A lunar polar region sequential image navigation method based on shadow features

Abstract

The application discloses a lunar polar region sequence image navigation method based on shadow features. Two adjacent sequence images are binarized, stable shadow regions are extracted and subjected to morphological erosion treatment, each connected domain formed by the shadow regions is obtained and a centroid thereof is calculated; each connected domain is screened by using the area size of the connected domain and the distance between the shadow centroid and the image edge, stable shadow features are obtained, and a descriptor is constructed therefrom; the shadow features are matched by using the descriptor, and the matching results are judged to reserve the centroid meeting the conditions as the final matching result; the relative position and pose of an aircraft between two adjacent images are solved by using a homographic matrix decomposition according to the matching results of the shadow features on the two adjacent images; and the real physical scale information of the relative position between the sequence images is recovered according to prior height information of a laser sensor; the above steps are repeated for two adjacent images in sequence images, and the position information of the aircraft at each imaging point is obtained.

Description

Technical Field

[0001] This invention relates to a lunar polar region sequence image navigation method based on shadow features, belonging to the field of visual navigation. Background Technology

[0002] Currently, high-precision landing exploration of extraterrestrial bodies generally employs visual navigation methods. Visual navigation positioning methods can be divided into two types based on whether prior data is used: one is a terrain-relative navigation method based on prior planetary surface topography (or images). This method obtains the spacecraft's position and attitude information by matching features between known planetary surface topography maps and images captured by the spacecraft's navigation camera; the other, without prior terrain data, matches adjacent frames of a sequence of navigation images captured by the spacecraft to obtain its relative position and attitude information. Due to limitations in the accuracy and resolution of prior terrain data, the sequential image frame-by-frame navigation method, which does not rely on prior data, has a wider range of applications. During sequential image frame-by-frame navigation, it is necessary to extract matching features between adjacent frames. This process presents the following problems:

[0003] (1) The low solar altitude angle in the lunar polar region results in a lot of shadow distribution in the captured images, which leads to insufficient stability and robustness of feature extraction and matching from the sequence images;

[0004] (2) The algorithm takes a long time, which is not conducive to real-time navigation. Summary of the Invention

[0005] The technical problem solved by this invention is to overcome the shortcomings of the prior art and provide a lunar polar region sequence image navigation method based on shadow features, which can improve the stability and matching robustness of feature extraction, while improving computational efficiency and meeting the needs of real-time navigation.

[0006] The technical solution of this invention is: a lunar polar region sequential image navigation method based on shadow features, comprising:

[0007] S1 binarizes two adjacent frame sequences of images at the start of navigation and extracts the shadow areas of the images;

[0008] S2 performs morphological erosion on the obtained shadow region and calculates the centroid of each connected region formed by the eroded shadow region.

[0009] S3 uses the area of ​​each connected region formed by the shadow region and the distance between the shadow centroid and the image edge as filtering conditions to filter each connected region and obtain stable shadow features.

[0010] S4 constructs a shadow feature descriptor for each stable shadow feature, resulting in a shadow feature description matrix for each shadow feature;

[0011] S5 uses the shadow feature description matrix to match the shadow features of two adjacent frame sequence images, and further filters the matching results, retaining the shadow feature matching results that meet the conditions as the final matching results;

[0012] S6 uses the final matching results and the homography matrix decomposition method to calculate the relative pose of the aircraft between two adjacent frames; and recovers the true physical scale information of the relative position between the sequential images based on the prior height information of the laser sensor.

[0013] S7 repeats the S1-S6 process for each subsequent two adjacent frames in the sequence image to obtain the position information of the spacecraft at each imaging point.

[0014] Preferably, the specific method for binarizing two adjacent frame sequence images is as follows:

[0015] The image is divided into blocks using a grid method. The Otsu method is used to determine the binarization threshold within each grid block. Areas smaller than the threshold are defined as shadow areas, thus obtaining a binarized image.

[0016] Perform a pyramid operation on the binarized image to obtain a multi-layered pyramid image;

[0017] The shadows are preserved in images of multiple pyramid layers.

[0018] Preferably, a corresponding shadow feature description matrix is ​​constructed for each shadow feature, and the construction method is as follows:

[0019] For a connected region containing a certain shaded feature A, first take its centroid A as the center of a circle, and let B be the centroid of the connected region containing the shaded feature closest to A, and the distance between them be r; establish concentric rings with r as the minimum radius and R as the maximum radius, where the value of R is an integer multiple of r, and the number of concentric rings is g = R / r;

[0020] Using the vector from centroid A to centroid B as the +X axis and the +Y axis as the direction of a 90° clockwise rotation around centroid A, the concentric rings are divided into four quadrants of the coordinate system, with each ring divided into four sub-rings.

[0021] Record the shadow features falling into each sub-ring: when a shadow feature falls into the sub-ring, the sub-ring is recorded as 1, and when no shadow feature falls into the sub-ring, the sub-ring is recorded as 0, thus forming a g×4 shadow feature description matrix, where g represents the ring number of the sub-ring and 4 represents the four quadrants of the ring.

[0022] Preferably, the specific method for matching shadow features is as follows:

[0023] S5-1. For a certain shadow feature i in the first image, subtract its shadow feature description matrix from the shadow feature description matrix of shadow feature j in the second image. The positions with 0 in the subtracted matrix are scored as G1; the positions with -1 or 1 are scored as G2. Finally, the total score of all positions in the subtracted matrix is ​​calculated as the matching score.

[0024] S5-2. Use the method of S5-1 to traverse the shadow feature description matrix of other shadow features in the second image, select and record the shadow features corresponding to the shadow feature description matrix with a total score greater than 85%×g×4×G1, and use them as candidate matching features for the current shadow feature.

[0025] S5-3. Repeat the methods of S5-1 to S5-2 to traverse all shadow features on the first image and obtain all candidate matching features of all shadow features in the first image in the second image.

[0026] Preferably, G1 is a positive integer, G2 is a negative integer, and G1 = 2 × |G2|.

[0027] Preferably, when further filtering the matching results, the reprojection error method is first used to perform preliminary filtering on the matching results of all shadow features, and finally RANSAC is used to remove mismatches from the shadow feature matching pairs retained after preliminary filtering.

[0028] Preferably, the preliminary screening process using the reprojection error method is as follows:

[0029] Based on the criterion that the distance between all pairs of shadow features in the first image should be proportional to the distance between the corresponding pairs of matching features in the second image, H pairs of shadow feature matching pairs are randomly selected from all matching features.

[0030] The initial homography between two images is solved using the selected H pairs of shadow feature matching pairs. The reprojection error of all shadow features is then solved using the initial homography. When the reprojection error is less than K pixels, the shadow features are considered to be a correct match and are retained; otherwise, they are discarded.

[0031] The values ​​of H and K are determined based on navigation mission indicators.

[0032] Compared with the prior art, the present invention has the following advantages:

[0033] (1) The present invention extracts the shadow region by means of the grid pyramid method, which effectively improves the stability of shadow feature extraction;

[0034] (2) This invention improves the accuracy and robustness of feature matching by constructing a shadow feature descriptor with scale and rotation invariance. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the algorithm of the present invention;

[0036] Figure 2 This is a schematic diagram illustrating the effect of the shadow feature preprocessing process of the present invention; where a is the generated binary image, b is the shadow image after morphological processing, and c is a schematic diagram of the shadow connectivity and centroid extraction results.

[0037] Figure 3 This is a schematic diagram of the shadow feature filtering results of the present invention; where a is a schematic diagram of the effect before filtering, and b is a schematic diagram of the effect after filtering.

[0038] Figure 4 This is a schematic diagram illustrating the construction of the shadow feature descriptor of the present invention;

[0039] Figure 5 This is a schematic diagram of the matching score determination process of the present invention. Detailed Implementation

[0040] The present invention will be further described below with reference to the accompanying drawings. The process of performing inter-frame matching of sequential images using shadow features includes the following steps:

[0041] 1. In the process of inter-frame matching of sequential images using shadow features, a grid pyramid method is used to binarize the navigation image and extract the shadow region of the image in order to obtain stable shadow features. The specific method is as follows:

[0042] First, the image is divided into blocks using a grid method. For each block, Otsu's method is used to determine the binarization threshold; areas below the threshold are considered shadow regions. Furthermore, due to scale differences between navigation images, a pyramid operation is performed to ensure the stability of extracted shadows across different scales, retaining shadows that are stably extracted at all scales. Each image layer is divided into a 3×3 grid, and Otsu's method is used for binarization. This paper sets up a three-layer pyramid, downsampling the image at a 0.5x sampling interval, and only retaining shadow regions that are stably present in all three pyramid layers and corresponding to the original image.

[0043] 2. For example Figure 2 As shown, morphological erosion is performed on the shadow region obtained in step 1 to remove smaller shadow regions, thus achieving the purpose of segmenting unstable shadows. After the erosion process is completed, the area of ​​the connected region containing the remaining shadow region and its centroid are calculated to form the shadow feature.

[0044] 3. The shadow features are filtered using the size of the connected component area and the distance between the shadow centroid and the image edge. The relevant parameter settings and process are as follows:

[0045] Set the minimum area of ​​connected components to 100 pixels and the maximum to 20,000 pixels, and remove shadow features that are less than 100 pixels away from the image edge. The filtering effect is as follows. Figure 3 As shown.

[0046] 4. After obtaining stable shadow features, it is necessary to construct a descriptor for each shadow feature, i.e., a shadow feature description matrix. The construction process mainly consists of the following steps:

[0047] (1) For shadow feature A, first take the centroid of A as the center of the circle, let the centroid of the shadow feature closest to A be B, and the distance between them be r. With r as the minimum radius and R as the maximum radius, the value of which is an integer multiple of r, establish concentric rings. The number of concentric rings is g = R / r; in the embodiment provided by the present invention, R = 5r.

[0048] →

[0049] (2) With BA as the X-axis and +Y as the +X-axis, rotate 90° counterclockwise around the centroid B. Clockwise rotation is the positive direction. Figure 4 As shown, the concentric rings are divided into four sub-rings according to the four quadrants of the coordinate system: 0°–90°, 90°–180°, 180°–270°, and 270°–360°. The number of shadow features falling into each sub-ring is calculated; when a shadow feature falls into a sub-ring, it is recorded as 1; when no shadow feature falls into a sub-ring, it is recorded as 0, thus forming a g×4 shadow feature description matrix, where g represents the ring number of the sub-ring and 4 represents the four quadrants of the ring.

[0050] Finally, the shadow features and their corresponding shadow feature description matrices are obtained.

[0051] 5. After obtaining the shadow features and their feature descriptors on two adjacent frames, a matching score is set as a threshold for matching the shadow features. Then, conditions such as RANSAC and reprojection error are used to judge the candidate matching features, and only the centroids of the shadow features that meet the conditions are retained as the final matching result.

[0052] The matching method is as follows:

[0053] S5-1, such as Figure 5 As shown, for a certain shadow feature i in the first image, its shadow feature description matrix is ​​subtracted from the shadow feature description matrix of shadow feature j in the second image. Positions with a value of 0 in the subtracted matrix are scored 10; positions with a value of -1 or 1 are scored -5. Finally, the total score of all positions is calculated as the matching score.

[0054] S5-2. Use the method in S5-1 to traverse the shadow feature description matrix of other shadow features in the second image, select and record the shadow features corresponding to the shadow feature description matrix with a total score greater than 85% × g × 4 × 10, and use them as candidate matching features for the current shadow feature.

[0055] S5-3. Return to S5-1 and repeat the methods of S5-1 to S5-2 to traverse all shadow features on the first image to obtain all candidate matching features of all shadow features in the first image.

[0056] When judging the matching results, RANSAC and reprojection error are used. The specific method is as follows:

[0057] For all candidate matching features obtained from the two images, the distance between each pair of shaded features in the first image should be proportional to the distance between each pair of matching features in the corresponding second image. Based on this criterion, six pairs of shaded feature matching pairs are randomly selected from all matching features.

[0058] The initial homography between two images is solved using the selected shadow feature matching pairs. The reprojection error of all shadow features is then calculated using the initial homography. If the reprojection error is less than 5 pixels, the shadow feature pair is considered a correct match and retained; otherwise, it is discarded. This process is repeated iteratively until all shadow feature matching pairs have been traversed.

[0059] Finally, RANSAC is used to remove mismatches from the obtained matching point pairs.

[0060] 6. After obtaining the final shadow feature matching results on two adjacent images, the relative pose of the aircraft between two adjacent frames is calculated using homography matrix decomposition; and the true physical scale information of the relative position between the sequential images is recovered based on the prior height information of the laser sensor.

[0061] 7. Repeat steps 1-6 for each subsequent two adjacent frames in the image sequence to obtain the position information of the spacecraft at each imaging point.

[0062] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

Claims

1. A lunar polar region image sequence navigation method based on shadow features, characterized in that... Includes the following steps: S1 binarizes two adjacent frame sequences of images at the start of navigation and extracts the shadow areas of the images; S2 performs morphological erosion on the obtained shadow region and calculates the centroid of each connected region formed by the eroded shadow region. S3 uses the area of ​​each connected region formed by the shadow region and the distance between the shadow centroid and the image edge as filtering conditions to filter each connected region and obtain stable shadow features. S4 constructs a shadow feature descriptor for each stable shadow feature, resulting in a shadow feature description matrix for each shadow feature; S5 uses the shadow feature description matrix to match the shadow features of two adjacent frame sequence images, and further filters the matching results, retaining the shadow feature matching results that meet the conditions as the final matching results; S6 uses the final matching results and the homography matrix decomposition method to calculate the relative pose of the aircraft between two adjacent frames; and recovers the true physical scale information of the relative position between the sequential images based on the prior height information of the laser sensor. S7 repeats the process of S1-S6 for the next two adjacent frames in the sequence image to obtain the position information of the spacecraft at each imaging point. The specific method for binarizing two adjacent frames of images is as follows: The image is divided into blocks using a grid method. The Otsu method is used to determine the binarization threshold within each grid block. Areas smaller than the threshold are defined as shadow areas, thus obtaining a binarized image. Perform a pyramid operation on the binarized image to obtain a multi-layered pyramid image; Shadows that can be extracted from multiple pyramid images are preserved; For each shadow feature, a corresponding shadow feature description matrix is ​​constructed. The construction method is as follows: For a connected region containing a certain shaded feature A, first take its centroid A as the center of a circle, let B be the centroid of the connected region containing the shaded feature closest to A, and let the distance between the two be r; establish concentric rings with r as the minimum radius and R as the maximum radius, where the value of R is an integer multiple of r, and the number of concentric rings is g = R / r. Using the vector from centroid A to centroid B as the +X axis and the +Y axis as the direction of a 90° clockwise rotation around centroid A, the concentric rings are divided into four quadrants of the coordinate system, with each ring divided into four sub-rings. Record the shadow features falling into each sub-ring: when a shadow feature falls into the sub-ring, the sub-ring is recorded as 1, and when no shadow feature falls into the sub-ring, the sub-ring is recorded as 0, thus forming a g×4 shadow feature description matrix, where g represents the ring number of the sub-ring and 4 represents the four quadrants of the ring.

2. The lunar polar region sequence image navigation method based on shadow features according to claim 1, characterized in that: The specific method for matching shadow features is as follows: S5-1. For a certain shadow feature i in the first image, subtract its shadow feature description matrix from the shadow feature description matrix of shadow feature j in the second image. The positions with 0 in the subtracted matrix are scored as G1; the positions with -1 or 1 are scored as G2. Finally, the total score of all positions in the subtracted matrix is ​​calculated as the matching score. S5-2. Use the method of S5-1 to traverse the shadow feature description matrix of other shadow features in the second image, select and record the shadow features corresponding to the shadow feature description matrix with a total score greater than 85%×g×4×G1, and use them as candidate matching features for the current shadow feature. S5-3. Repeat the methods of S5-1 to S5-2 to traverse all shadow features on the first image and obtain all candidate matching features of all shadow features in the first image in the second image.

3. The lunar polar region sequence image navigation method based on shadow features according to claim 2, characterized in that: G1 is a positive integer, G2 is a negative integer, and G1 = 2 × |G2|.

4. The lunar polar region image sequence navigation method based on shadow features according to claim 1, characterized in that: When further filtering the matching results, the reprojection error method is first used to perform preliminary filtering on the matching results of all shadow features, and finally RANSAC is used to remove mismatches from the shadow feature matching pairs retained after preliminary filtering.

5. A lunar polar region sequential image navigation method based on shadow features according to claim 4, characterized in that: The process of preliminary screening using the reprojection error method is as follows: Based on the criterion that the distance between all pairs of shadow features in the first image should be proportional to the distance between the corresponding pairs of matching features in the second image, H pairs of shadow feature matching pairs are randomly selected from all matching features. The initial homography between two images is solved using the selected H pairs of shadow feature matching pairs. The reprojection error of all shadow features is then solved using the initial homography. When the reprojection error is less than K pixels, the shadow features are considered to be a correct match and are retained; otherwise, they are discarded. The values ​​of H and K are determined based on navigation mission indicators.

Images