Deep space probe optical navigation pose decoupling estimation method

Abstract

The deep space probe optical navigation position and posture decoupling estimation method disclosed by the application belongs to the field of deep space exploration. The method of the application is as follows: a navigation camera is used to image a target celestial body surface, and a pixel, a line coordinate and an image point observation error covariance matrix of a navigation landmark are extracted; according to the covariance propagation law of a nonlinear function, the image point observation error covariance is propagated to a line of sight (LOS) angle, and the observation error variance of the LOS angle is calculated; a weight is used to represent the relative accuracy between different observation angles, and an observation error weight matrix W for optical navigation position and posture decoupling estimation is constructed A ; by performing least squares weighting on the angle observation equation to incorporate the angle observation uncertainty, the weight of the navigation landmark with different observation qualities is adjusted, the angle observation error uncertainty is weighted and optimized in the absolute position estimation process, and the probe position is obtained; and the probe position and the LOS observation quantity obtained by separate decoupling estimation are used to decouple and estimate the probe posture through a multi-vector posture determination principle.

Description

Technical Field

[0001] The present invention relates to a method for determining the position and posture of a deep space probe, and in particular to a method for estimating the position and posture of a deep space probe by observing navigation landmarks with a camera, and belongs to the field of deep space exploration. Background Art

[0002] Flying close to a target celestial body will be one of the most complex tasks in future deep space exploration. However, due to communication delays, the complexity of the deep space dynamics, and the remote location of the exploration mission, traditional ground-based navigation and control methods suffer from significant communication delays, are technically difficult to operate, and are costly, making them incapable of meeting the requirements for high-precision exploration. This necessitates autonomous navigation capabilities for the probe. Although inertial navigation does not require ground-based measurement and control support, the inertial measurement unit (IMU) suffers from constant bias and drift, resulting in low navigation accuracy. With advances in computer hardware technology and optical sensors, pose determination methods based on optical information have become a research hotspot. Craters, as natural topographical features on the surface of target celestial bodies, offer high orbit determination performance through autonomous navigation methods that utilize these features as navigation landmarks. Improving pose estimation accuracy through the use of optical information from navigation landmarks has become a key research focus in the aerospace industry worldwide.

[0003] However, optical cameras are subject to sensor errors, environmental disturbances, target motion, and other factors during imaging observations. Consequently, observation errors are unavoidable when extracting the center of navigation landmarks in the imaging plane. These errors are anisotropic and non-independently distributed, severely impacting the accuracy of the detector's pose determination. Furthermore, the strong coupling and nonlinear solution of the pose in the image point observation equations complicate the pose estimation process, significantly impacting the computational efficiency of the onboard computer.

[0004] Among the developed pose estimation methods, the prior art [1] (Zhu Shengying, Xiu Yi, Cui Pingyuan, et al. A weighted determination method for the optical navigation pose of deep space probes: China, ZL201810766656.2[P], 2021-06-15.) proposed a method for obtaining the uncertainty of the navigation landmark center error by using the crater fitting ellipse. The weighted matrix was constructed by performing singular value decomposition on the measurement error covariance matrix, and the weighted matrix was integrated into the image point observation equation to realize the weighted estimation of the absolute pose of the detector, thereby improving the pose estimation accuracy. However, this method uses the perspective projection observation equation. In the process of solving the pose, the observation equation has a strong nonlinearity and the pose variables have a coupling relationship, resulting in high computational complexity.

[0005] Prior art [2] (Zhu S, Liu D, Liu Y, et al. Observability-based visual navigation using landmarks measuring angle for pinpoint landing [J], ActaAstronaut. 155 (2019) 313–324.) addresses the strong coupling and nonlinearity issues in solving pose estimation using the image point observation equation. Considering the angular invariance under Euclidean transformation, the line-of-sight angle formed by the navigation landmark and the detector is used as the observation quantity, replacing the pixel line observation in the traditional image point observation equation. A new angle observation equation is constructed, which realizes the decoupled estimation of position and attitude variables, reduces the estimation dimension, and improves the pose estimation accuracy. However, this method does not consider the impact of navigation landmark observation error on pose estimation, and assumes that the observation error is an independent and identically distributed error model.

[0006] Among the developed landmark uncertainty evaluation methods, the prior art [3] (Zhou Run, Zhang Zhengyu, Huang Xuhui. Weighted orthogonal iterative algorithm for camera pose estimation [J]. Acta Optica Sinica, 2018, 38(05):193-199.) proposed a weighted orthogonal iterative algorithm for camera pose estimation. This algorithm uses the weighted collinearity error as the objective function and automatically adjusts the weight coefficient according to the image plane reprojection error to optimize the camera pose estimation result. However, this method adds weights based on the collinearity error equation and achieves nonlinear optimization by minimizing the reprojection error. It also has problems such as not considering observation errors, slow solution speed, and complex solution. Summary of the Invention

[0007] The main purpose of the present invention is to provide a method for decoupling and estimating the optical navigation attitude of a deep space probe. Based on the extraction of navigation landmark observation errors and the calculation of uncertainties in captured images, the angle observation equation is constructed using the navigation landmark sight angle as the observation quantity. The uncertainty of the observation errors of different image points is integrated into the angle observation error according to the error propagation law. Through hyper-toroid positioning and multi-vector attitude determination, the position and attitude are decoupled and estimated, and the decoupled weighted estimation of the absolute attitude of the probe under the angle observation error is realized, thereby further reducing the difficulty of attitude estimation and improving the accuracy of the probe's attitude estimation.

[0008] The absolute position and attitude refers to the position and attitude of the probe relative to the fixed coordinate system of the target celestial body.

[0009] The purpose of the present invention is achieved through the following technical solutions.

[0010] The present invention discloses a method for decoupling the optical navigation pose of a deep space probe. The probe uses a navigation camera to image the surface of a target celestial body, extracts the pixels of navigation landmarks, image line coordinates, and the image point observation error covariance matrix; propagates the observation error covariance of the image point to the line of sight angle according to the covariance propagation law of a nonlinear function, and calculates the observation error variance of the line of sight angle; uses weights to characterize the relative accuracy between different observation angles, and constructs the observation error weight matrix W for decoupling the optical navigation pose. A ; By performing least square weighted method on the angle observation equation to incorporate the angle observation uncertainty, the weights of navigation landmarks with different observation qualities are adjusted to achieve weighted and optimized angle observation error uncertainty in the absolute position estimation process. Based on the angle observation equation that incorporates the angle observation uncertainty, the detector position is separately decoupled and estimated to obtain the detector position; using the separately decoupled estimated detector position and line of sight observation, the detector attitude is decoupled and estimated through the multi-vector attitude determination principle.

[0011] The present invention discloses a method for decoupling and estimating the optical navigation posture of a deep space probe, comprising the following steps:

[0012] Step 1: The detector uses the navigation camera to image the surface of the target celestial body and extract the pixels, image line coordinates and image point observation error covariance matrix of the navigation landmark.

[0013] After the detector uses the navigation camera to image the surface of the target celestial body, the pixel and image line coordinates of navigation landmark i and navigation landmark j in the target celestial body surface image are extracted through the image detection algorithm. i ,l i ,] T ,[p j ,l j ] T The covariance matrix of the image observation error of the navigation landmark is calculated by the onboard computer and is:

[0014]

[0015]

[0016] Among them, R pp and R ll Represents the covariance matrix R pl The variance of p and l in ; by the symmetry of the covariance matrix, R pl =R lp , represents the covariance matrix R pl where p and l are the covariances of the two variables, and the subscript i or j is the landmark number.

[0017] Step 2: According to the covariance propagation law of nonlinear functions, propagate the observation error covariance of the image point to the line of sight angle, and calculate the observation error variance of the line of sight angle.

[0018] Given two navigation landmark sight vectors [p i ,l i ,f] T ,[p j ,l j ,f] T , then the sight angle A formed by the two sight vectors is ij for

[0019]

[0020] Where f is the focal length of the camera, (p i ,l i ) and (p j ,l j ) are independent observations, but there is a correlation between pixel p and image line l, and its error covariance matrix R pl Follow the error model established in step 1.

[0021] According to the covariance propagation law of nonlinear function, the sight angle A ij Variance for

[0022]

[0023] in,

[0024]

[0025] Step 3: Based on the observation error variance of the line of sight angle obtained in step 2, the relative accuracy between different observation angles is characterized by weights, and the observation error weight matrix W is constructed for optical navigation pose decoupling estimation. A .

[0026] Based on the observation error variance of the sight angle obtained in step 2, the weight is used to characterize the relative accuracy between different observation angles. For the sight angle A ij , the weight of the observation angle is For the case where n navigation landmarks are observed to form n(n–1) / 2 sight angles, the observation error weight matrix W for optical navigation pose decoupling estimation is constructed: A for

[0027]

[0028] Step 4: Incorporate the angle observation uncertainty into the angle observation equation by performing least squares weighting, adjust the weights of navigation landmarks with different observation qualities, and achieve weighted and optimized angle observation error uncertainty in the absolute position estimation process. Based on the angle observation equation incorporating the angle observation uncertainty, the detector position is separately decoupled and estimated to obtain the detector position.

[0029] Let n i ,n j are the unit sight line vectors of the i-th and j-th navigation landmarks respectively, and the landmark sight line angle observation equation shown in formula (7) is established:

[0030] δA ij =h ij δr (7)

[0031] Among them, δA ij is the deviation of the sight angle formed by the i-th and j-th landmarks.

[0032] Calculate the row vector h of the observation matrix for the i-th and j-th landmark combination ij for

[0033]

[0034] Among them, m ij and m ji is the auxiliary vector, and the calculation formula is

[0035]

[0036] Perform the least square weighting on Equation (7), incorporate the uncertainty of the angle observation error, and decouple the detector position separately to estimate

[0037] r=r * +(H T W A H) -1 H T W A δA (10)

[0038] Among them, r * is the orbit prediction position, the observation angle deviation δA and the linearized observation matrix H are

[0039]

[0040] Step 5: Using the detector position and line of sight observations estimated separately in step 4, the detector attitude C is estimated by decoupling the multi-vector attitude determination principle. bL .

[0041] Given the detector position r estimated in step 4 and the kth (k=1,2,…,n) landmark unit sight vector n k , when reading the position of the navigation landmark, according to the multi-vector attitude determination principle, the decoupling estimation of the detector's attitude is

[0042]

[0043] in

[0044]

[0045] Among them, r k is the distance from the kth landmark to the detector calculated using the read landmark position and the estimated detector position, r k is the corresponding sight vector, n k Calculate the unit sight vector of the kth landmark using pixel line coordinates.

[0046]

[0047] Beneficial effects:

[0048] 1. The present invention discloses a method for decoupling the pose estimation of deep space probes using optical navigation. This method addresses the problem of limited pose estimation accuracy due to image point observation error uncertainty. By propagating image point observation errors, an angle observation error weight matrix is ​​constructed, and the weights of navigation landmarks with different observation qualities are adjusted. This achieves weighted and optimized angular observation error uncertainty in the absolute pose estimation process, significantly improving the accuracy of absolute pose estimation.

[0049] 2. The present invention discloses a method for decoupling and estimating the optical navigation posture of a deep space probe. It aims to solve the problem of strong coupling of the posture of the image point observation equation and nonlinearity. It incorporates the uncertainty of the angle observation error into the angle observation equation by least square weighting. The position and attitude are decoupled and estimated separately based on the angle observation equation that incorporates the angle observation uncertainty, avoiding the complex linearization process of the traditional posture estimation algorithm, reducing the difficulty of posture estimation, and simplifying the complexity of the posture solution problem. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flow chart of the method for decoupling and estimating the optical navigation posture of a deep space probe according to the present invention;

[0051] Figure 2 Schematic diagram of the propagation of image point observation error to angle observation error in the present invention;

[0052] Figure 3 This is a schematic diagram of the angular position plane positioning principle of the present invention;

[0053] Figure 4 Schematic diagram of the multi-vector attitude determination principle of the present invention;

[0054] Figure 5 The four navigation landmark observation data and their error ellipses constructed in the simulation example of the present invention are:

[0055] Figure 6 is the decoupling weighted estimation result of the detector position error in the simulation of the present invention;

[0056] Figure 7 It is the decoupling weighted estimation result of the detector attitude error in the example simulation of the present invention. DETAILED DESCRIPTION

[0057] In order to better illustrate the purpose and advantages of the present invention, the invention is further described below with reference to the accompanying drawings and examples.

[0058] In order to verify the feasibility of the present invention, the camera parameters used in the example and the detector parameters under the target planet fixed connection system are first given, as shown in Table 1.

[0059] Table 1 Example simulation parameters

[0060]

[0061] Taking the detector observing four navigation landmarks for absolute pose estimation as an example, based on the image point projection coordinates of the navigation landmarks on the image plane, 1000 target shootings are performed by customizing the error ellipse parameters to generate the image point observation data for the Monte Carlo experiment. The image plane simulated by 1000 image points is as follows: Figure 5 The three-dimensional positions of the navigation landmarks in the target planet fixed coordinate system and the corresponding image point error ellipse parameters are shown in Table 2.

[0062] Table 2 Observation error parameters of navigation landmarks and their image points

[0063]

[0064] like Figure 1 As shown, the method for decoupling and estimating the optical navigation posture of a deep space probe disclosed in this embodiment is specifically implemented in the following steps:

[0065] Step 1: The detector uses the navigation camera to image the surface of the target celestial body and extract the pixels, image line coordinates and image point observation error covariance matrix of the navigation landmark.

[0066] After the probe uses the navigation camera to image the surface of the target celestial body, the image detection algorithm is used to extract the pixel and image line coordinates of the four navigation landmarks in the target celestial body surface image. The onboard computer calculates the image point observation error covariance matrix of the navigation landmarks, as shown in Table 2.

[0067] Step 2: According to the covariance propagation law of nonlinear functions, propagate the observation error covariance of the image point to the line of sight angle, and calculate the observation error variance of the line of sight angle.

[0068] Given any two of the four navigation landmarks, the sight vectors [p i ,l i ,f] T ,[p j ,l j ,f] T , then the sight angle A formed by the two sight vectors is ij for

[0069]

[0070] Where f is the focal length of the camera, (p i ,l i ) and (p j ,l j ) are independent observations, but there is a correlation between pixel p and image line l, and its error covariance matrix R pl Follow the error model established in step 1.

[0071] According to the covariance propagation law of nonlinear functions, the error propagation is shown as follows: Figure 2 As shown, the sight angle A ij Variance for

[0072]

[0073] in,

[0074]

[0075] Step 3: Based on the observation error variance of the line of sight angle obtained in step 2, the relative accuracy between different observation angles is characterized by weights, and the observation error weight matrix W is constructed for optical navigation pose decoupling estimation. A .

[0076] Based on the observation error variance of the sight angle obtained in step 2, in order to measure the accuracy between different observation angles, the weight is used to represent the relative accuracy between different observation angles. For a certain sight angle A ij , the weight of the observation angle is For the case of 6 sight angles formed by 4 navigation landmarks, the observation error weight matrix W is constructed for optical navigation pose decoupling estimation A for

[0077]

[0078] Step 4: Incorporate the angle observation uncertainty into the angle observation equation by performing least squares weighting, adjust the weights of navigation landmarks with different observation qualities, and achieve weighted and optimized angle observation error uncertainty in the absolute position estimation process. Based on the angle observation equation incorporating the angle observation uncertainty, the detector position is separately decoupled and estimated to obtain the detector position.

[0079] Let n i ,n j are the unit sight line vectors of the i-th and j-th navigation landmarks among the four observed navigation landmarks, respectively. The landmark sight line angle observation equation shown in formula (19) is established.

[0080] δA ij =h ij δr (19)

[0081] Among them, δA ij is the deviation of the sight angle formed by the i-th and j-th landmarks.

[0082] Calculate the row vector h of the observation matrix for the i-th and j-th landmark combination ij for

[0083]

[0084] Among them, m ij and m ji is the auxiliary vector, and the calculation formula is

[0085]

[0086] The least squares weighting is performed on Equation (19), the uncertainty of the angle observation is incorporated, and the detector position is estimated based on the angle position.

[0087] r=r * +(H T W A H) -1 H T W A δA (22)

[0088] Among them, the principle of determining the position of the angle position plane is as follows Figure 3 As shown, r * is the orbit prediction position, the observation angle deviation δA and the linearized observation matrix H are

[0089]

[0090] Step 5: Using the detector position and line of sight observations estimated separately in step 4, the detector attitude C is estimated by decoupling the multi-vector attitude determination principle. bL .

[0091] Given the detector position r estimated in step 4 and the kth (k=1,2,3) landmark unit sight vector n k , when reading the position of the navigation landmark, according to the multi-vector attitude determination principle, such as Figure 4 As shown, the decoupled estimated detector pose is

[0092]

[0093] in

[0094]

[0095] Among them, r k is the distance from the kth landmark to the detector calculated using the read landmark position and the estimated detector position, r k is the corresponding sight vector, n k Calculate the unit sight vector of the kth landmark using pixel line coordinates.

[0096]

[0097] From the generated observation data table 2, the observation error covariance of navigation landmark No. 3 is the largest. Correspondingly, the error ellipse of this landmark in the image plane is also the largest, as shown in Figure 5 As shown in the upper left corner, the uncertainty of the line of sight angle formed by landmark 3 and other landmarks will become larger. In the pose decoupling weighted estimation algorithm, the weight of the line of sight angle observation formed by this landmark should be reduced to ensure the accuracy of the final pose estimation. Based on the above analysis, 1000 Monte Carlo simulation experiments were conducted using the six observation angles formed by these four navigation landmarks. The position estimation error is shown in Figure 6 As shown, the attitude estimation error results are as follows Figure 7 The statistical results of the mean and standard deviation of the error are shown in Table 3.

[0098] Table 3. Statistical comparison of pose estimation errors from angle observations

[0099]

[0100] The pose estimation error results show that by incorporating the observation error uncertainty caused by the image points into the angle observation, the position accuracy estimated using angle error weighted least squares is significantly better than the direct decoupling estimation method that does not consider the observation error. Compared with the direct decoupling estimation method, the proposed method for optical navigation pose estimation for deep space probes improves position accuracy by 43.69% and attitude estimation accuracy by 15.83%, thus verifying the effectiveness of the proposed method for optical navigation pose estimation for deep space probes.

[0101] The above specific description further illustrates the purpose, technical solutions and beneficial effects of the invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for decoupling and estimating the optical navigation pose of a deep space probe, characterized by: The following steps are included: Step 1: The detector uses the navigation camera to image the surface of the target celestial body and extract the pixels, image line coordinates and image point observation error covariance matrix of the navigation landmarks; Step 2: According to the covariance propagation law of nonlinear functions, propagate the observation error covariance of the image point to the line of sight angle, and calculate the observation error variance of the line of sight angle; Step 3: Based on the observation error variance of the line of sight angle obtained in step 2, the relative accuracy between different observation angles is characterized by weights, and the observation error weight matrix W is constructed for optical navigation pose decoupling estimation. A ; Step 4: Incorporate the angle observation uncertainty into the angle observation equation through least squares weighting. Adjust the weights of navigation landmarks with different observation qualities to achieve weighted and optimized angle observation error uncertainty in the absolute position estimation process. Decouple the detector position based on the angle observation equation incorporating the angle observation uncertainty to obtain the detector position. Step 5: Using the detector position and line of sight observations estimated separately in step 4, the detector attitude C is estimated by decoupling the multi-vector attitude determination principle. bL .

2. The method for decoupling and estimating the optical navigation pose of a deep space probe according to claim 1, wherein: The implementation method of step one is: After the detector uses the navigation camera to image the surface of the target celestial body, the pixel and image line coordinates of navigation landmark i and navigation landmark j in the target celestial body surface image are extracted through the image detection algorithm. i ,l i ,] T ,[p j ,l j ] T The covariance matrix of the image observation error of the navigation landmark is calculated by the onboard computer and is: Among them, R pp and R ll Represents the covariance matrix R pl The variance of p and l in ; by the symmetry of the covariance matrix, R pl =R lp , represents the covariance matrix R pl where p and l are the covariances of the two variables, and the subscript i or j is the landmark number.

3. The method for decoupling and estimating the optical navigation posture of a deep space probe according to claim 2, wherein: The implementation method of step 2 is: Given two navigation landmark sight vectors [p i ,l i ,f] T ,[p j ,l j ,f] T , then the sight angle A formed by the two sight vectors is ij for Where f is the focal length of the camera, (p i ,l i ) and (p j ,l j ) are independent observations, but there is a correlation between pixel p and image line l, and its error covariance matrix R pl Follow the error model established in step 1; According to the covariance propagation law of nonlinear function, the sight angle A ij Variance for in, 4. The method for decoupling and estimating the optical navigation posture of a deep space probe according to claim 3, wherein: The implementation method of step three is: Based on the observation error variance of the sight angle obtained in step 2, the weight is used to characterize the relative accuracy between different observation angles. For the sight angle A ij , the weight of the observation angle is For the case where n navigation landmarks are observed to form n(n–1) / 2 sight angles, the observation error weight matrix W for optical navigation pose decoupling estimation is constructed: A for 5. The method for decoupling and estimating the optical navigation posture of a deep space probe according to claim 4, wherein: The implementation method of step 4 is: Let n i ,n j are the unit sight line vectors of the i-th and j-th navigation landmarks respectively, and the landmark sight line angle observation equation shown in formula (7) is established: δA ij =h ij δr (7) Among them, δA ij is the deviation of the sight angle formed by the i-th and j-th landmarks; Calculate the row vector h of the observation matrix for the i-th and j-th landmark combination ij for Among them, m ij and m ji is the auxiliary vector, and the calculation formula is Perform the least square weighting on Equation (7), incorporate the uncertainty of the angle observation error, and decouple the detector position separately to estimate r=r * +(H T W A H) -1 H T W A δA (10) where r * is the orbit prediction position, the observation angle deviation δA and the linearized observation matrix H are 6. The method for decoupling and estimating the optical navigation posture of a deep space probe according to claim 5, wherein: The implementation method of step five is: Given the detector position r estimated in step 4 and the kth (k=1,2,…,n) landmark unit sight vector n k , when reading the position of the navigation landmark, according to the multi-vector attitude determination principle, the decoupling estimation of the detector's attitude is in Among them, r k is the distance from the kth landmark to the detector calculated using the read landmark position and the estimated detector position, r k is the corresponding sight vector, n k To calculate the unit sight vector of the kth landmark using the pixel line coordinates;

Images