CN-121999312-A - Construction method and system for pose estimation dataset of spacecraft

CN121999312ACN 121999312 ACN121999312 ACN 121999312ACN-121999312-A

Abstract

The invention relates to a method and a system for constructing a pose estimation dataset for a spacecraft, wherein the method comprises the steps of deriving a dynamic equation based on rigid body assumption, free rotation assumption and inertial coupling assumption, constructing a 7-dimensional nonlinear state equation based on the dynamic equation, solving the 7-dimensional nonlinear state equation by using a high-order numerical integration method according to an acquired inertial matrix, angular velocity and initial pose to obtain a pose sequence, traversing the pose sequence, rendering and generating a corresponding pose estimation dataset image, traversing the pose sequence, marking key points of the pose to be marked, creating an array and storing coordinates of all marked key points in the pose, and carrying out noise adding processing and fuzzy processing on the pose estimation dataset image and storing the processed image according to preset triggering noise adding probability and fuzzy processing triggering probability. Compared with the prior art, the method has the advantage of good diversity of the generated data set.

Inventors

XU LIANG
CHEN XIUYANG
XUE YUAN
LIU ZIHAO
WANG XIAOFAN

Assignees

上海大学

Dates

Publication Date: 20260508
Application Date: 20251217

Claims (10)

1. A method of constructing a pose estimation dataset for a spacecraft, comprising: The method comprises the steps of obtaining a dynamic equation based on a rigid body hypothesis, a free rotation hypothesis and an inertial coupling hypothesis, constructing a 7-dimensional nonlinear state equation based on the dynamic equation, obtaining an inertial matrix, an angular velocity and an initial pose of a spacecraft, solving the 7-dimensional nonlinear state equation by using a high-order numerical integration method according to the inertial matrix, the angular velocity and the initial pose, and obtaining a pose sequence; Traversing the pose sequence, rendering and generating a pose estimation dataset image corresponding to each pose, traversing the pose sequence, marking key points of the poses needing marking, creating an array and storing coordinates of all marked key points in the poses; and carrying out noise adding processing and blurring processing on the pose estimation dataset image according to the preset probability of triggering noise adding processing and the preset probability of triggering blurring processing, and storing the processed image.
2. The method for constructing a pose estimation dataset for a spacecraft of claim 1, wherein said rigid body assumption specifically comprises treating the spacecraft as an absolute rigid body with a fixed mass distribution, with its centroid fixed to the origin of the body coordinate system; The free rotation assumption specifically includes considering only the spacecraft motion phases without external control or disturbance, when the torque is combined ; The inertial coupling assumption specifically comprises the steps of introducing a complete inertial tensor and considering a main moment of inertia and an inertia product so as to reflect the asymmetry of the actual structure of the spacecraft.
3. The method for constructing a pose estimation dataset for a spacecraft of claim 1, wherein the 7-dimensional nonlinear state equation specifically comprises: Wherein, the To be the rate of change of system state over time, As a state vector for the system, Representing a transpose of the quaternion, Is a transpose of the angular velocity of the light, Is the inverse of the inertial matrix; is the derivative of the quaternion, Representing a quaternion number, To describe a matrix of differential relations between quaternions and angular velocities, As a component of the angular velocity in the x-axis, As a component of the angular velocity in the y-axis, As a component of the angular velocity in the z-axis, Representing the derivative of the angular velocity.
4. The method for constructing a pose estimation dataset for a spacecraft of claim 1, wherein the derivation process of the dynamics equation specifically comprises: defining angular momentum: Wherein, the Representing an inertia matrix of the device, In order to be able to achieve an angular velocity, Is the second-order dominant moment of inertia, Is the product of inertia; Using the transport theorem, the inertial derivative of angular momentum is obtained: Wherein, the Representing the derivative of the angular momentum, Representing the resultant torque; Representing the derivative of angular momentum in an inertial coordinate system; based on the free rotation assumption, it is further derived that: Wherein, the Representing the derivative of angular momentum in the body coordinate system; Will be Substituting Euler equation to obtain the dynamics equation of spacecraft during free rotation: Wherein, the Representing the derivative of the angular velocity.
5. The method for constructing a pose estimation dataset for a spacecraft of claim 1, wherein the process of rendering to generate a pose estimation dataset image for each pose specifically comprises: firstly, reading a spacecraft simulation model, and carrying out initialization setting; The pose sequence is traversed, wherein the pose is firstly applied to a spacecraft when the pose is traversed each time, and then one number is randomly selected in sequence in each environment parameter range to be used as the environment parameter and applied; Rendering the constructed simulation environment into an image, storing the image, and then traversing the next pose.
6. The method for constructing a pose estimation dataset for a spacecraft of claim 5, wherein the process of rendering to generate a pose estimation dataset image for each pose specifically comprises: The plurality of environmental parameter ranges are specifically an irradiation angle range of environmental illumination, an environmental illumination intensity range, a brightness range of an environmental background and a rotation angle range of the environmental background; and sequentially and randomly selecting one number in the irradiation angle range of the ambient light, the ambient light intensity range, the brightness range of the ambient background and the rotation angle range of the ambient background, and respectively applying the number to the ambient light irradiation angle, the ambient light intensity, the brightness of the ambient background and the rotation of the ambient background.
7. The method for constructing a pose estimation dataset for a spacecraft according to claim 1, wherein the process of labeling the key points of the pose to be labeled specifically comprises: firstly, reading a spacecraft simulation model, then reading a pose sequence of the spacecraft and setting an array of the spacecraft to be marked so as to store key point index numbers; Firstly, applying the pose to a spacecraft, then creating an array for storing the coordinates of all set key points on an image coordinate system, traversing the array for storing key point index numbers, finding the coordinates of the key points corresponding to the key point index numbers when traversing to one key point index number, and storing the coordinates into the array for storing the coordinates of all set key points on the image coordinate system; After the traversal is completed, the array for storing the coordinates of all the set key points on the image coordinate system is stored into a file and the traversal of the next pose is entered.
8. The method for constructing a pose estimation dataset for a spacecraft of claim 7, wherein the process of labeling key points of the pose to be labeled further comprises: And traversing the array for storing the index numbers of the key points, finding the key points corresponding to the index numbers of the key points when traversing to one index number of the key points each time, acquiring the coordinates of the key points on an image coordinate system, and storing the coordinates into the array for storing the coordinates of all the set key points on the image coordinate system.
9. The method for constructing a pose estimation dataset for a spacecraft according to claim 1, wherein the process of performing noise adding and blurring processing on the pose estimation dataset image specifically comprises: reading the pose estimation data set image, and setting the probability of triggering noise adding processing and the probability of triggering fuzzy processing, wherein the value is in the interval of [0, 1 ]; Randomly taking a value in the interval of [0,1], and if the value is smaller than or equal to the probability value of triggering the noise adding process, carrying out image blurring process; And randomly taking a value again in the interval of [0,1], if the value is smaller than or equal to the probability value triggering the blurring process, carrying out image noise adding process, and finally storing the processed image.
10. A construction system for a pose estimation dataset for a spacecraft, characterized by comprising a memory storing a computer program and a processor invoking the computer program to perform the steps of the method of any of claims 1 to 9.

Description

Construction method and system for pose estimation dataset of spacecraft Technical Field The invention relates to the field of spacecraft dynamics and control research, in particular to a method and a system for constructing a pose estimation dataset for a spacecraft. Background For supporting autonomous space tasks such as on-orbit service, rendezvous and docking of non-cooperative spacecraft, a spacecraft pose estimation dataset is used as a core training resource of a deep learning algorithm, the prior art gradually develops around an initial target of 'reducing an algorithm development threshold and covering a main stream satellite model', and a key support is provided for early pose estimation research. The method takes a Tango satellite dataset proposed by Shalma et al in 2018 as a starting point, realizes standardized output of a satellite model and a pose tag for the first time, enables an algorithm developer to avoid constructing data from zero, greatly reduces technical entry cost, further adopts an OpenGL rendering technology to generate continuous pose images of the Tango satellite by a subsequent derived SPEED dataset and an extended version SPEED+, becomes reference data of a SPEC series international challenge race, directly promotes the landing of special estimation algorithms such as SPNv, UDA and the like, tries to expand application scenes such as SPADES, focuses Proba-2 satellites and adapts novel sensing equipment such as event cameras in recent datasets such as SPARS, SPARK and the like, brings in AcrimSat, aquarius and the like NASA satellite models, introduces real space background, initially shows advantages in the aspects of expanding model coverage and adapting multiple sensing requirements, and forms a technical path from single model to multimode and adapting from universal to scenerization. However, the prior art gradually exposes a plurality of limitations in practical application, and is difficult to meet the specific requirements of the autonomous aerospace task in China. Firstly, the existing data sets are all built around foreign satellites such as Tango and Proba-2, and the number four eastern red spacecraft which does not relate to the third-generation geostationary orbit of China is completely avoided, so that the algorithm such as SPNv and UDA cannot be applied to related tasks of the number four eastern red spacecraft because the algorithm is trained on the SPEED+ data set only comprising the Tango spacecraft. The eastern red No. four spacecraft is used as a main stream platform of a high-orbit commercial communication satellite, and the pose estimation task of a non-cooperative target (such as a disabled middle satellite No. two) derived from the eastern red No. four spacecraft is lack of special data support, so that the existing algorithm cannot adapt to the related task of a Chinese satellite infrastructure, and a technical blank is formed. Secondly, although data sets such as speed+, SPARK and the like try to simulate the movement of the spacecraft, a sequence is generated by adopting simple gesture transformation, rigid body dynamics modeling of physical consistency (such as real characteristics of inertial coupling, torque balance under free rotation and the like are not considered) is not integrated, complex rotation behaviors of the non-cooperative spacecraft are difficult to capture, the data authenticity is insufficient, and an algorithm is easy to generate adaptation deviation in an actual task. Moreover, in the prior art, data enhancement in a single dimension is concentrated, for example SPADES only focuses on data simulation of an event camera, the SPARK only improves background authenticity, and dual enhancement of a 3D domain (such as random illumination direction and background rotation) and an image domain (such as Gaussian blur and pretzel noise) is not realized at the same time, so that diversity of a data set is limited, and generalization capability of an algorithm is difficult to improve. The invention discloses an integrated modeling method of a flexible spacecraft, which is disclosed by the publication number CN107992660B, and comprises a first step of calculating the dual momentum of the flexible accessory relative to the mass center of the flexible spacecraft under the dynamic condition by utilizing a finite element method and an integration method under the flexible spacecraft body coordinate system, a second step of calculating the dual momentum of the central rigid body relative to the mass center of the flexible spacecraft under the dynamic condition by utilizing the finite element method and the integration method under the flexible spacecraft body coordinate system, a third step of adding the dual momentum of the flexible accessory relative to the mass center of the flexible spacecraft in the first step and the dual momentum of the central rigid body relative to the mass center of the flexible spacecraft in the second step, and calcul