Method for quickly auditing oblique photography model

Abstract

The invention provides a method for quickly checking an oblique photography model, and relates to the technical field of image processing. Comprising the following steps: S1, acquiring an oblique photography model, a digital orthoimage, a digital surface model and high-precision control point data, judging a topographic type of a measured area based on DSM elevation mutation parameters, and correspondingly executing format conversion, noise removal and hierarchical data registration to form a standardized auditing data set adapted to topographic features; s2, based on the preprocessed data set, synchronously executing time-space cooperation precision detection and feature fusion integrity and compliance detection, and introducing a three-dimensional evaluation model to complete dynamic risk level evaluation; s3, generating a visual auditing report containing a dynamic label; the problems of complex terrain adaptation, risk assessment and closed-loop efficiency in the prior art are solved, the multi-dimensional detection coverage rate is increased, the risk identification accuracy is improved, the auditing and rectification closed-loop efficiency is improved, the labor cost is reduced, and the scene adaptability is improved.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method for rapidly reviewing oblique photography models. Background Technology

[0002] With the in-depth development of digital twins, smart cities, intelligent construction and other fields, oblique photogrammetry models have been upgraded from static scene restoration to dynamic decision support. Their application scenarios cover key areas such as urban renewal, emergency mapping, and land planning, which put forward three stringent requirements for the real-time performance, accuracy and scene adaptability of model data.

[0003] Existing oblique photogrammetry model review techniques have significant bottlenecks, specifically: First, existing techniques (such as CN120374682B and CN117522981A) employ a preprocessing mode that combines fixed format conversion, uniform noise removal, and a single registration algorithm, which cannot adapt to diverse terrain features. In flat areas, redundant data formats lead to low processing efficiency; in mountainous areas, fixed registration parameters make it difficult to control errors in steep slope areas; and in densely built-up urban areas, simple repetitive patch deletion logic mistakenly deletes building detail vertices. Ultimately, this results in a disconnect between the standardized dataset and actual terrain features, creating potential accuracy risks for subsequent detection. II. Traditional review methods only achieve single-dimensional detection of planar / elevation accuracy combined with basic integrity, which has two major drawbacks: 1) They do not associate multi-source data features. For example, texture misalignment issues rely solely on DOM visual judgment without combining DSM elevation gradient verification, leading to the omission of high-risk issues such as the superposition of texture misalignment and elevation abrupt changes in urban building clusters; 2) They lack a risk classification mechanism, treating 1m² holes and 100m² holes, and deviations in ordinary areas and deviations in sensitive areas such as airports / railways, equally, failing to provide priority guidance for rectification. Especially in emergency mapping scenarios, the ambiguity of review priorities can easily delay decision-making. III. Existing technologies (such as CN118067088A) only stop at outputting a visual report of the detection results, failing to form a closed loop: On the one hand, the review report only marks the coordinates and type of the problem area, without providing a rectification plan adapted to the terrain, requiring repeated manual trial and error, with a rectification cycle of several hours; on the other hand, after rectification, the model needs to restart the entire process of manual review, failing to achieve automated connection between detection, rectification, and re-inspection. In large-scale model review, the overall efficiency is reduced by more than 50%. Fourth, existing methods do not take into account the differences in accuracy requirements of different application scenarios: For example, the planar accuracy requirements of airport runway areas are much higher than those of ordinary residential areas, but traditional review uses a uniform threshold, resulting in excessive accuracy in sensitive areas or over-detection in non-sensitive areas; at the same time, emergency mapping scenarios require rapid pre-inspection of high-risk areas, but the existing full-process review model cannot meet the timeliness requirements, further limiting the application boundaries of oblique photogrammetry models. Summary of the Invention

[0004] The main objective of this invention is to provide a method for rapidly reviewing oblique photogrammetry models, thereby addressing the problems of poor adaptability to complex terrain, limited preprocessing accuracy, low rectification efficiency, lack of scene sensitivity adaptation, and significant application limitations in existing technologies.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for quickly reviewing oblique photogrammetry models, comprising the following steps: S1. Data preprocessing: Obtain oblique photogrammetry model, digital orthophoto (DOM), digital surface model (DSM) and high-precision control point data. Determine the terrain type of the survey area based on the DSM elevation change parameters. Perform format conversion, noise removal and hierarchical data registration to form a standardized audit dataset adapted to terrain features. S2, Multi-dimensional detection: Based on the preprocessed dataset, spatiotemporal collaborative accuracy detection, feature fusion integrity and compliance detection are performed simultaneously, and a three-dimensional evaluation model is introduced to complete dynamic risk level assessment; S3. Output Results: Generate a visual audit report with dynamic annotations, including automatically outputting color annotations that match the problem type and the problem area in the oblique photogrammetry model.

[0006] In the preferred embodiment, step S1, which involves standardizing the data based on terrain features, includes: The data format conversion strategy is dynamically selected based on the terrain type, which is determined by the elevation parameters of the digital surface model. Based on the mapping relationship between terrain type and noise type, the corresponding filtering algorithm is used to perform noise removal; Perform hierarchical data registration, including coarse registration and fine registration, to ensure spatial consistency among multi-source data.

[0007] In the preferred embodiment, the terrain types include flat areas, mountainous areas, and urban areas, wherein: Flat areas are determined by a slope of less than 15°. Mountainous areas are determined by a slope of ≥15°; Urban areas are defined by a building density greater than 30%.

[0008] In the preferred embodiment, the oblique photogrammetry model is converted to PLY format and vertex coordinates, texture coordinates and triangular facet topology are extracted. The DOM and DSM are converted to TIFF format, and the high-precision control point data is converted to JSON format.

[0009] In a preferred embodiment, the noise removal is performed based on terrain type, including: Vegetation cover area: A bilateral filtering method combined with isolated point clustering and removal is adopted. First, vegetation edge vertices are retained, and then isolated vertices that are far from the cluster center by a preset distance threshold are removed. In densely built areas: a combination of topological consistency check and duplicate face deletion is used. Duplicate faces are identified by the number of shared edges of triangular faces, and duplicate faces that are not detected for the first time are deleted. Water area: First, locate the water area using DOM water reflectivity, then remove noise points with signal strength < 0.5 * preset threshold within the water area.

[0010] In the preferred embodiment, the hierarchical data registration includes coarse registration and fine registration, specifically as follows: Coarse registration involves matching affine transformation parameters corresponding to the terrain type from a preset parameter library. This includes matching parameter sets according to terrain type from a preset affine transformation image geometric correction parameter library. For mountainous areas, parameter sets that conform to the first rotation angle and the first translation amount are selected, while for urban areas, parameter sets that conform to the second rotation angle and the second translation amount are selected. Fine registration employs an iterative nearest-point algorithm and introduces auxiliary data's feature outlines or elevation change lines as constraints to control registration errors. This includes: using a source-constraint iterative nearest-point algorithm, based on high-precision control points, adding collaborative constraints between DOM feature outlines and DSM elevation change lines, calculating their overlap, and increasing the weight of the outlines when the overlap is insufficient; the registration error is controlled to be ≤0.05m in flat areas, ≤0.1m in mountainous areas, and ≤0.03m in urban areas.

[0011] In the preferred embodiment, the multi-dimensional detection step includes: The spatiotemporal coordinated accuracy detection includes planar accuracy detection and elevation accuracy detection. Planar accuracy detection incorporates time-series data comparison, while elevation accuracy detection dynamically adjusts the deviation threshold based on terrain slope. The integrity detection identifies model voids and associates them with texture loss by fusing digital surface model and digital orthophoto data; The compliance inspection uses a machine learning model to analyze texture images, identify texture anomalies, and associate them with geometric defects.

[0012] In the preferred scheme, the planar accuracy detection introduces time series data comparison, including: combining the projection correction algorithm, adding time series DOM comparison, and marking the planar deviation of digital orthophotos of the same area at different periods as a dynamic deformation area when the preset digital orthophoto deviation threshold is set; The elevation accuracy detection is based on the dynamic adjustment of deviation thresholds according to terrain slope, including: constructing a dynamic mapping between terrain slope and elevation thresholds, preset deviation thresholds for flat areas, gentle slope areas and steep slope areas, calculating the elevation deviation values ​​of control points in the model and digital surface model, and marking areas with abnormal elevation accuracy as such if the deviation exceeds the corresponding threshold.

[0013] In the preferred scheme, the integrity detection includes: constructing an association model of holes and texture loss by fusing digital surface model and digital orthophoto; firstly, locating potential holes by the gaps between triangular facets of digital surface model; then verifying the texture coverage of the corresponding digital orthophoto area; and classifying risks according to hole area to complete the integrity detection. The compliance test includes: the machine learning model used is a convolutional neural network model, whose input includes texture images and elevation gradient maps, and the weights are assigned through a feature fusion layer to output the recognition results of texture misalignment, blurring or color distortion; The results of integrity checks and compliance checks are linked and integrated to generate check output data.

[0014] In the preferred embodiment, the visual audit report uses color coding and dynamic annotation to highlight problem areas, and includes problem statistics and comparison charts; The data in the visualized audit report is automatically generated based on the problem type and terrain features, and the rectification effect is simulated through digital twin technology; It supports closed-loop data processing, enabling the rectified model to be reconnected to the data preprocessing step for iterative review.

[0015] This invention provides a method for rapidly reviewing oblique photogrammetry models. By acquiring terrain-adaptive related data, preprocessing it, performing multi-dimensional detection, introducing a three-dimensional evaluation model to complete dynamic risk level assessment, and finally generating a visual review report, including an automatically output visual review report matching the problem type and terrain features, this method solves the problems of existing technologies in complex terrain adaptation, risk assessment, and closed-loop efficiency. It improves the coverage of multi-dimensional detection, enhances the accuracy of risk identification, improves the efficiency of review and rectification closed-loop, reduces labor costs, and improves scene adaptability. Attached Figure Description

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram showing the labeled problem area of ​​the present invention. Detailed Implementation

[0017] Example 1 like Figure 1-2 As shown, a method for quickly reviewing oblique photogrammetry models includes the following steps: S1. Data Preprocessing: Obtain oblique photogrammetry model, digital orthophoto (DOM), digital surface model (DSM) and high-precision control point data. Determine the terrain type of the survey area based on the DSM elevation change parameters. Perform format conversion, intelligent noise removal and hierarchical data registration to form a standardized audit dataset adapted to terrain features.

[0018] S2. Multi-dimensional detection: Based on the preprocessed dataset, spatiotemporal collaborative accuracy detection, feature fusion integrity and compliance detection are performed simultaneously, and a three-dimensional evaluation model is introduced to complete dynamic risk level assessment.

[0019] S3. Output Results: Generate a visual audit report with dynamic annotations, including automatically outputting rectification parameters that match the problem type and terrain features, as well as a pre-simulation of the rectification effect. The rectification parameters and pre-simulation results include automatically outputting color annotations that match the problem type and the problem area in the oblique photogrammetry model.

[0020] In this embodiment, by acquiring terrain-adaptive related data, preprocessing it, performing multi-dimensional detection, introducing a three-dimensional evaluation model to complete dynamic risk level assessment, and finally generating a visualized audit report, including automatically outputting rectification parameters matching the problem type and terrain features, as well as the pre-simulation results of rectification effects, it solves the problems of existing technologies in complex terrain adaptation, risk assessment, and closed-loop efficiency, improves the multi-dimensional detection coverage, enhances the risk identification accuracy, improves the audit and rectification closed-loop efficiency, reduces labor costs, and improves scenario adaptability.

[0021] In the preferred embodiment, step S1, which involves standardizing the data based on terrain features, includes: Dynamically select the data format conversion strategy based on the data type; Based on the mapping relationship between terrain type and noise type, the corresponding filtering algorithm is used to perform noise removal; Perform hierarchical data registration, including coarse registration and fine registration, to ensure spatial consistency among multi-source data.

[0022] In this embodiment, step S1, forming a standardized audit dataset adapted to terrain features, includes: Terrain types include plains, mountains, and urban areas, among which: Terrain type is determined by the DSM elevation change parameter: slope < 5° in flat areas, slope ≥ 15° in mountainous areas, and building density > 30% in urban areas.

[0023] The format conversion specifically involves converting the oblique photogrammetry model to PLY format and extracting vertex coordinates, texture coordinates, and triangular facet topology; converting DOM and DSM to TIFF format (0.1m pixel resolution); and converting high-precision control point data to JSON format.

[0024] In this embodiment, after data format conversion, redundant data (PLY format) is reduced, and format compatibility is improved.

[0025] In this embodiment, automatic conversion between OSGB / 3DTiles / PLY is supported without the need for manual parameter settings, adapting to the data habits of different surveying and mapping units; it improves the customized processing capability for sensitive areas / emergency scenarios and increases scene coverage.

[0026] In the preferred embodiment, noise removal is performed based on a mapping model between terrain type and noise type, including: Vegetation cover area (DSM vegetation index > 0.6): A bilateral filtering method combined with isolated point clustering and removal is used to first retain vegetation edge vertices and then remove isolated vertices that are at a preset distance threshold from the cluster center. In densely built areas: a combination of topological consistency check and duplicate face deletion is used. Duplicate faces are identified by the number of shared edges of triangular faces (≥2 edges are considered adjacent), and duplicate faces that are not detected for the first time are deleted. Water area: First, locate the water area using the DOM water reflectivity (<20%), then remove noise points in the water area with signal strength <0.5*preset threshold.

[0027] This embodiment reduces the false deletion rate of vegetation areas through bilateral filtering and clustering removal, improves the recognition rate of duplicate facets in building areas through topology consistency verification, and enhances the efficiency of water area noise removal through reflectivity localization, thereby reducing interference from invalid data.

[0028] In the preferred scheme, hierarchical data registration includes coarse registration and fine registration, specifically as follows: Coarse registration includes: matching affine transformation parameters corresponding to the terrain type from the preset affine transformation image geometric correction parameter library, including: matching parameter groups according to terrain type from the preset affine transformation image geometric correction parameter library, selecting parameter groups that conform to the first rotation angle and the first translation amount for mountainous areas, and selecting parameter groups that conform to the second rotation angle and the second translation amount for urban areas.

[0029] In this embodiment, the first rotation angle and the first translation amount are rotation angle ±3° and translation amount ±10m, and the second rotation angle and the second translation amount are rotation angle ±0.5° and translation amount ±2m.

[0030] Matching parameter groups by terrain type: for mountainous areas, select parameter group with rotation angle ±3° and translation ±10m; for urban areas, select parameter group with rotation angle ±0.5° and translation ±2m.

[0031] Fine registration includes: using the iterative nearest point algorithm and introducing auxiliary data's feature outlines or elevation change lines as constraints to control registration errors. This includes: using the source constraint iterative nearest point algorithm, based on high-precision control points, adding collaborative constraints between DOM feature outlines and DSM elevation change lines, calculating their overlap, and increasing the weight ratio of the outlines when the overlap is insufficient; in this embodiment, an overlap of 85% is preset as acceptable.

[0032] Based on the above operations, this embodiment controls the registration error to ≤0.05m in flat areas, ≤0.1m in mountainous areas, and ≤0.03m in urban areas.

[0033] In this embodiment, coarse registration reduces the initial deviation in mountainous areas by matching terrain parameter groups; fine registration adds contour line-elevation line collaborative constraints, which improves the accuracy of traditional single registration.

[0034] In the preferred solution, in step S2, the multi-dimensional detection step: Spatiotemporal collaborative accuracy detection includes planar accuracy detection and elevation accuracy detection. Planar accuracy detection incorporates time-series data comparison, while elevation accuracy detection dynamically adjusts the deviation threshold based on terrain slope.

[0035] Integrity inspection identifies model voids and correlates them with texture loss by fusing digital surface model (DSM) and digital orthophoto (DOM) data.

[0036] The compliance inspection uses a machine learning model to analyze texture images, identify texture anomalies, and correlate them with geometric defects.

[0037] In the preferred scheme, spatiotemporal coordinated accuracy detection includes planar accuracy detection and elevation accuracy detection, specifically: Planar accuracy detection includes: combining projection correction algorithm, adding time series DOM comparison, and marking the DOM plane deviation of the same area at different times as >0.1m (>0.05m in urban areas) as a dynamic deformation area.

[0038] Elevation accuracy detection includes: constructing a dynamic mapping between terrain slope and elevation thresholds, with elevation deviation thresholds of ≤0.05m for flat areas, ≤0.08m for gentle slope areas (5°≤slope<15°), and ≤0.12m for steep slope areas (≥15°); calculating the elevation deviation values ​​of control points in the model and DSM, and marking areas exceeding the corresponding thresholds as areas with abnormal elevation accuracy.

[0039] Furthermore, fine registration includes: using high-precision control point data as a benchmark, spatially registering the oblique photogrammetry model with digital orthophotos and digital surface models through an iterative nearest-point algorithm, including: High-precision control point data is acquired as reference coordinates; the registration transformation matrix is ​​initialized for the vertex coordinates of the oblique photogrammetry model and the pixel coordinates of the digital orthophoto and digital surface model.

[0040] The nearest point pair is calculated using the iterative nearest point algorithm, and the transformation matrix is ​​updated until convergence.

[0041] If the registration error is lower than the preset threshold, the registered dataset will be output.

[0042] The coordinate information in the registered dataset is associated with the standardized audit dataset and used as input for the multi-dimensional automated detection module.

[0043] In this embodiment, time-series DOM comparison improves the recognition rate of dynamically deformed areas, such as settlement areas, reduces the misjudgment rate of steep slopes in mountainous areas due to slope-elevation threshold mapping, and improves the accuracy of plane and elevation precision detection.

[0044] In the preferred scheme, the feature fusion integrity and compliance detection specifically includes: Integrity testing includes: constructing a correlation model of holes and texture loss by integrating DSM and DOM; firstly, locating potential holes using the gaps between DSM triangles (>0.03m); then verifying the texture coverage of the corresponding DOM region (<90% confirms a hole); and classifying risks according to hole area: <1m² is low risk, 1-10m² is medium risk, and >10m² is high risk, to complete the integrity testing. See Table 1 for details. Table 1. Core Architecture of the Model and Data Mapping Relationship

[0045] The compliance inspection includes: using a visual ensemble geometric cross-modal feature fusion model, the input layer of which uses a pre-trained convolutional neural network model to input DOM texture images and DSM elevation gradient maps, the feature fusion layer assigns weights through an attention mechanism (60% weight for geometric features in the building area and 70% weight for visual features in the vegetation area), and the output layer identifies texture misalignment, blurring, and color distortion, and associates them with geometric defects. If the two are superimposed, the risk level is increased by 1 level to complete the compliance inspection.

[0046] The results of integrity checks and compliance checks are linked and integrated to generate check output data.

[0047] In the preferred embodiment, the step of using a pre-trained convolutional neural network model to input a texture image and output texture misalignment, blurring, and color distortion recognition results includes: Extract texture images from the preprocessed dataset as input; Texture features are extracted using the convolutional layers of a convolutional neural network model; The misaligned areas are identified based on the extracted texture features. If the feature offset exceeds a preset threshold, it is marked as a texture misalignment. Calculate the sharpness index of the texture image; if the sharpness is lower than a preset threshold, it is marked as texture blurry. Analyze the color distribution; if the color deviation exceeds a preset threshold, it is marked as color distortion.

[0048] In this embodiment, the hole-texture loss association model improves the accuracy of hole confirmation, enhances the recognition rate of texture misalignment and elevation change superposition problems caused by cross-modal feature fusion of vision and geometry, and improves the multi-dimensional detection coverage.

[0049] In the preferred embodiment, step S3 includes the following output: Visualized report: It adopts a two-dimensional dynamic labeling system that corresponds to the risk level of the problem type. High-risk problems are marked in red and flashing, while medium-risk problems are marked in yellow and static. A comparison chart between the problem area and the standard template has also been added.

[0050] Automatic rectification plan: recommend interpolation methods for hollow areas (linear interpolation for flat areas, kriging interpolation for mountainous areas), recommend rectification range for texture misalignment based on misalignment distance (local remapping for <0.1m, remapping for >0.1m), and preview the rectification effect through digital twin technology.

[0051] Data closed loop: After rectification, the model is reconnected to the terrain adaptive data preprocessing step, and the detection process is repeated until the risk level is reduced to low risk.

[0052] This embodiment improves the response speed to high-risk issues in sensitive areas, such as airport runway deviations, by using a three-dimensional model of severity, scope of impact, and sensitivity; it solves the resource waste caused by traditional non-priority rectification and improves decision-making efficiency in emergency scenarios.

[0053] In the preferred embodiment, after automated testing is completed, the test results are integrated to form a visual audit report, including: The detection output data, including accuracy deviation information, hole location information and texture recognition results, is obtained from the multi-dimensional automated detection module. The detection output data is statistically summarized to calculate the number and distribution of problem areas; Based on the statistical summary results, color-coded markers are used on the model to highlight problem areas; Generates a visual audit report that includes audit result statistics and color-coded annotations, and supports coordinate positioning and comparison chart display.

[0054] Furthermore, the visual audit report uses color coding and dynamic annotations to highlight problem areas, and includes problem statistics and comparison charts.

[0055] The rectification parameters are automatically generated based on the problem type and terrain features, and the rectification effect is simulated through digital twin technology.

[0056] It supports closed-loop data processing, enabling the rectified model to be reconnected to the data preprocessing step for iterative review.

[0057] In this embodiment, the automated rectification scheme shortens the rectification time for gaps and improves the success rate of rectification through digital twin pre-simulation; the closed-loop re-inspection realizes the city-level model, shortening the entire cycle and improving work efficiency; dynamic annotation improves the speed of problem location.

[0058] This embodiment improves the accuracy of complex terrain review, shortens the review cycle of large-scale models, and reduces manpower costs through an overall architecture of terrain adaptive preprocessing, cross-modal detection, and closed-loop output. It is applicable to multiple application scenarios such as emergency mapping, smart cities, and precision engineering.

[0059] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A method for rapidly reviewing oblique photogrammetry models, characterized in that, Includes the following steps: S1. Data preprocessing: Obtain oblique photogrammetry model, digital orthophoto (DOM), digital surface model (DSM) and high-precision control point data. Determine the terrain type of the survey area based on the DSM elevation change parameters. Perform format conversion, noise removal and hierarchical data registration to form a standardized audit dataset adapted to terrain features. S2, Multi-dimensional detection: Based on the preprocessed dataset, spatiotemporal collaborative accuracy detection, feature fusion integrity and compliance detection are performed simultaneously, and a three-dimensional evaluation model is introduced to complete dynamic risk level assessment; S3. Output Results: Generate a visual audit report with dynamic annotations, including automatically outputting color annotations that match the problem type and the problem area in the oblique photogrammetry model.

2. The method for rapidly reviewing oblique photogrammetry models according to claim 1, characterized in that, In step S1, the data standardization process based on terrain features includes: Dynamically select the data format conversion strategy based on the data type; Based on the mapping relationship between terrain type and noise type, the corresponding filtering algorithm is used to perform noise removal; Perform hierarchical data registration, including coarse registration and fine registration, to ensure spatial consistency among multi-source data.

3. The method for rapidly reviewing oblique photogrammetry models according to claim 2, characterized in that, The terrain types include plains, mountains, and urban areas, among which: Flat areas are determined by a slope of less than 15°. Mountainous areas are determined by a slope of ≥15°; Urban areas are defined by a building density greater than 30%.

4. The method for rapidly reviewing oblique photogrammetry models according to claim 2, characterized in that, The oblique photogrammetry model was converted to PLY format and vertex coordinates, texture coordinates and triangular topological relationships were extracted. The DOM and DSM were converted to TIFF format, and the high-precision control point data was converted to JSON format.

5. The method for rapidly reviewing oblique photography models according to claim 2, characterized in that, The noise removal is performed based on terrain type and includes: Vegetation cover area: A bilateral filtering method combined with isolated point clustering and removal is adopted. First, vegetation edge vertices are retained, and then isolated vertices that are far from the cluster center by a preset distance threshold are removed. In densely built areas: a combination of topological consistency check and duplicate face deletion is used. Duplicate faces are identified by the number of shared edges of triangular faces, and duplicate faces that are not detected for the first time are deleted. Water area: First, locate the water area using DOM water reflectivity, then remove noise points with signal strength < 0.5 * preset threshold within the water area.

6. The method for rapidly reviewing oblique photogrammetry models according to claim 2, characterized in that, The hierarchical data registration includes coarse registration and fine registration, specifically: Coarse registration involves matching affine transformation parameters corresponding to the terrain type from a preset parameter library. This includes matching parameter sets according to terrain type from a preset affine transformation image geometric correction parameter library. For mountainous areas, parameter sets that conform to the first rotation angle and the first translation amount are selected, while for urban areas, parameter sets that conform to the second rotation angle and the second translation amount are selected. Fine registration employs an iterative nearest-point algorithm and introduces auxiliary data's feature outlines or elevation change lines as constraints to control registration errors. This includes: using a source-constraint iterative nearest-point algorithm, based on high-precision control points, adding collaborative constraints between DOM feature outlines and DSM elevation change lines, calculating their overlap, and increasing the weight of the outlines when the overlap is insufficient; the registration error is controlled to be ≤0.05m in flat areas, ≤0.1m in mountainous areas, and ≤0.03m in urban areas.

7. The method for rapidly reviewing oblique photogrammetry models according to claim 1, characterized in that, In the multi-dimensional detection step: The spatiotemporal coordinated accuracy detection includes planar accuracy detection and elevation accuracy detection. Planar accuracy detection incorporates time-series data comparison, while elevation accuracy detection dynamically adjusts the deviation threshold based on terrain slope. The integrity detection identifies model holes and correlates texture loss by fusing DSM and DOM data; The compliance inspection uses a machine learning model to analyze texture images, identify texture anomalies, and associate them with geometric defects.

8. The method for rapidly reviewing oblique photogrammetry models according to claim 7, characterized in that, Planar accuracy detection introduces time-series data comparison, including: combining projection correction algorithm, adding time-series DOM comparison, and marking the DOM plane deviation of the same area at different periods as a dynamic deformation area when the preset DSM deviation threshold is used; The elevation accuracy detection is based on the dynamic adjustment of deviation thresholds according to terrain slope, including: constructing a dynamic mapping between terrain slope and elevation thresholds, preset deviation thresholds for flat areas, gentle slope areas and steep slope areas, calculating the elevation deviation values ​​of control points in the model and DSM, and marking areas with abnormal elevation accuracy as such if the deviation exceeds the corresponding threshold.

9. The method for rapidly reviewing oblique photogrammetry models according to claim 7, characterized in that, Integrity detection includes: integrating DSM and DOM to construct an association model of holes and texture loss, first locating potential holes through the gaps between DSM triangles, then verifying the texture coverage of the corresponding DOM region, and classifying risks according to the hole area to complete the integrity detection; The compliance test includes: the machine learning model used is a convolutional neural network model, whose input includes texture images and elevation gradient maps, and the weights are assigned through a feature fusion layer to output the recognition results of texture misalignment, blurring or color distortion; The results of integrity checks and compliance checks are linked and integrated to generate check output data.

10. The method for rapidly reviewing oblique photogrammetry models according to claim 1, characterized in that, The visual audit report uses color coding and dynamic annotation to highlight problem areas, and includes problem statistics and comparison charts; The data in the visualized audit report is automatically generated based on the problem type and terrain features, and the rectification effect is simulated through digital twin technology; It supports closed-loop data processing, enabling the rectified model to be reconnected to the data preprocessing step for iterative review.

Images