CN-121999179-A - Eave intelligent recognition and modification system for oblique photography model

CN121999179ACN 121999179 ACN121999179 ACN 121999179ACN-121999179-A

Abstract

The invention discloses an intelligent eave recognition and modification system for an oblique photography model, which relates to the technical field of intelligent eave recognition, and is characterized in that an original point cloud containing an eave area is acquired based on a data acquisition module and preprocessed to generate a preprocessed point cloud, an ideal point cloud model and a clean point cloud model are generated by utilizing a double-channel recognition module aiming at the preprocessed point cloud, a proper point cloud model is selected and output as a preferable point cloud model by utilizing a point cloud arbitration module, a three-dimensional grid model of the eave is generated by utilizing a contour guiding reconstruction module based on the preferable point cloud model, and a final three-dimensional model of the eave is generated by utilizing a texture remapping module aiming at the three-dimensional grid model.

Inventors

Sun Jingcong
Feng Luoyi
BAO ZHENGHONG
YANG ZEYAN
ZHAO JIANBO

Assignees

昆明麦普空间科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260326

Claims (8)

1. Eave intelligent identification and modification system towards oblique photography model, characterized by comprising: The data acquisition module is used for extracting an original point cloud containing an eave area from the oblique photography model, and preprocessing the original point cloud to generate a preprocessed point cloud; The double-path identification module is used for carrying out parallel processing on the preprocessed point cloud based on the double-path identification model, generating an ideal point cloud model through the first path model, and generating a cleaning point cloud model through the second path model; The point cloud arbitration module is connected with the double-path identification module, matches the ideal point cloud model with the cleaning point cloud model, and selects one model from the ideal point cloud model and the cleaning point cloud model to output as a preferred point cloud model based on the matching result; The contour guiding reconstruction module is connected with the point cloud arbitration module, extracts boundary lines of eave in the preferred point cloud model based on the scanning probe, generates a series of ordered two-dimensional section contours among the boundary lines by utilizing a two-dimensional section sequence generation mechanism, and outputs a three-dimensional grid model of the eave through a curved surface lofting algorithm; the texture remapping module is connected with the contour guiding reconstruction module, acquires texture samples from the oblique photography model through multi-view projection, generates a final texture map matched with the three-dimensional grid model after weighted fusion and texture synthesis, and generates a final three-dimensional model of the eave based on the final texture map.
2. The intelligent eave recognition and modification system for oblique photography models according to claim 1, wherein the data acquisition module comprises: The acquisition unit is used for receiving the oblique photography model file and converting the oblique photography model file into an original point cloud; the point cloud filtering unit is used for removing outlier noise points in the original point cloud through a radius filtering algorithm; the semantic pre-segmentation unit primarily identifies and extracts candidate regional point clouds containing eave through rules based on color, intensity and spatial distribution characteristics aiming at the original point clouds with outlier noise removed, and outputs the candidate regional point clouds as preprocessed point clouds.
3. The intelligent eave recognition and modification system for oblique photography models according to claim 1, wherein the dual-path recognition module comprises: generating a completion unit, as a first path model, identifying a missing part of an eave region in the preprocessing point cloud, carrying out deterministic completion on the missing part based on a pre-trained structure priori reasoning model, and outputting an ideal point cloud model with continuous structure on a global scale; And the iterative attention unit is used as a second path model, performs surface interpolation repair on the missing part based on missing part boundary point cloud data, and outputs a clean point cloud model with real details.
4. The intelligent eave recognition and modification system for oblique photography models according to claim 1, wherein the point cloud arbitration module comprises: the matching degree calculating unit is used for calculating the overall matching degree of the ideal point cloud model and the cleaning point cloud model, and presetting a matching degree threshold; The decision unit selects an output model based on the matching result; if the overall matching degree exceeds the matching degree threshold, a point cloud model is randomly output in the ideal point cloud model and the cleaning point cloud model to serve as a preferred point cloud model; If the overall matching degree is lower than the matching degree threshold, the method is used for evaluating an ideal point cloud model and cleaning the point cloud model based on a repairing cost evaluation mechanism; Constructing a standard test model, wherein the standard test model is a simplified geometric model constructed based on a convex hull generation algorithm; Respectively converting the ideal point cloud model and the cleaning point cloud model into standard test models, and calculating geometric consistency errors in the conversion process; The geometric consistency error is the average distance from the point cloud in the two point cloud models to the surface of the standard test model; And taking the geometric consistency error as an evaluation index of the repair cost, and outputting the point cloud model with low geometric consistency error as a preferred point cloud model.
5. The intelligent eave recognition and modification system for oblique photography models according to claim 1, wherein the contour guide reconstruction module comprises: The probe scanning unit is used for generating a virtual probe, and the probe is used for detecting the normal vector change gradient of the point cloud in the edge area of the preferred point cloud model and presetting a change gradient threshold; marking a point cloud as a candidate boundary point when the normal vector change gradient between the point cloud and the adjacent point cloud exceeds a change gradient threshold value; Connecting adjacent candidate boundary points to generate rough granularity boundaries of eave; A parallel section generating unit for generating a two-dimensional section profile sequence based on the coarse grain boundary; And the curved surface lofting unit remodels the boundary of the eave point cloud by taking the two-dimensional section profile sequence as a key section, and generates a three-dimensional grid model of the eave.
6. The intelligent eave recognition and modification system for oblique photography models according to claim 5, wherein the step of generating the two-dimensional profile sequence based on coarse grain boundaries is: determining the extending direction of the eave as a main slicing direction based on the coarse granularity boundary line; generating a series of parallel vertical cutting planes with high density along the direction perpendicular to the main slicing direction as point cloud slices; For one point cloud slice, calculating the intersection point of the point cloud slice and the coarse-granularity boundary line, and connecting all the intersection points to generate a two-dimensional section profile of the point cloud slice; and orderly arranging all the two-dimensional section profiles to generate a two-dimensional section profile sequence.
7. The intelligent eave recognition and modification system for an oblique photography model according to claim 6, wherein the step of generating the three-dimensional grid model of the eave by taking the two-dimensional profile sequence as a key cross section to remodel the boundary of the eave point cloud comprises the steps of: matching corresponding points of two adjacent two-dimensional section contours, and establishing point correspondence between the contours; Generating a smooth transition curved surface piece between adjacent contours by using a spline curved surface interpolation algorithm based on the point correspondence; and converting the generated curved surface sheet into triangular grids through a triangulation algorithm, and carrying out accurate matching on boundary vertexes of adjacent triangular grids and stitching to generate the three-dimensional grid model of the eave.
8. The intelligent eave recognition and modification system for oblique photography models according to claim 1, wherein the texture remapping module comprises: the UV calculation module is used for projecting the three-dimensional grid model to a two-dimensional plane to generate a reconstructed UV unfolding diagram; The texture generation module is used for sampling pixels of a corresponding area from the oblique photographing original image based on the reconstructed UV unfolded image to generate a brand new texture map matched with the three-dimensional grid model; and the texture binding module is used for endowing the brand new texture map to the three-dimensional grid model to finish texture mapping.

Description

Eave intelligent recognition and modification system for oblique photography model Technical Field The invention relates to the technical field of intelligent eave recognition, in particular to an intelligent eave recognition and modification system for an oblique photography model. Background With the wide application of the oblique photography measurement technology in the fields of smart cities, digital twinning and the like, a three-dimensional model based on oblique photography has become an important spatial data base, however, the prior art has obvious limitations in processing building detail structures, particularly eave features; The existing mainstream three-dimensional reconstruction method mostly adopts global optimization strategies, such as poisson reconstruction, voxelization and the like, when facing eave areas with uneven point cloud density and serious shielding, the existing method is difficult to maintain the feature integrity and structural rationality, and due to the fact that the eave areas are easily influenced by factors such as shielding and shading in the oblique photography acquisition process, point cloud data are lost and noise, the traditional reconstruction method is easy to generate geometric distortion, eave edge blurring and structural collapse and the like, meanwhile, in the aspect of texture mapping, texture coordinates of an original model are usually used in the prior art, when the geometric structure is corrected, the texture is distorted due to the simple coordinate synchronization mechanism, on the other hand, the existing processing method is mostly dependent on manual intervention for local restoration, and the existing processing method lacks an automatic processing flow aiming at eave features, so that the processing efficiency is low and the quality is inconsistent, and the technical defects limit the application effect of the oblique photography model in the refinement management to a certain extent; Therefore, developing an eave intelligent recognition and modification system oriented to an oblique photography model has important significance. Disclosure of Invention The invention aims to provide an intelligent eave recognition and modification system for an oblique photography model, which aims to solve the problem of the defects in the background technology. In order to achieve the purpose, the invention provides the following technical scheme that the intelligent eave recognition and modification system for the oblique photography model comprises the following steps: The data acquisition module is used for extracting an original point cloud containing an eave area from the oblique photography model, and preprocessing the original point cloud to generate a preprocessed point cloud; The double-path identification module is used for carrying out parallel processing on the preprocessed point cloud based on the double-path identification model, generating an ideal point cloud model through the first path model, and generating a cleaning point cloud model through the second path model; The point cloud arbitration module is connected with the double-path identification module, matches the ideal point cloud model with the cleaning point cloud model, and selects one model from the ideal point cloud model and the cleaning point cloud model to output as a preferred point cloud model based on the matching result; The contour guiding reconstruction module is connected with the point cloud arbitration module, extracts boundary lines of eave in the preferred point cloud model based on the scanning probe, generates a series of ordered two-dimensional section contours among the boundary lines by utilizing a two-dimensional section sequence generation mechanism, and outputs a three-dimensional grid model of the eave through a curved surface lofting algorithm; the texture remapping module is connected with the contour guiding reconstruction module, acquires texture samples from the oblique photography model through multi-view projection, generates a final texture map matched with the three-dimensional grid model after weighted fusion and texture synthesis, and generates a final three-dimensional model of the eave based on the final texture map. In a preferred embodiment, the data acquisition module comprises: The acquisition unit is used for receiving the oblique photography model file and converting the oblique photography model file into an original point cloud; the point cloud filtering unit is used for removing outlier noise points in the original point cloud through a radius filtering algorithm; the semantic pre-segmentation unit primarily identifies and extracts candidate regional point clouds containing eave through rules based on color, intensity and spatial distribution characteristics aiming at the original point clouds with outlier noise removed, and outputs the candidate regional point clouds as preprocessed point clouds. In a preferred embodiment, the dual path