A spatial intelligent three-dimensional reconstruction method and system based on cross-scale topological consistency

Abstract

The application discloses a kind of spatial intelligence three-dimensional reconstruction method and system based on cross-scale topological consistency, it is related to spatial intelligence field, solve the problem that existing spatial intelligence three-dimensional reconstruction technology exists three-dimensional reconstruction effect is not good, including steps S1: acquisition area plan image is carried out to modeling region space marking, obtain modeling region plan image, to target modeling region preliminary creation three-dimensional model, obtain target region preliminary three-dimensional model, step S2: target region preliminary three-dimensional model is verified with target modeling region scene geometry topological, according to the topological element repair of target region preliminary three-dimensional model according to verification result, obtain target region repair three-dimensional model, step S3: according to the real-time color data of target modeling region corresponding to target region preliminary three-dimensional model is carried out real-time color restoration, obtain spatial region three-dimensional reconstruction model, the application improves the structural integrity and geometric accuracy of three-dimensional model under complex scene.

Description

Technical Field

[0001] This invention belongs to the field of spatial intelligence and relates to three-dimensional reconstruction technology, specifically a spatial intelligent three-dimensional reconstruction method and system based on cross-scale topological consistency. Background Technology

[0002] Spatial intelligent 3D reconstruction technology has the following drawbacks when performing 3D reconstruction of spatial regions:

[0003] 1. Existing 3D reconstruction technologies are difficult to effectively maintain multi-scale consistency and the continuity of topological structure. In complex spatial scenes, geometric features at different scales often have significant differences. Existing algorithms are prone to geometric distortion or topological breaks when fusing across scales, resulting in local distortion or structural loss in the reconstructed model.

[0004] 2. Existing 3D reconstruction technologies have limited ability to enhance the geometric details of complex scenes and struggle to guarantee semantic consistency. On the one hand, due to limitations in sensor resolution or algorithm robustness, reconstructed models often lose minute geometric features, resulting in insufficient detail representation.

[0005] To this end, we propose a spatial intelligent 3D reconstruction method and system based on cross-scale topological consistency. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the present invention aims to provide a spatial intelligent 3D reconstruction method and system based on cross-scale topological consistency. The present invention aims to improve the geometric accuracy and detail representation of the spatial intelligent 3D reconstruction process.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a spatial intelligent 3D reconstruction method based on cross-scale topological consistency, the specific steps of which are as follows:

[0008] Step S1: Acquire a planar image of the area to spatially mark the modeling area, obtain a planar image of the modeling area, and perform preliminary 3D model creation on the target modeling area to obtain a preliminary 3D model of the target area;

[0009] Step S2: Perform scene geometry and topology verification between the preliminary 3D model of the target area and the target modeling area. Based on the verification results, perform topology element repair on the preliminary 3D model of the target area to obtain the repaired 3D model of the target area.

[0010] Step S3: Perform real-time color restoration of the preliminary 3D model of the target area based on the real-time color data corresponding to the target modeling area to obtain the 3D reconstruction model of the spatial area.

[0011] Furthermore, in step S1, the specific steps are as follows:

[0012] Step S11: Acquire planar images of the spatial region that needs to be modeled in three dimensions to obtain a planar image of the modeling region. Divide the planar image of the modeling region into several spatial modeling regions and arbitrarily select a target modeling region from the acquired spatial modeling regions.

[0013] Step S12: Use a lidar to perform a laser scan on the target modeling area, obtain multiple laser sampling points based on the scan results, and obtain the laser emission positions of the lidar in the target modeling area to obtain multiple laser emission positions in the area.

[0014] Step S13: Randomly select a sample laser emission position from the multiple acquired regional laser emission positions, collect the laser beam emitted by the sample laser emission position to obtain the sample emitted laser beam, obtain the laser sampling points in the different sample emitted laser beam travel paths, and obtain the straight-line distance value between each laser sampling point and the corresponding sample laser emission position, compare the obtained straight-line distance values, and set the laser sampling point corresponding to the maximum straight-line distance value as the edge point of the travel path corresponding to the sample emitted laser beam.

[0015] Furthermore, in step S1, the specific steps are as follows:

[0016] Step S14: Obtain the travel path edge points corresponding to the laser emission position of each region, and obtain multiple travel path edge points. Connect the travel path edge points that are in an adjacent state to obtain multiple adjacent edge point lines. Fill the closed polygon enclosed by the adjacent edge point lines with pixels to obtain the initial three-dimensional contour model of the target region.

[0017] Step S15: Divide the spatial region covered by the initial three-dimensional contour model of the target region into several voxel spaces to obtain multiple region segmentation voxel spaces;

[0018] Step S16: Randomly select a sample voxel space from the multiple region segmentation voxel spaces obtained, and perform noise point screening on the laser sampling points covered in the sample voxel space. Obtain the sample voxel processing space based on the screening results.

[0019] Step S17: Perform noise point screening on each region segmented voxel space, and set the initial three-dimensional contour model of the target region after completing the noise point screening of the region segmented voxel space as the preliminary three-dimensional model of the target region.

[0020] Furthermore, in step S16, the specific steps are as follows:

[0021] The laser sampling points existing in the sample voxel space are acquired, and a feature sampling point is arbitrarily selected from the acquired laser sampling points. The feature sampling point is used as the ray starting point to draw several spatial rays at different angles. The distance value between each laser sampling point in the spatial ray travel path and the ray starting point is obtained, and the obtained distance values ​​are compared. The laser sampling point with the smallest value is set as the nearest neighbor sampling point. The nearest neighbor sampling points in the adjacent state are connected to obtain multiple adjacent sampling point lines. The closed polygon enclosed by the adjacent sampling point lines is filled with pixels to obtain the sample nearest neighbor pixel volume.

[0022] The volume of the nearest neighbor pixels of the sample is collected to obtain the volume value of the adjacent sampling space. The number of laser sampling points contained in the nearest neighbor pixels of the sample is counted to obtain the number of adjacent sampling points. The ratio of the number of adjacent sampling points to the volume value of the adjacent sampling space is calculated to obtain the sampling point density of the neighborhood space corresponding to the feature sampling point.

[0023] Furthermore, in step S16, the specific steps are as follows:

[0024] Create a preset range for the sampling density of the neighborhood space. If the density of the sampling points in the neighborhood space is within the preset range, the feature sampling points are classified as sampling points with reasonable density. If the density of the sampling points in the neighborhood space is not within the preset range, the feature sampling points are classified as abnormal sampling points.

[0025] Distance values ​​are collected for the nearest neighbor sampling points corresponding to reasonable density sampling points, resulting in multiple nearest neighbor sampling distance values. These values ​​are then compared, and the nearest neighbor sampling point corresponding to the largest nearest neighbor sampling distance value is marked as the peak distance sampling point. The distance values ​​between the peak distance sampling point and its corresponding nearest neighbor sampling point are obtained, resulting in multiple nearest neighbor sampling distance values. The standard deviation of these multiple nearest neighbor sampling distance values ​​is calculated, and a preset value for the nearest neighbor sampling standard deviation is set. If the standard deviation of the nearest neighbor sampling distance is less than the preset value, the reasonable density sampling point is still considered a reasonable density sampling point. If the standard deviation of the nearest neighbor sampling distance is greater than or equal to the preset value, the reasonable density sampling point is still considered an abnormal sampling point.

[0026] Within the sample voxel space, abnormal sampling points are removed to obtain the sample voxel processing space.

[0027] Furthermore, in step S2, the specific steps are as follows:

[0028] Step S21: Obtain a preliminary 3D model of the target area, and collect the spatial geometric center point of the preliminary 3D model of the target area to obtain the first geometric center point. Mark the actual scene position corresponding to the first geometric center point in the target modeling area to obtain the second geometric center point.

[0029] Step S22: In the preliminary 3D model of the target area, create a spatial coordinate system with the first geometric center point as the origin to obtain the model spatial coordinate system. Map the model spatial coordinate system onto the target modeling area and keep the origin of the model spatial coordinate system coincide with the second geometric center point to obtain the scene spatial coordinate system.

[0030] Step S23: Obtain the topological elements appearing in the target modeling region to obtain multiple region topological elements, and arbitrarily select a sample topological element from the multiple obtained region topological elements, and perform element repair on the sample topological element.

[0031] Step S24: Perform element repair on each region's topology element to obtain a repaired 3D model of the target region.

[0032] Furthermore, in step S23, the specific steps are as follows:

[0033] In the target modeling region, the spatial region covered by the sample topology element is marked to obtain the sample element spatial region, and the sample element spatial region is discretized into several regional element points to obtain the regional element point set.

[0034] In the preliminary 3D model of the target area, the model area covered by the sample topology element is marked using the model space analysis algorithm to obtain the sample element model area. The laser sampling points covered by the sample element model area are marked as valid topology points, and the laser sampling points not covered by the sample element model area are marked as invalid topology points.

[0035] Furthermore, in step S23, the specific steps are as follows:

[0036] For each region element point in the set of region element points, acquire the sample element point from the acquired multiple region element points, collect the three-dimensional coordinates of the sample element point in the scene space coordinate system, obtain the three-dimensional coordinates of the sample element, and acquire the laser sampling point corresponding to the three-dimensional coordinates of the sample element in the model space coordinate system. If the corresponding laser sampling point is a valid topological point, retain the laser sampling point; if the corresponding laser sampling point is an invalid topological point, discard the laser sampling point.

[0037] Furthermore, in step S3, the specific steps are as follows:

[0038] Each region topology element in the target modeling region is labeled to obtain multiple element-labeled spatial regions;

[0039] The element marking spatial region is discretized into several spatial element points. The three-dimensional coordinates corresponding to each spatial element point are collected according to the scene spatial coordinate system to obtain the three-dimensional coordinates of the real scene element. The real-time color score corresponding to each spatial element point is obtained using the RGB color model to obtain the color score of the real scene element.

[0040] In the three-dimensional reconstruction model of the spatial region, the model space points corresponding to the three-dimensional coordinates of each real scene element are obtained through the three-dimensional coordinate system of the model space, and the color scores of the real scene elements are used to restore the color of the model space points.

[0041] Color restoration is performed on the spatial region marked by each element to obtain a three-dimensional reconstruction model of the spatial region.

[0042] A spatial intelligent 3D reconstruction system based on cross-scale topological consistency includes:

[0043] Model creation module: Acquire planar images of the region, spatially mark the modeling region to obtain a planar image of the modeling region, and perform preliminary 3D model creation on the target modeling region to obtain a preliminary 3D model of the target region;

[0044] Topology Repair Module: Performs scene geometry topology verification between the preliminary 3D model of the target region and the target modeling region. Based on the verification results, it repairs the topology elements of the preliminary 3D model of the target region to obtain the repaired 3D model of the target region.

[0045] Color restoration module: Based on the real-time color data corresponding to the target modeling area, the initial 3D model of the target area is restored in real time to obtain the 3D reconstruction model of the spatial area.

[0046] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0047] 1. This invention verifies the scene geometry and topology of the preliminary 3D model against the actual modeling area, accurately identifying topological errors in the model and automatically repairing topological elements based on the verification results. This significantly improves the structural integrity of 3D models in complex scenes;

[0048] 2. This invention uses real-time color data collected from the target area to dynamically color the preliminary 3D model, which can achieve accurate mapping of lighting, material and environmental colors. Compared with the traditional static texture mapping method, this technology can simultaneously reflect the lighting changes and material characteristics in the scene, so that the reconstructed model presents a higher visual realism in the virtual scene. Attached Figure Description

[0049] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0050] Figure 1 This is a diagram illustrating the implementation steps of the present invention;

[0051] Figure 2 This is an overall system block diagram of the present invention;

[0052] Figure 3 This is a schematic diagram of the sample nearest neighbor pixels of the present invention. Detailed Implementation

[0053] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Example 1

[0055] Please see Figure 1 This invention provides a technical solution: a spatial intelligent 3D reconstruction method based on cross-scale topological consistency, the specific steps of which are as follows:

[0056] Step S1: Acquire a planar image of the area to spatially mark the modeling area, obtain a planar image of the modeling area, and perform preliminary 3D model creation on the target modeling area to obtain a preliminary 3D model of the target area;

[0057] In step S1, the specific steps are as follows:

[0058] Step S11: Acquire planar images of the spatial region that needs to be modeled in three dimensions to obtain a planar image of the modeling region. Divide the planar image of the modeling region into several spatial modeling regions and arbitrarily select a target modeling region from the acquired spatial modeling regions.

[0059] Step S12: Use a lidar to perform a laser scan on the target modeling area, obtain multiple laser sampling points based on the scan results, and obtain the laser emission positions of the lidar in the target modeling area to obtain multiple laser emission positions in the area.

[0060] Step S13: Randomly select a sample laser emission position from the multiple acquired regional laser emission positions, collect the laser beam emitted through the sample laser emission position to obtain the sample emitted laser beam, obtain the laser sampling points in the different sample emitted laser beam travel paths, and obtain the straight-line distance value between each laser sampling point and the corresponding sample laser emission position, and compare the obtained straight-line distance values, and set the laser sampling point corresponding to the maximum straight-line distance value as the edge point of the travel path corresponding to the sample emitted laser beam;

[0061] Step S14: Obtain the travel path edge points corresponding to the laser emission position of each region, and obtain multiple travel path edge points. Connect the travel path edge points that are in an adjacent state to obtain multiple adjacent edge point lines. Fill the closed polygon enclosed by the adjacent edge point lines with pixels to obtain the initial three-dimensional contour model of the target region.

[0062] Step S15: Divide the spatial region covered by the initial three-dimensional contour model of the target region into several voxel spaces to obtain multiple region segmentation voxel spaces;

[0063] Step S16: Randomly select a sample voxel space from the multiple region segmentation voxel spaces obtained, and perform noise point screening on the laser sampling points covered in the sample voxel space. Obtain the sample voxel processing space based on the screening results.

[0064] In step S16, the specific steps are as follows:

[0065] The laser sampling points existing in the sample voxel space are acquired, and a feature sampling point is arbitrarily selected from the acquired laser sampling points. The feature sampling point is used as the ray starting point to draw several spatial rays at different angles. The distance value between each laser sampling point in the spatial ray travel path and the ray starting point is obtained, and the obtained distance values ​​are compared. The laser sampling point with the smallest value is set as the nearest neighbor sampling point. The nearest neighbor sampling points in the adjacent state are connected to obtain multiple adjacent sampling point lines. The closed polygon enclosed by the adjacent sampling point lines is filled with pixels to obtain the sample nearest neighbor pixel volume.

[0066] The volume of the nearest neighbor pixels of the sample is collected to obtain the volume value of the adjacent sampling space. The number of laser sampling points contained in the nearest neighbor pixels of the sample is counted to obtain the number of adjacent sampling points. The ratio of the number of adjacent sampling points to the volume value of the adjacent sampling space is calculated to obtain the sampling point density of the neighborhood space corresponding to the feature sampling point.

[0067] Create a preset range for the sampling density of the neighborhood space. If the density of the sampling points in the neighborhood space is within the preset range, the feature sampling points are classified as sampling points with reasonable density. If the density of the sampling points in the neighborhood space is not within the preset range, the feature sampling points are classified as abnormal sampling points.

[0068] Distance values ​​are collected for the nearest neighbor sampling points corresponding to reasonable density sampling points, resulting in multiple nearest neighbor sampling distance values. These values ​​are then compared, and the nearest neighbor sampling point corresponding to the largest nearest neighbor sampling distance value is marked as the peak distance sampling point. The distance values ​​between the peak distance sampling point and its corresponding nearest neighbor sampling point are obtained, resulting in multiple nearest neighbor sampling distance values. The standard deviation of these multiple nearest neighbor sampling distance values ​​is calculated, and a preset value for the nearest neighbor sampling standard deviation is set. If the standard deviation of the nearest neighbor sampling distance is less than the preset value, the reasonable density sampling point is still considered a reasonable density sampling point. If the standard deviation of the nearest neighbor sampling distance is greater than or equal to the preset value, the reasonable density sampling point is still considered an abnormal sampling point.

[0069] Within the sample voxel space, outlier sampling points are removed to obtain the sample voxel processing space;

[0070] Step S17: Perform noise point screening on each region segmented voxel space, and set the initial three-dimensional contour model of the target region after completing the noise point screening of the region segmented voxel space as the preliminary three-dimensional model of the target region.

[0071] Step S2: Perform scene geometry and topology verification between the preliminary 3D model of the target area and the target modeling area. Based on the verification results, perform topology element repair on the preliminary 3D model of the target area to obtain the repaired 3D model of the target area.

[0072] In step S2, the specific steps are as follows:

[0073] Step S21: Obtain a preliminary 3D model of the target area, and collect the spatial geometric center point of the preliminary 3D model of the target area to obtain the first geometric center point. Mark the actual scene position corresponding to the first geometric center point in the target modeling area to obtain the second geometric center point.

[0074] Step S22: In the preliminary 3D model of the target area, create a spatial coordinate system with the first geometric center point as the origin to obtain the model spatial coordinate system. Map the model spatial coordinate system onto the target modeling area and keep the origin of the model spatial coordinate system coincide with the second geometric center point to obtain the scene spatial coordinate system.

[0075] Step S23: Obtain the topological elements appearing in the target modeling region to obtain multiple region topological elements, and arbitrarily select a sample topological element from the multiple obtained region topological elements, and perform element repair on the sample topological element.

[0076] In step S23, the specific steps are as follows:

[0077] In the target modeling region, the spatial region covered by the sample topology element is marked to obtain the sample element spatial region, and the sample element spatial region is discretized into several regional element points to obtain the regional element point set.

[0078] In the preliminary 3D model of the target area, the model area covered by the sample topology element is marked using the model space analysis algorithm to obtain the sample element model area. The laser sampling points covered by the sample element model area are marked as valid topology points, and the laser sampling points not covered by the sample element model area are marked as invalid topology points.

[0079] For each region element point in the set of region element points, acquire the sample element point from the acquired multiple region element points, collect the three-dimensional coordinates of the sample element point in the scene space coordinate system, obtain the three-dimensional coordinates of the sample element, and acquire the laser sampling point corresponding to the three-dimensional coordinates of the sample element in the model space coordinate system. If the corresponding laser sampling point is a valid topological point, retain the laser sampling point; if the corresponding laser sampling point is an invalid topological point, discard the laser sampling point.

[0080] Step S24: Perform element repair on each region's topology element to obtain a repaired 3D model of the target region;

[0081] Step S3: Perform real-time color restoration of the preliminary 3D model of the target area based on the real-time color data corresponding to the target modeling area to obtain the 3D reconstruction model of the spatial area;

[0082] The specific steps in step S3 are as follows:

[0083] Each region topology element in the target modeling region is labeled to obtain multiple element-labeled spatial regions;

[0084] The element marking spatial region is discretized into several spatial element points. The three-dimensional coordinates corresponding to each spatial element point are collected according to the scene spatial coordinate system to obtain the three-dimensional coordinates of the real scene element. The real-time color score corresponding to each spatial element point is obtained using the RGB color model to obtain the color score of the real scene element.

[0085] In the three-dimensional reconstruction model of the spatial region, the model space points corresponding to the three-dimensional coordinates of each real scene element are obtained through the three-dimensional coordinate system of the model space, and the color scores of the real scene elements are used to restore the color of the model space points.

[0086] Color restoration is performed on the spatial region marked by each element to obtain a three-dimensional reconstruction model of the spatial region.

[0087] Example 2

[0088] Please see Figure 2 Based on another concept of the same invention, a spatial intelligent 3D reconstruction system based on cross-scale topological consistency is proposed. The specific working process of each module is as follows:

[0089] The initial model creation module acquires planar images of the region, spatially marks the modeling region, obtains a planar image of the modeling region, and performs preliminary 3D model creation on the target modeling region, resulting in a preliminary 3D model of the target region.

[0090] Specifically as follows:

[0091] The planar image of the spatial region that needs to be three-dimensional is acquired to obtain the planar image of the modeling region. The planar image of the modeling region is divided into several spatial modeling regions, and a target modeling region is arbitrarily selected from the acquired spatial modeling regions.

[0092] It should be noted here that:

[0093] In this application, the planar image of the modeling area referred to herein is specifically a spatial region where a spatial intelligent 3D reconstruction system is deployed.

[0094] The target modeling area is scanned using a lidar, and multiple laser sampling points are obtained based on the scanning results. The laser emission positions of the lidar in the target modeling area are then obtained, resulting in multiple laser emission positions in the area.

[0095] It should be noted here that:

[0096] In this application, the laser sampling points referred to herein are specifically the data points that make up the point cloud;

[0097] In this application, the lidar referred to herein is specifically a mechanical lidar;

[0098] In this application, the laser emission position refers to the specific starting point coordinates in space when the lidar emits a laser beam.

[0099] Arbitrarily select a sample laser emission position from the multiple acquired regional laser emission positions, collect the laser beam emitted by the sample laser emission position to obtain the sample emitted laser beam, obtain laser sampling points in the different sample emitted laser beam travel paths, and obtain the straight-line distance value between each laser sampling point and the corresponding sample laser emission position. Compare the obtained straight-line distance values, and set the laser sampling point corresponding to the maximum straight-line distance value as the edge point of the travel path corresponding to the sample emitted laser beam.

[0100] The edge points of the travel path corresponding to the laser emission position in each region are obtained, resulting in multiple travel path edge points. The adjacent travel path edge points are connected to obtain multiple adjacent edge point lines. The closed polygon enclosed by the adjacent edge point lines is filled with pixels to obtain the initial three-dimensional contour model of the target region.

[0101] The spatial region covered by the initial three-dimensional contour model of the target region is divided into several voxel spaces, resulting in multiple region segmentation voxel spaces.

[0102] It should be noted here that:

[0103] In this application, the voxel space referred to herein is specifically a cube space.

[0104] Arbitrarily select a sample voxel space from the multiple region segmentation voxel spaces obtained, and perform noise point screening on the laser sampling points covered in the sample voxel space. Obtain the sample voxel processing space based on the screening results.

[0105] Specifically as follows:

[0106] Please see Figure 3 The laser sampling points existing in the sample voxel space are acquired, and a feature sampling point is randomly selected from the acquired laser sampling points. The feature sampling point is used as the ray starting point to draw several spatial rays at different angles. The distance value between each laser sampling point in the spatial ray travel path and the ray starting point is obtained. The obtained distance values ​​are compared, and the laser sampling point with the smallest value is set as the nearest neighbor sampling point. The nearest neighbor sampling points in the adjacent state are connected to obtain multiple adjacent sampling point lines. The closed polygon enclosed by the adjacent sampling point lines is filled with pixels to obtain the sample nearest neighbor pixel volume.

[0107] The volume of the nearest neighbor pixels of the sample is collected to obtain the volume value of the adjacent sampling space. The number of laser sampling points contained in the nearest neighbor pixels of the sample is counted to obtain the number of adjacent sampling points. The ratio of the number of adjacent sampling points to the volume value of the adjacent sampling space is calculated to obtain the sampling point density of the neighborhood space corresponding to the feature sampling point.

[0108] Create a preset range for the sampling density of the neighborhood space. If the density of the sampling points in the neighborhood space is within the preset range, the feature sampling points are classified as sampling points with reasonable density. If the density of the sampling points in the neighborhood space is not within the preset range, the feature sampling points are classified as abnormal sampling points.

[0109] It should be noted here that:

[0110] In this application, the density of neighboring spatial sampling points is collected for laser sampling points that have been historically divided into reasonable density sampling points. Multiple neighboring spatial sampling point densities are obtained, and the values ​​of the multiple neighboring spatial sampling point densities are compared. The neighboring spatial sampling point density with the largest value is marked as the upper limit of the preset interval of the neighboring spatial sampling density, and the neighboring spatial sampling point density with the smallest value is marked as the lower limit of the preset interval of the neighboring spatial sampling density, thus obtaining the preset interval of the neighboring spatial sampling density.

[0111] Distance values ​​are collected for the nearest neighbor sampling points corresponding to reasonable density sampling points, resulting in multiple nearest neighbor sampling distance values. These values ​​are then compared, and the nearest neighbor sampling point corresponding to the largest nearest neighbor sampling distance value is marked as the peak distance sampling point. The distance values ​​between the peak distance sampling point and its corresponding nearest neighbor sampling point are obtained, resulting in multiple nearest neighbor sampling distance values. The standard deviation of these multiple nearest neighbor sampling distance values ​​is calculated, and a preset value for the nearest neighbor sampling standard deviation is set. If the standard deviation of the nearest neighbor sampling distance is less than the preset value, the reasonable density sampling point is still considered a reasonable density sampling point. If the standard deviation of the nearest neighbor sampling distance is greater than or equal to the preset value, the reasonable density sampling point is still considered an abnormal sampling point.

[0112] It should be noted here that:

[0113] In this application, the standard deviation of the nearest neighbor sampling distance is collected for laser sampling points that have been historically divided into reasonable density sampling points, and multiple standard deviations of the nearest neighbor sampling distance are obtained. The values ​​of the multiple standard deviations of the nearest neighbor sampling distance are compared, and the standard deviation of the nearest neighbor sampling distance with the largest value is set as the preset value of the standard deviation of the nearest neighbor sampling distance.

[0114] Within the sample voxel space, outlier sampling points are removed to obtain the sample voxel processing space;

[0115] Repeat the process of acquiring the sample voxel processing space, perform noise point screening for each region segmentation voxel space, and set the initial three-dimensional contour model of the target region after completing the noise point screening of the region segmentation voxel space as the preliminary three-dimensional model of the target region.

[0116] The topology repair module performs scene geometry topology verification between the preliminary 3D model of the target region and the target modeling region. Based on the verification results, it repairs the topology elements of the preliminary 3D model of the target region to obtain the repaired 3D model of the target region.

[0117] Specifically as follows:

[0118] A preliminary 3D model of the target area is obtained, and the spatial geometric center point of the preliminary 3D model of the target area is collected to obtain the first geometric center point. The actual scene position corresponding to the first geometric center point is marked in the target modeling area to obtain the second geometric center point.

[0119] In the initial 3D model of the target area, a spatial coordinate system is created by taking the first geometric center point as the origin, and the model spatial coordinate system is obtained. The model spatial coordinate system is then mapped onto the target modeling area, and the origin of the model spatial coordinate system is kept to coincide with the second geometric center point, thus obtaining the scene spatial coordinate system.

[0120] It should be noted here that:

[0121] In this application, the coordinate axes in the model space coordinate system and the scene space coordinate system extend in the same direction.

[0122] The topological elements appearing in the target modeling region are acquired to obtain multiple region topological elements. Then, a sample topological element is randomly selected from the acquired multiple region topological elements, and the sample topological element is repaired.

[0123] It should be noted here that:

[0124] In this application, the topological elements referred to herein are specifically volume topological elements, and the regional topological elements referred to herein include, but are not limited to, walls, roofs, and seats.

[0125] In the target modeling region, the spatial region covered by the sample topology element is marked to obtain the sample element spatial region, and the sample element spatial region is discretized into several regional element points to obtain the regional element point set.

[0126] In the preliminary 3D model of the target area, the model area covered by the sample topology element is marked using the model space analysis algorithm to obtain the sample element model area. The laser sampling points covered by the sample element model area are marked as valid topology points, and the laser sampling points not covered by the sample element model area are marked as invalid topology points.

[0127] It should be noted here that:

[0128] In this application, the model space analysis algorithm specifically referred to herein is the MCluster-Hic algorithm.

[0129] For each region element point in the set of region element points, acquire the sample element point from the acquired multiple region element points, collect the three-dimensional coordinates of the sample element point in the scene space coordinate system, obtain the three-dimensional coordinates of the sample element, and acquire the laser sampling point corresponding to the three-dimensional coordinates of the sample element in the model space coordinate system. If the corresponding laser sampling point is a valid topological point, retain the laser sampling point; if the corresponding laser sampling point is an invalid topological point, discard the laser sampling point.

[0130] Repeat the process of element repair on the sample topology elements, and repair the topology elements of each region separately to obtain the repaired 3D model of the target region.

[0131] The color restoration module performs real-time color restoration on the preliminary 3D model of the target area based on the real-time color data corresponding to the target modeling area, and obtains the 3D reconstruction model of the spatial area.

[0132] Specifically as follows:

[0133] Each region topology element in the target modeling region is labeled to obtain multiple element-labeled spatial regions;

[0134] The element marking spatial region is discretized into several spatial element points. The three-dimensional coordinates corresponding to each spatial element point are collected according to the scene spatial coordinate system to obtain the three-dimensional coordinates of the real scene element. The real-time color score corresponding to each spatial element point is obtained using the RGB color model to obtain the color score of the real scene element.

[0135] It should be noted here that:

[0136] In this application, real-time color score generally refers to a numerical index that reflects the current color state of an object or image, obtained through real-time measurement or calculation.

[0137] In the three-dimensional reconstruction model of the spatial region, the model space points corresponding to the three-dimensional coordinates of each real scene element are obtained through the three-dimensional coordinate system of the model space, and the color scores of the real scene elements are used to restore the color of the model space points.

[0138] Color restoration is performed on the spatial region marked by each element to obtain a three-dimensional reconstruction model of the spatial region.

[0139] Compared to the problems described in the background art, this invention, by performing scene geometric topology verification between the preliminary 3D model and the actual modeling area, can accurately identify topological errors in the model and automatically repair topological elements based on the verification results. This mechanism significantly improves the structural integrity of 3D models in complex scenes;

[0140] Furthermore, this invention dynamically colors the preliminary 3D model based on color data collected in real time from the target area, which can achieve accurate mapping of lighting, materials and environmental colors. Compared with the traditional static mapping method, this technology can simultaneously reflect the lighting changes and material characteristics in the scene, making the reconstructed model present a higher visual realism in the virtual scene.

[0141] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A spatial intelligent 3D reconstruction method based on cross-scale topological consistency, characterized in that, include: Step S1: Acquire a planar image of the area to spatially mark the modeling area, obtain a planar image of the modeling area, and perform preliminary 3D model creation on the target modeling area to obtain a preliminary 3D model of the target area; In step S1, the specific steps are as follows: Step S11: Acquire planar images of the spatial region that needs to be reconstructed in three dimensions to obtain a planar image of the modeling region. Divide the planar image of the modeling region into several spatial modeling regions and arbitrarily select a target modeling region from the acquired spatial modeling regions. Step S12: Use a lidar to perform a laser scan on the target modeling area, obtain multiple laser sampling points based on the scan results, and obtain the laser emission positions of the lidar in the target modeling area to obtain multiple laser emission positions in the area. Step S13: Randomly select a sample laser emission position from the multiple acquired regional laser emission positions, collect the laser beam emitted by the sample laser emission position to obtain the sample emitted laser beam, obtain the laser sampling points in the different sample emitted laser beam travel paths, obtain the straight-line distance value between each laser sampling point and the corresponding sample laser emission position, and set the laser sampling point corresponding to the maximum straight-line distance value as the edge point of the travel path corresponding to the sample emitted laser beam. Step S14: Obtain the travel path edge points corresponding to the laser emission position of each region, and obtain multiple travel path edge points. Connect the travel path edge points that are in an adjacent state to obtain multiple adjacent edge point lines. Fill the closed polygon enclosed by the adjacent edge point lines with pixels to obtain the initial three-dimensional contour model of the target region. Step S15: Divide the spatial region covered by the initial three-dimensional contour model of the target region into several voxel spaces to obtain multiple region segmentation voxel spaces; Step S16: Randomly select a sample voxel space from the multiple region segmentation voxel spaces obtained, and perform noise point screening on the laser sampling points covered in the sample voxel space. Obtain the sample voxel processing space based on the screening results. Step S17: Perform noise point screening on the voxel space of each region, and set the initial three-dimensional contour model of the target region of the spatial noise point screening as the preliminary three-dimensional model of the target region. Step S2: Perform scene geometry and topology verification between the preliminary 3D model of the target area and the target modeling area, and repair the topology elements of the preliminary 3D model of the target area to obtain the repaired 3D model of the target area; In step S2, the specific steps are as follows: Step S21: Obtain a preliminary 3D model of the target area, collect the spatial geometric center point of the preliminary 3D model of the target area to obtain the first geometric center point, and mark the actual scene position corresponding to the first geometric center point in the target modeling area to obtain the second geometric center point; Step S22: In the preliminary 3D model of the target area, create a spatial coordinate system with the first geometric center point as the origin to obtain the model spatial coordinate system, and map the model spatial coordinate system onto the target modeling area, while keeping the origin of the model spatial coordinate system coincide with the second geometric center point to obtain the scene spatial coordinate system; Step S23: Obtain the topological elements appearing in the target modeling area to obtain multiple regional topological elements. Randomly select a sample topological element from the multiple obtained regional topological elements and perform element repair on the sample topological element. In step S23, the specific steps are as follows: In the target modeling region, the spatial region covered by the sample topology element is marked to obtain the sample element spatial region. The sample element spatial region is then discretized into several regional element points to obtain the set of regional element points. In the preliminary 3D model of the target area, the model area covered by the sample topology element is marked using the model space analysis algorithm to obtain the sample element model area. The laser sampling points covered by the sample element model area are marked as valid topology points, and the laser sampling points not covered by the sample element model area are marked as invalid topology points. Step S24: Perform element repair on each region's topology element to obtain a repaired 3D model of the target region; Step S3: Perform real-time color restoration of the preliminary 3D model of the target area based on the real-time color data corresponding to the target modeling area to obtain the 3D reconstruction model of the spatial area.

2. The spatial intelligent 3D reconstruction method based on cross-scale topological consistency according to claim 1, characterized in that, The specific steps in step S16 are as follows: Obtain laser sampling points in the sample voxel space. Arbitrarily select a feature sampling point and use the feature sampling point as the ray starting point to draw several spatial rays at different angles. Obtain the distance value between each laser sampling point in the spatial ray travel path and the ray starting point. Set the laser sampling point with the smallest value as the nearest neighbor sampling point. Connect the nearest neighbor sampling points in the adjacent state to obtain multiple adjacent sampling point lines. Fill the closed polygon enclosed by the adjacent sampling point lines with pixels to obtain the sample nearest neighbor pixel volume. The volume of the nearest neighbor pixels of the sample is collected to obtain the volume value of the adjacent sampling space. The number of laser sampling points contained in the nearest neighbor pixels of the sample is counted to obtain the number of adjacent sampling points. The ratio of the number of adjacent sampling points to the volume value of the adjacent sampling space is calculated. The density of sampling points in the neighborhood space corresponding to the feature sampling point is obtained.

3. The spatial intelligent 3D reconstruction method based on cross-scale topological consistency according to claim 2, characterized in that, The specific steps in step S16 are as follows: Create a preset interval for the sampling density of the neighborhood space. If the density of the sampling points in the neighborhood space is within the preset interval, the feature sampling points are classified as sampling points with reasonable density. If not, the feature sampling points are classified as abnormal sampling points. Distance values ​​are collected for the nearest neighbor sampling points corresponding to the sampling points with reasonable density, resulting in multiple nearest neighbor sampling distance values. The obtained nearest neighbor sampling distance values ​​are compared, and the nearest neighbor sampling point corresponding to the maximum nearest neighbor sampling distance value is marked as the peak distance sampling point. The distance values ​​between the peak distance sampling point and the corresponding nearest neighbor sampling point are obtained, resulting in multiple nearest neighbor sampling distance values. The standard deviation of the obtained multiple nearest neighbor sampling distance values ​​is calculated to obtain the standard deviation of the nearest neighbor sampling distance. A preset value for the standard deviation of the nearest neighbor sampling distance is set. If the standard deviation of the nearest neighbor sampling distance is greater than or equal to the preset value, the sampling point with reasonable density is still judged as an abnormal sampling point. Within the sample voxel space, abnormal sampling points are removed to obtain the sample voxel processing space.

4. The spatial intelligent 3D reconstruction method based on cross-scale topological consistency according to claim 1, characterized in that, In step S23, the specific steps are as follows: For each region element point in the set of region element points, acquire the sample element point from the acquired multiple region element points, collect the three-dimensional coordinates of the sample element point in the scene space coordinate system, obtain the three-dimensional coordinates of the sample element, and acquire the laser sampling point corresponding to the three-dimensional coordinates of the sample element in the model space coordinate system. If the corresponding laser sampling point is a valid topological point, retain the laser sampling point; otherwise, discard the laser sampling point.

5. The spatial intelligent 3D reconstruction method based on cross-scale topological consistency according to claim 1, characterized in that, The specific steps in step S3 are as follows: Each region topology element in the target modeling region is labeled to obtain multiple element-labeled spatial regions; The element marking spatial region is discretized into several spatial element points. The three-dimensional coordinates corresponding to each spatial element point are collected according to the scene spatial coordinate system to obtain the three-dimensional coordinates of the real scene element. The real-time color score corresponding to each spatial element point is obtained using the RGB color model to obtain the color score of the real scene element. In the three-dimensional reconstruction model of the spatial region, the model space points corresponding to the three-dimensional coordinates of each real scene element are obtained through the three-dimensional coordinate system of the model space, and the color scores of the real scene elements are used to restore the color of the model space points. Color restoration is performed on the spatial region marked by each element to obtain a three-dimensional reconstruction model of the spatial region.

6. A spatial intelligent 3D reconstruction system based on cross-scale topological consistency, applicable to the spatial intelligent 3D reconstruction method based on cross-scale topological consistency as described in any one of claims 1-5, characterized in that, The spatial intelligent 3D reconstruction system includes: Model creation module: Acquire planar images of the region, spatially mark the modeling region to obtain a planar image of the modeling region, and perform preliminary 3D model creation on the target modeling region to obtain a preliminary 3D model of the target region; Topology Repair Module: Verifies the scene geometry and topology of the preliminary 3D model of the target area with the target modeling area, repairs the topology elements of the preliminary 3D model of the target area, and obtains the repaired 3D model of the target area. Color restoration module: Based on the real-time color data corresponding to the target modeling area, the initial 3D model of the target area is restored in real time to obtain the 3D reconstruction model of the spatial area.

Images