An AI technology fusion-based high-precision 3D scanner system and method

Abstract

The application relates to the technical field of cultural heritage building scanning detection, and discloses a high-precision 3D scanner system and method based on AI technology fusion. The system comprises a data acquisition module, an AI data preprocessing module, a structural feature analysis module, a self-adaptive laser detection module and an AI simulation modeling and detection result fusion module. The method obtains cultural heritage building point cloud data through laser scanning, extracts structural feature parameters and determines morphological tendency categories after AI multi-stage noise reduction and segmentation processing, adaptively adjusts the laser incidence direction and the scanning path, combines AI three-dimensional reconstruction and texture optimization to generate a high-precision simulation model, and simultaneously completes structural defect detection and classification. The application clearly defines the dynamic setting, calibration method and feature sub-region adaptive division strategy of key parameters, solves the problem of insufficient precision caused by environmental interference and complex structure in cultural heritage building scanning, significantly improves the scanning detection precision and efficiency, and provides technical support for cultural heritage protection.

Description

Technical Field

[0001] This invention relates to the field of cultural heritage building scanning and inspection technology, specifically a high-precision 3D scanner system and method based on AI technology fusion. Background Technology

[0002] As important carriers of historical civilization, cultural heritage buildings require protection that simultaneously meet the dual demands of high-precision digital archiving and structural safety inspection. Traditional scanning and inspection methods for cultural heritage buildings have many limitations: on the one hand, relying on manual measurement or simple scanning equipment makes it difficult to capture the fine forms of complex structures (such as brackets, wood carvings, and curved beams), and is easily affected by factors such as ambient lighting and surface material reflection, resulting in noisy point cloud data and insufficient modeling accuracy; on the other hand, existing laser inspection systems lack the ability to adaptively adjust to the structural form of cultural heritage buildings, have weak targeting for key area identification, and cannot dynamically optimize inspection strategies based on changes in structural curvature, resulting in low efficiency and poor accuracy in defect detection.

[0003] In existing technologies, some solutions focus on laser inspection of construction beams, improving inspection accuracy through feature recognition and laser adjustment, but fail to fully consider the complex environmental interference and high-precision modeling requirements of cultural heritage buildings. Other solutions use AI visual recognition to achieve simulated modeling of objects, optimizing point cloud data processing and model reconstruction, but lack specialized inspection functions for building structures, thus failing to meet the structural safety assessment needs of cultural heritage buildings. Therefore, there is an urgent need for an integrated system that combines AI data processing, adaptive laser inspection, and high-precision modeling to solve the technical problems of insufficient accuracy and low efficiency in the scanning and inspection of cultural heritage buildings.

[0004] In view of this, the applicant conducted in-depth research on the above-mentioned issues, which led to this case. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a high-precision 3D scanner system and method based on AI technology fusion, which overcomes the aforementioned deficiencies of existing technologies and enables high-precision scanning, modeling, and structural inspection of cultural heritage buildings.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A high-precision 3D scanner system based on AI technology fusion includes: The data acquisition module is used to perform laser scanning on preset acquisition points of cultural heritage buildings to obtain raw three-dimensional point cloud data; The AI ​​data preprocessing module, connected to the data acquisition module, is used to perform multi-stage noise reduction on the original 3D point cloud data to obtain noise-reduced point cloud data; the noise-reduced point cloud data is input into a preset point cloud segmentation model, which is divided into several detection areas according to the building structure type, and feature point cloud data is extracted at the same time. The structural feature analysis module is connected to the data acquisition module and the AI ​​data preprocessing module, respectively. It is used to calculate morphological fluctuation parameters based on the segmented detection area point cloud data to screen key structural regions; and to determine the arc characterization coefficient by the angle between the feature sub-region normal vector and the laser axis, thereby determining the structural morphological tendency category. An adaptive laser detection module is connected to the data acquisition module, the AI ​​data preprocessing module and the structural feature analysis module respectively, and is used to adaptively adjust the laser incident direction or simultaneously adjust the laser scanning path and incident direction according to the structural morphology tendency category. The AI ​​simulation modeling and detection result fusion module is connected to the AI ​​data preprocessing module, structural feature analysis module, and adaptive laser detection module, respectively. It is used to perform 3D reconstruction and texture optimization based on the detected point cloud data to generate a simulated 3D model and a simulated texture model. The structural defect detection results are marked in the model, and the defect classification is completed by the AI ​​visual recognition system to output a high-precision model of fused detection results.

[0007] Furthermore, the multi-stage denoising process of the AI ​​data preprocessing module includes: performing preliminary denoising on the original 3D point cloud data to obtain preliminary denoised point cloud data; performing point cloud density homogenization on the preliminary denoised point cloud data using a mean filtering algorithm to obtain homogenized point cloud data; performing fine denoising on the homogenized point cloud data using a region growing algorithm to obtain finely denoised point cloud data; estimating the normals of the finely denoised point cloud data using the Boltzmann method to obtain a set of normal vectors; processing the set of normal vectors using nonlocal mean filtering to obtain smoothed point cloud data; performing edge-preserving denoising on the smoothed point cloud data using nonlocal mean filtering to obtain edge-preserving denoised point cloud data; and performing data fusion on the edge-preserving denoised point cloud data to obtain denoised point cloud data.

[0008] Furthermore, the structural feature analysis module calculates the morphological fluctuation parameter as follows: based on the point cloud data of the detection area, several building structure widths are determined, and the maximum and minimum building structure widths within the detection area are respectively determined as the first morphological factor and the second morphological factor; the difference between the first morphological factor and the second morphological factor is calculated, and the difference is determined as the morphological fluctuation parameter; if the morphological fluctuation parameter exceeds a preset fluctuation reference value, the detection area is selected as a key structural area.

[0009] Furthermore, the structural feature analysis module determines the structural morphological tendency category as follows: the key area of ​​the structure is divided into several feature sub-regions, and the normal vector and laser axis of each feature sub-region are determined based on point cloud data; the angle between the direction of the normal vector and the direction of the laser axis is calculated, and the angle is determined as the radian characterization coefficient of the feature sub-region; if the variance of several radian characterization coefficients in the key area of ​​the structure does not exceed a preset variance threshold, it is determined as the first morphological tendency category; otherwise, it is determined as the second morphological tendency category.

[0010] Furthermore, the adjustment method of the adaptive laser detection module is as follows: if the structural morphology tendency category is the first morphology tendency category, a reference normal vector is determined based on several normal vectors in the key area of ​​the structure, and the laser incident direction for laser scanning of the key area of ​​the structure is adjusted, wherein the laser incident direction is parallel to the reference normal vector; if the structural morphology tendency category is the second morphology tendency category, the angle between the normal vector of the feature sub-region and the horizontal plane is calculated, and the scanning path is determined by sorting the angle values, while the laser incident direction is adjusted to be parallel to the normal vector of each feature sub-region.

[0011] Furthermore, the reference normal vector is a vector obtained by adding several normal vectors within the key structural region; the key structural region is the area in the building structure where the morphological fluctuation parameter exceeds the preset fluctuation reference value, specifically the area enclosed by the line segment of the first morphological factor, the line segment of the second morphological factor, and the edge contour along the length direction of the cultural heritage building.

[0012] Furthermore, the AI ​​simulation modeling and detection result fusion module generates the simulated 3D model in the following way: the target matching point set is reconstructed by Poisson reconstruction to obtain a preliminary simulated 3D model; the preliminary simulated 3D model is smoothed by a global optimization algorithm based on the feature vector set to obtain an optimized surface; the optimized surface is shaped by an iterative nearest point algorithm based on the matching accuracy value to obtain the simulated 3D model of the object.

[0013] Furthermore, the AI ​​simulation modeling and detection result fusion module generates the simulation texture model in the following way: The UV unwrapping algorithm is used to calculate the texture coordinates of the simulation 3D model to obtain texture coordinate data, normal vectors, and texture lighting; the Laplacian smoothing algorithm is used to perform preliminary texture mapping on the texture coordinate data to obtain a preliminary simulation texture model; a high dynamic range imaging algorithm is used to enhance the texture details of the preliminary simulation texture model based on the normal vectors to obtain a detail-enhanced simulation texture model; the Hough transform illumination correction algorithm is used to correct the illumination of the detail-enhanced simulation texture model to obtain an illumination-corrected simulation texture model; and a texture optimization algorithm based on a multi-scale convolutional neural network is used to optimize the illumination-corrected simulation texture model to obtain the final simulation texture model.

[0014] Furthermore, the defect classification method of the AI ​​simulation modeling and detection result fusion module is as follows: principal component analysis is applied to extract features from the simulated 3D model and the simulated texture model to obtain feature extraction data; the feature extraction data is fused using a weighted average method to obtain fused feature data; the fused feature data is input into a preset neural network model for preliminary classification to obtain preliminary classification results; the confidence of the preliminary classification results is calculated using Bayesian inference to obtain confidence data; the preliminary classification results are optimized using an optimization algorithm to obtain optimized classification results; the optimized classification results are verified for accuracy using cross-validation to obtain accuracy verification data; and the accuracy verification data is corrected for category using post-processing techniques to obtain corrected defect classification results.

[0015] A high-precision 3D scanning method based on AI technology fusion includes the following steps: S1. Laser scanning is performed on the preset collection points of the cultural heritage buildings through the data acquisition module to obtain the original three-dimensional point cloud data; S2. The original 3D point cloud data is subjected to multi-stage noise reduction processing through the AI ​​data preprocessing module to obtain the noise-reduced point cloud data. The noise-reduced point cloud data is input into the preset point cloud segmentation model, which divides the data into several detection areas according to the building structure type, and extracts feature point cloud data at the same time. S3. The structural feature analysis module calculates morphological fluctuation parameters based on the point cloud data of the detection area to screen key structural regions; the arc characterization coefficient is determined by the angle between the normal vector of the feature sub-region and the laser axis, thereby determining the structural morphological tendency category. S4. The adaptive laser detection module adjusts the laser incident direction or simultaneously adjusts the laser scanning path and incident direction according to the structural morphology category to accurately scan the key areas of the structure. S5. Based on the scanned point cloud data, the AI ​​simulation modeling and detection result fusion module performs 3D reconstruction and texture optimization to generate a simulated 3D model and a simulated texture model. The structural defect detection results are marked in the model, and the defect classification is completed by the AI ​​visual recognition system to output a high-precision model of fused detection results.

[0016] This invention provides a high-precision 3D scanner system and method based on AI technology fusion, comprising the following advantages: 1. The present invention significantly improves data quality by employing a multi-stage AI noise reduction algorithm, which effectively eliminates environmental noise and equipment errors, solves the problem of high data noise in the scanning of cultural heritage buildings, and provides a high-quality data foundation for subsequent detection and modeling. 2. The detection accuracy of this invention is greatly improved. Key areas are accurately identified through structural feature analysis, and the laser detection strategy is adaptively adjusted according to the morphological tendency category to avoid insufficient accuracy caused by one-size-fits-all scanning, thus achieving accurate detection of complex structures. 3. The invention has a more comprehensive functional coverage: it simultaneously meets the needs of high-precision digital archiving and structural safety inspection of cultural heritage buildings; the simulated 3D model and texture model can realize the permanent archiving of building form; and the defect detection results provide a scientific basis for protection and restoration. 4. This invention effectively improves detection efficiency by reducing invalid scanning time through key area screening and optimizing the scanning path and incident direction with an adaptive detection strategy. It improves detection efficiency while ensuring accuracy and is suitable for scanning and detecting large and complex cultural heritage buildings. Detailed Implementation

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] This invention provides a high-precision 3D scanner system based on AI technology fusion, comprising: The data acquisition module employs a high-precision laser scanning device. Based on precision adaptation logic, it sets preset acquisition points to scan key areas of cultural heritage buildings, such as beams, brackets, walls, and decorative components. Higher precision requirements result in smaller acquisition point intervals, ranging from [20, 50] cm, with 30 cm being preferred. For intricate structures like wood carvings and brick carvings, the interval can be further reduced to 10-20 cm. For ultra-fine structures like carvings and openwork, the interval can be reduced to 5-10 cm to ensure the integrity and precision of the acquired data. The scanning device emits a laser beam and measures the laser's round-trip time, combining this with angle information to calculate the three-dimensional coordinates of the target point and obtain raw three-dimensional point cloud data.

[0019] The AI ​​data preprocessing module, connected to the data acquisition module, is used to perform multi-stage noise reduction on the original 3D point cloud data to obtain noise-reduced point cloud data; the noise-reduced point cloud data is input into a preset point cloud segmentation model, which is divided into several detection areas according to the building structure type, and feature point cloud data is extracted at the same time. Specifically, the multi-stage denoising process of the AI ​​data preprocessing module includes: using Gaussian filtering to perform preliminary denoising on the original 3D point cloud data, removing obvious noise points caused by light reflection and surface dust; obtaining preliminary denoised point cloud data; adjusting the point cloud distribution through a mean filtering algorithm to make the distance between each point more uniform and avoid local point cloud density or sparseness; using a region growing algorithm to remove irregular or isolated points based on the similarity of the point cloud, obtaining finely denoised point cloud data; using the Boltzmann method to estimate the normals of the finely denoised point cloud data, obtaining a set of normal vectors, and then using nonlocal mean filtering to optimize the stability of the normal direction, obtaining smooth point cloud data; using nonlocal mean filtering to perform edge preservation processing on the smooth point cloud data, removing redundant noise while avoiding blurring the edge features of the building structure; and fusing the edge-preserved denoised point cloud data to obtain high-quality denoised point cloud data. The denoised point cloud data is input into a preset point cloud segmentation model (such as PointNet++), and segmented into several detection areas based on the building structure type (beam frame, wall, decorative components). The preferred size of the detection area is 2m×2m, which can be adjusted to 0.5m×0.5m for fine structures. At the same time, feature point cloud data, such as crack edges, structural splicing points, and decorative pattern outlines, are extracted.

[0020] The structural feature analysis module, connected to both the data acquisition module and the AI ​​data preprocessing module, calculates morphological fluctuation parameters based on the segmented point cloud data of the detection area to screen key structural regions. It determines the radian characterization coefficient by the angle between the normal vector of the feature sub-region and the laser axis, thereby determining the structural morphological tendency category. The core function of the structural feature analysis module is to complete the screening of key areas and morphological category determination of cultural heritage building structures through precise parameter calculation, dynamic calibration, and region adaptation, providing data support for subsequent adaptive laser detection. The specific implementation logic is as follows: The parameter calibration unit is used to dynamically set and calibrate morphological fluctuation reference values ​​and variance thresholds to ensure that the parameters are adapted to the structural characteristics of different types of cultural heritage buildings. Regarding the setting and calibration of morphological fluctuation reference values, the buildings are first categorized into five types: wooden structures, brick and stone structures, grotto structures, modern concrete structures, and finely decorated structures. At least 30 sets of historical scanning data are collected for each type of building, and the average fluctuation value μi of the structural width for each type of building is calculated. For example, the average fluctuation value μ1 for the beam frame of a wooden structure is 5cm, the average fluctuation value μ2 for the wall of a brick and stone structure is 8cm, and the average fluctuation value μ3 for finely decorated structures is 2cm. Then, the fluctuation coefficient k is set according to the required detection accuracy. The higher the accuracy requirement, the smaller the value of k. The value range of k is [0.5, 0.5]. [7] The preferred value is 0.6. The initial morphological fluctuation reference value of various types of buildings is obtained by multiplying μi and k, that is, the reference value of the beam frame of the wooden building is 5×0.6=3cm, the reference value of the wall of the brick and stone building is 8×0.6=4.8cm, and the reference value of the finely decorated building is 2×0.6=1.2cm. In the actual detection process, the fluctuation mean μi is updated after every 10 groups of the same type of building are detected, and the reference value is dynamically adjusted. If the missed detection rate of the key area in the detection results exceeds 5% or the false detection rate exceeds 8%, the k value is recalibrated. When the missed detection rate is too high, the k value is reduced, and when the false detection rate is too high, the k value is increased. Regarding the setting and calibration of variance thresholds, at least 50 sets of arc characterization coefficient samples of key areas of similar building structures need to be collected, and the sample variance σ² needs to be calculated. For example, the sample variance σ1² = 0.02 for the brackets of wooden buildings, σ2² = 0.05 for the surface of grotto buildings, and σ3² = 0.01 for fine carvings. An adjustment coefficient λ is set, with a value range of [1.2, 1.5]. The higher the complexity of the structural surface, the larger the value of λ. For brackets, λ1 = 1.3, λ2 = 1.5 for the surface of grottoes, and λ3 = 1.2 for carvings. The variance threshold is obtained by multiplying σ² and λ, i.e., the threshold for brackets = 0.02 × 1.3 = 0.026, the threshold for the surface of grottoes = 0.05 × 1.5 = 0.075, and the threshold for carvings = 0.01 × 1.2 = 0.012. The calibration process is consistent with the morphological fluctuation reference value, and adjustments are made based on feedback from actual test results.

[0021] The region division adaptation unit is used to dynamically adjust the size of the feature sub-regions according to the fineness of the building structure. It is preset with three levels of division standards: regular (30cm×30cm), fine (10cm×10cm), and ultra-fine (5cm×5cm). The regular level is suitable for areas with relatively flat structural surfaces and few details, such as the main beams of wooden buildings, the walls of brick and stone buildings, and the walls of modern concrete buildings. The fine level is suitable for areas with rich structural details and moderate curvature changes, such as the brackets of wooden buildings, the reliefs of brick and stone buildings, and the main body of grotto statues. The ultra-fine level is suitable for areas with extremely fine structures and tiny details, such as wood carvings, brick carvings, metal decorative patterns, and miniature components. This unit features automatic adaptation logic. The system uses AI image recognition to initially determine the building structure type and level of detail, and automatically matches the corresponding classification standard. For example, when scanning a wall, it automatically selects the regular level; when scanning a bracket, it automatically switches to the fine level; and when scanning a carving, it automatically switches to the ultra-fine level. At the same time, users can manually adjust the classification size through the operation interface. The adjustment range is [3cm×3cm, 50cm×50cm]. After classification, the system automatically counts the number of point clouds in each feature sub-region. If the number of point clouds is less than 50, the classification size is reduced; if it is more than 300, the classification size is increased to ensure that the number of point clouds in each sub-region is between 80 and 300, thus ensuring the accuracy of feature extraction.

[0022] In the specific feature analysis process, the calculation of morphological fluctuation parameters is based on the segmented point cloud data of the detection area. The width of the building structure (such as beam width and wall thickness) is determined by point cloud segmentation. The maximum width (first morphological factor) and minimum width (second morphological factor) of each detection area are extracted, and the difference between the two is the morphological fluctuation parameter. This parameter can intuitively reflect the degree of change in the structural morphology within the detection area. The screening of key structural areas involves comparing the morphological fluctuation parameter with the morphological fluctuation reference value set by the parameter calibration unit. If the morphological fluctuation parameter exceeds the reference value, the area is determined to be a key structural area, i.e., a stress concentration or damage-prone location. The calculation of the curvature characterization coefficient requires first dividing the key structural areas into regions. The dimensions determined by the adaptation unit are divided into several feature sub-regions. The normal vector of each feature sub-region is determined by local surface fitting (such as the least squares method). The laser axis is determined by the centroid and principal direction analysis of the point cloud. The angle between the normal vector and the laser axis is the radian characterization coefficient, which is used to characterize the local curvature of the feature sub-region. The determination of the morphological tendency category is to calculate the variance of the radian characterization coefficients of all feature sub-regions in the key area of ​​the structure and compare it with the variance threshold set by the parameter calibration unit. If the variance does not exceed the threshold, it indicates that the curvature of the structure surface changes gently and is determined to be the first morphological tendency category. If the variance exceeds the threshold, it indicates that the curvature of the structure surface changes significantly and the curvature of each sub-region is large, and is determined to be the second morphological tendency category.

[0023] An adaptive laser detection module, connected to the data acquisition module, AI data preprocessing module, and structural feature analysis module, is used to adaptively adjust the laser incident direction or simultaneously adjust the laser scanning path and incident direction according to the structural morphology tendency category. The adjustment method of the adaptive laser detection module is as follows: If the structural morphology tendency category is the first morphology tendency category, for key areas of the structure with gentle curvature changes, determine the beam frame morphology fluctuation area (the area enclosed by the line segment of the first morphology factor, the line segment of the second morphology factor, and the edge contour of the building length direction); calculate the vector sum of all normal vectors in this area to obtain the reference normal vector; adjust the laser incident direction to be parallel to the reference normal vector to ensure that the laser is perpendicular to the structural surface, reduce distance measurement deviation and data noise, and improve detection accuracy; If the structural morphology tendency category is the second morphology tendency category, for key areas of the structure with significant curvature changes, first calculate the angle between the normal vector of each feature sub-region and the horizontal plane, sort them according to the angle values ​​from small to large or from large to small, determine the scanning path, and avoid frequent and large changes in the laser incident direction; at the same time, adjust the laser incident direction of each feature sub-region to be parallel to its normal vector to ensure the highest laser energy reflection efficiency, stable reflection signal, and accurate acquisition of detection data for each sub-region.

[0024] The reference normal vector is the vector obtained by adding several normal vectors within the key structural region; the key structural region is the area in the building structure where the morphological fluctuation parameter exceeds the preset fluctuation reference value, specifically the area enclosed by the line segment of the first morphological factor, the line segment of the second morphological factor, and the edge contour along the length direction of the cultural heritage building.

[0025] The AI ​​simulation modeling and detection result fusion module is connected to the AI ​​data preprocessing module, structural feature analysis module, and adaptive laser detection module, respectively. It is used to perform 3D reconstruction and texture optimization based on the detected point cloud data to generate a simulated 3D model and a simulated texture model. The structural defect detection results are marked in the model, and the defect classification is completed by the AI ​​visual recognition system to output a high-precision model of fused detection results.

[0026] Specifically, the AI ​​simulation modeling and detection result fusion module generates the simulated 3D model as follows: Feature matching is performed on the point cloud data after adaptive laser detection. Geometric feature parameter sets and feature point cloud data are processed through min-max normalization and mean-variance normalization. The target matching point set is obtained through feature descriptor algorithm description and RANSAC algorithm filtering. The feature vector set is calculated using the K-nearest neighbor algorithm, and the matching accuracy value is obtained based on the root mean square error. Based on the target matching point set, surface reconstruction is performed using Poisson reconstruction to obtain a preliminary simulated 3D model. The preliminary model is smoothed using a global optimization algorithm, and the shape is optimized based on the matching accuracy value using an iterative nearest point algorithm to obtain a high-precision simulated 3D model. For ultra-fine structural regions, the surface subdivision density is increased during reconstruction to ensure detail reproduction.

[0027] Specifically, the AI ​​simulation modeling and detection result fusion module generates the simulation texture model in the following way: The UV unwrapping algorithm is used to calculate the texture coordinates of the simulation 3D model to obtain texture coordinate data, normal vectors, and texture lighting; a preliminary texture mapping is performed using the Laplacian smoothing algorithm to obtain a preliminary simulation texture model; based on the normal vectors, a high dynamic range imaging algorithm is used to enhance texture details; the Hough transform lighting correction algorithm is used to adjust the lighting effect; finally, a multi-scale convolutional neural network is used to optimize the texture to obtain a high-fidelity simulation texture model; for areas with fine decoration, the texture sampling density is increased to ensure that the texture details are consistent with the actual structure.

[0028] Specifically, the defect classification method of the AI ​​simulation modeling and detection result fusion module is as follows: principal component analysis is applied to extract features from the simulated 3D model and the simulated texture model. The feature data is then fused using a weighted average method and input into a neural network model for preliminary classification. The confidence level is calculated using Bayesian inference, the classification results are optimized using an optimization algorithm, cross-validation is performed to verify accuracy, and post-processing techniques are used for correction to obtain the corrected classification results for structural defects. The defect location, level, type, and other information are labeled in the simulated 3D model. Combined with the simulated texture model, a high-precision model and defect detection report are output to support the digital archiving of cultural heritage buildings and subsequent structural monitoring.

[0029] A high-precision 3D scanning method based on AI technology fusion includes the following steps: S1. Data Acquisition: Using high-precision laser scanning equipment, the preset acquisition points of cultural heritage buildings are scanned to obtain raw three-dimensional point cloud data. The acquisition points are reasonably distributed according to the building structure type and accuracy requirements. The acquisition is intensified in fine structure areas and the acquisition interval is further reduced in ultra-fine structure areas.

[0030] S2. AI Data Preprocessing: The original 3D point cloud data is subjected to preliminary denoising, density homogenization, fine denoising, normal optimization, edge preservation denoising and data fusion in sequence to obtain denoised point cloud data; the denoised point cloud data is input into the point cloud segmentation model, segmented into several detection areas according to the building structure type, and feature point cloud data is extracted.

[0031] S3. Structural Feature Analysis: S31. Parameter setting and calibration: Through the parameter calibration unit, based on the current scanned cultural heritage building type (such as wooden building or brick and stone building), the mean fluctuation value μi and adjustment coefficients k and λ in the preset parameter table are called to calculate and set the morphological fluctuation reference value and variance threshold; if it is a new type of building, at least 30 sets of sample data are collected first to calculate the initial parameters before detection. S32. Feature Sub-region Division: Through the region division adaptation unit, the feature sub-region division size is automatically matched according to the building structure refinement, or the size is adjusted by manual intervention, and the system automatically optimizes the number of point clouds in the sub-region. S33. Key Region Screening and Morphological Classification: Calculate the morphological fluctuation parameters of each detection region and compare them with the morphological fluctuation reference values ​​to screen key structural regions; calculate the arc characterization coefficient of each feature sub-region and determine the structural morphological tendency category by comparing the variance with the variance threshold.

[0032] S4. The adaptive laser detection module adjusts the laser incident direction or simultaneously adjusts the scanning path and incident direction according to the structural morphology category to accurately scan key areas of the structure and obtain high-quality detection data. For ultra-fine structural areas, the laser scanning resolution is optimized to ensure the accuracy of detail detection.

[0033] S5. Based on the scanned point cloud data, the AI ​​simulation modeling and detection result fusion module generates a simulated 3D model and a simulated texture model. The ultra-fine structural region adopts high subdivision density reconstruction and high sampling density texture mapping. The AI ​​visual recognition system completes the classification of structural defects, fuses the detection results with the model, and outputs a high-precision model and detection report of the fused detection results.

[0034] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A high-precision 3D scanner system based on AI technology fusion, characterized in that, include: The data acquisition module is used to perform laser scanning on preset acquisition points of cultural heritage buildings to obtain raw three-dimensional point cloud data; The AI ​​data preprocessing module, connected to the data acquisition module, is used to perform multi-stage noise reduction on the original 3D point cloud data to obtain noise-reduced point cloud data; the noise-reduced point cloud data is input into a preset point cloud segmentation model, which is divided into several detection areas according to the building structure type, and feature point cloud data is extracted at the same time. The structural feature analysis module is connected to the data acquisition module and the AI ​​data preprocessing module respectively. It is used to calculate morphological fluctuation parameters based on the segmented point cloud data of the detection area to screen key structural regions. The arcuate characterization coefficient is determined by the angle between the normal vector of the feature sub-region and the laser axis, thereby determining the structural morphology tendency category; An adaptive laser detection module is connected to the data acquisition module, the AI ​​data preprocessing module and the structural feature analysis module respectively, and is used to adaptively adjust the laser incident direction or simultaneously adjust the laser scanning path and incident direction according to the structural morphology tendency category. The AI ​​simulation modeling and detection result fusion module is connected to the AI ​​data preprocessing module, structural feature analysis module, and adaptive laser detection module, respectively. It is used to perform 3D reconstruction and texture optimization based on the detected point cloud data to generate a simulated 3D model and a simulated texture model. The structural defect detection results are marked in the model, and the defect classification is completed by the AI ​​visual recognition system to output a high-precision model of fused detection results.

2. The high-precision 3D scanner system based on AI technology fusion according to claim 1, characterized in that, The multi-stage noise reduction process of the AI ​​data preprocessing module includes: The original 3D point cloud data undergoes preliminary denoising to obtain preliminary denoised point cloud data. Then, a mean filtering algorithm is used to homogenize the point cloud density, resulting in homogenized point cloud data. Next, a region growing algorithm is used to further refine the denoising, yielding finely denoised point cloud data. The Boltzmann method is then used to estimate the normals of the finely denoised point cloud data, resulting in a set of normal vectors. This set of normal vectors is then processed using nonlocal mean filtering to obtain smoothed point cloud data. Finally, edge-preserving denoising is applied to the smoothed point cloud data using nonlocal mean filtering, resulting in edge-preserving denoised point cloud data. Finally, the edge-preserving denoised point cloud data is fused to obtain the denoised point cloud data.

3. The high-precision 3D scanner system based on AI technology fusion according to claim 1, characterized in that, The structural feature analysis module calculates the morphological fluctuation parameters in the following way: Based on the point cloud data of the detection area, determine the width of several building structures, and determine the maximum and minimum widths of the building structures within the detection area as the first morphological factor and the second morphological factor, respectively; calculate the difference between the first morphological factor and the second morphological factor, and determine the difference as the morphological fluctuation parameter; If the morphological fluctuation parameters exceed the preset fluctuation reference value, the detection area will be selected as a critical structural area.

4. The high-precision 3D scanner system based on AI technology fusion according to claim 1, characterized in that, The structural feature analysis module determines the structural morphological tendency category in the following way: The key structural region is divided into several feature sub-regions. The normal vector and laser axis of each feature sub-region are determined based on point cloud data. The angle between the direction of the normal vector and the direction of the laser axis is calculated, and the angle is determined as the radian characterization coefficient of the feature sub-region. If the variance of several radian characterization coefficients in the key structural region does not exceed a preset variance threshold, it is determined as the first morphological tendency category; otherwise, it is determined as the second morphological tendency category.

5. A high-precision 3D scanner system based on AI technology fusion according to claim 4, characterized in that, The adjustment method of the adaptive laser detection module is as follows: If the structural morphology tendency category is the first morphology tendency category, a reference normal vector is determined based on several normal vectors within the key area of ​​the structure, and the laser incident direction for laser scanning of the key area of ​​the structure is adjusted, wherein the laser incident direction is parallel to the reference normal vector; if the structural morphology tendency category is the second morphology tendency category, the angle between the normal vector of the feature sub-region and the horizontal plane is calculated, and the scanning path is determined by sorting the angle values, while the laser incident direction is adjusted to be parallel to the normal vector of each feature sub-region.

6. A high-precision 3D scanner system based on AI technology fusion according to claim 5, characterized in that, The reference normal vector is the vector obtained by adding several normal vectors within the key structural region; the key structural region is the area in the building structure where the morphological fluctuation parameter exceeds the preset fluctuation reference value, specifically the area enclosed by the line segment of the first morphological factor, the line segment of the second morphological factor, and the edge contour along the length direction of the cultural heritage building.

7. A high-precision 3D scanner system based on AI technology fusion according to claim 1, characterized in that, The AI ​​simulation modeling and detection result fusion module generates the simulated 3D model in the following way: the target matching point set is reconstructed by Poisson reconstruction to obtain a preliminary simulated 3D model; the preliminary simulated 3D model is smoothed by a global optimization algorithm based on the feature vector set to obtain an optimized surface; the optimized surface is shaped by the iterative nearest point algorithm based on the matching accuracy value to obtain the simulated 3D model of the object.

8. A high-precision 3D scanner system based on AI technology fusion according to claim 1, characterized in that, The AI ​​simulation modeling and detection result fusion module generates the simulation texture model in the following way: The UV unwrapping algorithm is used to calculate the texture coordinates of the simulation 3D model to obtain texture coordinate data, normal vectors, and texture lighting; the Laplacian smoothing algorithm is used to perform preliminary texture mapping on the texture coordinate data to obtain a preliminary simulation texture model; a high dynamic range imaging algorithm is used to enhance the texture details of the preliminary simulation texture model based on the normal vectors to obtain a detail-enhanced simulation texture model; the Hough transform lighting correction algorithm is used to correct the lighting of the detail-enhanced simulation texture model to obtain a lighting-corrected simulation texture model; and a texture optimization algorithm based on a multi-scale convolutional neural network is used to optimize the lighting-corrected simulation texture model to obtain the final simulation texture model.

9. A high-precision 3D scanner system based on AI technology fusion according to claim 1, characterized in that, The defect classification method of the AI ​​simulation modeling and detection result fusion module is as follows: principal component analysis is applied to extract features from the simulation 3D model and the simulation texture model to obtain feature extraction data; The feature extraction data is fused using a weighted average method to obtain fused feature data. The fused feature data is input into a preset neural network model for preliminary classification to obtain preliminary classification results. The confidence of the preliminary classification results is calculated using Bayesian inference to obtain confidence data. The preliminary classification results are optimized using an optimization algorithm to obtain optimized classification results. The optimized classification results are then verified for accuracy using cross-validation to obtain accuracy verification data. Finally, the accuracy verification data is corrected for categories using post-processing techniques to obtain corrected classification results for defects.

10. A high-precision 3D scanning method based on AI technology fusion as described in any one of claims 1-9, characterized in that, Includes the following steps: S1. Laser scanning is performed on the preset collection points of the cultural heritage buildings through the data acquisition module to obtain the original three-dimensional point cloud data; S2. The original 3D point cloud data is subjected to multi-stage noise reduction processing through the AI ​​data preprocessing module to obtain the noise-reduced point cloud data. The noise-reduced point cloud data is input into the preset point cloud segmentation model, which divides the data into several detection areas according to the building structure type, and extracts feature point cloud data at the same time. S3. The structural feature analysis module calculates morphological fluctuation parameters based on the point cloud data of the detection area to screen key structural regions. The arcuate characterization coefficient is determined by the angle between the normal vector of the feature sub-region and the laser axis, thereby determining the structural morphology tendency category; S4. The adaptive laser detection module adjusts the laser incident direction or simultaneously adjusts the laser scanning path and incident direction according to the structural morphology category to accurately scan the key areas of the structure. S5. Based on the scanned point cloud data, the AI ​​simulation modeling and detection result fusion module performs 3D reconstruction and texture optimization to generate a simulated 3D model and a simulated texture model. The structural defect detection results are marked in the model, and the defect classification is completed by the AI ​​visual recognition system to output a high-precision model of fused detection results.