Island reef three-dimensional real scene surveying and mapping method based on unmanned equipment

Abstract

The application relates to the technical field of surveying and mapping engineering, and discloses an island reef three-dimensional real scene mapping method based on unmanned equipment, which comprises the following steps: analyzing high-resolution satellite remote sensing images to obtain a survey area sketch map, arranging ground control points to obtain high-precision geographic coordinates, planning adaptive flight routes for the unmanned equipment, collecting island reef multi-angle original image data, performing uniform light and color processing on the images to obtain a standardized image sequence, extracting a feature point set, completing multi-baseline multi-view matching, constructing a connection point network, taking the high-precision geographic coordinates as constraints, performing joint regional network adjustment by the beam method to obtain high-precision exterior orientation elements and encrypted point coordinates, performing pixel-by-pixel depth estimation based on the high-precision exterior orientation elements and the encrypted point coordinates, obtaining island reef three-dimensional point cloud data, constructing an irregular triangle network for the three-dimensional point cloud, and generating a measurable three-dimensional real scene model of the target island reef through texture mapping. The application can improve the efficiency of island reef three-dimensional real scene mapping.

Description

Technical Field

[0001] This invention relates to the field of surveying and mapping engineering technology, and in particular to a method for three-dimensional real-scene surveying and mapping of islands and reefs based on unmanned equipment. Background Technology

[0002] The island and reef areas have complex topography and features such as steep slopes and diverse rock textures. Existing island and reef surveying and mapping technologies are unable to accurately obtain the topographic information of the entire island and reef area in the early survey area planning stage. This results in a lack of targeted flight route planning, and the images collected by unmanned equipment are prone to coverage blind spots or redundant acquisition problems. The efficiency and effectiveness of acquiring raw image data from multiple angles are insufficient, and it is impossible to provide comprehensive and high-quality data source support for subsequent 3D modeling.

[0003] Existing methods for 3D real-scene mapping of islands and reefs lack precise extraction and matching mechanisms for image feature points in the image processing and 3D reconstruction stages. The constraints in the adjustment calculation are not reasonably configured, resulting in low accuracy of exterior orientation elements and densified point coordinates. Furthermore, the accuracy of depth estimation and the effectiveness of point cloud data are poor during the 3D point cloud construction process. The subsequent triangulation construction and texture mapping are not well-fitted, and the final generated 3D island and reef models have poor accuracy and measurability, making it difficult to meet the practical application needs of island and reef mapping. Therefore, how to improve the accuracy of 3D real-scene mapping of islands and reefs has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a method for three-dimensional real-scene mapping of islands and reefs based on unmanned equipment, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for three-dimensional real-scene mapping of islands and reefs based on unmanned equipment, comprising:

[0006] S1. Perform geometric analysis on the high-resolution satellite remote sensing image of the target island to obtain a general map of the survey area of ​​the target island, and set up ground control points on the target island according to the general map of the survey area to obtain high-precision geographic coordinate data of the ground control points.

[0007] S2. Based on the survey area outline map, formulate an adaptive flight path for the unmanned equipment to collect multi-angle raw image data of the target islands and reefs;

[0008] S3. Perform uniform light and color processing on the multi-angle original image data to obtain a standardized image sequence of the target island and reef, and extract features from the standardized image sequence to obtain the feature point set of the standardized image sequence.

[0009] S4. Perform multi-baseline multi-view matching on the feature point set to obtain the corresponding point matching relationship between the standardized image sequences. Construct a connection point network of the standardized image sequences with the feature point set as nodes and the corresponding point matching relationship as edges.

[0010] S5. Using the high-precision geographic coordinate data as a constraint, substitute it into the connection point network to perform bundle adjustment of the joint regional network to obtain the high-precision exterior orientation elements and densification point coordinates of the standardized image sequence.

[0011] S6. Based on the high-precision exterior orientation elements, and using the encrypted point coordinates as seed points to constrain the search range, perform pixel-by-pixel depth estimation on the standardized image sequence to obtain the three-dimensional point cloud data of the target island / reef;

[0012] S7. Construct an irregular triangular mesh from the three-dimensional point cloud data, and perform texture mapping on the surface of the triangular mesh to obtain a measurable three-dimensional real-world model of the target island / reef.

[0013] In a preferred embodiment, the step of performing geometric analysis on the high-resolution satellite remote sensing image of the target island / reef to obtain a rough map of the survey area of ​​the target island / reef includes:

[0014] Acquire high-resolution satellite stereo image pairs covering the target islands and reefs, eliminate geometric distortions in the high-resolution satellite stereo image pairs, and obtain epipolar image pairs of the target islands and reefs;

[0015] Dense matching is performed on the epipolar image pairs to obtain the digital surface model of the target island / reef. The digital surface model is then filtered to obtain the initial elevation data of the target island / reef.

[0016] Multi-scale segmentation is performed on the panchromatic image in the epipolar image pair to extract the lithological texture boundary of the target island reef, and combined with the initial elevation data, the stable exposed rock mass area on the target island reef is identified.

[0017] The stable exposed rock mass area is processed into a grid to obtain a rough map of the survey area of ​​the target island reef.

[0018] In a preferred embodiment, the step of formulating an adaptive flight path for the unmanned equipment based on the survey area outline map to collect multi-angle raw image data of the target islands and reefs includes:

[0019] Read the digital surface model data contained in the survey area outline map, resample the digital surface model data, and obtain the rasterized elevation matrix of the target island / reef;

[0020] Traverse the grid cells in the rasterized elevation matrix to obtain the elevation difference between the grid cells, and determine the slope steepness coefficient of the target island / reef based on the elevation difference.

[0021] Based on the slope gradient coefficient, the island and reef areas in the survey area outline map are adaptively partitioned to obtain the refined collection area and the conventional collection area of ​​the target island and reef.

[0022] Differentiated trajectory planning is performed on the refined data collection area and the conventional data collection area to obtain the three-dimensional route planning for the target island / reef;

[0023] The three-dimensional flight path plan is imported into the flight control system of the unmanned equipment to generate the exposure trigger command and gimbal angle adjustment command of the unmanned equipment.

[0024] Based on the exposure trigger command and the gimbal angle adjustment command, the camera tilt angle of the unmanned equipment is adjusted during flight to collect multi-angle raw image data covering the target island reef.

[0025] In a preferred embodiment, the step of performing uniform illumination and color processing on the multi-angle original image data to obtain a standardized image sequence of the target island / reef, and extracting features from the standardized image sequence to obtain a feature point set of the standardized image sequence, includes:

[0026] Extract the global grayscale mean and grayscale standard deviation of the original image from the multi-angle original image data to obtain the brightness statistical characteristics of the multi-angle original image data;

[0027] The image covering the main area of ​​the island and reef and with uniform illumination in the multi-angle original image data is used as the reference image. The brightness statistical characteristics of the reference image are used as a reference to perform pixel-by-pixel grayscale linear transformation on the remaining original images to obtain the initial corrected image sequence of the target island and reef.

[0028] The initial calibrated image sequence is divided into grids to obtain the local average brightness of the initial calibrated image sequence;

[0029] Based on the local brightness mean, gamma correction is performed on the pixels of the images in the initial corrected image sequence to obtain the standardized image sequence of the target island.

[0030] The three-dimensional local extreme points of the images in the standardized image sequence are used as candidate feature points, and the candidate feature points are precisely located at the sub-pixel level. Unstable points located in low-contrast regions and edge regions are eliminated to obtain stable feature points of the standardized image sequence.

[0031] Using the stable feature point as the center, a neighborhood window is defined, and the gradient direction distribution of pixels within the neighborhood window is statistically analyzed to obtain the main direction of the stable feature point.

[0032] In the main direction, the neighborhood window is rotated and normalized, and the gradient direction distribution is statistically analyzed in blocks to obtain the feature descriptor of the stable feature point;

[0033] Stable feature points, including location, scale, main direction, and feature descriptor, are aggregated into the feature point set of the standardized image sequence.

[0034] In a preferred embodiment, the step of performing multi-baseline, multi-view matching on the feature point set to obtain corresponding point matching relationships between the standardized image sequences, and constructing a connection point network of the standardized image sequences using the feature point set as nodes and the corresponding point matching relationships as edges, includes:

[0035] The image with the largest image overlap area is selected from the standardized image sequence as the matching basis set, and the feature points of the matching basis set are extracted from the feature point set as the set of points to be matched.

[0036] Based on the feature descriptors in the set of points to be matched, an approximate nearest neighbor search is performed on the matching basis set to obtain the initial matching pair set of the normalized image sequence;

[0037] Consistency verification is performed on the corresponding points in the initial matching pair set to obtain the corresponding point matching relationship of the standardized image sequence;

[0038] Using the corresponding point matching relationship as a seed, and through the adjacency relationship between images, corresponding point tracking is extended to other images in the standardized image sequence to obtain the dense corresponding point trajectory of the standardized image sequence.

[0039] The feature points in the dense corresponding point trajectory are abstracted as nodes, and the correspondence between corresponding feature points on the image in the dense corresponding point trajectory is abstracted as edges, thus constructing the connection point network of the standardized image sequence.

[0040] In a preferred embodiment, the step of substituting the high-precision geographic coordinate data as constraints into the connect point network for bundled joint regional network adjustment to obtain the high-precision exterior orientation elements and densification point coordinates of the standardized image sequence includes:

[0041] Based on the connection relationships of the nodes in the connection point network, the initial exterior orientation elements of the images in the standardized image sequence are extracted, and the image point coordinate residuals of the connection points in the connection point network on different images are obtained.

[0042] Based on the high-precision geographic coordinate data, coordinate constraints are constructed for the ground control points, and the weight matrix configuration of the bundle adjustment method combined with the regional network is adjusted according to the coordinate constraints.

[0043] The image point coordinate residuals and the adjusted weight matrix are substituted into the bundle joint regional network adjustment to update the exterior orientation elements of the standardized image sequence and the ground coordinates of the connection points;

[0044] The exterior orientation elements and the ground coordinates are iteratively updated to obtain the high-precision exterior orientation elements and densification point coordinates of the standardized image sequence.

[0045] In a preferred embodiment, the step of performing pixel-by-pixel depth estimation on the standardized image sequence based on the high-precision exterior orientation elements, using the encrypted point coordinates as seed points to constrain the search range, to obtain the three-dimensional point cloud data of the target island / reef includes:

[0046] Based on the high-precision exterior orientation elements, the coordinates of the encrypted point are back-projected onto the image data in the standardized image sequence to obtain the image point position and seed depth of the encrypted point on the image.

[0047] A reference image is selected from the standardized image sequence. A matching window is defined with the coordinates of the pixel to be estimated in the reference image as the center. The depth search range and candidate depth values ​​of the standardized image sequence are determined according to the distribution of the seed depth in the neighborhood of the pixel to be estimated.

[0048] Based on the high-precision exterior orientation elements, the object point corresponding to the candidate depth value is projected onto a neighboring image that overlaps with the reference image, and the coordinates of the projected point on the neighboring image are obtained.

[0049] Extract the grayscale value sequence of the pixels within the matching window in the neighborhood image, and determine the photometric consistency measure of the candidate depth value in the neighborhood image based on the grayscale value sequence;

[0050] The aggregation cost of the candidate depth values ​​is obtained by fusing the photometric consistency metric.

[0051] The candidate depth value with the lowest aggregation cost is determined as the depth value of the pixel to be estimated;

[0052] By combining the depth value and the high-precision exterior orientation elements, and back-projecting them into the object space through collinearity, the three-dimensional point cloud data of the target island / reef is obtained.

[0053] In a preferred embodiment, the aggregation cost is calculated using the following formula:

[0054] ;

[0055] in, This represents the aggregation cost. This represents the candidate depth value. This represents the set of neighboring images. This indicates the preset viewpoint weights. Represented in pixels The matching window centered on the center, Represents pixels on the reference image In color channels grayscale values ​​on Representing neighboring images Upper projection point In color channels grayscale values ​​on The preset smoothness constraint weights, For pixels The neighborhood pixel set, Represents neighboring pixels The depth estimate.

[0056] In a preferred embodiment, the step of constructing an irregular triangular mesh from the 3D point cloud data and performing texture mapping on the surface of the triangular mesh to obtain a measurable 3D reality model of the target island / reef includes:

[0057] Morphological filtering is performed on the three-dimensional point cloud data to obtain clean three-dimensional point cloud data of the target island / reef;

[0058] The clean 3D point cloud data is subjected to adaptive simplification based on curvature features. High-density point clouds are retained in areas with drastic curvature changes, while point clouds are thinned in areas with gentle curvature, resulting in a non-uniform density point cloud of the target island reef.

[0059] The non-uniform density point cloud is projected onto a two-dimensional plane, and the projected point cloud is triangulated in two dimensions to obtain the triangular network topology of the target island reef.

[0060] The triangular mesh topology is mapped back to three-dimensional space, and a spatial triangular mesh is constructed based on the three-dimensional coordinates of the three-dimensional point cloud data. The narrow triangles and intersecting facets in the spatial triangular mesh are detected.

[0061] The elongated triangle is optimized by edge swapping, and the intersecting facets are locally subdivided to obtain the irregular triangular network of the target island / reef.

[0062] Texture mapping is performed on the irregular triangular mesh to obtain a measurable three-dimensional reality model of the target island reef.

[0063] In a preferred embodiment, the step of texture mapping the irregular triangular mesh to obtain a measurable 3D reality model of the target island / reef includes:

[0064] Candidate images covering the irregular triangular mesh are selected from the standardized image sequence. Based on the imaging tilt angle and image sharpness of the candidate images, the comprehensive visibility score between the irregular triangular mesh and the candidate images is determined.

[0065] Based on the comprehensive visibility score, the optimal texture image is selected for the irregular triangular mesh using a Markov random field optimization model, and the texture seam position of the irregular triangular mesh is determined.

[0066] The optimal texture image is projected and transformed to obtain the initial texture block of the irregular triangular mesh. At the texture seam position, the initial texture block is subjected to multi-band fusion processing to obtain the fused texture block of the irregular triangular mesh.

[0067] Global color consistency adjustment is performed on the fused texture block to obtain the texture mapping result of the irregular triangular mesh;

[0068] The texture mapping result is applied to the irregular triangular mesh to obtain a measurable three-dimensional reality model of the target island / reef.

[0069] Compared with the prior art, the present invention has the following beneficial effects:

[0070] 1. This invention utilizes the dual foundations of high-resolution satellite remote sensing image analysis and ground control point deployment to plan adaptive flight routes for unmanned equipment, enabling precise acquisition of original images of islands and reefs from multiple angles. Through a series of standardized processes such as uniform illumination and color matching, feature extraction, and multi-baseline multi-view matching, a precise network of connecting points is constructed. Combined with high-precision geographic coordinate constraints, bundle adjustment is performed using the bundle method, significantly improving the accuracy of image exterior orientation elements and densified point coordinates. Based on this, pixel-by-pixel depth estimation accurately acquires 3D point cloud data of islands and reefs. The entire process from data acquisition to processing achieves improved accuracy, significantly enhancing the overall accuracy of 3D real-world mapping of islands and reefs.

[0071] 2. This invention improves the image acquisition efficiency of unmanned equipment by adaptively partitioning and differentiating the trajectory planning of island and reef terrain features. At the same time, in the 3D point cloud processing stage, point cloud optimization is completed through morphological filtering and adaptive simplification based on curvature features. Then, through irregular triangular mesh construction and refined texture mapping, a measurable 3D real scene model is efficiently generated. The technical design of the whole process realizes the efficiency optimization of each stage of surveying and mapping, effectively improving the overall operational efficiency of 3D real scene surveying and mapping of islands and reefs. Moreover, the final generated 3D real scene model has measurable attributes, further enhancing the practical value of surveying and mapping results. Attached Figure Description

[0072] Figure 1 This is a flowchart illustrating a method for three-dimensional real-scene mapping of islands and reefs based on unmanned equipment, provided in an embodiment of the present invention.

[0073] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0074] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0075] This application provides a method for 3D real-scene mapping of islands and reefs based on unmanned equipment. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method for 3D real-scene mapping of islands and reefs based on unmanned equipment can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0076] Reference Figure 1 The diagram shown is a flowchart illustrating a method for three-dimensional real-scene mapping of islands and reefs based on unmanned equipment, according to an embodiment of the present invention. In this embodiment, the method for three-dimensional real-scene mapping of islands and reefs based on unmanned equipment includes:

[0077] S1. Perform geometric analysis on the high-resolution satellite remote sensing image of the target island to obtain a general map of the survey area of ​​the target island, and set up ground control points on the target island according to the general map of the survey area to obtain high-precision geographic coordinate data of the ground control points.

[0078] In this embodiment of the invention, the step of performing geometric analysis on the high-resolution satellite remote sensing image of the target island / reef to obtain a rough map of the survey area of ​​the target island / reef includes:

[0079] Acquire high-resolution satellite stereo image pairs covering the target islands and reefs, eliminate geometric distortions in the high-resolution satellite stereo image pairs, and obtain epipolar image pairs of the target islands and reefs;

[0080] Dense matching is performed on the epipolar image pairs to obtain the digital surface model of the target island / reef. The digital surface model is then filtered to obtain the initial elevation data of the target island / reef.

[0081] Multi-scale segmentation is performed on the panchromatic image in the epipolar image pair to extract the lithological texture boundary of the target island reef, and combined with the initial elevation data, the stable exposed rock mass area on the target island reef is identified.

[0082] The stable exposed rock mass area is processed into a grid to obtain a rough map of the survey area of ​​the target island reef.

[0083] High-resolution satellite stereo image pairs that can fully cover the geographical area of ​​the target islands and reefs are acquired through satellite imagery acquisition channels. Based on the internal and external orientation elements of satellite imaging and the imaging deviation rules caused by topographic relief, the geometric distortions such as tilt distortion, projection distortion, and scaling distortion in the high-resolution satellite stereo image pairs are corrected pixel by pixel. After correction, the epipolar image pairs of the target islands and reefs that retain only the epipolar direction image information are obtained.

[0084] A pixel-by-pixel search and matching operation is performed on the epipolar image pairs. The successfully matched corresponding points are converted into object space coordinates by combining the imaging geometry. Based on these object space coordinates, a digital surface model that can reflect the topographic relief of the target island reef is constructed. The digital surface model is filtered by the statistical method of neighborhood pixel gray values ​​to remove abnormal elevation points caused by image noise in the model. Finally, the initial elevation data that can reflect the basic topographic elevation of the target island reef is obtained.

[0085] Multi-scale image segmentation was performed on the panchromatic images in the epipolar image pair, from large to small. The contour edge information of lithological texture in the images was extracted at different scales. The contour edge information of lithological texture at different scales was fused to accurately extract the lithological texture boundary of the target island. The extracted lithological texture boundary was spatially superimposed and matched with the initial elevation data. Based on the stability of the elevation data and the integrity of the lithological texture, stable exposed rock mass areas without vegetation cover and loose deposits were identified within the geographical range of the target island.

[0086] The identified stable exposed rock mass area is divided into regular grids according to a fixed spatial size. Each grid cell is assigned a corresponding spatial geographic coordinate and lithological elevation attribute. The grid cells with complete attribute information are combined into a whole to obtain a general map of the target island reef survey area.

[0087] The beneficial effects of this implementation process are that by performing pixel-by-pixel geometric distortion correction on high-resolution satellite stereo image pairs, the imaging quality and spatial accuracy of epipolar image pairs are guaranteed from the source. Based on the dense matching of epipolar image pairs, a digital surface model is constructed and filtered, so that the initial elevation data can truly and accurately reflect the topographic elevation characteristics of the target islands and reefs. By combining the lithological texture boundary extracted by multi-scale segmentation with the initial elevation data for spatial overlay analysis, stable exposed rock areas on the islands and reefs suitable for setting up control points can be accurately identified. Then, through gridding processing, standardized spatial attribute information is given to the area. The final survey area outline map has the characteristics of high accuracy and strong targeting, which can provide detailed and reliable basic geographic reference for the subsequent accurate deployment of ground control points and the scientific planning of adaptive flight routes for unmanned equipment. This lays a solid foundation for the accuracy of three-dimensional real-scene surveying of islands and reefs from the early stage of surveying and mapping work, and improves the targeting and effectiveness of subsequent surveying and mapping operations.

[0088] S2. Based on the survey area outline map, formulate an adaptive flight path for the unmanned equipment to collect multi-angle raw image data of the target islands and reefs;

[0089] In this embodiment of the invention, the step of formulating an adaptive flight path for the unmanned equipment based on the survey area outline map to collect multi-angle raw image data of the target islands and reefs includes:

[0090] Read the digital surface model data contained in the survey area outline map, resample the digital surface model data, and obtain the rasterized elevation matrix of the target island / reef;

[0091] Traverse the grid cells in the rasterized elevation matrix to obtain the elevation difference between the grid cells, and determine the slope steepness coefficient of the target island / reef based on the elevation difference.

[0092] Based on the slope gradient coefficient, the island and reef areas in the survey area outline map are adaptively partitioned to obtain the refined collection area and the conventional collection area of ​​the target island and reef.

[0093] Differentiated trajectory planning is performed on the refined data collection area and the conventional data collection area to obtain the three-dimensional route planning for the target island / reef;

[0094] The three-dimensional flight path plan is imported into the flight control system of the unmanned equipment to generate the exposure trigger command and gimbal angle adjustment command of the unmanned equipment.

[0095] Based on the exposure trigger command and the gimbal angle adjustment command, the camera tilt angle of the unmanned equipment is adjusted during flight to collect multi-angle raw image data covering the target island reef.

[0096] The built-in digital surface model data is retrieved from the geographic attribute data of the survey area outline map. The digital surface model data is resampled according to the preset grid size, and the continuous elevation surface is converted into discrete elevation data with grid cells as the basic unit. Each grid cell corresponds to a unique elevation value. The elevation values ​​of all grid cells are arranged according to their spatial location, and finally the rasterized elevation matrix of the target island is obtained.

[0097] The system iterates through all grid cells in the rasterized elevation matrix in row and column order, calculates the elevation difference between adjacent grid cells, calculates the terrain slope around each grid cell based on the elevation difference and the actual spatial spacing of the grid cells, and then calculates the slope steepness coefficient that reflects the degree of abrupt change in the terrain slope of the island and reef based on the change range of the terrain slope value, thus completing the determination of the overall slope steepness coefficient of the target island and reef.

[0098] Using the slope steepness coefficient as a zoning criterion, a corresponding coefficient judgment threshold is set, and the coefficients of the island and reef areas in the survey area outline map are compared region by region. Areas with slope steepness coefficients exceeding the judgment threshold and large topographic relief are designated as refined collection areas, while areas with slope steepness coefficients not exceeding the judgment threshold and gentle topographic relief are designated as regular collection areas, thus completing the adaptive zoning of the target island and reef.

[0099] Different flight path planning standards are formulated for the refined data collection area and the regular data collection area. For the refined data collection area, a planning method of increasing the density of waypoints and reducing the spacing of flight strips is adopted, while for the regular data collection area, a planning method of using regular waypoints and standard spacing of flight strips is adopted. At the same time, the flight altitude is designed in combination with the overall topography and elevation of the island and reef, and the flight paths of the two areas are seamlessly connected and integrated as a whole to obtain the three-dimensional flight path planning of the target island and reef.

[0100] All data information from the integrated 3D flight path planning is imported into the flight control system of the unmanned equipment. The system automatically generates exposure trigger commands for each waypoint based on the spatial location, flight altitude, and image acquisition requirements of each waypoint on the flight path. At the same time, it generates gimbal angle adjustment commands adapted to the flight status of each flight segment based on the orientation and acquisition angle requirements of the island and reef terrain.

[0101] The unmanned equipment flies along the preset route of the flight control system. During the flight, it receives and executes exposure trigger commands and gimbal angle adjustment commands in real time. Based on the commands, it precisely adjusts the tilt angle of the camera and completes the precise exposure acquisition of images at each preset waypoint. This achieves full coverage of the refined acquisition area and the conventional acquisition area of ​​the target island and reef, and finally acquires multi-angle raw image data of the target island and reef.

[0102] The beneficial effects are that this implementation process obtains a rasterized elevation matrix by resampling the digital surface model data of the survey area's general map, providing a standardized data foundation for terrain analysis. The slope steepness coefficient determined based on the elevation difference can accurately reflect the topographic undulation characteristics of the islands and reefs. Based on this, adaptive zoning makes image acquisition more targeted. Differentiated trajectory planning for different areas, combined with precise command generation and execution, enables unmanned equipment to complete the acquisition of multi-angle raw image data at the optimal angle. This ensures the image acquisition density and accuracy in complex terrain areas while avoiding resource waste in conventional areas, significantly improving the efficiency and targeting of unmanned equipment image acquisition, and laying a high-quality data foundation for subsequent island and reef image data processing.

[0103] S3. Perform uniform light and color processing on the multi-angle original image data to obtain a standardized image sequence of the target island and reef, and extract features from the standardized image sequence to obtain the feature point set of the standardized image sequence.

[0104] In this embodiment of the invention, the step of performing uniform illumination and color processing on the multi-angle original image data to obtain a standardized image sequence of the target island / reef, and extracting features from the standardized image sequence to obtain a feature point set of the standardized image sequence, includes:

[0105] Extract the global grayscale mean and grayscale standard deviation of the original image from the multi-angle original image data to obtain the brightness statistical characteristics of the multi-angle original image data;

[0106] The image covering the main area of ​​the island and reef and with uniform illumination in the multi-angle original image data is used as the reference image. The brightness statistical characteristics of the reference image are used as a reference to perform pixel-by-pixel grayscale linear transformation on the remaining original images to obtain the initial corrected image sequence of the target island and reef.

[0107] The initial calibrated image sequence is divided into grids to obtain the local average brightness of the initial calibrated image sequence;

[0108] Based on the local brightness mean, gamma correction is performed on the pixels of the images in the initial corrected image sequence to obtain the standardized image sequence of the target island.

[0109] The three-dimensional local extreme points of the images in the standardized image sequence are used as candidate feature points, and the candidate feature points are precisely located at the sub-pixel level. Unstable points located in low-contrast regions and edge regions are eliminated to obtain stable feature points of the standardized image sequence.

[0110] Using the stable feature point as the center, a neighborhood window is defined, and the gradient direction distribution of pixels within the neighborhood window is statistically analyzed to obtain the main direction of the stable feature point.

[0111] In the main direction, the neighborhood window is rotated and normalized, and the gradient direction distribution is statistically analyzed in blocks to obtain the feature descriptor of the stable feature point;

[0112] Stable feature points, including location, scale, main direction, and feature descriptor, are aggregated into the feature point set of the standardized image sequence.

[0113] For each original image in the multi-angle original image data, the gray value is statistically analyzed pixel by pixel. The global gray value is obtained by calculating the average value of all pixel gray values. The gray standard deviation is obtained by calculating the square root of the sum of squares of the deviations of all pixel gray values ​​from the global gray value. The global gray value and gray standard deviation of each original image are integrated to form the brightness statistical characteristics of the multi-angle original image data.

[0114] From the multi-angle raw image data, images that can completely cover the main geographical area of ​​the island and reef, and whose light intensity is uniform in each area without obvious overexposure or underexposure areas are selected as reference images. The brightness statistical features of the reference images are extracted as a correction reference. Pixel-by-pixel grayscale linear transformation is performed on all other raw images to adjust the grayscale value of each pixel to match the brightness features of the reference images. All images after grayscale adjustment are arranged in the order of acquisition to obtain the initial corrected image sequence of the target island and reef.

[0115] Each image in the initial calibrated image sequence is divided into a regular grid of equal size, so that a single image is divided into multiple independent grid regions. The average gray value of all pixels in each grid region is calculated, and the calculation results corresponding to each grid region are integrated to obtain the local brightness average value of the initial calibrated image sequence.

[0116] Based on the average local brightness of the initial image sequence, gamma correction is performed on each pixel of each image in the initial image sequence to adjust the grayscale response of pixels in different brightness areas, so that the local brightness of the image is consistent with the overall brightness. All images after gamma correction are arranged in the original order to obtain the standardized image sequence of the target island.

[0117] In each image of the standardized image sequence, three-dimensional local extrema points in the image space are retrieved and used as candidate feature points. Sub-pixel precise positioning of the spatial coordinates of the candidate feature points is performed by pixel interpolation. The contrast and spatial position of the region where the candidate feature points are located are detected one by one. Unstable points located in low-contrast areas and physical edge areas of the image are eliminated, and effective points with clear features are retained to obtain stable feature points of the standardized image sequence.

[0118] Using each stable feature point as the geometric center, a square neighborhood window of fixed size is defined in the image. The gray-level change trend of all pixels in the neighborhood window is calculated, the gradient direction distribution of each pixel in the window is statistically analyzed, and the gradient direction with the highest distribution ratio is selected as the main direction of the stable feature point.

[0119] The neighborhood window is rotated and adjusted around the stable feature point along the main direction of the feature point to complete the rotation normalization process. The normalized neighborhood window is divided into multiple sub-blocks of equal size. The pixel gradient direction distribution in each sub-block is statistically analyzed. The gradient direction distribution features of all sub-blocks are integrated into a set of feature information to obtain the feature descriptor of the stable feature point.

[0120] The spatial location information, scale information, main direction information, and feature descriptor corresponding to each stable feature point are extracted. The above four types of information of all stable feature points are systematically collected and organized, and arranged in order according to image ownership and feature point number to finally obtain the feature point set of the standardized image sequence.

[0121] The beneficial effects of this implementation process are that, through phased light and color homogenization, first, a linear correction of global brightness is completed using a reference image, and then gamma correction is performed in combination with the local brightness mean. This achieves dual optimization of global and local image brightness, effectively eliminating brightness deviations caused by differences in illumination in the original images from multiple angles. The resulting standardized image sequence has uniform brightness characteristics. The stable feature points obtained by subsequent screening based on three-dimensional local extreme points and through precise positioning and removal of unstable points have high recognition and reliability. Through a series of operations such as defining neighborhood windows, determining the main direction, and generating feature descriptors, the feature information of the feature points is made more complete. The final feature point set can accurately reflect the image characteristics of the standardized image sequence, providing a high-precision and high-matching feature foundation for subsequent image matching and connection point network construction, and significantly improving the accuracy and efficiency of subsequent image processing steps.

[0122] S4. Perform multi-baseline multi-view matching on the feature point set to obtain the corresponding point matching relationship between the standardized image sequences. Construct a connection point network of the standardized image sequences with the feature point set as nodes and the corresponding point matching relationship as edges.

[0123] In this embodiment of the invention, the step of performing multi-baseline, multi-view matching on the feature point set to obtain the corresponding point matching relationships between the standardized image sequences, and constructing a connection point network of the standardized image sequences using the feature point set as nodes and the corresponding point matching relationships as edges, includes:

[0124] The image with the largest image overlap area is selected from the standardized image sequence as the matching basis set, and the feature points of the matching basis set are extracted from the feature point set as the set of points to be matched.

[0125] Based on the feature descriptors in the set of points to be matched, an approximate nearest neighbor search is performed on the matching basis set to obtain the initial matching pair set of the normalized image sequence;

[0126] Consistency verification is performed on the corresponding points in the initial matching pair set to obtain the corresponding point matching relationship of the standardized image sequence;

[0127] Using the corresponding point matching relationship as a seed, and through the adjacency relationship between images, corresponding point tracking is extended to other images in the standardized image sequence to obtain the dense corresponding point trajectory of the standardized image sequence.

[0128] The feature points in the dense corresponding point trajectory are abstracted as nodes, and the correspondence between corresponding feature points on the image in the dense corresponding point trajectory is abstracted as edges, thus constructing the connection point network of the standardized image sequence.

[0129] From all images in the standardized image sequence, the pixel overlap area between any two images is compared one by one. The image combination with the largest proportion of pixel overlap area in all image combinations is selected and determined as the matching base set. Then, from the feature point set of the standardized image sequence, all feature points corresponding to each image in the matching base set are extracted and integrated into a set of points to be matched.

[0130] For each feature point in the set of matching points, information about the feature descriptor is extracted. Based on the similarity measurement rules of the feature descriptor, an approximate nearest neighbor search is performed on the images in the matching basis set. The feature point that is most similar to the feature descriptor of the current feature point is found in another image. The feature points that are successfully matched in pairs are combined to form the initial matching pair set of the standardized image sequence.

[0131] Based on the imaging geometry and spatial location association rules of the images, the geometric consistency and spatial correlation of each pair of corresponding points in the initial matching pair set are checked. Incorrect matching pairs caused by misjudgment of feature descriptor similarity are eliminated, and the correct pairs of corresponding points that pass the check are retained, thereby determining the corresponding point matching relationship of the standardized image sequence.

[0132] Using the verified corresponding point matching relationships as basic seed data, we sort out the spatial adjacency and shooting temporal adjacency relationships between all images in the standardized image sequence. Taking the corresponding points in the seed data as clues, we continuously search for corresponding corresponding feature points in adjacent images, gradually complete the corresponding point matching expansion of all images, and finally obtain dense corresponding point trajectories covering all images in the standardized image sequence.

[0133] All feature points contained in the dense homonymous point trajectory are abstracted, and each feature point is treated as an independent node. The association between homonymous feature points with corresponding matching relationships on different images in the dense homonymous point trajectory is abstracted, and each set of corresponding relationships is treated as an edge connecting the nodes. All the abstracted nodes and edges are combined and constructed according to the association logic of image matching, and finally a connection point network of standardized image sequence is constructed.

[0134] The beneficial effects are that this implementation process uses images of the largest overlapping area as the matching basis, which makes the initial matching more successful and accurate. The approximate nearest neighbor search based on feature descriptors realizes the rapid matching of feature points. The corresponding point matching relationship verified by consistency eliminates the interference of erroneous matching. The corresponding point tracking expansion based on this relationship realizes dense corresponding point matching of the entire sequence of images. Finally, the connection point network constructed by the abstraction of nodes and edges has a clear structure and accurate correlation, which can fully reflect the feature point matching correlation of the standardized image sequence. It provides an accurate and complete topological relationship foundation for subsequent bundle adjustment and joint regional network adjustment, effectively improving the accuracy and efficiency of subsequent adjustment calculations.

[0135] S5. Using the high-precision geographic coordinate data as a constraint, substitute it into the connection point network to perform bundle adjustment of the joint regional network to obtain the high-precision exterior orientation elements and densification point coordinates of the standardized image sequence.

[0136] In this embodiment of the invention, the step of substituting the high-precision geographic coordinate data as a constraint into the connective point network for bundle joint regional network adjustment to obtain the high-precision exterior orientation elements and densification point coordinates of the standardized image sequence includes:

[0137] Based on the connection relationships of the nodes in the connection point network, the initial exterior orientation elements of the images in the standardized image sequence are extracted, and the image point coordinate residuals of the connection points in the connection point network on different images are obtained.

[0138] Based on the high-precision geographic coordinate data, coordinate constraints are constructed for the ground control points, and the weight matrix configuration of the bundle adjustment method combined with the regional network is adjusted according to the coordinate constraints.

[0139] The image point coordinate residuals and the adjusted weight matrix are substituted into the bundle joint regional network adjustment to update the exterior orientation elements of the standardized image sequence and the ground coordinates of the connection points;

[0140] The exterior orientation elements and the ground coordinates are iteratively updated to obtain the high-precision exterior orientation elements and densification point coordinates of the standardized image sequence.

[0141] Based on the matching connection relationship between nodes in the connection point network and combined with the spatial geometric correlation law of image imaging, the initial exterior orientation element corresponding to each image in the standardized image sequence is extracted. At the same time, the difference between the observed and calculated values ​​of the image point coordinates of each connection point in the connection point network on different images is calculated, and the image point coordinate residuals corresponding to all connection points are obtained in full.

[0142] High-precision geographic coordinate data corresponding to ground control points are retrieved. Based on the spatial coordinate accuracy and geographic reference attributes of this data, spatial coordinate constraints for ground control points are established, and the fixed range of coordinate values ​​of ground control points in object space is defined. Then, according to the constraint strength of the coordinate constraints, the element values ​​of the weight matrix corresponding to the bundle adjustment method and the regional network adjustment are adjusted to make the weight matrix compatible with the coordinate constraints.

[0143] All the acquired image point coordinate residual data and the adjusted weight matrix are substituted into the calculation system of bundle adjustment for joint regional network adjustment. According to the spatial geometric solution rules of adjustment, the original exterior orientation elements in the standardized image sequence are corrected and updated. At the same time, the ground coordinates corresponding to each connection point in the connection point network are recalculated and updated to obtain the updated exterior orientation elements and ground coordinates.

[0144] The updated exterior orientation elements and ground coordinates are then substituted back into the bundle adjustment calculation system for joint regional network adjustment. The operations of calculating image point coordinate residuals and correcting exterior orientation elements and ground coordinates are repeated. The exterior orientation elements and ground coordinates of the tie points of the standardized image sequence are iteratively updated until the correction values ​​of the exterior orientation elements and ground coordinates reach a stable state. Finally, high-precision exterior orientation elements and densification point coordinates of the standardized image sequence are obtained.

[0145] The beneficial effects are that the implementation process accurately extracts the initial exterior orientation elements and image point coordinate residuals based on the node connection relationship of the connection point network, laying a precise foundation of data for adjustment calculation. The constraints constructed with high-precision geographic coordinate data, combined with the appropriate weight matrix configuration, give the adjustment calculation strong geographic reference constraints. Through multiple iterations and updates, the exterior orientation elements and ground coordinates are continuously optimized, effectively eliminating calculation errors. The final high-precision exterior orientation elements and densified point coordinates have high accuracy and strong stability, which can provide accurate spatial geometric parameters for subsequent pixel-by-pixel depth estimation, ensuring the solution accuracy of the island and reef 3D point cloud data from the core parameter level.

[0146] S6. Based on the high-precision exterior orientation elements, and using the encrypted point coordinates as seed points to constrain the search range, perform pixel-by-pixel depth estimation on the standardized image sequence to obtain the three-dimensional point cloud data of the target island / reef;

[0147] In this embodiment of the invention, the step of performing pixel-by-pixel depth estimation on the standardized image sequence based on the high-precision exterior orientation elements, using the encrypted point coordinates as seed points to constrain the search range, to obtain the three-dimensional point cloud data of the target island / reef includes:

[0148] Based on the high-precision exterior orientation elements, the coordinates of the encrypted point are back-projected onto the image data in the standardized image sequence to obtain the image point position and seed depth of the encrypted point on the image.

[0149] A reference image is selected from the standardized image sequence. A matching window is defined with the coordinates of the pixel to be estimated in the reference image as the center. The depth search range and candidate depth values ​​of the standardized image sequence are determined according to the distribution of the seed depth in the neighborhood of the pixel to be estimated.

[0150] Based on the high-precision exterior orientation elements, the object point corresponding to the candidate depth value is projected onto a neighboring image that overlaps with the reference image, and the coordinates of the projected point on the neighboring image are obtained.

[0151] Extract the grayscale value sequence of the pixels within the matching window in the neighborhood image, and determine the photometric consistency measure of the candidate depth value in the neighborhood image based on the grayscale value sequence;

[0152] The aggregation cost of the candidate depth values ​​is obtained by fusing the photometric consistency metric.

[0153] The candidate depth value with the lowest aggregation cost is determined as the depth value of the pixel to be estimated;

[0154] By combining the depth value and the high-precision exterior orientation elements, and back-projecting them into the object space through collinearity, the three-dimensional point cloud data of the target island / reef is obtained.

[0155] The formula for calculating the aggregation cost is as follows:

[0156] ;

[0157] in, This represents the aggregation cost. This represents the candidate depth value. This represents the set of neighboring images. This indicates the preset viewpoint weights. Represented in pixels The matching window centered on the center, Represents pixels on the reference image In color channels grayscale values ​​on Representing neighboring images Upper projection point In color channels grayscale values ​​on The preset smoothness constraint weights, For pixels The neighborhood pixel set, Represents neighboring pixels The depth estimate.

[0158] Based on the image imaging geometric parameters corresponding to the high-precision exterior orientation elements, the coordinates of the encryption points are projected onto each image data of the standardized image sequence according to the back projection rule from object space to image space, so as to accurately determine the two-dimensional image point position of each encryption point on the corresponding image, and at the same time, the object space elevation information corresponding to the encryption point is extracted as the seed depth.

[0159] Images with high imaging clarity and wide coverage are selected from the standardized image sequence as reference images. Using the two-dimensional coordinates of each pixel to be estimated in the reference image as the geometric center, a square matching window of fixed size is delineated on the image. The distribution range and numerical changes of the seed depth in the neighborhood of the pixel to be estimated are statistically analyzed. Based on the distribution characteristics, the depth search interval corresponding to the standardized image sequence is delineated, and several values ​​are selected as candidate depth values ​​at fixed intervals within this interval.

[0160] Referring to the spatial projection parameters contained in the high-precision exterior orientation elements, each candidate depth value is converted into a corresponding object point by combining the pixel coordinates to be estimated. Then, the object point is projected onto all neighboring images with pixel overlap with the reference image according to the image-side projection rules. The two-dimensional projection point coordinates of the object point on each neighboring image are accurately obtained through spatial coordinate conversion.

[0161] Using a matching window centered on the pixel to be estimated in the reference image as the range, extract the gray values ​​of all pixels within the window around the corresponding projection points in each neighboring image. Arrange these gray values ​​into an ordered gray value sequence according to the pixel arrangement order. Based on the similarity of the gray value sequences, calculate and determine the photometric consistency measure of each candidate depth value in the corresponding neighboring image.

[0162] The photometric consistency measure corresponding to the same candidate depth value on all neighboring images is aggregated. These measures are then comprehensively calculated according to the preset fusion rules. The calculation result is used as the aggregation cost corresponding to the candidate depth value, and the aggregation cost of all candidate depth values ​​is calculated and determined.

[0163] The set of neighboring images is directly selected from neighboring images that overlap with the reference image; the preset viewpoint weights are fixed values ​​pre-set based on the actual scene of island and reef mapping and the angle characteristics of image acquisition; the pixel-centered matching window is a specific area delineated in the reference image with the coordinates of the pixel to be estimated as the center; the grayscale value of the pixel in the color channel of the reference image is the pixel value directly extracted from the corresponding pixel position and corresponding color channel of the reference image; the grayscale value of the projection point in the neighboring image is the pixel value extracted from the projection point position and corresponding color channel after projecting the object point corresponding to the candidate depth value onto the neighboring image based on the high-precision exterior orientation elements; the preset smoothness constraint weights are fixed values ​​pre-set based on the accuracy requirements of the island and reef 3D point cloud construction; the set of neighboring pixels is a set of pixels around the pixel to be estimated selected from the reference image; the depth estimate of the neighboring pixels is the depth result value corresponding to the neighboring pixels calculated first during the pixel-by-pixel depth estimation process of the standardized image sequence.

[0164] By quantifying the photometric consistency between the reference image and the neighboring image, and combining it with the smoothness constraint of pixel depth for comprehensive calculation, a quantization result that reflects the adaptability of candidate depth values ​​is obtained. Based on this, the optimal depth value of the pixel to be estimated is determined, providing an accurate depth determination basis for obtaining the 3D point cloud data of the target island and reef through collinear back projection, thus ensuring the accuracy and effectiveness of the 3D point cloud data.

[0165] In the photometric consistency calculation, the smaller the sum of the grayscale differences between corresponding pixels in the reference image and the neighboring image, the smaller the corresponding calculation result, and the lower the overall calculation result will be. The preset viewpoint weights will weight the photometric consistency calculation results of the corresponding neighboring image according to the set values. The larger the weight value, the greater the influence on the overall calculation result. In the smoothness constraint calculation, the smaller the sum of the squares of the differences between the candidate depth value and the estimated depth value of the neighboring pixels, the smaller the corresponding calculation result, and the lower the overall calculation result will be. The preset smoothness constraint weights will weight the smoothness constraint calculation results according to the set values. The larger the weight value, the greater the influence on the overall calculation result. The photometric consistency calculation result and the smoothness constraint calculation result are summed. Changes in the values ​​of the two calculation results will directly lead to changes in the overall calculation result in the same direction. The smaller the final overall calculation result value, the higher the matching degree between the corresponding candidate depth value and the actual depth of the pixel to be estimated.

[0166] The aggregated costs of all candidate depth values ​​corresponding to the pixel to be estimated are compared numerically, and the candidate depth value with the smallest aggregated cost value is selected. This value is directly determined as the final depth value of the pixel to be estimated. The determination of the depth values ​​of all pixels to be estimated in the reference image is completed in turn.

[0167] By combining the final depth value of each pixel to be estimated in the reference image with high-precision exterior orientation elements, and based on the collinearity equation relationship of photogrammetry, the image coordinates and depth values ​​of all pixels are back-projected to the object space 3D space, the object space 3D coordinates corresponding to each pixel are calculated, and all object space 3D coordinates are integrated into a point set, finally obtaining the 3D point cloud data of the target island reef.

[0168] The beneficial effects are that this implementation process completes the back projection of encrypted points based on high-precision exterior orientation elements, providing accurate seed depth constraints for depth estimation. By defining the matching window and determining the depth search interval, the search range of depth estimation is made more targeted, effectively improving computational efficiency. Combined with photometric consistency measurement and aggregation cost screening, the depth value of the pixel to be estimated can be accurately determined. The 3D point cloud data obtained by back projection through collinearity can accurately reflect the 3D topographic features of the target island and reef. The spatial accuracy and density of the point cloud data can meet the needs of 3D real-scene mapping of islands and reefs, providing a high-quality 3D data foundation for subsequent irregular triangulation construction and texture mapping.

[0169] S7. Construct an irregular triangular mesh from the three-dimensional point cloud data, and perform texture mapping on the surface of the triangular mesh to obtain a measurable three-dimensional real-world model of the target island / reef.

[0170] In this embodiment of the invention, the step of constructing an irregular triangular mesh from the three-dimensional point cloud data and performing texture mapping on the surface of the triangular mesh to obtain a measurable three-dimensional reality model of the target island / reef includes:

[0171] Morphological filtering is performed on the three-dimensional point cloud data to obtain clean three-dimensional point cloud data of the target island / reef;

[0172] The clean 3D point cloud data is subjected to adaptive simplification based on curvature features. High-density point clouds are retained in areas with drastic curvature changes, while point clouds are thinned in areas with gentle curvature, resulting in a non-uniform density point cloud of the target island reef.

[0173] The non-uniform density point cloud is projected onto a two-dimensional plane, and the projected point cloud is triangulated in two dimensions to obtain the triangular network topology of the target island reef.

[0174] The triangular mesh topology is mapped back to three-dimensional space, and a spatial triangular mesh is constructed based on the three-dimensional coordinates of the three-dimensional point cloud data. The narrow triangles and intersecting facets in the spatial triangular mesh are detected.

[0175] The elongated triangle is optimized by edge swapping, and the intersecting facets are locally subdivided to obtain the irregular triangular network of the target island / reef.

[0176] Texture mapping is performed on the irregular triangular mesh to obtain a measurable three-dimensional reality model of the target island reef.

[0177] The process of performing texture mapping on the irregular triangular mesh to obtain a measurable 3D reality model of the target island / reef includes:

[0178] Candidate images covering the irregular triangular mesh are selected from the standardized image sequence. Based on the imaging tilt angle and image sharpness of the candidate images, the comprehensive visibility score between the irregular triangular mesh and the candidate images is determined.

[0179] Based on the comprehensive visibility score, the optimal texture image is selected for the irregular triangular mesh using a Markov random field optimization model, and the texture seam position of the irregular triangular mesh is determined.

[0180] The optimal texture image is projected and transformed to obtain the initial texture block of the irregular triangular mesh. At the texture seam position, the initial texture block is subjected to multi-band fusion processing to obtain the fused texture block of the irregular triangular mesh.

[0181] Global color consistency adjustment is performed on the fused texture block to obtain the texture mapping result of the irregular triangular mesh;

[0182] The texture mapping result is applied to the irregular triangular mesh to obtain a measurable three-dimensional reality model of the target island / reef.

[0183] Morphological filtering is performed on the 3D point cloud data. The neighborhood traversal of the 3D point cloud data is performed by setting the structuring element to remove isolated noise points, redundant points and abnormal points caused by surveying errors. The effective point cloud data that can truly reflect the topographic features of the target island and reef is retained, and finally clean 3D point cloud data of the target island and reef is obtained.

[0184] Curvature calculation is performed on each point cloud unit in the clean 3D point cloud data to obtain the curvature feature distribution of the entire point cloud data. Based on the magnitude of the curvature value, regions with drastic curvature changes and regions with gentle curvature changes are distinguished. In regions with drastic curvature changes, the original high-density point cloud is completely preserved without processing. In regions with gentle curvature changes, some point cloud units are removed at a fixed ratio to achieve point cloud thinning, and finally, a non-uniform density point cloud of the target island reef is obtained.

[0185] The non-uniform density point cloud is projected onto a two-dimensional plane according to the set projection rules, so that the three-dimensional point cloud data is converted into a two-dimensional plane point set. The two-dimensional plane point set is then subjected to a two-dimensional triangulation operation, which divides the plane point set into multiple interconnected and non-overlapping triangular units. The vertex connection relationship and adjacency relationship of each triangular unit are determined, and the triangular network topology of the target island and reef is obtained.

[0186] The triangular mesh topology obtained on the two-dimensional plane is reverse-mapped back to three-dimensional space according to the original projection rules. Combined with the three-dimensional coordinate information of the non-uniform density point cloud, each triangular unit is assigned a corresponding three-dimensional spatial coordinate to construct the spatial triangular mesh of the target island. The triangular units and faces in the mesh are checked one by one by the spatial geometric detection algorithm to accurately identify the narrow triangles and intersecting faces in the spatial triangular mesh.

[0187] An edge swap optimization operation is performed on the detected elongated triangles to readjust the vertex connection method of the triangles and convert the elongated triangles into regular triangles that conform to geometric norms. A local subdivision operation is performed on the detected intersecting facets to decompose the intersecting facets into independent facets without intersections and reconstruct the triangular connection relationship. After the optimization and adjustment are completed, the irregular triangular network of the target island and reef is obtained.

[0188] Images that can completely cover all regions of the irregular triangular network are selected from the standardized image sequence as candidate images. The imaging tilt angle of each candidate image is quantitatively evaluated and the image sharpness is detected at the pixel level. The two evaluation results are combined according to the preset weight ratio to obtain the comprehensive visibility score between each region of the irregular triangular network and the corresponding candidate image.

[0189] The comprehensive visibility score of each region is used as the basic data input into the Markov random field optimization model. Through the neighborhood association optimization calculation of the model, the image with the best comprehensive visibility score is matched for each region of the irregular triangular mesh and determined as the optimal texture image of that region. At the same time, the texture seam position of the irregular triangular mesh is accurately delineated according to the coverage boundary of different optimal texture images.

[0190] The selected optimal texture image is subjected to projection transformation according to the spatial geometry of the irregular triangular mesh, so that the two-dimensional texture image and the three-dimensional triangular mesh surface are accurately fitted, generating initial texture blocks corresponding to each region of the irregular triangular mesh. For the defined texture seam positions, multi-band fusion processing is performed on adjacent initial texture blocks, fusing texture information of different frequency bands layer by layer, eliminating texture differences at the seams, and obtaining fused texture blocks of the irregular triangular mesh.

[0191] Global color consistency adjustment is performed on the merged texture blocks, and the color, brightness and contrast of the texture blocks are corrected pixel by pixel to ensure that the texture blocks on the entire surface of the irregular triangular mesh maintain a coordinated and unified color performance without obvious color banding and brightness differences. After the adjustment is completed, the texture mapping result of the irregular triangular mesh is obtained.

[0192] The final texture mapping result is assigned to each triangular unit of the irregular triangular mesh according to the topological relationship of the triangular mesh, so that the surface of the three-dimensional irregular triangular mesh is covered with real image texture, while preserving the three-dimensional spatial measurement attributes of the triangular mesh, and finally obtaining a measurable three-dimensional real scene model of the target island reef.

[0193] The beneficial effects are that this implementation process, through morphological filtering and adaptive simplification based on curvature features, removes point cloud noise while preserving point cloud details in key areas of island and reef topography, resulting in higher quality base data for subsequent triangulation construction. The irregular triangulation constructed through 2D subdivision, 3D mapping, and geometric optimization has a standardized geometric shape and can accurately fit the 3D topographic features of the island and reef. During texture mapping, the optimal texture image is selected by comprehensively considering visibility scores. Combined with multi-band fusion and global color consistency adjustment, texture seams and color differences are effectively eliminated, allowing the texture to accurately fit the triangulation. The final measurable 3D reality model has both high-precision 3D spatial morphology and realistic texture representation, while retaining complete measurable attributes. It can meet the requirements of accuracy, visualization, and measurability in 3D reality mapping of islands and reefs, greatly improving the practicality and application value of island and reef mapping results.

[0194] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0195] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0196] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for three-dimensional real-scene mapping of islands and reefs based on unmanned equipment, characterized in that, The method includes: S1. Perform geometric analysis on the high-resolution satellite remote sensing image of the target island to obtain a general map of the survey area of ​​the target island, and set up ground control points on the target island according to the general map of the survey area to obtain high-precision geographic coordinate data of the ground control points. S2. Based on the survey area outline map, formulate an adaptive flight path for the unmanned equipment to collect multi-angle raw image data of the target islands and reefs; S3. Perform uniform light and color processing on the multi-angle original image data to obtain a standardized image sequence of the target island and reef, and extract features from the standardized image sequence to obtain the feature point set of the standardized image sequence. S4. Perform multi-baseline multi-view matching on the feature point set to obtain the corresponding point matching relationship between the standardized image sequences. Construct a connection point network of the standardized image sequences with the feature point set as nodes and the corresponding point matching relationship as edges. S5. Using the high-precision geographic coordinate data as a constraint, substitute it into the connection point network to perform bundle adjustment of the joint regional network to obtain the high-precision exterior orientation elements and densification point coordinates of the standardized image sequence. S6. Based on the high-precision exterior orientation elements, and using the encrypted point coordinates as seed points to constrain the search range, perform pixel-by-pixel depth estimation on the standardized image sequence to obtain the three-dimensional point cloud data of the target island / reef; S7. Construct an irregular triangular mesh from the three-dimensional point cloud data, and perform texture mapping on the surface of the triangular mesh to obtain a measurable three-dimensional real-world model of the target island / reef.

2. The method for three-dimensional real-scene mapping of islands and reefs based on unmanned equipment as described in claim 1, characterized in that, The geometric analysis of the high-resolution satellite remote sensing imagery of the target islands and reefs to obtain a rough map of the survey area of ​​the target islands and reefs includes: Acquire high-resolution satellite stereo image pairs covering the target islands and reefs, eliminate geometric distortions in the high-resolution satellite stereo image pairs, and obtain epipolar image pairs of the target islands and reefs; Dense matching is performed on the epipolar image pairs to obtain the digital surface model of the target island / reef. The digital surface model is then filtered to obtain the initial elevation data of the target island / reef. Multi-scale segmentation is performed on the panchromatic image in the epipolar image pair to extract the lithological texture boundary of the target island reef, and combined with the initial elevation data, the stable exposed rock mass area on the target island reef is identified. The stable exposed rock mass area is processed into a grid to obtain a rough map of the survey area of ​​the target island reef.

3. The method for three-dimensional real-scene mapping of islands and reefs based on unmanned equipment as described in claim 1, characterized in that, The step of formulating an adaptive flight path for the unmanned equipment based on the survey area outline map to collect multi-angle raw image data of the target islands and reefs includes: Read the digital surface model data contained in the survey area outline map, resample the digital surface model data, and obtain the rasterized elevation matrix of the target island / reef; Traverse the grid cells in the rasterized elevation matrix to obtain the elevation difference between the grid cells, and determine the slope steepness coefficient of the target island / reef based on the elevation difference. Based on the slope gradient coefficient, the island and reef areas in the survey area outline map are adaptively partitioned to obtain the refined collection area and the conventional collection area of ​​the target island and reef. Differentiated trajectory planning is performed on the refined data collection area and the conventional data collection area to obtain the three-dimensional route planning for the target island / reef; The three-dimensional flight path plan is imported into the flight control system of the unmanned equipment to generate the exposure trigger command and gimbal angle adjustment command of the unmanned equipment. Based on the exposure trigger command and the gimbal angle adjustment command, the camera tilt angle of the unmanned equipment is adjusted during flight to collect multi-angle raw image data covering the target island reef.

4. The method for three-dimensional real-scene mapping of islands and reefs based on unmanned equipment as described in claim 1, characterized in that, The process of performing uniform illumination and color matching on the multi-angle original image data to obtain a standardized image sequence of the target island / reef, and extracting features from the standardized image sequence to obtain a feature point set of the standardized image sequence, includes: Extract the global grayscale mean and grayscale standard deviation of the original image from the multi-angle original image data to obtain the brightness statistical characteristics of the multi-angle original image data; The image covering the main area of ​​the island and reef and with uniform illumination in the multi-angle original image data is used as the reference image. The brightness statistical characteristics of the reference image are used as a reference to perform pixel-by-pixel grayscale linear transformation on the remaining original images to obtain the initial corrected image sequence of the target island and reef. The initial calibrated image sequence is divided into grids to obtain the local average brightness of the initial calibrated image sequence; Based on the local brightness mean, gamma correction is performed on the pixels of the images in the initial corrected image sequence to obtain the standardized image sequence of the target island. The three-dimensional local extreme points of the images in the standardized image sequence are used as candidate feature points, and the candidate feature points are precisely located at the sub-pixel level. Unstable points located in low-contrast regions and edge regions are eliminated to obtain stable feature points of the standardized image sequence. Using the stable feature point as the center, a neighborhood window is defined, and the gradient direction distribution of pixels within the neighborhood window is statistically analyzed to obtain the main direction of the stable feature point. In the main direction, the neighborhood window is rotated and normalized, and the gradient direction distribution is statistically analyzed in blocks to obtain the feature descriptor of the stable feature point; Stable feature points, including location, scale, main direction, and feature descriptor, are aggregated into the feature point set of the standardized image sequence.

5. The method for three-dimensional real-scene mapping of islands and reefs based on unmanned equipment as described in claim 1, characterized in that, The step of performing multi-baseline, multi-view matching on the feature point set to obtain the corresponding point matching relationships between the standardized image sequences, and constructing a connection point network of the standardized image sequences using the feature point set as nodes and the corresponding point matching relationships as edges, includes: The image with the largest image overlap area is selected from the standardized image sequence as the matching basis set, and the feature points of the matching basis set are extracted from the feature point set as the set of points to be matched. Based on the feature descriptors in the set of points to be matched, an approximate nearest neighbor search is performed on the matching basis set to obtain the initial matching pair set of the normalized image sequence; Consistency verification is performed on the corresponding points in the initial matching pair set to obtain the corresponding point matching relationship of the standardized image sequence; Using the corresponding point matching relationship as a seed, and through the adjacency relationship between images, corresponding point tracking is extended to other images in the standardized image sequence to obtain the dense corresponding point trajectory of the standardized image sequence. The feature points in the dense corresponding point trajectory are abstracted as nodes, and the correspondence between corresponding feature points on the image in the dense corresponding point trajectory is abstracted as edges, thus constructing the connection point network of the standardized image sequence.

6. The method for three-dimensional real-scene mapping of islands and reefs based on unmanned equipment as described in claim 1, characterized in that, The step of using the high-precision geographic coordinate data as constraints and substituting it into the connective point network for bundle adjustment to obtain the high-precision exterior orientation elements and densification point coordinates of the standardized image sequence includes: Based on the connection relationships of the nodes in the connection point network, the initial exterior orientation elements of the images in the standardized image sequence are extracted, and the image point coordinate residuals of the connection points in the connection point network on different images are obtained. Based on the high-precision geographic coordinate data, coordinate constraints are constructed for the ground control points, and the weight matrix configuration of the bundle adjustment method combined with the regional network is adjusted according to the coordinate constraints. The image point coordinate residuals and the adjusted weight matrix are substituted into the bundle joint regional network adjustment to update the exterior orientation elements of the standardized image sequence and the ground coordinates of the connection points; The exterior orientation elements and the ground coordinates are iteratively updated to obtain the high-precision exterior orientation elements and densification point coordinates of the standardized image sequence.

7. The method for three-dimensional real-scene mapping of islands and reefs based on unmanned equipment as described in claim 1, characterized in that, The process involves performing pixel-by-pixel depth estimation on the standardized image sequence based on the high-precision exterior orientation elements, using the encrypted point coordinates as seed points to constrain the search range, to obtain the three-dimensional point cloud data of the target island / reef, including: Based on the high-precision exterior orientation elements, the coordinates of the encrypted point are back-projected onto the image data in the standardized image sequence to obtain the image point position and seed depth of the encrypted point on the image. A reference image is selected from the standardized image sequence. A matching window is defined with the coordinates of the pixel to be estimated in the reference image as the center. The depth search range and candidate depth values ​​of the standardized image sequence are determined according to the distribution of the seed depth in the neighborhood of the pixel to be estimated. Based on the high-precision exterior orientation elements, the object point corresponding to the candidate depth value is projected onto a neighboring image that overlaps with the reference image, and the coordinates of the projected point on the neighboring image are obtained. Extract the grayscale value sequence of the pixels within the matching window in the neighborhood image, and determine the photometric consistency measure of the candidate depth value in the neighborhood image based on the grayscale value sequence; The aggregation cost of the candidate depth values ​​is obtained by fusing the photometric consistency metric. The candidate depth value with the lowest aggregation cost is determined as the depth value of the pixel to be estimated; By combining the depth value and the high-precision exterior orientation elements, and back-projecting them into the object space through collinearity, the three-dimensional point cloud data of the target island / reef is obtained.

8. The method for three-dimensional real-scene mapping of islands and reefs based on unmanned equipment as described in claim 7, characterized in that, The formula for calculating the aggregation cost is as follows: ； in, This represents the aggregation cost. This represents the candidate depth value. This represents the set of neighboring images. This indicates the preset viewpoint weights. Represented in pixels The matching window centered on the center, Represents pixels on the reference image In color channels grayscale values ​​on Representing neighboring images Upper projection point In color channels grayscale values ​​on The preset smoothness constraint weights, For pixels The neighborhood pixel set, Represents neighboring pixels The depth estimate.

9. The method for three-dimensional real-scene mapping of islands and reefs based on unmanned equipment as described in claim 1, characterized in that, The process of constructing an irregular triangular mesh from the 3D point cloud data and performing texture mapping on the surface of the triangular mesh to obtain a measurable 3D reality model of the target island / reef includes: Morphological filtering is performed on the three-dimensional point cloud data to obtain clean three-dimensional point cloud data of the target island / reef; The clean 3D point cloud data is subjected to adaptive simplification based on curvature features. High-density point clouds are retained in areas with drastic curvature changes, while point clouds are thinned in areas with gentle curvature, resulting in a non-uniform density point cloud of the target island reef. The non-uniform density point cloud is projected onto a two-dimensional plane, and the projected point cloud is triangulated in two dimensions to obtain the triangular network topology of the target island reef. The triangular mesh topology is mapped back to three-dimensional space, and a spatial triangular mesh is constructed based on the three-dimensional coordinates of the three-dimensional point cloud data. The narrow triangles and intersecting facets in the spatial triangular mesh are detected. The elongated triangle is optimized by edge swapping, and the intersecting facets are locally subdivided to obtain the irregular triangular network of the target island / reef. Texture mapping is performed on the irregular triangular mesh to obtain a measurable three-dimensional reality model of the target island reef.

10. The method for three-dimensional real-scene mapping of islands and reefs based on unmanned equipment as described in claim 9, characterized in that, The process of performing texture mapping on the irregular triangular mesh to obtain a measurable 3D reality model of the target island / reef includes: Candidate images covering the irregular triangular mesh are selected from the standardized image sequence. Based on the imaging tilt angle and image sharpness of the candidate images, the comprehensive visibility score between the irregular triangular mesh and the candidate images is determined. Based on the comprehensive visibility score, the optimal texture image is selected for the irregular triangular mesh using a Markov random field optimization model, and the texture seam position of the irregular triangular mesh is determined. The optimal texture image is projected and transformed to obtain the initial texture block of the irregular triangular mesh. At the texture seam position, the initial texture block is subjected to multi-band fusion processing to obtain the fused texture block of the irregular triangular mesh. Global color consistency adjustment is performed on the fused texture block to obtain the texture mapping result of the irregular triangular mesh; The texture mapping result is applied to the irregular triangular mesh to obtain a measurable three-dimensional reality model of the target island / reef.

Images