CN-121999159-A - Three-dimensional terrain reconstruction method and system based on remote sensing image

CN121999159ACN 121999159 ACN121999159 ACN 121999159ACN-121999159-A

Abstract

The invention relates to the technical field of remote sensing mapping, in particular to a three-dimensional terrain reconstruction method and system based on remote sensing images, comprising the following steps of: the method comprises the steps of collecting visible light and near infrared three-dimensional multispectral data, carrying out pixel level registration based on geometric calibration, constructing a geometric registration multispectral image matrix, generating a wave band difference, extracting a structural edge, constructing a spectrum signal diffusion impedance field, forming a terrain self-adaptive matching support domain, calculating multispectral matching parallax, and resolving a three-dimensional terrain reconstruction model based on space front intersection.

Inventors

LUO JIANHONG
ZHAO YANWEN
LONG YUTING
XU YONGLONG
FANG WEI
WU HUAN
ZHAO PEIFENG
WU JUNFANG

Assignees

云南省测绘地理信息科技发展有限公司

Dates

Publication Date: 20260508
Application Date: 20260127

Claims (10)

1. The three-dimensional terrain reconstruction method based on the remote sensing image is characterized by comprising the following steps of: S1, acquiring visible light wave band and near infrared wave band data of a stereopair by a multispectral remote sensing sensor, performing pixel-level geometric registration on the wave band data based on geometric calibration parameters of the sensor, and constructing a geometric registration multispectral image matrix; s2, extracting the geometric registration multispectral image matrix to construct a wave band differential spectrum, performing convolution operation on the wave band differential spectrum by using a Sobel operator to calculate differential gradient amplitude, screening pixels larger than a preset structural threshold value, and generating a topographic structure edge feature map; S3, calling the edge feature map of the topographic structure to lock the edge pixels, calculating the spectrum feature difference of adjacent pixels of the image, mapping the spectrum feature difference to be a basic impedance parameter, adjusting the basic impedance parameter at the edge pixels to a maximum blocking value, and constructing a spectrum signal diffusion impedance field; s4, calculating the residual signal intensity of a central pixel diffusion signal to be matched by using the spectrum signal diffusion impedance field, and aggregating pixel points larger than an intensity threshold value to construct a terrain self-adaptive matching support domain; And S5, calculating multispectral matching cost in a stereo image pair based on the terrain self-adaptive matching support domain to obtain a parallax value, inputting the parallax value into a space front intersection model to execute three-dimensional coordinate calculation, and constructing a three-dimensional terrain reconstruction model.
2. The remote sensing image based three-dimensional terrain reconstruction method according to claim 1, wherein the geometrical registration multispectral image matrix comprises multiband radiance values and geospatial coordinate information, the terrain structure edge feature map comprises a gradient magnitude matrix and an edge binary mask, the spectrum signal diffusion impedance field comprises spectrum similarity weights and edge stop function values, the terrain adaptive matching support domain comprises a connected domain pixel set and an adaptive aggregation window, and the three-dimensional terrain reconstruction model comprises three-dimensional point cloud coordinates and digital elevation model data.
3. The three-dimensional terrain reconstruction method based on remote sensing images according to claim 1, wherein the specific steps of S1 are as follows: S101, collecting visible light and near infrared data of a stereopair by a multispectral remote sensing sensor, performing analog-to-digital conversion, analyzing spectral radiance response values, dividing the spectral radiance response values into independent frequency spectrum subsets, performing normalization coding, and establishing an original multispectral band data frame; S102, calling the original multispectral band data frame, reading geometric calibration parameters, calculating relative displacement vectors between imaging planes of a visible light band and a near infrared band, constructing a coordinate transformation equation according to the geometric calibration parameters to solve the deviation, and obtaining a pixel-level geometric transformation parameter set; And S103, performing spatial resampling on the original multispectral band data frame based on the pixel-level geometric transformation parameter set, mapping the near-infrared channel pixel coordinates to a visible light channel reference grid, performing bilinear interpolation, performing matrix splicing on the aligned band data along the depth direction, and generating a geometric registration multispectral image matrix.
4. The three-dimensional terrain reconstruction method based on remote sensing images according to claim 3, wherein the specific steps of S2 are as follows: s201, calling the geometric registration multispectral image matrix, separating a near infrared band layer from a visible light band layer, performing subtraction operation and absolute value processing on the same-coordinate pixel points, and recombining difference values according to original coordinates to generate a multispectral band gray level difference matrix; s202, based on the multispectral band gray scale difference matrix, configuring a discrete differential convolution kernel to execute sliding window operation, respectively calculating gray scale change rates of pixel points in horizontal and vertical directions, and executing vector synthesis and module length calculation on gradient components to obtain a spatial gradient amplitude distribution matrix; And S203, traversing the pixel point gradient modular length and executing numerical comparison according to a preset structure threshold value aiming at the spatial gradient amplitude distribution matrix, extracting the pixel points larger than the preset structure threshold value and eliminating the background, and establishing a topographic structure edge feature map.
5. The three-dimensional terrain reconstruction method based on remote sensing images according to claim 4, wherein the preset structural threshold is determined based on statistical characteristics of a spatial gradient amplitude distribution matrix, specifically, traversing statistics is performed on gradient modular lengths of all pixels in the spatial gradient amplitude distribution matrix, a maximum value, a minimum value and a mean value of the gradient modular lengths are obtained, linear calculation is performed on a difference value between the mean value and the maximum value according to a preset proportionality coefficient on the basis, and the preset structural threshold is set to be a numerical value corresponding to a linear calculation result.
6. The three-dimensional terrain reconstruction method based on remote sensing images according to claim 4, wherein the specific steps of S3 are as follows: S301, invoking the topographic structure edge feature map, traversing pixels to retrieve geostructure boundary target points, extracting target point plane row and column position parameters, and serializing according to a spatial adjacent relation to generate an edge pixel coordinate index set; S302, calling the geometric registration multispectral image matrix, constructing a pixel four-neighborhood window aiming at each pixel point in the matrix, calculating the Euclidean distance average value of a central pixel vector and a neighborhood pixel vector, inputting the average value into a preset positive correlation mapping function, calculating a signal transmission resistance value, and establishing an initial spectral impedance distribution matrix; S303, based on the initial spectrum impedance distribution matrix and the edge pixel coordinate index set, overwriting edge impedance to be the maximum blocking value according to coordinate positioning, reserving impedance values of other positions in the matrix, and performing global data integration to generate a spectrum signal diffusion impedance field.
7. The three-dimensional terrain reconstruction method based on remote sensing images according to claim 6, wherein the specific steps of S4 are as follows: S401, utilizing the spectrum signal diffusion impedance field, setting a central pixel as a diffusion source, configuring initial pulse energy, executing energy dissipation operation to calculate accumulated attenuation, utilizing the initial pulse energy value to subtract the accumulated attenuation, calculating residual energy value of a pixel point, and generating a residual signal intensity matrix of the pixel; S402, calling the pixel residual signal intensity matrix, performing logic comparison with an intensity threshold, screening pixel units with residual energy values larger than the intensity threshold, extracting row and column indexes, and generating an effective signal response pixel set; S403, based on the effective signal response pixel set, mapping the pixels back to an original coordinate system, detecting and aggregating a spatial adjacent relation, merging adjacent pixels into an irregular area, determining a boundary, and constructing a terrain self-adaptive matching supporting domain.
8. The three-dimensional terrain reconstruction method based on remote sensing images according to claim 7, wherein the intensity threshold is calculated statistically based on residual energy value distribution of all pixel units in a pixel residual signal intensity matrix, minimum and maximum values of residual energy values are extracted, an energy value interval is constructed, the energy value interval is divided linearly according to a preset proportionality coefficient, and a demarcation energy value located on the upper limit side of the energy value interval is determined as the intensity threshold.
9. The three-dimensional terrain reconstruction method based on remote sensing images according to claim 7, wherein the specific steps of S5 are as follows: S501, calling the terrain self-adaptive matching support domain, defining a search range according to a regional boundary, calculating the multi-dimensional spectrum gray scale difference of the feature points, determining the geometric displacement of the corresponding pixels based on a difference minimization criterion, and generating a binocular parallax mapping matrix; S502, calling the binocular parallax mapping matrix, calculating a target pixel depth value by utilizing the inverse relation of parallax and depth in combination with binocular baseline length, camera focal length and optical axis deviation parameters, and resolving a three-dimensional position vector by combining plane coordinates to establish a discrete point cloud coordinate set; And S503, performing triangular mesh subdivision of the space point cloud based on the discrete point cloud coordinate set, constructing a topological connection relation among the point clouds, connecting discrete three-dimensional coordinate points into a continuous and closed geometric polygonal surface patch, mapping spectrum texture information of an original image to the surface of the surface patch for rendering, and generating a three-dimensional terrain reconstruction model.
10. A three-dimensional terrain reconstruction system based on remote sensing images, wherein the system is used for realizing the three-dimensional terrain reconstruction method based on remote sensing images as set forth in any one of claims 1 to 9, and the system comprises: The image registration module is used for acquiring visible light wave band and near infrared wave band data of the stereopair through the multispectral remote sensing sensor, performing pixel-level geometric registration on the wave band data based on the geometric calibration parameters of the sensor, and constructing a geometric registration multispectral image matrix; the differential edge module is used for extracting the geometric registration multispectral image matrix to construct a wave band differential spectrum, performing convolution operation on the wave band differential spectrum by adopting a Sobel operator to calculate differential gradient amplitude, screening pixels larger than a preset structural threshold value and generating a topographic structure edge feature map; The impedance construction module is used for calling the edge feature map of the topographic structure to lock the edge pixels, calculating the spectrum feature difference of adjacent pixels of the image and mapping the spectrum feature difference to be a basic impedance parameter, adjusting the basic impedance parameter at the edge pixels to a maximum blocking value, and constructing a spectrum signal diffusion impedance field; The support domain generation module calculates the residual signal intensity of the center pixel diffusion signal to be matched by utilizing the spectrum signal diffusion impedance field, aggregates pixel points larger than an intensity threshold value, and constructs a terrain self-adaptive matching support domain; And the three-dimensional resolving module is used for calculating multispectral matching cost in the stereo image pair based on the terrain self-adaptive matching support domain to obtain a parallax value, inputting the parallax value into the space front intersection model to execute three-dimensional coordinate resolving, and constructing a three-dimensional terrain reconstruction model.

Description

Three-dimensional terrain reconstruction method and system based on remote sensing image Technical Field The invention relates to the technical field of remote sensing mapping, in particular to a three-dimensional terrain reconstruction method and system based on remote sensing images. Background The technical field of remote sensing mapping relates to non-contact detection and data acquisition of the earth surface by utilizing a satellite aircraft or an unmanned aerial vehicle carried sensor, and further position shape and attribute information of the earth surface are acquired through a physical model and a geometric processing means. The traditional three-dimensional terrain reconstruction method is characterized in that a space geometrical relationship at shooting moment is restored according to a photogrammetry principle by utilizing an aerial or satellite remote sensing image overlapping region with multiple views, in the specific implementation, homonymous feature points in the image are firstly identified by adopting a feature extraction operator, epipolar constraint conditions are built by combining internal and external azimuth elements of a camera to carry out three-dimensional matching, the space three-dimensional coordinates of homonymous point pairs are calculated according to a triangulation principle, sparse point clouds are generated, then a parallax map is calculated pixel by adopting a dense matching strategy, parallax data is mapped into a three-dimensional dense point cloud or grid model containing elevation information through coordinate conversion, and the fluctuation form and the ground feature structure of the earth surface are restored. The traditional three-dimensional terrain reconstruction method relies on texture features of visible light images for matching, but under complex environments such as illumination change, ground feature material difference or vegetation coverage, the texture stability of the images is poor, so that matching of homonymous feature points is easy to make mistakes. Under the constraint of polar lines, the pixel matching range is fixed, and the change of complex terrain or structural boundary is difficult to adapt to, so that when the terrain mutation or edge region is processed, the matching support domain easily spans different terrain types, the accuracy of parallax estimation is influenced, the problem on the elevation precision and structural continuity of a reconstruction result is caused, and particularly in the complex or variable terrain region, the integrity and precision of point cloud are greatly influenced. Disclosure of Invention In order to solve the problem that the traditional three-dimensional terrain reconstruction method relies on texture features of visible light images for matching, but under complex environments such as illumination change, ground feature material difference or vegetation coverage, the texture stability of the images is poor, so that matching of homonymous feature points is easy to make mistakes. Under the constraint of polar lines, the pixel matching range is fixed, and the change of complex terrain or structural boundary is difficult to adapt, so that when the terrain mutation or edge region is processed, the matching support domain easily spans different terrain types, the accuracy of parallax estimation is influenced, and the problem on the elevation precision and structural continuity of a reconstruction result is caused, particularly the technical problem that the integrity and precision of point cloud are greatly influenced in the complex or changeable terrain region is solved. In order to achieve the above purpose, the invention adopts a three-dimensional terrain reconstruction method based on remote sensing images, which comprises the following steps: S1, acquiring visible light wave band and near infrared wave band data of a stereopair by a multispectral remote sensing sensor, performing pixel-level geometric registration on the wave band data based on geometric calibration parameters of the sensor, and constructing a geometric registration multispectral image matrix; s2, extracting the geometric registration multispectral image matrix to construct a wave band differential spectrum, performing convolution operation on the wave band differential spectrum by using a Sobel operator to calculate differential gradient amplitude, screening pixels larger than a preset structural threshold value, and generating a topographic structure edge feature map; S3, calling the edge feature map of the topographic structure to lock the edge pixels, calculating the spectrum feature difference of adjacent pixels of the image, mapping the spectrum feature difference to be a basic impedance parameter, adjusting the basic impedance parameter at the edge pixels to a maximum blocking value, and constructing a spectrum signal diffusion impedance field; s4, calculating the residual signal intensity of a central pixel diffusion sign