A method and device for generating a visual image sequence of a large deformation structure

Abstract

A method and device for generating a visual image sequence of a large deformation structure belongs to the field of computer simulation, and in particular to the field of simulation of rigid-flexible coupling technology of multi-body systems. It solves the problem that when the convolutional neural network technology is applied to the aerospace field, the original image data in the aerospace field is small and cannot meet the requirements of the convolutional neural network training data volume, and the problem that the accuracy of the visual algorithm itself is difficult to evaluate. The method and device for generating a visual image sequence of a large deformation structure are suitable for multi-body dynamics simulation of large deformation targets.

Description

Technical Field

[0001] The present invention relates to the field of computer simulation, and in particular to the field of simulation of rigid-flexible coupling technology of multi-body systems. Background Art

[0002] In the aerospace field, it is difficult to conduct direct on-site observations of large deformation targets such as satellites, so it is necessary to acquire images and then analyze and process them to study the dynamic problems of large deformation target scenes.

[0003] In recent years, convolutional neural network technology has developed rapidly and has gradually been applied to the analysis and processing of images. Convolutional neural network is an important part of computer vision intelligence. In the process of using convolutional neural network to analyze and process images, it is necessary to use large-scale labeled data, i.e. training set, to train the initial convolutional neural network in order to obtain a trained convolutional neural network that can process image problems.

[0004] In the training process of convolutional neural networks, the size of the training set data plays a crucial role. The larger the training set, the more conducive it is to fully train the parameters in the convolutional neural network. Only sufficient training can ensure the accuracy of the convolutional neural network. On the contrary, when the amount of training set data is small, model training may have the following problems:

[0005] 1. The model may not converge during training: A small amount of training data may not provide enough information for model learning, resulting in the model not being able to converge;

[0006] 2. Overfitting problem: The model incorporates completely irrelevant information in the training data into the learning content, which is easily disturbed by noise and leads to overfitting;

[0007] 3. It is easy to fall into local optimal value;

[0008] 4. It is difficult to measure model quality: Both training data and test data are relatively small, and the accuracy of a small amount of data may have a significant impact on the results.

[0009] When the convolutional neural network technology is applied to the aerospace field, the amount of on-orbit image data containing large deformation target scenes in the aerospace field is relatively scarce, resulting in a lack of large-scale labeled data for convolutional neural network training. In other words, the original image data in the aerospace field is relatively small and cannot meet the requirements of the convolutional neural network training data volume. To address this problem, technicians in this field have tried some solutions, such as using transfer learning methods or data enhancement methods:

[0010] Among them, the method of using transfer learning can use an already trained network to solve new problems, but transfer learning itself also has certain problems, namely, the domain adaptation problem; the domain adaptation problem is caused by the large gap that may exist between the target domain of the original network and the target domain that needs to be processed currently. How to adjust and optimize the source domain knowledge according to the domain differences is a challenge; especially in the aerospace industry, there are few pre-trained networks for specific target domains in this industry, and transfer learning is more difficult.

[0011] The use of data augmentation methods can indeed solve the problem of poor generalization ability caused by the small amount of network data to a certain extent; however, noise errors may be introduced in the data augmentation process, thereby greatly reducing network performance and may also damage the performance of unenhanced data.

[0012] To sum up, when applying convolutional neural network technology to the aerospace field, there is a problem that the original image data in the aerospace field is insufficient and cannot meet the requirements of the convolutional neural network training data volume. Technical personnel in this field have not yet provided a very good solution. This has become a technical problem that technical personnel in this field have always wanted to solve but have not solved.

[0013] In addition, for such large spatial deformation targets, there is also the problem of identifying and measuring their dynamic information. The existing identification and measurement methods are divided into visual algorithms with identification points and visual algorithms without identification points:

[0014] The visual algorithm with markers refers to a visual algorithm that introduces markers on a large deformation target in space, and then uses different camera groups with higher precision to measure the movement of the markers. This visual algorithm can well evaluate the accuracy of its own algorithm, but the introduction of markers will affect the dynamic information of the model itself, thereby affecting the measurement accuracy. Therefore, people prefer to use visual algorithms without markers.

[0015] The visual algorithm without identification points does not need to introduce identification points, so the accuracy of measurement will not be affected by the introduction of identification points. However, the accuracy of the algorithm itself of this visual algorithm is difficult to evaluate. Summary of the invention

[0016] The present invention solves the problem that when the convolutional neural network technology is applied to the aerospace field, the original image data in the aerospace field is insufficient and cannot meet the requirements of the convolutional neural network training data volume, and the problem that the accuracy of the visual algorithm itself is difficult to evaluate.

[0017] The present invention discloses a method for generating a visual image sequence of a large deformation structure, and its technical solution is as follows:

[0018] The visual image sequence generation method comprises the following steps:

[0019] S1. Drawing a three-dimensional model of a large deformation target; the three-dimensional model includes a rigid body part and a flexible body part;

[0020] S2, performing finite element division on the flexible body part;

[0021] S3, meshing the flexible body part according to the finite element partitioning result, and storing the meshed flexible body part as a .msh file;

[0022] S4. Use MBDyn to couple the rigid part and the flexible part to obtain a rigid-flexible coupling model of a large deformation target.

[0023] S5. Perform dynamic simulation on the rigid-flexible coupling model of the large deformation target to obtain a simulation result of the large deformation target;

[0024] S6, displaying the simulation results in a model visualization to obtain a simulation image;

[0025] S7: setting a camera viewing angle in the simulation image, rendering the simulation image after the camera viewing angle is set, and obtaining an image sequence under the camera viewing angle.

[0026] Further, a preferred embodiment is provided, wherein the method further comprises step S5.1;

[0027] S5.1. According to the simulation results, the spatial motion information of the nodes of the rigid-flexible coupling model of the large deformation target is obtained.

[0028] The present invention also proposes a visual image sequence generation device for a large deformation structure, and its technical scheme is as follows: the visual image sequence generation device includes a three-dimensional modeling module, a finite element division module, a mesh division module, a coupled modeling module, a simulation module, a model visualization module and an image sequence generation module;

[0029] The three-dimensional modeling module is used to draw a three-dimensional model of a large deformation target; the three-dimensional model includes a rigid body part and a flexible body part;

[0030] The finite element division module is used to perform finite element division on the flexible body part;

[0031] The meshing module is used to mesh the flexible body part according to the finite element meshing result, and store the meshed flexible body part as a .msh file;

[0032] The coupling modeling module is used to couple the rigid part and the flexible part using MBDyn to obtain a rigid-flexible coupling model of a large deformation target;

[0033] The simulation module is used to perform dynamic simulation on the rigid-flexible coupling model of the large deformation target to obtain the simulation result of the large deformation target;

[0034] The model visualization module is used to visualize the simulation results and obtain simulation images;

[0035] The image sequence generation module is used to set the camera viewing angle in the simulation image, and is also used to render the simulation image after the camera viewing angle is set to obtain an image sequence under the camera viewing angle.

[0036] Further, a preferred embodiment is provided, wherein the coupling modeling module includes a rigid body configuration unit, a flexible body configuration unit, and a rigid-flexible coupling model configuration;

[0037] The rigid body configuration unit is used to configure the rigid body part in a RIGIDBODY form;

[0038] The flexible body configuration unit is used to configure the flexible body part in the form of GMSHELE;

[0039] The rigid-flexible coupling model configuration is used to couple the rigid body part and the flexible body part in the form of connecting the flexible node to the rigid body.

[0040] Furthermore, a preferred embodiment is provided, wherein the simulation module is also used to obtain spatial motion information of nodes of the rigid-flexible coupling model of the large deformation target according to the simulation results.

[0041] Further, a preferred embodiment is provided, wherein the image sequence generation module includes a camera viewing angle setting unit, a model appearance setting unit, and a rendering unit;

[0042] The camera viewing angle setting unit is used to add a camera to the simulation image, and to define the position and camera parameters of the added camera to complete the camera viewing angle setting;

[0043] The model appearance setting unit is used to define the appearance parameters of the rigid-flexible coupling model in the simulation image;

[0044] The rendering unit is used to render the simulation image whose camera viewing angle is set and whose appearance parameters are defined, so as to obtain an image sequence under the camera viewing angle.

[0045] The present invention also proposes a visual image sequence generation device for a large deformation structure, and its technical solution is as follows: a visual image sequence generation device for a large deformation structure includes a processor and a memory, the memory is used to store executable instructions of the processor, and the processor is configured to execute the above-mentioned visual image sequence generation method for a large deformation structure by executing the executable instructions.

[0046] The present invention also proposes a computer storage medium, and its technical solution is as follows:

[0047] The storage medium stores a computer program, and when the computer program is run, the above-mentioned method for generating a visual image sequence of a large deformation structure is executed.

[0048] The present invention also proposes an accuracy evaluation method for a visual algorithm without a marking point, and its technical solution is as follows:

[0049] The method comprises the following steps:

[0050] ST1, select nodes as measurement points in the large deformation target;

[0051] ST2. Use a visual algorithm without identification points to calculate the motion information of the measurement points and obtain the calculation data to be evaluated;

[0052] ST3. Using the above-mentioned visual image sequence generation method for a large deformation structure, obtain the spatial motion information of the nodes corresponding to the measurement points of the rigid-flexible coupling model of the large deformation target, and use the spatial motion information of the nodes corresponding to the measurement points as accurate reference data;

[0053] ST4. Compare and analyze the calculated data to be evaluated with the precise reference data to obtain the accuracy evaluation result of the visual algorithm without identification points.

[0054] The present invention has the following beneficial effects:

[0055] 1. The method for generating a visual image sequence of a large deformation structure described in the present invention can simulate the actual shooting situation of a large deformation target by acquiring the image sequence under the camera perspective, and further can simulate the actual motion state of the large deformation target under different lighting conditions, different excitation disturbances and other scenarios.

[0056] 2. The method for generating a visual image sequence of a large deformation structure described in the present invention can replace the real shooting data of the large deformation target by obtaining the image sequence under the camera's perspective, and provide a large amount of labeled data for the training of the convolutional neural network, thereby solving the problem that the original image data in the aerospace field is insufficient and cannot meet the requirements of the convolutional neural network training data volume, that is, it solves the technical problem that technicians in this field have always wanted to solve but have not solved.

[0057] 3. The method for generating a visual image sequence of a large deformation structure described in the present invention can simulate the spatial motion of each node of a large deformation target in a real scene by acquiring the spatial motion information of the node; the spatial motion information of the node can be used as a basis for evaluating the precision (or accuracy) of various visual algorithms (such as visual algorithms without identification points).

[0058] The method and device for generating a visual image sequence of a large deformation structure described in the present invention are suitable for performing multi-body dynamics simulation on a large deformation target. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0060] Figure 1 A schematic diagram of a three-dimensional model of a satellite and a single-sided solar wing in one embodiment of the present invention;

[0061] Figure 2 This is a flow chart of a method for obtaining an image sequence of a space large flexible solar wing under a camera viewing angle by using a visual image sequence generating device of a large deformation structure in one embodiment of the present invention;

[0062] Figure 3 A schematic diagram of a MBDyn rigid body configuration in one embodiment of the present invention;

[0063] Figure 4 A schematic diagram of the configuration of the MBDyn flexible body in one embodiment of the present invention;

[0064] Figure 5 A schematic diagram of the configuration of the MBDyn rigid-flexible coupling model in one embodiment of the present invention;

[0065] Figure 6 A schematic diagram of a simulated image of a satellite and a single-sided solar wing in one embodiment of the present invention;

[0066] Figure 7 A schematic diagram of setting the camera viewing angle in Blender in one embodiment of the present invention;

[0067] Figure 8 A schematic diagram of an image sequence of a flexible solar wing in the left camera perspective in one embodiment of the present invention;

[0068] Fig. 9A schematic diagram of an image sequence of a flexible solar wing in the perspective of a right camera in one embodiment of the present invention;

[0069] Fig.10 is a schematic diagram of nodes corresponding to measurement points in one embodiment of the present invention;

[0070] Fig.11 It is a schematic diagram of spatial motion information of a node corresponding to a measurement point in one embodiment of the present invention. DETAILED DESCRIPTION

[0071] In order to make the technical solutions and advantages of the present invention more clearly described, the specific implementation methods of the present invention will be further described in detail and completely in conjunction with the accompanying drawings. The various implementation methods described below are only part of the preferred solutions of the present invention, rather than all implementation methods; the various implementation methods described below are intended to explain the present invention and cannot be understood as limitations on the present invention; the reasonable combination of technical features defined in the various implementation methods of the present invention, and all other implementation methods obtained by ordinary technicians in the field without creative work based on the implementation methods of the present invention, are within the scope of protection of the present invention.

[0072] Implementation Method 1: Combination Figure 1 -9 describes this embodiment. This embodiment provides a method for generating a visual image sequence of a large deformation structure. The specific implementation contents are as follows:

[0073] The visual image sequence generation method comprises the following steps:

[0074] S1. Drawing a three-dimensional model of a large deformation target; the three-dimensional model includes a rigid body part and a flexible body part;

[0075] S2, performing finite element division on the flexible body part;

[0076] S3, meshing the flexible body part according to the finite element partitioning result, and storing the meshed flexible body part as a .msh file;

[0077] S4. Use MBDyn to couple the rigid part and the flexible part to obtain a rigid-flexible coupling model of a large deformation target.

[0078] S5. Perform dynamic simulation on the rigid-flexible coupling model of the large deformation target to obtain a simulation result of the large deformation target;

[0079] S6, displaying the simulation results in a model visualization to obtain a simulation image;

[0080] S7: setting a camera viewing angle in the simulation image, rendering the simulation image after the camera viewing angle is set, and obtaining an image sequence under the camera viewing angle.

[0081] In this implementation, the three-dimensional model of the large deformation target can be drawn using three-dimensional mapping software.

[0082] In this implementation, finite element software may be used to perform finite element division; when performing finite element division, the flexible body portion is divided into a plurality of nodes and a plurality of units, wherein each unit may be a tetrahedron.

[0083] In this implementation, the Gmsh preprocessing finite element software may be used to perform mesh division on the flexible body portion to obtain a .msh file.

[0084] The .msh file can be used in MBDyn to perform (multi-body) dynamics simulation analysis.

[0085] In this embodiment, MBDyn is a multi-body dynamics simulation software and an integrated modeling and simulation platform for multi-body dynamic systems. It can be used to describe multi-body systems of arbitrary topological structures, closed-loop systems including controllers, flexible multi-body systems considering object deformation, etc.

[0086] MBDyn has complete dynamic modeling, simulation solution and post-processing functions, and integrates a series of large deformation flexible bodies and fluid units based on the ANCF method for dynamic modeling of complex mechanical physical systems.

[0087] In order to better realize platform integration and data interaction,

[0088] MBDyn can connect with external modeling and solving software such as Matlab / Simulink, Python, etc. through API application programming interfaces developed using standard C or C++, complete joint simulation of multi-physical systems and multiple simulation platforms, and realize integration and data interaction of multiple platforms.

[0089] MBDyn is mainly used for (multi-body) dynamic simulation and analysis of spacecraft objects, such as space robots, space flying nets, tethered satellites, space flexible electric sails, and truss cable net antennas (generally multi-body systems); in addition, MBDyn can also be applied to the dynamic simulation and analysis of robots, automobiles, precision equipment, biomechanics, and other complex multi-body systems.

[0090] In this implementation, MBDyn may be used to perform (multi-body) dynamics simulation on a rigid-flexible coupling model of a large deformation target.

[0091] When using MBDyn to perform multibody dynamics simulation on a rigid-flexible coupling model of a large deformation target, it is necessary to first configure the solver parameters, which include mass, quality, specified external force, and field force. After configuring the solver parameters, MBDyn will calculate the motion of the entire rigid-flexible coupling model under the influence of specified external force, field force, and other influencing factors according to the solver settings.

[0092] In this implementation, Paraview may be used to visualize the simulation results and obtain a simulation image.

[0093] like Figure 6 As shown in , through Paraview, the simulation images of the rigid-flexible coupling model of the large deformation target in continuous time frames can be viewed, and the movement of the rigid-flexible coupling model of the large deformation target can be observed from the continuous simulation images.

[0094] In this implementation, Blender may be used to set a camera viewing angle in the simulation image, and the simulation image after the camera viewing angle is set may be rendered to obtain an image sequence under the camera viewing angle.

[0095] Implementation Method 2: Combination Fig.10 and 11 This embodiment is described. This embodiment further limits the method for generating a visual image sequence of a large deformation structure described in the first embodiment. The specific implementation contents are as follows:

[0096] The method further comprises step S5.1;

[0097] S5.1. According to the simulation results, the spatial motion information of the nodes of the rigid-flexible coupling model of the large deformation target is obtained.

[0098] In this implementation, the node spatial motion information can simulate the spatial motion of each node of a large deformation target in a real scene.

[0099] The node spatial motion information can be used as a basis for evaluating the precision (or accuracy) of various visual algorithms (such as visual algorithms without marker points).

[0100] Implementation Method 3: Combination Figure 1 -9 describes this embodiment. This embodiment provides a visual image sequence generation device with a large deformation structure. The specific implementation contents are as follows:

[0101] The visual image sequence generation device includes a three-dimensional modeling module, a finite element division module, a mesh division module, a coupling modeling module, a simulation module, a model visualization module and an image sequence generation module;

[0102] The three-dimensional modeling module is used to draw a three-dimensional model of a large deformation target; the three-dimensional model includes a rigid body part and a flexible body part;

[0103] The finite element division module is used to perform finite element division on the flexible body part;

[0104] The meshing module is used to mesh the flexible body part according to the finite element meshing result, and store the meshed flexible body part as a .msh file;

[0105] The coupling modeling module is used to couple the rigid part and the flexible part using MBDyn to obtain a rigid-flexible coupling model of a large deformation target;

[0106] The simulation module is used to perform dynamic simulation on the rigid-flexible coupling model of the large deformation target to obtain the simulation result of the large deformation target;

[0107] The model visualization module is used to visualize the simulation results and obtain simulation images;

[0108] The image sequence generation module is used to set the camera viewing angle in the simulation image, and is also used to render the simulation image after the camera viewing angle is set to obtain an image sequence under the camera viewing angle.

[0109] In this implementation, the device is used to implement the method for generating a visual image sequence of a large deformation structure described in the first implementation.

[0110] Implementation Method 4: Combination Figure 3 -5 Description of this embodiment. This embodiment is a further limitation of the visual image sequence generation device of a large deformation structure described in the third embodiment. The specific implementation contents are as follows:

[0111] The coupling modeling module includes a rigid body configuration unit, a flexible body configuration unit and a rigid-flexible coupling model configuration;

[0112] The rigid body configuration unit is used to configure the rigid body part in a RIGIDBODY form;

[0113] The flexible body configuration unit is used to configure the flexible body part in the form of GMSHELE;

[0114] The rigid-flexible coupling model configuration is used to couple the rigid body part and the flexible body part in the form of connecting the flexible node to the rigid body.

[0115] Implementation Method 5: Combination Fig.10 and 11This embodiment is described as a further limitation of the visual image sequence generation device of a large deformation structure described in the third embodiment. The specific implementation contents are as follows:

[0116] The simulation module is also used to obtain the node spatial motion information of the rigid-flexible coupling model of the large deformation target according to the simulation results.

[0117] Implementation Method 6: Combination Figure 6 -9 This embodiment is a further limitation of the visual image sequence generation device of a large deformation structure described in the third embodiment. The specific implementation contents are as follows:

[0118] The image sequence generation module includes a camera viewing angle setting unit, a model appearance setting unit and a rendering unit;

[0119] The camera viewing angle setting unit is used to add a camera to the simulation image, and to define the position and camera parameters of the added camera to complete the camera viewing angle setting;

[0120] The model appearance setting unit is used to define the appearance parameters of the rigid-flexible coupling model in the simulation image;

[0121] The rendering unit is used to render the simulation image whose camera viewing angle is set and whose appearance parameters are defined, so as to obtain an image sequence under the camera viewing angle.

[0122] In this implementation, the camera parameters include sensor size, spectrum, lens focal length, lens aperture, etc., as shown in the following table:

[0123] Table 1 Blender camera parameters

[0124] Sensor(inch) spectrum Lens focal length Lens aperture 1.1, Global Shutter Color RGB 16mm 6

[0125] In this implementation manner, the appearance parameters of the rigid-flexible coupling model in the simulation image include material parameters.

[0126] In this implementation, the image sequence under the camera's viewing angle is used to simulate the actual shooting situation of a large deformation target, and can simulate the actual motion state of a large deformation target under different lighting conditions, different excitation disturbances and other scenarios.

[0127] The image sequence under the camera's perspective can replace the real shooting data of the large deformable target and provide a large amount of labeled data for the training of the convolutional neural network.

[0128] Implementation Method 7: Combination Fig.10 and 11 This embodiment describes a method for evaluating the accuracy of a visual algorithm without a marker point. The specific implementation contents are as follows:

[0129] The method comprises the following steps:

[0130] ST1, select nodes as measurement points in the large deformation target;

[0131] ST2. Use a visual algorithm without identification points to calculate the motion information of the measurement points and obtain the calculation data to be evaluated;

[0132] ST3. Using the visual image sequence generation method of the large deformation structure described in Implementation Method 2, obtain the spatial motion information of the nodes corresponding to the measurement points of the rigid-flexible coupling model of the large deformation target, and use the spatial motion information of the nodes corresponding to the measurement points as accurate reference data;

[0133] ST4. Compare and analyze the calculated data to be evaluated with the precise reference data to obtain the accuracy evaluation result of the visual algorithm without identification points.

[0134] In this embodiment, the accuracy evaluation method of the visual algorithm without identification points is implemented based on the visual image sequence generation method of a large deformation structure described in the second embodiment; in fact, the accuracy evaluation method of the visual algorithm without identification points can also be implemented based on the visual image sequence generation method of a large deformation structure described in the first embodiment, and step ST3 is changed as follows:

[0135] ST3. Using the visual image sequence generation method of a large deformation structure described in Implementation Method 1, obtain the simulation result of the large deformation target;

[0136] According to the simulation results, the spatial motion information of the nodes of the rigid-flexible coupling model of the large deformation target is obtained, and the spatial motion information of the nodes corresponding to the measurement points is used as accurate reference data.

[0137] Implementation Method 8: Combination Figure 1 -9 describes this embodiment. This embodiment is an example of the method for generating a visual image sequence of a large deformation structure described in the first embodiment. The specific implementation contents are as follows:

[0138] Taking the space large flexible solar wing as a large deformation target, the image sequence under the camera perspective is obtained by using the visual image sequence generation device of a large deformation structure:

[0139] The first step is to build a rigid-flexible coupling model of a large flexible solar wing in space:

[0140] First, a 3D model of the satellite and the single-sided solar wing is drawn using 3D mapping software, such as Figure 1 As shown; in the three-dimensional model, the satellite body and the connecting bracket are rigid parts, and the solar wing is a flexible part;

[0141] Then, the flexible body part, i.e., the solar wing part, is imported into the finite element software for finite element division;

[0142] Then, the mesh file output by the finite element software is preprocessed by Gmsh to obtain a .msh file, which can be used for dynamic simulation analysis by MBDyn multi-body dynamics simulation software;

[0143] Finally, MBDyn is used to couple the rigid part and the flexible part to obtain the rigid-flexible coupling model of the large flexible solar wing in space. Figure 3 The rigid body is configured in the RIGIDBODY form shown in Figure 4 The GMSHELE form shown in the figure configures the flexible body part and Figure 5 The form of connecting the flexible node to the rigid body is shown to couple the rigid body and the flexible body parts.

[0144] The second step is to perform dynamic simulation on the rigid-flexible coupling model of the large flexible solar wing in space to obtain the simulation results of the large flexible solar wing in space:

[0145] Firstly, the rigid-flexible coupling model of the large flexible solar wing in space is imported into MBDyn software as a configuration file;

[0146] Then, set up the solver, and calculate the motion of the entire model under the influence of specified external forces, field forces and other influencing factors according to the solver settings; set the solver condition to fix the satellite center rigid body, apply a constant field force perpendicular to the plane where the solar wing is located, and obtain the simulation results at this time; in this process, the spatial motion information of each node on the entire model can also be obtained and output;

[0147] Finally, Paraview is used to visualize the above simulation results and obtain simulation images, such as Figure 6 As shown;

[0148] Step 3: Get the image sequence from the camera perspective of the large flexible solar wing in space:

[0149] 3. Satellite rigid-flexible coupling model image rendering generation:

[0150] First, the simulation image obtained by MBDyn simulation is imported into Blender;

[0151] Then, in Blender you can add a camera to the simulation image and define the camera pose and camera parameters, such as Figure 7 As described above, two cameras are mounted on the satellite center rigid body, and the two cameras in the figure can be defined as the left camera and the right camera of the binocular camera group respectively;

[0152] Then, in Blender, you can also define the appearance parameters of each part of the rigid-flexible coupling model (such as material settings);

[0153] Finally, rendering is performed to obtain a more realistic and beautiful image sequence from the camera perspective, such as Figure 8 and 9 As shown. :

[0154] Among them, the image sequence of the flexible solar wing in the left camera perspective is as follows: Figure 8 As shown; the image sequence of the flexible solar wing in the right camera perspective is as follows Fig. 9 shown.

[0155] As can be seen from the figure, the image well records the movement of the flexible solar wing under the action of constant field force, which is similar to the shooting effect of the physical model. The generated image can provide a large amount of data for the training of convolutional neural networks, and can also be used to verify the accuracy of visual algorithms.

[0156] Implementation Method 9: Combination Fig.10 and 11 This embodiment is described as an example of the accuracy evaluation method of the visual algorithm without identification points described in Embodiment 7:

[0157] The large flexible solar wing in space is used as a large deformation target to evaluate the accuracy of the visual algorithm without marker points.

[0158] First, the nodes of the flexible solar wing are selected as measurement points;

[0159] Then, the motion information of the measuring points is calculated using a visual algorithm without identification points to obtain the calculation data to be evaluated;

[0160] Then, the visual image sequence generation method of a large deformation structure described in the above embodiment is used to obtain the nodes corresponding to the measurement points in the rigid-flexible coupling model of the large flexible solar wing in space (such as Fig.10 The spatial motion information of grid 1 and grid 5 in Fig.11 The relationship between the Y-axis position of the grid 1 and the grid 5 in the middle and the time change), and the spatial motion information of the node corresponding to the measurement point is used as the precise reference data;

[0161] Finally, the calculated data to be evaluated is compared and analyzed with the precise reference data to obtain the precision (accuracy) evaluation result of the visual algorithm without identification points.

[0162] The technical solution provided by the present invention is further described in detail above through several specific implementation modes in order to highlight the advantages and benefits of the technical solution provided by the present invention. However, the several specific implementation modes described above are not intended to be used as limitations on the present invention. Any reasonable changes and improvements to the present invention, reasonable combinations of implementation modes and equivalent substitutions based on the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for generating a visual image sequence of a large deformation structure, characterized in that: The visual image sequence generation method comprises the following steps: S1. Drawing a three-dimensional model of a large deformation target; the three-dimensional model includes a rigid body part and a flexible body part; S2, performing finite element division on the flexible body part; S3, meshing the flexible body part according to the finite element partitioning result, and storing the meshed flexible body part as a .msh file; S4. Use MBDyn to couple the rigid part and the flexible part to obtain a rigid-flexible coupling model of a large deformation target. S5. Perform dynamic simulation on the rigid-flexible coupling model of the large deformation target to obtain a simulation result of the large deformation target; S6, displaying the simulation results in a model visualization to obtain a simulation image; S7, setting a camera viewing angle in the simulation image, rendering the simulation image after the camera viewing angle is set, and obtaining an image sequence under the camera viewing angle; The method further comprises step S5.1; S5.

1. Obtaining spatial motion information of nodes of the rigid-flexible coupling model of the large deformation target according to the simulation results; The step S7 comprises: Adding a camera to the simulation image, defining the position and camera parameters of the added camera, and completing the camera viewing angle setting; Defining appearance parameters of the rigid-flexible coupling model in the simulation image; Rendering the simulation image with the camera viewing angle set and appearance parameters defined to obtain an image sequence under the camera viewing angle; When MBDyn is used to perform multi-body dynamics simulation on a rigid-flexible coupling model of a large deformation target, the MBDyn is used to calculate the motion of the entire rigid-flexible coupling model under the action of specified external forces and field forces; By acquiring an image sequence under the camera's perspective, the image sequence is used to replace the real shooting data of the large deformable target, so as to provide annotation data for the training of the convolutional neural network; The node spatial motion information is used as a basis for evaluating the accuracy of the visual algorithm.

2. A device for generating a visual image sequence of a large deformation structure, characterized in that: The visual image sequence generation device includes a three-dimensional modeling module, a finite element division module, a mesh division module, a coupling modeling module, a simulation module, a model visualization module and an image sequence generation module; The three-dimensional modeling module is used to draw a three-dimensional model of a large deformation target; the three-dimensional model includes a rigid body part and a flexible body part; The finite element division module is used to perform finite element division on the flexible body part; The meshing module is used to mesh the flexible body part according to the finite element meshing result, and store the meshed flexible body part as a .msh file; The coupling modeling module is used to couple the rigid part and the flexible part using MBDyn to obtain a rigid-flexible coupling model of a large deformation target; The simulation module is used to perform dynamic simulation on the rigid-flexible coupling model of the large deformation target to obtain the simulation result of the large deformation target; The model visualization module is used to visualize the simulation results and obtain simulation images; The image sequence generation module is used to set the camera viewing angle in the simulation image, and is also used to render the simulation image after the camera viewing angle is set to obtain an image sequence under the camera viewing angle; The simulation module is also used to obtain the node spatial motion information of the rigid-flexible coupling model of the large deformation target according to the simulation results; The image sequence generation module includes a camera viewing angle setting unit, a model appearance setting unit and a rendering unit; The camera viewing angle setting unit is used to add a camera to the simulation image, and to define the position and camera parameters of the added camera to complete the camera viewing angle setting; The model appearance setting unit is used to define the appearance parameters of the rigid-flexible coupling model in the simulation image; The rendering unit is used to render the simulation image whose camera viewing angle is set and whose appearance parameters are defined, to obtain an image sequence under the camera viewing angle; When MBDyn is used to perform multi-body dynamics simulation on a rigid-flexible coupling model of a large deformation target, the MBDyn is used to calculate the motion of the entire rigid-flexible coupling model under the action of specified external forces and field forces; By acquiring an image sequence under the camera's perspective, the image sequence is used to replace the real shooting data of the large deformation target, so as to provide annotation data for the training of the convolutional neural network; The node spatial motion information is used as a basis for evaluating the accuracy of the visual algorithm.

3. The device for generating a visual image sequence of a large deformation structure according to claim 2, characterized in that: The coupling modeling module includes a rigid body configuration unit, a flexible body configuration unit and a rigid-flexible coupling model configuration; The rigid body configuration unit is used to configure the rigid body part in a RIGIDBODY form; The flexible body configuration unit is used to configure the flexible body part in the form of GMSHELE; The rigid-flexible coupling model configuration is used to couple the rigid body part and the flexible body part in the form of connecting the flexible node to the rigid body.

4. A device for generating a visual image sequence of a large deformation structure, comprising: A processor and a memory, characterized in that the memory is used to store executable instructions of the processor, and the processor is configured to execute the method for generating a visual image sequence of a large deformation structure as described in claim 1 by executing the executable instructions.

5. A computer storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is run, the method for generating a visual image sequence of a large deformation structure according to claim 1 is executed.

6. A method for evaluating the accuracy of a visual algorithm without a marker, characterized in that: The method comprises the following steps: ST1, select nodes as measurement points in the large deformation target; ST2. Use a visual algorithm without identification points to calculate the motion information of the measurement points and obtain the calculation data to be evaluated; ST3. Using the visual image sequence generation method of a large deformation structure as described in claim 1, obtain the spatial motion information of the nodes corresponding to the measurement points of the rigid-flexible coupling model of the large deformation target, and use the spatial motion information of the nodes corresponding to the measurement points as accurate reference data; ST4. Compare and analyze the calculated data to be evaluated with the precise reference data to obtain the accuracy evaluation result of the visual algorithm without identification points.

Images