Selective up-sampling method based on indoor parking lot point cloud

Abstract

The invention provides a selective up-sampling method based on an indoor parking lot point cloud, and the method comprises the steps: inputting an original sparse point cloud into a general input preprocessing module, and carrying out the global normalization, fixed-size dicing, rarefaction and online enhancement operation, thereby obtaining a preprocessed sparse point cloud; and inputting the preprocessed sparse point cloud into a PU-Transform network structure based on a self-attention and local fusion mechanism to carry out final feature fusion and optimization, and outputting a dense point cloud with high resolution, structure fidelity and uniform distribution. According to the invention, high-quality up-sampling of sparse point clouds in an indoor parking lot scene is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional point cloud processing, and in particular to a selective upsampling method based on indoor parking lot point cloud. Background Art

[0002] Indoor parking lots are crucial infrastructure in smart city transportation systems. High-precision mapping and spatial structure perception are crucial for enabling automated parking, path planning, and intelligent navigation. However, due to complex parking environments, severe point cloud occlusion, reflective surface materials, and dramatic changes in spatial scale, traditional point cloud reconstruction and upsampling methods face significant challenges in these scenarios. This is particularly true for thin structures (such as parking space lines), local density variations, and detail continuity, making reconstruction accuracy and robustness difficult to guarantee.

[0003] Existing point cloud upsampling methods mostly rely on regular scenes or synthetic data for training, lacking model adaptation and data validation for real-world parking environments. This results in weak generalization and incomplete structural reconstruction in real-world deployments. Furthermore, traditional methods struggle to uniformly handle areas of varying sparsity, occlusion, and scale, failing to balance reconstruction quality with inference efficiency.

[0004] In recent years, with the development of deep learning technology in the field of point clouds, deep learning architectures have gradually been introduced into point cloud upsampling tasks. However, these methods still suffer from insufficient perception, structural ambiguity, and high computational costs when faced with the non-uniform distribution, multi-scale structure, and local noise of point clouds in real scenes.

[0005] Therefore, a global structure perception mechanism and local feature extraction, as well as a high-quality point cloud upsampling method that can handle complex indoor scenes, are needed to provide a higher quality data foundation for automatic parking, structure recognition, and three-dimensional map construction. Summary of the Invention

[0006] In view of this, the present invention is provided to at least solve the above technical problems.

[0007] According to a first aspect of an embodiment of the present invention, a selective upsampling method based on an indoor parking lot point cloud is provided, including: step S101, inputting the original sparse point cloud into a universal input preprocessing module, performing global normalization, fixed-size slicing, sparsification and online enhancement operations to obtain a preprocessed sparse point cloud; step S102, inputting the preprocessed sparse point cloud into a PU-Transformer network structure based on a self-attention and local fusion mechanism, performing final feature fusion and optimization, and outputting a dense point cloud with high resolution, structural fidelity and uniform distribution.

[0008] Optionally, step S101 specifically includes: global normalization: obtaining a normalized point cloud by performing centroid translation and unit sphere scaling on the original sparse point cloud; fixed-size blocking and sparsification: obtaining a downsampled point cloud by extracting local sub-blocks from the normalized point cloud and downsampling; online data enhancement: obtaining a preprocessed sparse point cloud by performing rotation, jitter and scaling operations on the downsampled point cloud.

[0009] Optionally, the step S102 specifically includes: a local feature fusion mechanism for relative position perception: by introducing the relative coordinate difference of neighbors and fusing it with the embedded features of the point itself, the Transformer's perception of local geometric structure is effectively enhanced; a Transformer encoder design for global context modeling: the Transformer encoder for global context modeling establishes long-range dependencies between points through a multi-head self-attention mechanism, effectively capturing the structural continuity and global geometric patterns in the point cloud; an efficient feature expansion mechanism based on channel rearrangement: by rearranging the point feature dimensions into multiple sub-features, the number of points is rapidly expanded, taking into account both upsampling efficiency and feature consistency; a patch-wise reasoning strategy and a full-scene point cloud reconstruction fusion mechanism: the large-scale point cloud is divided into processable small blocks for independent upsampling, and then integrated into a continuous point cloud through a fusion mechanism to output a high-resolution, structurally fidelity and uniformly distributed dense point cloud, effectively improving the reasoning efficiency and overall reconstruction quality.

[0010] According to a second aspect of an embodiment of the present invention, a selective upsampling system based on an indoor parking lot point cloud is provided, including: a universal input preprocessing module, which performs global normalization, fixed-size slicing, sparsification, and online enhancement operations on the input original sparse point cloud to obtain a preprocessed sparse point cloud; a PU-Transformer network structure based on self-attention and local fusion mechanism, which performs final feature fusion and optimization on the preprocessed sparse point cloud to output a high-resolution, structure-fidelity, and uniformly distributed dense point cloud.

[0011] According to a third aspect of an embodiment of the present invention, there is provided an electronic device comprising a processor and a memory storing a program, wherein the program comprises instructions that, when executed by the processor, cause the processor to perform the steps of the method according to the first aspect.

[0012] According to a fourth aspect of an embodiment of the present invention, a computer storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the method of the first aspect is implemented.

[0013] In summary, this invention can efficiently and stably achieve high-quality upsampling of sparse point clouds in indoor parking scenarios, effectively improving the density, structural integrity, and detail fidelity of the point clouds. Compared to existing upsampling methods based on a single network structure or synthetic data training, this invention, by constructing a unified preprocessing process and structural innovation model (PU-Transformer), demonstrates greater adaptability and robustness in the face of occlusion, noise, density mutations, and structural diversity in real scenes.

[0014] By incorporating an innovative self-attention mechanism, this method significantly improves the ability to restore thin structures (such as parking space lines), edge contours, and complex geometry, while balancing upsampling accuracy and inference efficiency. The resulting high-resolution, evenly distributed, and structurally continuous point cloud output can be widely used in tasks such as 3D segmentation, object detection, semantic mapping, and automated parking. It enhances the expressive power of high-precision indoor maps and provides reliable data support and engineering application value for intelligent transportation systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0016] Figure 1 Flow chart of the steps of the method of the present invention.

[0017] Figure 2 Schematic diagram of the improved PU-Transformer network structure. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0019] See also Figure 1 The present invention provides a selective upsampling method based on indoor parking lot point cloud, comprising:

[0020] S101, inputting the original sparse point cloud into the universal input preprocessing module, performing global normalization, fixed-size slicing, sparsification, and online enhancement operations to obtain a preprocessed sparse point cloud;

[0021] S102: The pre-processed sparse point cloud is input into the PU-Transformer network structure based on the self-attention and local fusion mechanism for final feature fusion and optimization, and a high-resolution, structurally fidelity, and evenly distributed dense point cloud is output.

[0022] Optionally, step S101 specifically includes: global normalization: obtaining a normalized point cloud by performing centroid translation and unit sphere scaling on the original sparse point cloud; fixed-size blocking and sparsification: obtaining a downsampled point cloud by extracting local sub-blocks from the normalized point cloud and downsampling; online data enhancement: obtaining a preprocessed sparse point cloud by performing rotation, jitter and scaling operations on the downsampled point cloud.

[0023] Optionally, the step S101 specifically includes: a local feature fusion mechanism for relative position perception: by introducing the relative coordinate difference of neighbors and fusing it with the embedded features of the point itself, the Transformer's perception of local geometric structure is effectively enhanced; a Transformer encoder design for global context modeling: the Transformer encoder for global context modeling establishes long-range dependencies between points through a multi-head self-attention mechanism, effectively capturing the structural continuity and global geometric patterns in the point cloud; an efficient feature expansion mechanism based on channel rearrangement: by rearranging the point feature dimensions into multiple sub-features, the number of points can be rapidly expanded, taking into account both upsampling efficiency and feature consistency; a patch-wise reasoning strategy and a full-scene point cloud reconstruction fusion mechanism: the large-scale point cloud is divided into processable small blocks for independent upsampling, and then integrated into a continuous point cloud through a fusion mechanism to output a high-resolution, structurally fidelity and uniformly distributed dense point cloud, effectively improving the reasoning efficiency and overall reconstruction quality.

[0024] In summary, this invention can efficiently and stably achieve high-quality upsampling of sparse point clouds in indoor parking scenarios, effectively improving the density, structural integrity, and detail fidelity of the point clouds. Compared to existing upsampling methods based on a single network structure or synthetic data training, this invention, by constructing a preprocessing process and an innovative structural model (PU-Transformer), demonstrates greater adaptability and robustness in the face of occlusion, noise, density mutations, and structural diversity in real scenes.

[0025] By incorporating an innovative self-attention mechanism, this method significantly improves the ability to restore thin structures (such as parking space lines), edge contours, and complex geometry, while balancing upsampling accuracy and inference efficiency. The resulting high-resolution, evenly distributed, and structurally continuous point cloud output can be widely used in tasks such as 3D segmentation, object detection, semantic mapping, and automated parking. It enhances the expressive power of high-precision indoor maps and provides reliable data support and engineering application value for intelligent transportation systems.

[0026] Specifically, the solution of the present invention is further described according to the following examples:

[0027] This paper proposes a selective upsampling method based on indoor parking lot point clouds, combining deep learning structure optimization with high-precision data construction, which is widely used in urban space understanding tasks such as object detection, 3D segmentation, semantic mapping, and intelligent navigation. The present invention adopts the following technical solutions:

[0028] A selective upsampling method based on an indoor parking lot point cloud according to an embodiment of the present invention includes:

[0029] S1. The general input preprocessing module standardizes the scale, structure and enhancement strategy of the sparse point cloud, providing a consistent, standardized and robust input basis for the densification model;

[0030] S2. PU-Transformer network structure innovation and point cloud upsampling optimization design based on self-attention and local fusion mechanism.

[0031] Furthermore, step S1 specifically includes the following sub-steps:

[0032] S11. The universal input preprocessing module provides a stable and consistent input basis for various densification networks by standardizing the scale, structure and enhancement strategy of sparse point clouds, thereby improving training effects and model robustness.

[0033] To improve the robustness of our model to varying scanning conditions, density variations, and noise perturbations in indoor parking lot point clouds, we employ a preprocessing process within the network architecture (PU-Transformer). This module not only standardizes the model's input structure but also serves as a key innovation throughout the entire process, improving training stability and enhancing generalization capabilities.

[0034] In the specific preprocessing stage, the original point cloud is subjected to global normalization, patch cutting, sparse sampling and online enhancement operations, aiming to unify the scale, simulate the sparsity of real sensors, and improve the model's adaptability to rotational disturbances, noise interference and scale changes.

[0035] S12. Global normalization unifies the spatial scale and position distribution of the point cloud through centroid translation and unit sphere scaling, providing a stable input basis for subsequent feature extraction and densification modeling.

[0036] Calculate the point cloud centroid:

[0037]

[0038] Calculate the maximum Euclidean distance from all points to the centroid:

[0039]

[0040] Normalize each point and map it to the unit sphere:

[0041]

[0042] Global normalization effectively unifies the spatial scale and positional distribution of the input point cloud by calculating the point cloud's centroid and scaling all points to fit within a unit sphere. This step eliminates scale and displacement differences between scanned regions, providing a standardized and stable geometric foundation for subsequent feature extraction and densification modeling. It is an essential pre-process in the entire point cloud processing workflow.

[0043] S13, fixed-size blocking and sparsification achieves consistency in local structure modeling by randomly extracting local sub-blocks containing 1024 points from the normalized point cloud and uniformly downsampling them to 256-point input, while effectively simulating the sparse sampling characteristics of real lidar.

[0044] Randomly crop local blocks from the normalized point cloud, each block contains:

[0045] N patch =1024

[0046] Uniformly downsample the local block as network input:

[0047]

[0048] Fixed-size tiling and sparsification: Randomly extract local sub-blocks of 1024 points from the normalized point cloud and uniformly downsample them to 256 points. This process not only ensures scale consistency in local structure modeling but also effectively simulates the sampling characteristics of real lidar under spatial sparsity conditions.

[0049] S14. Online data enhancement effectively improves the model's robustness to direction changes, sensor noise, and scale differences by applying random rotation, coordinate jitter, and scale scaling to the input point cloud during the training phase.

[0050] Random rotation around the vertical axis (simulating inconsistent scan directions):

[0051] p i (1) =R Z (θ)·p i (0) ,θ~U(-π,π)

[0052]

[0053] Coordinate jitter (simulating sensor noise):

[0054] pi (2) =p i (1) +ε,ε~N(0,σ 2 I),σ=0.01m

[0055] Random scaling (simulating structural scale differences):

[0056] p i (aug) =s·p i (2) ,s~U[0.8,1.2]

[0057] Online data enhancement enhances the model's robustness to orientation changes, noise interference, and scale differences through random rotation, coordinate jittering, and scale scaling operations, making it more adaptable in diverse point cloud scenarios.

[0058] This process exposes the model to diverse point distributions, directional perturbations, and sampling density variations during training, maintaining good performance under occlusion, sparse, and noisy conditions. In the model proposed in this paper (PU-Transformer), this preprocessing module is adopted as the standard input process.

[0059] Furthermore, step S2 specifically includes the following sub-steps:

[0060] See also Figure 2 Schematic diagram of the improved PU-Transformer network structure.

[0061] S21. The relative position-aware local feature fusion mechanism effectively enhances the Transformer's ability to perceive local geometric structures by introducing the relative coordinate difference of neighbors and fusing it with the embedded features of the point itself;

[0062] In order to improve the geometric perception ability of Transformer in sparse point clouds, this paper proposes a PositionalFusion module. To embed:

[0063]

[0064] Then calculate the relative positions of its k neighbors:

[0065]

[0066] And then through another MLP mapping and max pooling aggregation into local descriptors:

[0067]

[0068] Finally, the original embedding is fused with the local structure information:

[0069] f i =MLP3([f i (0) ||g i ])

[0070] This structure significantly enhances the ability of each point to model its local geometric contour and is an important supplement to the original Transformer smooth attention mechanism.

[0071] S22, the Transformer encoder for global context modeling establishes long-range dependencies between points through a multi-head self-attention mechanism, effectively capturing structural continuity and global geometric patterns in point clouds;

[0072] The present invention uses a standard Transformer encoder with 1=5 layers, each layer containing multi-head self-attention and feed-forward sublayers. The attention mechanism is as follows:

[0073]

[0074] where q i , k j , v j The query, key, and value vectors are obtained by linear mapping. Self-attention helps points establish global associations with other points and learn structural relationships such as plane extension, edge alignment, and symmetry.

[0075] Each layer is followed by a residual connection and LayerNorm to ensure training stability and gradient flow:

[0076] f i (l+1) =LayerNorm(f i (l) +FFN(Attention i ))

[0077] S23, an efficient feature expansion mechanism based on channel rearrangement, achieves rapid expansion of the number of points by rearranging the point feature dimension into multiple sub-features, taking into account both upsampling efficiency and feature consistency;

[0078] Transformer outputs an N×D feature map (e.g. D=256), which we rearrange into rN sub-features to complete 4× upsampling

[0079] The channel rearrangement rules are as follows:

[0080]

[0081] Each sub-feature is then mapped to a 3D coordinate by a shared MLP:

[0082]

[0083] Finally, rN upsampling points are generated, which maintains structural consistency and has low computational overhead, significantly outperforming the repeated replication strategy.

[0084] S24, Patch-wise reasoning strategy divides large-scale point clouds into small, processable blocks for independent upsampling, and then integrates them into continuous point clouds through a fusion mechanism, effectively improving reasoning efficiency and overall reconstruction quality.

[0085] For large-scale point cloud scenarios such as parking lots, this paper adopts a patch-wise reasoning approach. It processes only one local patch (256 or 1024 points) at a time, completing feature extraction and upsampling within the GPU's controllable memory, making it suitable for distributed or incremental reasoning processes.

[0086] The results of multiple overlapping patches are merged in the post-processing stage through farthest point sampling (FPS) or overlapping area average fusion method to avoid redundancy and dense aggregation:

[0087] P final =Merge(P1,P2,P3,…,P M ), M is the number of patches

[0088] This strategy achieves consistent point distribution output at the entire scene level while maintaining high-quality local reconstruction, making it suitable for deployment in complex indoor environments.

[0089] In summary, the present invention discloses a selective upsampling method for indoor parking lot point clouds, which covers an input preprocessing framework and a structural innovation model (PU-Transformer). It also constructs a measured high-precision dataset for typical scenes such as planes, slender parking lines, occluding columns, and density mutations for the first time, filling the generalization gap of existing synthetic data in complex environments. The method first standardizes the scale and structure of the input point cloud through global normalization, slicing, sparsification, and data enhancement, providing a stable and consistent training basis for the model; in terms of model design, PU-Transformer integrates relative position perception and global self-attention mechanism, and improves the processing efficiency of large scenes through channel rearrangement upsampling and Patch-wise reasoning. Ultimately, the method can output high-resolution, structurally fidelity, and uniformly distributed point cloud results, which are suitable for a variety of downstream tasks such as target detection, three-dimensional segmentation, and navigation mapping.

[0090] An embodiment of the present invention further provides a selective upsampling system based on an indoor parking lot point cloud, comprising:

[0091] The general input preprocessing module performs global normalization, fixed-size slicing, sparsification, and online enhancement operations on the input original sparse point cloud to obtain the preprocessed sparse point cloud;

[0092] Based on the PU-Transformer network structure of self-attention and local fusion mechanism, the final feature fusion and optimization are performed on the pre-processed sparse point cloud, outputting a dense point cloud with high resolution, structural fidelity and uniform distribution.

[0093] It should be understood that the system of this embodiment is used to implement the corresponding methods in the aforementioned multiple method embodiments and has the beneficial effects of the corresponding method embodiments.

[0094] As another example, the present invention also provides an electronic device, which will now be described as an electronic device that can serve as a server or client of the present invention, which is an example of a hardware device that can be applied to various aspects of the present invention. The electronic device is intended to represent various forms of digital electronic computer equipment, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0095] The electronic device may include: a processor, a communication interface, a memory, and a communication bus.

[0096] The processor, communication interface and memory communicate with each other through a communication bus. The communication interface is used to communicate with other electronic devices or servers.

[0097] The processor is used to execute programs, and specifically can execute the relevant steps in the above method embodiments.

[0098] Specifically, the program may include program codes including computer operation instructions.

[0099] The processor may be a CPU, an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present invention. The one or more processors included in the smart device may be processors of the same type, such as one or more CPUs, or processors of different types, such as one or more CPUs and one or more ASICs.

[0100] The memory is used to store programs and may include high-speed RAM memory or non-volatile memory, such as at least one disk storage.

[0101] When executed by a processor, the program enables the electronic device to perform a selective upsampling method based on an indoor parking lot point cloud according to the present invention.

[0102] In addition, the specific implementation of each step in the program can refer to the corresponding description of the corresponding steps and units in the above method embodiments, and will not be repeated here. Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working process of the above-described devices and modules can refer to the corresponding process description in the above method embodiments, and will not be repeated here.

[0103] An exemplary embodiment of the present invention further provides a computer storage medium storing a computer program, wherein when the computer program is executed by a processor, the methods of the various embodiments of the present invention are implemented. The corresponding process descriptions in the aforementioned method embodiments can be referred to and will not be repeated here.

[0104] The method according to the embodiment of the present invention described above can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD ROM, RAM, floppy disk, hard disk or magneto-optical disk), or as computer code that is originally stored in a remote recording medium or a non-temporary machine-readable medium downloaded via a network and will be stored in a local recording medium, so that the method described herein can be stored in such software processing on a recording medium using a general-purpose computer, a dedicated processor or programmable or dedicated hardware (such as an ASIC or FPGA). It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code, and when the software or computer code is accessed and executed by a computer, a processor or hardware, the method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the method shown here, the execution of the code converts the general-purpose computer into a dedicated computer for executing the method shown here.

[0105] Thus far, specific embodiments of the present invention have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing may be advantageous.

[0106] It should be understood that although this specification is described according to various embodiments, not every embodiment contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

[0107] Finally, it should be noted that the above implementation methods are only used to illustrate the embodiments of the present invention, and are not limitations on the embodiments of the present invention. Ordinary technicians in the relevant technical field can make various changes and modifications without departing from the spirit and scope of the embodiments of the present invention. Therefore, all equivalent technical solutions also fall within the scope of the embodiments of the present invention, and the scope of patent protection of the embodiments of the present invention should be defined by the claims.

Claims

1. A selective upsampling method based on indoor parking lot point cloud, characterized in that: include: Step S101: Input the original sparse point cloud into the universal input preprocessing module, perform global normalization, fixed-size dicing, thinning, and online enhancement operations to obtain a preprocessed sparse point cloud; Step S102: Input the preprocessed sparse point cloud into the PU-Transformer network structure based on self-attention and local fusion mechanism to perform final feature fusion and optimization, and output a dense point cloud with high resolution, structural fidelity and uniform distribution.

2. The method according to claim 1, characterized in that The step S01 specifically includes: Global normalization: The normalized point cloud is obtained by performing centroid translation and unit sphere scaling on the original sparse point cloud. Fixed-size patching and sparsification: extract local sub-blocks from the normalized point cloud and downsample them to obtain the downsampled point cloud; Online data enhancement: The preprocessed sparse point cloud is obtained by rotating, jittering and scaling the downsampled point cloud.

3. The method according to claim 1, characterized in that The step S102 specifically includes: Relative position-aware local feature fusion mechanism: By introducing the relative coordinate difference of neighbors and fusing it with the embedded features of the point itself, it effectively enhances the Transformer's ability to perceive local geometric structures; Transformer encoder design for global context modeling: The Transformer encoder for global context modeling establishes long-range dependencies between points through a multi-head self-attention mechanism, effectively capturing structural continuity and global geometric patterns in point clouds; Efficient feature expansion mechanism based on channel rearrangement: By rearranging the point feature dimensions into multiple sub-features, the number of points can be rapidly expanded, balancing upsampling efficiency and feature consistency. Patch-wise reasoning strategy and full-scene point cloud reconstruction fusion mechanism: Divide large-scale point clouds into small, processable blocks and independently upsample them. Then, integrate them into a continuous point cloud through a fusion mechanism to output a high-resolution, structurally fidelity, and evenly distributed dense point cloud, effectively improving reasoning efficiency and overall reconstruction quality.

4. A selective upsampling system based on indoor parking lot point cloud, characterized by: include: The general input preprocessing module performs global normalization, fixed-size slicing, sparsification, and online enhancement operations on the input original sparse point cloud to obtain the preprocessed sparse point cloud; Based on the PU-Transformer network structure of self-attention and local fusion mechanism, the final feature fusion and optimization are performed on the pre-processed sparse point cloud, outputting a dense point cloud with high resolution, structural fidelity and uniform distribution.

5. An electronic device, characterized in that: include: processor; Memory for storing programs; The program includes instructions, which, when executed by the processor, cause the processor to perform the steps of the method according to any one of claims 1 to 3.

6. A computer storage medium, characterized in that A computer program is stored thereon, and when the program is executed by a processor, the method according to any one of claims 1 to 3 is implemented.

Images