CN-121982236-A - Three-dimensional scene reconstruction method based on aerial image data and laser point cloud

CN121982236ACN 121982236 ACN121982236 ACN 121982236ACN-121982236-A

Abstract

The invention relates to a three-dimensional scene reconstruction method based on aerial image data and laser point cloud, which comprises the steps of carrying out data preprocessing on the aerial image data and the laser point cloud data to obtain corrected image data and classified point cloud data; performing feature extraction and registration processing on the corrected image data and the classified point cloud data to obtain registered image data and registered point cloud data, fusing the registered image data and the registered point cloud data to generate feature fusion data, performing point cloud triangularization processing and texture mapping to reconstruct a three-dimensional scene surface, and performing geometric optimization and detail enhancement on the three-dimensional scene surface to obtain a three-dimensional scene model. The method can perform data preprocessing on aerial image data and laser point cloud data, and then perform feature extraction, registration processing, data fusion, triangularization processing, texture mapping, geometric optimization and detail enhancement, so that the aerial image data and the laser point cloud data can be fully and effectively utilized, and the effect of three-dimensional scene reconstruction is improved.

Inventors

CHEN YULONG
ZHANG YING
JU JINJUN
WANG HUI
Song Wanma
ZHENG PENGLONG
LI FENG

Assignees

中国人民解放军陆军工程大学

Dates

Publication Date: 20260505
Application Date: 20260123

Claims (10)

1. The three-dimensional scene reconstruction method based on aerial image data and laser point cloud is characterized by comprising the following steps of: collecting aerial image data and laser point cloud data, and performing data preprocessing on the aerial image data and the laser point cloud data to obtain corrected image data and classified point cloud data; Feature extraction is carried out on the corrected image data and the classified point cloud data, and registration processing is carried out, so that registration image data and registration point cloud data are obtained; fusing the registration image data and the registration point cloud data to generate feature fusion data; performing point cloud triangularization processing on the feature fusion data, performing texture mapping, and reconstructing a three-dimensional scene surface; and performing geometric optimization and detail enhancement on the three-dimensional scene surface to obtain a three-dimensional scene model.
2. The three-dimensional scene reconstruction method based on aerial image data and laser point cloud according to claim 1, wherein the acquiring aerial image data and laser point cloud data specifically comprises: acquiring aerial photography image parameters and planning an aerial photography route; Acquiring aerial images according to the aerial image parameters and the aerial photographing route to acquire aerial image data; Acquiring laser radar parameters and flight parameters; And acquiring laser point cloud according to the laser radar parameters and the flight parameters, and acquiring laser point cloud data.
3. The method for reconstructing a three-dimensional scene based on aerial image data and laser point clouds according to claim 2, wherein the aerial image parameters comprise camera resolution, camera focal length, camera aperture and camera shutter speed, the laser radar parameters comprise laser radar ranging accuracy, laser radar scanning frequency and laser radar point cloud density, and the flight parameters comprise flight altitude, flight speed and scanning angle.
4. The three-dimensional scene reconstruction method based on aerial image data and laser point cloud according to claim 1, wherein the performing data preprocessing on the aerial image data and the laser point cloud data to obtain corrected image data and classified point cloud data specifically comprises: performing radiation correction and geometric correction on the aerial image data to obtain corrected image data; Denoising the laser point cloud data to obtain denoised point cloud data; Performing point cloud classification on the denoising point cloud data to obtain classified point cloud data; The specific algorithm formula of the denoising process is as follows: ; Wherein, the In order to remove the noise of the data, Is the original data point; Is the filter window size; is rounded downwards; Is the post-denoising step Data; Is the original first Data.
5. The three-dimensional scene reconstruction method based on aerial image data and laser point cloud as claimed in claim 4, wherein the radiation correction is used for eliminating radiation non-uniformity in aerial image data to enable brightness and contrast of different images to be consistent, the geometric correction is used for eliminating geometric distortion of aerial image data to enable correct geographic coordinates and projection relations, the denoising process is used for removing abnormal flying points and isolated points and improving quality of point cloud data, and the point cloud classification is used for classifying data into ground points, building points and vegetation points according to elevation, intensity and geometric shape of denoising point cloud data.
6. The three-dimensional scene reconstruction method based on aerial photography image data and laser point cloud according to claim 1, wherein the feature extraction and registration processing are performed on the corrected image data and the classified point cloud data, and the obtaining registered image data and registered point cloud data specifically includes: Extracting features of the corrected image data to obtain image feature data; Extracting characteristics of the classified point cloud data to obtain point cloud characteristic data; Performing coarse registration on the image characteristic data and the point cloud characteristic data, and recording a coarse registration result; And carrying out fine registration on the image characteristic data and the point cloud characteristic data based on the coarse registration result to obtain registration image data and registration point cloud data.
7. The three-dimensional scene reconstruction method based on aerial image data and laser point clouds according to claim 1, wherein the feature fusion data is obtained through a preset deep learning fusion model.
8. The method for reconstructing a three-dimensional scene based on aerial image data and laser point clouds according to claim 1, wherein the performing the point cloud triangularization processing and the texture mapping on the feature fusion data, reconstructing a three-dimensional scene surface specifically comprises: Performing triangulation on the feature fusion data, and converting a triangular network model; Grid optimization is carried out on the triangular network model to obtain a triangular optimization model; extracting a plurality of texture region images corresponding to the triangular optimization model from the registered image data; And mapping a plurality of texture region images to the triangular optimization model, and reconstructing the three-dimensional scene surface.
9. The method for reconstructing a three-dimensional scene based on aerial image data and laser point clouds according to claim 8, wherein the performing grid optimization on the triangular network model to obtain a triangular optimization model specifically comprises: Removing redundant triangles in the triangle network model; repairing holes and cracks in the triangular network model; adjusting the topological structure of the triangular network model; And optimizing to generate a triangular optimizing model.
10. The method for reconstructing a three-dimensional scene based on aerial image data and laser point clouds according to claim 1, wherein the performing geometric optimization and detail enhancement on the three-dimensional scene surface to obtain a three-dimensional scene model specifically comprises: Performing error analysis and correction on the three-dimensional scene surface to obtain a corrected scene surface; smoothing the corrected scene surface to generate a smooth scene surface; Extracting and enhancing details of different scales on the smooth scene surface by utilizing a multi-scale analysis technology to obtain an enhanced scene surface; identifying a plurality of texture miss areas of the enhanced scene surface; And performing texture synthesis and mapping on the texture missing areas by adopting a texture synthesis technology to obtain a three-dimensional scene model.

Description

Three-dimensional scene reconstruction method based on aerial image data and laser point cloud Technical Field The invention relates to the technical field of three-dimensional scene reconstruction, in particular to a three-dimensional scene reconstruction method based on aerial image data and laser point cloud. Background Three-dimensional scene reconstruction is a process of recovering three-dimensional geometric structure and appearance information of a real-world scene from data such as two-dimensional images, videos or three-dimensional point clouds by using computer vision, graphics and sensor technology. The method has the core aim of converting discrete observation data into a continuous, complete and accurate three-dimensional model through an algorithm and a calculation model, thereby realizing the digital representation of the physical world. In the prior art, for three-dimensional scene reconstruction of aerial image data and laser point cloud, obvious defects exist in aspects of feature extraction, matching and fusion, and the advantages of the two data are difficult to fully develop, so that the three-dimensional scene reconstruction effect is not ideal. Disclosure of Invention Aiming at the technical problems in the prior art, the invention provides a three-dimensional scene reconstruction method based on aerial image data and laser point cloud. The technical scheme for solving the technical problems is as follows, the three-dimensional scene reconstruction method based on aerial image data and laser point cloud comprises the following steps: collecting aerial image data and laser point cloud data, and performing data preprocessing on the aerial image data and the laser point cloud data to obtain corrected image data and classified point cloud data; Feature extraction is carried out on the corrected image data and the classified point cloud data, and registration processing is carried out, so that registration image data and registration point cloud data are obtained; fusing the registration image data and the registration point cloud data to generate feature fusion data; performing point cloud triangularization processing on the feature fusion data, performing texture mapping, and reconstructing a three-dimensional scene surface; and performing geometric optimization and detail enhancement on the three-dimensional scene surface to obtain a three-dimensional scene model. As a further limitation of the technical solution of the embodiment of the present invention, the acquiring aerial image data and laser point cloud data specifically includes: acquiring aerial photography image parameters and planning an aerial photography route; Acquiring aerial images according to the aerial image parameters and the aerial photographing route to acquire aerial image data; Acquiring laser radar parameters and flight parameters; And acquiring laser point cloud according to the laser radar parameters and the flight parameters, and acquiring laser point cloud data. By further limiting the technical scheme of the embodiment of the invention, the aerial image parameters comprise camera resolution, camera focal length, camera aperture and camera shutter speed, the laser radar parameters comprise laser radar ranging precision, laser radar scanning frequency and laser radar point cloud density, and the flight parameters comprise flight altitude, flight speed and scanning angle. As a further limitation of the technical solution of the embodiment of the present invention, the performing data preprocessing on the aerial image data and the laser point cloud data to obtain corrected image data and classified point cloud data specifically includes: performing radiation correction and geometric correction on the aerial image data to obtain corrected image data; Denoising the laser point cloud data to obtain denoised point cloud data; Performing point cloud classification on the denoising point cloud data to obtain classified point cloud data; The specific algorithm formula of the denoising process is as follows: ; Wherein, the In order to remove the noise of the data,Is the original data point; Is the filter window size; is rounded downwards; Is the post-denoising step Data; Is the original first Data. The method is further limited by the technical scheme of the embodiment of the invention, the radiation correction is used for eliminating radiation non-uniformity in aerial image data to enable brightness and contrast of different images to be consistent, the geometric correction is used for eliminating geometric distortion of aerial image data to enable the aerial image data to have correct geographic coordinates and projection relations, the denoising processing is used for removing abnormal flying points and isolated points and improving the quality of point cloud data, and the point cloud classification is used for dividing the data into ground points, building points and vegetation points according to the elevation,