Intelligent impact crater extraction method, system and equipment based on binocular stereoscopic image reconstruction and medium

Abstract

The invention discloses an intelligent impact crater extraction method, system and device based on binocular stereoscopic image reconstruction and a medium. The method comprises the following steps: acquiring left and right images of a moon patroller; inputting the left and right images into a binocular stereo matching deep learning model, and outputting a representation disparity map; calculating the coordinate of any point of the pixel coordinate system in the camera coordinate system through combining the disparity map with the internal reference of the camera so as to reconstruct and generate a three-dimensional point cloud; preprocessing the three-dimensional point cloud to obtain preprocessed point cloud data, performing projection and contour detection on the preprocessed point cloud data, and respectively extracting complete and incomplete impact crater data information from contour detection results through minimum circumcircle and Hough circle transformation; through binocular stereo image matching and three-dimensional reconstruction of deep learning, reliable data support is provided for accurate extraction of impact crater information, the problem of limited view caused by the height of a low mast of the patroller can be effectively solved, a small-size impact crater is accurately extracted, and reliable landmark data is provided for navigation and positioning of the patroller.

Description

Technical Field

[0001] This invention relates to the field of deep space exploration perception and positioning technology, and in particular to an intelligent method, system, device and medium for extracting impact craters based on binocular stereo image reconstruction. Background Technology

[0002] In lunar exploration missions, the high-precision autonomous navigation and positioning technology of rovers is crucial for future long-distance, large-scale lunar surface scientific exploration missions. The lunar south pole region, due to its potential water ice resources and unique geological structure, has become a key exploration area. This region has rugged and complex terrain and variable lighting, while impact craters, a typical feature of the lunar surface, have a density exceeding 10 at the lunar south pole. 3 units / km 2 With an average spacing of about 30 meters, they can provide important landmark references for the absolute positioning of the patrol vehicle.

[0003] Impact crater information extracted from rover images is crucial for their absolute localization. Existing impact crater extraction methods are primarily based on orbiter images. However, rover observations of the lunar surface from a horizontal or Earth-tilted perspective, with the mast only 1-2 meters above the ground, limit the field of view, allowing only small-sized impact craters to be captured in rover images. Furthermore, due to the significant difference between the rover's and orbiter's imaging perspectives, impact craters exhibit different morphologies under these two perspectives. Current impact crater identification algorithms are mostly applicable to orbiter imaging perspectives and large-sized impact craters. Extraction of small-sized impact craters from horizontal / Earth-tilted rover perspectives still suffers from perspective distortion, incomplete morphology leading to inaccurate extraction, and the inapplicability of traditional algorithms. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an intelligent method and system for extracting impact craters based on binocular stereo image reconstruction. This solves the problem that current rover impact crater identification algorithms are not applicable to small-sized impact craters and still suffer from perspective distortion and incomplete morphology in the extraction of impact craters from horizontal or tilted observation angles on the lunar surface.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an intelligent method for extracting impact craters based on binocular stereo image reconstruction, comprising: acquiring left and right images of a lunar rover;

[0008] The left and right images are input into a binocular stereo matching deep learning model, and a stereo matching result disparity map representing the disparity of corresponding pixels in the left and right images is output.

[0009] The coordinates of any point in the pixel coordinate system in the camera coordinate system are calculated by combining the disparity map with the camera intrinsic parameters, so as to perform three-dimensional reconstruction and generate a three-dimensional point cloud.

[0010] The three-dimensional point cloud is preprocessed to obtain point cloud data with the impact crater depression removed;

[0011] Projection and contour detection were performed on the point cloud data after removing the concave part of the impact crater. Complete and incomplete impact crater data information were extracted from the contour detection results by using the minimum circumcircle and Hough circle transform, respectively.

[0012] As a preferred embodiment of the intelligent impact crater extraction method based on binocular stereo image reconstruction described in this invention, the method includes: calculating the coordinates of any point in the pixel coordinate system in the camera coordinate system using the disparity map and camera intrinsic parameters to perform 3D reconstruction and generate a point cloud, comprising:

[0013] Transform the coordinates in the pixel coordinate system to the image plane coordinate system;

[0014] Perform a parallax-to-depth transformation on each pixel;

[0015] Based on the obtained depth, the coordinates in the image plane coordinate system are transformed to the coordinates in the camera coordinate system.

[0016] As a preferred embodiment of the intelligent impact crater extraction method based on binocular stereo image reconstruction described in this invention, the preprocessing of the three-dimensional point cloud includes:

[0017] The 3D point cloud is filtered using statistical filtering methods, specifically as follows:

[0018] Calculate the neighborhood mean distance and standard deviation of each point in the point cloud;

[0019] Outliers exceeding the standard deviation threshold are removed to reduce noise.

[0020] It also includes cropping the point cloud within a preset range from the camera after denoising.

[0021] As a preferred embodiment of the intelligent impact crater extraction method based on binocular stereo image reconstruction described in this invention, the method further includes: preprocessing the three-dimensional point cloud, and:

[0022] The cropped point cloud data was fitted to a plane using the RANSAC algorithm to obtain point cloud data excluding the impact crater depressions.

[0023] The beneficial effects of this preferred solution are: RANSAC can fit the continuous plane with the largest area in a noisy point cloud containing a large number of depressions and convexities, and can make the formed ring structure highly consistent with the real edge of the original impact crater without relying on the complete data of the depression area.

[0024] As a preferred embodiment of the intelligent impact crater extraction method based on binocular stereo image reconstruction described in this invention, the projection of point cloud data excluding the impact crater depression includes: projecting the point cloud data excluding the impact crater depression onto the X coordinate system of the camera. CCS -Z CCS flat.

[0025] As a preferred embodiment of the intelligent impact crater extraction method based on binocular stereo image reconstruction described in this invention, the method involves extracting complete and incomplete impact crater data information from the contour detection results using both minimum circumcircle and Hough circle transforms, including:

[0026] For complete impact craters with intact annular edge features, the minimum circumcircle is used for impact crater detection, and the coordinates of the impact crater center in the image plane coordinate system and the diameter information are extracted.

[0027] As a preferred embodiment of the intelligent impact crater extraction method based on binocular stereo image reconstruction described in this invention, the method involves extracting complete and incomplete impact crater data information from the contour detection results using both minimum circumcircle and Hough circle transforms, including:

[0028] For incomplete impact craters on the outer contour of the terrain, Hough circle transform is used to detect impact craters and extract the coordinates of the impact crater center in the image plane coordinate system and the diameter information.

[0029] The beneficial effects of this preferred solution are: by combining projection with the detection methods of the minimum circumscribed circle and Hough circle that allow for seamless switching, the sensitivity and accuracy of circle detection are improved.

[0030] Secondly, the present invention provides an intelligent impact crater extraction system based on binocular stereo image reconstruction, comprising:

[0031] The acquisition module is used to acquire left and right images of the lunar rover;

[0032] The feature output module is used to input the left and right images into the binocular stereo matching deep learning model and output a stereo matching result disparity map representing the disparity of corresponding pixels in the left and right images.

[0033] The point cloud reconstruction module is used to calculate the coordinates of any point in the pixel coordinate system in the camera coordinate system by combining the disparity map with the camera intrinsic parameters, so as to perform three-dimensional reconstruction and generate a three-dimensional point cloud.

[0034] The data processing module is used to preprocess the three-dimensional point cloud to obtain point cloud data with the impact crater depression removed.

[0035] The detection and extraction module is used to project and detect the contours of the point cloud data after removing the impact crater depressions. It extracts complete and incomplete impact crater data information from the contour detection results through the minimum circumcircle and Hough circle transformations, respectively.

[0036] Thirdly, the present invention provides a computer device, comprising:

[0037] Memory and processor;

[0038] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the intelligent extraction method for impact crater based on binocular stereo image reconstruction.

[0039] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the intelligent crater extraction method based on binocular stereo image reconstruction.

[0040] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention provides reliable data support for the accurate extraction of impact crater information through binocular stereo image matching and 3D reconstruction based on deep learning, realizing intelligent and high-precision extraction of impact crater information. It can effectively deal with the limited field of view caused by the low mast height of the rover, accurately extract small-sized impact craters, provide reliable landmark data for the rover's navigation and positioning, and provide a technical foundation for future rover exploration and high-precision navigation and positioning in areas with complex terrain and lighting conditions such as the lunar south pole. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a schematic diagram of the overall process of the intelligent extraction method for impact craters based on binocular stereo image reconstruction according to an embodiment of the present invention.

[0043] Figure 2 This is a schematic diagram of the framework of an intelligent impact crater extraction method based on binocular stereo image reconstruction according to an embodiment of the present invention.

[0044] Figure 3This is an example diagram of intelligent impact crater extraction in the impact crater intelligent extraction method based on binocular stereo image reconstruction according to an embodiment of the present invention.

[0045] Figure 4 This is a schematic diagram of the coordinate system setting in the intelligent crater extraction method based on binocular stereo image reconstruction according to an embodiment of the present invention. Detailed Implementation

[0046] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0047] Example 1, referring to Figure 1 As one embodiment of the present invention, a method for intelligent extraction of impact craters based on binocular stereo image reconstruction is provided, comprising:

[0048] S100: Acquire left and right images of the lunar rover;

[0049] S200: Input the left and right images into the binocular stereo matching deep learning model and output a stereo matching result disparity map representing the disparity of corresponding pixels in the left and right images;

[0050] S300: Calculates the coordinates of any point in the pixel coordinate system in the camera coordinate system by combining the disparity map with the camera intrinsic parameters, in order to perform 3D reconstruction and generate a 3D point cloud.

[0051] S400: Preprocess the 3D point cloud to obtain point cloud data with the impact crater depression removed;

[0052] S500: Projects and detects the contours of the point cloud data after removing the impact crater depressions, and extracts complete and incomplete impact crater data information from the contour detection results through minimum circumcircle and Hough circle transform, respectively.

[0053] It should be noted that lunar rovers in deep space have relatively low mast heights, and their cameras typically observe from a horizontal or tilted angle towards the ground. The limited field of view of the cameras means that the impact craters captured by the rovers are mainly small in size. Their visual features, such as edges, textures, brightness contrast, etc., are usually weaker, blurry, or discontinuous, and they often appear as incomplete shapes. Existing mainstream impact crater recognition algorithms, after optimization, are mainly designed for large impact craters with relatively obvious features. When processing these small, blurry, and incomplete impact craters in the rover images, their performance drops significantly or they fail.

[0054] Therefore, through steps S100-S500, from the perspective of the rover (horizontally / tilted relative to the Earth), the information of the impact crater is intelligently and accurately extracted based on the binocular stereo image matching and 3D reconstruction using deep learning. This enables the rover to perform binocular image impact crater detection, matching, navigation, and positioning in regions such as the lunar south pole.

[0055] Example 2, refer to Figures 1-4 As an embodiment of the present invention, based on the above embodiment, a method for intelligent extraction of impact craters based on binocular stereo image reconstruction is provided.

[0056] In this embodiment of the application, step S100 involves acquiring left and right images of the lunar rover;

[0057] Specifically, images are captured using a mast-mounted binocular stereo camera. The camera is typically mounted at the top of the rover's mast and consists of two independent optical cameras (i.e., a left-eye camera and a right-eye camera) fixed side by side.

[0058] It should be noted that the two cameras have a fixed relative positional relationship and their intrinsic parameters, such as focal length, principal point, and distortion coefficient, are determined through rigorous geometric calibration; as well as extrinsic parameters, namely, the transformation parameters between coordinates in the world coordinate system and coordinates in the camera coordinate system, including rotation matrix and translation vector.

[0059] In this embodiment of the application, in step S200, the left and right images are input into the binocular stereo matching deep learning model, and the stereo matching result disparity map representing the disparity of corresponding pixels in the left and right images is output.

[0060] In an alternative implementation, the stereo matching deep learning model in S200 can be PSMNet, which can improve the matching robustness of low-texture regions such as flat lunar surfaces by aggregating context through feature pyramids and 3D convolutions in the model, and finally generate sub-pixel disparity maps based on the weighted summation of probability distributions.

[0061] In another alternative implementation, the stereo matching deep learning model in S200 can be IGEV-Stereo, which can improve disparity accuracy through iterative optimization and geometric encoding, such as impact crater contours, and finally output the disparity map of the last iteration.

[0062] It should be noted that the above model can be selected based on actual needs and image conditions, or other suitable binocular stereo matching deep learning models can be chosen. By using the deep neural network architecture in this step to match and calculate pixel features between the left and right images, three-dimensional information can be extracted from two-dimensional images.

[0063] It should also be noted that the introduction of intelligent stereo matching and point cloud processing improves the reliability of 3D information. For the complex lighting conditions and weakly textured terrain of the lunar south pole, it enhances the accuracy of parallax calculation, showing a significant advantage over traditional stereo matching methods. The intelligently generated parallax map, combined with camera intrinsic parameters, achieves high-precision reconstruction from 2D images to 3D point clouds, providing reliable data support for the accurate extraction of impact crater information.

[0064] In this embodiment of the application, step S300 calculates the coordinates of any point in the pixel coordinate system in the camera coordinate system by combining the disparity map with the camera intrinsic parameters, in order to perform 3D reconstruction and generate a point cloud, including the following steps A1-A3:

[0065] A1: Transform the coordinates in the Pixel Coordinate System (PCS) to the Image Plane Coordinate System (IPCS);

[0066] Specifically, it can be expressed as:

[0067] x image =x pixel -c x

[0068] y image =c y -y pixel

[0069] Among them, (x pixel ,y pixel (x) is the coordinate in the pixel coordinate system. image ,y image (c) is a coordinate system like a plane coordinate system. x ,c y () represents the coordinates of the principal point of the image in the pixel coordinate system.

[0070] A2: Perform a parallax-to-depth transformation on each pixel;

[0071] Specifically, it can be expressed as:

[0072]

[0073] Where D is the depth (distance) from the camera to the object, f is the focal length of the camera in pixels, B is the baseline length between the two cameras, and d is the parallax.

[0074] A3: Based on the obtained depth, transform the coordinates in the image plane coordinate system to the coordinates in the camera coordinate system; specifically, from image plane coordinates (x...)... image ,y imageThe coordinates (X) from the camera coordinate system (CCS) to the camera coordinate system (CCS) camera ,Y camera Z camera The transformation of ) can be expressed as:

[0075]

[0076] Z camera =D

[0077] In this embodiment of the application, step S400 involves preprocessing the 3D point cloud, including the following steps B1-B2:

[0078] B1: The 3D point cloud is filtered using statistical filtering methods, specifically:

[0079] B1-1: Calculate the neighborhood mean distance and standard deviation of each point in the point cloud;

[0080] B1-2: Remove outliers that exceed the standard deviation threshold to reduce noise.

[0081] Furthermore, since the point cloud of distant targets is sparse and insufficient for effective impact crater detection, step B2 is also included.

[0082] B2: Extract point clouds from the denoised data within a preset range from the camera for cropping;

[0083] For example, the preset range can be within 50 meters.

[0084] In this embodiment of the application, the impact crater depression still exists in the point cloud after the above processing in step S400. Therefore, the preprocessing of the three-dimensional point cloud also includes step B3.

[0085] B3: The clipped point cloud data is fitted to a plane using the RANSAC (Random Sample Consensus) algorithm to obtain point cloud data with the impact crater depression removed.

[0086] It should be noted that the RANSAC algorithm, based on a random sampling iterative mechanism, can fit the largest continuous plane in a noisy point cloud containing a large number of depressions (impact craters) and protrusions (rocks). That is, the original lunar surface plane that has not been destroyed by impact craters. Then, by setting a distance threshold, points below the fitted plane are classified as depression regions and removed. Compared with some other algorithms, such as traditional least squares, which are severely interfered with by depression regions, RANSAC does not rely on the complete data of the depression regions. It can make the formed ring structure highly consistent with the real edge of the original impact crater, achieving targeted separation of the impact crater structure and providing ideal input for subsequent contour extraction.

[0087] In this embodiment of the application, the projection of the point cloud data after removing the impact crater depression in step S500 includes: projecting the point cloud data after removing the impact crater depression onto the X coordinate system of the camera. CCS -Z CCS flat.

[0088] Specifically, after projecting all 3D points onto the XZ plane of the camera coordinate system, the output is a 2D point set (image), which is a black terrain formed by the aggregation of scattered points from the point cloud projected onto this plane, such as... Figure 3 As shown in (j); where the X-axis points to the right, the Z-axis points upward, and the unshown Y-axis is perpendicular to the paper and faces outward, forming a right-handed system. For details, please refer to [reference needed]. Figure 4 Schematic diagram of coordinate system setup.

[0089] It should be noted that by using projection, the viewing angle can be changed from the horizontal / ground-tilted observation angle of the rover to an aerial view. This change in perspective can avoid the perspective distortion problem caused by low angle.

[0090] In this embodiment of the application, step S500 extracts complete and incomplete impact crater data information from the contour detection results using the minimum circumcircle and Hough circle transforms, respectively, including:

[0091] For complete impact craters with intact annular edge features, the minimum circumcircle is used for impact crater detection, and the coordinates of the impact crater center in the image plane coordinate system and the diameter information are extracted.

[0092] Specifically, in this step, the complete set of annular projection points is input, the minimum circumcircle is solved, the radius and center coordinates are obtained, and the center of the circle in the camera coordinate system is transformed into image plane coordinates by combining the camera intrinsic parameters.

[0093] It should be noted that impact crater detection using the smallest circumcircle can ignore sporadic noise within the crater, relying solely on the topological continuity of the crater edge. This method offers high accuracy but low computational complexity, can directly output the diameter, and supports rapid navigation and positioning.

[0094] In this embodiment of the application, step S500 extracts complete and incomplete impact crater data information from the contour detection results using the minimum circumcircle and Hough circle transforms, respectively, including:

[0095] For incomplete impact craters on the outer contour of the terrain, Hough circle transform is used to detect impact craters and extract the coordinates of the impact crater center in the image plane coordinate system and the diameter information.

[0096] Specifically, this step involves inputting the entire outer contour of the terrain, that is, the projection points of the edge contour, such as... Figure 3(k) The outer circle is a fan-shaped green outline (i.e., the outline detection result image). The Hough circle transformation steps include: constructing a 3D accumulator with the center and radius; for each edge point, traversing possible circles along the gradient direction, and obtaining votes for the circle parameters that satisfy the arc curvature constraint; counting the circle parameter with the highest number of votes as the detection result, and outputting the center and radius.

[0097] It should be noted that impact crater detection using Hough circle transform can achieve stable detection of arc segments, and is applicable to scenarios where the crater edge is not fully displayed due to the limited field of view of the camera, thus ensuring recognition accuracy.

[0098] It should also be noted that step S500, through a combination of projection and seamless switching between the detection of the minimum circumscribed circle and the Hough circle, can project the height difference of the pit edge onto the XZ plane to make it appear as a sine curve, thereby improving the sensitivity and accuracy of circle detection.

[0099] In summary, this invention supports the extraction of impact crater information from both horizontal and tilted observation perspectives of the rover on the lunar surface. Through binocular stereo image matching and 3D reconstruction, high-precision extraction of impact crater geometric parameters is achieved for both horizontal and tilted observation perspectives of the rover on the lunar surface. Unlike traditional orbiter imagery-based impact crater identification methods that rely on a vertical perspective, this invention effectively addresses the limited field of view caused by the rover's low mast height, accurately extracting small-sized impact craters and providing reliable landmark coordinate data for the rover's autonomous absolute positioning.

[0100] Example 3, referring to Figure 3 This embodiment is based on the above embodiment of the intelligent extraction method for impact craters based on binocular stereo image reconstruction, and provides a practical application example to demonstrate the operability and beneficial effects of the present invention.

[0101] like Figure 3 As shown, Figure 3 (a) and Figure 3 (b) is a binocular simulation image from the lunar rover's navigation camera. Figure 3 (c) Disparity map generated using a deep learning binocular stereo matching algorithm. Figure 3 (d) is the point cloud for 3D reconstruction based on disparity map. Figure 3 (e) illustrates the point cloud filtering and denoising process, where gray dots represent valid point cloud data retained after filtering and denoising, and red dots represent outliers that have been identified and marked. Figure 3 (f) shows the point cloud results from the rovers' perspective after filtering and cropping, demonstrating that these processes can improve the quality of the point cloud data. Figure 3 (g) shows the filtered and cropped point cloud data from a vertical perspective, with the red arrow pointing to the impact crater depressions that need to be removed; Figure 3(h) shows the outliers identified and removed using the RANSAC algorithm, demonstrating that the algorithm can effectively identify the impact crater depressions. Figure 3 (i) represents the point cloud retained after plane fitting. Figure 3 (j) is the X-axis of projecting the point cloud onto the camera coordinate system. CCS -Z CCS The result after planarization, from a top-down perspective, shows the outline of the impact crater. Figure 3 (k) shows the impact crater detection results. The green curve in the figure represents the contour detection result, circle ① represents the detection result for complete impact craters, and circle ② represents the detection result for incomplete impact craters. It can be seen that the impact crater detection method based on 3D point clouds proposed in this invention can effectively and accurately detect both complete and incomplete impact craters.

[0102] Example 4 illustrates a schematic scheme for an intelligent impact crater extraction method based on binocular stereo image reconstruction. It should be noted that the technical solution of this intelligent impact crater extraction system based on binocular stereo image reconstruction belongs to the same concept as the aforementioned intelligent impact crater extraction method based on binocular stereo image reconstruction. Details not described in detail in this embodiment can be found in the description of the aforementioned intelligent impact crater extraction method based on binocular stereo image reconstruction.

[0103] This embodiment also provides another intelligent impact crater extraction system based on binocular stereo image reconstruction, including:

[0104] The acquisition module is used to acquire left and right images of the lunar rover;

[0105] The feature output module is used to input the left and right images into the binocular stereo matching deep learning model and output a stereo matching result disparity map representing the disparity of corresponding pixels in the left and right images.

[0106] The point cloud reconstruction module is used to calculate the coordinates of any point in the pixel coordinate system in the camera coordinate system by combining the disparity map with the camera intrinsic parameters, so as to perform 3D reconstruction and generate a 3D point cloud.

[0107] The data processing module is used to preprocess the 3D point cloud to obtain point cloud data with the impact crater depression removed.

[0108] The detection and extraction module is used to project and detect the contours of the point cloud data after removing the impact crater depressions. It extracts complete and incomplete impact crater data information from the contour detection results through the minimum circumcircle and Hough circle transformations, respectively.

[0109] This embodiment also provides a computer device suitable for intelligent extraction of impact craters based on binocular stereo image reconstruction, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the intelligent extraction method for impact craters based on binocular stereo image reconstruction as proposed in the above embodiment.

[0110] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the intelligent crater extraction method based on binocular stereo image reconstruction as proposed in the above embodiments.

[0111] The storage medium proposed in this embodiment and the intelligent extraction method for impact crater based on binocular stereo image reconstruction proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0112] Based on the above description of the implementation methods, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0113] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not 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, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent extraction of impact craters based on binocular stereo image reconstruction, characterized in that, include: Acquire left and right images of the lunar rover; The left and right images are input into a binocular stereo matching deep learning model, and a stereo matching result disparity map representing the disparity of corresponding pixels in the left and right images is output. The coordinates of any point in the pixel coordinate system in the camera coordinate system are calculated by combining the disparity map with the camera intrinsic parameters, so as to perform three-dimensional reconstruction and generate a three-dimensional point cloud. The three-dimensional point cloud is preprocessed to obtain point cloud data with the impact crater depression removed; Projection and contour detection were performed on the point cloud data after removing the concave part of the impact crater. Complete and incomplete impact crater data information were extracted from the contour detection results by using the minimum circumcircle and Hough circle transform, respectively.

2. The intelligent impact crater extraction method based on binocular stereo image reconstruction as described in claim 1, characterized in that, The coordinates of any point in the pixel coordinate system are calculated in the camera coordinate system using the disparity map and camera intrinsic parameters, in order to perform 3D reconstruction and generate a point cloud, including: Transform the coordinates in the pixel coordinate system to the image plane coordinate system; Perform a parallax-to-depth transformation on each pixel; Based on the obtained depth, the coordinates in the image plane coordinate system are transformed to the coordinates in the camera coordinate system.

3. The intelligent crater extraction method based on binocular stereo image reconstruction as described in claim 2, characterized in that, Preprocessing of the 3D point cloud includes: The 3D point cloud is filtered using statistical filtering methods, specifically as follows: Calculate the neighborhood mean distance and standard deviation of each point in the point cloud; Outliers exceeding the standard deviation threshold are removed to reduce noise. It also includes cropping the point cloud within a preset range from the camera after denoising.

4. The intelligent crater extraction method based on binocular stereo image reconstruction as described in claim 3, characterized in that, Preprocessing the three-dimensional point cloud further includes: The cropped point cloud data was fitted to a plane using the RANSAC algorithm to obtain point cloud data excluding the impact crater depressions.

5. The intelligent crater extraction method based on binocular stereo image reconstruction as described in claim 4, characterized in that, Projecting the point cloud data after removing the impact crater depressions includes: projecting the point cloud data after removing the impact crater depressions onto the X coordinate system of the camera. CCS -Z CCS flat.

6. The intelligent impact crater extraction method based on binocular stereo image reconstruction as described in claim 5, characterized in that, Complete and incomplete impact crater data information were extracted from the contour detection results using the minimum circumcircle and Hough circle transforms, respectively, including: For complete impact craters with intact annular edge features, the minimum circumcircle is used for impact crater detection, and the coordinates of the impact crater center in the image plane coordinate system and the diameter information are extracted.

7. The intelligent crater extraction method based on binocular stereo image reconstruction as described in claim 5 or 6, characterized in that, Complete and incomplete impact crater data information were extracted from the contour detection results using the minimum circumcircle and Hough circle transforms, respectively, including: For incomplete impact craters on the outer contour of the terrain, Hough circle transform is used to detect impact craters and extract the coordinates of the impact crater center in the image plane coordinate system and the diameter information.

8. An intelligent impact crater extraction system based on binocular stereo image reconstruction, employing the method described in any one of claims 1-7, characterized in that, include: The acquisition module is used to acquire left and right images of the lunar rover; The feature output module is used to input the left and right images into the binocular stereo matching deep learning model and output a stereo matching result disparity map representing the disparity of corresponding pixels in the left and right images. The point cloud reconstruction module is used to calculate the coordinates of any point in the pixel coordinate system in the camera coordinate system by combining the disparity map with the camera intrinsic parameters, so as to perform three-dimensional reconstruction and generate a three-dimensional point cloud. The data processing module is used to preprocess the three-dimensional point cloud to obtain point cloud data with the impact crater depression removed. The detection and extraction module is used to project and detect the contours of the point cloud data after removing the impact crater depressions. It extracts complete and incomplete impact crater data information from the contour detection results through the minimum circumcircle and Hough circle transformations, respectively.

9. A computer device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the intelligent crater extraction method based on binocular stereo image reconstruction as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the intelligent crater extraction method based on binocular stereo image reconstruction as described in any one of claims 1 to 7.

Images