Autonomous estimation method of shape and motion of space targets based on monocular sequence images

Abstract

A method for autonomously estimating the shape and motion of a space target based on monocular sequence images comprises the following steps: establishing a basic equation for binocular vision measurement, a three-dimensional reconstruction equation for a space target point, and a target motion state measurement equation based on sequence images; setting a conversion matrix from a camera system to a principal axis system of a previous frame equal to a unit matrix, matching adjacent frame images, and obtaining a transformation matrix of a target inertial principal axis coordinate system between adjacent frame imaging moments; using the three-dimensional reconstruction equation for the space target point, obtaining a position vector set of a target feature point in a left camera coordinate system and a position vector set of a target feature point in a right camera coordinate system; then unifying the scales of adjacent frames; re-determining a position vector set of a target feature point in a left camera coordinate system and a right camera coordinate system; calculating and obtaining a posture quaternion, a spin angular velocity, and a spin axis direction between adjacent frames; and obtaining a shape estimation of the target by matching and calculating a sequence image of a selected cycle of the target.

Description

Technical Field

[0001] The invention relates to a method for autonomously estimating the shape and motion of a space target based on monocular sequence images, and belongs to the technical field of space relative navigation. Background Art

[0002] With the advancement of major engineering projects such as manned space flight, deep space exploration, and on-orbit services in my country, it is necessary to complete complex tasks such as the construction and operation support of space stations, satellite assembly and services, and planetary surface exploration and testing. Spacecraft must be able to provide various on-orbit operation services, such as target capture, target transfer and release, on-orbit assembly, on-orbit services (fuel filling, module replacement, etc.), on-orbit manufacturing, assisting astronauts in extravehicular activities, and even the surface exploration of unknown extraterrestrial bodies. It can be foreseen that future on-orbit operation tasks will have the characteristics of uncertain environment, many constraints, high reliability requirements, diverse and complex tasks, and the objects of operation are often a type of space non-cooperative targets without identification, response, and prior knowledge. The current mode based on program control or man-in-the-loop teleoperation for deterministic tasks has been difficult to meet the flexible, agile, and precise motion requirements of on-orbit control for such high-risk and uncertain tasks due to problems such as action determination or large delay. This requires spacecraft to autonomously perceive targets, autonomously plan actions, and autonomously control operations. Among them, autonomous navigation for space non-cooperative targets is the basic premise and key technology for spacecraft to provide on-orbit operation services.

[0003] The autonomous navigation capability of the navigation system is mainly limited by navigation sensors, onboard computers and onboard navigation algorithms. In order to improve navigation capabilities, spacecraft usually use a multi-sensor configuration scheme that backs up each other, such as microwave radar, lidar, and optical camera. However, due to the extremely complex fusion structure and filtering algorithm, and the strict constraints of onboard resources, it can only be switched in segments, making it difficult to achieve true multi-source information fusion navigation. As a standard configuration of spacecraft, the optical sensor it obtains has the characteristics of full applicability and rich information, and is an ideal source of information for autonomous navigation of non-cooperative targets in space. At present, there are four main categories of visual navigation methods for non-cooperative targets in space: (1) Navigation methods based on binocular vision, such as Tweddle BE, a scholar from MIT in the United States, who proposed a three-dimensional reconstruction and attitude measurement method for non-cooperative and spinning targets based on a binocular camera, and conducted on-orbit experiments in the cabin of the International Space Station; (2) Monocular vision navigation methods based on terrain registration, such as the US "Perseverance" Mars rover, which uses image matching technology based on the landing area terrain model to achieve obstacle avoidance and fixed-point landing in the landing area; (3) Monocular vision navigation methods based on landmarks, such as Japan's "Hayabusa-1" dropped a luminous cooperative landmark ball onto the surface of a small celestial body during the attachment phase, and used a monocular optical camera to achieve relative posture measurement; (4) Modeling and motion estimation methods based on SLAM (simultaneous localization and mapping), such as Japanese scholar Takeishi studied modeling and motion estimation methods based on particle filtering and nonlinear optimization in the "Hayabusa-2" mission.

[0004] The main problems of the existing technology are:

[0005] (1) The navigation method based on binocular vision has a range and measurement accuracy that is heavily dependent on the baseline length. It is only applicable to short-distance measurements within a few meters and cannot be used for long-distance navigation ranging from hundreds of meters to tens of kilometers.

[0006] (2) Monocular vision navigation based on terrain registration requires a high-precision 3D model of the target to be known in advance. This usually requires circumnavigation of the target to obtain a large number of multi-viewpoint images, which are then transmitted back to the ground for processing. This model takes several months to several years to complete. It is mostly suitable for tasks where the target is determined and the boundary conditions are known, but cannot meet the requirements for rapid navigation tasks for highly maneuverable targets with unknown physical properties.

[0007] (3) The monocular vision navigation method based on markers requires the placement of luminous markers on the target, which greatly limits the scope of application and is difficult to apply to targets such as failed satellites and space debris.

[0008] (4) The current SLAM technology is mainly applicable to the field of ground mobile robots, which requires the scene to be static. However, for the application scenario of relative navigation of non-cooperative targets in space, both the target and the spacecraft are moving objects. The traditional methods cannot be directly applied to the satellite. The research algorithms currently given are mainly used in the scenario of small celestial body detection. The algorithms are complex and have high computational overhead, which makes it difficult to meet the stringent resource constraints on the satellite. Summary of the invention

[0009] The technical problem to be solved by the present invention is: to overcome the shortcomings of the prior art and provide a method for autonomously estimating the shape and motion of a space target based on a monocular sequence image. Aiming at the problem of estimating the three-dimensional shape and motion parameters of a space non-cooperative target, the target shape and motion information contained in the monocular sequence image is utilized, and the registration technology between adjacent images is used to realize the measurement of the three-dimensional shape, angular velocity, and spin axis direction of the target. The algorithm is mainly matrix algebraic operations, does not require a complex filtering estimation process, has low computational overhead, fast convergence speed, and is easy to implement on a satellite.

[0010] The purpose of the present invention is achieved through the following technical solutions:

[0011] A method for autonomously estimating the shape and motion of a space target based on monocular sequence images comprises the following steps:

[0012] S1. Establish the basic equations of binocular vision measurement;

[0013] S2. Using the basic equations of binocular vision measurement, establish the three-dimensional reconstruction equations of the spatial target points;

[0014] S3. Using the basic equation of binocular vision measurement, establish the target motion state measurement equation based on the sequence image;

[0015] S4, setting the conversion matrix from the camera system to the main axis system of the previous frame equal to the unit matrix, matching the adjacent frame images, and obtaining the transformation matrix of the target inertial main axis coordinate system between the adjacent frame imaging moments;

[0016] Using the three-dimensional reconstruction equation of the spatial target point, a position vector set of the target feature point in the left camera coordinate system and a position vector set of the target feature point in the right camera coordinate system are obtained;

[0017] The scales of the adjacent frames are unified by using the position vector sets of the target feature points in the left camera coordinate system and the right camera coordinate system;

[0018] According to the unified scales of adjacent frames, the position vector sets of the target feature points in the left camera coordinate system and the right camera coordinate system are re-determined;

[0019] The transformation matrix of the target inertial principal axis coordinate system between the imaging moments of adjacent frames is used to calculate the attitude quaternion, spin angular velocity and spin axis direction between adjacent frames;

[0020] By using the position vector set of the re-determined target feature points in the left camera coordinate system and the right camera coordinate system, the shape estimation of the target can be obtained by matching and calculating the sequence images of a circle of the target.

[0021] In one embodiment of the present invention, the basic equation of binocular vision measurement is:

[0022]

[0023]

[0024] Where P l is the position of the spatial target point in the left camera coordinate system, P r is the position of the spatial target point in the right camera coordinate system; T is the position of the left camera coordinate system in the right camera coordinate system, R is the transformation matrix from the left camera coordinate system to the right camera coordinate system, p l The image point of the spatial target point on the left camera, p r The image point of the spatial target point imaged by the left camera, E is the essential matrix describing the limit constraint of binocular vision, and F is the basic matrix describing the limit constraint of binocular vision; A l and A r are the known intrinsic parameter matrices of the left camera and the right camera respectively, It is a skew-symmetric matrix composed of T.

[0025] In one embodiment of the present invention, the three-dimensional reconstruction equation of the spatial target point is:

[0026]

[0027] Where V i Determined by the following formula:

[0028]

[0029]

[0030]

[0031]

[0032]

[0033]

[0034] mp3=mp2(span(SM1,N,1)-span(SM2,N,1))

[0035] mp3(4(i-1)+1:4i,:)=U i S i V i T ,i=1,2,...,N

[0036] In the formula, A l and A r is the known intrinsic parameter matrix of the left camera and the right camera; for any matrix M, M(n,m) represents the element of the nth row and mth column of the matrix M. When n=: or m=:, it means selecting all columns or rows. span(M,n,m) represents a larger matrix composed of the matrix M as elements. n represents a row arrangement of n matrices, and m represents a column arrangement of m M matrices. N is the number of spatial target points, p li and p ri , i = 1, 2, ..., N represents the image point of the i-th spatial target point imaged by the left camera and the right camera, P l,i , i=1,2,…,N represents the position of the i-th spatial target point in the left camera coordinate system, U i S i V i T It is the result of the matrix singular value operation of mp3(4(i-1)+1:4i,:).

[0037] In one embodiment of the present invention, the target motion state measurement equation based on the sequence image is:

[0038] q e =quat(R e )

[0039] θ=π-2sin -1 (q e (4)

[0040]

[0041]

[0042] In the formula, R e is the attitude transformation matrix of the target inertial axis coordinate system between adjacent frame imaging moments obtained by image matching, θ is the spin angle of the target between adjacent image imaging moments, is the target spin angular velocity, e is the target spin axis direction between adjacent image imaging moments, ΔT is the time interval between adjacent frame imaging moments; π is pi, and quat is the quaternion inverse transformation operator symbol.

[0043] In one embodiment of the present invention, the space target is a non-cooperative target.

[0044] A device for autonomously estimating the shape and motion of a space target based on monocular sequence images, comprising a first module and a second module, wherein:

[0045] The first module is used to establish the basic equations of binocular vision measurement; using the basic equations of binocular vision measurement, establish the three-dimensional reconstruction equations of spatial target points; using the basic equations of binocular vision measurement, establish the target motion state measurement equation based on sequence images;

[0046] In the second module, the conversion matrix from the camera system to the main axis system of the previous frame is set to be equal to the unit matrix, and the adjacent frame images are matched to obtain the transformation matrix of the target inertial main axis coordinate system between the adjacent frame imaging moments; the three-dimensional reconstruction equation of the spatial target point is used to obtain the position vector set of the target feature points in the left camera coordinate system and the position vector set of the target feature points in the right camera coordinate system; the position vector set of the target feature points in the left camera coordinate system and the right camera coordinate system is used to unify the scales of the adjacent frames; based on the unified scales of the adjacent frames, the position vector set of the target feature points in the left camera coordinate system and the right camera coordinate system is re-determined; the transformation matrix of the target inertial main axis coordinate system between the adjacent frame imaging moments is used to calculate the attitude quaternion, spin angular velocity and spin axis direction between the adjacent frames; the shape estimation of the target can be obtained by matching and calculating the sequence images of one cycle of the target selection using the re-determined position vector set of the target feature points in the left camera coordinate system and the right camera coordinate system.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] (1) The non-cooperative target shape and motion estimation method obtained by the present invention only uses the sequence images provided by the monocular camera, breaking the constraints of the traditional binocular camera-based measurement algorithm that is limited by the baseline length and the inability to improve the working distance and measurement accuracy. The present method has no constraints on the target size and working distance, and is suitable for small celestial body detection missions with large targets and long observation distances.

[0049] (2) The non-cooperative target shape and motion estimation method obtained by the present invention does not require prior information such as the three-dimensional model of the target and the cooperative mark, and can obtain the shape and motion estimation of the target by only using the sequence images of the monocular camera;

[0050] (3) The non-cooperative target shape and motion estimation method obtained by the present invention is different from the traditional filtering estimation method. It can obtain the target motion estimation by only using the image matching technology, avoiding the stability and convergence time problems existing in the filtering algorithm. In theory, only two images of the initial adjacent frames are needed to obtain the target motion estimation value;

[0051] (4) The non-cooperative target shape and motion estimation method obtained by the present invention only uses image matching technology, has a small amount of calculation and a stable algorithm. Different from the traditional algorithm that only estimates the target motion state, the present method can not only realize motion estimation, but also obtain the shape estimation of the target, truly realizing the simultaneous modeling and simultaneous estimation of space moving targets. Moreover, the algorithm only uses basic matrix-vector operations, which is convenient for code transplantation and embedded implementation, and can well meet the real-time computing requirements on the satellite. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flowchart of the method of the present invention;

[0053] Figure 2 Basic geometric relationships for visual measurement;

[0054] Figure 3 Target measurement geometric relationship based on sequential images. DETAILED DESCRIPTION

[0055] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0056] The present invention discloses a method for autonomously estimating the shape and motion of a space target based on a monocular sequence image, and the steps are as follows: (1) establishing a basic equation for binocular vision measurement; (2) establishing a three-dimensional reconstruction equation for a space target point based on the basic equation for binocular vision measurement in step (1); (3) establishing a target motion state measurement equation based on the sequence image based on the basic equation for binocular vision measurement in step (1); (4) determining the target shape and motion estimation based on the basic equation for binocular vision measurement in step (1), the three-dimensional reconstruction equation established in step (2), and the target motion state measurement equation based on the sequence image established in step (3), such as Figure 1 shown.

[0057] Aiming at the problem of estimating the three-dimensional shape and motion parameters of non-cooperative targets in space, an autonomous estimation method for the shape and motion of space targets based on monocular sequence images was proposed. The target shape and motion information contained in the monocular sequence images was utilized, and the registration technology between adjacent images was used to realize the measurement of the target's three-dimensional shape, angular velocity, and spin axis direction. The algorithm mainly involves matrix algebraic operations and does not require complex filtering estimation processes. It has low computational overhead, fast convergence speed, and is easy to implement on board.

[0058] Depend on Figure 1 As shown in FIG. 1 , a method for autonomously estimating the shape and motion of a spatial target based on a monocular sequence image is shown in FIG. 1 . The steps are as follows:

[0059] (1) Establish the basic equations of binocular vision measurement;

[0060] (2) establishing a three-dimensional reconstruction equation for the spatial target point based on the basic equation of binocular vision measurement in step (1);

[0061] (3) Based on the basic equation of binocular vision measurement in step (1), establish a target motion state measurement equation based on the sequence image;

[0062] (4) Determine the target shape and perform motion estimation based on the basic equation of binocular vision measurement in step (1), the three-dimensional reconstruction equation established in step (2), and the target motion state measurement equation based on the sequence image established in step (3).

[0063] Step 1: Establish the basic equations for binocular vision measurement

[0064] Figure 2 Describes the basic geometric relationship of visual measurement, coordinate system O l X l Y l Z l is the left camera coordinate system, O r X r Y r Z r is the right camera coordinate system, and the position of the spatial target point in the left camera coordinate system and the right camera coordinate system is described as

[0065] P l =PR l +T l

[0066] P r =PR r +T r

[0067] Among them, P is the position of the spatial target point in the world coordinate system, T l and R l is the coordinate transformation from the world coordinate system to the left camera coordinate system, T r and R r is the coordinate transformation from the world coordinate system to the right camera coordinate system, P l is the position of the spatial target point in the left camera coordinate system, P r is the position of the spatial target point in the right camera coordinate system.

[0068] When the transformation matrices T and R from the left camera coordinate system to the right camera coordinate system are known, the following relationship can also be obtained:

[0069] P r =RP l +T

[0070] The above relationship forms a closed operation, and the following relationship can be obtained

[0071]

[0072]

[0073] R r =RR l

[0074]

[0075] Then, the image points of the spatial target point on the left camera and the right camera are

[0076] p l =P l A l

[0077] p r =P r A r

[0078] According to the measurement geometric relationship, we can get

[0079]

[0080]

[0081] in, and E and F are the essential matrix and basic matrix describing the limit constraints of binocular vision, A l and A r is the known internal parameter matrix of the left camera and the right camera. The basic matrix F can be obtained by imaging the image point p of the left camera and the right camera. l and p r Matching is obtained, and then through the formula Get the essential matrix and further decompose it Get T and R, T is the position of the left camera coordinate system in the right camera coordinate system, R is the transformation matrix from the left camera coordinate system to the right camera coordinate system, It is a skew-symmetric matrix composed of T.

[0082] Step 2: Based on the basic equation of binocular vision measurement in step (1), establish the three-dimensional reconstruction equation of the spatial target point.

[0083] After the left and right images are registered to obtain T and R, the three-dimensional reconstruction of the spatial target point can be achieved. Assuming that the left camera is in the world coordinate system, then the imaging projection matrix from the world coordinate system to the left and right cameras is

[0084]

[0085]

[0086]

[0087]

[0088]

[0089]

[0090] mp3=mp2(span(SM1,N,1)-span(SM2,N,1))

[0091] mp3(4(i-1)+1:4i,:)=U i S i V i T ,i=1,2,...,N

[0092]

[0093] Among them, A l and A r is the known intrinsic parameter matrix of the left camera and the right camera; for any matrix M, M(n,m) represents the element of the nth row and mth column of the matrix M. When n=: or m=:, it means selecting all columns or rows. span(M,n,m) represents a larger matrix composed of the matrix M as elements. n represents a row arrangement of n matrices, and m represents a column arrangement of m M matrices. N is the number of spatial target points, p li and p ri , i = 1, 2, ..., N represents the image point of the i-th spatial target point imaged by the left camera and the right camera, P l,i , i=1,2,…,N represents the position of the i-th spatial target point in the left camera coordinate system, U i S i V i T It is the result of the matrix singular value operation of mp3(4(i-1)+1:4i,:).

[0094] Step 3: Based on the basic equation of binocular vision measurement in step (1), establish the target motion state measurement equation based on the sequence image.

[0095] For the sequence of images of a spinning target obtained by a monocular camera, the photographic geometry between the previous frame and the next frame can be equivalent to the binocular measurement geometry, that is, "the target moves, the camera does not move" is equivalent to "the target does not move, the camera moves in the target coordinate system". Figure 3 As shown, the body coordinate system is the inertial principal axis coordinate system. For a spinning target, the rotation transformation matrix R between the target inertial axis coordinate system is equal to the adjacent time intervals of the previous and next frame imaging moments.e are the same, then, using the basic equation of binocular vision measurement established in step (1), we can get the following relationship:

[0096] R l =R0

[0097] T l =T0

[0098] R r =R e R l

[0099] T r =T e R l +T l

[0100]

[0101]

[0102] Among them, T e and R e is the transformation matrix of the target inertial principal axis coordinate system between adjacent frame imaging moments, which is the parameter to be estimated.

[0103] According to the above relationship, the relationship between posture and movement is analyzed as follows

[0104] For the relationship between the first frame and the second frame, we have:

[0105] R l =R0

[0106] R r =RR0

[0107]

[0108] For the relationship between the second and third frames, we have:

[0109] R l =R r =R1R0

[0110] R r =RR l =RRR0

[0111]

[0112] Therefore, for the case where the inter-frame transformation amount R remains unchanged (uniform rotation), each time R e The Euler axis angle of the spin motion is described as θ·e, where the attitude quaternion can be obtained by R e To obtain, that is

[0113] q e =quat(R e )

[0114] Then, the spin axis direction and spin angle are:

[0115] θ=π-2sin -1 (q e (4)

[0116]

[0117]

[0118] Among them, q e is the transformation matrix R obtained by matching adjacent frames e The corresponding attitude quaternion, R e is the attitude transformation matrix of the target inertial axis coordinate system between adjacent frame imaging moments obtained by image matching, θ is the spin angle of the target between adjacent image imaging moments, is the target spin angular velocity, e is the target spin axis direction between adjacent image imaging moments, ΔT is the time interval between adjacent frame imaging moments; π is pi, and quat is the quaternion inverse transformation operator symbol.

[0119] Step 4: Estimate the target shape and motion based on the basic equation of binocular vision measurement in step (1), the 3D reconstruction equation established in step (2), and the target motion state measurement equation based on the sequence images established in step (3).

[0120] For the matched matrix, there is an expression Where R e is the transformation matrix obtained by image matching, R pre is the transformation matrix from the camera system to the main axis system of the previous frame, then the transformation matrix from the camera system to the main axis system of the current frame is R cur =RR pre =R pre R e .

[0121] Since the initial posture is unknown, that is, the matrix R0 is unknown, it is assumed that R0 = I3 (3×3 unit matrix). Then, under the assumption that R0 = I3, only R e The information obtained should be the information of a new coordinate system relative to the camera system. cur =R0R e , which is equivalent to defining a new target system (called the measurement system), and the transformation matrix from the measurement system to the principal axis system is R0.

[0122] Then, the specific process is:

[0123] 1) Start setting R pre =I3;

[0124] 2) Match adjacent frame images to get R e and T e ;

[0125] 3) Using the 3D reconstruction equation in step (2), we can obtain the position vector set of the target feature points in the left camera coordinate system: The position of the target feature point in the right camera coordinate system is

[0126]

[0127] 4) Standardize

[0128] First frame:

[0129] Subsequent frames:

[0130] Among them, s is the scale factor.

[0131] 5) Re-determine the position vector set of the target feature point in the left camera coordinate system and the right camera coordinate system:

[0132] P l =sP l

[0133]

[0134] 6) Using R e , calculate the attitude quaternion between adjacent frames

[0135] q e =quat(R e )

[0136] The spin angular velocity and spin axis direction are:

[0137] θ=π-2sin -1 (q e (4)

[0138]

[0139]

[0140] Among them, ΔT is the imaging time interval between adjacent frame images. It can be seen that for a uniformly spinning target, as long as the adjacent frame images can be matched to obtain a stable R e and T e , the target's spin angular velocity can be obtained through one image matching And the spin axis direction e. By matching and calculating the target's selected sequence of images for one cycle, the target shape estimation P can be obtained. l .

[0141] Example:

[0142] A method for autonomously estimating the shape and motion of a space target based on monocular sequence images comprises the following steps:

[0143] S1. Establish the basic equations of binocular vision measurement;

[0144] S2. Using the basic equations of binocular vision measurement, establish the three-dimensional reconstruction equations of the spatial target points;

[0145] S3. Using the basic equation of binocular vision measurement, establish the target motion state measurement equation based on the sequence image;

[0146] S4, setting the conversion matrix from the camera system to the main axis system of the previous frame equal to the unit matrix, matching the adjacent frame images, and obtaining the transformation matrix of the target inertial main axis coordinate system between the adjacent frame imaging moments;

[0147] Using the three-dimensional reconstruction equation of the spatial target point, a position vector set of the target feature point in the left camera coordinate system and a position vector set of the target feature point in the right camera coordinate system are obtained;

[0148] The scales of the adjacent frames are unified by using the position vector sets of the target feature points in the left camera coordinate system and the right camera coordinate system;

[0149] According to the unified scales of adjacent frames, the position vector sets of the target feature points in the left camera coordinate system and the right camera coordinate system are re-determined;

[0150] The transformation matrix of the target inertial principal axis coordinate system between the imaging moments of adjacent frames is used to calculate the attitude quaternion, spin angular velocity and spin axis direction between adjacent frames;

[0151] By using the position vector set of the re-determined target feature points in the left camera coordinate system and the right camera coordinate system, the shape estimation of the target can be obtained by matching and calculating the sequence images of the target in one cycle.

[0152] A device for autonomously estimating the shape and motion of a space target based on monocular sequence images, comprising a first module and a second module, wherein:

[0153] The first module is used to establish the basic equations of binocular vision measurement; using the basic equations of binocular vision measurement, establish the three-dimensional reconstruction equations of spatial target points; using the basic equations of binocular vision measurement, establish the target motion state measurement equation based on sequence images;

[0154] In the second module, the conversion matrix from the camera system to the main axis system of the previous frame is set to be equal to the unit matrix, and the adjacent frame images are matched to obtain the transformation matrix of the target inertial main axis coordinate system between the adjacent frame imaging moments; the three-dimensional reconstruction equation of the spatial target point is used to obtain the position vector set of the target feature points in the left camera coordinate system and the position vector set of the target feature points in the right camera coordinate system; the position vector set of the target feature points in the left camera coordinate system and the right camera coordinate system is used to unify the scales of the adjacent frames; based on the unified scales of the adjacent frames, the position vector set of the target feature points in the left camera coordinate system and the right camera coordinate system is re-determined; the transformation matrix of the target inertial main axis coordinate system between the adjacent frame imaging moments is used to calculate the attitude quaternion, spin angular velocity and spin axis direction between the adjacent frames; the shape estimation of the target can be obtained by matching and calculating the sequence images of one cycle of the target selection using the re-determined position vector set of the target feature points in the left camera coordinate system and the right camera coordinate system.

[0155] Preferably, the basic equation of binocular vision measurement is:

[0156]

[0157]

[0158] Where P l is the position of the spatial target point in the left camera coordinate system, P r is the position of the spatial target point in the right camera coordinate system; T is the position of the left camera coordinate system in the right camera coordinate system, R is the transformation matrix from the left camera coordinate system to the right camera coordinate system, p l The image point of the spatial target point on the left camera, p r The image point of the spatial target point imaged by the left camera, E is the essential matrix describing the limit constraint of binocular vision, and F is the basic matrix describing the limit constraint of binocular vision; A l and A r are the known intrinsic parameter matrices of the left camera and the right camera respectively, It is a skew-symmetric matrix composed of T.

[0159] Preferably, the three-dimensional reconstruction equation of the spatial target point is:

[0160]

[0161] Where V i Determined by the following formula:

[0162]

[0163]

[0164]

[0165]

[0166]

[0167]

[0168] mp3=mp2(span(SM1,N,1)-span(SM2,N,1))

[0169] mp3(4(i-1)+1:4i,:)=U i S i V i T ,i=1,2,...,N

[0170] In the formula, A l and A r is the known intrinsic parameter matrix of the left camera and the right camera; for any matrix M, M(n,m) represents the element of the nth row and mth column of the matrix M. When n=: or m=:, it means selecting all columns or rows. span(M,n,m) represents a larger matrix composed of the matrix M as elements. n represents a row arrangement of n matrices, and m represents a column arrangement of m M matrices. N is the number of spatial target points, p li and p ri , i = 1, 2, ..., N represents the image point of the i-th spatial target point imaged by the left camera and the right camera, P l,i , i=1,2,…,N represents the position of the i-th spatial target point in the left camera coordinate system, U i S i V i T It is the result of the matrix singular value operation of mp3(4(i-1)+1:4i,:).

[0171] Preferably, the target motion state measurement equation based on the sequence image is:

[0172] q e =quat(R e )

[0173] θ=π-2sin -1 (q e (4)

[0174]

[0175]

[0176] In the formula, R eis the attitude transformation matrix of the target inertial axis coordinate system between adjacent frame imaging moments obtained by image matching, θ is the spin angle of the target between adjacent image imaging moments, is the target spin angular velocity, e is the target spin axis direction between adjacent image imaging moments, ΔT is the time interval between adjacent frame imaging moments; π is pi, and quat is the quaternion inverse transformation operator symbol.

[0177] Preferably, the space target is a non-cooperative target.

[0178] The contents not described in detail in the specification of the present invention belong to the common knowledge of those skilled in the art.

[0179] Although the present invention has been disclosed as above in the form of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art may make possible changes and modifications to the technical solution of the present invention by using the methods and technical contents disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope of the technical solution of the present invention.

Claims

1. A method for autonomously estimating the shape and motion of a spatial target based on monocular sequence images, characterized in that: The steps include: S1. Establish the basic equations of binocular vision measurement; S2. Using the basic equations of binocular vision measurement, establish the three-dimensional reconstruction equations of the spatial target points; The three-dimensional reconstruction equation of the spatial target point is: Where V i Determined by the following formula: mp3=mp2(span(SM1,N,1)-span(SM2,N,1)) mp3(4(i-1)+1:4i,:)=U i S i V i T ,i=1,2,...,N In the formula, A l and A r are the known intrinsic parameter matrices of the left and right cameras; for any matrix M, M(n,m) represents the elements of the nth row and mth column of the matrix M. When n=: or m=:, it means selecting all columns or rows. span(M,n,m) represents a larger matrix composed of the matrix M as elements. n represents a row arrangement of n matrices, and m represents a column arrangement of m M matrices. N is the number of spatial target points, p li and p ri , i = 1, 2, ..., N represents the image point of the i-th spatial target point imaged by the left camera and the right camera, P l,i , i=1,2,…,N represents the position of the i-th spatial target point in the left camera coordinate system, U i S i V i T is the result of the matrix singular value operation of mp3(4(i-1)+1:4i,:); S3. Using the basic equation of binocular vision measurement, establish the target motion state measurement equation based on the sequence image; The target motion state measurement equation based on the sequence image is: what e =quat(R e ) θ=π-2sin -1 (q e (4)) In the formula, R e is the attitude transformation matrix of the target inertial axis coordinate system between adjacent frame imaging moments obtained by image matching, θ is the spin angle of the target between adjacent image imaging moments, is the target spin angular velocity, e is the target spin axis direction between adjacent image imaging moments, ΔT is the time interval between adjacent frame image imaging moments; π is pi, quat is the quaternion inverse transformation operator symbol; S4, setting the conversion matrix from the camera system to the main axis system of the previous frame equal to the unit matrix, matching the adjacent frame images, and obtaining the transformation matrix of the target inertial main axis coordinate system between the adjacent frame imaging moments; Using the three-dimensional reconstruction equation of the spatial target point, a position vector set of the target feature point in the left camera coordinate system and a position vector set of the target feature point in the right camera coordinate system are obtained; The scales of the adjacent frames are unified by using the position vector sets of the target feature points in the left camera coordinate system and the right camera coordinate system; According to the unified scales of adjacent frames, the position vector sets of the target feature points in the left camera coordinate system and the right camera coordinate system are re-determined; The transformation matrix of the target inertial principal axis coordinate system between adjacent frame imaging moments is used to calculate the attitude quaternion, spin angular velocity and spin axis direction between adjacent frames; By using the position vector set of the re-determined target feature points in the left camera coordinate system and the right camera coordinate system, the target shape estimation can be obtained by matching and calculating the target selection sequence images. Space targets are non-cooperative targets.

2. The method for autonomously estimating the shape and motion of a space target according to claim 1, characterized in that: The basic equation of binocular vision measurement is: Where P l is the position of the spatial target point in the left camera coordinate system, P r is the position of the spatial target point in the right camera coordinate system; T is the position of the left camera coordinate system in the right camera coordinate system, R is the transformation matrix from the left camera coordinate system to the right camera coordinate system, p l The image point of the spatial target point on the left camera, p r The image point of the spatial target point imaged by the left camera, E is the essential matrix describing the limit constraint of binocular vision, and F is the basic matrix describing the limit constraint of binocular vision; A l and A r are the known intrinsic parameter matrices of the left camera and the right camera respectively, It is a skew-symmetric matrix composed of T.

3. A device for autonomously estimating the shape and motion of a space target based on monocular sequence images, characterized in that: It includes a first module and a second module, wherein: The first module is used to establish the basic equations of binocular vision measurement; using the basic equations of binocular vision measurement, establish the three-dimensional reconstruction equations of spatial target points; using the basic equations of binocular vision measurement, establish the target motion state measurement equation based on sequence images; The second module sets the conversion matrix from the camera system to the main axis system of the previous frame equal to the unit matrix, matches the adjacent frame images, and obtains the transformation matrix of the target inertial main axis coordinate system between the adjacent frame imaging moments; uses the three-dimensional reconstruction equation of the spatial target point to obtain the position vector set of the target feature point in the left camera coordinate system and the position vector set of the target feature point in the right camera coordinate system; uses the position vector set of the target feature point in the left camera coordinate system and the right camera coordinate system to unify the scales of the adjacent frames; based on the unified scales of the adjacent frames, redetermines the position vector set of the target feature point in the left camera coordinate system and the right camera coordinate system; uses the transformation matrix of the target inertial main axis coordinate system between the adjacent frame imaging moments to calculate the attitude quaternion, spin angular velocity and spin axis direction between the adjacent frames; uses the re-determined position vector set of the target feature point in the left camera coordinate system and the right camera coordinate system to obtain the shape estimation of the target by matching and calculating a sequence of images of the target for one cycle; The three-dimensional reconstruction equation of the spatial target point is: Where V i Determined by the following formula: mp3=mp2(span(SM1,N,1)-span(SM2,N,1)) mp3(4(i-1)+1:4i,:)=U i S i V i T ,i=1,2,...,N In the formula, A l and A r is the known intrinsic parameter matrix of the left camera and the right camera; for any matrix M, M(n,m) represents the element of the nth row and mth column of the matrix M. When n=: or m=:, it means selecting all columns or rows. span(M,n,m) represents a larger matrix composed of the matrix M as elements. n represents a row arrangement of n matrices, and m represents a column arrangement of m M matrices. N is the number of spatial target points, p li and p ri , i = 1, 2, ..., N represents the image point of the i-th spatial target point imaged by the left camera and the right camera, P l,i , i=1,2,…,N represents the position of the i-th spatial target point in the left camera coordinate system, U i S i V i T is the result of the matrix singular value operation of mp3(4(i-1)+1:4i,:); The target motion state measurement equation based on the sequence image is: what e =quat(R e ) θ=π-2sin -1 (q e (4)) In the formula, R e is the attitude transformation matrix of the target inertial axis coordinate system between adjacent frame imaging moments obtained by image matching, θ is the spin angle of the target between adjacent image imaging moments, is the target spin angular velocity, e is the target spin axis direction between adjacent image imaging moments, ΔT is the time interval between adjacent frame image imaging moments; π is pi, quat is the quaternion inverse transformation operator symbol; Space targets are non-cooperative targets.

4. The device for autonomously estimating the shape and motion of a space target according to claim 3, characterized in that: The basic equation of binocular vision measurement is: Where P l is the position of the spatial target point in the left camera coordinate system, P r is the position of the spatial target point in the right camera coordinate system; T is the position of the left camera coordinate system in the right camera coordinate system, R is the transformation matrix from the left camera coordinate system to the right camera coordinate system, p l The image point of the spatial target point on the left camera, p r The image point of the spatial target point imaged by the left camera, E is the essential matrix describing the limit constraint of binocular vision, and F is the basic matrix describing the limit constraint of binocular vision; A l and A r are the known intrinsic parameter matrices of the left camera and the right camera respectively, It is a skew-symmetric matrix composed of T.

Images