Three-dimensional point cloud target layered accurate extraction method based on multi-feature fusion

Abstract

The invention discloses a three-dimensional point cloud target layering accurate extraction method based on multi-feature fusion, and aims to improve the precision and anti-noise capability of target object extraction and effectively restore detail features. The method comprises the following steps of data preprocessing, hierarchical processing, feature extraction, AOI extraction and restoration processing. By adopting voxel-based downsampling and combining hierarchical processing and threshold screening methods, noise points of non-target objects are effectively distinguished and removed, and efficient extraction of target object point clouds in a complex scene is realized; in order to ensure the detail integrity of the target object, a point cloud restoration step is designed, detail restoration is carried out on the target object after crude extraction in an index search and neighborhood query mode, and subtle information possibly lost in downsampling is supplemented; according to the innovation, the target object is accurately extracted while the miscellaneous points are efficiently removed, the point cloud extraction precision of the target object is greatly improved, and the point cloud data processing requirements in a complex environment are met.

Description

Technical Field

[0001] The present invention relates to the technical field of point cloud target extraction, and in particular to a hierarchical and precise extraction method of three-dimensional point cloud targets based on multi-feature fusion. Background Art

[0002] Point cloud data is a data representation method that describes the surface of an object in three-dimensional space. In the fields of scene reconstruction, equipment inspection, and defect monitoring, three-dimensional point cloud data has been widely used because it can fully and accurately describe the spatial structure. With the development of LiDAR and point cloud technology, the amount of collected point cloud data has increased dramatically, containing a large amount of spatial information. In particular, point cloud data in complex scenes usually contains a large amount of non-target information, which makes it a challenging task to accurately extract target point clouds from massive point clouds. Multi-feature extraction of target point clouds is one of the core technologies of point cloud processing, which aims to accurately identify and segment the point cloud information of specific target objects from complex backgrounds. In recent years, point cloud target extraction technology has played a key role in the fields of autonomous driving and robot navigation. Especially in the field of industrial automation inspection, extracting the various components of target objects from large-scale disordered point cloud data provides important support for tasks such as quality inspection and defect identification. However, due to the complex spatial structure and noise interference of point cloud data, effective target point cloud extraction still faces great challenges.

[0003] In terms of target point cloud extraction, there are mainly methods based on geometric features, methods based on deep learning, and methods based on multi-feature fusion. Among them, the method based on geometric features extracts local features such as normal vectors and curvature of point clouds for segmentation, but has poor robustness when processing noisy point clouds; although the method based on deep learning has strong feature expression capabilities, it requires a large amount of labeled data for training and has high computational overhead. These methods generally have problems such as low accuracy and low computational efficiency when processing point cloud data in complex scenes, and lack the ability to perform multi-level analysis of target objects, making it difficult to meet actual application needs.

[0004] Feature fusion technology has important application value in point cloud data processing because of its advantages such as being able to comprehensively utilize the multi-dimensional feature information of point clouds and improve the accuracy of target recognition. However, traditional feature fusion methods still have the following problems when processing target point cloud data: first, it is difficult to effectively process point cloud data with significant differences in feature distribution; second, there is a lack of a reasonable hierarchical extraction mechanism, which cannot reflect the hierarchical structural characteristics of the target object; third, the feature fusion process is computationally complex and it is difficult to meet the real-time requirements of large-scale point cloud data processing. These problems seriously restrict the actual application effect of feature fusion methods in target point cloud extraction.

[0005] Compared with the traditional feature fusion method, by introducing a hierarchical extraction strategy and combining it with the fusion analysis of multi-dimensional features, it can better adapt to the feature change law of target point clouds at different levels, thereby improving the accuracy and robustness of target point cloud extraction. The solution of the present invention is proposed. Summary of the invention

[0006] The purpose of the present invention is to provide a three-dimensional point cloud target hierarchical accurate extraction method based on multi-feature fusion to solve the problems raised in the above background technology.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] A hierarchical and precise extraction method of three-dimensional point cloud targets based on multi-feature fusion, the method includes:

[0009] S100, downsampling the original point cloud data, and simplifying the data using a voxel grid downsampling method; removing ground points in the point cloud using a ground filtering algorithm based on cloth simulation;

[0010] S200, setting an elevation threshold, and dividing the point cloud after removing the ground points of the point cloud into an upper and lower layer;

[0011] S300, extracting features from the upper and lower layers of point clouds, and setting relevant feature thresholds to remove point clouds irrelevant to the target;

[0012] S400, based on the point cloud after eliminating the irrelevant points, respectively performing target positioning and target clustering;

[0013] S500 , performing a neighborhood query on the original point cloud data based on the target point cloud data after the target object is clustered, and restoring the point cloud near the target object.

[0014] Preferably, S100 comprises:

[0015] S101. Use voxel grid downsampling method to simplify data:

[0016] Obtaining original point cloud data, calculating the bounding box of the point cloud data, and discretizing the bounding box into small voxels, the length, width and height of the voxels can be freely set according to the desired downsampling degree;

[0017] Get the point set that falls in each voxel, and take a point closest to the voxel center in each voxel as a sampling point to replace the original point set;

[0018] S102. Remove ground points from point cloud using ground filtering algorithm based on cloth simulation:

[0019] Based on the downsampled point cloud data, the point cloud is inverted and the size and position of the initial cloth grid are determined;

[0020] Project the inverted point cloud and grid particles to the same horizontal plane, find the nearest neighbor point cp of each particle, and record its elevation IHV before projection;

[0021] For each movable grid particle, calculate its displacement X(t+Δt) due to gravity:

[0022]

[0023] Where m is the mass of the particle, Δt is the time step, and G is a constant;

[0024] Compare X(t) with the IHV of the cp point corresponding to the current particle; if the height of the particle is less than or equal to IHV, set the height of the particle to IHV and set it as an immovable point;

[0025] For each grid particle, calculate the displacement caused by the internal driving factor

[0026]

[0027] Where b is set to 1 when the particle is movable, otherwise it is 0. Respectively represent the current position vectors of a particle and its neighboring particles, represents the unit vector in the vertical direction;

[0028] Repeat the above displacement calculation steps until the maximum elevation change of all particles is small enough, or the number of iterations reaches a preset value, then terminate the simulation process;

[0029] The height difference between the point cloud and the grid particles is calculated, and the ground points and non-ground points are distinguished according to the pre-set threshold.

[0030] Preferably, S200 includes:

[0031] S201, point cloud regular grid division:

[0032] The preprocessed point cloud space is divided into a grid of M×N×1. The regular grid of the point cloud is divided into:

[0033]

[0034] Among them, x min 、x max are the minimum and maximum x-coordinates of the preprocessed point cloud data; y min ,y max They are the minimum and maximum values ​​of the y coordinates in the preprocessed point cloud data;

[0035] Get the x-coordinate and y-coordinate of a point i in the point cloud data, and calculate the row and column number of the point in the grid:

[0036]

[0037] Among them, (x i ,y i ) is the coordinate value of point i; m and n are the row and column numbers of point i in the regular grid;

[0038] S202. Statistical analysis:

[0039] The lowest point elevation in the grid is defined as the DEM feature, the highest point elevation in the grid is defined as the DSM feature, and the maximum height information of the objects in the grid compared to the ground is defined as the nDSM feature. DSM, DEM, and nDSM can be expressed as follows:

[0040]

[0041] Where z(m,n,i) is the z coordinate of the i-th point in the m-th row and n-th column grid;

[0042] S203. Set the elevation threshold to split the point cloud into two layers:

[0043] According to the statistical analysis of nDSM, an appropriate height difference threshold is set to segment the preprocessed point cloud into upper and lower layers.

[0044] Preferably, S300 includes:

[0045] S301, upper point cloud feature extraction:

[0046] Traverse all points in the upper point cloud, calculate the covariance matrix of its neighborhood point set, and then calculate the eigenvalues ​​λ1, λ2, λ3 of the covariance matrix (λ1≥λ2≥λ3≥0), and calculate the linearity f according to the eigenvalues l :

[0047] S302, setting a linearity threshold according to the linearity:

[0048] According to the statistical analysis of linearity, an appropriate linearity threshold is set to remove non-target point clouds in the upper point cloud;

[0049] S303, lower layer point cloud feature extraction:

[0050] Traverse all points in the lower point cloud and calculate the neighborhood density feature d of each point s :

[0051] Among them, n s is the number of neighborhood points in the sphere neighborhood, rs is the radius of the sphere neighborhood; after calculating the density of all points, the eigenvalues ​​need to be normalized to the [0,1] interval;

[0052] S304, setting density feature threshold: according to the statistical analysis of the normalized density features, setting a suitable density feature threshold, and removing non-target points in the lower layer point cloud.

[0053] Preferably, S400 includes:

[0054] S401, Target Positioning:

[0055] Based on the point cloud after non-target points have been removed, the elevation threshold is set according to the elevation characteristics of the target object, and the grids that meet the requirements are retained;

[0056] S402, target object clustering:

[0057] Traverse each retained point and search its neighborhood. If the number of neighborhood points is greater than the set threshold, it is considered as a core point, and the neighborhood points of the core point are considered as density reachable points.

[0058] Traverse each core point and find all clusters by checking its density reachable points; set a threshold based on the elevation characteristics of the target object and delete the clusters that do not meet the requirements.

[0059] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for hierarchical and precise extraction of three-dimensional point cloud targets based on multi-feature fusion.

[0060] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in a method for hierarchical and precise extraction of three-dimensional point cloud targets based on multi-feature fusion are implemented.

[0061] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0062] The present invention can accurately extract the target object while efficiently removing clutter, greatly improving the accuracy of point cloud extraction of the target object, adapting to the needs of point cloud data processing in complex environments, and providing accurate and reliable data support for equipment inspection and structural monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0064] Figure 1It is a method flow chart of the hierarchical accurate extraction method of three-dimensional point cloud target based on multi-feature fusion of the present invention;

[0065] Figure 2 is a point cloud schematic diagram of original data in an embodiment of the present invention;

[0066] Figure 3 is a schematic diagram of a point cloud after voxel grid downsampling in an embodiment of the present invention;

[0067] Figure 4 is a schematic diagram of a point cloud after the ground is removed in an embodiment of the present invention;

[0068] Figure 5 is a schematic diagram of a point cloud after layered processing in an embodiment of the present invention;

[0069] Figure 6 is a schematic diagram of a point cloud extracted by AOI in an embodiment of the present invention;

[0070] Figure 7 Schematic diagram of the point cloud after restoration processing in an embodiment of the present invention. DETAILED DESCRIPTION

[0071] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0072] The following is combined with Figure 1-7 The specific implementation mode of the present invention is described in detail with Example 1:

[0073] Embodiment 1: The present invention provides the following technical solution: a three-dimensional point cloud target layered precise extraction method based on multi-feature fusion, which can accurately extract the target object while efficiently removing noise, greatly improving the accuracy of target object point cloud extraction, adapting to the point cloud data processing needs in complex environments, and providing accurate and reliable data support for equipment inspection and structural monitoring.

[0074] like Figure 1 As shown, the flowchart of the method of the present invention comprises the following steps:

[0075] S100, data preprocessing;

[0076] like Figure 2 , which is a schematic diagram of a point cloud of original data in an embodiment of the present invention;

[0077] Downsampling of voxel grid; calculating the bounding box of the point cloud, discretizing the bounding box into small voxels, the length, width and height of the voxel can be freely set according to the desired downsampling degree; obtaining the point set falling in each voxel, taking a point closest to the center of the voxel in each voxel as the sampling point, replacing the original point set;

[0078] like Figure 3 The figure shows a schematic diagram of the point cloud after downsampling of the voxel grid. It is obvious that the point cloud data is significantly reduced after downsampling, but the effective features of the scene are still retained;

[0079] Ground filtering based on cloth simulation; inverting the point cloud and initializing the size and position of the cloth grid;

[0080] Project the inverted point cloud and grid particles to the same horizontal plane, find the nearest neighbor point cp of each particle, and record its elevation IHV before projection;

[0081] For each movable grid particle, calculate its displacement X(t+Δt) caused by gravity, and solve the equation:

[0082]

[0083] Where m is the mass of the particle (usually set to 1), Δt is the time step, and G is a constant;

[0084] And compare X(t) with the IHV of the cp point corresponding to the current particle; if the height of the particle is less than or equal to IHV, set the height of the particle to IHV and set it as an immovable point;

[0085] For each grid particle, calculate the displacement (vector) caused by the internal driving factor It is expressed as:

[0086]

[0087] Where b is set to 1 when the particle is movable, otherwise it is 0. Respectively represent the current position (coordinate) vectors of a particle and its neighboring particles, represents the unit vector in the vertical direction;

[0088] Repeat the above displacement calculation steps until the maximum elevation change of all particles is small enough, or the number of iterations reaches a preset value, then terminate the simulation process;

[0089] The height difference between the point cloud and the grid particles is calculated, and the ground points and non-ground points are distinguished according to the pre-set threshold.

[0090] like Figure 4The figure shows the point cloud after ground removal. It is obvious that the ground points are removed and the non-ground point cloud is retained.

[0091] S200, layered processing;

[0092] Point cloud regular grid division: The preprocessed point cloud space is divided into M×N×1 grids. The point cloud regular grid division is:

[0093]

[0094] In the formula, x min 、x max are the minimum and maximum x-coordinates of the preprocessed point cloud data; y min ,y max are the minimum and maximum y coordinates of the preprocessed point cloud data. Based on the x and y coordinates of point i, the row and column number of the point in the grid can be calculated:

[0095]

[0096] In the formula, (x i ,y i ) is the coordinate value of point i; m and n are the row and column numbers of point i in the regular grid;

[0097] Statistical analysis: The lowest point elevation in the grid is defined as the DEM feature, the highest point elevation in the grid is defined as the DSM feature, and the maximum height information of the objects in the grid compared to the ground is defined as the nDSM feature. DSM, DEM, and nDSM can be expressed as follows:

[0098]

[0099] Where z(m,n,i) is the z coordinate of the i-th point in the m-th row and n-th column grid;

[0100] Set the elevation threshold; according to the statistical analysis of nDSM, set the appropriate height difference threshold to split the preprocessed point cloud into two layers.

[0101] S300, feature extraction;

[0102] Extract features from the upper point cloud; traverse all points in the upper point cloud, calculate the covariance matrix of its neighborhood point set, and then calculate the eigenvalues ​​λ1, λ2, λ3 (λ1≥λ2≥λ3≥0) of the covariance matrix, and calculate the linearity f according to the eigenvalues l , expressed as:

[0103]

[0104] Set the linearity threshold; according to the statistical analysis of linearity, set the appropriate linearity threshold to remove non-target point clouds in the upper point cloud;

[0105] Extract the features of the lower point cloud; traverse all the points in the lower point cloud and calculate the neighborhood density feature d of each point s , expressed as:

[0106] Where n s is the number of neighborhood points in the sphere neighborhood, r s is the radius of the sphere neighborhood; after calculating the density of all points, the eigenvalues ​​need to be normalized to the [0,1] interval;

[0107] Set the density feature threshold; according to the statistical analysis of the normalized density features, set the appropriate density feature threshold to remove non-target points in the lower point cloud.

[0108] like Figure 5 The figure shows a schematic diagram of the point cloud after layered processing. It is obvious that the non-target object points with high linearity are removed from the upper layer point cloud, and the non-target object points with high density features are removed from the lower layer point cloud, and the point cloud of the target object is completely retained.

[0109] S400, AOI extraction;

[0110] Target positioning: Based on the point cloud after non-target points have been removed, the elevation threshold is set according to the elevation characteristics of the target object, and the grids that meet the requirements are retained;

[0111] Clustering of objects; traversing each retained point, searching the neighborhood of the point. If the number of neighborhood points is greater than the set threshold, it is considered as a core point, and the neighborhood points of the core point are considered as density-reachable points.

[0112] Traverse each core point and find all clusters by checking its density reachable points; set a threshold based on the elevation characteristics of the target object and delete the clusters that do not meet the requirements;

[0113] like Figure 6 The figure shows a schematic diagram of the point cloud extracted by AOI. It is obvious that the grids containing the point clouds related to the target object are retained and marked with different colors, while other non-target point cloud data are removed.

[0114] S500, restoration processing;

[0115] The point cloud near the target object is restored by querying the neighborhood of the original point cloud data with the extracted target point cloud data.

[0116] like Figure 7The figure shows a schematic diagram of the point cloud after restoration processing. It can be clearly seen that the data of the point cloud related to the target object is completely restored and the original detailed features are retained.

[0117] Embodiment 2:

[0118] The computer-readable storage medium of this embodiment stores a computer program thereon, which, when executed by a processor, implements the steps of a method for hierarchical and precise extraction of three-dimensional point cloud targets based on multi-feature fusion in Embodiment 1.

[0119] The computer-readable storage medium of this embodiment may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal; the computer-readable storage medium of this embodiment may also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card, a secure digital card, a flash memory card, etc. equipped on the terminal; further, the computer-readable storage medium may also include both an internal storage unit of the terminal and an external storage device.

[0120] The computer-readable storage medium of this embodiment is used to store computer programs and other programs and data required by the terminal. The computer-readable storage medium can also be used to temporarily store data that has been output or is to be output.

[0121] Embodiment 3:

[0122] The computer device of this embodiment includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of a method for hierarchical and precise extraction of three-dimensional point cloud targets based on multi-feature fusion in Embodiment 1 are implemented.

[0123] In this embodiment, the processor may be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, readily available programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0124] Those skilled in the art will appreciate that the disclosed content of the embodiments may be provided as methods, systems, or computer program products. Therefore, the present solution may adopt the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Moreover, the present solution may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program codes.

[0125] The present solution is described with reference to the method according to the embodiment of the present solution and the flowchart and / or block diagram of the computer program product. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions; these computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 one or more processes and / or methods Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0126] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 one or more processes and / or methods Figure 1 A function specified in one or more boxes.

[0127] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 one or more processes and / or methods Figure 1 A step that specifies a function in one or more boxes.

[0128] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).

[0129] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A hierarchical and precise extraction method of three-dimensional point cloud targets based on multi-feature fusion, characterized by: The method comprises: S100, downsampling the original point cloud data, and simplifying the data using a voxel grid downsampling method; removing ground points in the point cloud using a ground filtering algorithm based on cloth simulation; S200, setting an elevation threshold, and dividing the point cloud after removing the ground points of the point cloud into an upper and lower layer; S300, extracting features from the upper and lower layers of point clouds, and setting relevant feature thresholds to eliminate point clouds irrelevant to the target; S400, based on the point cloud after eliminating the irrelevant points, respectively performing target positioning and target clustering; S500 , performing a neighborhood query on the original point cloud data based on the target point cloud data after the target object is clustered, and restoring the point cloud near the target object.

2. The method for accurate layered extraction of three-dimensional point cloud targets based on multi-feature fusion according to claim 1, characterized in that: The S100 includes: S101. Use voxel grid downsampling method to simplify data: Obtaining original point cloud data, calculating the bounding box of the point cloud data, and discretizing the bounding box into small voxels, the length, width and height of the voxels can be freely set according to the desired downsampling degree; Get the point set that falls in each voxel, and take a point closest to the voxel center in each voxel as a sampling point to replace the original point set; S102. Remove ground points from point cloud using ground filtering algorithm based on cloth simulation: Based on the downsampled point cloud data, the point cloud is inverted and the size and position of the initial cloth grid are determined; Project the inverted point cloud and grid particles to the same horizontal plane, find the nearest neighbor point cp of each particle, and record its elevation IHV before projection; For each movable grid particle, calculate its displacement X(t+Δt) due to gravity: Where m is the mass of the particle, Δt is the time step, and G is a constant; Compare X(t) with the IHV of the cp point corresponding to the current particle; if the height of the particle is less than or equal to IHV, set the height of the particle to IHV and set it as an immovable point; For each grid particle, calculate the displacement caused by the internal driving factor Where b is set to 1 when the particle is movable, otherwise it is 0. Respectively represent the current position vectors of a particle and its neighboring particles, represents the unit vector in the vertical direction; Repeat the above displacement calculation steps until the maximum elevation change of all particles is small enough, or the number of iterations reaches a preset value, then terminate the simulation process; The height difference between the point cloud and the grid particles is calculated, and the ground points and non-ground points are distinguished according to the pre-set threshold.

3. The method for accurate layered extraction of three-dimensional point cloud targets based on multi-feature fusion according to claim 1, characterized in that: The S200 includes: S201, point cloud regular grid division: The preprocessed point cloud space is divided into a grid of M×N×1. The regular grid of the point cloud is divided into: Among them, x min 、x max are the minimum and maximum x-coordinates of the preprocessed point cloud data; y min ,y max They are the minimum and maximum values ​​of the y coordinates in the preprocessed point cloud data; Get the x-coordinate and y-coordinate of a point i in the point cloud data, and calculate the row and column number of the point in the grid: Among them, (x i ,y i ) is the coordinate value of point i; m and n are the row and column numbers of point i in the regular grid; S202. Statistical analysis: The lowest point elevation in the grid is defined as the DEM feature, the highest point elevation in the grid is defined as the DSM feature, and the maximum height information of the objects in the grid compared to the ground is defined as the nDSM feature. DSM, DEM, and nDSM can be expressed as follows: Where z(m,n,i) is the z coordinate of the i-th point in the m-th row and n-th column grid; S203. Set the elevation threshold to split the point cloud into two layers: According to the statistical analysis of nDSM, an appropriate height difference threshold is set to segment the preprocessed point cloud into upper and lower layers.

4. The method for accurate layered extraction of three-dimensional point cloud targets based on multi-feature fusion according to claim 1, characterized in that: The S300 includes: S301, upper point cloud feature extraction: Traverse all points in the upper point cloud, calculate the covariance matrix of its neighborhood point set, and then calculate the eigenvalues ​​λ1, λ2, λ3 of the covariance matrix (λ1≥λ2≥λ3≥0), and calculate the linearity f according to the eigenvalues l : S302, setting a linearity threshold according to the linearity: According to the statistical analysis of linearity, an appropriate linearity threshold is set to remove non-target point clouds in the upper point cloud; S303, lower layer point cloud feature extraction: Traverse all points in the lower point cloud and calculate the neighborhood density feature d of each point s : Among them, n s is the number of neighborhood points in the sphere neighborhood, r s is the radius of the sphere neighborhood; after calculating the density of all points, the eigenvalues ​​need to be normalized to the [0,1] interval; S304, setting density feature threshold: according to the statistical analysis of the normalized density features, setting a suitable density feature threshold, and removing non-target points in the lower layer point cloud.

5. The method for accurate layered extraction of three-dimensional point cloud objects based on multi-feature fusion according to claim 1, characterized in that: The S400 includes: S401, Target Positioning: Based on the point cloud after non-target points have been removed, the elevation threshold is set according to the elevation characteristics of the target object, and the grids that meet the requirements are retained; S402, target object clustering: Traverse each retained point and search its neighborhood. If the number of neighborhood points is greater than the set threshold, it is considered as a core point, and the neighborhood points of the core point are considered as density reachable points. Traverse each core point and find all clusters by checking its density reachable points; set a threshold based on the elevation characteristics of the target object and delete the clusters that do not meet the requirements.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the method for hierarchical and precise extraction of three-dimensional point cloud targets based on multi-feature fusion as described in any one of claims 1 to 5 are implemented.

7. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the method for hierarchical and precise extraction of three-dimensional point cloud targets based on multi-feature fusion as described in any one of claims 1 to 6 are implemented.

Images