An absolute scale 3D modeling method for non-cooperative rotating targets in space

Abstract

The invention discloses an absolute scale three-dimensional modeling method for a space non-cooperative rotating target, which comprises the following steps: calculating a rotation matrix from a world coordinate system of the non-cooperative target to a target coordinate system, a position vector from the world coordinate system of the non-cooperative target to a target coordinate system, calculating an average value of a cross product matrix of a rotation angular velocity of the non-cooperative target, measuring the acceleration and angular velocity of a spacecraft, calculating a virtual rotation matrix and a virtual translation of a camera of the non-cooperative target during a maneuver of the spacecraft, calculating a real rotation matrix and a real camera translation of the spacecraft itself during the maneuver, calculating an average value of a cross product matrix of a real angular velocity of the spacecraft itself during the maneuver, and calculating a scale factor; obtaining a three-dimensional point cloud model of the non-cooperative target; the invention does not need to fly around the non-cooperative target, and can complete real-scale point cloud modeling of the non-cooperative target by using only sensors of the spacecraft platform itself.

Description

Technical Field

[0001] The present invention belongs to the field of point cloud modeling of non-cooperative targets, and in particular relates to an absolute scale three-dimensional modeling method of a spatial non-cooperative rotating target. Background Art

[0002] As the space environment becomes increasingly complex and on-orbit services become more mature, on-orbit operations on non-cooperative targets have become the core of a large number of space missions and have received widespread attention. In most on-orbit services, operating spacecraft to safely and stably approach mission targets is a prerequisite for the success of the mission. However, the approach and approach of unknown non-cooperative rotating targets is a recognized technical problem in on-orbit services. The main reasons include:

[0003] 1. A communication link cannot be established between the non-cooperative target and the operating spacecraft. The operating spacecraft needs to rely entirely on its own sensors to complete the approach mission and cannot use the target's cooperative identification to complete the approach;

[0004] 2. Critical prior information such as the structure and size of non-cooperative targets is often not available in advance;

[0005] 3. Non-cooperative targets usually lose stability and are in a rotating or tumbling state, which makes it difficult to identify and estimate the state of the target.

[0006] Although a lot of research has been conducted on the approach and approach process of non-cooperative rotating targets at home and abroad, and some progress has been made, most of the research is based on large spacecraft platforms flying around the target completely to obtain the complete state of the target at multiple angles. Generally, large-scale ranging equipment such as laser radar is used to obtain real distance information, which facilitates the approach of non-cooperative targets. However, this will inevitably bring a great burden to the navigation, guidance and control of the platform, and increase the mission cost. Especially for tiny spacecraft, the process of approaching and approaching unknown rotating targets is obviously more challenging. On the one hand, the cost and volume constraints of tiny spacecraft limit the installation of high-performance sensors. The main ranging means such as laser ranging radar and millimeter wave radar are usually difficult to be directly installed on tiny spacecraft; on the other hand, the limited fuel and maneuverability of tiny spacecraft make it difficult for them to maneuver significantly, and it is impossible to fly around and model the target in a large range. Summary of the invention

[0007] The purpose of the present invention is to provide an absolute scale three-dimensional modeling method for a spatial non-cooperative rotating target, so as to solve the problem that point cloud modeling of a non-cooperative target requires large-scale flying and is costly.

[0008] The present invention adopts the following technical solution: a method for absolute scale three-dimensional modeling of a space non-cooperative rotating target, which consists of the following steps:

[0009] Step S1:

[0010] Under the assumption that the non-cooperative target is stationary and the spacecraft rotates periodically, the rotation matrix from the non-cooperative target world coordinate system to the target coordinate system and the position vector from the non-cooperative target world coordinate system to the target coordinate system are calculated by combining the SFM algorithm.

[0011] According to the rotation matrix from the non-cooperative target world coordinate system to the target coordinate system, the average value of the non-cooperative target's rotation angular velocity cross product matrix is ​​calculated.

[0012] Step S2:

[0013] Drive the spacecraft to perform limited angle and position maneuvers within the orbital plane, and use the spacecraft's inertial measurement unit to measure the spacecraft's acceleration and angular velocity.

[0014] The SFM algorithm is used to calculate the camera virtual rotation matrix and camera virtual translation of non-cooperative targets during the maneuvering of the spacecraft.

[0015] Calculate the spacecraft's true rotation matrix and true camera translation during maneuvers.

[0016] Calculate the average value of the cross product matrix of the spacecraft's true angular velocity during the maneuver.

[0017] Step S3:

[0018] The scale factor is calculated based on the acceleration and angular velocity of the spacecraft measured by the inertial measurement unit;

[0019] Step S4:

[0020] Using the scale factor and based on the point cloud model of the non-cooperative target at a relative scale, the coordinates of each point in the point cloud model are restored from the coordinates under the visual normalized scale to the real coordinates to obtain the three-dimensional point cloud model of the non-cooperative target.

[0021] Furthermore, the rotation matrix from the non-cooperative target world coordinate system to the target coordinate system in step S1 is The position vector from the non-cooperative target world coordinate system to the target coordinate system W The calculation formula of t0 is:

[0022]

[0023]

[0024] In the formula, is the rotation matrix from the non-cooperative target world coordinate system to the target coordinate system, W t0 is the position vector from the non-cooperative target world coordinate system to the target coordinate system, I3 is the 3×3 unit matrix, It is the virtual translation vector of the camera calculated by the SFM algorithm; It is the virtual rotation motion of the camera calculated by the SFM algorithm.

[0025] Furthermore, the average value of the cross product matrix of the rotation angular velocity of the non-cooperative target in step S1 is The calculation formula is:

[0026]

[0027] Furthermore, in step S3, the scale factor is calculated s The calculation formula is:

[0028]

[0029] in: W t c O is the origin of the world coordinate system W w To the origin O of the camera coordinate system C c The displacement vector of

[0030] Where: W x i is the absolute displacement of the body measured by the inertial measurement unit; represents the rotation matrix from the world coordinate system W to the camera coordinate system C, is the rotation matrix from the inertial measurement unit sensor coordinate system to the visual coordinate system, S a i It is the acceleration value directly output by the inertial measurement unit; S b i is the total noise of the angular velocity meter, The rotation matrix that transforms the gravitational acceleration into the selected world coordinate system; W g is the gravitational acceleration at the orbit, t f The final time point of the movement process.

[0031] The beneficial effect of the present invention is that the present invention only uses a monocular camera and an inertial measurement unit to complete the real-scale point cloud modeling of a non-cooperative rotating target, which can not only take advantage of the rich information of the point cloud model, but also avoid the time and cost brought by flying around the non-cooperative target, and can greatly assist the positioning and abutment of the non-cooperative target during the approach process. There is no need to fly around the non-cooperative target, and the real-scale point cloud modeling of the non-cooperative target can be completed using only the sensors of the spacecraft platform itself. DETAILED DESCRIPTION

[0032] The present invention is described in detail below in conjunction with specific implementation modes.

[0033] It should be noted that the structures, proportions, sizes, etc. described in this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the conditions under which the present invention can be implemented. Therefore, they have no substantive technical significance. Any structural modification, change in proportional relationship or adjustment of size should still fall within the scope of the technical content disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0034] The present invention discloses an absolute scale three-dimensional modeling method for a non-cooperative rotating target in space. The premise of the present invention is that: a spacecraft and a non-cooperative target are in the same orbit, the non-cooperative target presents a periodic rotation state of the inertial principal axis, and the precession, nutation and perturbation on the orbit are ignored; the entire morphology of the target falls within the field of view of a camera throughout the whole process, and feature points can be clearly acquired and identified by the camera without distortion; the same camera is always used throughout the entire approximation and modeling process, and all images are at the same normalized scale; the influence of other satellites, the earth or starlight that may fall into the field of view is not considered in the modeling process; and the spacecraft is equipped with only an ordinary monocular camera and an inertial measurement unit (IMU).

[0035] The spacecraft is first in the initial state and remains relatively still with the centroid of the non-cooperative target. The rotation state estimation of the non-cooperative target is completed in a purely visual way by utilizing the periodic rotation characteristics of the non-cooperative target. Since the classic incremental Structure from Motion (SFM) method only involves discrete images and does not require the camera position, trajectory, etc. to be solved, the SFM method can correctly obtain the relative three-dimensional point cloud model of the target at the visual normalized scale. Since the information of the inertial measurement unit has not been used in the parking stage, the drift problem of the inertial measurement unit will not have an impact, allowing this stage to last for a longer time.

[0036] The present invention uses the following coordinate system:

[0037] Camera coordinate system C(O c ,x c ,y c ,z c ):

[0038] Origin O c Located at the camera optical center, y c The axis points down the camera shape, z c Axis is along the optical axis, x c The axis is determined by the right-hand principle, and the camera coordinate system C is a moving system bound to the camera body.

[0039] World coordinate system W(O w ,x w ,y w ,zw ):

[0040] The camera coordinate system C0 at the initial moment (i.e., before the spacecraft performs maneuvers) is the world coordinate system W(O w ,x w ,y w ,z w ). It is easy to know that the transformation of the camera coordinate system C relative to the world coordinate system W is the movement of the spacecraft relative to the initial position, which can be obtained by the rotation matrix With displacement vector W t c Fully described.

[0041] Target coordinate system T(O t ,x t ,y t ,z t ):

[0042] The origin is located at the centroid of the target, and x is defined in the initial state t ,y t ,z t The x-axis w ,y w ,z w The target coordinate system is fixed to the target and is a rotating coordinate system, and the coordinates rotate with time.

[0043] The sensor coordinate system S(O) of the inertial measurement unit s ,x s ,y s ,z s ):

[0044] The origin and coordinate axes are determined by the inertial measurement unit itself and have been calibrated in advance.

[0045] For a moving non-cooperative target, it is impossible to distinguish the movement of itself from that of the target by vision alone. Traditional visual algorithms usually assume that a stationary target is used as a reference, and when applied to the present invention, calculation errors will occur. Introducing IMU information in advance to judge movement will lead to a lack of comparison. At present, the field of ground robots usually uses a large number of static feature points on the ground, buildings, etc. to solve this problem. Obviously, this processing cannot be extended to space. To this end, the present invention will obtain a visual solution result that can be fused with IMU information through coupling analysis, thereby completing absolute scale reconstruction.

[0046] In the present invention, the non-cooperative target rotates continuously, which is different from the classic SFM usage scenario. However, it can be proved that the two situations of the target being stationary and the camera moving and the target moving and the camera being stationary are equivalent. In addition, if the problem of fusion with the inertial measurement unit is not considered, the camera position estimation error does not affect any link in the modeling, and in the usual ground application scenario, the background is not empty, and there are always a large number of available static feature points on the ground for the camera to use as a position reference. Therefore, the conventional visual SFM field usually does not conduct further research on the use of rotating targets to restore the true position of the camera. Because the above problems cannot be ignored in the application environment of the present invention, it is necessary to provide an analysis of the camera position error solution and an equivalence proof.

[0047] If the target is stationary and the camera is moving:

[0048] It is easy to know at this time For feature point P,

[0049]

[0050] In the formula, c p is the coordinate of the feature point of the non-cooperative target in the camera coordinate system, W t0 is the position vector from the non-cooperative target world coordinate system to the target coordinate system, is the rotation matrix from the non-cooperative target world coordinate system to the target coordinate system, T p is the coordinate of the feature point of the non-cooperative target in the target coordinate system.

[0051] Therefore, the camera imaging model of spacecraft-rotating target can be obtained as follows:

[0052]

[0053] In the formula, W t c Represents the world coordinate system W with the origin O w To the origin O of the camera coordinate system C c The displacement vector, is the rotation matrix from the world coordinate system to the target coordinate system, W p is the coordinate of the feature point of the non-cooperative target in the world coordinate system, W t c Represents the world coordinate system W with the origin O w To the origin O of the camera coordinate system C c The conventional SFM method is based on the above situation to model and estimate the camera position.

[0054] If the non-cooperative target rotates periodically, the camera and the target centroid remain relatively stationary:

[0055] At this time, we know that W t c = 03, then the camera imaging model of the spacecraft-rotating target is

[0056]

[0057] The SFM method cannot distinguish between the above two situations. When the camera sees that the feature point is moving in the field of view, the algorithm still thinks that the target is stationary and the movement is caused by the camera itself, resulting in an incorrect estimation of the camera motion. This is an important condition for modeling and true trajectory recovery, so this misestimation needs to be modeled in detail.

[0058] The algorithm assumes that the target does not rotate and can be expressed as

[0059] T p= W p- W t0

[0060] In summary, we can get

[0061]

[0062] Where: The virtual rotation matrix of the camera calculated by the SFM algorithm; is the virtual translation vector of the camera calculated by the SFM algorithm; I3 is the 3×3 unit matrix.

[0063] At this time, the position movement calculated by the algorithm is virtual. Comparing the two sides of the above equation, we can see that the condition for the equation to hold is:

[0064]

[0065]

[0066] From the above two equations, we can see that when the camera is stationary and the target is rotating, the SFM algorithm will mistakenly believe that it is flying around the target. The angle of the fly-around is opposite to the target rotation angle, but the value is the same. The virtual moving distance is the difference between the position vectors of the initial state and the final state after the target is rotated. When the target rotates completely, it is equivalent to the camera flying around the target with a radius of Wt0. Although this feature makes it difficult to solve visual motion, it provides conditions for modeling spacecraft under limited maneuvers, making it possible to establish a full point cloud model of the target without a complete fly-around. So far, the above has mathematically strictly proved the equivalence of the two processes of target rotation-self-stationary and self-flying-target stationary in visual imaging.

[0067] After proving the equivalence of motion, the conventional SFM method can be used directly to model non-cooperative targets. Although the camera does not move at this stage, the algorithm can output the virtual pose change value of the camera at this time. and Like all algorithms based on monocular vision, the 3D point cloud model is based on the normalized scale of the camera, which loses the real scale information. The real distance between the camera and the target cannot be obtained either.

[0068] Combined with the SFM algorithm, the output camera pose estimation is given under the assumption that the non-cooperative target is stationary. These two data can be used to obtain the rotation matrix of the non-cooperative target. The position vector from the non-cooperative target world coordinate system to the target coordinate system W t0, that is

[0069]

[0070]

[0071] In the formula, is the rotation matrix from the non-cooperative target world coordinate system to the target coordinate system, W t0 is the position vector from the non-cooperative target world coordinate system to the target coordinate system, I3 is the 3×3 unit matrix, It is the virtual translation vector of the camera calculated by the SFM algorithm; It is the virtual rotation motion of the camera calculated by the SFM algorithm.

[0072] Getting the rotation matrix Then, combined with the transformation relationship between the derived rotation matrix and the rotation angular velocity, the rotation angular velocity ω of the non-cooperative target can be further obtained: T for:

[0073]

[0074] Due to the differential operation, the calculated angular velocity results will be affected by noise and numerical errors in the calculation, resulting in more obvious errors, especially for visual measurement results with lower effective signal frequencies. Therefore, the angular velocity needs to be averaged.

[0075]

[0076] Where: The target rotation angular velocity cross product matrix is ​​averaged, and the smoothed angular velocity signal is further smoothed using a Rauch-Tung-Striebel (RTS) filter.

[0077] After the parking decoupling phase is completed, the point cloud model of the non-cooperative target at the relative scale, the distance from the camera and the angular velocity of rotation are obtained. After that, the spacecraft begins to perform limited angle and position maneuvers in the orbital plane. The present invention does not have too many requirements for maneuvers, and there is no need to fly around in a large range. It only needs to generate displacement so that the inertial measurement unit can produce effective output. Since this process is not modeled, the trajectory of the maneuver is different from the modeling process from the algorithm perspective. Even if the angular velocity is the same as the target and the radius is equal to the initial distance, which makes the modeling lose its observability, the special condition can still be combined with the target rotation information obtained in the first stage to correctly obtain its own motion trajectory.

[0078] In order to avoid collision with the target and make the image change more obvious and obtain better observation effect, the spacecraft maneuvers mainly along the y w Direction. The angle and position changes ensure that the target always falls within the camera's field of view. Considering the inherent drift problem of inertial components, the time of the maneuvering process should be minimized. Also, since this process is not modeled, there is no need to obtain a complete image of the target's rotation, and only the reference is needed to solve its own position, so the demand for short maneuvering time can be met. Record the video information recorded by the camera and the acceleration and angular velocity information output by the inertial measurement unit at this stage.

[0079] In order to fuse the information from the inertial measurement unit, it is necessary to solve the real movement trajectory of the camera under visual measurement. Then, the camera position estimation part of the SFM algorithm can be used again to obtain the virtual camera pose change that the SFM algorithm mistakenly believes during the maneuver. and The SFM algorithm mistakenly believes that the target is stationary and all visual changes are caused by camera motion. and Remove the part caused by the target from the image, and separate the real rotation matrix caused by the spacecraft itself and the real camera translation. Then we can get

[0080]

[0081] Similarly, in order to make both sides of the above equation equal, we can get

[0082]

[0083]

[0084] In the formula, Calculated from the angular velocity combined with time, Represents the rotation matrix from the world coordinate system W to the camera coordinate system C.

[0085] So far, the real motion of the spacecraft itself has been separated from the motion coupled with visual measurements.

[0086] When the spacecraft platform performs active maneuvers, the inertial measurement unit of the camera's close-range strapdown considers it to have the same motion as the camera. Since the inertial measurement unit can provide real new scale information, the visual-inertial fusion is used to restore the scale information missing in the visual measurement.

[0087] When using accelerometers and visual measurements for fusion, time synchronization of two completely separate data streams is a practical problem. Since the angular velocity given by vision is an unnormalized true value, the angular velocity measured by the camera and accelerometer can be used for time coordination first. Let the visual measurement angular velocity be W ω c (t), and the corresponding relationship between the rotation matrix and the angular velocity is also used to obtain

[0088]

[0089] Similarly, in order to avoid the error introduced by the differential operation, the visual angular velocity is also averaged

[0090]

[0091] Where: is the average value of the cross product matrix of its own real angular velocity under visual measurement. And this result also needs to be smoothed using the RTS filter.

[0092] Then time coordination can be summarized as the following optimization problem:

[0093]

[0094] Where: td is the signal delay between the inertial measurement unit and the visual measurement; S ω i (t) is the angular velocity measured by the inertial measurement unit; is the rotation matrix from the inertial measurement unit sensor coordinate system to the visual coordinate system; S b i is the total noise of the angular velocity meter.

[0095] It should be noted that visual measurement and inertial measurement have different errors and noises. In order to improve the optimization effect, the two angular velocity signals need to be smoothed using RTS filters respectively. This processing is not performed in real time during the maneuver, but is processed uniformly after the camera maneuvers are completed. Therefore, it will not increase the computing burden of the spacecraft.

[0096] After the time coordination is completed, the scale factor can be recovered by comparing the visual measurement with the accelerometer measurement. The actual comparison parameter can be acceleration or displacement, and their respective advantages and disadvantages are:

[0097] Acceleration comparison does not have the problem of cumulative error and is not constrained by maneuvering time. However, acceleration comparison requires two differentiations of the visual signal to convert the moving distance into angular velocity. This process is extremely sensitive to errors, and a small disturbance will cause a huge difference in the signal after two differentiations.

[0098] When comparing the displacement, the inertial measurement unit is integrated twice, which results in drift error. However, considering that the maneuvering time of the spacecraft in this problem is limited (less than 20 seconds), the drift effect is still within an acceptable range. In addition, scholars have conducted a lot of research on the drift problem of the inertial measurement unit and have achieved certain research results, which can be used to correct IMU data and minimize the drift effect. In addition, the integral operation is less sensitive to noise and is easier to implement than the visual differential operation.

[0099] Therefore, the present invention chooses to sacrifice a certain degree of accuracy and greatly reduces the computational complexity by comparing the displacement. At the same time, since the maneuvering time of the spacecraft is short, the drift problem will not have a high accuracy impact on the algorithm. After selecting the moving distance as the comparison quantity, the scale factor can be expressed as

[0100]

[0101] in:

[0102]

[0103] Where: W x i is the absolute displacement of the body measured by the inertial measurement unit; represents the rotation matrix from the world coordinate system W to the camera coordinate system C, is the rotation matrix from the inertial measurement unit sensor coordinate system to the visual coordinate system, Sa i It is the acceleration value directly output by the inertial measurement unit; S b i is the total noise of the angular velocity meter, The rotation matrix that transforms the gravitational acceleration into the selected world coordinate system; W g is the gravitational acceleration in orbit, transmitted from the parent star to the spacecraft; t f is the final time point of the movement process, W t c O is the origin of the world coordinate system W w To the origin O of the camera coordinate system C c The displacement vector.

[0104] By solving the above two equations, the scale factor s of the visual normalized scale can be restored to the actual scale.

[0105] Using the obtained scale factor, a new point cloud model can be generated based on the established relative scale point cloud model, and the coordinates of each point in the point cloud can be restored from the coordinates under the visual normalized scale to the real coordinates.

[0106] The point cloud model is essentially a structure containing several points. The spatial position, color and intensity information of each point are recorded, which can be expressed as P(x, y, z, R, G, B, I). The boundary value of the model and the number of points included in the overall model are recorded, which can be expressed as M(x lim ,y lim ,z lim ,N). Then for a single point i=1:N, the scale correction can be expressed as

[0107] W x i =s N x i , W y i =s N y i , W z i =s N z i

[0108] W R i = N R i , W G i = N G, W B i = N B i , W I i = N I i

[0109] In the formula, W x i , W y i , W z i is the three-dimensional coordinate of the point on the point cloud at the real scale; N x i , N y i , N z i are the three-axis coordinates of the corresponding points on the point cloud at the normalized scale; W Ri , W G i , W B i The color information of each point in the real-scale point cloud; N R i , N G i , N B i is the color information of each point in the original point cloud; W I i is the intensity of each point in the true-scale point cloud; N I i is the intensity of each point in the original point cloud.

[0110] At the same time, the boundary size of the entire point cloud model is expanded accordingly W x lim =s N x lim , W y lim =s N y lim , W z lim =s N z lim ,

[0111] Where: W x lim , W y lim , W z lim are the model boundary coordinates of the true-scale point cloud; N x lim , N y lim , N z lim is the normalized point cloud model boundary coordinate.

[0112] After obtaining the necessary information of the new point cloud, the discrete point cloud sequence is repackaged into a structure to complete the generation of a new point cloud at the real scale, while keeping the color and intensity information of each point unchanged. At this point, a three-dimensional point cloud model of the periodically rotating target at the absolute scale is obtained.

[0113] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for absolute scale three-dimensional modeling of a non-cooperative rotating target in space, characterized in that: It consists of the following steps: Step S1: Under the assumption that the non-cooperative target is stationary and the spacecraft rotates periodically, the rotation matrix from the non-cooperative target world coordinate system to the target coordinate system and the position vector from the non-cooperative target world coordinate system to the target coordinate system are calculated by combining the SFM algorithm. According to the rotation matrix from the non-cooperative target world coordinate system to the target coordinate system, the average value of the non-cooperative target's rotation angular velocity cross product matrix is ​​calculated. Step S2: Drive the spacecraft to perform limited angle and position maneuvers within the orbital plane, and use the spacecraft's inertial measurement unit to measure the spacecraft's acceleration and angular velocity. The SFM algorithm is used to calculate the camera virtual rotation matrix and camera virtual translation of non-cooperative targets during the maneuvering of the spacecraft. Calculate the spacecraft's true rotation matrix and true camera translation during maneuvers. Calculate the average value of the cross product matrix of the spacecraft's true angular velocity during the maneuver. Step S3: The scale factor is calculated based on the acceleration and angular velocity of the spacecraft measured by the inertial measurement unit; Step S4: Using the scale factor and based on the point cloud model of the non-cooperative target at a relative scale, the coordinates of each point in the point cloud model are restored from the coordinates at the visual normalized scale to the real coordinates to obtain the three-dimensional point cloud model of the non-cooperative target.

2. The absolute scale three-dimensional modeling method of a spatial non-cooperative rotating target according to claim 1, characterized in that: The rotation matrix from the non-cooperative target world coordinate system to the target coordinate system in step S1 The position vector from the non-cooperative target world coordinate system to the target coordinate system W The calculation formula of t0 is: In the formula, is the rotation matrix from the non-cooperative target world coordinate system to the target coordinate system, W t0 is the position vector from the non-cooperative target world coordinate system to the target coordinate system, I3 is the 3×3 unit matrix, It is the virtual translation vector of the camera calculated by the SFM algorithm; It is the virtual rotation motion of the camera calculated by the SFM algorithm.

3. The absolute scale three-dimensional modeling method of a spatial non-cooperative rotating target according to claim 2, characterized in that: The average value of the cross product matrix of the rotation angular velocity of the non-cooperative target in step S1 The calculation formula is:

4. The absolute scale three-dimensional modeling method of a spatial non-cooperative rotating target according to claim 3, characterized in that: The calculation formula for calculating the scale factor s in step S3 is: in: W t c O is the origin of the world coordinate system W w To the origin O of the camera coordinate system C c The displacement vector of Where: W x i is the absolute displacement of the body measured by the inertial measurement unit; represents the rotation matrix from the world coordinate system W to the camera coordinate system C, is the rotation matrix from the inertial measurement unit sensor coordinate system to the visual coordinate system, S a i It is the acceleration value directly output by the inertial measurement unit; S b i is the total noise of the angular velocity meter, The rotation matrix that transforms the gravitational acceleration into the selected world coordinate system; W g is the gravitational acceleration at the orbit, t f The final time point of the movement process.

Images