Scene three-dimensional point cloud registration method, device and medium based on Gaussian mixture model

Abstract

The application discloses a scene three-dimensional point cloud registration method and device based on a Gaussian mixture model, and a medium, and belongs to the field of point cloud registration, and comprises the following steps: S1, collecting point clouds and processing the point clouds; S2, clustering the point clouds to generate a Gaussian mixture model; S3, performing ground point cloud elimination on the point clouds by using the generated Gaussian mixture model; S4, constructing a maximum likelihood function according to the Gaussian mixture model obtained after processing, and generating a response matrix of to-be-registered points to a source point cloud submodel; S5, optimizing a target function according to the obtained response matrix, and iteratively solving a rotation matrix and a translation vector; and S6, repeatedly iterating the steps S4 and S5 until a convergence condition is met, that is, the optimal rotation matrix and the translation vector are obtained, and the target point cloud is transformed by using the obtained optimal rotation matrix and the translation vector, so that the fast registration of the point cloud is completed. The application improves the accuracy and rapidity of scene point cloud registration.

Description

Technical Field

[0001] This invention relates to the field of point cloud registration, and more specifically, to a method, device, and medium for 3D point cloud registration of a scene based on a Gaussian mixture model. Background Technology

[0002] In recent years, 3D reconstruction technology has continued to develop, and more and more researchers have begun to focus on 3D point cloud registration of objects or scenes. Most existing point cloud registration techniques are based on the iterative nearest point method, which has problems such as poor robustness and sensitivity to initial values. When there are many noisy points or outliers in the point cloud, the method performs very poorly.

[0003] Registration methods based on probabilistic models transform point-to-point hard matching into probability density function matching, which effectively solves the problem of the iterative nearest-point method. The coherent point drift algorithm proposed by Myronenko et al., based on Gaussian mixture models for point cloud registration, has achieved good results. However, the coherent point drift method establishes a Gaussian probability model for each point in the point cloud, leading to excessively high computational complexity. Furthermore, the coherent point drift method assumes that each Gaussian model has an equal covariance matrix and is isotropic, resulting in decreased registration accuracy.

[0004] In addition, there are a large number of ground point clouds in the scene point cloud. Since ground point clouds cannot provide effective information for registration like other environments, they increase the number of points by a large amount. Some existing registration methods directly downsample and filter the point cloud to reduce the number of points, but the filtering process increases the amount of computation and removes valuable information from the point cloud, resulting in a worse registration effect. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, device and medium for 3D point cloud registration of a scene based on a Gaussian mixture model, which improves the accuracy and speed of scene point cloud registration.

[0006] The objective of this invention is achieved through the following solution:

[0007] A method for 3D point cloud registration of a scene based on a Gaussian mixture model includes the following steps:

[0008] S1, collect point cloud data and process the point cloud data;

[0009] S2, clusters the point cloud to generate a Gaussian mixture model;

[0010] S3, ground point cloud removal is performed on the point cloud using the generated Gaussian mixture model;

[0011] S4. Based on the Gaussian mixture model obtained after step S3, construct the maximum likelihood function and generate the response matrix of the cloud sub-model from the point to be registered to the source point.

[0012] S5. Based on the response matrix obtained in step S4, optimize the objective function and iteratively solve for the rotation matrix and translation vector.

[0013] S6. Repeat steps S4 and S5 until the convergence condition is met, i.e., the optimal rotation matrix and translation vector are obtained. Use the obtained optimal rotation matrix and translation vector to transform the target point cloud, thus completing the fast registration of the point cloud.

[0014] Furthermore, after step S6, the method also includes the step of running steps S1 to S6 on each frame of point cloud acquired, thereby completing the three-dimensional reconstruction of the scene.

[0015] Further, step S1 includes the following sub-steps: acquiring two frames of point cloud data at two positions and angles, and performing null point removal and downsampling processing on the original point cloud data.

[0016] Further, step S2 includes the sub-step of clustering the two point clouds to generate Gaussian mixture models, specifically including the following sub-steps:

[0017] Point clouds are clustered using k-means, and the cluster centers are used as initial values ​​for the mean of a Gaussian mixture model. The mean, covariance, and sub-model weights of the Gaussian mixture model are iteratively solved using the EM algorithm. The resulting probability density function of the Gaussian mixture model is expressed as:

[0018]

[0019] In the formula:

[0020]

[0021] Where K represents the number of sub-models in the Gaussian mixture model, y represents the coordinates of the target point cloud, and θ k The model parameters of the k-th sub-model include: μ k Σ represents the mean of the k-th sub-model. k Let α represent the covariance matrix of the k-th sub-model. k This represents the weight of the k-th sub-model.

[0022] Further, step S3 includes the following sub-steps: analyzing the covariance of the generated sub-models, considering those with covariance z-axis components less than a certain threshold as ground point sub-models, removing points from the ground point sub-models, and updating the weights of the non-ground sub-models.

[0023] Furthermore, step S4 includes the following sub-steps:

[0024] Based on the Gaussian mixture model obtained in step S3, a maximum likelihood function is constructed, and the response matrix of the cloud sub-model from the point to be registered to the source point is generated. The expression of the constructed maximum likelihood function is as follows:

[0025]

[0026] Where T is the transformation matrix of the target point cloud, containing the rotation matrix and translation vector, R represents the rotation matrix, t represents the translation vector, and M... ′ K represents the number of points in the target point cloud. ′ ∑ represents the number of sub-models. k This represents the covariance matrix of the k-th sub-model;

[0027] The elements in the response matrix represent the posterior probability that a point in the target point cloud belongs to a sub-model in the mixture model, and are calculated by the following formula:

[0028]

[0029] In the formula, α j μ represents the weight of the j-th sub-model. j Let ∑ represent the mean of the j-th sub-model. j Let J represent the covariance matrix of the j-th sub-model, and J represent the total number of sub-models.

[0030] Furthermore, step S5 includes the following sub-steps:

[0031] Based on the response matrix obtained in step S4, the objective function is optimized by iteratively solving for the rotation matrix and translation vector. After simplification, the objective function is transformed from the form of a probability density exponential function into the form of a Mahalanobis distance.

[0032]

[0033] By decoupling the model's parameters, the objective function is further simplified to:

[0034]

[0035] The point cloud is decentroided, and the rotation matrix is ​​solved using the SVD method to further obtain the translation vector; where M ′ Indicates the total number of target point clouds. This indicates the weighted average. This represents the weighted mean of the model.

[0036] Furthermore, the acquisition of two frames of point cloud data at two locations and angles is specifically achieved using lidar.

[0037] A computer device includes a processor and a memory, the memory storing a computer program that, when loaded by the processor, executes the method as described in any of the preceding claims.

[0038] A computer-readable storage medium storing a computer program therein, the computer program being loaded by a processor and executing the method as described in any of the preceding claims.

[0039] The beneficial effects of this invention include:

[0040] This invention utilizes the EM algorithm to generate a Gaussian mixture model from scene point clouds. By analyzing the covariance matrix, ground points in the point cloud are removed, thereby improving the speed and accuracy of point cloud registration. Existing registration methods based on Gaussian mixture models use complex log-likelihood functions as objective functions, while this invention reduces the computational cost of optimizing the objective function by transforming it into a simple Mahalanobis distance form. Furthermore, unlike existing methods, this invention uses anisotropic covariance matrices from the Gaussian mixture model, which more accurately describes the distribution of the point cloud. By decomposing the covariance matrix and applying the SVD method to solve for the transformation matrix between the point clouds to be registered, the accuracy and speed of scene point cloud registration are further improved. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart illustrating the steps of the method in an embodiment of the present invention;

[0043] Figure 2 These are two frames of point cloud maps to be registered in the method of this embodiment of the invention. Detailed Implementation

[0044] All features disclosed in all embodiments of this specification, or steps in all methods or processes implied in the disclosure, may be combined and / or extended or replaced in any way, except for mutually exclusive features and / or steps.

[0045] To address the shortcomings of existing technologies, this invention provides a method for 3D point cloud registration based on a Gaussian mixture model. The method involves clustering the point cloud to generate a Gaussian mixture model, followed by removing ground points. Finally, the expectation-maximization method is used to optimize the transformation matrix (including rotation and translation vectors) between the target and source point clouds, and this transformation is applied to the source point cloud to achieve point cloud registration, thus improving the speed and accuracy of the registration. In a specific implementation, the method includes the following steps:

[0046] The first step is to use a lidar to acquire two frames of point cloud data at two different locations and angles, and then perform descrambling and downsampling on the original point cloud data.

[0047] The second step is to cluster the two point cloud frames to generate Gaussian mixture models. First, the point cloud is clustered using k-means, and the cluster centers are used as the initial values ​​of the Gaussian mixture model mean. The mean, covariance, and sub-model weights of the Gaussian mixture model are then solved iteratively using the EM algorithm.

[0048] The probability density function of the generated Gaussian mixture model can be expressed as:

[0049]

[0050] In the formula:

[0051]

[0052] Where, μ k Let α be the mean of the k-th sub-model, Σ be the covariance matrix of the k-th sub-model, and α be the mean of the k-th sub-model. k The weights are for the k-th sub-model.

[0053] The third step involves removing ground point clouds from the generated Gaussian mixture model. The covariance of the generated sub-models is analyzed, and those with a covariance z-axis component less than a certain threshold are considered ground point sub-models. Points in these ground point sub-models are then removed, and the weights of the non-ground sub-models are updated.

[0054] The fourth step is to construct the maximum likelihood function based on the Gaussian mixture model obtained in the third step, and generate the response matrix of the cloud sub-model from the point to be registered to the source point.

[0055] The expression for the constructed maximum likelihood function is:

[0056]

[0057] Where T is the transformation matrix of the target point cloud, which includes the rotation matrix and translation vector.

[0058] The elements in the response matrix represent the posterior probability that a point in the target point cloud belongs to a sub-model in the mixture model, which can be calculated by the following formula:

[0059]

[0060] The fifth step involves optimizing the objective function based on the response matrix obtained in the fourth step, iteratively solving for the rotation matrix and translation vector.

[0061] After simplifying the objective function, the objective function is transformed from the form of a probability density exponential function into the form of a simple Mahalanobis distance:

[0062]

[0063] By decoupling the model's parameters, the objective function is further simplified to:

[0064]

[0065] The point cloud is decentrifuged, and the rotation matrix is ​​solved using the SVD method to obtain the translation vector.

[0066] Repeat steps four and five until the convergence condition is met, i.e., the optimal rotation matrix and translation vector are obtained. Using the obtained rotation matrix and translation vector to transform the target point cloud, rapid registration of the two point cloud frames can be achieved. Running the above algorithm on each frame of point cloud acquired by the radar from the scene completes the 3D reconstruction of the scene. The complete process is as follows: Figure 1 As shown.

[0067] It should be noted that, within the scope of protection defined in the claims of this invention, the following embodiments can be combined and / or extended or replaced in any logical manner from the above specific embodiments, such as the disclosed technical principles, disclosed technical features or implicitly disclosed technical features.

[0068] Example 1

[0069] This invention provides an automatic registration method for scene point clouds. First, a Gaussian mixture model of the point cloud is iteratively constructed using cluster centers obtained through k-means clustering. The parameters of the sub-probability models are analyzed to identify and eliminate ground sub-models, and the weights of non-ground sub-models are updated. Then, a maximum likelihood objective function is constructed based on the established mixture model. Using the EM algorithm as the overall optimization framework, the response matrix from the point to be registered to the source point cloud sub-model is first generated, and the objective function is transformed from probability density to Mahalanobis distance. Finally, the optimal transformation parameters are obtained using weighted singular value decomposition to complete the registration of the scene point cloud. The algorithm's input is any two overlapping frames of scene point clouds, and the output is the registered two frames of point clouds, the transformation parameters, and the registration error. The implementation process of this embodiment specifically includes the following steps:

[0070] Step 1: Perform void removal and downsampling on the acquired point cloud. Since point clouds acquired by lidar are typically dense, with a large number of points and some voids, this embodiment first removes voids and then downsamples the point cloud. The processed source point cloud coordinates are X... d×N = [x1,x2,x3,…,x N The target point cloud coordinates are Y. d×M =[y1,y2,y3,…,y M The diagram shows that 'd' represents the dimension of the point cloud, with each point having x, y, and z coordinates. M and N represent the number of points in the processed target point cloud and the number of points in the source point cloud, respectively. This example demonstrates the registration effect of the algorithm using two frames of point clouds collected by an autonomous vehicle equipped with a LiDAR system. The two frames of point clouds to be registered are shown below. Figure 2 As shown. In this example, the target point cloud has 122,227 points, and the source point cloud has 120,574 points, i.e., M = 122,227 and N = 120,574.

[0071] Step 2: Generate a Gaussian mixture model (GMM) of the two point clouds. The GMM parameters are generated using the EM algorithm. The probability density function of the GMM can be expressed as:

[0072]

[0073] In the formula:

[0074]

[0075] Where, μ k Let α be the mean of the k-th sub-model, Σ be the covariance matrix of the k-th sub-model, and α be the mean of the k-th sub-model. k The weights are for the k-th sub-model.

[0076] First, set the parameter K. x and K y K x K represents the number of sub-models in the source point cloud clustering Gaussian mixture model. y Let be the number of sub-models in the Gaussian mixture model for clustering the target point cloud. To avoid convergence to local optima, the k-means algorithm is first used to cluster the point cloud. Model parameters are initialized as follows: the initial mean of the sub-models is set to the cluster centers obtained by the k-means algorithm, the initial covariance matrix of the sub-models is set to the identity matrix, and the initial weight of each sub-model is set to... Where K is the number of Gaussian sub-models.

[0077] The model parameters are then solved iteratively using the EM algorithm. Each iteration consists of an E-step and an M-step. In the E-step, the latent variables are calculated using the model parameters obtained from the previous generation:

[0078]

[0079] in Let represent the response of the k-th sub-model to the j-th point. The practical significance of the latent variable is an estimate of the probability that the j-th point belongs to the k-th sub-model.

[0080] In the M-step, the model parameters are estimated using the maximum likelihood method. The log-likelihood function is as follows:

[0081]

[0082] Update model parameters:

[0083] Mean:

[0084] Covariance:

[0085] Weight:

[0086] Repeat the E-step and M-step iterations until the convergence condition is met. The convergence condition is set to either the number of iterations exceeds the maximum number of iterations or the change in model parameters is less than a set threshold.

[0087] In this example, the number of sub-models in the generated Gaussian mixture model is set to 50.

[0088] Step 3: Identify and remove ground points in the point cloud based on the covariance of the sub-models. Since the ground point cloud can be considered as a planar point cloud with relatively small fluctuations in the z-axis coordinate, this invention selects ground point sub-models and non-ground point sub-models by analyzing the sub-models of the hybrid model. Because the coordinates in the three dimensions of the point cloud distribution are mutually independent, the covariance matrix is ​​a diagonal matrix. The covariance matrix of the k-th sub-model can be expressed as:

[0089]

[0090] Where σ x ,σ y ,σ z The variances are the variances in the three dimensions x, y, and z, respectively.

[0091] Through σ z To determine whether a point is on the ground or not. When σ z Points smaller than a preset threshold are considered ground points, σ z Points exceeding a threshold are considered non-ground points. The sub-model containing these ground points is removed, and the sub-models containing the non-ground points are retained. The weights of the remaining sub-models are then updated using the following formula:

[0092]

[0093] Where, α k To update the weights of the non-ground submodel, α ′ k To update the weights of the non-ground submodel, α i This defines the weights of the ground sub-models before updating. We define the number of points in the source point cloud after removing ground points as N′, the number of sub-models as K′, and the number of points in the target point cloud as M′. In this example, the source point cloud retains 17733 non-ground points, the number of sub-models is 18, and the target point cloud retains 21748 non-ground points.

[0094] Step 4: Construct the maximum likelihood function and generate the response matrix from the point to be registered to the source point cloud sub-model. The rotation matrix is ​​obtained through parameter optimization using the maximum likelihood method. Specifically, by rotating and translating the target point cloud, the probability of points in the target point cloud appearing in the Gaussian mixture model generated from the source point cloud is maximized. The maximum likelihood function expression is:

[0095]

[0096] Here, T is the transformation matrix of the target point cloud, which includes the rotation matrix and translation vector. First, the rotation matrix and translation vector are initialized, and then the EM algorithm is used iteratively to solve for them.

[0097] In this embodiment, the rotation matrix is ​​initialized as follows:

[0098]

[0099] The translation vector is initialized as follows:

[0100]

[0101] Based on the rotation matrix and translation vector of the previous generation, the posterior probability of each point in the target point cloud belonging to a sub-model in the hybrid model is calculated in parallel:

[0102]

[0103] Step 5, according to c in step E ik The parameters are solved by maximizing the likelihood function. The objective function is simplified by ignoring terms independent of R and t, and becomes:

[0104]

[0105] Define the transformed target point cloud as:

[0106]

[0107] Calculate the weighted average of the sub-model means and the average of the latent variables:

[0108]

[0109]

[0110] and After transformation, it becomes:

[0111]

[0112] The objective function can be transformed to obtain:

[0113]

[0114]

[0115] Ignoring terms irrelevant to R and t, the objective function becomes:

[0116]

[0117] After the above processing, the original objective function of M′×K′ terms is transformed into the form of summation of Mahalanobis distances of K′ terms, which reduces the amount of computation in the process of optimizing the objective function.

[0118] Existing techniques directly assume that the covariance matrix of each Gaussian mixture model is the same and isotropic. However, the covariance of the real point cloud distribution does not satisfy the isotropic assumption. Therefore, this invention uses anisotropic covariance matrices and decomposes the covariance matrix as follows:

[0119]

[0120] The objective function can be transformed into:

[0121]

[0122] The SVD decomposition method is used to optimize and solve for R and t. and μ j Perform centroid removal:

[0123]

[0124]

[0125] The optimal rotation matrix can be obtained as follows:

[0126]

[0127] Ignoring terms irrelevant to R, the optimization problem becomes:

[0128]

[0129] Where W = diag(w1, w2, ..., w K ), Q=[q1,q2,…,q K ], P=[p1,p2,…,p K ]

[0130] Let S = PWQ T Perform SVD decomposition on S:

[0131] svd(S)=UΣV T

[0132] Then the rotation matrix R = VU T

[0133] To ensure that R is a rotation matrix and not a reflection matrix, and that R needs to satisfy the constraint that its determinant is 1, R is corrected as follows:

[0134] R = Vdiag(1,1,det(VU) T ))U T

[0135] The translation vector can be obtained from the rotation matrix:

[0136]

[0137] Repeat steps 4 and 5 until convergence. The convergence condition is set to reach the required number of iterations or the change in the rotation matrix between two generations is less than the set threshold.

[0138] In this example, the maximum number of convergence algebras is set to 200, the minimum threshold for rotation angle is set to 0.05°, and the minimum threshold for translation distance is set to 0.001m.

[0139] The optimal rotation matrix R is finally obtained by iterating through the EM algorithm until convergence. * With translation vector t * In this example, the final rotation matrix is:

[0140]

[0141] The translation vector is:

[0142]

[0143] The point cloud after registration is as follows Figure 2 As shown.

[0144] This example uses the RMSE of the two registered point clouds as a metric for registration accuracy:

[0145] RMSE represents the average distance between a point in the source point cloud and the nearest point in the target point cloud, and its definition is:

[0146]

[0147] Where I = min(M,N), p i ,q i The closest point pair between the source point cloud and the target point cloud.

[0148] In this example, the RMSE of the point cloud after registration was 0.261, and the registration time was 2.34 seconds. This verifies the accuracy and speed of the proposed method.

[0149] The units described in the embodiments of the present invention can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0150] According to one aspect of the present invention, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and executes the computer instructions, causing the computer device to perform the methods provided in the various optional implementations described above.

[0151] In another aspect, embodiments of the present invention also provide a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the methods described in the above embodiments.

[0152] In addition to the examples above, other embodiments may be obtained by those skilled in the art based on the above disclosure or by making modifications using knowledge or technology in related fields. The features of each embodiment may be interchanged or replaced. Modifications and changes made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for 3D point cloud registration of a scene based on a Gaussian mixture model, characterized in that, Includes the following steps: S1, collect point cloud data and process the point cloud data; S2, clusters the point cloud to generate a Gaussian mixture model; S3, ground point cloud removal is performed on the point cloud using the generated Gaussian mixture model; S4, based on the Gaussian mixture model obtained after step S3, construct the maximum likelihood function and generate the response matrix of the source point cloud sub-model from the point to be registered; step S4 includes the following sub-steps: Based on the Gaussian mixture model obtained in step S3, a maximum likelihood function is constructed, and the response matrix of the cloud sub-model from the point to be registered to the source point is generated. The expression of the constructed maximum likelihood function is as follows: in, It is a transformation matrix for the target point cloud, containing rotation matrices and translation vectors. R Represents the rotation matrix. t Represents the translation vector. This indicates the number of points in the target point cloud. Indicates the number of sub-models. This represents the covariance matrix of the k-th sub-model; The elements in the response matrix represent the posterior probability that a point in the target point cloud belongs to a sub-model in the mixture model, and are calculated by the following formula: In the formula, This represents the weight of the j-th sub-model. This represents the mean of the j-th sub-model. Let the covariance matrix of the j-th sub-model be denoted as . This represents the total number of sub-models; S5, based on the response matrix obtained in step S4, optimize the objective function by iteratively solving for the rotation matrix and translation vector; step S5 includes the following sub-steps: Based on the response matrix obtained in step S4, the objective function is optimized by iteratively solving for the rotation matrix and translation vector. After simplification, the objective function is transformed from the form of a probability density exponential function into the form of a Mahalanobis distance. By decoupling the model's parameters, the objective function is further simplified to: The point cloud is centroid-removed, and the rotation matrix is ​​solved using the SVD method to further obtain the translation vector; where, Indicates the total number of target point clouds. This indicates the weighted average. This represents the weighted mean of the model; S6. Repeat steps S4 and S5 until the convergence condition is met, that is, the optimal rotation matrix and translation vector are obtained. Use the obtained optimal rotation matrix and translation vector to transform the target point cloud, that is, complete the fast registration of the point cloud. Step S3 includes the following sub-steps: analyzing the covariance of the generated sub-models, considering those with a covariance z-axis component less than a certain threshold as ground point sub-models, removing points from the ground point models, and updating the weights of the non-ground sub-models.

2. The scene 3D point cloud registration method based on Gaussian mixture model according to claim 1, characterized in that, After step S6, the following steps are also included: running steps S1 to S6 on each frame of point cloud acquired, thereby completing the 3D reconstruction of the scene.

3. The scene 3D point cloud registration method based on Gaussian mixture model according to claim 1, characterized in that, Step S1 includes the following sub-steps: acquiring two frames of point cloud data at two positions and angles, and performing null point removal and downsampling processing on the original point cloud data.

4. The scene 3D point cloud registration method based on Gaussian mixture model according to claim 1, characterized in that, Step S2 includes the sub-step of clustering the two point clouds to generate Gaussian mixture models, specifically including the following sub-steps: Point clouds are clustered using k-means, and the cluster centers are used as initial values ​​for the mean of a Gaussian mixture model. The mean, covariance, and sub-model weights of the Gaussian mixture model are iteratively solved using the EM algorithm. The resulting probability density function of the Gaussian mixture model is expressed as: In the formula: in, K This indicates the number of sub-models in the Gaussian mixture model, and y represents the coordinates of the target point cloud. The model parameters of the k-th sub-model include: Indicates the first The mean of each sub-model Indicates the first The covariance matrix of each sub-model Indicates the first The weights of each sub-model.

5. The scene 3D point cloud registration method based on Gaussian mixture model according to claim 3, characterized in that, The acquisition of two frames of point cloud data at two different locations and angles is specifically achieved using lidar.

6. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing a computer program that, when loaded by the processor, executes the method as described in any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that, A computer program is stored in a readable storage medium, the computer program being loaded by a processor and executing the method as described in any one of claims 1 to 5.

Images