A sequential feature space construction method for autonomous optical navigation

Abstract

The application relates to a sequence feature space-time construction method of autonomous optical navigation, and belongs to the technical field of spacecraft navigation, guidance and control. System state equations of visual aided inertial navigation in an unknown environment are established. Image features corresponding to unknown land marks which have the largest contribution to navigation accuracy in a field of view are selected in space. Land marks are observed, and observation information is taken as input of a Kalman filter to obtain optimal estimation of each state variable. The optimal estimation of each state variable is taken as a navigation parameter to perform landing navigation of a lander. A next observation time is calculated. Land marks are re-observed to obtain optimal estimation of each state variable. The optimal estimation of each state variable is taken as a navigation parameter at the time to perform landing navigation of the lander. The application obtains an optimal observation interval between two observations by solving an observation interval time which makes a proposed depth error model minimum, can effectively reduce unnecessary observation, and greatly reduces a calculation burden caused by image processing.

Description

Technical Field

[0001] This invention belongs to the field of spacecraft navigation, guidance and control technology, and relates to a method for constructing a sequence feature spatiotemporal structure for autonomous optical navigation. Background Technology

[0002] To achieve precise landings in areas of scientific value and avoid obstacles such as hills and rocks, planetary landers require high-precision autonomous navigation. The lander uses its onboard inertial measurement unit (IMU) to integrate and calculate navigation status information such as position, velocity, and attitude. However, due to factors such as initial navigation errors, gravitational field model errors, and random measurement errors, the acceleration and angular velocity measured by the IMU exhibit deviations. As a result, the navigation status information obtained through integration gradually becomes biased over time. Therefore, inertial navigation often needs to be combined with other external measurements to correct the deviations in the inertial navigation integration, thereby improving navigation accuracy.

[0003] Currently, some progress has been made in navigation technology based on image sequences, but the following shortcomings still exist:

[0004] (1) Due to the limited computing and storage capacity on the lander, the lander can only track a small number of landmarks to reduce the matrix dimension of the navigation filter and take a small number of images to reduce the computational load of image processing, or directly transmit the images back to the ground for processing.

[0005] (2) Existing landmark observation planning methods mainly focus on selecting landmarks that have the strongest effect on improving navigation accuracy from a spatial perspective. The observability used in the planning is usually a function of the observability matrix, and the computational load is often large when optimizing it.

[0006] (3) Existing land landmark observation planning methods lack research on reducing the number of observations. If the number of observations can be reduced from a time perspective, the computational load of onboard image processing can be reduced, thereby improving the autonomous navigation capability. Summary of the Invention

[0007] The technical problem solved by this invention is to overcome the shortcomings of the prior art and propose a spatiotemporal construction method for sequence features of autonomous optical navigation. By solving for the observation interval time that minimizes the proposed depth error model, the optimal observation interval between two observations is obtained, which can effectively reduce unnecessary observations and significantly reduce the computational burden caused by image processing.

[0008] The solution of the present invention is:

[0009] A method for constructing the spatiotemporal sequence features of autonomous optical navigation includes:

[0010] Step 1: Establish the system state equations for vision-assisted inertial navigation in an unknown environment. The variables in the system state equations include the lander's position r in the landing coordinate system {L}, the lander's velocity v in the landing coordinate system {L}, the lander's acceleration a in the landing coordinate system {L}, the attitude quaternion q from the landing coordinate system {L} to the body coordinate system {B}, and the accelerometer drift deviation b. a Angular velocity meter drift deviation b ω Landmark p i The coordinates r in the landing coordinate system {L} pi ;

[0011] Step 2: Establish a spatial configuration model of landmarks; based on the spatial configuration model of landmarks, select the image features corresponding to the unknown landmarks in the field of view that contribute the most to navigation accuracy.

[0012] Step 3: Establish a Kalman filter based on the state equation of the vision-assisted inertial navigation system; observe landmarks and use the observation information as input to the Kalman filter to obtain the optimal estimates of each state variable in the state equation of the vision-assisted inertial navigation system; use the optimal estimates of each state variable as navigation parameters to perform landing navigation for the lander.

[0013] Step 4: Using the optimal estimates of each state variable in the state equation of the vision-assisted inertial navigation system as input to the depth error model, calculate the interval of the observed image sequence. And the next observation time; combined with the image features corresponding to the unknown landmarks that contribute the most to navigation accuracy selected spatially in step two, and the best observation time determined temporally in step four, the spatiotemporal construction of sequence features is obtained;

[0014] Step 5: At the next observation time, return to Step 3, re-observe the landmark, and use the observation information as input to the Kalman filter to obtain the optimal estimates of each state variable in the state equation of the vision-assisted inertial navigation system; use the optimal estimates of each state variable as the navigation parameters at that time for the landing navigation of the lander.

[0015] Step 6: Repeat steps 3 to 5 until the lander has landed successfully; if the land marker needs to be replaced, return to step 2; otherwise, return to step 4.

[0016] In the above-mentioned method for constructing the spatiotemporal sequence features of autonomous optical navigation, in step one, the system state equation for vision-assisted inertial navigation in an unknown environment is:

[0017]

[0018] In the formula, r, v, and a are the position, velocity, and acceleration of the lander in the landing coordinate system {L}, respectively;

[0019] q is the attitude quaternion from the landing coordinate system {L} to the body coordinate system {B};

[0020] Let {B} be the representation of the three-axis rotational angular velocity of the body coordinate system {B} relative to the landing coordinate system {L} in the body coordinate system {B}.

[0021] Ω(ω) is the coefficient matrix of the attitude motion equation;

[0022] b a and b ω These are the drift deviations of the accelerometer and the angular velocity meter, respectively.

[0023] n wa and n wω Driver b a and b ω Zero-mean white noise;

[0024] r pi For land standard p i The coordinates in the landing coordinate system {L}; set one known landmark p1 and any one unknown landmark p2; where p1 is at the origin of the landing coordinate system {L}.

[0025] In the above-mentioned method for constructing the spatiotemporal features of autonomous optical navigation, in step two, the landmark spatial configuration model is as follows:

[0026]

[0027] In the formula, δr is the estimation error bound of the lander's position in the landing coordinate system {L};

[0028] r represents the position of the lander in the landing coordinate system {L};

[0029] α represents the position r of the lander in the landing coordinate system {L} and the coordinates r of p2 in the landing coordinate system {L}. p2 The angle between them;

[0030] β is the angle between r1 and r2; r i Let i be the position of the i-th landmark relative to the lander;

[0031] δr p Let r be the coordinates of p2 in the landing coordinate system {L} p2 The estimation error bound;

[0032] ε θ Let be the attitude estimation error bound.

[0033] In the aforementioned method for constructing a sequence feature spatiotemporal framework for autonomous optical navigation, the method for spatially selecting the unknown landmarks in the field of view that contribute the most to navigation accuracy is as follows:

[0034] Find the optimal value of α in the landmark spatial configuration model. * The optimal value of β * ;

[0035]

[0036] When the triangle formed by the position of the unknown landmark p2, the lander position, and the position of p1 satisfies α * and β * When considering the relationship, the position of the unknown landmark p2 contributes the most to the observability.

[0037] In the aforementioned method for constructing the spatiotemporal sequence features of autonomous optical navigation, the solved α * and β * The optimal position of p2 is relative to the direction of the lander. By extracting feature points from the image along this direction, the image features corresponding to p2 can be obtained.

[0038] In the aforementioned method for constructing a sequence feature spatiotemporal model for autonomous optical navigation, in step three, the state variables in the state equation of the vision-assisted inertial navigation system include the lander's position r in the landing coordinate system {L}, the lander's velocity v in the landing coordinate system {L}, the attitude quaternion q from the landing coordinate system {L} to the body coordinate system {B}, and the accelerometer drift deviation b. a Angular velocity meter drift deviation b ω .

[0039] In the above-mentioned method for constructing the spatiotemporal sequence features of autonomous optical navigation, in step four, the depth error model is:

[0040]

[0041] In the formula, This represents the depth estimation error for unknown landmarks.

[0042] (k+s,k) represents the value from t k Time to t k +t s time;

[0043] ε v This serves as the error bound for velocity estimation;

[0044] t s This refers to the observation interval.

[0045] Let t be the current time. k r2 estimated by the time-kalman filter;

[0046] Let t be the observation time. k+t s r2 is then recursively estimated using the system state equations described in step one.

[0047] In the aforementioned method for constructing the spatiotemporal features of autonomous optical navigation, the observation image sequence interval t s The calculation method for * is as follows:

[0048]

[0049] In the formula, Δt is the sampling interval time;

[0050] t smax This represents the maximum observation interval.

[0051] In the aforementioned method for constructing the spatiotemporal features of autonomous optical navigation, the next observation time is:

[0052]

[0053] In the aforementioned method for constructing a sequence feature spatiotemporal framework for autonomous optical navigation, the interval between the observed image sequences is... The observation interval that minimizes the depth estimation error of the unknown landmark p2; the maximum observation interval t. smax This is used to limit the observation interval time, which may cause landmark tracking failure.

[0054] The advantages of this invention compared to the prior art are:

[0055] (1) The spatiotemporal construction method of sequence features proposed in this invention selects the image features corresponding to the unknown landmarks that contribute the most to navigation accuracy in the field of view by optimizing the proposed landmark spatial configuration model, and obtains the optimal sequence feature observation interval by optimizing the depth error model.

[0056] (2) This invention selects the best landmark by optimizing a one-dimensional bivariate convex function form landmark space configuration model, which greatly reduces the amount of computation required for landmark space configuration planning;

[0057] (3) By solving the observation interval time that minimizes the proposed depth error model, this invention obtains the optimal observation interval between two observations, which can effectively reduce unnecessary observations and significantly reduce the computational burden caused by image processing. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the observation geometry of the lander and landing site of the present invention;

[0059] Figure 2 This is a schematic diagram of the spatial configuration model of the landmarks of this invention;

[0060] Figure 3This is a schematic diagram illustrating the depth estimation error of the present invention;

[0061] Figure 4 This is a diagram showing the 3σ envelope curves of the location error for different land landmark spatial configuration planning methods of this invention;

[0062] Figure 5 This is a schematic diagram illustrating the algorithm consumption time of the different land landmark spatial configuration planning methods of the present invention;

[0063] Figure 6 This is a schematic diagram of the 3σ envelope curve of the position error at different observation intervals in this invention;

[0064] Figure 7 This is a schematic diagram illustrating the number of observations at different observation intervals according to the present invention. Detailed Implementation

[0065] The present invention will be further described below with reference to the embodiments.

[0066] For autonomous optical navigation based on sequential images, existing methods for landmark spatial configuration planning require complex matrix operations, resulting in high computational costs, and lack temporal planning methods to reduce image processing computation. This method proposes a sequential feature-based spatiotemporal construction approach for autonomous optical navigation, reducing the computational cost of landmark spatial configuration planning and decreasing the number of landmark observations to lower image processing computation.

[0067] The spatiotemporal construction method for sequence features includes the following steps:

[0068] Step 1: Establish the system state equation for vision-assisted inertial navigation in an unknown environment. The system state equation for vision-assisted inertial navigation in an unknown environment is as follows:

[0069]

[0070] In the formula, r, v, and a are the position, velocity, and acceleration of the lander in the landing coordinate system {L}, respectively;

[0071] q is the attitude quaternion from the landing coordinate system {L} to the body coordinate system {B};

[0072] Let {B} be the representation of the three-axis rotational angular velocity of the body coordinate system {B} relative to the landing coordinate system {L} in the body coordinate system {B}.

[0073] Ω(ω) is the coefficient matrix of the attitude motion equation;

[0074] b a and b ω These are the drift deviations of the accelerometer and the angular velocity meter, respectively.

[0075] n wa and nwω Driver b a and b ω Zero-mean white noise;

[0076] r pi For land standard p i The coordinates in the landing coordinate system {L}; set one known landmark p1 and any one unknown landmark p2; where p1 is at the origin of the landing coordinate system {L}.

[0077] Step 2: Establish a spatial configuration model of landmarks; based on the spatial configuration model of landmarks, select the image features corresponding to the unknown landmarks in the field of view that contribute the most to navigation accuracy.

[0078] The spatial configuration model of the landmark is as follows:

[0079]

[0080] In the formula, δr is the estimation error bound of the lander's position in the landing coordinate system {L};

[0081] r represents the position of the lander in the landing coordinate system {L};

[0082] α represents the position r of the lander in the landing coordinate system {L} and the coordinates r of p2 in the landing coordinate system {L}. p2 The angle between them;

[0083] β is the angle between r1 and r2; r i Let i be the position of the i-th landmark relative to the lander;

[0084] δr p Let r be the coordinates of p2 in the landing coordinate system {L} p2 The estimation error bound;

[0085] ε θ Let be the attitude estimation error bound.

[0086] The method for selecting the unknown landmark in the field of view that contributes the most to navigation accuracy is as follows:

[0087] Find the optimal value of α in the landmark spatial configuration model. * The optimal value of β * ;

[0088]

[0089] When the triangle formed by the position of the unknown landmark p2, the lander position, and the position of p1 satisfies α * and β * When considering the relationship, the position of the unknown landmark p2 contributes the most to the observability.

[0090] Solving for α * and β * The optimal position of p2 is relative to the direction of the lander. By extracting feature points from the image along this direction, the image features corresponding to p2 can be obtained.

[0091] Step 3: Establish a Kalman filter based on the state equation of the vision-assisted inertial navigation system; observe landmarks and use the observation information as input to the Kalman filter to obtain the optimal estimates of each state variable in the state equation of the vision-assisted inertial navigation system; use the optimal estimates of each state variable as navigation parameters to perform landing navigation for the lander.

[0092] The state variables in the state equation of a vision-assisted inertial navigation system include the lander's position r in the landing coordinate system {L}, the lander's velocity v in the landing coordinate system {L}, the attitude quaternion q from the landing coordinate system {L} to the body coordinate system {B}, and the accelerometer drift bias b. a Angular velocity meter drift deviation b ω .

[0093] Step 4: Using the optimal estimates of each state variable in the state equation of the vision-assisted inertial navigation system as input to the depth error model, calculate the interval of the observed image sequence. And the next observation time; combined with the image features corresponding to the unknown landmarks that contribute the most to navigation accuracy selected spatially in step two, and the best observation time determined temporally in step four, the spatiotemporal construction of sequence features is obtained.

[0094] The depth error model is as follows:

[0095]

[0096] In the formula, This represents the depth estimation error for unknown landmarks.

[0097] (k+s,k) represents the value from t k Time to t k +t s time;

[0098] ε v This serves as the error bound for velocity estimation;

[0099] t s This refers to the observation interval.

[0100] Let t be the current time. k r2 estimated by the time-kalman filter;

[0101] Let t be the observation time. k +t sr2 is then recursively estimated using the system state equations described in step one.

[0102] Observation image sequence interval The calculation method is as follows:

[0103]

[0104] In the formula, Δt is the sampling interval time;

[0105] t smax This represents the maximum observation interval.

[0106] The next observation time is:

[0107]

[0108] Observation image sequence interval The observation interval that minimizes the depth estimation error of the unknown landmark p2; the maximum observation interval t. smax This is used to limit the observation interval time, which may cause landmark tracking failure.

[0109] Step 5: At the next observation time, return to Step 3, re-observe the landmark, and use the observation information as input to the Kalman filter to obtain the optimal estimates of each state variable in the state equation of the vision-assisted inertial navigation system; use the optimal estimates of each state variable as the navigation parameters at that time for the landing navigation of the lander.

[0110] Step 6: Repeat steps 3 to 5 until the lander has landed successfully; if the land marker needs to be replaced, return to step 2; otherwise, return to step 4.

[0111] Example 1:

[0112] A method for constructing the spatiotemporal sequence features of autonomous optical navigation includes the following steps:

[0113] (1) Establish the state equation of the visual-assisted inertial navigation system in an unknown environment.

[0114] Define the following coordinate systems: landing coordinate system {L}, lander body coordinate system {B}, as follows: Figure 1 As shown. The landing coordinate system is defined as the northeast celestial reference coordinate system fixed on the planetary surface. The lander body coordinate system {B} is fixed on the lander, with its origin at the lander's center of mass, denoted by C.

[0115] The state equation of a vision-assisted inertial navigation system in an unknown environment is:

[0116]

[0117] In the formula: r, v, and a are the position, velocity, and acceleration of the lander in the landing coordinate system {L}, q is the attitude quaternion from the landing coordinate system {L} to the body coordinate system {B}, and ω = [ω x ω y ω z ] T Let {B} be the rotational angular velocity of this system {B} relative to the landing coordinate system {L} in {B}. [ω×] is the antisymmetric cross product of ω. b a and b ω These are the drift deviations of the accelerometer and angular velocity meter, respectively, n wa and n wω To drive b a and b ω Zero-mean white noise. One known landmark (hereinafter referred to as known landmark) p1 and one unknown landmark (hereinafter referred to as unknown landmark) p2 were observed in the landing system {L}. pi For land standard p i The coordinates in the landing coordinate system {L}. p1 is located at the origin of the landing coordinate system {L}.

[0118] (2) Optimize the spatial configuration model of landmarks and select the image features corresponding to the unknown landmarks that contribute the most to navigation accuracy in the field of view.

[0119] Depend on Figure 2 It can be seen that the lander position estimation error is caused by both the attitude estimation error and the position error of p2. The attitude estimation error bound ε θ The position error limit δr of p2 p It can be obtained from the covariance matrix of the extended Kalman filter.

[0120] The positional error δr along the r1 direction is used as an indicator of the observability of the spatial landmark configuration:

[0121]

[0122] Optimal landmark orientation α * and β * This can be obtained by minimizing the observability δr of the spatial landmark configuration:

[0123]

[0124] Solving for α * and β * The optimal position of p2 relative to the lander's direction corresponds to the direction in which feature points are extracted from the image along this direction to obtain the features of p2.

[0125] (3) Observe the landmarks and combine them with the system state equations described in (1) to perform Kalman filtering to obtain the optimal state estimate.

[0126] The system state vector can be defined as follows:

[0127]

[0128] Define the lander error state as The estimated value of x is The error state vector is represented as

[0129]

[0130] In the formula: δθ represents the attitude angle error. The continuous-time state equation for the lander's error state can be expressed as:

[0131]

[0132] In the formula: system noise is defined as n a and n ω Measurement noise is in the form of zero-mean Gaussian white noise. I3∈R 3×3 It is a 3×3 identity matrix, 0 3×3 ∈R 3×3 It is a 3×3 zero matrix.

[0133] In observing n s Known landmarks and n p When there is an unknown landmark The discrete-time state equation from the i-th to the j-th sampling time can be expressed as:

[0134]

[0135] In the formula: the upper right subscript (i) represents the i-th sampling time, (j,i) represents the sampling times from the i-th to the j-th sampling time, and the lower right subscript n s ,n p For observing n s Known landmarks and n p An unknown landmark, It is by Received

[0136]

[0137] In the formula: Δt is the sampling period, t i Let i be the time corresponding to the i-th sampling moment.

[0138]

[0139]

[0140] The measurement at the k-th sampling time is the representation of the orientation of all observed landmarks relative to the lander in {B}:

[0141]

[0142]

[0143] In the formula: r i For the landing system {L}, the lander is positioned relative to the land reference p. i The location. For z (k) Regarding the estimated value The error is such that the measurement equation can be linearized as follows:

[0144]

[0145]

[0146] In the formula: η (k) To measure noise, By using an extended Kalman filter to filter the discrete-time state equation and measurement equation, the optimal state estimates of the lander and the unknown landmark can be estimated simultaneously.

[0147] (4) Optimize the depth error model based on the system state equation described in (1) to obtain the sequence feature observation interval and the next observation time in time.

[0148] First, the geometric definitions in the depth error model are given, such as Figure 3 As shown. Because the location of landmark p2 is unknown, there is an error in the depth estimation along the line of sight. The lander at time t... k from The direction vector of the landmass p2 relative to itself was observed at point 1, i.e. The direction vector. After the observation interval t... s Afterwards, the estimated displacement of the motion is:

[0149]

[0150] In the formula: This can be viewed as the baseline estimated in triangulation. The lander from... After the second observation of the land marker p2, the solution can be obtained using the geometric relations of the triangle. model That is, the line-of-sight depth. However, due to errors in velocity estimation, the actual displacement is:

[0151]

[0152] In the formula: b (k+s,k) This can be considered as the actual baseline in triangulation. The error is caused by baseline estimation error. The estimation error is the depth estimation error.

[0153] The error bound between the actual baseline and the estimated baseline is calculated below. The difference between the actual baseline and the estimated baseline is:

[0154]

[0155] A radius of ε caused by velocity error v t s Error sphere, where: ε v The error bound for velocity estimation can also be obtained from the state covariance matrix of the extended Kalman filter. Therefore, the error bound between the actual baseline and the estimated baseline is a bound with radius ε. v t s Error ball.

[0156] Finally, the depth estimation error is calculated. At time t k +t s The estimated observation direction is

[0157]

[0158] From actual location Line of sight to p2 This constitutes a width of ε v t s And parallel to A cylinder. Therefore, from along The direction is correct Projection, and with The width of the intersecting line segments is the depth estimation error.

[0159]

[0160] In the formula: at t s ∈[Δt,t smax Internal optimization The optimal observation interval can then be obtained. It is the observation interval that minimizes the depth estimation error of the unknown landmark p2.

[0161]

[0162] In the formula: t smax This is the maximum observation interval, used to prevent landmark tracking failure due to excessively long observation intervals.

[0163] (5) At the next observation time, the landmark is re-observed and Kalman filtering is performed in combination with the system state equation described in (1) to obtain the optimal state estimate.

[0164] The observation time t obtained in (4) k +t s The landmarks were re-observed and filtered.

[0165] (6) If the same landmark is observed three times, jump to (2); otherwise jump to (4).

[0166] After three consecutive observations, you need to jump to (2) to select a landmark that contributes significantly to navigation accuracy for observation; otherwise, you can jump to (4) to calculate the next observation time.

[0167] Example 2:

[0168] The method of Example 1 was used to simulate the landing process of the lander on the Martian surface. The orbit was generated by a polynomial guidance law, the landing process lasted 198 seconds, the sampling period of the navigation filter was Δt = 1 second, and the landmarks were uniformly distributed on the Martian surface. The initial position of the lander in the landing coordinate system was [3300 5870 6570]. T m, initial velocity is [-56 -66 -90] T m / s, initial attitude [45 180 0] T °. The correctness of the observability analysis was verified using Monte Carlo simulation. The initial position estimation error range was set to ±1000m, the initial velocity estimation error range to ±10m / s, and the initial attitude estimation error range to ±2°, all following a uniform distribution. 300 Monte Carlo simulations were performed.

[0169] The land landmark spatial configuration model proposed in this patent is compared with traditional observability indices such as singular values, geometric precision factors, and condition numbers.

[0170] like Figure 4 As shown in the 3σ envelope curve of the navigation error, when using the observability index δr proposed in this patent for landmark spatial configuration planning, the error envelope curve decreases faster than when using the traditional observability method. This indicates that when using the method proposed in this paper to plan the landmark spatial configuration, it is possible to select landmarks that contribute more to improving navigation accuracy from the sequence images, making the navigation state solution more accurate and the error convergence faster.

[0171] like Figure 5As shown, when different methods are run in the same simulation environment to plan the landmark space configuration, it can be seen that since the landmark space configuration model proposed in this patent is a single-variable elementary function, it has lower algorithm complexity and higher computational efficiency than the other three traditional indexes containing matrix operations. Therefore, the landmark space configuration model proposed in this patent is significantly better than the other three methods in terms of solution speed and is easier to plan the landmark configuration online on the lander.

[0172] The observation intervals of the land landmarks in each sampling period are compared with the adaptive observation interval optimized by the interval of the observed image sequence.

[0173] like Figure 6 As shown, the positional error of the adaptive observation interval is almost the same as that of the equal observation interval, and even converges faster.

[0174] like Figure 7 The diagram shows the number of observations at each sampling time after 300 Monte Carlo simulations. For any given time, if a landmark is observed at that time in every simulation, the number of observations is 300. The navigation process with equal observation intervals involves observations at each sampling time, resulting in 300 observations at each sampling time. The adaptive observation interval extends the interval between observations. The earlier the landing begins, the fewer the observations, because the state estimation error is larger at the start of landing, requiring a larger baseline length to reduce depth estimation error. As time progresses, the state estimation error gradually decreases, and an excessively large baseline length can actually increase depth estimation error, necessitating continuous landmark observations to maintain navigation accuracy. Therefore, it can be seen that the observation timing planning method can adaptively extend the time interval for landmark observations without significantly affecting navigation accuracy, thereby reducing the computational burden on the lander for image processing. In practical engineering, the observation interval is designed to fit the computational constraints on the lander. Using an adaptive observation interval can theoretically further reduce unnecessary observations and save more computational resources.

[0175] In summary, the above embodiments verify that the proposed sequence feature spatiotemporal construction method can find the landmarks that contribute the most to navigation accuracy with less computation, and can adaptively increase the time interval for observing landmarks, thereby further reducing the computational burden of image processing on the lander.

[0176] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications to the technical solutions of the present invention by utilizing the methods and techniques disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall fall within the protection scope of the technical solutions of the present invention.

Claims

1. A method for constructing the spatiotemporal sequence features of autonomous optical navigation, characterized in that: include: Step 1: Establish the system state equations for vision-assisted inertial navigation in an unknown environment; the variables in the system state equations include the position of the lander in the landing coordinate system {L}. r The lander's velocity in the landing coordinate system {L} v、 acceleration of the lander in the landing coordinate system {L} a Attitude quaternions from landing coordinate system {L} to body coordinate system {B} accelerometer drift deviation Angular velocity meter drift deviation Landmark Coordinates in the landing coordinate system {L} ; Step 2: Establish a spatial configuration model of landmarks; based on the spatial configuration model of landmarks, select the image features corresponding to the unknown landmarks in the field of view that contribute the most to navigation accuracy. The spatial configuration model of the landmark is as follows: In the formula, This is the error bound for estimating the lander's position in the landing coordinate system {L}; Let L be the position of the lander in the landing coordinate system {L}; Let {L} be the position of the lander in the landing coordinate system. and Coordinates in the landing coordinate system {L} The angle between them; for and The included angle; Let i be the position of the i-th landmark relative to the lander; for Coordinates in the landing coordinate system {L} The estimation error bound; This serves as the attitude estimation error bound. Step 3: Establish a Kalman filter based on the state equation of the vision-assisted inertial navigation system; The system observes landmarks and uses the observation information as input to a Kalman filter to obtain the optimal estimates of each state variable in the state equation of the vision-assisted inertial navigation system. The optimal estimates of each state variable are then used as navigation parameters to perform landing navigation for the lander. Step 4: Using the optimal estimates of each state variable in the state equation of the vision-assisted inertial navigation system as input to the depth error model, calculate the interval of the observed image sequence. And the next observation time; combined with the image features corresponding to the unknown landmarks that contribute the most to navigation accuracy selected spatially in step two, and the best observation time determined temporally in step four, the spatiotemporal construction of sequence features is obtained; Step 5: At the next observation time, return to Step 3, re-observe the landmark, and use the observation information as input to the Kalman filter to obtain the optimal estimates of each state variable in the state equation of the vision-assisted inertial navigation system; use the optimal estimates of each state variable as the navigation parameters at that time for the landing navigation of the lander. Step 6: Repeat steps 3 to 5 until the lander has landed successfully; if the land marker needs to be replaced, return to step 2; otherwise, return to step 4.

2. The method for constructing a sequence feature spatiotemporal structure for autonomous optical navigation according to claim 1, characterized in that: In step one, the system state equation for vision-assisted inertial navigation in an unknown environment is: In the formula, r , v and a These represent the lander's position, velocity, and acceleration in the landing coordinate system {L}, respectively. Let {L} be the attitude quaternion from the landing coordinate system {L} to the body coordinate system {B}; Let {B} be the representation of the three-axis rotational angular velocity of the body coordinate system {B} relative to the landing coordinate system {L} in the body coordinate system {B}. This is the coefficient matrix of the attitude motion equation; and These are the drift deviations of the accelerometer and the angular velocity meter, respectively. and drivers respectively and Zero-mean white noise; Landmark Coordinates in the landing coordinate system {L}; 1 known landmark is defined. And any 1 unknown landmark ;in, At the origin of the landing coordinate system {L}.

3. The method for constructing a sequence feature spatiotemporal structure for autonomous optical navigation according to claim 2, characterized in that: The method for selecting the unknown landmark in the field of view that contributes the most to navigation accuracy is as follows: Solving the spatial configuration model of landmarks optimal value , optimal value ; When unknown landmarks The position and the lander position and The triangle formed by the positions satisfies and When the relationship is unknown, the landmass The location contributes the most to observability.

4. A method for constructing the spatiotemporal sequence features of autonomous optical navigation according to claim 3, characterized in that: The solution and Corresponding The optimal position is relative to the lander's orientation; feature points can be extracted from the image along this direction. The corresponding image features.

5. A method for constructing the spatiotemporal sequence features of autonomous optical navigation according to claim 4, characterized in that: In step three, the state variables in the state equation of the vision-assisted inertial navigation system include the lander's position r in the landing coordinate system {L}, the lander's velocity v in the landing coordinate system {L}, and the attitude quaternion from the landing coordinate system {L} to the body coordinate system {B}. accelerometer drift deviation Angular velocity meter drift deviation .

6. A method for constructing the spatiotemporal sequence features of autonomous optical navigation according to claim 2, characterized in that: In step four, the depth error model is as follows: In the formula, This represents the depth estimation error for unknown landmarks. From Time's up time; This serves as the error bound for velocity estimation; This refers to the observation interval. For the current moment Time Kalman filter estimation ; For the observation time The system state equation described in step one is used for recursive estimation. .

7. A method for constructing the spatiotemporal sequence features of autonomous optical navigation according to claim 6, characterized in that: Observation image sequence interval The calculation method is as follows: In the formula, This is the sampling interval time; This represents the maximum observation interval.

8. A method for constructing the spatiotemporal sequence features of autonomous optical navigation according to claim 7, characterized in that: The next observation time is: 。 9. A method for constructing the spatiotemporal sequence features of autonomous optical navigation according to claim 8, characterized in that: The interval of the observed image sequence Unknown land landmark The observation interval with the smallest depth estimation error; the maximum observation interval. This is used to limit the observation interval time, which may cause landmark tracking failure.

Images