Method and system for checking cadastral map pattern

Abstract

The application provides a real estate surveying and mapping figure verification method and system, and belongs to the technical field of surveying and mapping. Through deep estimation of matching cost of a multi-scale feature map set, a perception depth map is obtained. Through normal estimation of the depth constraint of the multi-scale feature map set, a cost normal map is obtained. Based on the cost normal map and the perception depth map, a distortion region is determined. The neighborhood subgraph of each change edge in the change time sequence graph on multiple time slices is extracted to obtain a verification subgraph set of each change edge. All nodes in each verification subgraph set are coded, fused and subjected to spatio-temporal anomaly detection to obtain multiple abnormal change events of associated coordinate precision deviation. The distortion coordinate point set is determined according to the distortion region and all abnormal change events. The surveying and mapping figure is calibrated based on the distortion coordinate point set. According to the geometric perception of the model of the real estate and the spatio-temporal anomaly detection at the change time, efficient calibration of the distorted coordinate points of the surveying and mapping figure can be realized.

Description

Technical Field

[0001] This application relates to the field of surveying and mapping technology, and more specifically, to a method and system for verifying the graphics of real estate surveying and mapping. Background Technology

[0002] Real estate surveying is a crucial foundation for real estate registration, transactions, and management. The accuracy of survey maps directly affects the effectiveness of real estate ownership delineation and spatial planning. With the application of technologies such as UAV oblique photography and 3D modeling in the field of real estate surveying, the efficiency and scope of surveying work have been significantly improved.

[0003] In existing technologies, real estate mapping graphic verification often relies solely on manual checking or single geometric feature detection. Some technologies only achieve simple 3D model construction through image feature extraction, lacking precise analysis of the geometric consistency of model depth and normals. This makes it difficult to automatically identify geometrically distorted areas in the 3D model and ignores the temporal correlation and spatiotemporal characteristics of historical changes in real estate, failing to effectively detect abnormal events during the change process, resulting in ambiguous localization of mapping graphic distortion points. Therefore, how to achieve efficient calibration of mapping graphic distortion coordinate points based on the geometric perception of the real estate model and the detection of spatiotemporal anomalies during changes has become a challenge for the industry. Summary of the Invention

[0004] This application provides a method and system for verifying the graphics of real estate surveying maps, which can achieve efficient calibration of distorted coordinate points of surveying maps based on the geometric perception of the real estate model and the detection of spatiotemporal anomalies during changes.

[0005] In a first aspect, this application provides a method for verifying the graphics of a real estate survey map, comprising the following steps: A multi-view image sequence of the real estate to be verified is obtained, aerial triangulation and geodetic coordinate correction are performed on the multi-view image sequence, and feature extraction is performed on the multi-view image sequence to obtain a multi-scale feature map set. Depth estimation of matching cost is performed on the multi-scale feature map set, and surface normal information generated during the matching process is integrated to obtain a perception depth map. Normal estimation under depth constraints is performed on the multi-scale feature map set to obtain a cost normal map. The distortion region of the real estate model is determined based on the cost normal map and the perception depth map. Obtain the historical change records of the real estate to be verified, construct the change time sequence diagram of the real estate, extract the neighborhood subgraphs of each change edge in the change time sequence diagram in multiple time slices, and obtain the verification subgraph set of each change edge; Encoding fusion and spatiotemporal anomaly detection are performed on all nodes in each verification sub-map set to obtain multiple abnormal change events with related coordinate accuracy deviations. Based on the distorted area and all abnormal change events, the distorted coordinate point set of the survey map of the real estate is determined. Using geodetic control points and cadastral control points as confidence references, the coordinates of the survey map are calibrated based on the distorted coordinate point set.

[0006] In some embodiments, depth estimation of the matching cost is performed on the multi-scale feature map set, and surface normal information generated during the matching process is integrated to obtain a perceptual depth map, specifically including: An initial depth map covering the entire depth search range is constructed based on the multi-scale feature map set; The local depth search range for the current stage is determined based on the initial depth map, and multiple depth planes are uniformly sampled within the local depth search range; Obtain the surface normal map estimated from the initial depth map, and upsample the surface normal map to a resolution that matches the preset feature map in the multi-scale feature map set; The sampled surface normal map and the matching cost map of each depth plane are concatenated to obtain a geometric perceptron with fused normal information; The geometric sensing volume is converted into a probability distribution, and the probability distribution is weighted and summed with all depth planes to obtain a sensing depth map.

[0007] In some embodiments, performing depth-constrained normal estimation on the multi-scale feature map set to obtain a cost normal map specifically includes: Using each pixel in the perceived depth map as the center, a preset depth sampling interval is set near the depth value of each pixel to construct multiple depth-first local sampling planes; The feature maps from the neighborhood perspective in the multi-scale feature map set are projected onto each local sampling plane and aggregated to obtain the local cost volume of the normal. The cost normal map is determined based on the local cost volume of the normal.

[0008] In some embodiments, determining the distortion region of the real estate model based on the cost normal map and the perceived depth map specifically includes: Perform differential calculations on the perceived depth map to obtain a derived normal map; The derived normal map and the cost normal map are compared pixel-by-pixel with the angle error to obtain a geometric consistency difference map. The difference map is segmented by thresholding, and pixel areas with angle errors exceeding a preset threshold are marked as distorted areas of the real estate model.

[0009] In some embodiments, obtaining historical change records of the real estate to be verified and constructing a change sequence diagram of the real estate specifically includes: Extract all real estate units from the historical change records as nodes; Extract all ownership and graphic change events from the historical change records, treat all events as directed edges connecting the nodes before and after the change, and assign an edge weight to each directed edge based on the timestamp of the occurrence. Construct a change sequence diagram of the real estate based on the nodes and all directed edges.

[0010] In some embodiments, extracting the neighborhood subgraphs of each change edge in the change time sequence graph across multiple time slices to obtain the verification subgraph set for each change edge specifically includes: Obtain the timestamp of each change edge to be detected in the change sequence graph and the time window to be examined; Select one of the changed edges as the selected changed edge; In each time slice within the time window of the selected change edge, the connection strength between each node and the center node in the change time sequence diagram is determined, taking the two nodes of the selected change edge as the center node. Based on all the connection strengths, determine multiple context nodes of the selected changed edge and construct a local neighborhood subgraph of the selected changed edge in that time slice; Merge the local neighborhood subgraphs on all time slices to obtain the verification subgraph set of the selected changed edge; Continue to determine the verification sub-graphite set of the remaining changed edges.

[0011] In some embodiments, encoding fusion and spatiotemporal anomaly detection are performed on all nodes in each verification sub-graph set to obtain multiple abnormal change events of associated coordinate precision deviation, specifically including: Determine the diffusion code, distance code, time code, and coordinate precision deviation code for all nodes in each verification sub-map set; All diffusion codes, distance codes, time codes, and coordinate precision deviation codes are added element by element to obtain the comprehensive coding vector of each node. All comprehensive coding vectors are stacked in chronological order to form a node coding matrix. The node encoding matrix is ​​input into an encoder based on a bidirectional self-attention mechanism, and the attention weight between any two nodes in the subgraph is calculated to obtain a node hidden representation that incorporates spatiotemporal context information. The hidden representation of the nodes is subjected to mean pooling to obtain the edge embedding vector of each changed edge; Anomaly probability determination is performed on the coordinate precision deviation association of all edge embedding vectors to obtain multiple abnormal change events of associated coordinate precision deviation.

[0012] In some embodiments, determining the set of distorted coordinate points of the survey map of the real estate based on the distorted region and all abnormal change events specifically includes: The distorted region is projected onto a two-dimensional plane to obtain the corresponding two-dimensional distorted region; Extract multiple graphical change events from all abnormal change events, locate the boundary points and boundary lines involved in all graphical change events, and use them as the distortion locations associated with the events; The two-dimensional distorted region is overlaid and merged with all the distorted locations to obtain the set of distorted coordinate points of the survey map of the real estate.

[0013] In some embodiments, calibrating the coordinates of the survey map based on the distorted coordinate point set, using geodetic control points and cadastral control points as confidence references, specifically includes: Obtain the geodetic control points and cadastral control points around each distorted point in the distorted coordinate point set, and extract the accurate geographic coordinates of all reference datums; Determine the coordinate correction amount for multiple distortion points based on all confidence reference points; The coordinates of the corresponding distorted points in the surveyed figure are calibrated based on all coordinate corrections.

[0014] Secondly, this application provides a real estate survey map graphic verification system, comprising: The acquisition module is used to acquire a multi-view image sequence of the real estate to be verified, perform aerial triangulation and geodetic coordinate correction on the multi-view image sequence, and extract features from the multi-view image sequence to obtain a multi-scale feature map set. The processing module is used to perform depth estimation of matching cost on the multi-scale feature map set, integrate the surface normal information generated during the matching process to obtain a perception depth map, perform normal estimation under depth constraints on the multi-scale feature map set to obtain a cost normal map, and determine the distortion region of the real estate model based on the cost normal map and the perception depth map. The processing module is also used to obtain historical change records of the real estate to be verified, construct a change time sequence diagram of the real estate, extract neighborhood subgraphs of each change edge in the change time sequence diagram in multiple time slices, and obtain a set of verification subgraphs for each change edge. The processing module is also used to perform encoding fusion and spatiotemporal anomaly detection on all nodes in each verification sub-map set to obtain multiple abnormal change events with related coordinate precision deviations, and to determine the distorted coordinate point set of the survey map of the real estate based on the distorted area and all abnormal change events. The execution module is used to calibrate the coordinates of the survey map based on the distorted coordinate point set, using geodetic control points and cadastral control points as confidence reference benchmarks.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The real estate mapping graphic verification method and system provided in this application first acquires a multi-view image sequence of the real estate to be verified, performs aerial triangulation and geodetic coordinate correction on the multi-view image sequence, and extracts features from the multi-view image sequence to obtain a multi-scale feature map set; performs depth estimation of matching cost on the multi-scale feature map set, and integrates the surface normal information generated during the matching process to obtain a perceived depth map; performs normal estimation under depth constraints on the multi-scale feature map set to obtain a cost normal map; and determines the distortion area of ​​the real estate model based on the cost normal map and the perceived depth map. The process involves: acquiring historical change records of the real estate to be verified; constructing a change time series diagram of the real estate; extracting neighborhood subgraphs of each change edge in the change time series diagram across multiple time slices to obtain a verification subgraph set for each change edge; performing encoding fusion and spatiotemporal anomaly detection on all nodes in each verification subgraph set to obtain multiple abnormal change events with associated coordinate accuracy deviations; determining the distorted coordinate point set of the survey map of the real estate based on the distorted area and all abnormal change events; and calibrating the coordinates of the survey map based on the distorted coordinate point set, using geodetic control points and cadastral control points as confidence reference benchmarks.

[0016] Therefore, in the process of a real estate surveying and mapping graphic verification method, this application first performs depth estimation of the matching cost on the multi-scale feature map set and integrates the surface normal information generated during the matching process to obtain a perceived depth map. Then, it performs normal estimation under depth constraints on the multi-scale feature map set to obtain a cost normal map. Based on the cost normal map and the perceived depth map, it determines the distortion region of the real estate model. The distortion region is a set of pixel regions in the 3D model with geometrical anomalies, facilitating the mapping of the 3D geometric verification results to the 2D surveying and mapping graphics for subsequent coordinate calibration, thus solving the 3D model quality problem. The first issue is the lack of correlation between two-dimensional graphic calibrations. Secondly, historical change records of the real estate to be verified are obtained, and a change time series diagram of the real estate is constructed. Neighborhood subgraphs of each change edge in the change time series diagram are extracted across multiple time slices to obtain a verification subgraph set for each change edge. Encoding fusion and spatiotemporal anomaly detection are performed on all nodes in each verification subgraph set to obtain multiple abnormal change events with associated coordinate accuracy deviations. Based on the distorted region and all abnormal change events, the distorted coordinate point set of the real estate's survey map is determined. The distorted region facilitates mapping the three-dimensional geometric verification results to the two-dimensional survey map for subsequent coordinate calibration. This scheme can achieve efficient calibration of distorted coordinate points in the survey map based on the geometric perception of the real estate model and the spatiotemporal anomaly detection during changes. Attached Figure Description

[0017] Figure 1 This is an exemplary flowchart of a method for verifying the shape of a real estate survey map according to some embodiments of this application; Figure 2 This is an exemplary flowchart illustrating the determination of a cost normal map according to some embodiments of this application; Figure 3 This is an exemplary flowchart illustrating the determination of a set of distorted coordinate points according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of a real estate surveying and mapping graphic verification system according to some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device for implementing a method for verifying real estate surveying and mapping graphics, according to some embodiments of this application. Detailed Implementation

[0018] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0019] refer to Figure 1 The figure is an exemplary flowchart of a method for verifying the shape of a real estate survey map according to some embodiments of this application. The method for verifying the shape of a real estate survey map mainly includes the following steps: In step 101, a multi-view image sequence of the real estate to be verified is obtained, aerial triangulation and geodetic coordinate correction are performed on the multi-view image sequence, and feature extraction is performed on the multi-view image sequence to obtain a multi-scale feature map set.

[0020] In specific implementation, the acquisition of a multi-view image sequence of the real estate to be verified, and the aerial triangulation and geodetic coordinate correction of the multi-view image sequence can be achieved in the following way: an aerial photograph is taken of the real estate area to be verified by a UAV equipped with a five-lens tilting camera to acquire original images in five directions: downward, forward, backward, left, and right. Each direction contains multiple consecutive images with spatial location information. All images are organized into a multi-view image sequence according to the shooting time sequence and camera position relationship, and the multi-view image sequence is then subjected to aerial triangulation and geodetic coordinate correction. Other embodiments may also use other methods, which are not limited here.

[0021] In some embodiments, feature extraction from the multi-view image sequence to obtain a multi-scale feature map can be achieved using the following steps: Each image in the multi-view image sequence is convolved, and deep feature maps at three scales are extracted through downsampling. By using skip connections, the lost information from each deep feature map during the downsampling process is fused together; All deep feature maps after fusion are restored to the original image size to obtain a multi-scale feature map set.

[0022] In specific implementation, each image in the multi-view image sequence undergoes convolution processing. Extracting deep feature maps at three scales through downsampling can be achieved as follows: Each image in the multi-view image sequence is input into a feature extraction network, which employs an encoder-decoder structure. In the encoder stage, the input image is downsampled four times using a convolutional layer with a stride of two. Each downsampling halves the feature map resolution, extracting feature maps at four levels: 1 / 2, 1 / 4, 1 / 8, and 1 / 16 of the original image size. The 1 / 16 level feature map has 32 channels, the 1 / 4 level has 16 channels, and the 1 / 16 level has 8 channels. In the decoder stage, skip connections are used to transfer the encoded data. The feature maps at each level of the encoder are passed to the corresponding resolution decoder layer. After the 1 / 16 level feature map is upsampled to restore the resolution to 1 / 8, it is concatenated with the 1 / 8 level feature map of the encoder. It is then upsampled again to 1 / 4 resolution and concatenated with the 1 / 4 level feature map of the encoder. It is then upsampled again to 1 / 2 resolution and concatenated with the 1 / 2 level feature map of the encoder. Finally, it is upsampled to the original image size. After each concatenation, a convolutional layer is used to refine the feature map, and finally, three scale feature maps are output. The resolutions of these three scale feature maps are 1 / 16, 1 / 4, and 1 times the original image size, respectively, and the number of channels are 32, 16, and 8, respectively. These three scale feature maps together constitute a multi-scale feature map set. Other embodiments may also use other methods, which are not limited here.

[0023] It should be noted that the deep feature map in this application is a feature matrix used to characterize the structural and texture information of an image at different scales. The lost information refers to the details such as edges and corners that are lost due to the reduction in resolution during the downsampling process. The multi-scale feature map set is a set of feature maps containing multiple resolution levels used for subsequent depth estimation and normal estimation.

[0024] In step 102, the depth of the matching cost is estimated for the multi-scale feature map set, and the surface normal information generated during the matching process is integrated to obtain a perception depth map. The normal under depth constraint is estimated for the multi-scale feature map set to obtain a cost normal map. The distortion region of the real estate model is determined based on the cost normal map and the perception depth map.

[0025] In some embodiments, the depth estimation of the matching cost of the multi-scale feature map set and the integration of surface normal information generated during the matching process to obtain the perceptual depth map can be achieved by the following steps: An initial depth map covering the entire depth search range is constructed based on the multi-scale feature map set; The local depth search range for the current stage is determined based on the initial depth map, and multiple depth planes are uniformly sampled within the local depth search range; Obtain the surface normal map estimated from the initial depth map, and upsample the surface normal map to a resolution that matches the preset feature map in the multi-scale feature map set; The sampled surface normal map and the matching cost map of each depth plane are concatenated to obtain a geometric perceptron with fused normal information; The geometric sensing volume is converted into a probability distribution, and the probability distribution is weighted and summed with all depth planes to obtain a sensing depth map.

[0026] In specific implementation, the initial depth map covering the complete depth search range based on the multi-scale feature map set can be constructed in the following way: In the first stage, at the lowest resolution level of the multi-scale feature map set, a depth search interval covering the entire scene undulation range is set. Multiple depth values ​​are uniformly sampled within this depth search interval as candidate depth planes. The feature map of the neighborhood view is projected onto each candidate depth plane through homography transformation to construct a global cost volume. The global cost volume is subjected to 3D convolution processing to smooth noise. Then, the cost volume is converted into a probability volume through the softmax function. Finally, the probability value of each pixel on all candidate depth planes is weighted and summed with the corresponding depth value to obtain the initial depth estimate of each pixel. The initial depth estimates of all pixels are combined into an initial depth map. Other embodiments can also be implemented in other ways, which are not limited here.

[0027] In specific implementation, determining the local depth search range for the current stage based on the initial depth map and uniformly sampling multiple depth planes within the local depth search range can be achieved in the following way: In the first stage, a fixed depth offset is set as the search radius, centered on the depth value of each pixel in the initial depth map, thereby determining the local depth search range corresponding to each pixel. Within this local depth search range, multiple depth values ​​are sampled at equal intervals, and these sampled depth values ​​are used as candidate depth planes for the current stage for subsequent fine depth estimation. Other embodiments may also use other methods, which are not limited here.

[0028] In specific implementation, obtaining the surface normal map estimated from the initial depth map and upsampling the surface normal map to a resolution matching the preset feature map in the multi-scale feature map set can be achieved in the following way: In the second stage, the initial depth map is input to the normal estimation branch, and local depth sampling planes are constructed near the depth values ​​of the initial depth map. The feature maps of the neighborhood view are projected onto each local depth sampling plane through homography transformation, and the local cost volume of normal inference is generated by aggregation. After performing three-dimensional convolution regularization processing on the local cost volume, it is compressed along the depth dimension and normalized along the feature channel to obtain an initial surface normal map that matches the resolution of the first stage. The resolution of this initial surface normal map is lower than the feature map resolution required in the current stage. The resolution of the surface normal map in the previous stage is enlarged by bilinear interpolation to make it consistent with the resolution of the feature map used in the current stage in the multi-scale feature map set, and the upsampled surface normal map is obtained for the construction of the geometric perception cost volume in the current stage. Other embodiments can also be implemented in other ways, which are not limited here.

[0029] In specific implementation, the sampled surface normal map and the matching cost map of each depth plane are concatenated to obtain a geometrically perceptual volume with fused normal information. This can be achieved in the following way: For each candidate depth plane, the feature map of the neighboring view is first projected onto the reference view through homography transformation to generate the matching cost map corresponding to that depth plane. Multiple copies of the upsampled surface normal map are made, making them the same number as the matching cost map. Then, the matching cost map on each depth plane is stitched together with one of the surface normal maps in the channel dimension to form a new multi-channel feature map. All the stitched feature maps corresponding to the depth planes are stacked together in depth order to form a three-dimensional geometrically perceptual cost volume. Other embodiments may also use other methods to achieve this, which are not limited here.

[0030] In a specific implementation, the geometric receptive volume is converted into a probability distribution, and the probability distribution is weighted and summed with all depth planes to obtain the receptive depth map. This can be achieved as follows: the geometric receptive cost volume is input into a 3D convolutional network for regularization, and then the regularized cost volume is converted into a probability distribution along the depth dimension using a softmax function. This probability distribution represents the probability of each pixel on each candidate depth plane. For each pixel, its probability value on all candidate depth planes is multiplied by its corresponding depth value, and then all the product results are summed to obtain the fine depth estimate of the pixel. The fine depth estimates of all pixels are combined to form the receptive depth map. Other embodiments may also use other methods, which are not limited here.

[0031] It should be noted that the initial depth map in this application is a two-dimensional matrix representing the initial distance estimate of each pixel in the scene from the camera; the local depth search range is a reduced depth interval centered on the initial depth value for fine depth estimation; the surface normal map is a two-dimensional vector field representing the orientation information of the surface where each pixel in the scene is located; the geometric perception cost volume is a three-dimensional data structure that integrates geometric structure information for fine depth estimation; and the perception depth map is a two-dimensional matrix representing the depth estimation result with higher geometric accuracy after optimization with the assistance of normal information. It is used as the depth constraint input for subsequent normal estimation and as the basic data for evaluating the geometric quality of the three-dimensional model. This facilitates the discovery of potential geometric distortions in the three-dimensional model through geometric consistency analysis of depth and normals, solving the problem that existing technologies cannot automatically identify geometric anomalies in the three-dimensional model.

[0032] In some embodiments, reference Figure 2 As shown, this figure is an exemplary flowchart for determining the cost normal map in some embodiments of this application. In this embodiment, the cost normal map is obtained by performing depth-constrained normal estimation on the multi-scale feature map set using the following steps: In step 1021, taking each pixel in the perceived depth map as the center, a preset depth sampling interval is set near the depth value of each pixel to construct multiple depth-first local sampling planes; In step 1022, the feature maps of the neighborhood view in the multi-scale feature map set are projected onto each local sampling plane and aggregated to obtain the local cost volume of the normal. In step 1023, a cost normal map is determined based on the local cost volume of the normal.

[0033] In a specific implementation, taking each pixel in the perceived depth map as the center, a preset depth sampling interval is set near the depth value of each pixel. The construction of multiple depth-first local sampling planes can be implemented in the following way: obtain the obtained perceived depth map, for each pixel in the perceived depth map, take its depth value as the center, set a fixed depth sampling interval, and uniformly sample multiple depth values ​​on both sides of the center value according to the set interval. These sampled depth values ​​are used as the local depth sampling planes corresponding to the pixel. Each pixel corresponds to a set of local depth sampling planes for subsequent normal estimation. Other embodiments can also be implemented in other ways, which are not limited here.

[0034] In specific implementation, projecting the feature maps from the neighborhood viewpoints in the multi-scale feature map set onto each local sampling plane and aggregating them to obtain the local cost body for normal inference can be achieved in the following way: For each local depth sampling plane of each pixel, project the feature maps from all neighborhood viewpoints in the multi-scale feature map set onto the reference viewpoint using a homography transformation matrix, generating a set of aligned feature maps on each depth sampling plane, aggregating these aligned feature maps along the channel dimension to obtain the cost map corresponding to that depth sampling plane, and stacking the cost maps corresponding to all depth sampling planes together in depth order to obtain a local cost body for normal inference; the cost method is determined based on the local cost body for normal inference. The line graph can be implemented as follows: the local cost volume of normal inference is input into a 3D convolutional network for regularization. The regularized cost volume still retains the depth dimension and feature channel dimension. The regularized cost volume is summed along the depth dimension to compress the 3D cost volume into a 2D feature map. The feature map is normalized along the feature channel dimension so that the output vector of each pixel has a length of one. The normalized 2D feature map is used as the cost normal map, where the three channel values ​​of each pixel in the cost normal map represent the three directional components of the surface normal of that pixel in 3D space. Other embodiments can also be implemented in other ways, which are not limited here.

[0035] It should be noted that the local depth sampling plane in this application is a set of candidate depth positions sampled near the depth value for normal estimation; the local cost volume for normal inference is a three-dimensional data structure of matching cost constructed based on depth constraints for surface normal estimation; the cost normal map is a two-dimensional vector field representing the estimation result of the orientation information of the surface where each pixel is located in the scene, used to perform pixel-by-pixel comparison with the normal map derived from the depth map to calculate the geometric consistency error, which facilitates the quantification of the surface smoothness and edge sharpness of the three-dimensional model from the local geometric structure level, and solves the problem of lack of quantitative basis for the detection of geometric distortion of three-dimensional models in the prior art.

[0036] In some embodiments, determining the distortion region of the real estate model based on the cost normal map and the perceived depth map can be achieved by the following steps: Perform differential calculations on the perceived depth map to obtain a derived normal map; The derived normal map and the cost normal map are compared pixel-by-pixel with the angle error to obtain a geometric consistency difference map. The difference map is segmented by thresholding, and pixel areas with angle errors exceeding a preset threshold are marked as distorted areas of the real estate model.

[0037] In specific implementation, the derived normal map is obtained by differential calculation of the perceived depth map in the following manner: For each pixel in the perceived depth map, its neighboring pixels in the four directions of up, down, left, and right are selected. The depth difference between the center pixel and the horizontally adjacent neighboring pixels is calculated as the horizontal gradient, and the depth difference between the center pixel and the vertically adjacent neighboring pixels is calculated as the vertical gradient. The horizontal and vertical gradients are combined into a two-dimensional gradient vector. This two-dimensional gradient vector is converted into a three-dimensional spatial vector in its orthogonal direction to obtain the surface normal estimate of the pixel. The surface normal estimates of all pixels are combined to form the derived normal map. The derived normal map is then compared with the cost normal map pixel by pixel in terms of angle error. The geometric consistency difference map can be obtained as follows: For each pixel position in the derived normal map and the cost normal map, the 3D normal vector of the derived normal map at that pixel and the 3D normal vector of the cost normal map at that pixel are multiplied by a dot product to obtain the cosine value of the angle between the two vectors. The angle between the two vectors is calculated using the inverse cosine function, and this angle value is used as the geometric consistency error of that pixel. The geometric consistency errors of the remaining pixel positions are then determined. The geometric consistency errors of all pixels are combined into a geometric consistency difference map, where a larger value for each pixel in the difference map indicates a more severe geometric contradiction between the depth and the normal at that location. Other implementation methods can also be used in other embodiments, which are not limited here.

[0038] In specific implementation, threshold segmentation of the difference map and marking pixel regions with angle errors exceeding a preset threshold as distorted regions of the real estate model can be achieved in the following way: set an angle error threshold, which is determined according to the application accuracy requirements of the 3D model; perform pixel-by-pixel judgment on the obtained geometric consistency difference map; mark pixels with angle errors greater than the set threshold as distorted pixels; connect all the marked distorted pixel positions into connected regions; and determine these connected regions as the geometrically distorted regions of the real estate 3D model for subsequent graphic calibration; other embodiments may also use other methods, which are not limited here.

[0039] It should be noted that the derived normal map in this application is a two-dimensional vector field of surface orientation information derived from the depth map through mathematical calculation; the geometric consistency difference map is an error distribution map used to represent the degree of geometric self-consistency between the depth map and the normal map; the distortion region is a set of pixel regions in the 3D model that have geometric structural anomalies, used to indicate the spatial locations in the 3D model that need to be focused on and calibrated, so as to facilitate the mapping of the 3D geometric verification results to the 2D mapping graphics for subsequent coordinate calibration, and solve the problem of lack of correlation between 3D model quality problems and 2D graphics calibration.

[0040] In step 103, the historical change records of the real estate to be verified are obtained, a change time sequence diagram of the real estate is constructed, and the neighborhood subgraphs of each change edge in the change time sequence diagram on multiple time slices are extracted to obtain the verification subgraph set of each change edge.

[0041] In some embodiments, obtaining historical change records of the real estate to be verified and constructing a change sequence diagram of the real estate can be achieved by the following steps: Extract all real estate units from the historical change records as nodes; Extract all ownership and graphic change events from the historical change records, treat all events as directed edges connecting the nodes before and after the change, and assign an edge weight to each directed edge based on the timestamp of the occurrence. Construct a change sequence diagram of the real estate based on the nodes and all directed edges.

[0042] In specific implementation, extracting all real estate units from the historical change records as nodes can be achieved in the following way: Obtain the historical change records of the real estate to be verified. These records contain the status information of multiple real estate units at different points in time. Parse the unique identifiers of all real estate units that have appeared in the historical change records. Each unique identifier corresponds to one real estate unit. Abstract each real estate unit as a node, and all nodes constitute a node set of a graph. Extract all ownership change and graphic change events from the historical change records. Treat all events as directed edges connecting nodes before and after the change, and assign edge weights based on the timestamp of the occurrence to each directed edge. This can be achieved in the following way: Parse each ownership change event and graphic change event from the historical change records. Each event record contains the identifier of the real estate unit before the change. The changes to the real estate unit identifier and the time of the event are used to abstract each event into a directed edge, with the edge pointing from the node before the change to the node after the change. The time of the event is used as the attribute value of the edge. All directed edges constitute the edge set of the graph. The change sequence graph of the real estate can be constructed based on the nodes and all directed edges in the following way: combine all nodes and all directed edges together to form a directed graph structure. This directed graph contains the node states and change relationships at multiple time points. Divide the graph into multiple static graph snapshots in time sequence. Each snapshot in time slice contains the nodes that exist at that time point and the changed edges that occurred at that time point. This directed graph with time dimension information is used as the change sequence graph of the real estate. Other embodiments can also be implemented in other ways, which are not limited here.

[0043] It should be noted that directed edges are temporally directed connections used to represent the change relationships between real estate units; the change sequence graph of real estate is a graph structure data representing the evolution relationship of real estate units over time. It is used to transform historical change records into a mathematical structure that can be processed by graph algorithms to extract the spatiotemporal context of change events. This facilitates capturing the degree of deviation between a single change event and historical patterns from a global evolutionary perspective, and solves the problem of missing anomalies due to ignoring the change sequence correlation.

[0044] In some embodiments, extracting the neighborhood subgraphs of each change edge in the change time sequence graph across multiple time slices to obtain the verification subgraph set for each change edge can be achieved using the following steps: Obtain the timestamp of each change edge to be detected in the change sequence graph and the time window to be examined; Select one of the changed edges as the selected changed edge; In each time slice within the time window of the selected change edge, the connection strength between each node and the center node in the change time sequence diagram is determined, taking the two nodes of the selected change edge as the center node. Based on all the connection strengths, determine multiple context nodes of the selected changed edge and construct a local neighborhood subgraph of the selected changed edge in that time slice; Merge the local neighborhood subgraphs on all time slices to obtain the verification subgraph set of the selected changed edge; Continue to determine the verification sub-graphite set of the remaining changed edges.

[0045] In specific implementation, obtaining the timestamp of each change edge to be detected in the change sequence graph and the time window to be examined can be achieved in the following way: Obtain all change edges to be detected from the change sequence graph. For each change edge to be detected, read the timestamp corresponding to that edge as the current time point. Set a time window length parameter, which indicates how many time units need to be backtracked. Subtract the time window length from the current time point to obtain the window start time. Use the current time point as the window end time, forming a continuous time interval. Within each time slice of the time window of the selected change edge, using the two nodes of the selected change edge as the center node, the connection strength between each node in the change sequence graph and the center node can be determined in the following way: For the two end nodes of the selected change edge as the center node, within each time slice of the time window, use a graph diffusion algorithm to calculate the connection strength between the nodes in the static graph of that time slice. The degree of association between a node and two central nodes is determined by iteratively calculating the access probability of each node relative to the central nodes using a personalized PageRank algorithm. The access probability value is used as the connection strength between the node and the central nodes, with higher connection strength indicating greater structural importance of the node. The determination of multiple context nodes for the selected change edge based on all connection strengths, and the construction of a local neighborhood subgraph of the selected change edge in that time slice, can be achieved as follows: In each time slice, based on all connection strength values, all nodes are sorted from high to low according to connection strength. A preset number of nodes with the highest connection strength are selected as context nodes. The two end nodes of the selected change edge, along with these context nodes, form the subgraph node set for that time slice. All edges between these nodes in that time slice are extracted, and these nodes and edges constitute the local neighborhood subgraph for that time slice. Other embodiments may also employ other methods, which are not limited here.

[0046] In specific implementation, merging the local neighborhood subgraphs on all time slices to obtain the verification subgraph set for the selected changed edge can be achieved in the following way: collect the local neighborhood subgraphs on all time slices within the time window, save the subgraphs on each time slice independently, keep their own timestamp information, and use these subgraph sets with time order as the verification subgraph set for the selected changed edge for subsequent spatiotemporal anomaly detection; other embodiments may also use other methods to achieve this, which are not limited here.

[0047] It should be noted that the connection strength in this application is a numerical indicator that measures the degree of connection between other nodes in the graph and the central node; the local neighborhood subgraph is a small-scale graph representing the local structure of the target changed edge in a single time slice; the verification subgraph set is a set of subgraphs used to represent the local structure of the target changed edge in multiple time slices, which is used to provide local structure data containing spatiotemporal context information as input for anomaly detection, so as to facilitate modeling the spatiotemporal coupling relationship between the change event and the surrounding environment at the graph structure level, and solve the problem of difficulty in simultaneously capturing spatial structural mutations and temporal evolution anomalies.

[0048] In step 104, encoding fusion and spatiotemporal anomaly detection are performed on all nodes in each verification sub-map set to obtain multiple abnormal change events with associated coordinate accuracy deviations. Based on the distorted area and all abnormal change events, the distorted coordinate point set of the survey map of the real estate is determined.

[0049] In some embodiments, the following steps can be used to perform encoding fusion and spatiotemporal anomaly detection on all nodes in each verification sub-graphetrating to obtain multiple abnormal change events with related coordinate precision deviations: Determine the diffusion code, distance code, time code, and coordinate precision deviation code for all nodes in each verification sub-map set; All diffusion codes, distance codes, time codes, and coordinate precision deviation codes are added element by element to obtain the comprehensive coding vector of each node. All comprehensive coding vectors are stacked in chronological order to form a node coding matrix. The node encoding matrix is ​​input into an encoder based on a bidirectional self-attention mechanism, and the attention weight between any two nodes in the subgraph is calculated to obtain a node hidden representation that incorporates spatiotemporal context information. The hidden representation of the nodes is subjected to mean pooling to obtain the edge embedding vector of each changed edge; Anomaly probability determination is performed on the coordinate precision deviation association of all edge embedding vectors to obtain multiple abnormal change events of associated coordinate precision deviation.

[0050] In specific implementation, the diffusion encoding, distance encoding, time encoding, and coordinate precision deviation encoding of all nodes in each verification sub-graphet can be determined in the following way: For each node in the verification sub-graphet, firstly, calculate the global diffusion value of the node in the change sequence graph. This diffusion value is obtained by calculating the access probability of the node relative to all nodes using a personalized PageRank algorithm. Map the ranking position of the diffusion value among all nodes into a fixed-length vector, which serves as the diffusion encoding of the node. For each node in the verification sub-graphet, calculate the shortest path length from the node to the two end nodes of the selected change edge, and take the middle distance value. The smaller value is used as the distance value of the node, and this distance value is mapped to a fixed-length vector as the distance code of the node; for each node in the verification sub-map, the absolute value of the time difference between the time slice where the node is located and the time when the selected changed edge occurs is calculated, and this time difference value is mapped to a fixed-length vector as the time code of the node; for each node in the verification sub-map, the coordinate accuracy deviation value of its associated changed edge is extracted, and the ratio of the deviation value to the surveying specification threshold is mapped to a fixed-length vector as the coordinate accuracy deviation code of the node; other embodiments may also use other methods to implement this, which are not limited here.

[0051] In specific implementation, all diffusion codes, distance codes, time codes, and coordinate precision deviation codes are added element by element to obtain the comprehensive coding vector of each node. The node coding matrix is ​​formed by stacking all the comprehensive coding vectors in chronological order. This can be achieved as follows: For each node in the verification sub-graph set, the node's diffusion code vector, distance code vector, coordinate precision deviation code vector, and time code vector are numerically added at the same dimensional position to obtain a new vector as the comprehensive coding vector of that node. All nodes are arranged according to the order of their time slices, and the comprehensive coding vector of each node is used as a row of the matrix, and they are stacked in order to form a two-dimensional node coding matrix. Other embodiments may also use other methods to achieve this, which are not limited here.

[0052] In a specific implementation, the node encoding matrix is ​​input into an encoder based on a bidirectional self-attention mechanism, and the attention weights between any two nodes in the subgraph are calculated to obtain the node hidden representation fused with spatiotemporal context information. This can be achieved in the following way: the node encoding matrix is ​​input into an encoder based on a bidirectional self-attention mechanism, which contains multiple attention heads. For each attention head, the node encoding matrix is ​​multiplied by three different weight matrices to obtain a query matrix, a key matrix, and a value matrix. The attention weight matrix is ​​obtained by calculating the dot product of the query matrix and the key matrix and then normalizing it using softmax. The attention weight matrix is ​​multiplied by the value matrix to obtain a weighted node representation. The outputs of multiple attention heads are concatenated and then subjected to a linear transformation to obtain a hidden representation vector for each node that incorporates information from other nodes in the subgraph. The hidden representations of all nodes constitute the node hidden representation matrix. Other implementation methods can also be used in other embodiments, which are not limited here.

[0053] In specific implementation, the node hidden representation is averaged and pooled to obtain the edge embedding vector of each changed edge. This can be achieved as follows: For the node hidden representation matrix, the average value of all node hidden representation vectors is calculated along the node dimension. That is, the values ​​in each feature dimension are summed and divided by the number of nodes to obtain a fixed-length vector as the edge embedding vector of the changed edge. This vector integrates the spatiotemporal information of all nodes in the subgraph. The abnormal probability determination of coordinate precision deviation association for all edge embedding vectors can be achieved as follows: The edge embedding vector of each changed edge is input into a fully connected layer. This fully connected layer maps the edge embedding vector to a scalar value. The sigmoid function is used to convert the scalar value into a probability value between 0 and 1. This probability value is used as the abnormal probability of the changed edge. Changed edges with an abnormal probability greater than a preset threshold are judged as abnormal change events. All changed edges judged as abnormal and their related information are output. Other implementation methods can also be used in other embodiments, which are not limited here.

[0054] It should be noted that the diffusion encoding in this application is a numerical vector used to represent the importance level of a node in the global structure; the distance encoding is a numerical vector used to represent the topological distance from a node to a target change edge; the time encoding is a numerical vector used to represent the relative time difference between the time slice where the node is located and the current time; the comprehensive encoding vector is a feature vector used to represent the comprehensive role of a node in the spatiotemporal structure; the node hidden representation is a deep feature vector used to represent the node after updating with context information; the abnormal change event is a change behavior in which the spatiotemporal pattern of the real estate model and landmarks is significantly inconsistent with the normal historical pattern; the coordinate precision deviation encoding is a numerical vector used to represent the degree of coordinate precision deviation of the node's associated change edge; the abnormal change event with associated coordinate precision deviation refers to a change behavior in which the spatiotemporal pattern of the real estate change behavior is significantly inconsistent with the normal historical pattern, and is accompanied by a boundary point coordinate precision deviation exceeding the surveying and mapping specifications.

[0055] In some embodiments, reference Figure 3 As shown, this figure is an exemplary flowchart for determining the set of distorted coordinate points in some embodiments of this application. In this embodiment, determining the set of distorted coordinate points of the survey map of the real estate based on the distorted area and all abnormal change events can be achieved by the following steps: In step 1041, the distorted region is projected onto a two-dimensional plane to obtain the corresponding two-dimensional distorted region; In step 1042, multiple graphic change events are extracted from all abnormal change events, and the boundary points and boundary lines involved in all graphic change events are located as the distortion locations associated with the events; In step 1043, the two-dimensional distorted region is overlaid and merged with all the distorted locations to obtain the set of distorted coordinate points of the survey map of the real estate.

[0056] In practice, multiple graphic change events are extracted from all abnormal change events. The boundary points and boundary lines involved in all graphic change events are located as the distortion locations associated with the events. This can be achieved in the following way: From the spatial location deviation area vector map with geographic coordinates, extract all candidate boundary point coordinates that conform to the real estate surveying and mapping specifications. Filter according to the minimum spacing requirements of boundary points in the "Real Estate Surveying and Mapping Specifications," eliminating redundant candidate points with spacing smaller than the specification requirements, retaining the discrete boundary point candidate coordinates that conform to the specifications, and recording the degree of spatial location deviation corresponding to the candidate points; From all abnormal change events, events related to graphic changes are selected. For each graphic change event, the event is parsed from the historical change records. For the corresponding changes, the coordinates of the boundary points and the equations of the boundary lines involved in the changes are extracted. The coordinates of these boundary points and the points on the boundary lines are used as the distortion locations associated with the events. The two-dimensional distortion area is overlaid and merged with all the distortion locations to obtain the distortion coordinate point set of the real estate map. This can be achieved in the following way: convert all the pixel coordinates contained in the two-dimensional distortion area into geographic coordinates to obtain a set of coordinate points. Add the coordinates of all the distortion locations associated with the events to this set of coordinate points, remove duplicate coordinate points, and organize all the coordinate points into a point set according to their spatial proximity. This point set is used as the distortion coordinate point set of the real estate map. Other embodiments may also use other methods, which are not limited here.

[0057] It should be noted that the distortion location associated with the event in this application is the location of the specific graphic element involved in the abnormal change event, and the set of distortion coordinate points is the set of all graphic coordinate points that need to be calibrated.

[0058] In step 105, the coordinates of the survey map are calibrated based on the distorted coordinate point set, using geodetic control points and cadastral control points as confidence references.

[0059] In some embodiments, calibrating the coordinates of the survey map based on the distorted coordinate point set, using geodetic control points and cadastral control points as confidence references, can be achieved through the following steps: Obtain the geodetic control points and cadastral control points around each distorted point in the distorted coordinate point set, and extract the accurate geographic coordinates of all reference datums; Determine the coordinate correction amount for multiple distortion points based on all confidence reference points; The coordinates of the corresponding distorted points in the surveyed figure are calibrated based on all coordinate corrections.

[0060] In specific implementation, the acquisition of geodetic control points and cadastral control points surrounding each distorted point in the distorted coordinate point set, and the extraction of accurate mapping geographic coordinates of all reference benchmarks, can be achieved in the following way: From the basic surveying results of the real estate surveying area, extract the national-level geodetic control points and local cadastral control points within the area to be verified. These control points have all undergone precise measurement, and their coordinate accuracy meets the requirements of the "National Geodetic Surveying Standard" and the "Real Estate Surveying Standard." Extract the accurate mapping geographic coordinates (2000 National Geodetic Coordinate System) of all control points as the sole confidence reference benchmark for coordinate calibration. For each distorted point in the distorted coordinate point set, search for the multiple unmarked boundary points closest to the distorted point in the original surveying map. Use these boundary points as confidence reference points for the distorted point. Each distorted point corresponds to a set of confidence reference points, and these reference points have high positional accuracy. Based on all confidence reference points, determine multiple distortion points. The coordinate correction of a point can be implemented as follows: for each distorted point, the coordinates of its corresponding multiple confidence reference points are weighted and averaged. The weights are determined based on the distance between the reference points and the distorted point; the closer the distance, the greater the weight. This yields a target position coordinate. The current coordinates of the distorted point are subtracted from the target position coordinates to obtain the coordinate correction vector of the distorted point. This vector contains offsets in the horizontal and vertical directions. The calibration of the coordinates of the corresponding distorted points in the surveying map based on all the coordinate corrections can be implemented as follows: for each distorted point, its current coordinates are added to the coordinate correction vector to obtain a new coordinate value for the point. The new coordinate value replaces the coordinates of the corresponding point in the original map. After all the coordinates of the distorted points are updated, the edges connected to the distorted points in the map are reconnected and a topology check is performed. The resulting map is used as the calibrated real estate surveying map. Other implementation methods can also be used in other embodiments, which are not limited here.

[0061] In another aspect, in some embodiments, this application provides a real estate survey map graphic verification system, with reference to... Figure 4 The figure is a schematic diagram of the structure of a real estate surveying and mapping graphic verification system according to some embodiments of this application. The real estate surveying and mapping graphic verification system includes: an acquisition module 401, a processing module 402, and an execution module 403, which are described below: The acquisition module 401 in this application is mainly used to acquire the multi-view image sequence of the real estate to be verified, perform aerial triangulation and geodetic coordinate correction on the multi-view image sequence, and extract features from the multi-view image sequence to obtain a multi-scale feature map set. Processing module 402, in this application, is used to perform depth estimation of matching cost on the multi-scale feature map set, integrate the surface normal information generated during the matching process to obtain a perception depth map, perform normal estimation under depth constraints on the multi-scale feature map set to obtain a cost normal map, and determine the distortion region of the real estate model based on the cost normal map and the perception depth map. It should be noted that the processing module 402 in this application is also used to obtain the historical change records of the real estate to be verified, construct the change time sequence diagram of the real estate, extract the neighborhood subgraph of each change edge in the change time sequence diagram in multiple time slices, and obtain the verification subgraph set of each change edge. In addition, it should be noted that the processing module 402 in this application is also used to perform encoding fusion and spatiotemporal anomaly detection on all nodes in each verification sub-map set to obtain multiple abnormal change events with related coordinate precision deviations, and to determine the set of distorted coordinate points of the survey map of the real estate based on the distorted area and all abnormal change events. The execution module 403 in this application is mainly used to calibrate the coordinates of the survey map based on the distorted coordinate point set, using geodetic control points and cadastral control points as confidence reference benchmarks.

[0062] In addition, this application also provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described real estate surveying and mapping graphic verification method.

[0063] In some embodiments, reference Figure 5 The figure is a schematic diagram of the structure of a computer device for implementing a method for verifying the shape of real estate survey maps according to some embodiments of this application. The method for verifying the shape of real estate survey maps in the above embodiments can be implemented through... Figure 5 The computer device shown is used to implement this, and the computer device includes at least one processor 501, a communication bus 502, a memory 503, and at least one communication interface 504.

[0064] Processor 501 can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0065] The communication bus 502 can be used to transmit information between the aforementioned components.

[0066] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CDROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 503 may exist independently and be connected to processor 501 via communication bus 502. Memory 503 may also be integrated with processor 501.

[0067] The memory 503 stores program code for executing the scheme of this application, and its execution is controlled by the processor 501. The processor 501 executes the program code stored in the memory 503. The program code may include one or more software modules. The method used in the above embodiments can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0068] Communication interface 504 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0069] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single CPU) processor or a multi-core (multi CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0070] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0071] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for verifying real estate surveying and mapping graphics.

[0072] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0073] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for verifying the graphics of real estate survey maps, characterized in that, Includes the following steps: A multi-view image sequence of the real estate to be verified is obtained, aerial triangulation and geodetic coordinate correction are performed on the multi-view image sequence, and feature extraction is performed on the multi-view image sequence to obtain a multi-scale feature map set. Depth estimation of matching cost is performed on the multi-scale feature map set, and surface normal information generated during the matching process is integrated to obtain a perception depth map. Normal estimation under depth constraints is performed on the multi-scale feature map set to obtain a cost normal map. The distortion region of the real estate model is determined based on the cost normal map and the perception depth map. Obtain the historical change records of the real estate to be verified, construct the change time sequence diagram of the real estate, extract the neighborhood subgraphs of each change edge in the change time sequence diagram in multiple time slices, and obtain the verification subgraph set of each change edge; Encoding fusion and spatiotemporal anomaly detection are performed on all nodes in each verification sub-map set to obtain multiple abnormal change events with related coordinate accuracy deviations. Based on the distorted area and all abnormal change events, the distorted coordinate point set of the survey map of the real estate is determined. Using geodetic control points and cadastral control points as confidence references, the coordinates of the survey map are calibrated based on the distorted coordinate point set.

2. The method as described in claim 1, characterized in that, The depth estimation of the matching cost for the multi-scale feature map set, and the integration of surface normal information generated during the matching process, to obtain the perceptual depth map specifically includes: An initial depth map covering the entire depth search range is constructed based on the multi-scale feature map set; The local depth search range for the current stage is determined based on the initial depth map, and multiple depth planes are uniformly sampled within the local depth search range; Obtain the surface normal map estimated from the initial depth map, and upsample the surface normal map to a resolution that matches the preset feature map in the multi-scale feature map set; The sampled surface normal map and the matching cost map of each depth plane are concatenated to obtain a geometric perceptron with fused normal information; The geometric sensing volume is converted into a probability distribution, and the probability distribution is weighted and summed with all depth planes to obtain a sensing depth map.

3. The method as described in claim 1, characterized in that, The process of estimating the normals of the multi-scale feature map under depth constraints to obtain the cost normal map specifically includes: Using each pixel in the perceived depth map as the center, a preset depth sampling interval is set near the depth value of each pixel to construct multiple depth-first local sampling planes; The feature maps from the neighborhood perspective in the multi-scale feature map set are projected onto each local sampling plane and aggregated to obtain the local cost volume of the normal. The cost normal map is determined based on the local cost volume of the normal.

4. The method as described in claim 1, characterized in that, Determining the distortion region of the real estate model based on the cost normal map and the perceived depth map specifically includes: Perform differential calculations on the perceived depth map to obtain a derived normal map; The derived normal map and the cost normal map are compared pixel-by-pixel with the angle error to obtain a geometric consistency difference map. The difference map is segmented by thresholding, and pixel areas with angle errors exceeding a preset threshold are marked as distorted areas of the real estate model.

5. The method as described in claim 1, characterized in that, Obtaining the historical change records of the real estate to be verified and constructing a change sequence diagram of the real estate specifically includes: Extract all real estate units from the historical change records as nodes; Extract all ownership and graphic change events from the historical change records, treat all events as directed edges connecting the nodes before and after the change, and assign an edge weight to each directed edge based on the timestamp of the occurrence. Construct a change sequence diagram of the real estate based on the nodes and all directed edges.

6. The method as described in claim 1, characterized in that, Extracting the neighborhood subgraphs of each change edge in the change time sequence graph across multiple time slices to obtain the verification subgraph set for each change edge specifically includes: Obtain the timestamp of each change edge to be detected in the change sequence graph and the time window to be examined; Select one of the changed edges as the selected changed edge; In each time slice within the time window of the selected change edge, the connection strength between each node and the center node in the change time sequence diagram is determined, taking the two nodes of the selected change edge as the center node. Based on all the connection strengths, determine multiple context nodes of the selected changed edge and construct a local neighborhood subgraph of the selected changed edge in that time slice; Merge the local neighborhood subgraphs on all time slices to obtain the verification subgraph set of the selected changed edge; Continue to determine the verification sub-graphite set of the remaining changed edges.

7. The method as described in claim 1, characterized in that, Encoding fusion and spatiotemporal anomaly detection are performed on all nodes in each verification sub-graphetrating to obtain multiple abnormal change events with related coordinate precision deviations, specifically including: Determine the diffusion code, distance code, time code, and coordinate precision deviation code for all nodes in each verification sub-map set; All diffusion codes, distance codes, time codes, and coordinate precision deviation codes are added element by element to obtain the comprehensive coding vector of each node. All comprehensive coding vectors are stacked in chronological order to form a node coding matrix. The node encoding matrix is ​​input into an encoder based on a bidirectional self-attention mechanism, and the attention weight between any two nodes in the subgraph is calculated to obtain a node hidden representation that incorporates spatiotemporal context information. The hidden representation of the nodes is subjected to mean pooling to obtain the edge embedding vector of each changed edge; Anomaly probability determination is performed on the coordinate precision deviation association of all edge embedding vectors to obtain multiple abnormal change events of associated coordinate precision deviation.

8. The method as described in claim 1, characterized in that, The specific steps for determining the set of distorted coordinate points of the survey map of the real estate based on the distorted area and all abnormal change events include: The distorted region is projected onto a two-dimensional plane to obtain the corresponding two-dimensional distorted region; Extract multiple graphical change events from all abnormal change events, locate the boundary points and boundary lines involved in all graphical change events, and use them as the distortion locations associated with the events; The two-dimensional distorted region is overlaid and merged with all the distorted locations to obtain the set of distorted coordinate points of the survey map of the real estate.

9. The method as described in claim 1, characterized in that, Using geodetic control points and cadastral control points as confidence references, the calibration of the coordinates of the survey map based on the distorted coordinate point set specifically includes: Obtain the geodetic control points and cadastral control points around each distorted point in the distorted coordinate point set, and extract the accurate geographic coordinates of all reference datums; Determine the coordinate correction amount for multiple distortion points based on all confidence reference points; The coordinates of the corresponding distorted points in the surveyed figure are calibrated based on all coordinate corrections.

10. A real estate surveying and mapping graphic verification system, characterized in that, include: The acquisition module is used to acquire a multi-view image sequence of the real estate to be verified, perform aerial triangulation and geodetic coordinate correction on the multi-view image sequence, and extract features from the multi-view image sequence to obtain a multi-scale feature map set. The processing module is used to perform depth estimation of matching cost on the multi-scale feature map set, integrate the surface normal information generated during the matching process to obtain a perception depth map, perform normal estimation under depth constraints on the multi-scale feature map set to obtain a cost normal map, and determine the distortion region of the real estate model based on the cost normal map and the perception depth map. The processing module is also used to obtain historical change records of the real estate to be verified, construct a change time sequence diagram of the real estate, extract neighborhood subgraphs of each change edge in the change time sequence diagram in multiple time slices, and obtain a set of verification subgraphs for each change edge. The processing module is also used to perform encoding fusion and spatiotemporal anomaly detection on all nodes in each verification sub-map set to obtain multiple abnormal change events with related coordinate precision deviations, and to determine the distorted coordinate point set of the survey map of the real estate based on the distorted area and all abnormal change events. The execution module is used to calibrate the coordinates of the survey map based on the distorted coordinate point set, using geodetic control points and cadastral control points as confidence reference benchmarks.

Images