High-precision real-scene three-dimensional reconstruction method for building heritage

Abstract

The invention provides a high-precision live-action three-dimensional reconstruction method for building heritage, and relates to the technical field of data processing, and the method comprises the steps: obtaining image data, and dividing a to-be-reconstructed region into a plurality of sub-regions according to the overall structure of the building heritage; carrying out point cloud acquisition on each sub-region by utilizing acquisition equipment, adopting a first sampling interval in a conventional region, and dynamically shortening to a second sampling interval in an engraved concave or narrow space region; performing coarse registration processing, and executing iterative correction by using the feature corresponding points of the overlapping region to generate a first sub-region point cloud model; quality detection is carried out, and a missing area with the point cloud density not meeting the density judgment standard is determined; extracting local edge contour and gray gradient information in combination with the image data, and projecting the local edge contour and the gray gradient information to the missing area to generate a compensation point; carrying out weighted fusion processing, and executing curved surface interpolation in a fusion region; performing boundary feature alignment and global splicing to obtain a three-dimensional reconstruction model; according to the invention, the accuracy of three-dimensional reconstruction is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a high-precision real scene three-dimensional reconstruction method for building heritage. BACKGROUND

[0002] The existing real scene three-dimensional reconstruction method for building heritage usually adopts the way of unmanned aerial vehicle aerial image combined with ground photogrammetry to generate point cloud data, and realizes the splicing of the overall model through the conventional three-dimensional reconstruction algorithm. This kind of method can obtain a relatively complete external contour in large-scale building appearance reconstruction, but in the collection of complex structure or internal space, it often relies on fixed interval sampling strategy and single path scanning mode to construct point cloud data.

[0003] However, in specific applications, such as the reconstruction of historical buildings with a large number of carved details or narrow passages, the existing technology often causes insufficient point cloud density in recessed parts due to fixed sampling interval, resulting in local missing or blurred details. For example, in the arch or stone carving patterns, insufficient sampling points will cause the edge features to be difficult to reproduce completely, so that the final spliced three-dimensional model appears hollow or surface discontinuity in the local area. This situation not only affects the accuracy and authenticity of the overall three-dimensional model, but also causes the loss and misdirection of historical information when used for subsequent cultural heritage digital protection or virtual display. SUMMARY

[0004] The purpose of the present application is to provide a high-precision real scene three-dimensional reconstruction method for building heritage, which aims to solve the problems mentioned in the background.

[0005] To solve the above technical problems, the technical scheme of the present application is as follows: The high-precision real scene three-dimensional reconstruction method for building heritage comprises: Obtaining multi-angle image data of the exterior and interior of the building heritage, and dividing the area to be reconstructed into multiple sub-regions according to the overall structure of the building heritage to obtain image data; According to the image data, point cloud collection is performed on each sub-region using a collection device, a first sampling interval is used in the conventional region, a second sampling interval is dynamically shortened in the carved recessed or narrow space region, and a multi-path cross collection method is used to generate sub-region point cloud data; Performing coarse registration processing on the sub-region point cloud data, and performing iterative correction on the corresponding points of the features in the overlapping region to generate a first sub-region point cloud model; Performing quality detection on the first sub-region point cloud model, identifying the point cloud density distribution and integrity, and determining the missing area where the point cloud density does not meet the density judgment standard; In the missing area, the local edge contour and gray gradient information are extracted by combining the image data, the information is projected to the missing area to generate compensation points, and a second sub-area point cloud model is generated; The second sub-area point cloud model is subjected to weighted fusion processing, and a curved surface interpolation is performed in the fusion area to form a sub-area three-dimensional model; The sub-area three-dimensional model is subjected to boundary feature alignment and global splicing, and smooth fusion is realized through a multi-scale transition surface to obtain a high-precision real scene three-dimensional reconstruction model covering the whole building heritage.

[0006] The above scheme of the present application at least includes the following beneficial effects: Firstly, by acquiring multi-angle image data of the external and internal of the building heritage, and dividing the to-be-reconstructed area into multiple sub-areas according to the overall structure, the reconstruction object can be differentiated before collection. Compared with the fixed interval sampling method in the prior art, this division strategy based on image data can ensure the integrity of the overall framework in a large-scale area, and realize fine modeling in complex areas or parts with rich details, thereby avoiding the coexistence of data redundancy and local missing. For example, for the flat wall of the palace facade, it can be divided into a large-scale sub-area to improve the collection efficiency, and for the complex details such as arches and stone carvings, it can be divided into a small-scale sub-area to enhance the point cloud density.

[0007] Then, by using the collection device to collect point clouds in the sub-area, the first sampling interval is used in the conventional area, and the second sampling interval is dynamically shortened in the carved recess or narrow space area, and combined with the multi-path cross collection method, the uniformity and integrity of the point cloud data can be effectively improved. This dynamic sampling combined with multiple paths overcomes the limitations of traditional fixed interval and single path, ensuring that high-density point clouds can still be obtained in complex curved surfaces or occluded spaces. For example, in the interior of the cloister of the ancient temple, by using the dynamically shortened sampling interval and cross-path collection, the missing data of the corner caused by occlusion can be avoided.

[0008] Secondly, by performing rough registration on the sub-area point cloud data and executing iterative correction on the feature corresponding points of the overlapping area, the splicing error between the data collected by different paths can be significantly reduced. Compared with the conventional registration algorithm in the prior art, the multi-stage registration strategy of the present application can ensure that the sub-area point cloud model has higher accuracy when globally spliced, thereby balancing between detail expression and overall structure preservation. For example, in the multi-angle scanning of the stone tower, the iterative correction can gradually eliminate the misplacement, so that the corner structure of the tower body is complete and clear.

[0009] Further, by performing quality detection on the sub-region point cloud model, identifying the point cloud density distribution and integrity situation, and determining the missing region that does not meet the density judgment standard, potential defects can be found and marked in advance. This detection method avoids exposing the problem of holes after the final model is generated, so that dynamic repair can be performed during modeling. For example, in the dome part of the cave murals, if the point cloud is sparse due to occlusion, the subsequent compensation operation can be triggered by detecting and finding it in the detection stage.

[0010] Finally, by combining image data in the missing region to extract local edge contour and gray gradient information, and projecting the information to generate compensation points, and then performing weighted fusion and surface interpolation on the compensated point cloud model, the boundary feature alignment and smooth splicing can be completed in the global range, and a high-precision real scene three-dimensional reconstruction model covering the whole building heritage can be formed. This modeling mechanism based on the complementarity of image and point cloud can ensure the continuity of large-scale structure and the authenticity of small-scale details at the same time. For example, in virtual display, the audience can not only feel the complete contour of the palace roof, but also observe the exquisite texture of the eave carving, avoiding the information loss and misdirection caused by point cloud missing in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 is a flowchart of a building heritage high-precision real scene three-dimensional reconstruction method provided by an embodiment of the present application. DETAILED DESCRIPTION

[0012] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0013] As shown in Figure 1 An embodiment of the present application proposes a building heritage high-precision real scene three-dimensional reconstruction method, which comprises: acquiring multi-angle image data of the building heritage outside and inside, and dividing the to-be-reconstructed region into a plurality of sub-regions according to the overall structure of the building heritage to obtain image data; According to the image data, the point cloud of each sub-region is collected by using a collection device, a first sampling interval is used in a conventional region, a second sampling interval is dynamically shortened in a sculpture recess or narrow space region, and a multi-path cross collection method is used to generate sub-region point cloud data; performing coarse registration processing on the sub-region point cloud data, and performing iterative correction by using the corresponding points of the features of the overlapping region to generate a first sub-region point cloud model; Quality detection is performed on the first sub-region point cloud model to identify the point cloud density distribution and integrity, and to determine a missing region in which the point cloud density does not meet the density judgment standard; In the missing region, local edge contour and gray gradient information are extracted in combination with image data, and the information is projected to the missing region to generate compensation points, thereby generating a second sub-region point cloud model; The second sub-region point cloud model is subjected to weighted fusion processing, and surface interpolation is performed in the fusion region to form a sub-region three-dimensional model; The sub-region three-dimensional model is subjected to boundary feature alignment and global splicing, and smooth fusion is achieved through a multi-scale transition surface, thereby obtaining a high-precision real-scene three-dimensional reconstruction model covering the entire building heritage.

[0014] In the embodiment of the present application, multi-angle image data of the building heritage outside and inside is acquired, and the to-be-reconstructed region is divided into sub-regions in combination with the overall structure, so that a comprehensive and hierarchical spatial data basis can be established in the initial stage, and data redundancy or local collection deficiency caused by overall scanning can be avoided. Large-scale division is adopted for the external facade, and small-scale division is adopted for the carved details or recessed regions, so that differential processing is achieved.

[0015] Then, point cloud collection is performed based on the sub-region information using a collection device, a first sampling interval is used in a conventional region, a second sampling interval is dynamically shortened in a complex or narrow region, and multi-path cross collection is combined, so that the integrity and density distribution of the point cloud are improved, the overall structure is preserved, and the details are clear.

[0016] Secondly, rough registration is performed on the sub-region point cloud data, and iterative correction is performed on the feature corresponding points of the overlapping region, so that the error caused by the path difference is gradually reduced, and a reliable sub-region point cloud model is generated.

[0017] Further, point cloud density deficiency or missing regions are identified through quality detection, so that holes in the final model can be avoided, and a basis for subsequent compensation is provided.

[0018] Subsequently, in the missing region, edge contour and gray gradient information are extracted in combination with image data, and the information is projected to generate compensation points, so that a complete point cloud model is constructed, and the complementarity of two-dimensional texture and three-dimensional data is achieved.

[0019] Then, the compensated point cloud model is subjected to weighted fusion, and surface interpolation is performed in the sparse region to ensure surface continuity and smoothness and reduce redundant interference.

[0020] Finally, the boundary alignment and global splicing of the plurality of sub-region three-dimensional models are performed, and a multi-scale transition surface is introduced at the splicing position, so that overall continuity and local details are considered, thereby obtaining a high-precision three-dimensional model covering the entire building heritage.

[0021] The density determination criterion can be set according to the average characteristics of the point cloud distribution. For example, when the number of points in a certain grid unit is less than one-half of the global average number of points, it is determined as a low-density candidate region; further, when the local average distance between points exceeds twice the global average distance, the region is determined as a potential missing region; if there is an obvious break in the image texture feature comparison, it is finally determined as a missing region. Through the above multi-level standard setting, false positives and false negatives can be effectively avoided.

[0022] The multi-angle image data of the building heritage outside and inside is acquired, and the region to be reconstructed is divided into a plurality of sub-regions according to the overall structure of the building heritage, and the image data is obtained, specifically comprising: Firstly, multi-angle data of the building heritage outside and inside is acquired by using a plurality of image acquisition methods. The external image can be acquired by a high-resolution camera mounted on a collection device for aerial collection, and complete appearance data is acquired from multiple directions such as the top, side and inclined angle of the building; the internal image can be acquired by a ground camera, a panoramic camera or a mobile scanning device to acquire multi-angle images of internal spaces such as corridors, halls and vaults. Through this combined acquisition method, the external contour and internal details are ensured to be covered.

[0023] Then, the acquired image data is preprocessed, including distortion correction, brightness equalization and image stitching. The distortion correction is mainly used to eliminate the deformation caused by the camera lens, the brightness equalization is used to improve the difference in brightness caused by uneven lighting, and the image stitching is used to form a wider angle image scene under the same view angle.

[0024] Secondly, the region to be reconstructed is divided into a plurality of sub-regions based on the overall structural features of the building heritage. The division process depends on the extraction of structural features in the image, such as detecting the edges of the building outer wall, the ridge lines of the roof and the partition walls of the internal space to form a partition result with complete boundaries. In the division process, the regions can be differentiated according to the geometric complexity: regions with small curvature variation and flat geometric shape are divided into large-scale sub-regions, and regions with dense carving details and complex spatial curvature are divided into small-scale sub-regions.

[0025] Finally, the division result is associated with the corresponding image data, that is, an exclusive image data set is generated for each sub-region. This partitioned image data provides targeted basic data for subsequent point cloud acquisition and three-dimensional reconstruction, which can ensure the efficiency of the overall modeling and the accuracy of the acquisition of complex detail regions.

[0026] In a preferred embodiment of the present application, according to the image data, point cloud acquisition is performed on each sub-region by using a collection device, comprising: According to the acquired image data, the curvature characteristics of the building surface are extracted, the sub-region is divided into a curvature gentle region and a curvature mutation region, and a curvature division result is generated; According to the curvature division result, in the curvature gentle region, the acquisition device is controlled to perform image acquisition at a first motion speed and a first sampling interval, and first sampling point cloud data is obtained; According to the curvature division result, in the curvature mutation region, by monitoring the image gray scale change rate in real time, the acquisition device is triggered to reduce the motion speed, increase the sampling frequency, and shorten to a second sampling interval, and second sampling point cloud data is obtained; Based on the feedback of the obstacle avoidance sensor of the acquisition device in the narrow space, the posture and the collection angle of the acquisition device are adjusted, the compensation collection is performed on the shielding region, and compensation point cloud data is generated; The first sampling point cloud data, the second sampling point cloud data and the compensation point cloud data are merged, and the redundant points are weighted by the overlapping data generated by the multi-path cross scanning, and sub-region point cloud original data is obtained.

[0027] In the embodiment of the application, by extracting the curvature characteristics of the building surface based on the image data, the region is divided into a curvature gentle region and a curvature mutation region, differential control can be realized before acquisition, and sampling deficiency or redundancy can be avoided. In the gentle region, the first motion speed and the sampling interval are used, which can ensure the efficiency; in the mutation region, by monitoring the image gray scale change in real time, the acquisition parameters are dynamically adjusted and shortened to the second sampling interval, which can improve the point cloud density and ensure that the complex details are captured completely. At the same time, based on the feedback of the obstacle avoidance sensor, the posture and the scanning angle of the device are adjusted, the shielding region can be effectively covered, the different path data fusion and the redundant weighting are combined, and finally the complete and consistent sub-region point cloud original data is obtained.

[0028] Among them, the conventional region refers to a region with small curvature change and smooth geometric shape of the building surface, for example, a flat wall surface or a column surface; the carved recessed region refers to a part with obvious curvature change and complex geometric details, such as stone carving patterns, arch edge, etc.; the narrow space region is a space limited position determined by the feedback of the obstacle avoidance sensor, such as the inside of the passageway, the gap between the beams and columns or the back of the window lattice.

[0029] Among them, according to the acquired image data, the curvature characteristics of the building surface are extracted, the sub-region is divided into a curvature gentle region and a curvature mutation region, and a curvature division result is generated, which specifically includes: First, the surface feature extraction is performed on the image data corresponding to the sub-region. By analyzing the image gray scale change, the continuity and mutation of the surface shape are identified. The region with smooth gray scale distribution usually corresponds to the gentle part of the building surface, that is, the conventional region; and the region with sharp gray scale change is often related to the carving edge, the broken line or the complex texture.

[0030] Then, based on the extracted image features, curvature estimation is performed in combination with the preliminary point cloud data. The curvature estimation uses the geometric relationship between neighboring points for analysis, that is, by statistically analyzing the height difference and direction difference between neighboring pixels or point clouds, a parameter reflecting the degree of surface fluctuation is formed. The smaller the height difference between adjacent points, the flatter the surface; the greater the height difference and direction difference between adjacent points, the greater the surface curvature.

[0031] Secondly, according to the calculated curvature parameter, the sub-region is divided into a curvature gentle region and a curvature mutation region. The gentle region corresponds to a large area of flat structure of the building surface, such as the smooth surface of the outer wall facade or stone column; the curvature mutation region corresponds to the local complex part, such as the edge of the arch, the concave and convex area of the carved decoration or the small grid of the window lattice.

[0032] Finally, the curvature division result is generated and bound with the sub-region image data, providing a basis for the dynamic adjustment of the subsequent sampling interval and scanning strategy. Through this process, the complex areas can be optimized while maintaining the efficiency of large area acquisition.

[0033] Among them, according to the curvature division result, in the curvature gentle region, the acquisition device is controlled to perform image acquisition at a first motion speed and a first sampling interval to obtain first sampling point cloud data, specifically including: First, under the guidance of the curvature division result, the curvature gentle region is taken as the key area of motion path planning. For these regions, the control system generates motion instructions according to the set first motion speed to ensure that the acquisition device runs at a uniform speed above or around the region.

[0034] Then, during the motion process, the camera and the acquisition module are triggered for data acquisition according to the preset first sampling interval. The first sampling interval is a larger interval set based on the characteristics of the gentle region, which can ensure the completeness of the point cloud coverage and avoid unnecessary redundant sampling, thereby improving the overall acquisition efficiency.

[0035] Secondly, during the acquisition process, the motion control module adjusts the posture of the acquisition device in real time to ensure that the camera optical axis maintains a suitable angle with the region surface, avoiding the acquisition blind area caused by the angle deviation. For example, in the vertical plane region such as the palace outer wall, the control system will automatically adjust the pitch angle of the acquisition device to keep the camera at a vertical viewing angle.

[0036] Finally, the point cloud data and image data obtained through this process are preliminarily paired to generate first sampling point cloud data. These point cloud data cover the flat region of the building heritage, forming the basic framework of three-dimensional reconstruction, laying a data foundation for the subsequent fine acquisition of the curvature mutation region.

[0037] Wherein, based on the feedback of the obstacle avoidance sensor of the collection device in the narrow space, the posture and collection angle of the collection device are adjusted, the compensation collection is performed on the blocked area, and the compensation point cloud data is generated, specifically including: Firstly, when the collection device enters the narrow space or the area with obstacles, the obstacle avoidance sensor carried by the collection device starts to collect environmental information in real time. The obstacle avoidance sensor usually includes ultrasonic sensor, infrared sensor or laser radar, etc., which is used to detect the relative distance and direction between the collection device and the surrounding obstacles. The feedback information of the sensor can form a space obstacle distribution map, which provides a reference for the path adjustment of the collection device.

[0038] Then, the control system calculates the posture adjustment parameters of the collection device according to the obstacle distribution map. These parameters are used to control the roll angle, pitch angle and yaw angle of the collection device, so that it can continue to collect without colliding with obstacles. For example, when the collection device moves inside the arch-shaped passageway of ancient buildings, if the sensor detects that the top distance is too close, the control system will adjust the pitch angle of the collection device to avoid collision and ensure the camera view angle covers the area.

[0039] Secondly, combined with the obstacle information, the collection angle is dynamically adjusted to ensure that the blocked area can be effectively scanned. The adjustment method is to change the horizontal emission angle or vertical emission angle of the laser, so that the area that cannot be directly covered is compensated. For example, in the narrow space behind the carved window lattice, the laser can cover the original blind area through angle adjustment, so as to fill the point cloud data.

[0040] Finally, the data obtained in this process is marked as compensation point cloud data. The compensation point cloud not only makes up for the missing caused by obstacles or narrowness, but also enhances the overall spatial integrity when merging with normal point cloud data. For example, in the area under the beam structure inside the ancient temple, the three-dimensional information of the connection between the beam and the roof can be effectively restored through compensation collection.

[0041] Wherein, the first sampling point cloud data, the second sampling point cloud data and the compensation point cloud data are merged, and the redundant points are weighted through the overlapping data generated by the multi-path cross scanning to obtain the sub-area point cloud original data, specifically including: Firstly, the first sampling point cloud data obtained in the flat curvature area, the second sampling point cloud data obtained in the curvature mutation area, and the compensation point cloud data collected in the blocked area are data summarized. The summary process needs to unify the coordinate system and perform preliminary spatial alignment to ensure that the data from different paths and different ways can be superimposed in the same three-dimensional reference system.

[0042] Then, for data from different sources, redundant points existing in multi-path acquisition are identified through spatial overlap detection. Specifically, the spatial position difference between points in the same area is compared, and when the difference is less than the set distance threshold, it is determined as a redundant point.

[0043] Secondly, when processing redundant points, instead of directly deleting, optimization is performed through weighting. The weighting strategy is based on two factors: one is the sampling density of the point, that is, the local point cloud density of the area where the point is located, and the higher the density, the smaller the weight of the point; the second is the path confidence, that is, the stability and angle superiority of the acquisition path, and the higher the confidence, the greater the weight of the point. In this way, while ensuring that the data redundancy is compressed, the points with higher quality are preferentially retained.

[0044] Further, a fused point set is formed through the weighted processing, and a smoothing operation is performed on the fusion result to avoid surface roughness caused by small differences between point clouds from different sources. For example, in multi-path acquisition of a stone tower wall, point clouds obtained by different paths may show slight misalignment on the surface, and weighted fusion and smoothing can eliminate these inconsistencies.

[0045] Finally, the fused point set is used as the original data of the sub-area point cloud for subsequent registration and model generation. These original data not only ensure complete coverage, but also have high precision and high consistency, thereby providing a solid data foundation for subsequent three-dimensional reconstruction.

[0046] In a preferred embodiment of the present application, the sub-area point cloud data is subjected to coarse registration processing, and iterative correction is performed on the corresponding points of the feature points in the overlapping area to generate a first sub-area point cloud model, comprising: According to the sub-area point cloud original data, the geometric center and the normal vector distribution are calculated to form an initial pose parameter; According to the initial pose parameter, the point clouds collected by different paths are subjected to coarse registration to obtain preliminarily aligned sub-area point cloud data; According to the preliminarily aligned sub-area point cloud data, feature points in the overlapping area are extracted, and the feature points are selected according to the normal vector consistency and local curvature extreme value to obtain a corresponding point pair; According to the corresponding point pair, the Euclidean distance and angle difference between points are calculated to form an error distribution result; According to the error distribution result, iterative correction processing is performed to gradually update the pose parameter until the overall registration error is lower than the preset error threshold, and a first sub-area point cloud model is obtained.

[0047] In the embodiment of the present application, by rough registration of the sub-region point cloud data and forming the initial pose parameters with the geometric center and normal vector, a unified coordinate reference can be quickly established, and the misplacement of different path data can be reduced. Then, through the feature point extraction of the overlapping region and the screening of the corresponding point pairs, the stability of the matching points is ensured. Subsequently, the error distribution is formed by combining the Euclidean distance and the direction difference, and the pose parameters are gradually optimized in the iteration process until the overall error is lower than the threshold. This method can achieve high-precision alignment between multi-path data and obtain a reliable first sub-region point cloud model.

[0048] wherein, according to the sub-region point cloud original data, the geometric center and the normal vector distribution are calculated to form the initial pose parameters, specifically including: Firstly, the sub-region point cloud original data is imported into the point cloud processing system, and the three-dimensional coordinates of the points are read as a whole. In order to obtain the overall position distribution of the point cloud, the mean value operation is performed on the coordinates of all points to obtain the geometric center of the region point cloud. The geometric center can reflect the overall position deviation of the region point cloud, and is the initial reference point for subsequent registration.

[0049] Then, the normal vector of the local region in the point cloud is estimated. Specifically, a plurality of adjacent points are selected in the neighborhood range of each point, and the local orientation direction of the point cloud surface is determined by using the spatial distribution of these points. The overall distribution of the normal vector can be used to describe the spatial orientation characteristics of the sub-region point cloud.

[0050] Finally, the geometric center and the normal vector distribution are taken as inputs to form the initial pose parameters, which are used as the reference coordinate system for rough registration. In this way, a reasonable initial pose can be ensured for subsequent point cloud alignment, reducing the calculation amount and error range of subsequent iterations.

[0051] wherein, according to the initial pose parameters, the point clouds collected by different paths are roughly registered to obtain the preliminary aligned sub-region point cloud data, specifically including: Firstly, the initial pose parameters are applied to the point cloud data collected by different paths to unify their spatial reference system. Specifically, the geometric centers of the point clouds of each path are translated to the same position, and the rotation angle of the point cloud is adjusted according to the normal vector distribution to make them consistent in the general direction.

[0052] Then, the overall position of the preliminary aligned point cloud is corrected. The correction method is to compare the overlapping parts between the point clouds to ensure that the overall shape has basically matched on the macro level. At this time, although there may still be deviations in the details, the overall geometric contour has already been overlapped.

[0053] Finally, the preliminary aligned sub-region point cloud data is obtained. These data provide a basis for subsequent feature point extraction and fine iterative correction.

[0054] Among them, according to the preliminary alignment of the sub-region point cloud data, the feature points of the overlapping region are extracted, and the corresponding point pairs are obtained according to the consistency of the normal vector and the local curvature extreme, specifically including: Firstly, in the preliminary alignment of the sub-region point cloud data, the spatial regions existing overlapping between different paths are identified. These regions are the focus of subsequent fine registration.

[0055] Then, the feature points are extracted in the overlapping region. The selection of feature points is based on two conditions: one is the normal vector consistency, that is, when the orientations of adjacent points are similar, it is considered that the point can reflect the surface continuity feature; The second is the local curvature extreme, that is, the points where the surface bends or mutates obviously, which can reflect the geometric difference.

[0056] Secondly, the feature points extracted in the two paths are compared, and the corresponding relationship is established according to the similarity of spatial position and geometric feature, and the corresponding point pairs are obtained.

[0057] Finally, the corresponding point pairs obtained provide reference data for subsequent error calculation and iterative correction.

[0058] Among them, according to the corresponding point pairs, the Euclidean distance between points and the angle difference are calculated to form the error distribution result, specifically including: Firstly, for each pair of corresponding points, the spatial distance in the three-dimensional coordinate system is calculated. The smaller the straight-line distance between the corresponding points, the better the registration effect; the larger the distance, the greater the deviation.

[0059] Then, in addition to the spatial distance, the angle difference between the normal vectors is also calculated. The size of the angle difference can reflect the consistency of the two points in the surface direction. The smaller the difference, the more consistent the surface; the larger the difference, the greater the local registration error.

[0060] Secondly, the above two indicators are combined to form the error distribution result. This result is used to reflect the matching situation of the entire point cloud in the spatial position and direction, which can be intuitively represented in a two-dimensional heat map or numerical matrix.

[0061] Through the analysis of the error distribution, it can be found out which areas have larger errors, providing a reference direction for subsequent iterative correction.

[0062] Among them, according to the error distribution result, iterative correction processing is performed to gradually update the attitude parameters until the overall registration error is lower than the preset error threshold, and the first sub-region point cloud model is obtained, specifically including: Firstly, the error distribution result is taken as the input to adjust the translation and rotation parameters of the point cloud. Each adjustment aims to reduce the distance and angle difference between the corresponding point pairs, so that the point clouds gradually tend to be consistent.

[0063] Then, after one adjustment, the new error distribution result is recalculated, and the parameter is updated again. This process is iterated continuously, gradually approaching the optimal solution.

[0064] Secondly, a preset error threshold is set. When the overall registration error is lower than the threshold, it is considered that the point cloud registration reaches a satisfactory accuracy, and the iteration process stops. If the error is still higher than the threshold, a new iteration is continued.

[0065] Finally, the point cloud data obtained after multiple rounds of iteration and updating has achieved high-precision alignment in spatial position and direction, thereby forming a first sub-region point cloud model. The model not only retains the complete data collected by different paths, but also ensures the consistency in geometry and surface direction, thereby laying a reliable foundation for subsequent sub-region splicing and global reconstruction.

[0066] In a preferred embodiment of the present application, quality detection is performed on the first sub-region point cloud model to identify the point cloud density distribution and integrity, and to determine a missing region where the point cloud density does not meet the density judgment standard, including: According to the first sub-region point cloud model, a spatial division structure based on a three-dimensional grid is established to obtain a point cloud density matrix; According to the point cloud density matrix, the number of points in each grid element is compared with the global average value to form a low-density candidate region; According to the low-density candidate region, the average distance between points is calculated and compared with the global average distance to generate a potential missing region; According to the potential missing region, boundary continuity detection is performed in combination with the normal consistency of boundary points and the texture change of the corresponding image to obtain a missing region.

[0067] In the embodiment of the present application, through quality detection on the first sub-region point cloud model, a point cloud density matrix is formed by three-dimensional grid division, and the local and global average values are compared to quickly identify a low-density candidate region. Further calculation of the average distance between points in combination with the uniformity of the distribution can screen out a potential missing region. Finally, joint determination is performed in combination with the normal consistency of boundary points and the texture change of the image to obtain an actual missing region. This multi-level detection method can not only avoid misjudgment, but also accurately locate the region that needs to be compensated.

[0068] According to the first sub-region point cloud model, a spatial division structure based on a three-dimensional grid is established to obtain a point cloud density matrix, specifically including: Firstly, the first sub-region point cloud model is imported into a point cloud processing platform, and a grid structure is established according to the three-dimensional space range thereof. The structure divides the overall space into multiple regular cubic small elements, so that the point cloud distribution can be finely represented in space.

[0069] Then, the number of points contained in each small unit is counted, and the counting result is taken as the point cloud density value of the unit. In this way, the entire sub-region point cloud can be represented by a three-dimensional matrix, and the value of each unit corresponds to its point cloud density.

[0070] Secondly, in order to avoid the blurring of density information caused by too low resolution, the size of the unit is dynamically adjusted according to the amount of point cloud data when the grid is divided. For example, in the area with a large number of point clouds, smaller grid units are divided to ensure accuracy; in the area with fewer point clouds, larger grid units are divided to improve efficiency.

[0071] The final point cloud density matrix can intuitively reflect the point distribution of each spatial position, providing data support for subsequent low-density area detection.

[0072] Among them, according to the point cloud density matrix, the number of points of each grid unit is compared with the global average value to form a low-density candidate area, which specifically includes: First, the point cloud density matrix is statistically analyzed as a whole, and the average value of the number of points of all units is calculated. This average value represents the overall acquisition density level of the current sub-region point cloud.

[0073] Then, the number of points of each grid unit is compared with the average value one by one. If the number of points of a unit is significantly lower than the global average value, it is determined to be a low-density candidate area. This comparison avoids misjudgment caused by differences in overall density.

[0074] Secondly, in order to improve the accuracy of the judgment, a relative ratio judgment method can also be introduced in the comparison process. For example, when the number of points of a unit is less than half of the average value, it is considered as a candidate area. In this way, normal slight density fluctuations can be filtered out, and only areas with real point cloud sparseness are retained.

[0075] Finally, the marked low-density candidate area can provide a preliminary range for subsequent missing area detection, avoid large-scale redundant analysis, and thus improve the overall calculation efficiency.

[0076] Among them, according to the low-density candidate area, the average distance between points is calculated and compared with the global average distance to generate a potential missing area, which specifically includes: First, in the low-density candidate area, the point cloud data is selected, and the average distance between these points is calculated. The average distance can reflect the sparseness of the point cloud in this area, and the greater the distance, the more dispersed the points are.

[0077] Then, the local average distance is compared with the global average distance of the entire sub-region point cloud. If the local average distance is significantly greater than the global average distance, it means that the candidate area indeed has a problem of point cloud sparseness, and it can be further confirmed as a potential missing area.

[0078] Secondly, in the comparison process, in addition to the distance between points, the uniformity of point distribution will also be considered. For example, when the point cloud is sparse but uniformly distributed, it can be determined as a normal situation; and when the point cloud is both sparse and unevenly distributed, it is more inclined to be identified as a missing area.

[0079] Finally, after this determination step, the initially obtained potential missing area provides basic data for subsequent boundary continuity detection, ensuring that the identification of the missing area has both global reference and local detail difference.

[0080] In a preferred embodiment of the present application, in the missing area, the local edge contour and gray gradient information are extracted in combination with the image data, the information is projected into the missing area to generate compensation points, and a second sub-area point cloud model is generated, comprising: According to the missing area, a corresponding image data segment is extracted to obtain a missing area image input; According to the missing area image input, a local edge contour is extracted using an edge detection algorithm, and texture gradient information is extracted according to a gray gradient change rate to form edge gradient corresponding feature data; According to the edge gradient corresponding feature data, the contour trend and gray distribution trend of the missing area are calculated to obtain projection compensation parameters; According to the projection compensation parameters, the edge contour and gray gradient information are projected into the three-dimensional spatial position of the missing area in combination with the imaging model of the image acquisition device and its spatial position and attitude parameters in the acquisition process to generate a compensation point data set; According to the compensation point data set, the first sub-area point cloud model is merged to obtain a second sub-area point cloud model.

[0081] In the embodiment of the present application, by extracting the edge contour and gray gradient information of the image in the missing area and projecting these information to generate compensation points, the point cloud missing caused by insufficient sampling or occlusion can be effectively made up. This method makes full use of the texture and boundary features of two-dimensional images and complements three-dimensional point cloud data, so that the details of the building surface can still be restored under the condition of incomplete point cloud. For example, in stone column carving or arch rib edge, local gaps often occur due to the inability of laser to completely cover, and the continuity of patterns or edges can be restored through image gray gradient projection compensation. Then, the generated compensation points are fused with the first sub-area point cloud model to ensure natural transition of the compensation points without damaging the authenticity of the original point cloud. The finally formed second sub-area point cloud model is significantly improved in overall integrity and local continuity, making the texture and contour of complex buildings more clear, and providing high-quality data support for subsequent three-dimensional model splicing and overall display.

[0082] Wherein, according to the missing area image input, the local edge contour is extracted by using the edge detection algorithm, and the texture gradient information is extracted according to the gray gradient change rate, forming the edge gradient corresponding feature data, specifically including: Firstly, the image data segment corresponding to the missing area is input into the image processing module. By scanning the image gray value pixel by pixel, the boundary position with obvious brightness change is identified. At this time, the common edge detection method is adopted, such as the algorithm based on pixel difference or direction gradient, to extract the contour of the local area. In this way, the contour shape information of the missing area can be obtained.

[0083] Then, the change rate of pixel gray near the extracted edge is calculated. The greater the gray change rate, the more obvious the texture change in the region. In this way, the texture details in the two-dimensional image can be converted into quantifiable gradient information.

[0084] Secondly, the edge contour and the gray gradient are combined to form the edge gradient corresponding feature data. The data not only records the edge shape of the region, but also contains the direction and intensity of the texture change. These information provides complete spatial feature description for the subsequent projection of compensation points.

[0085] Wherein, according to the edge gradient corresponding feature data, the contour trend and gray distribution trend of the missing area are calculated to obtain the projection compensation parameters, specifically including: Firstly, the edge gradient corresponding feature data is structured. By sorting the edge points according to their spatial continuity, the overall trend of the contour can be determined. The contour trend reflects the extension direction of the surface structure of the missing area, such as straight line extension, arc transition or polygon change.

[0086] Then, based on the contour trend, the gray gradient is counted. By analyzing the size and direction of the gradient value, the distribution trend of the texture can be extracted. For example, when the gradient direction remains consistent in a certain area, it means that the texture of the area has extension; when the gradient direction changes suddenly, it means that there is a boundary or detail feature.

[0087] Secondly, the contour trend and the gray distribution trend are combined to generate the projection compensation parameters. These parameters contain three key contents: contour extension direction, texture intensity variation law, and spatial correspondence. They will guide the generation of compensation points in the subsequent steps, so that the point cloud distribution is consistent with the real building surface.

[0088] Wherein, according to the compensation point data set, the first sub-region point cloud model is merged to obtain the second sub-region point cloud model, specifically including: Firstly, the compensation point dataset generated by projection is loaded into the point cloud processing system and spatially aligned with the first sub-region point cloud model. During the alignment process, the boundary of the missing region is used as a reference to accurately fill the compensation points into the missing area without deviating from the actual structure.

[0089] Next, consistency checks are performed on the data during the merging process. Specifically, the spatial distance and normal direction between the compensation points and the original point cloud are compared. When the difference between the compensation points and the original point cloud surface is too large, automatic weight adjustment or rejection is performed to avoid surface abruptness caused by unreasonable compensation.

[0090] Secondly, the compensation points are gradually integrated into the original point cloud through weighted fusion. The weight setting is based on two aspects: one is the reliability of the compensation point source, for example, the more obvious the edge gradient information, the higher the weight of the point; the second is the consistency of the compensation points with the original point neighborhood, the higher the compensation points fit the original points, the higher the weight.

[0091] Finally, after the merging is completed, the second sub-region point cloud model is repaired in the missing area, which not only maintains the authenticity of the original point cloud, but also introduces the points generated by image feature compensation, thereby improving the overall continuity and integrity.

[0092] In a preferred embodiment of the present application, the second sub-region point cloud model is subjected to weighted fusion processing, and surface interpolation is performed in the fusion region to form a sub-region three-dimensional model, comprising: According to the second sub-region point cloud model, the overlapping region between adjacent sub-regions is extracted to form an overlapping region point cloud set; According to the overlapping region point cloud set, the density index and the acquisition path confidence of each point are calculated to obtain a weighted parameter matrix; According to the weighted parameter matrix, the point cloud of the overlapping region is subjected to weighted fusion to generate fused point cloud data; According to the fused point cloud data, surface interpolation is performed in the point cloud sparse area, and an interpolation strategy based on the consistency of the normal direction of adjacent points is adopted to generate interpolated surface data; According to the fused point cloud data and the interpolated surface data, the grid structure of the sub-region three-dimensional model is constructed, and surface continuity optimization processing is performed to obtain the sub-region three-dimensional model.

[0093] In the embodiment of the present application, by performing weighted fusion processing on the second sub-region point cloud model, the problem of excessive redundant points and uneven quality in multi-path acquisition can be effectively solved. The weighting strategy considers both the local density of the points and the confidence of the acquisition path, thereby ensuring that high-reliability points have greater weight in the fusion and the influence of low-confidence points is weakened. This mechanism can improve the overall accuracy of the fused data and reduce surface roughness caused by redundant points. Subsequently, surface interpolation is performed in the overlapping region, which can compensate for the sparsity or small-scale absence of point clouds, and in particular, the interpolation strategy based on the consistency of adjacent point normals can ensure a natural transition of the surface and avoid geometric anomalies. In this way, not only the continuity of the point cloud is restored, but also the smoothness of complex curved structures is ensured. For example, in structures such as domes, sculptures or vaults, the interpolated surface can balance detail and overall smoothness. The final sub-region three-dimensional model is more stable at the local junction and overall performance is more consistent with the real structure of the architectural heritage.

[0094] wherein according to the second sub-region point cloud model, an overlapping region between adjacent sub-regions is extracted to form an overlapping region point cloud set, specifically comprising: Firstly, the second sub-region point cloud model is input into a point cloud management system to detect the spatial boundary of the overall point cloud. By identifying the boundary range of each sub-region point cloud, it is determined which regions have intersections in space.

[0095] Then, the points in the overlapping part are extracted. The specific method is to compare the boundary coordinates of the point clouds of different sub-regions, and when the boundaries of two regions coincide or are close within a certain spatial threshold, the points in these regions are extracted as overlapping candidate points.

[0096] Secondly, the candidate points are further screened to remove isolated points or noise points, and only the data that truly represents the overlapping characteristics of the regions is retained. The overlapping region point cloud set obtained in this way can not only reflect the common part between adjacent sub-regions, but also provide data support for subsequent fusion processing.

[0097] wherein according to the overlapping region point cloud set, the density index and the acquisition path confidence of each point are calculated to obtain a weighting parameter matrix, specifically comprising: Firstly, the points in the overlapping region are analyzed for density. The method for calculating the density index is to count the number of neighboring points within a certain radius range centered on each point. The more points there are, the denser the point cloud in that region.

[0098] Then, the path confidence is calculated in combination with the acquisition path information. The path confidence refers to the reliability of point cloud data in the acquisition process, such as whether the motion posture of the acquisition device is stable and whether the acquisition angle is ideal. Generally, points with an acquisition angle directly facing the target surface have higher confidence than points obtained by oblique angle scanning.

[0099] Secondly, the point density index is combined with the path confidence to generate a weighted parameter matrix. This matrix assigns a comprehensive weight value to each point, considering both the local density of the point cloud and the reliability of the data source.

[0100] Among them, according to the weighted parameter matrix, the point cloud in the overlapping area is weighted and fused to generate fused point cloud data, specifically including: Firstly, the point cloud data from different collection paths in the overlapping area is spatially matched to ensure that points at the same location can correspond to the same spatial range.

[0101] Then, these corresponding points are fused according to the weighted parameters. The weighting strategy is: in the overlapping area, if the confidence and density value of a point are high, its position and attributes will have a greater weight in the fusion; while points with low confidence or density will have a relatively weakened impact in the fusion.

[0102] Secondly, after weighted fusion, the point cloud data in the overlapping area will be optimized into a unified point set. This set avoids the problem of excessive points caused by repeated collection and eliminates local inconsistencies caused by errors from different paths.

[0103] Among them, according to the fused point cloud data, surface interpolation is performed in the point cloud sparse area, and an interpolation surface data is generated using an interpolation strategy based on the consistency of the normal direction of adjacent points, specifically including: Firstly, the density of the fused point cloud data is detected to find out the areas where the point cloud distribution is sparse or there are small range holes.

[0104] Then, surface interpolation is performed in these sparse areas. Specifically, the normal direction of adjacent points is found with the center of the sparse area, and the consistency of these normal directions is used to infer the surface trend of the missing area.

[0105] Secondly, according to the trend, supplementary points are generated and a continuous surface is gradually formed. The interpolated surface can maintain the natural extension of the surface and avoid abrupt geometric changes.

[0106] The final interpolated surface data can effectively compensate for the lack of point cloud, making the overall three-dimensional structure of the sub-area more complete and smooth.

[0107] Among them, according to the fused point cloud data and the interpolated surface data, the grid structure of the sub-area three-dimensional model is constructed, and the surface continuity optimization processing is performed to obtain the sub-area three-dimensional model, specifically including: Firstly, the fused point cloud data and the interpolated surface data are imported into the three-dimensional modeling module for integration. Through point and surface matching, the discrete point cloud is converted into a grid structure.

[0108] Then, after the completion of the mesh construction, the continuity of the model surface is detected to find out whether there are cracks, overlaps or irregular protrusions on the surface.

[0109] Secondly, the surface is optimized by a smoothing algorithm to make the transition between adjacent meshes natural and avoid the slight discontinuity caused by interpolation and fusion. The original geometric features are preferentially maintained during optimization to avoid loss of details caused by excessive smoothing.

[0110] Finally, the output sub-region three-dimensional model has complete mesh structure and continuous surface characteristics and can be directly used for subsequent sub-region splicing and global modeling.

[0111] In a preferred embodiment of the present application, the sub-region three-dimensional model is aligned with the boundary features and globally spliced, and smooth fusion is achieved through a multi-scale transition surface to obtain a high-precision real scene three-dimensional reconstruction model covering the whole building heritage, including: According to the sub-region three-dimensional model, a boundary feature point set is extracted to form a global boundary index; According to the global boundary index, boundary alignment processing is performed to calculate the difference vector between the boundary points of adjacent sub-regions and obtain alignment correction parameters; According to the alignment correction parameters, the boundary position of the sub-region three-dimensional model is adjusted to generate a preliminary splicing model; According to the preliminary splicing model, a multi-scale transition surface is constructed, which is composed of a large-scale overall smooth surface and a small-scale local detail surface; According to the multi-scale transition surface, smooth fusion processing is performed on the splicing area of the preliminary splicing model to eliminate boundary discontinuity and obtain a high-precision real scene three-dimensional reconstruction model.

[0112] In the embodiment of the present application, by aligning the boundary features and globally splicing the sub-region three-dimensional model, the problem of boundary discontinuity between different sub-regions in the traditional splicing method can be solved. First, the boundary feature point set is extracted to establish a global index to ensure a unified reference system during splicing. Then, the difference vector of the boundary points of adjacent sub-regions is calculated, and the correction parameters are generated to adjust the boundary, thereby reducing the misalignment and gap. For example, when splicing the temple corridor and courtyard models, the boundary may deviate, and the correction processing can make the splicing more natural. Then, based on the preliminary splicing model, a multi-scale transition surface is constructed, the large-scale surface ensures the smoothness of the whole, and the small-scale surface retains the detail features such as carving, thereby achieving natural fusion at both macro and micro levels. Finally, the smooth processing eliminates the boundary protrusion, making the overall model continuous and complete, which can not only restore the general structure of the building heritage, but also maintain the authenticity of the details, providing high-precision three-dimensional results for cultural protection and virtual display.

[0113] Wherein, according to each sub-region three-dimensional model, the boundary feature point set is extracted, and a global boundary index is formed, specifically comprising: Firstly, a plurality of sub-region three-dimensional models are imported into a unified processing platform, and the boundary region of each model is automatically detected. Boundary detection can be performed by scanning the point cloud or the neighborhood distribution of points in the grid model. When a point lacks enough adjacent points in its neighborhood, it is marked as a boundary point.

[0114] Then, the boundary points of each sub-region are extracted and numbered and stored according to the coordinate position, forming a boundary point set.

[0115] Secondly, the boundary points of all sub-regions are combined to establish an index structure in the global range. The index is equivalent to a global boundary map, providing a unified reference framework for subsequent alignment processing.

[0116] Wherein, according to the global boundary index, boundary alignment processing is performed, the difference vector between adjacent sub-region boundary points is calculated, and the alignment correction parameter is obtained, specifically comprising: Firstly, the adjacent sub-region boundary points in the global boundary index are matched. The matching method is to find the point pairs that are close to each other according to the spatial proximity relationship.

[0117] Then, the difference vector between these matched points is calculated. The difference vector reflects the spatial deviation of two sub-regions on the boundary, including position offset and direction difference.

[0118] Secondly, all difference vectors are counted and summarized to obtain a set of correction parameters. These parameters represent the translation and rotation amounts that need to be adjusted between sub-regions, and are the key basis for subsequent model splicing.

[0119] Wherein, according to the alignment correction parameter, the boundary position of the sub-region three-dimensional model is adjusted to generate a preliminary splicing model, specifically comprising: Firstly, the correction parameters obtained in the previous step are applied to the corresponding sub-region model to adjust the position and direction of the whole or part.

[0120] Then, the boundaries of adjacent sub-regions are adjusted one by one so that they gradually approach and eventually coincide with the corresponding regions.

[0121] Secondly, the adjusted sub-regions are combined together to form a preliminary splicing model. Although there may still be slight discontinuities in the boundary at this time, the overall building form has been basically restored in the global range.

[0122] Wherein, according to the preliminary splicing model, a multi-scale transition surface is constructed, the multi-scale transition surface is composed of a large-scale overall smooth surface and a small-scale local detail surface, specifically comprising: Firstly, the boundary transition area of the preliminary splicing model is detected, and the area where obvious discontinuity exists is marked.

[0123] Next, multi-scale transition surfaces are generated in these areas. Large-scale surfaces form smooth transitions by fitting a large number of boundary points to eliminate macro-geometric discontinuities; small-scale surfaces focus on detail areas and maintain the original fine structure by fitting local point cloud features.

[0124] Secondly, large-scale and small-scale surfaces are combined to form a layered structure. This can ensure the overall smoothness and continuity of the model, while preserving local carving details or angular features.

[0125] Among them, according to the multi-scale transition surface, the splicing area of the preliminary splicing model is subjected to smooth fusion processing to eliminate boundary discontinuity, and a high-precision real scene three-dimensional reconstruction model is obtained, which specifically includes: Firstly, the multi-scale transition surface is applied to the splicing area, and the fusion operation of the surface and the point cloud is performed to make the boundary transition of adjacent sub-areas natural and connected.

[0126] Next, the continuity of the fused splicing area is detected to ensure that there is no crack, overlap or abrupt transition.

[0127] Secondly, the boundary area is smoothed if necessary to ensure that the surface of the model is consistent in vision and geometry.

[0128] Finally, the obtained three-dimensional model not only covers the overall structure of the architectural heritage, but also has continuous and complete boundary features, which can provide reliable support for subsequent digital protection, virtual display and three-dimensional analysis.

[0129] In a preferred embodiment of the present application, according to the curvature division result, in the curvature mutation area, by monitoring the image gray scale change rate in real time, triggering the acquisition device to reduce the motion speed, improve the sampling frequency, and shorten the second sampling interval, the second sampling point cloud data is obtained, including: According to the image data of the curvature mutation area, the pixel gray scale difference value between the continuous image frames is extracted to generate a local gray scale difference matrix; The local gray scale difference matrix is subjected to time series calculation to obtain the gray scale change rate in unit time to form a local texture change parameter; According to the local texture change parameter, a high-detail area marking result is generated by comparing with a preset change rate threshold; According to the high-detail area marking result, the motion control module of the acquisition device is controlled to perform adjustment processing of reducing the motion speed and improving the sampling frequency to generate updated motion and sampling parameters; According to the updated motion and sampling parameters, the second sampling interval is dynamically shortened to generate a dense point cloud acquisition instruction; According to the dense point cloud acquisition instruction, a local scanning is performed and second sampling point cloud data is generated.

[0130] In the embodiment of the present application, by monitoring the real-time gray scale change of the image data of the curvature mutation area, the complex detail area can be dynamically identified, and the acquisition device is automatically triggered to adjust the acquisition strategy. Specifically, the motion speed is reduced, the sampling frequency is increased, and the laser acquisition interval is shortened, so that high-density point clouds are formed in these areas. This dynamic response mechanism avoids the shortcomings of fixed interval acquisition in complex structures, and can ensure that more data points are collected in the areas such as carving, vault and pattern. For example, at the edge of a stone arch, due to the sharp change of gray scale, the system will automatically increase the sampling, so that the curve edge is continuous and complete in the point cloud model. Further, this mechanism can significantly improve the data quality of high-detail areas without significantly increasing the overall acquisition time. The second sampling point cloud data obtained finally is superior to the traditional method in terms of local detail restoration, ensuring that the overall model is efficient and delicate, and is particularly suitable for digital reconstruction of complex architectural heritage.

[0131] Wherein, according to the image data of the curvature mutation area, the pixel gray scale difference value between the continuous image frames is extracted to generate a local gray scale difference matrix, specifically including: Firstly, the multiple images corresponding to the curvature mutation area are input into the image processing module. By comparing the brightness values of the pixel points in the adjacent images frame by frame, the difference degree is recorded.

[0132] Then, the difference value of each pixel point is stored according to the spatial position to form a two-dimensional matrix, and the value in the matrix represents the gray scale change amplitude of the point between different image frames.

[0133] The final local gray scale difference matrix can intuitively reflect which areas in the image have rapid brightness or texture changes, providing basic data for subsequent texture parameter calculation.

[0134] Wherein, the local gray scale difference matrix is calculated in time sequence to obtain the gray scale change rate in unit time to form a local texture change parameter, specifically including: Firstly, the acquisition time of the continuous image frames is combined with the data in the gray scale difference matrix to calculate the gray scale change amount of each pixel point in unit time.

[0135] Then, these change amounts are converted into rate indicators, i.e. the change speed of the brightness value of a certain pixel point in a continuous time period. The higher the change rate is, the more significant the texture or detail at that position is.

[0136] The finally formed local texture change parameter can quantitatively represent the detail complexity of the curvature mutation area.

[0137] Wherein, according to the local texture change parameter, the preset change rate threshold is compared to generate a high detail area marking result, which specifically includes: Firstly, the texture change parameter of each pixel point is compared with the pre-set change rate threshold. The threshold is derived from experimental statistics and is used to distinguish between ordinary areas and high detail areas.

[0138] Then, when the change rate of a certain pixel point exceeds the threshold, the point is marked as a high detail point; when the change rate is lower than the threshold, it is marked as an ordinary point.

[0139] Finally, by summarizing the marking results of all pixel points, a spatial distribution map of the high detail area is generated, providing a basis for subsequent collection device motion strategy adjustment.

[0140] Wherein, according to the high detail area marking result, the motion control module of the collection device is controlled to execute the adjustment process of reducing the motion speed and increasing the sampling frequency, and the updated motion and sampling parameters are generated, which specifically includes: Firstly, the high detail area marking result is input into the motion control system of the collection device. The control system dynamically adjusts the motion path parameters according to the location of these areas.

[0141] Then, above the high detail area, the motion speed of the collection device is automatically reduced, so that it can obtain more image frames in unit time.

[0142] At the same time, the sampling frequency of the camera is increased, that is, more images are collected in the same time to enhance the data coverage.

[0143] The updated motion speed and sampling parameters are generated to enable the collection device to adopt differentiated collection strategies in different areas.

[0144] Wherein, according to the updated motion and sampling parameters, the second sampling interval is dynamically shortened to generate a dense point cloud collection instruction, which specifically includes: Firstly, in the high detail area, the collection frequency of the collection device is controlled to shorten the time interval between adjacent laser pulses.

[0145] Then, by shortening the interval, more laser sampling points are generated in unit area to form a higher density of point cloud.

[0146] Finally, the control system will generate a dense point cloud collection instruction, which contains the coordination parameters of motion speed, camera sampling frequency and collection frequency, to ensure the data collection quality of the high detail area.

[0147] Wherein, according to the dense point cloud collection instruction, a local scan is performed to generate the second sampling point cloud data, which specifically includes: Firstly, the acquisition device enters the high-detail area for scanning according to the dense point cloud acquisition instruction. At this time, the motion speed is reduced, the camera sampling frequency is increased, and the acquisition interval is shortened.

[0148] Then, the number of data points generated by the acquisition device in the high-detail area increases significantly, covering the detail area that is prone to be missed.

[0149] Finally, the newly acquired data is arranged as second sampling point cloud data, which is complementary to the first sampling point cloud data of the normal area, ensuring that the overall modeling is efficient and fine.

[0150] In a preferred embodiment of the present application, according to the potential missing area, the boundary continuity is detected in combination with the normal consistency of the boundary points and the texture change of the corresponding image, and the missing area is obtained, including: According to the point cloud data of the potential missing area, the normal vector of the boundary point is extracted to form a normal distribution data set; According to the normal distribution data set, the angle difference between adjacent normal vectors is calculated to generate a direction consistency index matrix; According to the direction consistency index matrix, the boundary points with an angle greater than a threshold value are screened out to obtain an abnormal boundary point set; According to the abnormal boundary point set, the image texture gradient direction is extracted by mapping to the corresponding image area to form a texture direction distribution data set; According to the texture direction distribution data set, the texture normal matching degree result is generated by joint comparison with the normal consistency index; According to the texture normal matching degree result, when the matching degree is lower than a preset matching threshold, the corresponding area is determined as a missing area, and a missing area boundary contour is generated.

[0151] In the embodiment of the present application, the joint detection of point cloud geometric information and image texture features can effectively improve the accuracy of missing area identification. Firstly, the boundary point normal consistency calculation of the potential missing area can preliminarily screen out the position of normal mutation, which may indicate that there is a data break. Then, by comparing with the texture gradient direction in the image, it is further confirmed whether it is a real missing. This double determination method can significantly reduce the probability of misjudgment. For example, when the point cloud is sparse due to the material reflection of the stone tablet surface, but the image texture remains continuous, the system will not misjudge as a missing area. On the contrary, in the area blocked by the vault or window lattice, if the normal is discontinuous and there is a break in the image texture, the missing area can be accurately identified. The final missing area boundary contour is not only accurate, but also provides a clear range for the generation of compensation points, making the subsequent repair more efficient and reliable. Overall, this method exhibits high robustness and accuracy in complex environments of architectural heritage.

[0152] Wherein, according to the point cloud data of the potential missing area, the normal vector of the boundary point is extracted, and a normal distribution data set is formed, specifically including: Firstly, the point cloud data of the potential missing area is imported into the point cloud processing system. By selecting several adjacent points in the neighborhood range of each point, the spatial distribution trend of these points is used to estimate the surface orientation direction of the point, that is, the normal vector.

[0153] Then, the above calculation is performed on all boundary points in the region, and the obtained normal vector is uniformly recorded as a data set.

[0154] The data set can reflect the overall directional characteristics of the boundary points in space, and provide basic data for subsequent consistency detection.

[0155] Wherein, according to the normal distribution data set, the angle difference between adjacent normal vectors is calculated, and a direction consistency index matrix is generated, specifically including: Firstly, in the normal distribution data set, the normal vector direction of adjacent points is compared pair by pair. If the surface direction of two points is close, the angle difference is small; if the surface direction difference is large, the angle difference is obvious.

[0156] Then, the results of these angle differences are converted into direction consistency indicators and stored as a matrix according to the spatial position.

[0157] The index matrix can intuitively show the consistency degree of the boundary points in a region in space direction, and provide quantitative basis for subsequent identification of abnormal boundary points.

[0158] Wherein, according to the direction consistency index matrix, the boundary points with angle greater than the threshold value are screened out, and an abnormal boundary point set is obtained, specifically including: Firstly, a threshold value is preset to distinguish normal continuous boundary from abnormal abrupt boundary.

[0159] Then, the values in the direction consistency index matrix are compared point by point with the threshold value, and if the angle difference exceeds the threshold value, it is considered that the point has abnormal direction deviation.

[0160] Finally, these abnormal points are extracted to form an abnormal boundary point set, which provides a candidate area for subsequent image data comparison.

[0161] Wherein, according to the abnormal boundary point set, the corresponding image area is mapped, the image texture gradient direction is extracted, and a texture direction distribution data set is formed, specifically including: Firstly, the position of the abnormal boundary point in the three-dimensional point cloud is projected onto the corresponding two-dimensional image to determine the image area where it is located.

[0162] Then, the directionality of the image gray value in the region is detected. By analyzing the direction of gray value change, the texture gradient direction is extracted.

[0163] Secondly, the extracted texture direction data is uniformly arranged to form a texture direction distribution data set. The data set can correspond to normal information to help determine the authenticity of point cloud missing.

[0164] Among them, according to the texture direction distribution data set, the normal consistency index is jointly compared to generate a texture normal matching degree result, specifically including: First, the texture direction and the direction of the normal vector are compared point by point. If the directions are consistent or close, it means that the point cloud in this area is consistent with the image features; if the difference is obvious, it means that there is potential point cloud missing in this area.

[0165] Next, the comparison results of all points are summarized to generate the overall texture normal matching degree result, which is used to quantify the consistency degree between the image and the point cloud.

[0166] The result provides an objective basis for subsequent missing area determination, avoiding misjudgment caused by relying on a single data source.

[0167] Among them, according to the texture normal matching degree result, when the matching degree is lower than the preset matching threshold, the corresponding area is determined as a missing area, and a missing area boundary contour is generated, specifically including: First, set a matching degree threshold to distinguish between normal areas and missing areas.

[0168] Next, compare the texture normal matching degree result, and when the matching degree value of a certain position is lower than the threshold, mark it as a missing area.

[0169] Secondly, in the missing area, further track the boundary points and generate a continuous missing boundary contour.

[0170] The final missing area boundary can not only accurately locate the point cloud missing range, but also provide a clear spatial range for subsequent compensation point generation.

[0171] In a preferred embodiment of the present application, according to the projection compensation parameters, combined with the imaging model of the image acquisition device and its spatial position and attitude parameters in the acquisition process, the edge contour and gray gradient information are projected to the three-dimensional spatial position of the missing area to generate a compensation point data set, including: According to the image segment of the missing area, perform edge detection to extract edge pixel points and form an edge contour data set; According to the edge contour data set, calculate the gray gradient value of each edge pixel point, and generate a texture gradient vector field; According to the texture gradient vector field, combined with the imaging model of the image acquisition device and its spatial position and attitude parameters in the acquisition process, generate the projection path parameters of the pixel points in the three-dimensional space; According to the projection path parameters, the edge contour and the texture gradient are mapped to the three-dimensional space point by point to generate a preliminary compensation point set; According to the preliminary compensation point set, the gray gradient amplitude is assigned to each compensation point to generate a weighted compensation point data set; According to the weighted compensation point data set, the first sub-region point cloud model is fused to form a second sub-region point cloud model.

[0172] In the embodiment of the present application, by using the edge contour and gray gradient information of the image, and combining the imaging model and the attitude parameter for three-dimensional projection, reasonable compensation points can be generated in the missing area. This process ensures that the two-dimensional image features can be accurately mapped to the three-dimensional space, avoiding compensation errors caused by mismatched coordinate systems. For example, in the top area of a high-rise stone tower, even if the laser data is insufficient, correct compensation points can be generated through image projection. Then, by assigning weights to the compensation points based on the gray gradient amplitude, the boundary and texture features can be highlighted, making the compensation points more consistent with the real surface. Finally, the weighted compensation point data is fused with the first sub-region point cloud model to repair the missing area and maintain a natural transition, ensuring the overall authenticity and avoiding the surface being conspicuous due to excessive compensation. The final generated second sub-region point cloud model has continuity and integrity, laying a solid foundation for subsequent global stitching and three-dimensional reconstruction.

[0173] Among them, according to the image segment of the missing area, edge detection is performed, and edge pixel points are extracted and an edge contour data set is formed, specifically including: First, the original image segment corresponding to the missing area is imported into the image processing module, and the image is preprocessed, including noise removal and brightness equalization, to ensure the accuracy of subsequent feature extraction.

[0174] Then, an edge detection algorithm is used to identify pixels with significant brightness or texture changes in the image. These pixels are usually located at the contour edges or texture boundaries of the building surface.

[0175] Finally, the detected edge pixels are connected and grouped according to their positions in the image to form an edge contour data set, providing a basis for subsequent gray gradient calculation.

[0176] Among them, according to the edge contour data set, the gray gradient value of each edge pixel point is calculated, and a texture gradient vector field is generated, specifically including: First, in the edge contour data set, the neighborhood region of each edge pixel point is scanned to measure the amplitude and direction of its gray value change.

[0177] Then, the size of the gray change is taken as the gradient amplitude, and the change direction is taken as the gradient direction to generate the corresponding gradient features for each edge point.

[0178] Finally, the gradient amplitudes and directions of all edge points together constitute a texture gradient vector field, which can reflect the texture direction and detail distribution of the edge region in the image.

[0179] According to the texture gradient vector field, the imaging model of the image acquisition device, and the spatial position and attitude parameters of the device in the acquisition process, the projection path parameters of the pixel points in the three-dimensional space are generated, specifically including: Firstly, the imaging model parameters of the acquisition device are called, including focal length, imaging scale factor, and lens distortion correction parameters, which are used to establish the mapping relationship between the two-dimensional image and the three-dimensional space.

[0180] Then, the spatial position and attitude of the device in the image acquisition process, such as position coordinates, pitch angle, yaw angle, and roll angle, are input into the mapping model to ensure that the image features can correspond to the actual spatial scene.

[0181] Secondly, each edge point in the texture gradient vector field is combined with the imaging model and attitude parameters to generate its projection path parameters in the three-dimensional space. This parameter describes the possible three-dimensional space trajectory corresponding to each image point, providing directional basis for subsequent point cloud compensation.

[0182] According to the projection path parameters, the edge contour and texture gradient are mapped to the three-dimensional space point by point to generate a preliminary compensation point set, specifically including: Firstly, each image edge point is extended along its projection path to find a position consistent with the boundary of the missing area, and a compensation point is generated at this position.

[0183] Then, during the mapping process, the gray scale gradient direction and the spatial position of the compensation point are considered simultaneously, so that the distribution of the compensation point not only conforms to the spatial geometric structure, but also maintains consistency with the texture information of the image.

[0184] Finally, all the compensation points generated by mapping constitute the preliminary compensation point data, filling the three-dimensional spatial coordinates of the missing area.

[0185] According to the preliminary compensation point set, the gray scale gradient amplitude is used to assign weights to each compensation point to generate a weighted compensation point data set, specifically including: Firstly, for each compensation point in the preliminary compensation point set, the amplitude of the gray scale gradient in the corresponding image is read. The greater the gradient amplitude, the more prominent the texture or boundary features of the point in the image.

[0186] Then, the gray scale gradient amplitude is converted into a weight value to give the compensation point different importance. For example, points with high gradient amplitude have higher weights and are more representative in subsequent fusion; while points with small gradient amplitude have lower weights and only play a supporting role.

[0187] Finally, the formed weighted compensation point dataset not only retains the key texture and edge features in the image, but also avoids model distortion caused by too many low-quality points.

[0188] The second sub-region point cloud model is formed by fusing the weighted compensation point dataset with the first sub-region point cloud model, and specifically includes: Firstly, the weighted compensation point dataset is imported into the point cloud processing system and is coordinated with the first sub-region point cloud model in the coordinate system, so as to ensure that the two types of point cloud data are superimposed under the same spatial reference.

[0189] Then, in the fusion process, the system will preferentially retain the original point cloud and supplement it into the missing area according to the weight distribution principle of the compensation points. When the compensation points and the original points conflict, the point with higher weight is retained to ensure the final data quality.

[0190] Secondly, smoothing and consistency processing are performed on the fused region, so that the compensation points and the original points transition naturally and avoid geometric anomalies.

[0191] Finally, the formed second sub-region point cloud model is repaired in the missing area, and the overall continuity and local detail integrity are significantly improved, providing a solid data foundation for subsequent sub-region splicing and global modeling.

[0192] The above is the preferred embodiment of the present application. It should be noted that for ordinary skilled persons in the technical field, several improvements and refinements can be made without departing from the principles of the present application, and these improvements and refinements should also be considered within the scope of protection of the present application.

Claims

1. A method for high-precision real scene three-dimensional reconstruction of architectural heritage, characterized in that, The method comprises: Obtaining multi-angle image data of the exterior and interior of the building heritage, and dividing the to-be-reconstructed area into a plurality of sub-regions according to the overall structure of the building heritage to obtain image data; According to the image data, point cloud data of each sub-region is collected by using a collection device, a first sampling interval is used in a conventional region, a second sampling interval is dynamically shortened in a carved recess or a narrow space region, and the sub-region point cloud data is generated by a multi-path cross collection method; Performing coarse registration processing on the sub-region point cloud data, and performing iterative correction on corresponding points of features in the overlapping region to generate a first sub-region point cloud model; Performing quality detection on the first sub-region point cloud model, identifying the point cloud density distribution and integrity, and determining a missing region with point cloud density not meeting a density judgment standard; In the missing region, local edge contour and gray gradient information is extracted in combination with the image data, the information is projected into the missing region to generate compensation points, and a second sub-region point cloud model is generated; Performing weighted fusion processing on the second sub-region point cloud model, and performing curved surface interpolation in the fusion region to form a sub-region three-dimensional model; Aligning the sub-region three-dimensional model with boundary features and globally splicing, and realizing smooth fusion through a multi-scale transition surface to obtain a high-precision real scene three-dimensional reconstruction model covering the overall building heritage.

2. The method of claim 1, wherein, According to the image data, point cloud data of each sub-region is collected by using a collection device, including: According to the obtained image data, curvature features of the building surface are extracted, the sub-region is divided into a curvature gentle region and a curvature mutation region, and a curvature division result is generated; According to the curvature division result, in the curvature gentle region, the collection device is controlled to collect image data at a first motion speed and a first sampling interval to obtain first sampling point cloud data; According to the curvature division result, in the curvature mutation region, by monitoring the image gray scale change rate in real time, the collection device is triggered to reduce the motion speed, increase the sampling frequency, and shorten the second sampling interval to obtain second sampling point cloud data; Based on the feedback of the obstacle avoidance sensor of the collection device in the narrow space, the posture and collection angle of the collection device are adjusted, compensation collection is performed on the blocked region, and compensation point cloud data is generated; The first sampling point cloud data, the second sampling point cloud data and the compensation point cloud data are merged, and the overlapping data generated by the multi-path cross scanning is weighted to obtain sub-region point cloud raw data.

3. The method of claim 1, wherein, The coarse registration processing is performed on the sub-region point cloud data, and the corresponding points of the features in the overlapping region are used to perform iterative correction to generate a first sub-region point cloud model, including: According to the sub-region point cloud raw data, the geometric center and the normal vector distribution are calculated to form initial attitude parameters; According to the initial attitude parameters, the point clouds collected by different paths are coarsely registered to obtain preliminarily aligned sub-region point cloud data; According to the preliminarily aligned sub-region point cloud data, feature points in the overlapping region are extracted, the feature points are selected according to the normal vector consistency and local curvature extreme value to obtain corresponding point pairs; According to the corresponding point pairs, the Euclidean distance and the angle difference between the points are calculated to form an error distribution result; According to the error distribution result, an iterative correction process is performed to gradually update the attitude parameters until the overall registration error is lower than a preset error threshold, and a first sub-region point cloud model is obtained.

4. The method of claim 1, wherein, Quality detection is performed on the first sub-region point cloud model to identify the point cloud density distribution and integrity, and to determine a missing region where the point cloud density does not meet the density judgment standard, including: According to the first sub-region point cloud model, a three-dimensional grid-based space division structure is established to obtain a point cloud density matrix; According to the point cloud density matrix, the number of points in each grid cell is compared with the global average value to form a low-density candidate region; According to the low-density candidate region, the average distance between points is calculated and compared with the global average distance to generate a potential missing region; According to the potential missing region, boundary continuity detection is performed by combining the normal consistency of boundary points and the texture changes of the corresponding image to obtain the missing region.

5. The method of claim 1, wherein, In the missing region, local edge contour and gray gradient information are extracted from the image data, and the information is projected into the missing region to generate compensation points to generate a second sub-region point cloud model, including: According to the missing region, the corresponding image data segment is extracted to obtain the missing region image input; According to the missing region image input, edge detection algorithm is used to extract local edge contour, and texture gradient information is extracted according to the gray gradient change rate to form edge gradient corresponding feature data; According to the edge gradient corresponding feature data, the contour direction and gray distribution trend of the missing region are calculated to obtain the projection compensation parameters; According to the projection compensation parameters, the imaging model of the image acquisition device and its spatial position and attitude parameters in the acquisition process are combined to project the edge contour and gray gradient information into the three-dimensional space position of the missing region to generate a compensation point data set; According to the compensation point data set, the first sub-region point cloud model is merged to obtain the second sub-region point cloud model.

6. The method of claim 1, wherein, Weighted fusion processing is performed on the second sub-region point cloud model, and surface interpolation is performed in the fusion region to form a sub-region three-dimensional model, including: According to the second sub-region point cloud model, the overlapping region between adjacent sub-regions is extracted to form an overlapping region point cloud set; According to the overlapping region point cloud set, the density index and acquisition path confidence of each point are calculated to obtain a weighting parameter matrix; According to the weighting parameter matrix, the point cloud of the overlapping region is weighted and fused to generate fused point cloud data; According to the fused point cloud data, surface interpolation is performed in the point cloud sparse area, and an interpolation strategy based on the normal consistency of adjacent points is adopted to generate interpolation surface data; According to the fused point cloud data and the interpolation surface data, the grid structure of the sub-region three-dimensional model is constructed, and surface continuity optimization processing is performed to obtain the sub-region three-dimensional model.

7. The method of claim 1, wherein the method further comprises: The boundary feature alignment and global splicing of the sub-region three-dimensional model are performed, and smooth fusion is realized through multi-scale transition surfaces to obtain a high-precision real scene three-dimensional reconstruction model covering the whole building heritage, including: According to each sub-region three-dimensional model, a boundary feature point set is extracted to form a global boundary index; According to the global boundary index, boundary alignment processing is performed to calculate the difference vector between the boundary points of adjacent sub-regions to obtain alignment correction parameters; According to the alignment correction parameter, the boundary position of the sub-region three-dimensional model is adjusted, and a preliminary splicing model is generated; According to the preliminary splicing model, a multi-scale transition surface is constructed, and the multi-scale transition surface is composed of a large-scale overall smooth surface and a small-scale local detail surface; According to the multi-scale transition surface, a smoothing fusion process is performed on the splicing area of the preliminary splicing model to eliminate boundary discontinuity, and a high-precision real scene three-dimensional reconstruction model is obtained. 8.The method of claim 2, wherein, According to the curvature division result, in the curvature mutation area, the image gray change rate is monitored in real time, the motion speed of the acquisition device is reduced, the sampling frequency is increased, and the second sampling interval is shortened to obtain second sampling point cloud data, including: According to the image data of the curvature mutation area, the pixel gray difference value between the continuous image frames is extracted to generate a local gray difference matrix; The local gray difference matrix is subjected to time series calculation to obtain the gray change rate per unit time to form a local texture change parameter; According to the local texture change parameter, the preset change rate threshold is compared to generate a high-detail area marking result; According to the high-detail area marking result, the motion control module of the acquisition device is controlled to perform adjustment processing of reducing the motion speed and increasing the sampling frequency to generate updated motion and sampling parameters; According to the updated motion and sampling parameters, the second sampling interval is dynamically shortened to generate a dense point cloud acquisition instruction; According to the dense point cloud acquisition instruction, local scanning is performed to generate second sampling point cloud data. 9.The method of claim 4, wherein, According to the potential missing area, the boundary continuity is detected by combining the normal consistency of the boundary points and the texture change of the corresponding image to obtain the missing area, including: According to the point cloud data of the potential missing area, the normal vector of the boundary point is extracted to form a normal distribution data set; According to the normal distribution data set, the included angle difference between adjacent normal vectors is calculated to generate a direction consistency index matrix; According to the direction consistency index matrix, the boundary points with an included angle greater than a threshold value are screened out to obtain an abnormal boundary point set; According to the abnormal boundary point set, the corresponding image area is mapped to extract the image texture gradient direction to form a texture direction distribution data set; According to the texture direction distribution data set, the normal consistency index is jointly compared to generate a texture normal matching degree result; According to the texture normal matching degree result, when the matching degree is lower than a preset matching threshold, the corresponding area is determined as a missing area, and a missing area boundary contour is generated. 10.The method of claim 5, wherein, According to the projection compensation parameter, the edge contour and the gray gradient information are projected to the three-dimensional space position of the missing area by combining the imaging model of the image acquisition device and the spatial position and attitude parameters thereof in the acquisition process to generate a compensation point data set, including: According to the image segment of the missing area, edge detection is performed to extract edge pixel points and form an edge contour data set; According to the edge contour data set, the gray gradient value of each edge pixel point is calculated, and a texture gradient vector field is generated; According to the texture gradient vector field, the imaging model of the image acquisition device and the spatial position and attitude parameters thereof in the acquisition process are combined to generate projection path parameters of the pixel points in the three-dimensional space. According to the projection path parameter, the edge profile and the texture gradient are mapped to the three-dimensional space point by point to generate a preliminary compensation point set; According to the preliminary compensation point set, the gray scale gradient amplitude is used to assign weights to each compensation point to generate a weighted compensation point data set; According to the weighted compensation point data set, the first sub-region point cloud model is fused to form a second sub-region point cloud model.

Images