Methods, systems, and media for extracting obstacles from asteroid surfaces based on multi-curvature features

Abstract

This invention relates to a method, system, and medium for extracting obstacles from the surface of asteroids based on multi-curvature features. The method includes: acquiring three-dimensional terrain data of the asteroid to be processed; calculating the Gaussian curvature, mean curvature, and neighborhood normal difference curvature of each vertex on the asteroid surface; performing absolute value processing and normalization processing on each vertex sequentially; performing dimensionality reduction and fusion on the three curvatures after normalization processing to obtain the comprehensive curvature feature value of each triangular pixel in the three-dimensional terrain data, and selecting candidate obstacle pixels based on a percentile threshold; performing connectivity analysis on all candidate obstacle pixels to form multiple independent connected regions; selecting spatially closed regions from all connected regions and filling their interiors as obstacle candidate regions; performing planar fitting on each obstacle candidate region to filter out and remove falsely detected flat regions, and using the remaining obstacle candidate regions as the extracted obstacle regions. Compared with the prior art, this invention has the advantages of high detection accuracy.

Description

Technical Field

[0001] This invention relates to the field of asteroid surface obstacle extraction technology, and in particular to a method, system and medium for asteroid surface obstacle extraction based on multi-curvature features. Background Technology

[0002] As deep space exploration activities deepen, asteroid exploration, landing, and sample return missions have become important research topics in the international aerospace field. It is generally believed that asteroid surfaces are often composed of debris piles, and these obstacles pose a threat to the safe landing, surface sampling, and navigation planning of probes. Therefore, accurately identifying and extracting obstacle areas on the asteroid surface has become one of the key technologies to ensure the safe operation of probes under complex terrain conditions.

[0003] Currently, methods for extracting obstacle regions on asteroid surfaces mainly include methods based on thresholding of a single terrain parameter, methods based on image texture features, methods based on 3D point cloud processing, and methods based on deep learning.

[0004] Methods based on image texture features mainly rely on two-dimensional remote sensing images to determine obstacle areas through texture analysis or template matching. This method is easily affected by lighting conditions and changes in viewing angle, and it is difficult to accurately reflect the three-dimensional structural features of small-scale obstacles.

[0005] Methods based on 3D point cloud processing determine obstacle areas through geometric feature analysis of point cloud data, such as surface roughness and elevation changes. For example, invention CN111830534A discloses a method for selecting the optimal landing point using lidar. This method generates a 3D point cloud using lidar, extracts the terrain's horizontal plane, segments obstacles and safe zones, determines the optimal landing point using the maximum safe radius, and tracks the landing point using a rigid body transformation matrix. While this approach can reflect the spatial structure of obstacles relatively well, it suffers from susceptibility to noise and poor robustness.

[0006] Deep learning-based methods have become an important direction in the field of deep space image recognition in recent years for asteroid surface obstacle detection. Common deep learning algorithms include Mask R-CNN and YOLO, but these methods face challenges in terms of training samples and generalization ability. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the existing technology, such as single curvature features, insufficient detection accuracy, susceptibility to noise interference, and a large number of false detections and missed detections, and to provide a method, system and medium for asteroid surface obstacle extraction based on multiple curvature features, so as to achieve high-precision automatic extraction of obstacle regions on asteroid surfaces.

[0008] The objective of this invention can be achieved through the following technical solutions: A method for extracting obstacles on the surface of asteroids based on multi-curvature features includes: Acquire the three-dimensional terrain data of the asteroid to be processed, and calculate the Gaussian curvature, mean curvature, and neighborhood normal difference curvature of each vertex on the asteroid surface in the three-dimensional terrain data; The calculated Gaussian curvature, mean curvature, and neighborhood normal difference curvature are respectively subjected to absolute value processing and normalization processing. The normalized Gaussian curvature, average curvature, and neighborhood normal difference curvature are dimensionality reduced and fused to obtain the comprehensive curvature feature value of each triangular pixel in the three-dimensional terrain data. Triangular pixels with comprehensive curvature feature values ​​higher than the preset percentile threshold are selected as candidate obstacle pixels. Connectivity analysis is performed on all candidate obstacle pixels to form multiple independent connected regions; spatially closed regions are selected from all connected regions and their interiors are filled to serve as candidate obstacle regions. Plane fitting is performed on each candidate obstacle region. Based on the plane fitting results, falsely detected flat regions are filtered out and removed. The remaining candidate obstacle regions are used as the extracted obstacle regions.

[0009] Furthermore, the calculation process for the neighborhood normal difference curvature includes: The spatial index of each vertex on the surface of the asteroid in the three-dimensional terrain data is constructed by using the KD Tree spatial indexing algorithm. For each vertex, a corresponding neighborhood vertex set is determined, and the difference in normal vectors between the vertex and each neighborhood vertex in the corresponding neighborhood vertex set is calculated, thereby obtaining the neighborhood normal difference curvature of each vertex.

[0010] Furthermore, the expression for calculating the neighborhood normal difference curvature is as follows: In the formula, Represents vertices The curvature of the neighborhood normal at that location, The Euclidean norm of a vector. Represents vertices The normal vector at that point, This represents the normal vector of a neighboring vertex. Represents vertices The set of vertices in the domain, Indicates the range of the neighborhood search.

[0011] Furthermore, the method employs principal component analysis algorithm to perform dimensionality reduction and fusion of the normalized Gaussian curvature, mean curvature, and neighborhood normal difference curvature; The processing steps of the principal component analysis algorithm include: For each vertex on the acquired asteroid surface, the first principal component that contributes the most from the corresponding normalized Gaussian curvature, mean curvature, and neighborhood normal difference curvature is extracted as the dominant feature representation of the corresponding vertex. For each triangular pixel in the 3D terrain data, the arithmetic mean of the dominant feature expression of the three vertices is calculated to obtain the comprehensive curvature feature value of each triangular pixel.

[0012] Furthermore, the method employs a depth-first search algorithm or a breadth-first search algorithm to perform connectivity analysis on all candidate obstacle pixels, forming multiple independent connected regions.

[0013] Furthermore, the method excludes the largest non-closed loop regions from all connected regions, then retains spatially closed regions and fills their interiors with a flood-fill algorithm as the obstacle candidate regions.

[0014] Furthermore, the method employs a random sampling consistency plane fitting approach to fit each obstacle candidate region to obtain a corresponding fitting plane.

[0015] Furthermore, the flat regions for screening false detections based on plane fitting results include: For each triangular pixel in the obstacle candidate region, calculate the standard deviation of the angle between the normal vector of the triangular pixel and the normal vector of the corresponding fitting plane, as well as the overall tilt angle of the obstacle candidate region, and calculate the average elevation of the obstacle candidate region. Set the angle standard deviation threshold, tilt angle threshold, and elevation tolerance threshold for the obstacle candidate region. If the obstacle candidate region simultaneously meets the following conditions: the standard deviation is lower than the angle standard deviation threshold, the overall tilt angle is lower than the tilt angle threshold, and the average elevation is lower than the elevation threshold calculated from the elevation tolerance threshold, then the falsely detected flat region is removed.

[0016] The present invention also provides an asteroid surface obstacle extraction system based on multi-curvature features, including a memory and a processor, wherein the memory stores a computer program, and the processor calls the computer program to execute the steps of the method described above.

[0017] The present invention also provides a computer-readable storage medium on which a computer program is stored, the computer program being executed by a processor using the method described above.

[0018] Compared with the prior art, the present invention has the following advantages: (1) This invention proposes to first calculate three different types of curvature features for each vertex of the three-dimensional terrain data of the asteroid surface, namely Gaussian curvature, mean curvature and neighborhood normal difference curvature. Among them, Gaussian curvature describes the degree of concavity and convexity of the surface and reflects the product of the degree of curvature of the point in the two principal directions; mean curvature represents the average value of the overall curvature of the surface at the point and describes whether the surface tends to bend outward or inward; neighborhood normal difference curvature is achieved by establishing a spatial index to obtain the difference of the normal vector of the neighboring vertices; the three curvatures can comprehensively capture the local geometric information of the asteroid surface at different scales. Subsequently, the curvatures were normalized, and principal component analysis was used to reduce the dimensionality of features to obtain the main features of each pixel surface in the asteroid's 3D terrain data. This effectively fused the three curvature features, improved the robustness of the curvature feature expression, removed redundant or unimportant information between data, avoided feature redundancy, and improved the stability and efficiency of the method. The obtained comprehensive feature value reflects the importance of each pixel in the curvature space, providing a quantitative indicator for subsequent obstacle region extraction. Next, to further determine the spatial connectivity between these candidate pixels, surface pixels with high curvature features were selected for connectivity analysis, and flood filling was used to form closed-loop candidate obstacle regions, effectively identifying and eliminating falsely detected flat regions, thus effectively reducing false detections and false negatives. Finally, to eliminate falsely detected flat regions, random sampling consistency plane fitting is performed on the candidate regions. This, combined with pixel spatial features, removes flat falsely detected regions, thus determining the final output set of obstacle pixels. This method offers advantages such as improved obstacle region recognition accuracy, robustness, and automation.

[0019] (2) This invention does not require prior samples of asteroid obstacles, and can realize unsupervised identification of obstacles on the surface of asteroids, thus automating the obstacle extraction process.

[0020] (3) This invention provides an automatic extraction method for asteroid surface obstacles with high precision, good robustness and high degree of automation, which effectively solves the problems of insufficient precision and robustness in existing methods, and has good engineering application value and scientific significance. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating a method for extracting obstacles on the surface of an asteroid based on multi-curvature features, provided in an embodiment of the present invention. Figure 2 This is a three-dimensional obstacle extraction result image based on a simulation dataset provided in an embodiment of the present invention, showing the recognition results of spherical and pyramidal obstacle regions by the method of the present invention under different viewing angles; Figure 3This image shows the result of automatic 3D obstacle extraction based on the shape model of the asteroid Ryugu, as provided in an embodiment of the present invention. It illustrates the obstacle extraction results of the method of the present invention on the complex surface terrain of a real asteroid. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0023] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0024] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0025] Example 1 like Figure 1 As shown, this embodiment provides a method for extracting obstacles on the surface of asteroids based on multi-curvature features, including: S1: Obtain the three-dimensional terrain data of the asteroid to be processed, and calculate the Gaussian curvature, mean curvature and neighborhood normal difference curvature of each vertex on the surface of the asteroid in the three-dimensional terrain data; S2: The calculated Gaussian curvature, mean curvature, and neighborhood normal difference curvature are respectively subjected to absolute value processing and normalization processing; S3: The normalized Gaussian curvature, average curvature and neighborhood normal difference curvature are dimensionality reduced and fused to obtain the comprehensive curvature feature value of each triangular pixel in the three-dimensional terrain data. Triangular pixels with comprehensive curvature feature values ​​higher than the preset percentile threshold are selected as candidate obstacle pixels. S4: Perform connectivity analysis on all candidate obstacle pixels to form multiple independent connected regions; select spatially closed regions from all connected regions and fill their interiors to serve as candidate obstacle regions. S5: Perform plane fitting on each obstacle candidate region, filter out falsely detected flat regions based on the plane fitting results, and use the remaining obstacle candidate regions as the extracted obstacle regions.

[0026] Specifically, in step S1, the three-dimensional terrain mesh model data of the target asteroid was acquired. The model data includes a vertex set. and pixel collection The mesh model is preprocessed by scaling and other steps to meet the needs of subsequent calculations. Based on the asteroid's 3D terrain mesh model data, the Gaussian curvature of each vertex was calculated. Mean curvature and neighborhood normal difference curvature Three curvature features are included, among which the neighborhood normal difference curvature constructed in this invention is achieved by establishing a spatial index (KD Tree) to obtain the difference in normal vectors of neighboring vertices, specifically including: The spatial index of each vertex on the surface of an asteroid in 3D terrain data is constructed by using the KD Tree spatial indexing algorithm. For each vertex, a corresponding neighborhood vertex set is determined, and the difference in normal vectors between the vertex and each neighborhood vertex in the corresponding neighborhood vertex set is calculated, thereby obtaining the neighborhood normal difference curvature of each vertex.

[0027] The expression for calculating the neighborhood normal difference curvature is: In the formula, Represents vertices The curvature of the neighborhood normal at that location, The Euclidean norm of a vector. Represents vertices The normal vector at that point, This represents the normal vector of a neighboring vertex. Represents vertices The set of vertices in the domain, Indicates the range of the neighborhood search.

[0028] In step S2, the three curvature features obtained from the above calculations are subjected to absolute value normalization to unify the feature scale.

[0029] In step S3, principal component analysis algorithm is used to reduce the dimension of the normalized Gaussian curvature, mean curvature and neighborhood normal difference curvature and fuse them. The processing steps of the principal component analysis algorithm include: For each vertex on the acquired asteroid surface, the first principal component that contributes the most from the corresponding normalized Gaussian curvature, mean curvature, and neighborhood normal difference curvature is extracted as the dominant feature representation of the corresponding vertex. The corresponding expression is: For each triangular pixel in the 3D terrain data, the arithmetic mean of the dominant feature representations of the three corresponding vertices is calculated to obtain the comprehensive curvature feature value of each triangular pixel. The corresponding calculation expression is: In step S4, a depth-first search algorithm or a breadth-first search method is used to perform connectivity analysis on all candidate obstacle pixels to form multiple independent connected regions.

[0030] The largest non-loop region is excluded from all connected regions. Then, the spatially closed regions are retained and their interiors are filled using a flood-fill algorithm as candidate obstacle regions.

[0031] Specifically, based on the comprehensive curvature features of pixels, pixels with curvature features higher than a predetermined threshold are selected as candidate obstacle pixels, and a pixel adjacency graph is constructed. Subsequently, based on the connected component analysis method, the closed loop region formed by connected high curvature pixels is used as the initial candidate obstacle region, and the pixels in the closed loop region are closed and filled using the flood filling algorithm to obtain the complete candidate obstacle region.

[0032] In step S5, the random sampling consistency plane fitting method is used to fit the corresponding fitting plane for each obstacle candidate region. The plane fitting equation is: Flat regions that are used to filter false positives based on plane fitting results include: For each triangular pixel in the obstacle candidate region, calculate the standard deviation of the angle between the normal vector of the triangular pixel and the normal vector of the corresponding fitted plane. and the overall tilt angle of the obstacle candidate area. And calculate the average elevation of the candidate obstacle region; Standard deviation and overall tilt angle The calculation expression is: in, The radial unit vector representing the centroid of a pixel. To fit the plane normal vector; Set thresholds for the standard deviation of angle, tilt angle, and elevation tolerance for obstacle candidate regions. If an obstacle candidate region simultaneously meets the following conditions: its standard deviation is lower than the angle standard deviation threshold, its overall tilt angle is lower than the tilt angle threshold, and its average elevation is lower than the elevation threshold calculated from the elevation tolerance threshold, then the falsely detected flat region is removed. The corresponding judgment expression is: In the formula, The average elevation of the candidate region. The average elevation of the closed-loop boundary. and To set the threshold parameter.

[0033] The candidate regions processed by the above steps are the final obstacle region extraction results on the asteroid surface. The obstacle region pixel index is recorded in text form and output for subsequent analysis and visualization.

[0034] The following is a description of the method steps and results for implementing an example using 3D terrain data.

[0035] I. Experimental Data Acquisition and Preprocessing Steps This embodiment selects simulation data and 3D terrain data of a typical asteroid, Ryugu, for experimental verification. Specifically, the 3D mesh model data of the Ryugu asteroid used in the experiment has a scale of approximately 49,000 triangular pixels (faces). The data is stored in OBJ format, with the origin of the data coordinate system located at the centroid of the model, and the coordinate system of each vertex is a Cartesian 3D coordinate system.

[0036] II. Calculation and Normalization of Multiple Curvature Features Next, three different types of curvature features are calculated for each vertex of the mesh model: Gaussian curvature, mean curvature, and neighborhood normal difference curvature, in order to comprehensively capture the local geometric information of different scales on the asteroid surface.

[0037] First, calculate the Gaussian curvature of each vertex of the asteroid's 3D model. and mean curvature Gaussian curvature describes the degree of unevenness or roughness of the surface.

[0038] This is the principal curvature of the three-dimensional terrain data surface. and The product of the two principal directions represents the overall curvature of the surface at that point. For a given point, the Gaussian curvature reflects the product of the curvature at that point in the two principal directions.

[0039] The mean curvature is calculated as follows: This is the principal curvature. and The average value represents the average degree of curvature of the surface at that point. The average curvature primarily describes whether the surface tends to bend outward or inward.

[0040] Secondly, for the neighborhood normal difference curvature, this embodiment uses the KD Tree spatial indexing algorithm. Specifically, using this algorithm, the spatial index of all vertices is first constructed. For each vertex... Find the set of vertices in its spatially defined neighborhood. Next, calculate the difference in vertex normals. The specific calculation formula is as follows: in, The Euclidean norm of a vector. Represents vertices The normal vector at that point, Let represent the normal vector of the neighboring vertices, and k be the range of the neighborhood search. After obtaining the above three curvature features, they are respectively subjected to absolute value processing and normalization processing, i.e.: The above operations ensure that the scale of all features is uniform. Within this range, weight imbalances between features are effectively avoided.

[0041] III. Dimensionality Reduction and Fusion Steps for Curvature Features To effectively fuse the three features of Gaussian curvature, mean curvature, and neighborhood normal difference curvature, and further improve the robustness of curvature feature representation, this embodiment uses Principal Component Analysis (PCA) to perform dimensionality reduction and fusion on the normalized three curvature features. PCA is a statistical method primarily used to reduce data dimensionality, extract key information from the dataset, and thus remove redundant or unimportant information between data points.

[0042] The specific operation is as follows: First, organize the three normalized curvature features of all vertices into a matrix structure, with each vertex corresponding to one row of the matrix, and each column representing the three features: Gaussian curvature, mean curvature, and neighborhood normal difference curvature. Then, perform principal component analysis on this matrix data, extracting the first principal component that contributes the most, as the dominant feature representation for each vertex, denoted as . The specific calculation formula is as follows: Next, to further demonstrate the overall expression of curvature features at each pixel level, we summarize and calculate the dominant features of each vertex obtained above. Specifically, each triangular pixel consists of three vertices, and we take the arithmetic mean of the dominant features of these three vertices to obtain the comprehensive curvature feature value of each triangular pixel. The comprehensive curvature feature value is expressed as: This comprehensive feature value reflects the importance of each pixel in the curvature space, providing a quantitative indicator for subsequent obstacle region extraction.

[0043] IV. Preliminary Extraction Steps for High-Curvature Closed-Loop Regions In this embodiment, all pixels with high comprehensive curvature feature values ​​are marked as preliminary obstacle candidate pixels. Then, to further determine the spatial connectivity between these candidate pixels, the depth-first search (DFS) algorithm is used to analyze connected regions, thereby identifying spatially closed loop regions.

[0044] In its implementation, the depth-first search algorithm uses a stack structure, visiting pixel nodes sequentially according to the following logic: Initially, an unvisited pixel is selected as the starting point and placed on the stack. Each time, a node is removed from the stack and visited, with its unvisited adjacent nodes pushed onto the stack. This process continues until no further depth is possible. At this point, backtracking is initiated, and the search continues in other directions. The traversal order of the above algorithm satisfies the following traversal characteristic formula: Define the node set as The set of adjacent nodes is Let the function be... Represents a node The timestamp of the first visit. Depth-first search satisfies the following traversal rules: That is, nodes neighboring nodes Always at the node After being visited for the first time, it is visited in sequence, which is reflected in a path going to the end and then backtracking to explore the next path.

[0045] Through the depth-first search traversal process described above, this embodiment can effectively identify the spatial connectivity structure between various high-curvature feature cells, forming multiple independent connected regions. Then, the largest non-closed region is excluded from all connected regions, and the remaining spatially closed regions are retained as preliminary obstacle candidate regions. These regions are further filled using a flood-fill algorithm to obtain a spatially continuous and complete set of obstacle candidate regions.

[0046] V. RANSAC False Detection Removal Steps: After determining the initial candidate obstacle regions, in order to eliminate falsely detected flat regions, RANSAC plane fitting is performed individually for each candidate region. The resulting plane equation is expressed as: For a set of pixels within a candidate region, calculate the standard deviation of the angle between the pixel normal vector and the fitting plane normal vector. and the overall tilt angle of the region The specific calculation formula is as follows: in, Let be the normal vector of the pixel. The normal vector of the fitted plane, is the radial unit vector of the region's centroid.

[0047] Define the discrimination threshold condition: an angle standard deviation threshold can be set. Region tilt angle threshold Elevation tolerance threshold If the following conditions are met simultaneously, the region is determined to be a flat region that is a false positive: in, Indicates the average elevation of the candidate region. This represents the average elevation of the closed boundary. Regions that meet the criteria will be removed from the obstacle candidate region to effectively eliminate false detections.

[0048] VI. Based on the method of the present invention, the following verification experiments were conducted: To verify the effectiveness and robustness of the automatic 3D obstacle extraction method of this invention, a standard dataset based on an ellipsoid and additional obstacles was constructed as the first embodiment. In the verification experiment, a closed ellipsoid with a semi-major axis ratio of 1:0.85:0.75 was used as the substrate, and three spherical protrusions and five pyramidal protrusions were distributed on its surface at random angles. All obstacles and the substrate were generated into a single OBJ file through triangular mesh Booleanization, and the pixel indices of various obstacles were written into a TXT text file to ensure a quantifiable comparison between the actual values ​​and the algorithm output.

[0049] Subsequently, the sample mesh was automatically extracted according to the process of this invention. During the experiment, the neighborhood query adopted a KD Tree combined with a block strategy to reduce memory overhead; the curvature threshold, angle threshold, and elevation difference threshold were all set according to requirements and can be flexibly adjusted in practical applications.

[0050] The method of this invention can completely identify two major categories of obstacles, spheres and pyramids, in a single run, automatically outputting pixels of red obstacle areas and blue obstacle areas surrounded by red areas, achieving accurate correspondence with manually generated ground truth pixels. The evaluation metrics used are as follows: FP = DET - GT FN = GT - DET When evaluating 3D obstacle extraction algorithms, the total number of triangular pixels in the mesh is recorded as the total number of pixels. The set of obstacle ground truth values ​​generated manually or through simulation is denoted as GT; the set of obstacles output by the algorithm and merged after deduplication is denoted as DET. The intersection of the two, TP, represents the real obstacles correctly detected by the algorithm; the pixels FP in DET that are not in GT represent false alarms; and the pixels FN in GT that are not covered by the algorithm reflect missed detections. The parameters in the experiment are shown in Table 1.

[0051] Table 1 Test Evaluation Indicators like Figure 2 As shown, the algorithm in this embodiment can stably extract obstacle regions on an ideal asteroid model, and these regions are marked in red. Figure 2 Images (a)–(c) show the obstacle extraction results at different angles on the asteroid surface; Table 1 presents the quantitative evaluation results. Overall, in Of the 1578 obstacle pixels detected, 1542 were true (TP), 36 were false positives (FP), and 62 were missed. Correspondingly, the algorithm's hit rate was [percentage missing]. Coverage The method achieved an IoU of 0.9402 and an F1 score of 0.9692, indicating that the method in this embodiment has high accuracy and consistency in obstacle region extraction tasks.

[0052] To further illustrate the obstacle extraction effect of this method on real asteroid data, a second embodiment of the invention was also provided: Figure 3 Images (a)-(c) show the results of the algorithm of this invention extracting obstacles marked in red on the 49,000-facet high-resolution shape model of the asteroid Ryugu published by the Japan Space Agency. The shape model of Ryugu (data source: JAXAHayabusa2 DARTS, 2019) is used in the algorithm of this invention for qualitative analysis of the real terrain. It can be seen that the algorithm can provide coherent rock candidates across the entire surface; in the polar region of the southern hemisphere of Ryugu, the largest megalith, Otohime Saxum, is fully marked in red, as seen in... Figure 3 (a) The lower left position demonstrates that the algorithm also exhibits good sensitivity to large-scale geological units. Simultaneously, a large number of small and medium-sized scattered rock blocks in the equator and Northern Hemisphere are detected, while gently sloping, low-curvature areas remain largely white, achieving accurate separation of obstacle zones. This experiment qualitatively verifies the effectiveness and robustness of the proposed multi-curvature fusion obstacle extraction strategy under the complex terrain of real asteroids.

[0053] VII. Final Output Obstacle Area After the above steps, the obstacle region on the asteroid surface obtained in this embodiment is a precisely extracted set of pixels. Finally, the obstacle pixel index is output in the form of a text file for further analysis and 3D visualization, to assist in the subsequent safe landing and path planning of the asteroid exploration mission.

[0054] Example 2 This embodiment provides an asteroid surface obstacle extraction system based on multi-curvature features, including a memory and a processor. The memory stores a computer program, and the processor calls the computer program to execute the steps of an asteroid surface obstacle extraction method based on multi-curvature features as described in Embodiment 1.

[0055] This embodiment also provides a computer-readable storage medium storing a computer program, which is executed by a processor as described in Embodiment 1, a method for extracting asteroid surface obstacles based on multi-curvature features.

[0056] The computer program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This computer program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the computer program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0057] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium can be a machine-readable signal medium or a machine-readable storage medium. A computer-readable storage medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0058] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for extracting surface obstacles of an asteroid based on multi-curvature features, characterized in that, include: Acquire the three-dimensional terrain data of the asteroid to be processed, and calculate the Gaussian curvature, mean curvature, and neighborhood normal difference curvature of each vertex on the asteroid surface in the three-dimensional terrain data; The calculated Gaussian curvature, mean curvature, and neighborhood normal difference curvature are respectively subjected to absolute value processing and normalization processing. The normalized Gaussian curvature, average curvature, and neighborhood normal difference curvature are dimensionality reduced and fused to obtain the comprehensive curvature feature value of each triangular pixel in the three-dimensional terrain data. Triangular pixels with comprehensive curvature feature values ​​higher than the preset percentile threshold are selected as candidate obstacle pixels. Connectivity analysis is performed on all candidate obstacle pixels to form multiple independent connected regions; Select spatially closed regions from all connected regions and fill their interiors to create candidate obstacle regions. Plane fitting is performed on each candidate obstacle region. Based on the plane fitting results, falsely detected flat regions are filtered out and removed. The remaining candidate obstacle regions are used as the extracted obstacle regions.

2. The method for extracting the surface obstacle of the asteroid based on the multi-curvature feature according to claim 1, characterized in that, The calculation process for the neighborhood normal difference curvature includes: The spatial index of each vertex on the surface of the asteroid in the three-dimensional terrain data is constructed by using the KD Tree spatial indexing algorithm. For each vertex, a corresponding neighborhood vertex set is determined, and the difference in normal vectors between the vertex and each neighborhood vertex in the corresponding neighborhood vertex set is calculated, thereby obtaining the neighborhood normal difference curvature of each vertex.

3. The method of claim 2, wherein, The expression for calculating the neighborhood normal difference curvature is as follows: wherein denotes the neighborhood normal difference curvature at a vertex denotes the Euclidean norm of a vector denotes the normal vector at a vertex denotes the normal vector of a neighborhood vertex denotes the set of neighborhood vertices at a vertex denotes the range of the neighborhood search.​​​ 4. The method of claim 1, wherein, The method employs principal component analysis algorithm to perform dimensionality reduction and fusion of normalized Gaussian curvature, mean curvature and neighborhood normal difference curvature. The processing steps of the principal component analysis algorithm include: For each vertex on the acquired asteroid surface, the first principal component that contributes the most from the corresponding normalized Gaussian curvature, mean curvature, and neighborhood normal difference curvature is extracted as the dominant feature representation of the corresponding vertex. For each triangular pixel in the 3D terrain data, the arithmetic mean of the dominant feature expression of the three vertices is calculated to obtain the comprehensive curvature feature value of each triangular pixel.

5. The method of claim 1, wherein, The method employs a depth-first search algorithm or a breadth-first search algorithm to perform connectivity analysis on all candidate obstacle pixels, forming multiple independent connected regions.

6. The method of claim 1, wherein, The method excludes the largest non-closed loop regions from all connected regions, then retains spatially closed regions and fills their interiors with a flood-fill algorithm, which serve as the obstacle candidate regions.

7. The method of claim 1, wherein, The method employs a random sampling consistency plane fitting approach to fit each obstacle candidate region and obtain the corresponding fitting plane.

8. The method of claim 1, wherein, The flat regions used for screening false detections based on plane fitting results include: For each triangular pixel in the obstacle candidate region, calculate the standard deviation of the angle between the normal vector of the triangular pixel and the normal vector of the corresponding fitting plane, as well as the overall tilt angle of the obstacle candidate region, and calculate the average elevation of the obstacle candidate region. Set the angle standard deviation threshold, tilt angle threshold, and elevation tolerance threshold for the obstacle candidate region. If the obstacle candidate region simultaneously meets the following conditions: the standard deviation is lower than the angle standard deviation threshold, the overall tilt angle is lower than the tilt angle threshold, and the average elevation is lower than the elevation threshold calculated from the elevation tolerance threshold, then the falsely detected flat region is removed.

9. A multi-curvature feature based system for extracting surface obstacles of an asteroid, characterized in that, It includes a memory and a processor, the memory storing a computer program, the processor invoking the computer program to perform the steps of the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which is executed by a processor according to any one of claims 1 to 8.

Images