Three-dimensional temporal AI monitoring methods and systems for urban spatial governance

Abstract

This invention provides a three-dimensional temporal AI monitoring method and system for urban spatial governance. The method includes: preprocessing the current three-dimensional model data to remove ground interference data and acquire the current three-dimensional ground data; spatially calibrating the current three-dimensional ground data and previous three-dimensional ground data to obtain a differential three-dimensional model; dilating the differential three-dimensional model to obtain a differential stereo model; and identifying changed objects in the differential stereo model using an AI detection model. By removing interference points such as the ground and ground vegetation, clean modeling with ground interference elimination is achieved; the differential stereo model is constructed for areas where there are changes between two periods of data, which, compared with traditional sparse point sets, accurately represents the actual shape of the changed area, clarifies the subject of change, and improves the accuracy of detection; and by combining with the AI ​​detection model, identification results for changed areas are automatically and quickly generated.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a three-dimensional time-series AI monitoring method and system for urban spatial governance. Background Technology

[0002] Urban spatial governance is a fundamental term in the field of urban planning, referring to the holistic arrangement of a city's three-dimensional form and functional environment through systematic planning. Urban spatial change detection is one of the basic steps in achieving urban spatial governance.

[0003] Urban spatial change refers to the spatial morphological changes that occur in a fixed area within a certain period of time, usually manifested as the increase or decrease of buildings. The current methods for detecting urban spatial change mainly include detection based on two-dimensional image data comparison and detection based on three-dimensional point cloud data comparison.

[0004] For two-dimensional image data comparison and detection, it can only reflect two-dimensional changes in the ground surface and cannot accurately represent the three-dimensional changes in buildings, resulting in low detection accuracy. For three-dimensional point cloud data comparison, firstly, the point cloud data contains a lot of noise points, which can easily affect the accurate identification of buildings. Secondly, after registering the two periods of point cloud data, it directly calculates the distance difference between points or between points and surfaces, resulting in a sparse set of points that cannot clearly identify the subject of change. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a three-dimensional temporal AI monitoring method and system for urban spatial governance. This invention addresses the technical problems of existing technologies, such as low detection accuracy when using two-dimensional images for spatial change detection, and insufficient accuracy when using three-dimensional point clouds for spatial change detection due to excessive noise. Furthermore, these technologies only compare points in the point cloud data and cannot clearly identify the subject of the change.

[0006] To achieve the above objectives, firstly, embodiments of this application provide a three-dimensional temporal AI monitoring method for urban spatial governance, comprising the following steps:

[0007] Acquire the current three-dimensional model data, preprocess the current three-dimensional model data to remove ground interference data, and acquire the current three-dimensional ground data;

[0008] Spatial calibration is performed on the current and previous three-dimensional above-ground data to obtain a difference three-dimensional model;

[0009] The differential 3D model is dilated to obtain a differential stereo model, and the differential stereo model is used to identify the changed objects in the differential stereo model.

[0010] Furthermore, the current 3D model data includes several data points, and the step of preprocessing the current 3D model data to remove ground interference data and obtain the current 3D ground data includes:

[0011] Noise points are selected from a plurality of data points and removed to update the current three-dimensional model data to the current three-dimensional denoised data.

[0012] Perform coordinate transformation on the current 3D denoised data to obtain the current 3D transformed data;

[0013] The CSF algorithm is used to remove terrain points and ground vegetation points from the current 3D transformed data to obtain the current 3D ground data.

[0014] Furthermore, the step of selecting noise points from the plurality of data points includes:

[0015] One of the data points is selected as the reference point, and the remaining data points are selected as reference points. The Euclidean distance between the reference point and the reference point is obtained. Based on the Euclidean distance, several neighboring points are selected from several reference points.

[0016] The mean distance is obtained based on the Euclidean distance between the reference point and the neighboring points, and a distance threshold is constructed based on the mean distance.

[0017] The mean distance is compared with the distance threshold. If the mean distance is greater than the distance threshold, the reference point is determined to be a noise point.

[0018] Furthermore, the formula for obtaining the distance threshold is:

[0019] ,

[0020] in, Indicates the distance threshold. The mean of the distances from the mean. Represents the threshold multiplier. This represents the mean distance when the i-th data point is the reference point. This represents the total number of data points, where, .

[0021] Furthermore, the step of spatially calibrating the current 3D terrestrial data and previous 3D terrestrial data to obtain a difference 3D model includes:

[0022] Obtain the current descriptor of the current data point in the current three-dimensional aboveground data, and obtain the past descriptor of the past data point in the past three-dimensional aboveground data;

[0023] The current period descriptor and the previous period descriptor are compared to match the current period data points and the previous period data points as registration point pairs, and the registration transformation matrix is ​​obtained. Based on the registration transformation matrix, the current period three-dimensional ground data is mapped to the previous period three-dimensional ground data, and the overlapping ground data is obtained.

[0024] Core points are set in the overlapping ground data, and selection cylinders are constructed based on the core points. The selection cylinders are used to select the current point set from the current three-dimensional ground data and to select the past point set from the past three-dimensional ground data.

[0025] Based on the previous point set, obtain the normal vector corresponding to the core point, determine whether the core point is a change point based on the normal vector, and construct a difference 3D model based on several change points.

[0026] Furthermore, the step of obtaining the normal vector corresponding to the core point based on the previous point set is as follows:

[0027] Obtain the three-dimensional covariance matrix of the previous point set, and perform eigenvalue decomposition on the three-dimensional covariance matrix to obtain the normal vector corresponding to the core point;

[0028] The step of determining whether the core point is a point of change based on the normal vector includes:

[0029] Obtain the first average projection position of the past point set on the normal vector direction, and obtain the second average projection position of the current point set on the normal vector direction;

[0030] The normal distance is obtained based on the first average projection position and the second average projection position;

[0031] The normal distance is proportional to a distance threshold. If the normal distance is greater than the distance threshold, the core point is determined to be a point of change.

[0032] Furthermore, the formula for obtaining the three-dimensional covariance matrix is:

[0033] ,

[0034] in, This represents the three-dimensional covariance matrix of the set of past points corresponding to the m-th core point. This represents the total number of past point sets corresponding to the m-th core point. This represents the j-th point in the set of past points corresponding to the m-th core point. Indicates the transpose operator;

[0035] The formula for obtaining the first average projection position is:

[0036] ,

[0037] in, Let represent the first average projection position of the set of past points corresponding to the m-th core point along the normal vector direction of the m-th core point. This represents the m-th core point. Let m represent the normal vector of the m-th core point.

[0038] Furthermore, the step of dilating the differential 3D model to obtain a differential 3D model includes:

[0039] An initial mesh model is constructed based on the minimum bounding rectangle of the differential 3D model. The initial mesh model includes several initial meshes. The change points in the differential 3D model are mapped into the initial mesh model to distinguish the several initial meshes into initial voxel meshes and background meshes.

[0040] Determine whether the initial voxel grid exists in the adjacent grids of the background grid. If the initial voxel grid exists in the adjacent grids of the background grid, then select the background grid as the dilated voxel grid to complete a single dilation.

[0041] Perform several single dilations to combine the initial voxel mesh and the dilated voxel mesh into a differential stereo model.

[0042] Furthermore, the differential stereo model includes several groups of stereo voxel grids, and the AI ​​detection model includes a feature extraction module and a classification probability module. The step of performing object recognition on the differential stereo model using the AI ​​detection model includes:

[0043] The differential stereo model is photographed from different shooting angles to obtain several model snapshots;

[0044] The model snapshot is input into the feature extraction module to generate several candidate regions in the model snapshot, and the candidate regions correspond to the three-dimensional voxel mesh group;

[0045] The model snapshot of the generated candidate region is input into the classification probability module to determine whether the 3D voxel mesh group is a building.

[0046] Secondly, embodiments of this application provide a three-dimensional temporal AI monitoring system for urban spatial governance, applied to the three-dimensional temporal AI monitoring method for urban spatial governance as described in the first aspect above, the system comprising:

[0047] The acquisition module is used to acquire the current three-dimensional model data, preprocess the current three-dimensional model data to remove ground interference data, and acquire the current three-dimensional ground data.

[0048] The comparison module is used to perform spatial calibration on the current three-dimensional aboveground data and the previous three-dimensional aboveground data to obtain the difference three-dimensional model;

[0049] The execution module is used to perform dilation processing on the differential 3D model to obtain a differential stereo model, and to identify the changed objects in the differential stereo model through an AI detection model.

[0050] Thirdly, embodiments of this application provide a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the three-dimensional temporal AI monitoring method for urban spatial governance as described in the first aspect above.

[0051] Fourthly, embodiments of this application provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the three-dimensional temporal AI monitoring method for urban spatial governance as described in the first aspect above.

[0052] Compared with existing technologies, the beneficial effects of this invention are as follows: By preprocessing the current three-dimensional model data, the influence of noise points is removed, and interference points such as the ground and ground vegetation are eliminated, leaving only the points corresponding to the above-ground structures. This achieves clean modeling to eliminate ground interference, ensuring subsequent detection efficiency and accuracy. By spatially calibrating the data from the two periods, the regions where changes exist between the two periods are constructed into the differential 3D model. For the identification of changed objects, compared with traditional sparse point sets, the differential 3D model accurately represents the actual shape of the changed region, clarifies the subject of change, and further improves the accuracy of detection. By combining with the AI ​​detection model, recognition results for changed regions are automatically and quickly generated, and semantic-level judgments such as addition or disappearance can be made. Attached Figure Description

[0053] Figure 1 This is a flowchart of the three-dimensional temporal AI monitoring method for urban spatial governance in the first embodiment of the present invention;

[0054] Figure 2 This is a structural block diagram of a three-dimensional temporal AI monitoring system for urban spatial governance according to the second embodiment of the present invention;

[0055] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0056] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0057] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0059] Please see Figure 1 The first embodiment of the present invention provides a three-dimensional temporal AI monitoring method for urban spatial governance, comprising the following steps:

[0060] S10: Obtain the current three-dimensional model data, preprocess the current three-dimensional model data to remove ground interference data, and obtain the current three-dimensional ground data;

[0061] In this embodiment, the current 3D model data is acquired using a drone.

[0062] Step S10 includes:

[0063] S110: Select noise points from a plurality of data points and remove the noise points to update the current three-dimensional model data to the current three-dimensional denoised data;

[0064] Specifically, one of the data points is selected as a reference point, and the remaining data points are selected as reference points. The Euclidean distance between the reference point and the reference point is obtained, and several neighboring points are selected from the several reference points based on the Euclidean distance.

[0065] Understandably, after obtaining the Euclidean distance, the Euclidean distance is compared with a distance threshold, and the reference point corresponding to the Euclidean distance being less than the distance threshold is selected as the neighboring point.

[0066] The mean distance is obtained based on the Euclidean distance between the reference point and the neighboring points, and a distance threshold is constructed based on the mean distance.

[0067] The formula for obtaining the distance threshold is:

[0068] ,

[0069] in, Indicates the distance threshold. The mean of the distances from the mean. Represents the threshold multiplier. This represents the mean distance when the i-th data point is the reference point. This represents the total number of data points, where, .

[0070] The mean distance is compared with the distance threshold. If the mean distance is greater than the distance threshold, the reference point is identified as a noise point. By removing the noise points, the current 3D model data can more accurately represent objects in the actual space.

[0071] S120: Perform coordinate transformation on the current three-dimensional denoised data to obtain the current three-dimensional transformed data;

[0072] Drones typically use geographic coordinate systems such as WGS84 or UTM. To facilitate subsequent registration work, these coordinate systems are converted to Cartesian coordinates for easier geometric calculations.

[0073] S130: Based on the CSF algorithm, remove terrain points and ground vegetation points from the current 3D transformed data to obtain the current 3D ground data;

[0074] A cloth grid is set on top of the current 3D conversion data. The cloth grid includes several nodes and grid lines for connecting adjacent nodes. By simulating gravity, a node in the cloth grid is made to contact the top point of the current 3D conversion data. After the node is held up, it will no longer fall under the action of gravity. If this top point is the ground, there are many points of the same height nearby, which form a large area of ​​support for the cloth grid. If this top point is grass or a building, the neighboring nodes of the held-up node will still fall under the action of gravity and will be pulled downward by the grid lines. After several iterations, the cloth grid forms a smooth surface. Points in the current 3D conversion data whose height is lower than the smooth surface, whose height is higher than the smooth surface, and whose height difference is less than the height threshold are all removed to remove terrain points and ground vegetation points.

[0075] S20: Perform spatial calibration on the current three-dimensional above-ground data and the previous three-dimensional above-ground data to obtain a difference three-dimensional model;

[0076] Understandably, the acquisition methods for the previous three-dimensional aboveground data and the current three-dimensional aboveground data are the same, only the acquisition time differs, and will not be elaborated further here. Step S20 includes:

[0077] S210: Obtain the current descriptor of the current data point in the current three-dimensional aboveground data, and obtain the past descriptor of the past data point in the past three-dimensional aboveground data;

[0078] A current data point is selected from the current 3D ground data. Using this point as the center, a search is conducted within a spherical region of radius r to find nearby points. The angular feature values ​​between the current data point and these nearby points are obtained to form a feature histogram. Based on the feature histograms of the current data point and the nearby points, a current descriptor is obtained. This current descriptor describes the morphological features of the current data point. The method for obtaining previous period descriptors is the same as that for obtaining the current descriptor, and will not be elaborated here.

[0079] S220: Compare the current period descriptor and the previous period descriptor to match the current period data points and the previous period data points as registration point pairs, and obtain the registration transformation matrix. Based on the registration transformation matrix, map the current period three-dimensional ground data to the previous period three-dimensional ground data, and obtain the overlapping ground data.

[0080] The current and past data points corresponding to current and past descriptors with similarity higher than a similarity threshold are selected as initial point pairs. An initial transformation matrix corresponding to each initial point pair is obtained. This initial transformation matrix is ​​applied to all initial point pairs to determine if the initial transformation matrix meets the standard. Understandably, based on the initial transformation matrix, the positions of the current data points in all initial point pairs are changed, and the transformation distance between the changed current data points and their corresponding past data points is obtained. The number of transformation distances greater than a transformation threshold determines whether the initial transformation matrix meets the standard. If it meets the standard, the initial point pair corresponding to the initial transformation matrix is ​​selected as a registration point pair, and the initial transformation matrix is ​​selected as the registration transformation matrix.

[0081] S230: Set core points in the overlapping ground data, construct selection cylinders based on the core points, and select current point sets from the current three-dimensional ground data and select past point sets from the past three-dimensional ground data using the selection cylinders;

[0082] There are multiple core points, which can be directly identified in the overlapping ground data or selected directly from the data points in the previous 3D ground data. In this embodiment, the core points are directly identified in the overlapping ground data. After the core points are obtained, a projection scale (usually the radius and height of a cylinder) is set to form the selected cylinder. Data points within the range of the selected cylinder are combined into the current point set, and the same applies to the previous point set.

[0083] S240: Based on the previous point set, obtain the normal vector corresponding to the core point, determine whether the core point is a change point based on the normal vector, and construct a difference 3D model based on several change points;

[0084] Specifically, the three-dimensional covariance matrix of the previous point set is obtained, and eigenvalue decomposition is performed on the three-dimensional covariance matrix to obtain the normal vector corresponding to the core point. The formula for obtaining the three-dimensional covariance matrix is ​​as follows:

[0085] ,

[0086] in, This represents the three-dimensional covariance matrix of the set of past points corresponding to the m-th core point. This represents the total number of past point sets corresponding to the m-th core point. This represents the j-th point in the set of past points corresponding to the m-th core point. Indicates the transpose operator;

[0087] Furthermore, after obtaining the three-dimensional covariance matrix, eigenvalue decomposition is performed on the three-dimensional covariance matrix based on the following formula:

[0088] ,

[0089] in, Represents the eigenvector. Represents the eigenvalue.

[0090] The purpose of this formula is to find a special eigenvector that, when the three-dimensional covariance matrix is ​​applied to it, does not change its direction but only its length. For the three-dimensional covariance matrix, which is a 3*3 matrix, geometrically it can have at most three eigenvectors. After rearranging the formula, a cubic polynomial about the eigenvalues ​​can be formed, thus obtaining three eigenvalues. Substituting these three eigenvalues ​​back into the above formula, we can obtain three corresponding eigenvectors. The eigenvector corresponding to the smallest eigenvalue is selected as the normal vector of the core point.

[0091] After obtaining the normal vector of the core point, the first average projection position of the past point set on the direction of the normal vector is obtained, and the second average projection position of the current point set on the direction of the normal vector is obtained.

[0092] The formula for obtaining the first average projection position is:

[0093] ,

[0094] in, Let represent the first average projection position of the set of past points corresponding to the m-th core point along the normal vector direction of the m-th core point. This represents the m-th core point. Let m represent the normal vector of the m-th core point. The formula for obtaining the second average projection position is the same as the formula for obtaining the first average projection position, and will not be repeated here.

[0095] The normal distance is obtained based on the first average projection position and the second average projection position;

[0096] The normal distance is proportional to a distance threshold. If the normal distance is greater than the distance threshold, the core point is determined to be a point of change.

[0097] The normal distance is obtained based on the difference between the first average projection position and the second average projection position. The larger the value, the more significant the change at the core point. After identifying all the core points using the above steps, all the selected change points are combined into the differential 3D model. It should be noted that the normal distance can be positive or negative. When the first average projection position is greater than the second average projection position, it indicates a decrease in height, and the change point represents a cavity. Conversely, when the first average projection position is greater than the second average projection position, it indicates an increase in height, and the change point represents an entity with different attribute distinctions.

[0098] S30: Dilate the differential 3D model to obtain a differential stereo model, and use an AI detection model to identify objects in the differential stereo model.

[0099] Step S30 includes:

[0100] S310: Construct an initial mesh model based on the minimum bounding rectangle of the differential 3D model. The initial mesh model includes several initial meshes. Map the change points in the differential 3D model to the initial mesh model to distinguish the several initial meshes into initial voxel meshes and background meshes.

[0101] S320: Determine whether the initial voxel grid exists in the adjacent grids of the background grid. If the initial voxel grid exists in the adjacent grids of the background grid, then select the background grid as the dilated voxel grid to complete a single dilation.

[0102] S330: Perform several single expansions to combine the initial voxel mesh and the expanded voxel mesh into a differential solid model;

[0103] The number of single expansion cycles can be set as needed. When performing a single expansion cycle again, it is necessary to determine whether the initial voxel mesh or the expanded voxel mesh exists within the adjacent meshes of the background mesh. If the initial voxel mesh or the expanded voxel mesh exists, the background mesh is selected as the expanded voxel mesh. After completing several single expansion cycles, the background mesh is ignored, thus forming the differential 3D model. It should be noted that the 3D objects in the differential 3D model can be simultaneously formed based on the connectivity of the voxel meshes, and the volume, centroid, surface area, length, width, and height of the 3D objects can be simultaneously acquired.

[0104] S340: Take images of the differential stereo model from different shooting angles to obtain several model snapshots;

[0105] The differential stereo model is loaded into the rendering engine, and several virtual cameras are set around the differential stereo model. In this embodiment, virtual cameras are set at the top, bottom, front, back, left, and right. By setting fixed lighting and background color, it is ensured that the model snapshot generated later only highlights the shape and texture of the object. Then, the model snapshot is obtained through the virtual cameras. In this embodiment, the differential stereo model includes several stereo voxel mesh groups, and the AI ​​detection model includes a feature extraction module and a classification probability module. It can be understood that the model snapshot contains stereo voxel mesh groups under the corresponding viewpoint, and for the stereo voxel mesh groups, they exist in different model snapshots with different visual perspectives.

[0106] S350: Input the model snapshot into the feature extraction module to generate a number of candidate regions in the model snapshot, the candidate regions corresponding to the three-dimensional voxel mesh group;

[0107] When the model snapshot is input into the feature extraction module, its abstract features (edges, textures, parts, shapes) are extracted through a backbone CNN. On the extracted feature map, a window is slid through RPN to generate multiple anchor boxes of different sizes and proportions at each position. If an object exists within the width of the identification cap, then the candidate region is formed.

[0108] S360: Input the model snapshot of the generated candidate region into the classification probability module to determine whether the 3D voxel mesh group is a building;

[0109] When the model snapshot is input into the classification probability module, it is converted into a fixed-size feature map through the Roll Pooling operation. After the feature map is input into the fully connected layer, the recognition probability of whether the feature map is a building is output. If the recognition probability is greater than the probability threshold, the feature region is determined to be a building. The above operation is performed for each candidate region in each model snapshot. It can be understood that since the same 3D voxel mesh group may have candidate regions in different model snapshots, if any candidate region in the different candidate regions corresponding to the same 3D voxel mesh group is identified as a building, the corresponding 3D voxel mesh group is determined to be a building. Accordingly, combined with the description in step S240, it can be determined whether the building is a newly added building or a disappeared building.

[0110] By preprocessing the current 3D model data, the influence of noise points is removed, and interference points such as the ground and ground vegetation are eliminated, leaving only points corresponding to above-ground structures. This achieves clean modeling to eliminate ground interference, ensuring subsequent detection efficiency and accuracy. By spatially calibrating the data from two periods, the regions where changes occur between the two periods are constructed into the differential 3D model. For the identification of changed objects, compared to traditional sparse point sets, the differential 3D model accurately represents the actual shape of the changed region, clarifies the subject of change, and further improves detection accuracy. By combining with the AI ​​detection model, recognition results for changed regions are automatically and quickly generated, and semantic-level judgments such as additions or disappearances can be made.

[0111] Please see Figure 2 The second embodiment of the present invention provides a three-dimensional temporal AI monitoring system for urban spatial governance. This system is applied to the three-dimensional temporal AI monitoring method for urban spatial governance described in the above embodiments, and will not be repeated hereafter. As used below, the terms "module," "unit," "subunit," etc., can refer to a combination of software and / or hardware that performs a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0112] The system includes:

[0113] The acquisition module 10 is used to acquire the current three-dimensional model data, preprocess the current three-dimensional model data to remove ground interference data, and acquire the current three-dimensional ground data.

[0114] The acquisition module 10 includes:

[0115] The first unit is used to select noise points from a plurality of data points and remove the noise points in order to update the current three-dimensional model data to the current three-dimensional denoised data.

[0116] The first unit is specifically used to select one of the data points as a reference point and the remaining data points as reference points, obtain the Euclidean distance between the reference point and the reference point, and select several neighboring points from several reference points based on the Euclidean distance.

[0117] The mean distance is obtained based on the Euclidean distance between the reference point and the neighboring points, and a distance threshold is constructed based on the mean distance.

[0118] The mean distance is compared with the distance threshold. If the mean distance is greater than the distance threshold, the reference point is determined to be a noise point.

[0119] The second unit is used to perform coordinate transformation on the current three-dimensional denoised data to obtain the current three-dimensional transformed data.

[0120] The third unit is used to remove terrain points and ground vegetation points from the current three-dimensional transformation data based on the CSF algorithm in order to obtain the current three-dimensional ground data.

[0121] The comparison module 20 is used to perform spatial calibration on the current three-dimensional aboveground data and the previous three-dimensional aboveground data to obtain a difference three-dimensional model;

[0122] The comparison module 20 includes:

[0123] The fourth unit is used to obtain the current descriptor of the current data point in the current three-dimensional aboveground data, and to obtain the past descriptor of the past data point in the past three-dimensional aboveground data;

[0124] The fifth unit is used to compare the current descriptor and the previous descriptor to match the current data points and the previous data points as registration point pairs, and obtain the registration transformation matrix. Based on the registration transformation matrix, the current three-dimensional ground data is mapped to the previous three-dimensional ground data, and the overlapping ground data is obtained.

[0125] The sixth unit is used to set core points in the overlapping ground data, construct a selection cylinder based on the core points, and select the current point set from the current three-dimensional ground data and the previous point set from the previous three-dimensional ground data through the selection cylinder.

[0126] The seventh unit is used to obtain the normal vector corresponding to the core point based on the previous point set, determine whether the core point is a change point based on the normal vector, and construct a difference three-dimensional model based on several change points.

[0127] The seventh unit is specifically used to obtain the three-dimensional covariance matrix of the past point set, and to perform eigenvalue decomposition on the three-dimensional covariance matrix to obtain the normal vector corresponding to the core point.

[0128] The seventh unit is also used to obtain the first average projection position of the past point set on the normal vector direction, and to obtain the second average projection position of the current point set on the normal vector direction;

[0129] The normal distance is obtained based on the first average projection position and the second average projection position;

[0130] The normal distance is proportional to a distance threshold. If the normal distance is greater than the distance threshold, the core point is determined to be a point of change.

[0131] Execution module 30 is used to perform dilation processing on the differential 3D model to obtain a differential stereo model, and to identify changed objects in the differential stereo model through an AI detection model.

[0132] The execution module 30 includes:

[0133] The eighth unit is used to construct an initial mesh model based on the minimum bounding rectangle of the differential 3D model. The initial mesh model includes several initial meshes. The change points in the differential 3D model are mapped to the initial mesh model to distinguish the several initial meshes into initial voxel meshes and background meshes.

[0134] The ninth unit is used to determine whether the initial voxel grid exists in the adjacent grids of the background grid. If the initial voxel grid exists in the adjacent grids of the background grid, the background grid is selected as the dilated voxel grid to complete a single dilation.

[0135] The tenth unit is used to perform several single expansions to combine the initial voxel mesh and the expanded voxel mesh into a differential solid model.

[0136] The eleventh unit is used to capture images of the differential stereo model from different shooting angles to obtain several model snapshots.

[0137] The twelfth unit is used to input the model snapshot into the feature extraction module to generate several candidate regions in the model snapshot, and the candidate regions correspond to the three-dimensional voxel mesh group;

[0138] The thirteenth unit is used to input the model snapshot of the generated candidate region into the classification probability module to determine whether the 3D voxel mesh group is a building.

[0139] The present invention also provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the three-dimensional temporal AI monitoring method for urban spatial governance as described in the above technical solutions.

[0140] The present invention also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the three-dimensional temporal AI monitoring method for urban spatial governance as described in the above technical solution.

[0141] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0142] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A three-dimensional temporal AI monitoring method for urban spatial governance, characterized in that, Includes the following steps: Acquire the current three-dimensional model data, preprocess the current three-dimensional model data to remove ground interference data, and acquire the current three-dimensional ground data; Spatial calibration is performed on the current and previous three-dimensional above-ground data to obtain a difference three-dimensional model; The step of spatially calibrating the current 3D terrestrial data and previous 3D terrestrial data to obtain a difference 3D model includes: Obtain the current descriptor of the current data point in the current three-dimensional aboveground data, and obtain the past descriptor of the past data point in the past three-dimensional aboveground data; The current period descriptor and the previous period descriptor are compared to match the current period data points and the previous period data points as registration point pairs, and the registration transformation matrix is ​​obtained. Based on the registration transformation matrix, the current period three-dimensional ground data is mapped to the previous period three-dimensional ground data, and the overlapping ground data is obtained. Core points are set in the overlapping ground data, and selection cylinders are constructed based on the core points. The selection cylinders are used to select the current point set from the current three-dimensional ground data and to select the past point set from the past three-dimensional ground data. Based on the previous point set, obtain the normal vector corresponding to the core point, determine whether the core point is a change point based on the normal vector, and construct a difference 3D model based on several change points. The step of obtaining the normal vector corresponding to the core point based on the historical point set is as follows: Obtain the three-dimensional covariance matrix of the previous point set, and perform eigenvalue decomposition on the three-dimensional covariance matrix to obtain the normal vector corresponding to the core point; The step of determining whether the core point is a point of change based on the normal vector includes: Obtain the first average projection position of the past point set on the normal vector direction, and obtain the second average projection position of the current point set on the normal vector direction; The normal distance is obtained based on the first average projection position and the second average projection position; The normal distance is compared with a distance threshold. If the normal distance is greater than the distance threshold, the core point is determined to be a point of change. The differential 3D model is dilated to obtain a differential stereo model, which includes several groups of stereo voxel meshes. An AI detection model is used to identify changing objects in the differential stereo model, which includes a feature extraction module and a classification probability module. The step of identifying changed objects in the differential 3D model using an AI detection model includes: The differential stereo model is photographed from different shooting angles to obtain several model snapshots; The model snapshot is input into the feature extraction module to generate several candidate regions in the model snapshot, and the candidate regions correspond to the three-dimensional voxel mesh group; The model snapshot of the generated candidate region is input into the classification probability module to determine whether the 3D voxel mesh group is a building.

2. The three-dimensional temporal AI monitoring method for urban spatial governance according to claim 1, characterized in that, The current 3D model data includes several data points. The step of preprocessing the current 3D model data to remove ground interference data and obtain the current 3D ground data includes: Noise points are selected from a plurality of data points and removed to update the current three-dimensional model data to the current three-dimensional denoised data. Perform coordinate transformation on the current 3D denoised data to obtain the current 3D transformed data; The CSF algorithm is used to remove terrain points and ground vegetation points from the current 3D transformed data to obtain the current 3D ground data.

3. The three-dimensional temporal AI monitoring method for urban spatial governance according to claim 2, characterized in that, The step of selecting noise points from the plurality of data points includes: One of the data points is selected as the reference point, and the remaining data points are selected as reference points. The Euclidean distance between the reference point and the reference point is obtained. Based on the Euclidean distance, several neighboring points are selected from several reference points. The mean distance is obtained based on the Euclidean distance between the reference point and the neighboring points, and a distance threshold is constructed based on the mean distance. The mean distance is compared with the distance threshold. If the mean distance is greater than the distance threshold, the reference point is determined to be a noise point.

4. The three-dimensional temporal AI monitoring method for urban spatial governance according to claim 3, characterized in that, The formula for obtaining the distance threshold is: ， in, Indicates the distance threshold. The mean of the distances from the mean. Represents the threshold multiplier. This represents the mean distance when the i-th data point is the reference point. This represents the total number of data points, where, .

5. The three-dimensional temporal AI monitoring method for urban spatial governance according to claim 1, characterized in that, The formula for obtaining the three-dimensional covariance matrix is: ， in, This represents the three-dimensional covariance matrix of the set of past points corresponding to the m-th core point. This represents the total number of past point sets corresponding to the m-th core point. This represents the j-th point in the set of past points corresponding to the m-th core point. Indicates the transpose operator; The formula for obtaining the first average projection position is: ， in, This represents the first average projection position of the set of past points corresponding to the m-th core point along the normal vector direction of the m-th core point. This represents the m-th core point. Let m represent the normal vector of the m-th core point.

6. The three-dimensional temporal AI monitoring method for urban spatial governance according to claim 1, characterized in that, The step of performing dilation processing on the differential 3D model to obtain a differential stereo model includes: An initial mesh model is constructed based on the minimum bounding rectangle of the differential 3D model. The initial mesh model includes several initial meshes. The change points in the differential 3D model are mapped into the initial mesh model to distinguish the several initial meshes into initial voxel meshes and background meshes. Determine whether the initial voxel grid exists in the adjacent grids of the background grid. If the initial voxel grid exists in the adjacent grids of the background grid, then select the background grid as the dilated voxel grid to complete a single dilation. Perform several single dilations to combine the initial voxel mesh and the dilated voxel mesh into a differential stereo model.

7. A three-dimensional temporal AI monitoring system for urban spatial governance, applied to the three-dimensional temporal AI monitoring method for urban spatial governance as described in any one of claims 1 to 6, characterized in that, The system includes: The acquisition module is used to acquire the current three-dimensional model data, preprocess the current three-dimensional model data to remove ground interference data, and acquire the current three-dimensional ground data. The comparison module is used to perform spatial calibration on the current three-dimensional aboveground data and the previous three-dimensional aboveground data to obtain the difference three-dimensional model; The execution module is used to perform dilation processing on the differential 3D model to obtain a differential stereo model, and to identify the changed objects in the differential stereo model through an AI detection model.

Images