Three-dimensional geoscience modeling method based on multi-source data fusion
By using drones equipped with 3D scanners and oblique photography cameras, combined with multi-source data fusion technology, the shortcomings of building 3D models with a single technology were overcome, achieving high-precision 3D terrain modeling and improving the model's refinement and visualization effects.
Patent Information
- Application Number
- CN202510967894.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-14
- Publication Date
- 2025-11-21
AI Technical Summary
In existing technologies, both oblique photogrammetry and 3D laser scanning have their own shortcomings when constructing 3D models, making it difficult to obtain accurate models and affecting the integrity and visualization of 3D terrain modeling.
A drone equipped with a 3D scanner and a five-lens oblique photography camera was used to scan the area and take pictures from multiple angles. Through denoising, registration and correction processing, combined with ICP algorithm and manual registration method, the oblique photography point cloud and 3D laser point cloud were fused to form a high-precision fused point cloud model.
It improves the detail and realism of 3D models, obtains high-precision fused point cloud models, overcomes the shortcomings of single technologies, and enhances the integrity and visualization of models.
Smart Images
Figure CN120997420A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of three-dimensional geoscientific modeling methods, specifically a three-dimensional geoscientific modeling method based on multi-source data fusion. Background Technology
[0002] With rapid economic development and social progress, humanity has placed higher demands on spatial data in order to better understand and utilize the Earth and its resources, protect the environment, and improve the quality of life. Three-dimensional models can present complex real-world scenarios in an intuitive and visual way, making them easier for people to understand and master.
[0003] Oblique photogrammetry, as one of the most commonly used techniques for acquiring three-dimensional information of buildings, has gone through three development stages: analog photogrammetry, analytical photogrammetry, and digital photogrammetry. With the rapid development of computer technology and UAV technology, the data acquisition and processing capabilities of oblique photogrammetry have been comprehensively innovated. The oblique photogrammetry field acquisition platform has transitioned from manned aircraft to UAVs. UAVs can be divided into rotary-wing UAVs and fixed-wing UAVs according to their flight mode. Multi-rotor UAVs are relatively simple in design, have relatively low manufacturing and maintenance costs, and are easy to modify and upgrade. They can achieve vertical take-off and landing in confined spaces without runways or taxiing distances, which greatly increases their flexibility and convenience of use. However, due to factors such as terrain and tree obstruction and terrain surface texture, the model integrity is relatively low, and oblique photogrammetry models usually have a large amount of data, which is not conducive to model visualization.
[0004] 3D laser scanning technology can quickly acquire massive amounts of point data, making up for the shortcomings of traditional single-point mapping. It can perform all-round scanning of target objects around the clock, enabling rapid acquisition and processing of 3D surface data. It has advantages such as non-contact, high efficiency, and high precision, but it also has disadvantages such as high cost and large data volume. When using the laser point cloud 3D model as the data foundation for model visualization application development, it has high requirements for equipment and network transmission speed, and is not suitable for large-scale scene model visualization construction.
[0005] Since each surveying technique has its limitations, using only one technique to build a 3D model will not yield an accurate model, hindering the development of 3D terrain modeling planning. Therefore, to construct a more accurate 3D model, multi-source data should be fused.
[0006] Therefore, a three-dimensional geoscientific modeling method based on multi-source data fusion is proposed to address the above problems. Summary of the Invention
[0007] The purpose of this invention is to provide a three-dimensional geoscientific modeling method based on multi-source data fusion, the steps of which include:
[0008] Step 1: Divide the area to be modeled, set the flight trajectory in each area, use a drone equipped with a 3D scanner to scan the area in each area, and perform noise reduction and registration on the obtained scanned images to obtain a complete 3D terrain scan map.
[0009] Step 1 specifically includes:
[0010] Step 11: After exploring the area to be measured, determine the modeling range, divide the area according to the size and elevation difference, set the flight trajectory in each area, and use a drone equipped with a 3D scanner to scan the area;
[0011] Step 12: Denoise the acquired point cloud image using the sparse outlier removal method. This method calculates the average distance from each point to all its neighboring points. Assuming the result is a Gaussian distribution, points whose average distance is outside the standard range are defined as outliers and removed from the dataset.
[0012] Step 13: Apply the iterative nearest point automatic registration algorithm to the point cloud data from both the pre- and post-capture periods. This algorithm finds the closest point pair in each of the two point cloud sets, calculates the error after transformation based on the estimated transformation relationship, and iterates until the set objective function reaches its minimum value. This yields the optimal translation and rotation matrices to determine the final transformation relationship. The objective function is as follows:
[0013]
[0014] Where R represents the rotation matrix, T represents the translation matrix, k is the number of points in the point cloud to be registered, and a i For reference point cloud; b i The point cloud to be registered is then matched one-to-one with the reference point cloud data to complete the registration of the two point cloud data, thus obtaining a complete 3D topographic scan map.
[0015] Step 2: Set the flight path for each area, and use a drone equipped with a five-lens oblique photography camera to take multi-angle pictures of the area in each divided area. Correct and stitch the obtained oblique images to obtain a complete terrain oblique image.
[0016] Step 2 specifically includes:
[0017] Step 21: Within each defined region, use a drone equipped with a five-lens oblique photography camera to capture images of the region from multiple angles to obtain oblique images. Preprocess the oblique images and input the preprocessed oblique images into a convolutional neural network to obtain feature maps of the oblique images. Transform the feature maps to obtain the matrix of the oblique images. Normalize the matrix of the oblique images to obtain the feature vector of the oblique images.
[0018] Step 22: Use a feature matching algorithm to match the feature vectors of the tilted image to obtain the feature points of the tilted image. Calculate the fundamental matrix or essential matrix of the feature points of the tilted image to obtain the fundamental matrix or essential matrix of the feature points. Use the multi-view geometry principle to estimate the pose of the fundamental matrix or essential matrix of the feature points to obtain the pose information of the tilted image.
[0019] Step 23: Perform geometric correction calculations on the pose information of the tilted image to obtain geometric transformation parameters. Adjust the tilted image according to the geometric transformation parameters to obtain the corrected tilted image.
[0020] Step 24: Stitch together the corrected tilted images to obtain a complete terrain image model.
[0021] Step 3: Fuse the 3D terrain scan image with the terrain tilt image to obtain a fused point cloud model. Step 3 specifically includes:
[0022] Step 31: Convert the point cloud data from the complete 3D terrain scan map and the point cloud data from the complete terrain tilt image into a common point cloud format. Then extract the point cloud with obvious features from the terrain image model and fuse these point clouds with the laser point cloud obtained from the 3D terrain scan map.
[0023] Step 32: Using the 3D laser point cloud as a reference, the iterative nearest neighbor registration method, which combines the ICP algorithm with manual registration, is used to register the oblique photogrammetry point cloud with the 3D laser point cloud, thereby obtaining a high-precision fused point cloud model.
[0024] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention divides the area to be measured, plans the flight trajectory of the UAV in each small area, uses the UAV equipped with a 3D scanner to scan the area, and performs noise reduction and registration on the obtained scanned images to obtain a complete 3D terrain scan map. The UAV is equipped with a five-lens oblique photography camera to take pictures of the area from multiple angles, and the obtained oblique images are corrected and stitched to obtain a complete terrain oblique image. Finally, the point cloud with obvious features in the oblique image is extracted and fused with the point cloud of the terrain oblique image. The oblique photography point cloud is registered with the 3D laser point cloud, and the multi-source data is fused to obtain a high-precision fused point cloud model, thereby improving the refinement and realism of the final 3D model. Attached Figure Description
[0025] Figure 1 This is a flowchart of the steps of the present invention. Detailed Implementation
[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.
[0027] As attached Figure 1 As shown, the present invention provides a three-dimensional geoscientific modeling method based on multi-source data fusion. The steps of the three-dimensional geoscientific modeling method include:
[0028] Step 1: Divide the area to be modeled, set the flight trajectory in each area, use a drone equipped with a 3D scanner to scan the area in each area, and perform noise reduction and registration on the obtained scanned images to obtain a complete 3D terrain scan map.
[0029] Step 1 specifically includes:
[0030] Step 11: After exploring the area to be measured, determine the modeling range, divide the area according to the size and elevation difference, set the flight trajectory in each area, and use a drone equipped with a 3D scanner to scan the area;
[0031] Step 12: Denoise the acquired point cloud image using the sparse outlier removal method. This method calculates the average distance from each point to all its neighboring points. Assuming the result is a Gaussian distribution, points whose average distance is outside the standard range are defined as outliers and removed from the dataset.
[0032] Step 13: Apply the iterative nearest point automatic registration algorithm to the point cloud data from both the pre- and post-capture periods. This algorithm finds the closest point pair in each of the two point cloud sets, calculates the error after transformation based on the estimated transformation relationship, and iterates until the set objective function reaches its minimum value. This yields the optimal translation and rotation matrices to determine the final transformation relationship. The objective function is as follows:
[0033]
[0034] Where R represents the rotation matrix, T represents the translation matrix, k is the number of points in the point cloud to be registered, and a i For reference point cloud; b i The point cloud to be registered is then matched one-to-one with the reference point cloud data to complete the registration of the two point cloud data, thus obtaining a complete 3D topographic scan map.
[0035] Step 2: Set the flight path for each area, and use a drone equipped with a five-lens oblique photography camera to take multi-angle pictures of the area in each divided area. Correct and stitch the obtained oblique images to obtain a complete terrain oblique image.
[0036] Step 2 specifically includes:
[0037] Step 21: Within each defined region, use a drone equipped with a five-lens oblique photography camera to capture images of the region from multiple angles to obtain oblique images. Preprocess the oblique images and input the preprocessed oblique images into a convolutional neural network to obtain feature maps of the oblique images. Transform the feature maps to obtain the matrix of the oblique images. Normalize the matrix of the oblique images to obtain the feature vector of the oblique images.
[0038] Step 22: Use a feature matching algorithm to match the feature vectors of the tilted image to obtain the feature points of the tilted image. Calculate the fundamental matrix or essential matrix of the feature points of the tilted image to obtain the fundamental matrix or essential matrix of the feature points. Use the multi-view geometry principle to estimate the pose of the fundamental matrix or essential matrix of the feature points to obtain the pose information of the tilted image.
[0039] Step 23: Perform geometric correction calculations on the pose information of the tilted image to obtain geometric transformation parameters. Adjust the tilted image according to the geometric transformation parameters to obtain the corrected tilted image.
[0040] Step 24: Stitch together the corrected tilted images to obtain a complete terrain image model.
[0041] Step 3: Fuse the 3D terrain scan image with the terrain tilt image to obtain a fused point cloud model. Step 3 specifically includes:
[0042] Step 31: Convert the point cloud data from the complete 3D terrain scan map and the point cloud data from the complete terrain tilt image into a common point cloud format. Then extract the point cloud with obvious features from the terrain image model and fuse these point clouds with the laser point cloud obtained from the 3D terrain scan map.
[0043] Step 32: Using the 3D laser point cloud as a reference, the iterative nearest neighbor registration method, which combines the ICP algorithm and manual registration, is used to register the oblique photogrammetry point cloud with the 3D laser point cloud to obtain a high-precision fused point cloud model.
[0044] Suppose we have two point clouds, P and Q, where P is a 3D laser point cloud and Q is an oblique photographic point cloud. We need to find a rigid body transformation T (rotation matrix R and translation vector t) that aligns Q and P as closely as possible. Transformation T can be expressed as:
[0045] T(Q) = RQ + t;
[0046] The ICP algorithm solves for R and t by minimizing the following objective function;
[0047]
[0048] Where E(R, T) is the error function used to measure the degree of matching between two point clouds, n is the number of nearest neighbor pairs, and P... i It is a point in a 3D laser point cloud, Q i These are points in an oblique photographic point cloud, where R is the rotation matrix and t is the translation vector. This formula calculates the value of each pair of points (P). i Q i The sum of squares of the Euclidean distances between the two point clouds is the sum of squares of the errors of all point pairs. By minimizing this error function, we can find the optimal rotation and translation parameters that make the two point clouds overlap as much as possible.
[0049] Any technical solution that achieves the above-mentioned technical effects by utilizing the technical solutions described in this invention, or by designing similar technical solutions by those skilled in the art under the inspiration of the technical solutions described in this invention, falls within the protection scope of this invention.
Claims
1. A three-dimensional geoscientific modeling method based on multi-source data fusion, characterized in that: The steps of this 3D geoscientific modeling method include: Step 1: Divide the area to be modeled, set the flight trajectory in each area, use a drone equipped with a 3D scanner to scan the area in each area, and perform noise reduction and registration on the obtained scanned images to obtain a complete 3D terrain scan map. Step 2: Set the flight path for each area, and use a drone equipped with a five-lens oblique photography camera to take multi-angle pictures of the area in each divided area. Correct and stitch the obtained oblique images to obtain a complete terrain oblique image. Step 3: Fuse the 3D terrain scan image with the terrain tilt image to obtain a fused point cloud model.
2. The three-dimensional geoscientific modeling method based on multi-source data fusion according to claim 1, characterized in that: Step 1 specifically includes: Step 11: After exploring the area to be measured, determine the modeling range, divide the area according to the size and elevation difference, set the flight trajectory in each area, and use a drone equipped with a 3D scanner to scan the area; Step 12: Denoise the acquired point cloud image using the sparse outlier removal method. This method calculates the average distance from each point to all its neighboring points. Assuming the result is a Gaussian distribution, points whose average distance is outside the standard range are defined as outliers and removed from the dataset. Step 13: Apply the iterative nearest point automatic registration algorithm to the point cloud data from both the pre- and post-capture periods. This algorithm finds the closest point pair in each of the two point cloud sets, calculates the error after transformation based on the estimated transformation relationship, and iterates until the set objective function reaches its minimum value. This yields the optimal translation and rotation matrices to determine the final transformation relationship. The objective function is as follows: Where R represents the rotation matrix, T represents the translation matrix, k is the number of points in the point cloud to be registered, and a i For reference point cloud; b i The point cloud to be registered is then matched one-to-one with the reference point cloud data to complete the registration of the two point cloud data, thus obtaining a complete 3D topographic scan map.
3. The three-dimensional geoscientific modeling method based on multi-source data fusion according to claim 1, characterized in that: Step 2 specifically includes: Step 21: Within each defined region, use a drone equipped with a five-lens oblique photography camera to capture images of the region from multiple angles to obtain oblique images. Preprocess the oblique images and input the preprocessed oblique images into a convolutional neural network to obtain feature maps of the oblique images. Transform the feature maps to obtain the matrix of the oblique images. Normalize the matrix of the oblique images to obtain the feature vector of the oblique images. Step 22: Use a feature matching algorithm to match the feature vectors of the tilted image to obtain the feature points of the tilted image. Calculate the fundamental matrix or essential matrix of the feature points of the tilted image to obtain the fundamental matrix or essential matrix of the feature points. Use the multi-view geometry principle to estimate the pose of the fundamental matrix or essential matrix of the feature points to obtain the pose information of the tilted image. Step 23: Perform geometric correction calculations on the pose information of the tilted image to obtain geometric transformation parameters. Adjust the tilted image according to the geometric transformation parameters to obtain the corrected tilted image. Step 24: Stitch together the corrected tilted images to obtain a complete terrain image model.
4. The three-dimensional geoscientific modeling method based on multi-source data fusion according to claim 1, characterized in that: Step 3 specifically includes: Step 31: Convert the point cloud data from the complete 3D terrain scan map and the point cloud data from the complete terrain tilt image into a common point cloud format. Then extract the point cloud with obvious features from the terrain image model and fuse these point clouds with the laser point cloud obtained from the 3D terrain scan map. Step 32: Using the 3D laser point cloud as a reference, the iterative nearest neighbor registration method, which combines the ICP algorithm with manual registration, is used to register the oblique photogrammetry point cloud with the 3D laser point cloud, thereby obtaining a high-precision fused point cloud model.