An asteroid pinpoint landing optical navigation method based on three-dimensional navigation features
By employing a visual navigation method based on three-dimensional terrain features, the problems of large variations in light sources and image distortion on the asteroid surface were solved, achieving high-precision navigation and precise landing, adapting to different viewing angles and light conditions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TONGJI UNIV
- Filing Date
- 2023-06-30
- Publication Date
- 2026-07-21
AI Technical Summary
Existing asteroid visual navigation methods are difficult to adapt to environments with large changes in light source, have low accuracy in 2D navigation feature extraction, and cannot adapt to scenes with severe image distortion, resulting in insufficient accuracy in pinpoint landing.
A visual navigation method based on 3D terrain features is adopted. High-resolution terrain is created under different lighting conditions using the multi-view photometric method, and then rendered in combination with a lighting model. 3D feature vectors are extracted, and feature matching is performed through filters to adapt to different viewpoints and light source changes.
It improves the accuracy of extracting natural landmark features on the asteroid surface, adapts to changes in light source and image distortion, and achieves high-precision navigation and fixed-point landing.
Smart Images
Figure CN116843873B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of asteroid visual navigation, and in particular to an optical navigation method for asteroid fixed-point landing based on three-dimensional navigation features. Background Technology
[0002] Asteroid exploration is a new direction for deep space exploration [1]. Among them, high-precision pinpoint landing system (PLS) is one of the key core technologies to ensure the successful implementation of the entire attachment and sampling mission [2]. Due to the small size of asteroids, short rotation period, many surface obstacles, and difficulty in modeling microgravitational fields, pinpoint landing on the surface of asteroids is one of the most challenging and risky tasks [3]. Based on the successful asteroid attachment and sampling missions to date, visual navigation is an effective high-precision autonomous navigation and positioning method [4,5].
[0003] In asteroid visual navigation, the positioning of the lander is mostly based on feature acquisition and matching of measurement data, and is calculated by geometric calculation methods. Selecting landmarks that can effectively serve navigation is very important for accurate navigation in large-scale global areas and key local areas such as the sampling landing area [6]. Navigation landmarks selected in deep space exploration missions can be divided into natural landmarks and artificial landmarks. The Japanese Hayabusa1 and Hayabusa2 missions adopted a visual navigation positioning scheme assisted by artificial landmarks [7,8]. Easily identifiable navigation landmarks are deployed to the target celestial body during the attachment phase, and the spacecraft's attitude is estimated by tracking the landmarks in real time. This scheme has high overall accuracy, but the cost of carrying artificial landmarks is high and the deployment is uncertain.
[0004] Compared to artificially placed landmarks, the application of natural landmarks distributed on the surface of small celestial bodies is more extensive[9]. Natural landmarks refer to natural features distributed on the surface of planets, such as impact craters, rocks, etc., and may not be a specific type of feature. Among local feature descriptors, operators such as Harris
[10] , SIFT
[11] , SURF
[12] and BRIEF
[13] are less adaptable to changes in the spectrum or direction of the light source, making them difficult to apply to feature-based navigation needs in complex lighting environments of asteroids. In addition to extracting point features from images, some scholars have extracted line segment features from images[14,15], including straight lines, curves, and the edges of features, which are more robust to lighting than point features. However, the description and extraction of line features are more difficult, especially for asteroids with rubble piles, where line features are more difficult to define. In addition, considering that line features do not have in-plane or out-of-plane deformation robustness, they are difficult to apply to navigation landmarks of rapidly rotating asteroids. Furthermore, the aforementioned navigation methods only use two-dimensional navigation features and do not use three-dimensional features, making them difficult to adapt to scenarios where image features are severely distorted between images. Summary of the Invention
[0005] The purpose of this invention is to provide an optical navigation method for asteroid point-to-point landing based on three-dimensional navigation features, which is applicable to images with distortion and large changes in light source, and improves the extraction accuracy of features of natural landmarks distributed on the surface of small celestial bodies.
[0006] The objective of this invention can be achieved through the following technical solutions:
[0007] An optical navigation method for asteroid point-to-point landing based on three-dimensional navigation features, comprising the following steps:
[0008] S1. Based on the geographical location of the selected sampling area and the preset flight trajectory of the aircraft, the spatial distribution and lighting conditions of terrain features are selected as navigation features, and low-resolution terrain that meets the requirements of navigation features is selected.
[0009] S2. Acquire high-resolution images under different lighting conditions captured in actual shooting, and use photometric methods to optimize low-resolution terrain to obtain high-resolution terrain and corresponding albedo information.
[0010] S3. Based on the spacecraft's initial position and attitude, select possible high-resolution terrain from the high-resolution terrain in S2. Render the possible high-resolution terrain using real-time observation status and current time-estimated illumination conditions to obtain image blocks with texture information.
[0011] S4. Train filters based on image patches, and obtain feature vectors that match the texture features of image patches based on the trained filters;
[0012] S5. Obtain the initial position of the terrain features, determine the feature search range of the high-resolution image at the current time based on the initial position, correlate the feature search range with the filter, obtain the response function score of the feature vector, and the position with the largest response function score in the feature search range is the optimal feature matching position, and perform navigation based on the optimal feature matching position.
[0013] Furthermore, the filter training process is as follows:
[0014] The rendered high-resolution terrain is used as the input filter of the training set. The descriptive factors of the multi-features of the rendered high-resolution terrain are extracted. The rendered high-resolution terrain is densely sampled through a cyclic matrix. Ridge regression and kernel trick are used to construct a template matching objective function. The filter is trained iteratively by minimizing the template matching objective function. The trained filter is obtained when the number of iterations reaches a certain number.
[0015] Furthermore, the template matching objective function is:
[0016]
[0017] in, f Indicates a filter. y It is defined as a Gaussian function to train labels. For regularization parameters, w Here is the space regularization coefficient. x As a descriptor, C The feature dimension, c is the dimension index.
[0018] Furthermore, the features of the multi-feature descriptor include: fused image grayscale, HOG features, and graphic information entropy.
[0019] Furthermore, in the process of optimizing low-resolution terrain using photometric methods, optimization is achieved by minimizing the cost function, which is:
[0020]
[0021] in, μ and λ These are the coefficients of the error term. ( x , y ) is the pixel of high-resolution terrain ( x , y albedo, R ( () is a reflectivity model. I ( x , y ) represents the number of pixels in a high-resolution image. x , y The image intensity is z(x, y), where z(x, y) is the elevation of the pixel (x, y) of the high-resolution terrain. The elevation of the pixels (x, y) of the low-resolution terrain.
[0022] Furthermore, the cost function is solved using the ADMM algorithm, which yields high-resolution terrain and corresponding albedo information.
[0023] Furthermore, the relationship between image intensity and reflectance of high-resolution images is as follows:
[0024]
[0025] in,( x , y () represents the pixel position or pixel index in the image. I ( x , y ) represents a pixel ( x , y Image intensity, (x , y ) is a pixel ( x , y albedo; ( x , y () is the image exposure; R ( () is a reflectivity model. Let be the direction vector of the incident ray. Let be the direction vector of the emitted ray. is the normal vector of the reflection point.
[0026] Furthermore, the normal vector of the reflection point is:
[0027]
[0028] in, z x ( x , y ) and z y ( x , y )express z ( x , y The gradient of z is the elevation of pixel (x, y), and z(x, y) is the elevation of pixel (x, y).
[0029] Furthermore, the low-resolution terrain of S1 satisfies the following conditions for the spatial distribution of terrain features: at least three features of the terrain are visible in the field of view of the navigation camera, and are on any trajectory defined by the 3σ trajectory dispersion, where σ is determined by the uncertainty brought about by each maneuver in the tag sequence, plus the prior uncertainty of the spacecraft position.
[0030] Furthermore, the low-resolution terrain of S1 satisfies the following lighting conditions: emission angle less than 45°.
[0031] Compared with the prior art, the present invention has the following beneficial effects:
[0032] (1) This invention proposes a visual navigation method based on three-dimensional terrain features. It uses the multi-view photometric method to create a high-resolution terrain with a resolution comparable to that of the low-resolution terrain and high-resolution images taken under different lighting conditions. Under the set lighting model, the high-resolution terrain is rendered using the lighting and observation conditions at the current moment to obtain navigation features that are consistent with the texture distribution of the images taken at the current moment. Then, matching and navigation are performed. It can adapt to different viewing angles and has strong adaptability to changes in the light source spectrum or light source direction. At the same time, the feature vector of the three-dimensional feature extraction has high accuracy and is suitable for scenes with severe distortion between images.
[0033] (2) The filter of the present invention only requires a small amount of high-resolution terrain to be rendered for training. During training, high-dimensional multi-feature descriptive factors are extracted. In terms of the number of training sets, the convergence speed is fast. Attached Figure Description
[0034] Figure 1 This is a flowchart of the present invention;
[0035] Figure 2 This is a result map generated by the high-resolution terrain and albedo data of the present invention, wherein... Figure 2 (a) is the result diagram under a 40cm topographic relief. Figure 2 (b) is the result image under a 25cm topographic relief. Figure 2 (c) Albedo map;
[0036] Figure 3 These are renderings of the 40cm and 25cm terrain features of the present invention, wherein... Figure 3 (a) is an image. Figure 3 (b) The rendering result for a 40cm terrain. Figure 3 (c) is the rendering result of the terrain at 25cm;
[0037] Figure 4 The image shows the feature matching results for various methods. Detailed Implementation
[0038] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.
[0039] Definitions:
[0040] ADMM: Alternating Direction Multiplier Method;
[0041] DCF: Discriminant Correlation Filter
[0042] HOG: Oriented Gradient Histogram.
[0043] Visual navigation is an effective approach to achieving high-precision navigation and positioning of asteroids. Selecting effective navigation features and accurately matching them is crucial for the success of visual navigation missions. Asteroids are small, rotate rapidly, and exhibit significant changes in perspective between images, resulting in noticeable image feature distortion. Visual navigation methods based on two-dimensional feature extraction and matching are unsuitable for such scenarios. Therefore, this invention proposes a visual navigation method based on three-dimensional terrain features. First, a high-resolution terrain model with comparable resolution is created using a multi-view photometric method based on low-resolution terrain and high-resolution images taken under different lighting conditions. Then, under a defined lighting model, the high-resolution terrain is rendered using the current lighting and observation conditions to obtain navigation features consistent with the texture distribution of the currently captured images. Finally, considering the uncertainty of camera pose estimation, a perspective-adaptive navigation feature matching algorithm based on motion constraints is proposed, allowing for perspective and scale adaptation during asteroid landing and navigation.
[0044] The flowchart of this invention is as follows Figure 1 As shown, it includes the following steps:
[0045] S1. Based on the geographical location of the selected sampling area and the preset flight trajectory of the aircraft, the spatial distribution and lighting conditions of terrain features are selected as navigation features, and low-resolution terrain that meets the requirements of navigation features is selected.
[0046] S2. Acquire high-resolution images under different lighting conditions captured in actual shooting, and use photometric methods to optimize low-resolution terrain to obtain high-resolution terrain and corresponding albedo information.
[0047] S3. Based on the spacecraft's initial position and attitude, select possible high-resolution terrain from the high-resolution terrain in S2. Render the possible high-resolution terrain using real-time observation status and current time-estimated illumination conditions to obtain image blocks with texture information.
[0048] S4. Train filters based on image patches, and obtain feature vectors that match the texture features of image patches based on the trained filters;
[0049] S5. Obtain the initial position of the terrain features, determine the feature search range of the high-resolution image at the current time based on the initial position, correlate the feature search range with the filter, obtain the response function score of the feature vector, and the position with the largest response function score in the feature search range is the optimal feature matching position, and perform navigation based on the optimal feature matching position.
[0050] In S1, low-resolution terrain features need to meet navigation feature requirements. During navigation, key measurements provided to the navigation filter include laser rangefinder values and terrain feature matching results. Therefore, it is crucial that the terrain feature database contains sufficient features to ensure adequate measurements for each image. Furthermore, the features in the catalog must be easily identifiable and have accurate location definitions. The following are the terrain feature selection rules, i.e., the requirements that must be met:
[0051] 1) Spatial Location Distribution: The spatial distribution of terrain features is determined based on the geographical location of the selected sampling area and the preset spacecraft flight trajectory. The spatial location of terrain features is defined as: providing sufficient coverage for each image across the entire preset 3σ flight trajectory (where σ is determined by the uncertainty introduced by each maneuver in the TAG sequence, plus the prior uncertainty of the spacecraft's position). The terrain feature distribution is such that at least three features are visible within the field of view of the navigation camera, and simultaneously lie on any trajectory defined by the 3σ trajectory dispersion.
[0052] 2) Illumination Conditions: Incident angle, emission angle, and feature diffusion are all factors to consider during feature placement. The incident angle, defined as the angle between the asteroid surface normal and the solar vector, is ideally less than 75°. At larger incident angles, shadows will become the dominant image feature, making accurate terrain identification more difficult. There are usually no specific minimum requirements for the incident angle; it is typically determined by the trajectory sampled during the descent phase in the actual task. The emission angle, defined as the angle between the asteroid surface normal and the camera's line-of-sight, also needs to be kept small. Because terrain features are composed of facets, these angles are calculated relative to local planes, which are generally the surface normals. High emission angles can easily lead to edge effects in terrain feature rendering. This imaging condition will affect the accuracy of subsequent terrain feature matching. Through testing during the 3D model creation process, the proposed method found that the emission angle needs to be kept below 60°, ideally below 45°.
[0053] In S2, high-precision terrain feature generation combines low-resolution terrain and high-resolution imagery, using photometry to optimize the shape of local areas and obtain high-precision terrain and albedo information. The essence of photometry is to utilize the relationship between image intensity and reflectance; by minimizing the difference between the two, the terrain can be optimized. Specifically, the relationship between the two can be modeled as follows:
[0054] I(x,y)=Λ(x,y)×ρ(x,y)×R(s(x,y),v(x,y),n(x,y)) (1)
[0055] in,( x , y () represents the pixel position or pixel index in the image.I ( x , y ) represents a pixel ( x , y Image intensity, ( x , y ) is a pixel ( x , y albedo; ( x , y () is the image exposure; R ( This is a reflectivity model. Reflectivity is the ratio of radiation reflected by a surface to radiation incident on that surface. The reflectivity of a point can be modeled as the direction vector of the incident ray. s ( x , y ), the direction vector of the emitted light ray v ( x , y ) and the normal vector of that point n ( x , y The relationship between the normal vectors. Specifically, in the case of orthographic projection, the normal vectors... n ( x , y It can be defined as:
[0056] n(x,y)=[z x (x,y),z y (x,y),-1] T (2)
[0057] in, z x ( x , y ) and z y ( x , y )express z ( x , y The gradient of z(x, y) is the elevation of pixel (x, y). In terms of reflectivity, many empirical models have been constructed, such as Lommel-Seeliger
[19] , Lunar-Lambert
[20] , and the complex Hapke model
[21] .
[0058] In summary, to optimize the 3D shape and albedo information of local regions of an asteroid, the cost function is defined as follows, based on the minimization method:
[0059]
[0060] in, μ and λ These are the coefficients of the error terms. The first term is the brightness constraint, the second term is the smoothness constraint, which is achieved by minimizing the second derivative of the surface; the third term is the elevation constraint, which ensures that the generated terrain is consistent with the initial terrain. Based on the above framework, the ADMM algorithm
[22] is used to solve for high-resolution terrain and albedo.
[0061] The rendering process for S3 is as follows: First, based on the spacecraft's initial position and attitude, possible terrain features covering the current field of view are selected from the feature database. Then, the selected terrain is rendered based on the current observation and lighting conditions to obtain terrain features that approximate the image's perspective.
[0062] In S4, a robust and efficient template feature matching method is proposed to adapt to the deformation of terrain features that occur during asteroid visual navigation when the initial camera position or attitude differs significantly from the actual situation, thereby achieving navigation feature matching for visual navigation.
[0063] The proposed method is primarily based on motion-constrained DCF. Prior information provided by motion constraints between sequential images is used to provide a predicted search region for template feature matching.
[0064] A feature matching method based on correlation filtering is adopted. Considering the complexity and diversity of asteroid surface texture, a descriptive factor integrating multiple features such as image grayscale, HOG features, and image information entropy is established: x = [ x 1, x 2, …, x c ],in, C Let be the feature dimension. Considering the computational efficiency requirements of real-time aircraft navigation and the nonlinearity of image features, dense sampling of online-generated terrain features is performed using a cyclic matrix. Ridge regression and kernel trick are employed to construct a template matching objective function and obtain a kernel correlation filter. The objective function is defined as follows:
[0065]
[0066] in, f For correlation filtering, f = ( f 1, f 2, …, f c ), y It is defined as a Gaussian function to train labels. For regularization parameters, wLet be the space regularization coefficient. By minimizing equation (4), the relevant filter can be obtained. f After obtaining the correlation filter, step S5 is executed to determine the feature search range of the currently captured image based on the initial position of the terrain features provided by the navigation filter. The images within the search range are correlated with the filter, and a correlation score response function is established accordingly. The position with the highest response function score is the optimal feature matching position.
[0067] The following is an analysis of a practical example:
[0068] The proposed method was evaluated and analyzed using a shape model constructed and imagery captured by the NASA OSIRIS-Rex mission. The images used were from the OSIRIS-Rex navigation camera, with a resolution of 25 cm. The shape model used was a 40 cm resolution model constructed from its laser data. The coordinate system used was established by NASA based on ephemeris data. Figure 2 In (a) and (b), the local terrain is shown with resolutions of 40 cm and 25 cm, respectively.
[0069] First, from a qualitative perspective, the overall undulation trend of the optimized terrain at a resolution of 25 cm is consistent with that generated by laser data at a resolution of 40 cm. Furthermore, the terrain optimized using the proposed method exhibits more refined terrain texture. Figure 2 (c) The obtained albedo map shows almost no topographic components, indicating that the albedo calculation is reliable. From a quantitative perspective, the RMSE between the 40cm topographic area and the optimized 25cm topographic area is 0.5378, demonstrating that this method can effectively optimize local topography.
[0070] Furthermore, the initial position state of the images taken by the NavCam camera during the probe's orbital observation on April 21, 2020, and the sunlight information at the time of image capture, were used to render the constructed terrain. One result is shown below. Figure 3 As shown. Figure 3 (a) is an image. Figure 3 (b) and (c) show the rendering results for terrain with a depth of 40cm and 25cm, respectively. Clearly, the optimized terrain texture in this paper is more refined, and the rendered image texture is closer to the real navigation image, which is beneficial for subsequent matching work.
[0071] The proposed feature matching method was evaluated using three consecutive images taken by the navigation camera during spacecraft orbit observations on April 21, 2020. Five features were selected for each image. The proposed feature matching algorithm was compared with several common template matching algorithms, including the Normalized Cross-Correlation (NCC) and Best Partner Similarity (BBS) metrics.
[0072] The template image center is the geometric center of each terrain feature. Considering matching efficiency, this paper defines the template edge length as 99 pixels. Compared to the real image, the terrain features in the template image exhibit varying degrees of feature distortion, especially in areas with strong illumination or large shadows. Furthermore, since the Bennu asteroid is a rubble pile structure with various rubble structures distributed on its surface, the features exhibit structural repetition, posing a challenge to accurate matching. The matching results of these methods are as follows: Figure 4 As shown.
[0073] Furthermore, the matching accuracy of the proposed algorithm and its comparison algorithms is quantitatively evaluated. In the experiment, the actual location of the template image within the image was manually obtained. As shown in Table 1, the experimental results demonstrate that the proposed algorithm is the only one that can correctly match all five features, indicating its robustness. The accuracy verification for the template region uses the correct matching rate. = N um.( Acc .) / N ,in, Acc .Measure the overlap of common bounding boxes, i.e. Acc = area( B e B g ) / area( B e B g ), B e and B g These are the bounding boxes estimated based on the matching results and the manually obtained bounding boxes (as ground truth), respectively. Acc When the value is greater than 0.7, the current match is considered correct. N um.( Acc .) = #( Acc >0.7), N This represents the number of all features to be matched. The quantitative analysis of the method is shown in Table 1, where... The average time for matching each feature is shown in the table. As can be seen from the table, in 15 matching experiments for the five terrain features corresponding to these three images, the algorithm proposed in this paper achieved a 100% matching success rate despite various interference factors. Acc All scores were above 0.95. Compared to NCC, BBS has a relatively higher matching success rate and Acc However, its time efficiency is very low.
[0074] Table 1. Quantitative analysis of matching results of different algorithms
[0075]
[0076] This invention proposes a terrain-feature-based landing navigation and positioning method for future asteroid fixed-sampling-return missions. Experiments show that the proposed method, through real-time rendering of high-precision terrain, can achieve robust template matching with a matching success rate of 95%, and is adaptive to different imaging perspectives and lighting conditions.
[0077] The following are references in this invention:
[0078] [1]Lauretta D, DellaGiustina D, Bennett C, et al. The unexpected surface of asteroid (101955) Bennu. Nature, 2019, 568(7750): 55-60.
[0079] [2]Kubota T, Hashimoto T, Sawai S, et al. An autonomous navigation and guidance system for MUSES-C asteroid landing. Acta Astronautica, 2003, 52(2-6): 125-131.
[0080] [3]Hergenrother C, Maleszewski C, Nolan M, et al. The operational environment and rotational acceleration of asteroid (101955) Bennu from OSIRIS-REx observations. Nature communications, 2019, 10(1): 1291.
[0081] [4]Walsh K, Bierhaus E, Lauretta D, et al. Assessing thesampleability of Bennu’s surface for the OSIRIS-REx asteroid sample returnmission. Space Science Reviews, 2022, 218(4): 20.
[0082] [5]Pugliatti M, Franzese V, Topputo F. Data-driven image processingfor onboard optical navigation around a binary asteroid. Journal ofSpacecraft and Rockets, 2022, 59(3): 943-959.
[0083] [6]Vetrisano M, Vasile M. Autonomous navigation of a spacecraftformation in the proximity of an asteroid. Advances in Space Research, 2016,57(8): 1783-1804.
[0084] [7]Yano H, Kubota T, Miyamoto H, et al. Touchdown of the Hayabusaspacecraft at the Muses Sea on Itokawa. Science, 2006, 312(5778): 1350-1353.
[0085] [8]Watanabe S, Tsuda Y, Yoshikawa M, et al. Hayabusa2 missionoverview. Space Science Reviews, 2017, 208: 3-16.
[0086] [9]Barnouin O, Daly M, Palmer E, et al. Digital terrain mapping bythe OSIRIS-REx mission. Planetary and Space Science, 2020, 180: 104764.
[0087]
[10] Zhang Y, Liu R, Wang X, et al. Boosted binary Harris hawksoptimizer and feature selection. Engineering with Computers, 2021, 37: 3741-3770.
[0088]
[11] Ke Y, Sukthankar R. PCA-SIFT: A more distinctive representationfor local image descriptors. Proceedings of the 2004 IEEE Computer SocietyConference on Computer Vision and Pattern Recognition, 2004. CVPR 2004. IEEE,2004, 2: II-II.
[0089]
[12] Bay H, Ess A, Tuytelaars T, et al. Speeded-up robust features(SURF). Computer vision and image understanding, 2008, 110(3): 346-359.
[0090]
[13] Rublee E, Rabaud V, Konolige K, et al. ORB: An efficientalternative to SIFT or SURF. Proceedings of the 2011 International conferenceon computer vision. IEEE, 2011: 2564-2571.
[0091]
[14] Shao W, Gu T, Ma Y, et al. A novel approach to visual navigationbased on feature line correspondences for precision landing. The Journal ofNavigation, 2018, 71(6): 1413-1430.
[0092]
[15] Simard V, Neveu D, Bruneau S, et al. Pinpoint lunar landingnavigation using crater detection and matching: design and laboratoryvalidation. Proceedings of the AIAA Guidance, Navigation, and ControlConference. 2012: 5032.
[0093]
[16] Lauretta D, Balram-Knutson S, Beshore E, et al. OSIRIS-REx:sample return from asteroid (101955) Bennu. Space Science Reviews, 2017, 212:925-984.
[0094]
[17] Gaskell R, Barnouin‐Jha O, Scheeres D, et al. Characterizing andnavigating small bodies with imaging data. Meteoritics & Planetary Science,2008, 43(6): 1049-1061.
[0095]
[18] Lohse V, Heipke C, Kirk R L. Derivation of planetary topographyusing multi-image shape-from-shading. Planetary and space science, 2006, 54(7): 661-674.
[0096]
[19] Hapke, B. Bidirectional reflectance spectroscopy 4. Theextinction coefficient and the opposition effect. Icarus, 1986, 67, 264-280.
[0097]
[20] Hapke, B, Nelson, R, Smythe, W. The opposition effect of theMoon: the contribution of coherent backscatter. Science, 1993, 260, 509-511.
[0098]
[21] McEwen A S. Photometric functions for photoclinometry and otherapplications. Icarus, 1991, 92(2): 298-311.
[0099]
[22] Ghadimi E, Teixeira A, Shames I, et al. Optimal parameterselection for the alternating direction method of multipliers (ADMM):quadratic problems. IEEE Transactions on Automatic Control, 2014, 60(3): 644-658.
Claims
1. An optical navigation method for asteroid point-to-point landing based on three-dimensional navigation features, characterized in that, The method includes the following steps: S1. Based on the geographical location of the selected sampling area and the preset flight trajectory of the aircraft, the spatial distribution and lighting conditions of terrain features are selected as navigation features, and low-resolution terrain that meets the requirements of navigation features is selected. S2. Acquire high-resolution images under different lighting conditions captured in actual shooting, and use photometric methods to optimize low-resolution terrain to obtain high-resolution terrain and corresponding albedo information. S3. Based on the spacecraft's initial position and attitude, select possible high-resolution terrain from the high-resolution terrain in S2. Render the possible high-resolution terrain using real-time observation status and current time-estimated illumination conditions to obtain image blocks with texture information. S4. Train filters based on image patches, and obtain feature vectors that match the texture features of image patches based on the trained filters; S5. Obtain the initial position of terrain features, determine the feature search range of the high-resolution image at the current time based on the initial position, correlate the feature search range with the filter, obtain the response function score of the feature vector, and the position with the largest response function score in the feature search range is the optimal feature matching position. Navigation is performed based on the optimal feature matching position. The filter training process is as follows: The rendered high-resolution terrain is used as the input filter of the training set. The descriptive factors of the multi-features of the rendered high-resolution terrain are extracted. The rendered high-resolution terrain is densely sampled through a cyclic matrix. Ridge regression and kernel trick are used to construct a template matching objective function. The filter is trained iteratively by minimizing the template matching objective function. The trained filter is obtained when the number of iterations reaches a certain number.
2. The optical navigation method for asteroid point-to-point landing based on three-dimensional navigation features according to claim 1, characterized in that, The template matching objective function is: in, f Indicates a filter. y It is defined as a Gaussian function to train labels. For regularization parameters, w Here is the space regularization coefficient. x As a descriptor, C The feature dimension, c is the dimension index.
3. The optical navigation method for asteroid point-to-point landing based on three-dimensional navigation features according to claim 1, characterized in that, The features of the multi-feature descriptor include: fused image grayscale, HOG features, and graphic information entropy.
4. The optical navigation method for asteroid point-to-point landing based on three-dimensional navigation features according to claim 1, characterized in that, In the process of optimizing low-resolution terrain using photometric methods, optimization is achieved by minimizing the cost function, which is: in, μ and λ These are the coefficients of the error term. ( x , y ) is the pixel of high-resolution terrain ( x , y albedo, R ( () is a reflectivity model. I ( x , y ) represents the number of pixels in a high-resolution image. x , y The image intensity is z(x, y), where z(x, y) is the elevation of the pixel (x, y) of the high-resolution terrain. The elevation of the pixels (x, y) of the low-resolution terrain.
5. The optical navigation method for asteroid point-to-point landing based on three-dimensional navigation features according to claim 4, characterized in that, The ADMM algorithm is used to solve the cost function, which yields high-resolution terrain and corresponding albedo information.
6. The optical navigation method for asteroid point-to-point landing based on three-dimensional navigation features according to claim 4, characterized in that, The relationship between image intensity and reflectance in high-resolution images is as follows: in,( x , y () represents the pixel position or pixel index in the image. I ( x , y ) represents a pixel ( x , y Image intensity, ( x , y ) is a pixel ( x , y albedo; ( x , y () is the image exposure; R ( () is a reflectivity model. Let be the direction vector of the incident ray. Let be the direction vector of the emitted ray. is the normal vector of the reflection point.
7. The optical navigation method for asteroid point-to-point landing based on three-dimensional navigation features according to claim 6, characterized in that, The normal vector of the reflection point is: in, z x ( x , y ) and z y ( x , y )express z ( x , y The gradient of z is the elevation of pixel (x, y), and z(x, y) is the elevation of pixel (x, y).
8. The optical navigation method for asteroid point-to-point landing based on three-dimensional navigation features according to claim 1, characterized in that, The low-resolution terrain of S1 satisfies the following conditions for the spatial distribution of terrain features: at least three features of the terrain are visible in the field of view of the navigation camera, and they are on any trajectory defined by the 3σ trajectory dispersion, where σ is determined by the uncertainty brought about by each maneuver in the tag sequence, plus the prior uncertainty of the spacecraft position.
9. The optical navigation method for asteroid point-to-point landing based on three-dimensional navigation features according to claim 1, characterized in that, The low-resolution terrain of S1 meets the illumination conditions of having an emission angle of less than 45°.