High-generalization landing point positioning method based on extraterrestrial celestial body topographic image small sample

Abstract

The invention discloses a small sample high-generalization landing point positioning method based on an extraterrestrial celestial body topographic image. The method comprises the following steps: acquiring a spacecraft landing area DOM and a spacecraft landing original descent sequence image; dividing the DOM into small blocks, and recording the small blocks as sub-base maps; carrying out adaptive frame extraction on the spacecraft landing descending sequence image, calculating a homography matrix between adjacent frames of a descending image sequence, and recovering the relative pose of an inter-frame camera; obtaining the estimated height of each frame, and calculating the resolution corresponding to each frame in the descending image sequence; selecting a descending image with the physical resolution closest to the DOM as a descending image key frame; carrying out exhaustion matching on the key frame of the descending image and all the sub-base maps, and screening out an optimal sub-base map; matching point pairs between adjacent descending sequence images are obtained, and coarse positioning latitude and longitude coordinates of a landing point are obtained in combination with the descending image key frame and the matching point pair of the optimal sub-base image; and adding the optimal sub-base map into the descending image sequence to carry out cluster adjustment to obtain latitude and longitude coordinates of the fine positioning point.

Description

Technical Field

[0001] The invention relates to a highly generalized landing point positioning method based on a small sample of extraterrestrial body terrain images, belonging to the technical field of visual navigation. Background Art

[0002] The precise coordinates of the landing point play a key role in landing and returning, geological analysis, mineral positioning and other tasks. In view of the mission purpose of various probes landing on extraterrestrial bodies, the precise coordinates of the landing point are of great value to engineering implementation and scientific research, and are also of great significance to the analysis of the maneuvering of spacecraft during landing.

[0003] Due to the particularity of the landing location of extraterrestrial probes and the short mission implementation cycle, the data required for landing point positioning is scarce and the timeliness requirement is extremely high (for example, the sampling return mission on the back of the moon and the surface of Mars, etc.), while traditional landing point positioning algorithms often have limitations in matching accuracy and versatility. Summary of the invention

[0004] The technical problem solved by the present invention is: to overcome the shortcomings of the prior art and provide a highly generalized landing point positioning method based on a small sample of extraterrestrial body terrain images, thereby solving the demand for rapid realization of highly generalized landing positioning of extraterrestrial bodies under small sample conditions.

[0005] The technical solution of the present invention is: a highly generalized landing point positioning method based on a small sample of extraterrestrial body terrain images, wherein:

[0006] Obtain the DOM of a large area within the preset range of spacecraft landing and the original descent sequence images of the spacecraft landing;

[0007] The large area DOM is divided into small blocks, which are recorded as sub-base maps; the spacecraft landing and descent sequence images are adaptively extracted to obtain the descent sequence images after extraction, which are recorded as descent image sequences and are subsequently processed;

[0008] Calculate the homography matrix between adjacent frames of the descent image sequence, and use the homography matrix to restore the relative position of the camera between frames; obtain the estimated height of each frame through the physical trajectory constraints of the spacecraft landing process, and use the camera parameters to calculate the resolution corresponding to each frame in the descent image sequence;

[0009] According to the physical resolution corresponding to each frame of the descending image sequence, the descending image closest to the DOM physical resolution is selected as the descending image key frame; the descending image key frame is exhaustively matched with all sub-base images to screen out the best sub-base image;

[0010] Perform inter-frame matching on the descent image sequence to obtain matching point pairs between adjacent descent sequence images, combine the matching point pairs of the descent image key frame and the best sub-base map, obtain the inter-frame homography matrix through the image matching algorithm, and transfer the landing point from the last descent image frame to the descent image key frame through the homography transformation, and then transfer it to the best sub-base map to obtain the coarse positioning latitude and longitude coordinates of the landing point;

[0011] The best sub-base map is added to the descent image sequence for bundle adjustment. The position and posture of the best sub-base map relative to the descent sequence are obtained by continuously optimizing the reprojection error of the same-name points. The landing point is directly projected onto the best sub-base map according to the position and posture, thereby obtaining the longitude and latitude coordinates of the precise positioning point.

[0012] Preferably, the length and width of the large-area DOM initially acquired are both at the level of 10,000 meters, and the original descent sequence images of the spacecraft landing are no less than 500 frames.

[0013] Preferably, when the adaptive frame extraction algorithm is used to adaptively extract frames from the original descent sequence images of the spacecraft landing, the input parameters include the frame interval and the similarity threshold, and the frame sequence for trajectory reconstruction is output; wherein: the frame interval is used to accelerate the algorithm, and specifies the number of intermediate frames skipped between adjacent extracted frames; the similarity threshold is a parameter used to adaptively adjust the frame extraction interval;

[0014] The structural similarity index SSIM is used as an indicator to measure the similarity between adjacent landing images: when the structural similarity index SSIM between adjacent extracted frames meets the indicator requirements, frame extraction ends;

[0015] The calculation method of the structural similarity index SSIM is as follows:

[0016]

[0017] Where x and y are two non-negative image signals, C1 and C2 are stable constants used to avoid the denominator being zero; μ x and μ y are the means of images x and y, representing brightness information; σ x and σ y are the standard deviations of images x and y, representing contrast information; xy is the covariance of images x and y, representing structural information; l(x,y) is a brightness similarity measure, used to measure the similarity between the average brightness of two images; c(x,y) is a contrast similarity measure, used to measure the similarity between the contrasts of two images; s(x,y) is a structural similarity measure, used to measure the similarity between the structures of two images; SSIM(x,y) is the comprehensive result of brightness, contrast and structural similarity measures, and is the structural similarity index between two images.

[0018] Preferably, when calculating the homography matrix between adjacent frames of the descending image sequence after frame extraction, a matching algorithm is used and the matching result is optimized through bundle adjustment to obtain an optimized homography matrix, based on which the relative position and posture of the camera between frames are restored.

[0019] Preferably, the estimated height of each frame is obtained through the physical trajectory constraints of the spacecraft landing process, specifically:

[0020] In the final stage of landing, the lander does not perform any horizontal maneuvers and maintains a perpendicular direction to the surface of the extraterrestrial body. At this time, the camera position can be projected onto the vertical axis relative to the surface of the extraterrestrial body, and the estimated height of each frame can be obtained by combining the relative position of the camera between frames.

[0021] Preferably, the method of calculating the corresponding resolution of each frame of the descending image sequence using the camera parameters is:

[0022]

[0023] Among them, R x and R y are the physical resolutions in the x and y directions respectively; θ x and θ y Respectively represent the field of view in the x and y directions, and pix x and pix y They represent the pixel resolution in the x and y directions of the camera respectively; h is the height of the current descending sequence image.

[0024] Preferably, when exhaustively matching the descending image key frame with all sub-base images to select the best sub-base image:

[0025] Using pre-trained deep learning networks SuperPoint and SuperGlue, the SuperPoint network extracts key points from image pairs, and then the SuperGlue network performs pairwise matching of key points and evaluates the matching scores;

[0026] The predefined matching score and the threshold of the number of matching key point pairs are used as evaluation criteria. The key point pairs exceeding the score threshold are regarded as matching point pairs; the sub-images whose number of matching key points exceeds the threshold are regarded as potential best sub-images, and the potential best sub-images are sorted according to the matching scores, and the image with the highest matching score is selected as the best sub-image.

[0027] Preferably, during the image matching process:

[0028] The matched images are rotated at certain angle intervals to avoid mismatching caused by angle differences;

[0029] There is a significant illumination difference between DOM and landing images. Taking DOM as the reference image, histogram normalization is applied to the keyframes of the landing image to avoid the significant illumination difference between DOM and landing images.

[0030] Preferably, after determining the key frames of the descent image sequence and the optimal sub-base map, the key frames of the descent image sequence and their adjacent images contain a sufficient number of connection points projected to the optimal sub-base map; the aerial triangulation method is used to calculate the posture of the input image and the projections of several connection points on the corresponding image and their three-dimensional coordinates, and the projections of the connection points on the optimal sub-base map can be calculated by considering the posture of the optimal sub-image and the positions of the connection points in the world coordinate system, and finally the longitude and latitude of the landing point are determined according to its physical coordinate system.

[0031] Compared with the prior art, the present invention has the following advantages:

[0032] (1) The present invention designs an algorithm for matching adaptive frame extraction through structural similarity (SSIM) between landing images of an extraterrestrial probe during its descent, and obtains the probe pose sequence by extracting feature points and matching algorithms under small sample conditions, thereby improving computational efficiency;

[0033] (2) The present invention achieves high-robustness map feature point matching in a wider range of scenarios and realizes high-precision positioning calculation through pre-site selection and matching algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is a flow chart of the method of the present invention;

[0035] Figure 2 It is the software interface of the proposed framework of the present invention. DETAILED DESCRIPTION

[0036] In view of the significance of precise landing point positioning for spacecraft landing missions and the dependence of traditional landing point positioning algorithms on data sources and the limitations of their scope of application, the present invention proposes a method to address the needs for rapid implementation of high generalization of extraterrestrial landing positioning under small sample conditions.

[0037] In the first aspect, an algorithm for selecting landing image key frames is designed, the purpose of which is to obtain the landing image with the closest physical resolution to the base map in the landing sequence, so as to better match heterogeneous images. In order to further improve the efficiency of the algorithm, the present invention applies an adaptive frame extraction algorithm in this step. Thereafter, the sequence after frame extraction is matched with homologous data and trajectory reconstruction. Through physical constraints, this method converts the relative height between trajectories into the altitude relative to the lunar surface, and calculates the physical resolution in combination with camera parameters to select the landing image key frames.

[0038] Secondly, we designed an algorithm for selecting the best sub-base map, which aims to determine the coordinates of the sub-base map where the landing area is located. Based on the key frame of the landing image, we divided the large-scale base map into blocks and performed cyclic matching, and used the confidence score as a matching measure to select the best sub-base map. For the matching algorithm, we used a combination of preprocessing and deep learning to obtain more robust matching feature points.

[0039] Thirdly, we designed a rough positioning algorithm for the landing point, which mainly includes the landing point transfer between landing sequences and the landing point transfer of heterogeneous data. The former obtains the landing point coordinates on the key frame of the landing image, and the latter obtains the landing point coordinates with longitude and latitude information on the base map. The matching algorithm is consistent with the previous one, and the matching feature points are obtained by combining preprocessing and deep learning.

[0040] Fourthly, a landing point precise positioning algorithm is designed, and the best sub-base map is added to the descending image sequence for cluster adjustment. The position and posture of the best sub-base map relative to the descending sequence is obtained by continuously optimizing the reprojection error of feature matching points in different image frames. The landing point is directly projected onto the best sub-base map according to the position and posture, thereby obtaining the longitude and latitude coordinates of the precise positioning point.

[0041] The technical implementation scheme is described as follows.

[0042] Step 1: Obtain a large area DOM (length and width are both 10,000 meters) of the spacecraft landing preset range and a spacecraft landing and descent image sequence (usually more than 500 frames).

[0043] Step 2: Preprocess the large-area DOM and the spacecraft landing and descent image sequence: including dividing the DOM into small blocks and adaptively extracting frames from the spacecraft landing and descent image sequence. Finally, a small-area DOM (sub-base map) with a physical resolution of kilometers at a certain pixel resolution (such as 1080P, 4K, etc.) is formed. In addition, according to the adaptive extraction results, a descent image sequence of a certain proportion (such as 1 / 10, 1 / 5) of the original sequence is obtained for subsequent trajectory reconstruction tasks.

[0044] The adaptive frame extraction method for the image sequence captured by the landing camera is characterized in that the number of landing sequence frames can be reduced through a specific frame extraction strategy, thereby accelerating the trajectory reconstruction and landing point positioning process; the input parameters of the adaptive frame extraction algorithm include frame interval and similarity threshold. The frame interval is used to accelerate the algorithm and specifies the number of intermediate frames skipped between adjacent extracted frames; the similarity threshold is a parameter used to adaptively adjust the frame extraction interval, and the algorithm outputs a frame sequence for trajectory reconstruction; when the similarity is low, the interval is reduced until the interval is 1; the algorithm uses the structural similarity index (SSIM) as an indicator to measure the similarity between adjacent landing images; SSIM is based on the mathematical framework of signal processing and contains three main components: brightness (average intensity of pixels), contrast (dynamic range of pixel intensity) and structure (spatial pattern of pixel intensity); as shown in formula (1), where x and y are two non-negative image signals, C1 and C2 are stability constants used to avoid the situation where the denominator is zero. μ x and μ y are the mean values ​​of image x and y, respectively, representing brightness information. x and σ y are the standard deviations of images x and y, respectively, representing contrast information. xy is the covariance of images x and y, representing structural information. l(x,y) is a brightness similarity metric, which is used to measure the similarity between the average brightness of two images. c(x,y) is a contrast similarity metric, which is used to measure the similarity between the contrasts of two images. s(x,y) is a structural similarity metric, which is used to measure the similarity between the structures of two images. SSIM(x,y) is the combined result of brightness, contrast, and structural similarity metrics, and is the structural similarity index between two images. The adaptive frame extraction algorithm effectively saves computing resources while ensuring the accuracy of trajectory reconstruction and selection of key frames from landing images.

[0045]

[0046] Step 3: Calculation of the height resolution of the descending image sequence: Using the descending image sequence after frame extraction, the homography matrix between adjacent frames is obtained through the matching algorithm, and the matching results of the image sequence are optimized through bundle adjustment. The relative pose of the camera between frames is restored using the optimized homography matrix. Considering that the camera is stationary relative to the spacecraft, the relative pose is the pose of the spacecraft. In addition, the position of the spacecraft on the vertical lunar surface height axis is determined through the physical trajectory constraints of the spacecraft landing process to obtain the estimated height of each frame and calculate the corresponding resolution of each frame using the camera parameters.

[0047] The characteristic of the height resolution estimation of the descent image sequence is that the positioning of the landing point depends on the match between the landing image and the DOM. Overcoming the difference in physical resolution between the landing image and the DOM is one of the main challenges. The physical resolution of the landing image sequence decreases from large to small, so it is necessary to find a key frame close to the DOM resolution to bridge the resolution gap. After obtaining the external parameters of the camera corresponding to each image frame in the early stage, the relative position can be estimated and the relative trajectory can be reconstructed. After obtaining the relative trajectory, the physical information of the landing image is integrated to estimate the true height of each camera position relative to the surface of the extraterrestrial body. The physical information connotation is that in the final stage of landing, the lander does not perform any horizontal maneuvers and maintains a vertical direction with the surface of the extraterrestrial body. Therefore, in order to estimate the true height of each camera position, the camera position can be projected onto the vertical axis relative to the surface of the extraterrestrial body. Specifically, each camera position p is defined t (t is a subscript representing each frame. For example, the height of the 2nd and 5th frames is as shown in formula (2):

[0048]

[0049] Among them, p e-2 and p e-5 are the 3D position vectors of the camera position at frames 2 and 5 before landing, respectively; these frames were chosen to ensure that there was no significant horizontal maneuvering relative to the surface of the extraterrestrial body, making the landing essentially vertical. The vector z represents the height axis perpendicular to the surface of the extraterrestrial body in the current coordinate system; h ≤ 0 is the projection of the initial drop frame camera position on the height axis and is considered as a relative height. y The constant represents the absolute altitude at the start of the descent, which is a predetermined known value.

[0050] After determining the true height, the physical resolution of the landing image is calculated using the camera parameters. The calculation formula for the physical resolution is as follows:

[0051]

[0052] Among them, R x and R y are the physical resolutions in the x and y directions respectively. x and θ y Respectively represent the field of view in the x and y directions, and pix x and pix y They represent the pixel resolution in the x and y directions of the camera respectively; h is the height of the current descent sequence image. Taking into account the tolerance of the resolution estimation result and the extremely small ratio of the landing height to the radius of the extraterrestrial object, the curvature of the extraterrestrial object is not considered in the height estimation process; after the above calculation, the key frame of the landing image can be selected according to the known DOM resolution.

[0053] Step 4: Select the descending image sequence key frame and select the best sub-base map: According to the physical resolution of each frame, select the descending image closest to the DOM physical resolution as the descending image sequence key frame. By introducing the feature extraction and matching algorithm of preprocessing and deep learning, the matching score is obtained by exhaustively matching the descending image key frame and all sub-base maps. The sub-base map with the highest matching score is the best sub-base map.

[0054] The lunar surface image matching and DOM screening algorithm is characterized in that after determining the landing image key frame, the key frame of the landing image sequence can be used to obtain the best sub-base map in a small range for subsequent positioning; the pre-trained deep learning networks SuperPoint and SuperGlue are used to identify the sub-image containing the landing point by evaluating the matching score between the key frame of the landing image sequence and each sub-base image. The sub-image containing the landing point is called the best sub-base map here; the SuperPoint network extracts key points from the matching image pairs, and then the SuperGlue network matches the key points in pairs and evaluates the matching score; the pre-defined matching score and the threshold of the number of matched key point pairs are used as evaluation criteria, and the key point pairs exceeding the score threshold are regarded as matching point pairs; the sub-images with the number of matching key points exceeding the number threshold are regarded as potential best sub-images, and the potential best sub-images are ranked according to the average matching score to select the best sub-image;

[0055] During the matching process: To deal with the false matching caused by angle differences, each sub-image is rotated from 0 to 360 degrees at a certain angle interval; due to the significant illumination difference between DOM and landing images, histogram normalization is applied to the landing image keyframes with DOM as the reference image.

[0056] Step 5: Coarse positioning of the landing point. The coarse positioning in this framework depends on the transfer of the homography matrix sequence between image frames, including the homography matrix between the descending image sequence and the homography matrix between the descending image key frame and the best DOM. The process is to use the same preprocessing and deep learning feature point extraction and matching algorithm as in step 4 to perform inter-frame matching on the descending sequence images, and obtain the matching point pairs between the descending images. Combined with the matching point pairs between the descending image key frame and the best sub-base map, the landing point can be transferred from the last descending image frame to the key frame through the homography transformation, and then to the best sub-base map, thereby obtaining the coarse positioning latitude and longitude coordinates of the landing point.

[0057] The landing point rough positioning algorithm is characterized in that after determining the descent image key frame and the best sub-base map, the inter-frame homography matrix can be obtained through the image matching algorithm, and the camera posture can be transferred.

[0058] The camera pose transfer algorithm is characterized in that the transfer process includes two major stages. The first is the homologous image matching from the center point of the last frame descent image as the landing point to the descent image key frame, and the second is the heterologous image matching between the descent image key frame and the best sub-base image. Since there are significant illumination differences between the latter image pairs, histogram normalization is applied to the descent image key frame with DOM as the reference image.

[0059] Step 6: Precise positioning of the landing point. Compared with coarse positioning, the precise positioning adds an optimization method for inter-sequence bundle adjustment. The best sub-base map is added to the descending image sequence for bundle adjustment. The position of the best sub-base map relative to the descending sequence is obtained by continuously optimizing the reprojection error of the same-name points. The landing point is directly projected onto the best sub-base map according to the position, thereby obtaining the longitude and latitude coordinates of the precise positioning point.

[0060] The landing point precise positioning algorithm is characterized in that, after determining the descent image key frame and the best sub-base map, the descent image sequence key frame and its adjacent images contain a sufficient number of connection points projected to the best sub-base map, i.e., matching point pairs; aerial triangulation is used to calculate the posture of the input image (descent sequence image) and the projections of several connection points on the corresponding image and their three-dimensional coordinates; considering the posture of the best sub-image and the position of the connection points in the world coordinate system, the projections of the connection points on the best sub-base map can be calculated, and finally the longitude and latitude of the landing point are determined according to its physical coordinate system.

[0061] Example:

[0062] The specific application of the present invention is the landing of a lunar probe in the South Pole-Aitken Basin on the far side of the moon. The entire specific process can be found in Figure 1 .

[0063] Pre-organize the image data. After obtaining the lunar probe descent image and the base map with longitude and latitude of the landing point, the base map is divided into a small area to a certain pixel resolution, and an adaptive descent image frame extraction algorithm is used for the descent image sequence. The redundant data is removed by manually setting the descent image start and end frames and the frame extraction interval, and the frame extraction interval is flexibly adjusted according to the similarity index SSIM between descent images to achieve the self-adaptation of the frame extraction algorithm.

[0064] Landing trajectory reconstruction and height resolution estimation. Use the aerial triangulation method to perform feature matching and bundle adjustment on the descent image sequence after frame extraction to obtain the position and pose of the camera corresponding to the descent image. Based on the relative position and pose, integrate the physical information from the descent image to estimate the true height of each camera attitude relative to the lunar surface and the physical resolution of the image, and then obtain the descent image key frames that are closest to the base map resolution in turn.

[0065] Determination of the best sub-base map. Introducing deep learning feature extraction and matching algorithms, the matching score is obtained by exhaustively matching the descending image keyframes with all sub-base maps. The sub-base map with the highest matching score is the best sub-base map.

[0066] Coarse positioning algorithm for the landing point. Using the feature point extraction and matching algorithm, the SuperPoint algorithm is first used to extract feature points from the descent image sequence, and the SuperGlue feature matching algorithm is used to find matching point pairs between images. At the same time, feature point matching is performed on the descent image key frame and the best sub-base map to obtain matching point pairs between them. On the premise of ensuring a sufficient number of matching point pairs, points with lower matching scores are screened out to ensure the accuracy and robustness of the homography matrix calculation. Next, the landing point is transferred from the last frame of the descent image to the key frame using the calculated homography transformation matrix, and then further transferred to the best sub-base map. Finally, based on the landing point position transferred to the best sub-base map, the geographic coordinate information of the sub-base map is used to obtain the coarse positioning latitude and longitude coordinates of the landing point.

[0067] Precise positioning algorithm for the landing point. First, the best sub-base map is added to the descending image sequence to make it part of the entire sequence. Next, bundle adjustment is performed to gradually correct and adjust the relative position and posture of the entire image sequence by continuously optimizing the reprojection error of feature matching points in different image frames, thereby obtaining the precise position and posture of the best sub-base map relative to the descending sequence. After completing these steps, the landing point is directly projected onto the best sub-base map according to the optimized posture. In this way, the longitude and latitude coordinates of the precise positioning point are finally obtained.

[0068] Figure 2 The software interface of the proposed framework is designed in the present invention, and the software interface includes modular function windows such as data management, preprocessing, coarse positioning and fine positioning.

[0069] This embodiment is a good example of the application of the present invention, but is not limited to this. The extraterrestrial bodies to which this system is applicable include but are not limited to the moon and Mars. Other technicians in this field can make different extensions and changes based on the structure described in the present invention, but as long as they do not deviate from the invention described in the present invention, they are all within the protection scope of the present invention.

[0070] The contents not described in detail in the specification of the present invention belong to the prior art known to the professional and technical personnel in this field.

Claims

1. A highly generalizable landing point positioning method based on a small sample of extraterrestrial terrain images, characterized in that include: Obtain the DOM of a large area within the preset range of spacecraft landing and the original descent sequence images of the spacecraft landing; Divide the large area DOM into small blocks, which are called sub-base maps; Adaptively extract frames from the spacecraft landing and descent sequence images to obtain the descent sequence images after extracting frames, which are recorded as the descent image sequence and are subsequently processed accordingly; Calculate the homography matrix between adjacent frames of the descent image sequence, and use the homography matrix to restore the relative position of the camera between frames; obtain the estimated height of each frame through the physical trajectory constraints of the spacecraft landing process, and use the camera parameters to calculate the resolution corresponding to each frame in the descent image sequence; According to the physical resolution corresponding to each frame of the descending image sequence, the descending image closest to the DOM physical resolution is selected as the descending image key frame; Exhaustively match the descending image keyframe with all sub-base images to select the best sub-base image; Perform inter-frame matching on the descent image sequence to obtain matching point pairs between adjacent descent sequence images, combine the matching point pairs of the descent image key frame and the best sub-base map, obtain the inter-frame homography matrix through the image matching algorithm, and transfer the landing point from the last descent image frame to the descent image key frame through the homography transformation, and then transfer it to the best sub-base map to obtain the coarse positioning latitude and longitude coordinates of the landing point; The best sub-base map is added to the descent image sequence for bundle adjustment. The position and posture of the best sub-base map relative to the descent sequence are obtained by continuously optimizing the reprojection error of the same-name points. The landing point is directly projected onto the best sub-base map according to the position and posture, thereby obtaining the longitude and latitude coordinates of the precise positioning point.

2. The method for locating a landing point based on a small sample of extraterrestrial body terrain images according to claim 1, characterized in that: The length and width of the large-area DOM initially acquired are both at the level of 10,000 meters, and the original descent sequence images of the spacecraft landing are no less than 500 frames.

3. The method for locating a landing point based on a small sample of extraterrestrial body terrain images according to claim 1, characterized in that: When adaptive frame extraction algorithm is used to adaptively extract frames from the original descent sequence images of the spacecraft landing, the input parameters include frame interval and similarity threshold, and the frame sequence used for trajectory reconstruction is output; wherein: the frame interval is used to accelerate the algorithm and specifies the number of intermediate frames skipped between adjacent extracted frames; the similarity threshold is a parameter used to adaptively adjust the frame extraction interval; The structural similarity index SSIM is used as an indicator to measure the similarity between adjacent landing images: when the structural similarity index SSIM between adjacent extracted frames meets the indicator requirements, frame extraction ends; The calculation method of the structural similarity index SSIM is as follows: Where x and y are two non-negative image signals, C1 and C2 are stable constants used to avoid the denominator being zero; μ x and μ y are the means of images x and y, representing brightness information; σ x and σ y are the standard deviations of images x and y, representing contrast information; xy is the covariance of images x and y, representing structural information; l(x, y) is a brightness similarity measure, used to measure the similarity between the average brightness of two images; c(x, y) is a contrast similarity measure, used to measure the similarity between the contrasts of two images; s(x, y) is a structural similarity measure, used to measure the similarity between the structures of two images; SSIM(x, y) is the comprehensive result of brightness, contrast and structural similarity measures, and is the structural similarity index between two images.

4. The method for locating a landing point based on a small sample of extraterrestrial body terrain images according to claim 1, characterized in that: When calculating the homography matrix between adjacent frames of the descending image sequence after frame extraction, a matching algorithm is used and the matching results are optimized through bundle adjustment to obtain the optimized homography matrix, based on which the relative position and posture of the camera between frames are restored.

5. The method for locating a landing point based on a small sample of extraterrestrial body terrain images with high generalization according to claim 1, characterized in that: Through the physical trajectory constraints of the spacecraft landing process, the estimated height of each frame is obtained, specifically: In the final stage of landing, the lander does not perform any horizontal maneuvers and maintains a perpendicular direction to the surface of the extraterrestrial body. At this time, the camera position can be projected onto the vertical axis relative to the surface of the extraterrestrial body, and the estimated height of each frame can be obtained by combining the relative position of the camera between frames.

6. The method for locating a landing point based on a small sample of extraterrestrial body terrain images according to claim 1, characterized in that: The method of using camera parameters to calculate the corresponding resolution of each frame of the descending image sequence is: Among them, R x and R y are the physical resolutions in the x and y directions respectively; θ x and θ y Respectively represent the field of view in the x and y directions, and pix x and pix y They represent the pixel resolution in the x and y directions of the camera respectively; h is the height of the current descending sequence image.

7. The method for locating a landing point based on a small sample of extraterrestrial body terrain images according to claim 1, characterized in that: When exhaustively matching the descending image keyframe with all sub-basemaps to select the best sub-basemap: Using pre-trained deep learning networks SuperPoint and SuperGlue, the SuperPoint network extracts key points from image pairs, and then the SuperGlue network performs pairwise matching of key points and evaluates the matching scores; The predefined matching score and the threshold of the number of matching key point pairs are used as evaluation criteria. The key point pairs exceeding the score threshold are regarded as matching point pairs; the sub-images whose number of matching key points exceeds the threshold are regarded as potential best sub-images, and the potential best sub-images are sorted according to the matching scores, and the image with the highest matching score is selected as the best sub-image.

8. The method for locating a landing point based on a small sample of extraterrestrial body terrain images with high generalization according to claim 1, characterized in that: During the image matching process: The matched images are rotated at certain angle intervals to avoid mismatching caused by angle differences; There is a significant illumination difference between DOM and landing images. Taking DOM as the reference image, histogram normalization is applied to the keyframes of the landing image to avoid the significant illumination difference between DOM and landing images.

9. The method for locating a landing point based on a small sample of extraterrestrial body terrain images with high generalization according to claim 1, characterized in that: After determining the key frames of the descent image sequence and the best sub-base map, the key frames of the descent image sequence and their adjacent images contain a sufficient number of connection points projected to the best sub-base map; the aerotriangulation method is used to calculate the posture of the input image and the projections of several connection points on the corresponding image and their three-dimensional coordinates. Considering the posture of the best sub-image and the positions of the connection points in the world coordinate system, the projections of the connection points on the best sub-base map can be calculated, and finally the longitude and latitude of the landing point are determined according to its physical coordinate system.

Images