CN114925600A - Learning-based method for measuring bending hardness of suspended fabric - Google Patents

Abstract

The invention provides a learning-based measurement method for bending hardness of a suspended fabric, which comprises the following steps: acquiring the relation between the nonlinear bending modulus and each different bending modulus through real fabric data, and constructing a parameter data set; normalizing the parameters in the parameter data set to obtain a processed parameter data set; constructing a VAE subspace model by utilizing the processed parameter data set; acquiring the initial state of each parameter vector in the VAE subspace model, and generating a simulation data set; generating a multi-view depth map from the simulated dataset; obtaining a learned deep neural network through the learning of the deep neural network by using the multi-view depth map; and acquiring the bending hardness of the real fabric to be measured by utilizing the learned deep neural network. The invention simulates the real state of the cloth as much as possible and improves the measurement precision.

Description

Learning-based method for measuring bending hardness of suspended fabric

Technical Field

The invention relates to the technical field of fabric sample draping form simulation, in particular to a method for measuring the bending hardness of a draping fabric based on learning.

Background

It is well known that bending stiffness is very important for all fabrics, but it is extremely difficult to measure and simulate accurately. The performance and the accuracy are two important indexes for measuring the physical cloth simulation engine; although great success in simulation performance has been achieved in recent years, progress in simulation accuracy has been limited. This situation is especially acute for designers and developers of digital fashion, who need accurate simulation to create virtual garments that resemble real fashion garments.

Among the many factors that affect the accuracy of the simulation, stiffness in plane and bending (stiffness) may be the two most critical factors. Plane stiffness is generally only important for elastic fabrics because it exhibits strong extensibility in the resulting undergarment or athletic garment. In contrast, bending stiffness is important for almost all fabrics, as it determines the softness of the fabric and the details of the folds. Due to the non-linearity, anisotropy and diversity of the real world bending stiffness, accurate measurement becomes a huge challenge.

Over the past few decades, fabric engineers have developed various standardized test methods relating to flexural stiffness, including the cantilever method, the heart method, and the drape method. The cantilever approach is probably the most common and intuitive compared to other approaches. Assuming that the bending characteristics in each direction are not relevant, the cantilever method is: a flying cantilever was used to test how much the cloth strip could bend under its own weight, as shown in figure 1. In the field of graphics, the cantilever method also seems to be a de facto standard for obtaining flexural stiffness parameters.

However, when the cantilever method is used, it takes at least 15 minutes for even an experienced user to measure a single fabric. This includes the time to prepare the cloth sample and the time actually measured; if thousands of stored fabrics need to be digitalized, a large amount of time cost is needed; furthermore, many fabric strips exhibit a bending effect as in fig. 2, making accurate measurements using the cantilever method a difficult problem.

In short, the conventional cloth hardness measurement takes a long time and is difficult to measure.

Disclosure of Invention

In order to solve the problems in the prior art, the invention provides a method for measuring bending hardness of an overhanging type fabric based on learning, which can solve at least one technical problem in the prior art.

The technical scheme adopted by the invention for solving the technical problems is as follows:

a learning-based method for measuring bending hardness of a suspended fabric comprises the following steps:

acquiring the relation between the nonlinear flexural modulus and each anisotropic flexural modulus through real fabric data, and constructing a parameter data set;

normalizing the parameters in the parameter data set to obtain a processed parameter data set;

constructing a VAE subspace model by utilizing the processed parameter data set;

acquiring the initial state of each parameter vector in the VAE subspace model, and generating a simulation data set;

generating a multi-view depth map from the simulated dataset;

obtaining a learned deep neural network through the learning of the deep neural network by using the multi-view depth map;

and acquiring the bending hardness of the real fabric to be measured by utilizing the learned deep neural network.

The relation between the real fabric data acquisition and the nonlinear flexural modulus and the anisotropic flexural modulus comprises the following steps:

preparing a sample of a real fabric, and acquiring an image of the sample;

acquiring a curve sample point set on the image according to the image, and acquiring the moment of any curve sample point in the curve sample point set;

and acquiring the nonlinear flexural modulus and the anisotropic flexural modulus corresponding to the real fabric according to the moment of any curve sample point.

The "acquiring a curve sample point set on the image according to the image, and acquiring a moment of any curve sample point in the curve sample point set" includes:

selecting at least 1 control point on the curve corresponding to the sample on the image, and performing curveAfter fitting, the curve is uniformly sampled along the X-axis to obtain a set of curve sample points { r } ₀ ，...，r _N }；

At r is _i The moment of the points is:

wherein rho is the density of the fabric, g is the gravity acceleration, E is the width of the cloth strip, s is the arc length variable, and s _i And s _N Are respectively r _i Point and r _N The arc length of the spot; df(s) is the differential of the force at the s position, X(s) is the projection of the s position on the X-axis, and xi is the projection of the ith sample point on X.

The step of obtaining the nonlinear flexural modulus and the anisotropic flexural modulus corresponding to the real fabric according to the moment of any curve sample point comprises the following steps:

defining two bending variables simultaneously through the transverse direction, the vertical direction and the oblique direction of the sample to form six parameters so as to form the nonlinear bending modulus of the real fabric;

calculating a principal curvature and a principal curvature direction on all vertexes of the sample;

and estimating the average value of the maximum principal curvature directions of the two outermost vertexes as the bending direction of one dihedral angle element to form the anisotropic bending modulus of the real fabric.

The step of constructing the VAF subspace model by using the processed parameter data set comprises the following steps:

randomly selecting a part of parameter vectors from the parameter data set as a training set, and taking the other part of the parameter vectors as an evaluation set;

training a VAE model through an Adam optimizer to obtain a trained VAE model;

evaluating the trained VAE model through the evaluation set.

Before the step of constructing the VAE subspace model by using the processed parameter data set, the method further comprises the following steps: expanding the reference vector space:

sampling parameters in the parameter data set by adopting Gaussian distribution of N (mu, sigma) to enable the sampling to cover all real fabrics; where μ e [ -0.5, 0.5], σ e [0.8, 1.2] are two uniformly distributed random variables.

The "obtaining an initial state of each parameter vector in the VAE subspace model to generate a simulation dataset" includes:

determining a different initial state for each parameter vector in the VAE subspace model;

randomly generating eight initial states for each parameter vector, adding a random disturbance to the position of each initial state, and simulating the disturbance by using a simulation engine until the disturbance is calm to obtain the parameter vector of the initial state;

and adding the initial state parameter vector and the corresponding parameter vector into a data set formed by the simulation data.

The initial state comprises the following steps:

states generated by adding random sine waves to the planar fabric mesh; or the like, or, alternatively,

a state formed by intentionally folding the fabric mesh in randomly selected directions; or the like, or, alternatively,

and randomly selecting the initial state of other simulated samples in the simulation data set as the initial state of the current sample.

The "generating a multi-view depth map from the simulation dataset" includes:

layered random sampling is carried out to obtain at least 1 random orientation of each analog data in the analog data set;

and synthesizing at least 1 group of multi-view depth maps by random disturbance to the position or posture or the view of the camera by utilizing the random orientation.

The step of obtaining the learned deep neural network through learning of the deep neural network comprises the following steps:

defining a loss function in the deep neural network as ground route { g _i And the predicted result { p } _i RMSE error between }:

wherein N is the batch size;

and training the deep neural network by using an Adam optimizer to obtain a learned deep neural network.

Has the advantages that:

according to the method, a parameter data set is constructed by acquiring the relation between real fabric data and nonlinear flexural modulus and various different flexural moduli; carrying out normalization processing on the parameters to obtain a processed parameter data set; constructing a VAE subspace model by utilizing the processed parameter data set; acquiring the initial state of each parameter vector in the VAE subspace model, and generating a simulation data set; generating a multi-view depth map from the simulated dataset; obtaining a learned deep neural network through learning of the deep neural network by using the multi-view depth map; finally, acquiring the bending hardness of the real fabric to be measured by utilizing the learned deep neural network; the invention does not need to use a cantilever method to measure the data of each cloth, thereby saving a great deal of time; meanwhile, the real state of the cloth is simulated as much as possible by simulating the initial state of each parameter vector in the VAE subspace model, and the measurement precision is improved.

Drawings

FIG. 1 is a schematic diagram of a cantilever method;

FIG. 2 is an example of the bending of a fabric strip;

FIG. 3 is a flow chart of the method of the present invention;

FIG. 4 is a schematic view of the free stretch or compression of the fabric;

FIG. 5 is a schematic view of a bend along a connected edge;

FIG. 6 is a schematic view of an apparatus for measuring an actual fabric;

FIG. 7 is a schematic view of a bending curve of a cloth strip sample;

FIG. 8 is a schematic diagram of patterns generated by adding random sine waves to a planar fabric grid;

FIG. 9 is a schematic view of a fabric state resulting from intentionally folding a web of fabric in randomly selected directions;

FIG. 10 is an illustration of a depth map;

fig. 11 is a schematic diagram of a neural network structure.

Detailed Description

The technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

The present invention provides an embodiment:

as shown in fig. 3, a method for measuring bending hardness of a suspended fabric based on learning includes: acquiring the relation between the nonlinear bending modulus and each different bending modulus through real fabric data, and constructing a parameter data set; normalizing the parameters in the parameter data set to obtain a processed parameter data set;

constructing a VAE subspace model by utilizing the processed parameter data set; acquiring the initial state of each parameter vector in the VAE subspace model, and generating a simulation data set; generating a multi-view depth map from the simulated dataset; obtaining a learned deep neural network through the learning of the deep neural network by using the multi-view depth map; and acquiring the bending hardness of the real fabric to be measured by utilizing the learned deep neural network.

In order to better simulate the cloth, the application provides a co-rotational finite element model for processing plane hardness and a dihedral angle model for processing bending hardness; wherein, in the co-rotational finite element model for processing the plane hardness, the co-rotational finite element model is adopted to simulate the plane hardness of the cloth; however, the plane stiffness affects the simulation of the bending behavior, resulting in a locking problem; in order to solve the problem, the fabric is allowed to be (99%, 101%)]Freely pulled up or compressed, as shown in fig. 4; meanwhile, selecting a dihedral angle model to model the bending hardness of the fabric, and selecting the bending modulus as a parameter; in particular, it is assumed that a dihedral angle element is composed of two adjacent triangles which are planar in the reference state. The only way to deform the element is to bend along the connected edges, as shown in fig. 5. By definition, the bending energy is:

wherein tau (kappa, e) is a moment function tau (kappa, e) with a curvature kappa and a connection length e as parameters; using the above equation, the force at vertex i is calculated:

where we consider the dihedral angle θ as a function of vertex position; let τ (κ, e) be linearly proportional to e.

The acquisition process of the nonlinear bending modulus and the anisotropic bending modulus of the cloth is as follows:

to construct the non-linear bending behavior, we define the moment τ (κ, e) as a quadratic function of the curvature κ: τ (κ, e) ═ α κ + β κ ² ) e, the length of the e side; where α and β are the two flexural moduli. Defining alpha and beta in three horizontal, vertical and oblique directions to obtain six parameters; the force at vertex i becomes:

assuming that the face deformation is almost isometric and the dihedral elements are sufficiently small, θ ≈ 0. This dihedral angle element is then locally approximated by a cylinder and the curvature is estimated from the radius R shown in fig. 5:

for simplicity and high efficiency, consider H ₀ And H ₁ Calculated as the triangular height in the reference state. Thus, κ becomes a linear function of θ.

The real fabric can show different bending properties in different material directions. One typical method of simulating anisotropic bending is to assign different bending stiffness coefficients to the dihedral angle elements depending on the orientation of their connecting edges in the material space. This method, however, does not actually bend anisotropically correctly because, from geometric considerations, the edges are not necessarily oriented in the same direction as the bends. In the simulation, such errors would cause the warping effect to depend on the partitioning of the triangular mesh.

Therefore, the present embodiment calculates the principal curvature and the principal curvature direction at all vertices, and then estimates the average of the maximum principal curvature directions of the two edge vertices as the bending direction of one dihedral angle element. Let k ^warp ＝[α ^warp β ^warp ]And k is ^weft ＝[α ^weft β ^weft ]The flexural modulus in the warp and weft directions, respectively. Direction of any bending

Flexural modulus of

Similar to the curvature, both can be approximated as:

to make the anisotropic model more accurate, the flexural moduli in multiple sampling directions are taken as parameters k of the model ^warp ，...，k ^weft }. Order to

And

two sampling directions. We calculated the flexural modulus therebetween as:

wherein

Intuitively, equation 6 first predicts the flexural modulus in the warp and weft directions and then uses them to calculate

Three sampling directions are selected: 0, pi/4 and pi/2, the final parameter vector is defined as: k ═ k ^warp k ^diag k ^weft ]。

Specifically, in order to measure the real fabric, the device shown in fig. 6 is adopted, which comprises an adjustable inclined plane, a grid bottom plate and a telephoto single lens reflex camera shot from a distance; the slope of the inclined plane can be adjusted, and for each fabric, cloth strip samples of 200mm multiplied by 30mm are respectively prepared in the three material directions of warp yarns, oblique yarns and weft yarns; let the bending curve of a cloth sample be as shown in fig. 7; manually selecting four control points on the curve and fitting the control points by using a cubic Bezier curve; the curve is then uniformly sampled along the X-axis, resulting in a set of curve sample points { r } ₀ ，...，r _N -wherein N is 2000; as defined at r _i The moment of the points is:

where ρ is the density of the fabric, g is the acceleration of gravity, E is the width of the cloth strip, s is the arc length variable, and s is _i And s _N Are respectively r _i Point and r _N The arc length of the spot. Using sampling, the trapezoidal rule is applied to the above equation to calculate τ _i ：

At the same time, τ is (α κ + β κ) ² ) E, where α and β are two unknown flexural moduli. Given each r _i Point estimated kappa _i And τ _i Then, solving a quadratic regression problem to obtain alpha and beta, and obtaining six flexural moduli of a nonlinear anisotropic model aiming at any fabric; respectively as follows: alpha and beta are respectively corresponding to the horizontal direction, the vertical direction and the oblique direction; by the above method, on average, it takes 15 (one person) minutes to measure a fabric, which includes quasi-one personThe time to prepare the sample, the time to cantilever test and the time to parameter fit. The data set in this example contains a total of 618 parameters of a real fabric commonly used in garment production, while the entire measurement process takes over 150 (one person) hours.

Through the discussion above, it can be known that α and β are in a linear relationship with the density ρ of the fabric in the cantilever test; assuming that all fabrics have the same density:

the measured flexural modulus was then normalized:

wherein rho is the actual fabric density measured by the scale; to convert back to the actual parameters, they need only be multiplied by

Then the method is finished; the subspace is then constructed using the normalized parameters and the neural network is trained.

This embodiment uses a Variational Automatic Encoder (VAE) model to define the subspace, which essentially attempts to recover the same vector result at the output using one parameter vector as input. The encoder of this model consists of three complete connected layers of 2048, 1024 and 512 elements respectively. The size of the potential space is 64. The decoder is structurally opposite to the encoder; the VAE model was trained using an Adam optimizer with a decay weight of 10 ^-5 Learning rate of 10 ^-4 The batch size was 16. 494 parameter vectors were randomly picked as training data sets and the rest were taken as evaluation data sets. After training the VAE model, we can conveniently convert the random latent space vector satisfying Gaussian distribution N (0, 1) into a parameter vector as a sample through a decoder.

Due to limited training data and limited cantilever measurement accuracy, the subspace may not be sufficient to cover all real fabrics. To deal with this problem, this embodiment does not sample the pre-space variables by a gaussian distribution of N (0, 1), but uses a gaussian distribution of N (μ, σ), where μ e [ -0.5, 0, 5], σ e [0.8, 1.2] are two uniformly distributed random variables, which can effectively expand the transformed parameter vector space.

After the subspace is defined, the cloth needs to be simulated; since the actual draping result is related to how the person touches the sample, the whole draping process cannot be recorded in a precisely controllable way. Mathematically, there are multiple local minima for the simulation target, each corresponding to a possible overhang result. This embodiment employs three types of initial states, including a cloth state generated by adding random sine waves to the planar fabric mesh, as shown in fig. 8; or a cloth state formed by intentionally folding a mesh of the fabric in randomly selected directions, as shown in fig. 9; and randomly selecting overhanging grids of other samples which are well simulated in the existing data set as the initial state of the current sample.

For each parameter vector sample, eight initial states are randomly generated. Since the center of the fabric sample may not coincide exactly with the center of the cylinder, a small random perturbation may be added to the position of each initial state. Then simulating them to static equilibrium with a simulation engine; the grid resolution of the fabric sample is 100 x 100; each simulation is typically completed in 20 seconds. After simulating the six results, we add them to the simulation dataset with their corresponding parameter vectors.

The 12 random orientations of each simulated pendulous model are obtained by using hierarchical sampling, and then the 12 orientations are used to synthesize 12 sets of 240 x 180 multi-view depth maps by adding random disturbance to the camera position/posture/view. The depth map is shown in fig. 10, which includes four viewing angles. Finally, noise related to Kinect is added to the synthesized depth map to obtain a real depth map similar to the depth map containing noise and errors, and finally each group of depth maps and the corresponding parameter vector thereof are used as one data point in the synthesized data set.

In the present embodiment, 6000 parametric vector samples were generated and 48000 drape models were simulated. The simulated drape model was then converted into a synthetic depth map dataset comprising a total of 48000 × 12 × 4 ═ 2.3M tensor depth maps, taking about 2 hours.

As shown in fig. 11, the neural network described in this embodiment is composed of one ResNet-18 layer plus two complete connection layers. The whole network contains 12.8M variables; defining the loss function as normalized ground route { g } _i And the predicted result { p } _i RMSE error between }:

wherein N is the batch size; the network was trained using an Adam optimizer with a decay weight of 10 ^-4 Learning rate of 10 ^-4 The batch size is 128 and the learning rate decay is 0.995.

The invention also provides an embodiment:

a fabric bending hardness measuring device comprising: a storage medium and a processing unit; wherein the storage medium is used for storing a computer program; the processing unit exchanges data with the storage medium and is used for executing the computer program through the processing unit when the fabric bending hardness is measured so as to carry out the steps of the fabric bending hardness measuring method.

In the above test apparatus, the storage medium is preferably a storage device such as a mobile hard disk, a solid state disk, or a usb disk; a processing unit, preferably a CPU, for exchanging data with the storage medium, is used for executing the computer program by the processing unit when measuring the fabric bending hardness, so as to perform the steps of the fabric bending hardness measuring method.

The CPU described above can perform various appropriate actions and processes according to a program stored in a storage medium. The ATE test apparatus may also include peripherals, including input parts for a keyboard, a mouse, etc., and output parts such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), etc., and a speaker, etc.; in particular, according to the disclosed embodiment of the invention, the process as described in any of fig. 3 may be implemented as a computer software program.

An embodiment provided by the present invention comprises a computer program product comprising a computer program embodied on a computer-readable medium, the computer program comprising program code for performing the method shown in the flowchart of any one of fig. 3. The computer program may be downloaded and installed from a network. The computer program, when executed by the CPU, performs the above-described functions defined in the system of the present invention.

The present invention also provides a computer-readable storage medium having a computer program stored therein; the computer program, when executed, performs the steps of the fabric bending stiffness measurement method as described above. In the present invention, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.

In the present invention, a computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, fiber optic cable, RF, etc., or any suitable combination of the foregoing.

The above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily change or replace the technical scope of the present invention within the technical scope of the present invention. Therefore, the protection scope of the present invention is subject to the protection scope of the claims.

Claims (10)

1. A learning-based method for measuring bending hardness of a suspended fabric is characterized by comprising the following steps:

acquiring the relation between the nonlinear bending modulus and each different bending modulus through real fabric data, and constructing a parameter data set;

normalizing the parameters in the parameter data set to obtain a processed parameter data set;

constructing a VAE subspace model by utilizing the processed parameter data set;

acquiring the initial state of each parameter vector in the VAE subspace model, and generating a simulation data set;

generating a multi-view depth map from the simulated dataset;

obtaining a learned deep neural network through the learning of the deep neural network by using the multi-view depth map;

and acquiring the bending hardness of the real fabric to be measured by utilizing the learned deep neural network.

2. The method for measuring bending hardness of suspended fabric based on learning as claimed in claim 1, wherein the relation between real fabric data acquisition and nonlinear bending modulus and anisotropic bending modulus comprises:

preparing a sample of a real fabric, and acquiring an image of the sample;

acquiring a curve sample point set on the image according to the image, and acquiring the moment of any curve sample point in the curve sample point set;

and acquiring the nonlinear bending modulus and the anisotropic bending modulus corresponding to the real fabric according to the moment of any curve sample point.

3. The method for measuring bending hardness of suspended fabric based on learning according to claim 2, wherein the step of acquiring a curve sample point set on the image according to the image and acquiring a moment of any curve sample point in the curve sample point set comprises the following steps:

selecting at least 1 control point on the curve corresponding to the sample on the image, and after curve fittingUniformly sampling the curve along the X-axis to obtain a set of curve sample points { r } ₀ ，...，r _N }；

At r _i The moment of the points is:

wherein rho is the density of the fabric, g is the gravity acceleration, E is the width of the cloth strip, s is the arc length variable, and s _i And s _N Are respectively r _i Point and r _N The arc length of the spot; df(s) is the differential of the force at the s-position, x(s) is the projection of the s-position on the x-axis, x _i Is the projection of the ith sample point on X.

4. The method for measuring the bending hardness of the overhanging type fabric based on learning as claimed in claim 2, wherein the step of obtaining the nonlinear bending modulus and the anisotropic bending modulus corresponding to the real fabric according to the moment of any curve sample point comprises:

defining two bending variables simultaneously through the transverse direction, the vertical direction and the oblique direction of the sample to form six parameters so as to form the nonlinear bending modulus of the real fabric;

calculating a principal curvature and a principal curvature direction on all vertexes of the sample;

and estimating the average value of the maximum principal curvature directions of the two outermost vertexes as the bending direction of one dihedral angle element to form the anisotropic bending modulus of the real fabric.

5. The learning-based measurement method for bending stiffness of suspended fabric according to claim 1, wherein the step of constructing the VAE subspace model by using the processed parameter data set comprises the following steps:

randomly selecting a part of parameter vectors from the parameter data set as a training set, and taking the other part of the parameter vectors as an evaluation set;

training a VAE model through an Adam optimizer to obtain a trained VAE model;

evaluating the trained VAE model through the evaluation set.

6. The learning-based measurement method for bending stiffness of suspended fabric according to claim 5, wherein before the step of constructing the VAE subspace model by using the processed parameter data set, the method further comprises: expanding the reference vector space:

sampling parameters in the parameter data set by adopting Gaussian distribution of N (mu, sigma) to enable the sampling to cover all real fabrics; where μ e [ -0.5, 0.5], σ e [0.8, 1.2] are two uniformly distributed random variables.

7. The method for measuring bending hardness of suspended fabric based on learning according to claim 1, wherein the step of obtaining an initial state of each parameter vector in the VAE subspace model and generating a simulation data set comprises:

determining a different initial state for each parameter vector in the VAE subspace model;

randomly generating eight initial states for each parameter vector, adding a random disturbance to the position of each initial state, and simulating the disturbance by using a simulation engine until the disturbance is calm to obtain the parameter vector of the initial state;

and adding the initial state parameter vector and the corresponding parameter vector into a data set formed by the simulation data.

8. The learning-based measurement method for bending stiffness of suspended fabric according to claim 7, wherein the "initial state" comprises:

states generated by adding random sine waves to the planar fabric mesh; or the like, or a combination thereof,

a state formed by intentionally folding the fabric mesh in randomly selected directions; or the like, or, alternatively,

and randomly selecting the initial state of other simulated samples in the simulation data set as the initial state of the current sample.

9. The method for measuring bending hardness of suspended fabric based on learning as claimed in claim 1, wherein the step of generating a multi-view depth map through the simulation data set comprises:

layered random sampling is carried out to obtain at least 1 random orientation of each analog data in the analog data set;

and synthesizing at least 1 group of multi-view depth maps by random disturbance to the position or posture or the visual field of the camera by utilizing the random orientation.

10. The method for measuring bending hardness of suspended fabric based on learning according to claim 1, wherein the step of obtaining the learned deep neural network through learning of the deep neural network comprises the following steps:

defining a loss function in the deep neural network as ground route { g _i And the predicted result { p } _i RMSE error between }:

wherein N is the batch size;

and training the deep neural network by using an Adam optimizer to obtain a learned deep neural network.

Images