A method for optimizing three-dimensional reconstruction parameters of space targets based on on-orbit characteristics

Abstract

The application discloses a space target three-dimensional reconstruction parameter optimization method based on on-orbit characteristics, can solve the problems of misreconstruction and missing reconstruction caused by the factors such as no information exchange, no pre-design identification and uncontrollable attitude of space extraterrestrial objects in the three-dimensional reconstruction process, comprehensively considers the on-orbit geometry, spectrum, attitude and orbit characteristics of the target, improves the reconstruction authenticity and accuracy of the extraterrestrial object three-dimensional model, and thus assists in formulating subsequent detection, fly-around, close-in and landing navigation strategies.

Description

Technical Field

[0001] This invention relates to a method for optimizing the parameters of three-dimensional reconstruction of space targets based on on-orbit characteristics, belonging to the field of three-dimensional reconstruction technology. Background Technology

[0002] With the global surge in space resource development and the deepening of human exploration of deep space, the detection, identification, and 3D reconstruction technologies of extraterrestrial objects are increasingly influencing the success or failure of exploration missions, thus becoming a research hotspot. Extraterrestrial objects mainly refer to natural objects in space, whose characteristics are comprehensively reflected by their geometry, spectrum, attitude, orbit, and other properties.

[0003] In practical missions, the traditional 3D reconstruction process for extraterrestrial objects is generally divided into two main parts: sparse reconstruction and dense reconstruction. Sparse reconstruction first performs geometry-based feature extraction and matching, removes mismatches, then performs feature tracking, and finally uses the Structure of Motion (SFM) method to obtain the parameters of each camera and the corresponding object-space coordinates (feature points). During this process, bundle adjustment is used to optimize parameters and minimize secondary projection errors. Dense reconstruction uses the Patch-based Multi-View Stereo (PMVS) method. PMVS uses the results of sparse reconstruction to transform sparse point clouds into dense point clouds through patch expansion. However, these methods are not well-suited to various on-orbit scenarios in practical engineering applications. The reconstructed 3D models often suffer from scene degradation, attitude errors, and feature mismatches. Furthermore, factors such as changes in external environmental factors, lack of active information exchange during measurement, absence of pre-designed markers, and uncontrollable attitude lead to problems of misreconstruction and missed reconstruction. Summary of the Invention

[0004] The technical problem solved by this invention is to overcome the shortcomings of the prior art and propose a parameter optimization method for three-dimensional reconstruction of space targets based on on-orbit characteristics. By acquiring the geometric, spectral, attitude, and orbital characteristics of space targets, a highly accurate and realistic three-dimensional reconstruction of extraterrestrial objects can be achieved.

[0005] The technical solution of this invention is:

[0006] A method for optimizing the parameters of three-dimensional reconstruction of space targets based on on-orbit characteristics includes:

[0007] Based on the image of the space target, the layout of the space target's extended components is extracted, and the outer contour of the target is identified. The outer contour data is used as a subset of extrinsic parameters for the three-dimensional reconstruction geometric characteristics of the space target.

[0008] A mathematical model of the spectral characteristics of a space target is established, and inversion calculations are performed based on the mathematical model to obtain the spectral irradiance of the space target at the entrance pupil of the detector. The target spectrum is then fitted to form the extrinsic parameter subset of the spectral characteristics of the three-dimensional reconstruction of the space target.

[0009] Based on the spectral irradiance of the space target at the entrance pupil of the detector, the relative positional relationship between the target and the background radiation source is determined, and the space target is then classified and identified using orbital element information. The attitude information of the space target is determined by combining the irradiance at the entrance pupil of the detector when the space target is illuminated by the background light source and when it is visible to the detector. The orbital element information and the attitude information of the space target are used as the extrinsic parameter subset of the attitude and orbital characteristics of the space target in three-dimensional reconstruction.

[0010] The three subsets of extrinsic parameters are combined to form a set of extrinsic parameter constraints for 3D reconstruction, thereby generating a high-precision 3D model of the spatial target.

[0011] Preferably, the method for identifying the outer contour of the target is as follows: if the spatial target is a point target, differential detection and edge gradient extraction are used; if the spatial target is a surface target, the nearest neighbor algorithm is used for cluster analysis based on the geometric feature parameters of various target shapes in the collected and summarized prior knowledge base.

[0012] Preferably, the area target and background images segmented using Markov random field theory are used to perform target clustering analysis using grayscale histogram curves, including:

[0013] The gray-level histogram curve f1(i) of the target is compared with the gray-level histogram curve f2(i) of each target in the prior knowledge base. The correlation coefficient R between the two histogram curves is calculated one by one for identification.

[0014]

[0015] Where m represents the length of the shorter histogram sequence between f1(i) and f2(i);

[0016] Define the shape feature S of the target:

[0017]

[0018] Where P is the perimeter of the target in the image, and A is the area of ​​the target region;

[0019] By combining shape features and correlation coefficients, the spatial target to be reconstructed in three dimensions is identified, and the outer contour information of the target is obtained.

[0020] Preferably, a mathematical model of the spectral characteristics of the space target is established, and inversion calculations based on the mathematical model are performed to obtain the spectral irradiance of the space target at the entrance pupil of the detector, including:

[0021] By establishing a mathematical model of the spectral characteristics of a spatial target using the bidirectional reflectance distribution function of the target surface's spatial and spectral reflectance properties, the entrance pupil energy is calculated.

[0022]

[0023] In the formula, θ i , Let θ be the incident zenith angle and azimuth angle; r , λ represents the reflected zenith angle and azimuth angle; λ is the wavelength; dE i and dL r For incident irradiance and reflected irradiance;

[0024] Based on the different surface composition properties of the space target, the target surface is decomposed into regions and meshed. The spectral irradiance of each surface element at the detector entrance pupil is calculated according to the mathematical model of the spectral characteristics of the space target. Then, all surface element components are superimposed to obtain the spectral irradiance of the entire target at the detector entrance pupil.

[0025]

[0026] Among them, E sun (λ) represents the solar spectral irradiance at the surface element, and k is the correlation coefficient.

[0027] Preferably, the method of curve fitting is used to determine the BRDF database. And k, specifically:

[0028] The similarity between the target fitted spectrum X and the original spectrum Y in the BRDF database is evaluated using the spectral angle cosine.

[0029]

[0030] The closer the cosine value of the spectral angle is to 1, the higher the similarity.

[0031] Extract the value closest to 1 among all obtained spectral angle cosine values, and determine based on this value. and k.

[0032] Preferably, the relative positional relationship between the target and the background radiation source is determined. After initial orbit calculation, orbit matching, orbit improvement and improvement result analysis, the space target is classified and identified by orbital element information, which includes the positional parameters of the orbit, the shape parameters of the orbit, and the positional parameters of the satellite on the orbit.

[0033] Preferably, the extrinsic constraint set is subjected to feature point extraction, feature point matching, attitude estimation, beam adjustment optimization, and stereo matching to generate a dense point cloud. The generated dense point cloud is then subjected to point cloud mesh construction, mesh optimization, and texture mapping to generate a high-precision three-dimensional model of the spatial target.

[0034] Preferably, the accuracy of the high-precision spatial target 3D model is analyzed by terrain projection registration, and the root mean square error (RMS) of image grayscale value registration is statistically analyzed. If the RMS does not exceed a set threshold, the high-precision spatial target 3D model is considered to meet the accuracy requirements; otherwise, the 3D reconstruction extrinsic parameter constraint set data is changed, and the high-precision spatial target 3D model is reconstructed based on the changed 3D reconstruction extrinsic parameter set data. The accuracy analysis is performed again, and the process is iteratively updated until the RMS does not exceed the set threshold.

[0035] The advantages of this invention compared to the prior art are:

[0036] (1) The present invention can automatically extract spatial target information in a purposeful manner, distinguish different parts, and perform modular reconstruction. It has the characteristics of fast recognition speed and strong noise resistance.

[0037] (2) The present invention can flexibly input an optimized set of external parameters during the three-dimensional reconstruction process based on the geometric, spectral, attitude and orbital characteristics of the target space at the observation time, thereby improving the accuracy of three-dimensional reconstruction. Attached Figure Description

[0038] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0039] Figure 1 This is a flowchart of the method for optimizing the parameters of three-dimensional reconstruction of space targets based on on-orbit characteristics according to an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram of the space target orbit according to an embodiment of the present invention. Detailed Implementation

[0041] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0042] To address the problems of incorrect and missed reconstructions during 3D reconstruction caused by factors such as lack of information exchange, absence of pre-designed markers, and uncontrollable attitude of extraterrestrial objects, this invention proposes a parameter optimization method for 3D reconstruction of space targets based on on-orbit characteristics. When the target is a stably orbiting celestial body, this method comprehensively considers factors such as the target's shape, the peak value, slope, characteristic bands and morphology of its spectral characteristic curves, as well as its coordinates and orbital parameters in a specific space coordinate system. This comprehensive optimization yields a set of external parameters affecting 3D reconstruction, meeting the closed-loop analysis requirements for constructing high-precision 3D models of extraterrestrial objects. The main implementation methods are as follows: Figure 1 As shown, it includes:

[0043] (1) Geometric characteristic identification and classification to form a subset of extrinsic parameters for three-dimensional reconstruction based on geometric characteristics. The layout design of spacecraft extended components reflects the geometric appearance of the spacecraft. Extended components generally include orbit-changing engines, antennas, solar cell wings, attitude sensors, attitude control thrusters, coolers, thermal radiators or thermal radiator covers, etc. For example, the layout of solar cell wings is determined according to the on-orbit operating attitude and the performance of the payload. Generally, a symmetrical layout of two wings is adopted, which can reduce environmental interference forces and torques. However, some Earth observation satellites require the radiation cooler of the payload to be arranged on the shaded side, so a single solar cell wing facing the sun is adopted. Sometimes, in order to further reduce the influence of solar radiation pressure on attitude, a solar sail is also configured on the other side. The purpose of the shape, size and layout design of extended components is to ensure the realization of the satellite and its own functions. Usually, during the on-orbit operation phase, the satellite is mainly in the deployed state. Therefore, the inherent layout of extended components can be used to help identify space targets. Considering the shape, structure and size of the target, differential detection and edge gradient extraction methods are used for point targets; for face targets, the nearest neighbor algorithm is used for clustering. For example, in identifying small celestial bodies, we can first collect and summarize common target shapes as geometric feature parameters, then determine the categories after clustering based on these parameters, and further integrate shape factors and correlation coefficients to identify the basic shape of the small celestial body. The identification result is the subset of extrinsic parameter constraints for the geometric characteristics of the three-dimensional reconstruction.

[0044] After segmenting the target and background images using Markov random field theory, cluster analysis of the targets is performed using their gray-level histogram curves, and then category identification is performed using geometric features. Therefore, for targets of different shapes and sizes, identification can be performed by comparing the target's histogram curve f1(i) with the histogram curve f2(i) in the prior knowledge base, and calculating the correlation coefficient R between the two histogram curves.

[0045]

[0046] Where m represents the length of the shorter histogram sequence between f1(i) and f2(i).

[0047] Define the shape feature S (i.e., shape factor) of the target.

[0048]

[0049] Where P is the perimeter of the target in the image, which can be represented by the sum of the pixels at the outermost edge of the target area, and A is the area of ​​the target area, which is represented by the sum of the pixels.

[0050] Based on the geometric features of the target to be identified, and using an expert database, clustered targets can be identified. For example, to identify satellites, geometric feature parameters of common satellite body shapes and extended component layouts can be collected first, and then the clustered categories can be determined accordingly. By further integrating shape factors and correlation coefficients, the target can be identified. This yields the target's outer contour information, which serves as the geometric constraint for 3D reconstruction.

[0051] (2) Spectral characteristic identification and classification to form a subset of extrinsic parameters for 3D reconstruction based on geometric characteristics. A mathematical model of the target's spectral characteristics is established, and target feature extraction and identification are performed based on the inversion calculation of the mathematical model of spectral characteristics. Based on the concept of finite element method, the target surface is decomposed into regions and meshed according to the different compositional properties of the target surface.

[0052] The spectral irradiance of each surface element at the detector entrance pupil is calculated using the bidirectional reflectance distribution function, and these values ​​are superimposed to obtain the total spectral irradiance of the entire target at the detector entrance pupil. Then, considering the geometrical positional relationship between the target surface elements and the detector entrance pupil surface, the spectral irradiance of the entire target at the detector entrance pupil is integrated to obtain the total spectral irradiance of the target at the detector entrance pupil. The similarity in geometric spectrum between the measured spectral vector and the target vector in the potential spectral library is reflected by the spectral angle cosine; the closer the similarity is to 1, the higher the similarity. The identification result is the subset of extrinsic parameters for the 3D reconstructed spectral characteristics.

[0053] Since the spectral characteristics of a target can be described by the spectral irradiance generated at the detector entrance pupil by the background radiation reflected from the target surface, a two-way reflectance distribution function (BRDF) is introduced to calculate the entrance pupil energy, which combines the spatial and spectral reflectance characteristics of the target surface. The BRDF is defined as the ratio of the reflected irradiance to the incident irradiance.

[0054]

[0055] In the formula, θ i , Let θ be the incident zenith angle and azimuth angle; r , λ represents the reflected zenith angle and azimuth angle; λ is the wavelength; dE i and dL r These represent the incident irradiance and the reflected irradiance.

[0056] Based on the finite element method, the target surface is decomposed and meshed according to different material properties. BRDF is used to calculate the spectral irradiance (dA) of each surface element at the detector entrance pupil. Finally, all surface element components are superimposed to obtain the total spectral irradiance of the entire target at the detector entrance pupil. Then, considering the geometrical relationship between the target surface elements and the detector entrance pupil surface, integration is performed to obtain the total spectral irradiance of the entire target at the detector entrance pupil.

[0057]

[0058] Among them, E sun (λ) represents the solar spectral irradiance at surface element dA, and k is the correlation coefficient. The equivalent spectral reflectance is decomposed into a sum of polynomials, i.e.

[0059]

[0060] Next, based on the BRDF database, a curve fitting method is used to determine... and k i Since the types of target materials are limited, this method is feasible. Finally, the similarity between the target fitted spectrum and the original spectrum is evaluated using the spectral angle cosine. The degree of matching is characterized by calculating the angle cosine between the target fitted spectrum and the original spectrum. The angle cosine between the two spectral vectors X and Y is:

[0061]

[0062] The cosine of the spectral angle reflects the geometric similarity between two spectral vectors; the closer it is to 1, the higher the similarity. Spectral characteristics reflect the mechanism of image anomalies when encountering stray light interference during reconstruction and can serve as a constraint on the 3D reconstruction process based on illumination conditions.

[0063] (3) Attitude and orbital characteristics are identified and classified to form a subset of 3D reconstructed extrinsic parameters based on attitude and orbital characteristics. By establishing a relevant coordinate system and judging occlusion based on illumination and observation conditions, the irradiance generated by the target on the detector entrance pupil surface is obtained. Then, coordinate transformation is performed based on the target orbital parameters of the background radiation source to determine their relative positions. In addition, through initial orbit calculation, orbit matching, orbit improvement, and analysis of improvement results, information such as orbital elements is used to classify and identify space targets. The position of the satellite in its space orbit, such as... Figure 2As shown, the following parameters can be used to describe the orbit: (1) Position parameters of the orbit: right ascension of the ascending node Ω, orbital inclination i; (2) Shape parameters of the orbit: semi-major axis a, orbital eccentricity e, perigee angle ω; (3) Position of the satellite in the orbit: perigee time τ. These are collectively referred to as the six fundamental parameters describing the satellite orbit, which determine the law of target movement and can serve as the basis for target identification. The relevant identification results are the subset of extrinsic parameters of the attitude and orbit characteristics for three-dimensional reconstruction.

[0064] To determine the incident and observation direction vectors of light rays on the target surface, a relevant coordinate system is first established. Based on illumination and observation conditions, occlusion is assessed, resulting in a mathematical model describing the visible light characteristics as expressed by the irradiance generated by the target on the detector's entrance pupil. Then, a coordinate transformation is performed based on the target's orbital parameters from the background radiation source to determine their relative positions. This allows for the further calculation of the impact of target attitude changes on the irradiance of reflected background radiation at the detector's entrance pupil. With attitude changes, when the target is illuminated by the background light source and is visible to the detector, the irradiance at the detector's entrance pupil will exhibit different peak and zero values, indicating that the target's visible light characteristics possess strong directionality. This is consistent with the strong specular reflection characteristics of the target's surface coating material and solar panels.

[0065] The orbital parameters of a space target are the main data characterizing its motion patterns. Therefore, combining orbital characteristics allows for target classification and identification. By correlating observational data with known targets, initial orbit calculation, orbit matching, orbit improvement, and analysis of the improvement results are performed. The orbit improvement is used to verify the target matching results, improving orbit determination accuracy and achieving space target identification. Methods for space target identification generally consider the orbital patterns, distribution characteristics, and controllability of satellite orbits. By correlating and accumulating observational data, initial orbit calculation and orbit matching are performed, allowing for target identification using information such as orbital elements. This method offers good real-time performance and can also detect whether satellites deviate from their orbits due to perturbations.

[0066] (4) Based on the above classification of geometric, spectral, attitude and orbital characteristics, a set of external constraint parameters for three-dimensional reconstruction including geometric characteristics, spectral characteristics, attitude characteristics and orbital characteristics is formed, which can be used as constraints in the feature extraction and matching process of three-dimensional reconstruction.

[0067] (5) Formal three-dimensional reconstruction of space targets based on on-orbit characteristics is carried out. After feature extraction and feature matching based on the external constraint parameter set, attitude estimation, beam adjustment optimization and stereo matching are performed to generate dense point clouds. Finally, the generated dense point clouds are processed through point cloud mesh construction, mesh optimization and final texture mapping to generate a high-precision three-dimensional model of extraterrestrial objects. Based on further closed-loop verification of the conformity of the external constraint parameter set of three-dimensional reconstruction, such as by terrain projection registration, statistical image gray value registration average root mean square error (RMS), optimization iteration is performed until convergence to meet the accuracy index of model output, such as RMS≤0.2. Otherwise, the configuration is optimized and the geometric, spectral, attitude and orbital characteristic combination information is reselected by searching the characteristic parameter set, and the process (4) to (5) is repeated.

[0068] Complete the integration of the 3D model with the navigation control and data transmission drive system, and perform subsequent observation strategy analysis and other tasks.

[0069] This invention integrates the on-orbit geometry, spectrum, attitude, and orbital characteristics of the target to improve the realism and accuracy of the reconstruction of the three-dimensional model of the extraterrestrial body, thereby assisting in the formulation of subsequent exploration, fly-around, approach, and landing navigation strategies.

[0070] The embodiments described above are merely preferred embodiments of the present invention. Ordinary variations and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing parameters of three-dimensional reconstruction of space targets based on on-orbit characteristics, characterized in that, include: Based on the image of the space target, the layout of the space target's extended components is extracted, and the outer contour of the target is identified. The outer contour data is used as a subset of extrinsic parameters for the three-dimensional reconstruction geometric characteristics of the space target. A mathematical model of the spectral characteristics of a space target is established. Based on the mathematical model, inversion calculations are performed to obtain the spectral irradiance of the space target at the entrance pupil of the detector. The target spectrum is then fitted to form the extrinsic parameter subset of the spectral characteristics of the three-dimensional reconstruction of the space target. Based on the spectral irradiance of the space target at the entrance pupil of the detector, the relative positional relationship between the target and the background radiation source is determined, and the space target is then classified and identified using orbital element information. The attitude information of the space target is determined by combining the irradiance at the entrance pupil of the detector when the space target is illuminated by the background light source and when it is visible to the detector. The orbital element information and the attitude information of the space target are used as the extrinsic parameter subset of the attitude and orbital characteristics of the space target in three-dimensional reconstruction. The three subsets of extrinsic parameters are combined to form a three-dimensional reconstruction extrinsic parameter constraint set. Feature point extraction, feature point matching, attitude estimation, bundle adjustment optimization, and stereo matching are performed on the extrinsic parameter constraint set to generate a dense point cloud. The generated dense point cloud is then processed through point cloud mesh construction, mesh optimization, and texture mapping to generate a high-precision three-dimensional model of the spatial target.

2. The method for optimizing the parameters of three-dimensional reconstruction of spatial targets according to claim 1, characterized in that, The method for identifying the outer contour of a target is as follows: if the spatial target is a point target, differential detection and edge gradient extraction are used; if the spatial target is a surface target, the nearest neighbor algorithm is used for cluster analysis based on the geometric feature parameters of various target shapes in the collected and summarized prior knowledge base.

3. The method for optimizing the parameters of three-dimensional reconstruction of spatial targets according to claim 2, characterized in that, After segmenting the surface target and background images using Markov random field theory, cluster analysis of the targets is performed using gray-level histogram curves, including: Compare the grayscale histogram curve of the target and the grayscale histogram curves of each target in the prior knowledge base Calculate the correlation coefficient of the two histogram curves one-to-one. To perform identification: in, express and The length of shorter histogram sequences; Define the shape feature S of the target: Where P is the perimeter of the target in the image, and A is the area of ​​the target region; By combining shape features and correlation coefficients, the spatial target to be reconstructed in three dimensions is identified, and the outer contour information of the target is obtained.

4. The method for optimizing the parameters of three-dimensional reconstruction of spatial targets according to claim 1, characterized in that, A mathematical model of the spectral characteristics of the space target is established. Based on the mathematical model, inversion calculations are performed to obtain the spectral irradiance of the space target at the entrance pupil of the detector, including: By establishing a mathematical model of the spectral characteristics of a spatial target using the bidirectional reflectance distribution function of the target surface's spatial and spectral reflectance properties, the entrance pupil energy is calculated. In the formula, , These are the incident zenith angle and azimuth angle; , To reflect the zenith angle and azimuth angle; Wavelength; and For incident irradiance and reflected irradiance; Based on the different surface composition properties of the space target, the target surface is decomposed into regions and meshed. The spectral irradiance of each surface element at the detector entrance pupil is calculated according to the mathematical model of the spectral characteristics of the space target. Then, all surface element components are superimposed to obtain the spectral irradiance of the entire target at the detector entrance pupil. in, The solar spectral irradiance at the surface element is... The correlation coefficient.

5. The method for optimizing the parameters of three-dimensional reconstruction of spatial targets according to claim 4, characterized in that, Based on the BRDF database, curve fitting is used to determine... and Specifically: The similarity between the target fitted spectrum X and the original spectrum Y in the BRDF database is evaluated using the spectral angle cosine. The closer the cosine value of the spectral angle is to 1, the higher the similarity. Extract the value closest to 1 among all obtained spectral angle cosine values, and determine based on this value. and .

6. The method for optimizing the parameters of three-dimensional reconstruction of spatial targets according to claim 1, characterized in that, The relative positions of the target and the background radiation source are determined. After initial orbit calculation, orbit matching, orbit improvement and improvement result analysis, the space target is classified and identified by orbital element information, which includes the position parameters of the orbit, the shape parameters of the orbit, and the position parameters of the satellite in the orbit.

7. The method for optimizing the parameters of three-dimensional reconstruction of spatial targets according to claim 1, characterized in that, Accuracy analysis is performed on the high-precision 3D model of the spatial target by terrain projection registration. The root mean square error (RMS) of the image grayscale registration is statistically analyzed. If the RMS does not exceed the set threshold, the high-precision 3D model of the spatial target is considered to meet the accuracy requirements. Otherwise, the 3D reconstruction extrinsic parameter constraint set data is changed, and the high-precision 3D model of the spatial target is reconstructed based on the changed 3D reconstruction extrinsic parameter constraint set data. Accuracy analysis is performed again. Through iterative updates, the RMS does not exceed the set threshold.

Images