number of cracks has increased to three. Based on this comparison, the system outputs a damage comparison data table containing "newly added damage," "area growth rate," and "density change" as input data for trend analysis.

[0130] Step S472: Calculate the change rate and development speed of different damage types based on the damage comparison data to obtain the structural damage development trend;

[0131] The system of the embodiment of the present invention calculates the development rate of each type of damage (such as cracks, bulges, corrosion, etc.) within a specified time period based on the damage comparison data generated in step S471. The calculation method is: the key geometric parameters such as area and length are divided by the change amount according to the time difference to obtain the average annual growth value; and then the growth ratio is expressed as the ratio of the current value to the historical value. For example, the length of the crack in the middle section of the tower body increased from 0.75 meters to 1.2 meters in two years, and the annual growth rate is 0.225 meters / year, and the growth rate is 60%. Similarly, the area of the corrosion area increased from 0.02 square meters to 0.06 square meters, with an annual increase of 0.02 square meters / year. The system processes the growth rate and development speed data of all damage types in a centralized manner to form a structural damage development trend chart, which supports filtering and viewing by dimensions such as type, time, and location for subsequent priority evaluation.

[0132] Step S473: Calculating the maintenance priority index of each damage point based on the structural damage development trend;

[0133] After obtaining the development trend of various types of structural damage, the embodiment of the present invention evaluates the urgency of each damaged area by calculating the maintenance priority index. The index comprehensively considers three main factors: damage severity (such as area size and level), development speed (such as growth rate) and location importance (such as whether it is located in the main load-bearing component or key connection node). Each factor is assigned a weight and then weighted summed to form a maintenance priority index. For example, a crack located at the tower top support node has a current length of 1.5 meters, an annual growth rate of 0.4 meters / year, and a damage level of severe. The location weight of the point is 0.4, the development speed weight is 0.35, and the damage level weight is 0.25. The three are multiplied by their respective scores and summarized to obtain a total priority index of 0.86 (full score 1). Those above the set threshold of 0.75 will be automatically listed as priority processing objects. The system outputs a list of structural components and a sorted list with the priority index of each damage point as the basis for maintenance planning.

[0134] Step S474: Formulate a phased structural maintenance plan based on the maintenance priority index to form inspection decision support data, wherein the structural maintenance plan includes maintenance time nodes, maintenance methods and resource requirements.

[0135] The system in this embodiment of the present invention sorts all damaged points in descending order based on their maintenance priority index and then develops a phased structural maintenance plan based on available maintenance resources and inspection cycles. This plan includes three components: maintenance timelines (i.e., when to repair or review the point), maintenance methods (such as crack sealing grouting, localized steel plate reinforcement, rust removal and anti-corrosion treatment), and resource requirements (required manpower, materials, machinery, and time budget). For example, a high-priority crack in the middle of a transmission tower is scheduled for maintenance within the next 30 days using carbon fiber patch reinforcement. This requires two repair personnel, two meters of carbon fiber cloth, and one set of quick-drying resin, and takes approximately four hours. Mildly corroded areas at the tower base are scheduled for inspection and review in 120 days, with earlier inspections if the corrosion trend accelerates. The resulting inspection decision-support data is presented in a structured table and calendar format, facilitating the engineering department's development of annual maintenance plans and resource allocation strategies.

[0136] The present invention further provides a building model processing system for executing the above-mentioned building model processing method, wherein the building model processing system comprises:

[0137] The image acquisition and jitter compensation module is used to acquire initial image data of the tall structure through the aircraft and simultaneously record the motion status data; calculate the jitter characteristic parameters based on the initial image data and the motion status data; and establish a jitter compensation mechanism for long-range images based on the jitter characteristic parameters;

[0138] The image stabilization and coordinate system establishment module is used to process the initial image data using a jitter compensation mechanism to obtain stabilized image data; extract structural feature point information from the stabilized image data; and establish a towering structure reference coordinate system based on the structural feature point information;

[0139] The 3D modeling and anomaly detection module is used to perform high-precision registration of pre-acquired multi-angle image data of the towering structure with the towering structure reference coordinate system to obtain registered image data; construct a 3D point cloud model of the towering structure based on the registered image data; and extract abnormal feature information of the building surface from the 3D point cloud model according to the towering structure reference coordinate system;

[0140] The damage assessment and maintenance decision module is used to analyze the type of building surface damage based on abnormal characteristic information of the building surface, and perform graded processing to obtain structural damage assessment data; generate a health status report based on the structural damage assessment data; compare and analyze the health status report with preset historical data to obtain the development trend of structural damage; formulate a predictive maintenance plan based on the development trend of structural damage to form inspection decision support data.

[0141] The present invention also provides a computer medium storing a computer program, which implements the above-mentioned method for processing the building model when the computer program is executed.

[0142] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0143] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A method for processing a building model, characterized in that: The following steps are involved: Step S1: Acquire initial image data of a tall structure through an aircraft and simultaneously record motion state data; calculate jitter characteristic parameters based on the initial image data and the motion state data; and establish a jitter compensation mechanism for long-range images based on the jitter characteristic parameters; Step S2: Processing the initial image data using a jitter compensation mechanism to obtain stabilized image data; extracting structural feature point information from the stabilized image data; and establishing a towering structure reference coordinate system based on the structural feature point information; Step S3: High-precision registration of the pre-acquired multi-angle image data of the tall building structure with the tall building structure reference coordinate system to obtain registered image data; constructing a three-dimensional point cloud model of the tall building structure based on the registered image data; and extracting abnormal feature information of the building surface from the three-dimensional point cloud model according to the tall building structure reference coordinate system; Step S4: Analyze the building surface damage type based on the building surface abnormality feature information and perform classification processing to obtain structural damage assessment data; generate a health status report based on the structural damage assessment data; compare and analyze the health status report with preset historical data to obtain the structural damage development trend; Develop predictive maintenance plans based on structural damage development trends and generate inspection decision support data.

2. The method for processing a building model according to claim 1, wherein: Step S1 includes the following steps: Step S11: acquiring initial image data of the tall structure through an image acquisition device carried by the aircraft; Step S12: synchronously recording motion state data during the flight through the attitude sensor on the aircraft; Step S13: Compare consecutive frames in the initial image data, calculate the inter-frame displacement, and obtain image displacement data; Step S14: Calculating the jitter frequency and amplitude based on time domain analysis according to the image displacement data and the motion state data to obtain jitter characteristic parameters; Step S15: establishing jitter compensation strategies for different frequency bands according to jitter characteristic parameters; Step S16: combining jitter compensation strategies of different frequency bands to form a jitter compensation mechanism for long-range images.

3. The method for processing a building model according to claim 2, wherein: Step S14 includes the following steps: Step S141: extracting a displacement vector sequence from the image displacement data, thereby obtaining displacement change data in a two-dimensional plane; Step S142: extracting the attitude angle and acceleration information of the aircraft from the motion state data, thereby obtaining the flight attitude data of the aircraft; Step S143: converting the displacement change data into a time series, calculating the time derivative of the displacement, and obtaining image velocity data; Step S144: performing Fourier transform on the image velocity data and extracting its frequency components to obtain image spectrum data; Step S145: identifying the main frequency peak according to the image spectrum data and determining the dominant jitter frequency; Step S146: Calculating the jitter amplitude data corresponding to each frequency component according to the flight attitude data and the dominant jitter frequency; Step S147: establishing a jitter frequency-amplitude mapping table based on the jitter amplitude data and the dominant jitter frequency, and generating jitter characteristic parameters.

4. The method for processing a building model according to claim 3, wherein: Step S2 includes the following steps: Step S21: applying a remote image jitter compensation mechanism to the initial image data to obtain a preliminary corrected image; Step S22: performing low-frequency displacement compensation on the preliminary corrected image to correct the overall offset caused by aircraft drift and obtain stable image data; Step S23: performing edge enhancement processing on the stabilized image data to improve the clarity of the structure outline and obtain a building enhanced image; Step S24: using the building enhanced image to perform high contrast area detection based on gradient analysis to obtain candidate locations of structural feature points; Step S25: performing corner detection on the candidate positions of the structural feature points and screening the feature points with high stability, thereby obtaining the structural feature point information; Step S26: analyzing the spatial distribution of structural feature point information and identifying key structural nodes; Step S27: Using the key structural nodes as reference points, a three-dimensional rectangular coordinate system is established to form a reference coordinate system for the towering structure.

5. The method for processing a building model according to claim 4, wherein: The step S3 of high-precision registration of the pre-acquired multi-angle image data of the towering structure with the towering structure reference coordinate system includes: Acquire images of tall structures taken from different angles at intervals of 60-120 degrees to obtain multi-angle image data; Perform feature point extraction based on Harris corner detection on multi-angle image data to obtain multi-angle image feature points; Calculate the 128-dimensional descriptor vector of the multi-angle image feature points and set the matching threshold to 0.75 to obtain the feature point descriptor vector; The structural feature points in the towering structure reference coordinate system are projected onto each viewing plane with an accuracy of ±0.5 pixels to form a standard corresponding point set; The feature point descriptor vector is used to match the standard corresponding point mapping table, and the Euclidean distance threshold of 80 is used as the screening condition to establish the initial feature point correspondence relationship set; The initial feature point correspondence set is iteratively optimized, the error tolerance threshold is set to 2.5 pixels, and abnormal matching points exceeding 15% of the total number of matches are eliminated to obtain a reliable feature point correspondence matrix; Based on the reliable feature point correspondence matrix, a homography transformation matrix with an accuracy error of less than 0.8 pixels is calculated and applied to multi-angle image data for perspective correction, generating a spatially consistent image set with an edge error within 1.2 pixels. The bicubic interpolation sub-pixel deformation compensation based on 16×16 pixel grid density is implemented on the spatially consistent image set to obtain the registered image data with a maximum error of less than 0.3 pixels.

6. The method for processing a building model according to claim 5, characterized in that: Constructing a three-dimensional point cloud model of the towering structure based on the registered image data in step S3 includes: Identify the same-name points in the overlapping areas between adjacent images on the registered image data to obtain the spatial coordinates of the feature points; The spatial coordinates of the feature points are used to establish a preliminary sparse point cloud, and the validity of the spatial points is verified by viewing angles greater than 15° to obtain a skeleton point cloud; Determine the main plane and outline of the tall structure based on the skeleton point cloud, and divide the tall structure into different component areas to form a structural area index; Dense matching is performed on each region in the structure region index to generate a gridded dense point cloud, where the search window for dense matching is set to 7×7 pixels and the matching step size is 2 pixels; The gridded dense point cloud is merged according to the structural area index and local smoothing is performed based on radius query. The search radius is set to 0.05m, and points that deviate from the mean by 3 times the standard deviation are removed to obtain the filtered point cloud. Based on the color information of the filtered point cloud and the corresponding registered image, an RGB value is assigned to each point to generate a three-dimensional point cloud model of the towering structure.

7. The method for processing a building model according to claim 6, wherein: Extracting structural surface abnormality feature information from the three-dimensional point cloud model according to the towering structure reference coordinate system in step S3 includes: The 3D point cloud model is transformed and aligned based on the towering structure reference coordinate system to obtain a standardized 3D point cloud in the reference system; Based on the association between the towering structure reference coordinate system and the existing structural area index, the precise position mapping of each structural component in the reference coordinate system is established to form a reference system component mapping table; Calculate the surface deviation distribution of each component in the reference system component mapping table in the towering structure reference coordinate system, identify areas with deviations exceeding 5mm, and mark them as potential abnormal areas in the reference system; Perform spatial cluster analysis on potential abnormal areas to form abnormal characteristic patch data; The geometric properties are calculated based on the abnormal feature patch data and the reference coordinate system of the towering structure to generate abnormal feature information on the building surface.

8. The method for processing a building model according to claim 7, wherein: Step S4 includes the following steps: Step S41: determining the type of building surface damage based on the geometric form and distribution characteristics of the building abnormality feature information; Step S42: Calculate the number, area, and distribution density of damage based on the building surface damage type and building abnormality feature information to form quantitative index data; Step S43: performing damage classification processing on the quantitative index data to obtain structural damage assessment data; Step S44: classifying the structural damage assessment data based on location, type, and severity to generate a structural damage distribution map; Step S45: Calculate the key risk index for the structural damage distribution map, and generate a risk heat map based on the structural importance and damage severity of the damage location; Step S46: Based on the risk heat map and structural damage assessment data, damage statistics, risk level distribution and overall safety integration assessment are performed to generate a health status report; Step S47: Compare and analyze the health status report with the preset historical data to obtain the structural damage development trend; formulate a predictive maintenance plan based on the structural damage development trend to form inspection decision support data.

9. A building model processing system, characterized in that: A system for processing a building model according to claim 1, wherein the system comprises: The image acquisition and jitter compensation module is used to acquire initial image data of the tall structure through the aircraft and simultaneously record the motion status data; calculate the jitter characteristic parameters based on the initial image data and the motion status data; and establish a jitter compensation mechanism for long-range images based on the jitter characteristic parameters; The image stabilization and coordinate system establishment module is used to process the initial image data using a jitter compensation mechanism to obtain stabilized image data; extract structural feature point information from the stabilized image data; and establish a towering structure reference coordinate system based on the structural feature point information; The 3D modeling and anomaly detection module is used to perform high-precision registration of pre-acquired multi-angle image data of the towering structure with the towering structure reference coordinate system to obtain registered image data; construct a 3D point cloud model of the towering structure based on the registered image data; and extract abnormal feature information of the building surface from the 3D point cloud model according to the towering structure reference coordinate system; The damage assessment and maintenance decision module is used to analyze the type of building surface damage based on abnormal characteristic information of the building surface, and perform graded processing to obtain structural damage assessment data; generate a health status report based on the structural damage assessment data; compare and analyze the health status report with preset historical data to obtain the development trend of structural damage; formulate a predictive maintenance plan based on the development trend of structural damage to form inspection decision support data.

10. A computer medium storing a computer program, characterized in that: When the computer program is executed, the architectural model processing method according to any one of claims 1 to 8 is implemented.

Images