Three-dimensional point cloud segmentation method and device, electronic equipment and storage medium

Abstract

The invention relates to a three-dimensional point cloud segmentation method and device, electronic equipment and a storage medium, and belongs to the technical field of point cloud segmentation, and the method comprises the steps: obtaining three-dimensional point cloud data, and carrying out the feature extraction of the three-dimensional point cloud data, and obtaining a first feature point set; correcting features in the first feature point set to obtain a second feature point set; performing feature correction extraction and feature comparison extraction on the second feature point set to obtain a third feature point set and a fourth feature point set; and a three-dimensional point cloud segmentation result is obtained according to the fifth feature point set, so that three-dimensional point cloud data can be corrected, feature correction extraction and feature comparison extraction can be carried out, the extracted features can be filtered, and the ambiguity of the edge of an object in semantic prediction is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of point cloud segmentation, and in particular to a three-dimensional point cloud segmentation method, device, electronic device and storage medium. Background Art

[0002] With the technological breakthroughs of point cloud acquisition equipment such as LiDAR and RGB-D cameras, point cloud processing methods no longer need to rely on expensive mesh reconstruction or denoising processes, making point cloud data more and more widely used in multiple application fields of 3D scene understanding, such as robotics, autonomous driving, urban planning, infrastructure maintenance, and digital cultural relics protection. In these applications, point cloud semantic segmentation, as the core link in understanding and parsing 3D scenes, divides the original point cloud data into several subsets with different semantic labels. Therefore, it has received great attention and research in recent years. Among them, the segmentation of edges between 3D objects has always been the focus of semantic segmentation tasks. In recent years, some researchers have also made contributions in this direction, but they all need to introduce edge supervision in the model, which undoubtedly greatly increases the annotation work of point cloud data and is not conducive to practical engineering applications. Analyze the process of point cloud feature extraction. Due to the irregularity, sparsity and disorder of point cloud data, it has long been a technical challenge to directly operate point clouds, extract features and regress semantic labels.

[0003] The prior art proposes the Abstract Aggregation Module (SA), which is an efficient local feature extraction method. Since the SA module was proposed, it has become a key component of many models for local region feature learning and is used in combination with other modules to complete semantic segmentation tasks. However, when the SA module performs maximum pooling to aggregate all point features to the central representative point, it does not fully consider the diversity that may exist between neighborhood points. For example, if the central representative point is located at the edge of an object, its neighborhood may contain points of neighboring objects. In this case, the SA module directly aggregates the features of these points to the central representative point, which may have a negative impact on the semantic label prediction, resulting in blurring of the object edges in the semantic prediction.

[0004] Therefore, it is urgent to propose a three-dimensional point cloud segmentation method, device, electronic device and storage medium to solve the technical problem that the abstract aggregation module in the prior art does not fully consider the possible diversity between neighborhood points when performing semantic segmentation tasks, which may have a negative impact on semantic label prediction, thereby leading to blurred object edges in semantic prediction. Summary of the invention

[0005] In view of this, it is necessary to provide a three-dimensional point cloud segmentation method, device, electronic device and storage medium to solve the technical problem that the abstract aggregation module in the prior art does not fully consider the possible diversity between neighborhood points when performing semantic segmentation tasks, which may have a negative impact on semantic label prediction, thereby leading to blurred object edges in semantic prediction.

[0006] In order to solve the above problems, the present invention provides a three-dimensional point cloud segmentation method, comprising: Obtain 3D point cloud data; Performing feature extraction on the three-dimensional point cloud data to obtain a first feature point set; Modifying the features in the first feature point set to obtain a second feature point set; Performing feature correction extraction and feature comparison extraction on the second feature point set respectively to obtain a third feature point set and a fourth feature point set; A three-dimensional point cloud segmentation result is obtained according to the fifth feature point set.

[0007] In a possible implementation manner, the step of modifying the features in the first feature point set to obtain the second feature point set includes: Sampling the first feature point set to obtain a center point set; The first feature point set is grouped with the center point in the center point set as the center to obtain a neighborhood; According to the features of the domain points in the neighborhood, a local feature set is obtained; Based on the BALR module and the center point, the features of the domain points in the local feature set are corrected to obtain a target local feature set; Feature extraction is performed on the target local feature set to obtain a second feature point set.

[0008] In a possible implementation, after obtaining the local feature set according to the features of the domain points in the neighborhood, the method further includes: Determine whether there is a feature of a domain point in the local feature set that is different from a feature of the center point; If so, the features of the domain points in the local feature set are corrected based on the BALR module and the center point to obtain the target local feature set.

[0009] In a possible implementation, the domain points in the local feature set include neighborhood point features and neighborhood point spatial coordinates; the center point includes center point features and center point spatial coordinates; and the features of the domain points in the local feature set are corrected based on the BALR module and the center point to obtain a target local feature set, including: Calculating the neighborhood point features and the center point features to obtain feature differences; Calculating the spatial coordinates of the neighborhood point and the spatial coordinates of the center point to obtain a coordinate difference; Encoding the feature difference and the coordinate difference respectively to obtain a semantic weight and a spatial weight; The domain points in the local feature set are modified according to the semantic weight and the spatial weight to obtain a target local feature set.

[0010] In a possible implementation, performing feature comparison extraction on the second feature point set to obtain a fourth feature point set includes: Performing feature comparison and extraction on the second feature point set to obtain a fourth feature point set, including: Performing feature comparison on the second feature point set based on a cross-layer cross attention network to obtain feature connections; Determining the weight between the second feature point set and the representative point set according to the feature connection; A fourth feature point set is obtained according to the feature connection and the weight.

[0011] In a possible implementation, performing feature comparison on the second feature point set based on a cross-layer cross attention network to obtain feature connections includes: Selecting the second feature point set based on the farthest point sampling method to obtain a representative point set; The second feature point set and the representative point set are processed based on a cross-layer cross attention network to obtain feature connections.

[0012] In a possible implementation, extracting features from the target local feature set to obtain a second feature point set includes: Extracting features from the target local feature set to obtain a domain clustering feature centered on the center point; Obtaining, according to the domain aggregation features, the corrected features output by the BALR module; The modified features are concatenated with a set of neighborhood point spatial coordinates in the local feature set to obtain a second feature point set.

[0013] On the other hand, the present invention also provides a three-dimensional point cloud segmentation device, comprising: A data acquisition module, used to acquire three-dimensional point cloud data; A point set sampling module, used for performing feature extraction on the three-dimensional point cloud data to obtain a first feature point set; A first feature extraction module, used for correcting the features in the first feature point set to obtain a second feature point set; A second feature extraction module, used to perform feature correction extraction and feature comparison extraction on the second feature point set to obtain a third feature point set and a fourth feature point set; The segmentation result determination module is used to obtain a three-dimensional point cloud segmentation result according to the fifth feature point set.

[0014] On the other hand, an embodiment of the present invention discloses an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the computer program implements the various steps of the above-mentioned three-dimensional point cloud segmentation method embodiment when executed by the processor.

[0015] On the other hand, an embodiment of the present invention discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the various steps of the above-mentioned three-dimensional point cloud segmentation method embodiment are implemented.

[0016] The beneficial effects of the present invention are: three-dimensional point cloud data is acquired, and feature extraction is performed on the three-dimensional point cloud data to obtain a first feature point set; features in the first feature point set are corrected to obtain a second feature point set; feature correction extraction and feature comparison extraction are performed on the second feature point set respectively to obtain a third feature point set and a fourth feature point set; based on the fifth feature point set, a three-dimensional point cloud segmentation result is obtained, so that the three-dimensional point cloud data can be corrected, and feature correction extraction and feature comparison extraction can be performed, so that the extracted features can be filtered, thereby reducing the ambiguity of the edges of objects in semantic prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a flow chart of an embodiment of a three-dimensional point cloud segmentation method provided by the present invention; Figure 2 A schematic diagram of the structure of an embodiment of the BALR-NET model provided by the present invention; Figure 3 For the present invention Figure 1 A schematic flow chart of an embodiment of step S103; Figure 4 For the present invention Figure 3 A schematic flow chart of an embodiment of step S304; Figure 5 For the present invention Figure 1 A schematic flow chart of an embodiment of step S104; Figure 6 A schematic diagram of the structure of an embodiment of a three-dimensional point cloud segmentation device provided by the present invention; Figure 7 A schematic structural diagram of an embodiment of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0018] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0019] like Figure 1 As shown, a specific embodiment of the present invention discloses a three-dimensional point cloud segmentation method, comprising: S101, obtaining three-dimensional point cloud data; S102, extracting features from the three-dimensional point cloud data to obtain a first feature point set; S103, modifying the features in the first feature point set to obtain a second feature point set; S104, performing feature correction extraction and feature comparison extraction on the second feature point set respectively to obtain a third feature point set and a fourth feature point set; S105. Obtain a three-dimensional point cloud segmentation result according to the fifth feature point set.

[0020] It should be understood that the three-dimensional point cloud data obtained in step S101 may be obtained by obtaining a three-dimensional point cloud data set according to a radar detection device, or may be obtained by calling a historically stored three-dimensional point cloud data set from a storage medium.

[0021] In a specific embodiment of the present invention, a BALR-NET model is provided. The BALR-NET model introduces a BALR module with edge perception function to replace the SA module for local feature extraction. In addition, in order to break through the isolated state of the last layer of local features encoded by the PointNet++ model and enhance the information exchange between the last layer of local features, the BALR-NET model designs a cross-layer cross attention network. The network establishes an association between the representative point features of the next layer and the groups with a longer spatial distance in the previous layer that generate the features. The input of the BALR-NET model is three-dimensional point cloud data ,like Figure 2 As shown in Figure 2, the encoder of the BALR-NET model The layer input is , Indicates the number of input points, Represents the feature dimension of the input point, including 3D spatial information and Dimensional semantic information. After the farthest point sampling (FPS) ( Figure 2 The sample in The point set of points is ,in , Indicate point The Euclidean space position, Indicates the seed point of dimensional features. The output of the layer is , Represents the feature dimension of the output point. Considering that semantic information gradually becomes richer with the increase of network depth, while fine-grained spatial information is gradually lost, the BALR-NET model uses edge-aware BALR modules in the second and third layers of encoding to extract local features from the downsampled point cloud. In particular, in order to break the isolated state of the local features in the last layer (fourth layer) of the PointNet++ model B encoding, the BALR-NET model incorporates a cross-layer cross-attention network in the third-layer BALR module to extract local features from the downsampled point cloud.

[0022] The first layer of the encoder of the BALR-NET model is the SA module, and the input is , the point set after farthest point sampling (FPS) is , through the SA module Perform feature extraction to obtain the first feature point set , wherein the process of the SA module extracting features from the 3D point cloud data can be set according to the actual situation, and the embodiment of the present invention is not limited here. The second layer is the BALR module, which The point set after farthest point sampling (FPS) is , through the BALR module The features in are modified to obtain the second feature point set The third layer is the BALR module and the cross-layer attention network. The point set after farthest point sampling (FPS) is , through the BALR module and the cross-layer cross attention network CLCA (ie CLCA-BALR), respectively Perform feature extraction to obtain the third feature point set and the fourth feature point set, and concatenate the third feature point set and the fourth feature point set to obtain the fifth feature point set Specifically, the second feature point set can be feature extracted by the BALR module to obtain a third feature point set, and the second feature point set can be feature extracted by the cross-layer cross attention network to obtain a fourth feature point set, and then the fourth feature point set and the fifth feature point set are concatenated to obtain the fifth feature point set. The fourth layer is the SA module, which is used to calculate the fifth feature point set The point set after farthest point sampling (FPS) is , through the SA module Perform feature extraction and obtain , the output of the fourth layer After the first, second and third layer feature extraction, the point sets obtained by feature extraction can be respectively connected to , and ) to send a copy to the decoder (Decoder), you can , , and After merging, it is decoded by the decoder and then segmented (Segmentation Head) to obtain the three-dimensional point cloud segmentation result.

[0023] Compared with the prior art, the present embodiment provides a method for extracting features from three-dimensional point cloud data to obtain a first feature point set; correcting features in the first feature point set to obtain a second feature point set; extracting features from the second feature point set based on a BALR module and a cross-layer cross-attention network to obtain a third feature point set and a fourth feature point set; and obtaining a three-dimensional point cloud segmentation result based on the fifth feature point set, so that three-dimensional features can be extracted from the three-dimensional point cloud data according to the BALR module and the cross-layer cross-attention network, and the extracted features are filtered, thereby reducing the ambiguity of object edges in semantic prediction.

[0024] In some embodiments of the present invention, Figure 3 As shown, step S103 includes: S301, sampling the first feature point set to obtain a center point set; S302, grouping the first feature point set with the center point in the center point set as the center to obtain a neighborhood; S303, obtaining a local feature set according to the features of the domain points in the neighborhood; S304, correcting the features of the domain points in the local feature set based on the BALR module and the center point to obtain a target local feature set; S305 : Extract features from the target local feature set to obtain a second feature point set.

[0025] In a specific embodiment of the present invention, the input of the second layer is the first feature point set output by the first layer. , and then the farthest point sampling (FPS) can be used to Sampling is performed to obtain the center point set , Include center point , i Indicates icenter point, point Centered on Grouping to form a neighborhood, that is, each center point i The corresponding neighborhood is then constructed with the features of the domain points in the neighborhood, as shown in formula (1): (1) In the formula, Center point No. The characteristics and spatial coordinates of each field point, is the number of points in the neighborhood.

[0026] The SA module extracts features from all domain points in the local feature set, as shown in formula (2): (2) In the formula, represents a shared multilayer perceptron, represents the maximum pooling, For The local area features are centered.

[0027] In the SA module, if the center point If it is located at the boundary between objects, the center point Constructed collection May contain The characteristics of semantically heterogeneous points directly affect , which leads to blurred edge segmentation. To solve this problem, the BALR-NET model proposes a BALR module to Specifically, when the When the semantics of a domain point is different from that of the center point, the BALR module will By performing correction and updating, a target local feature set can be obtained, and then the target local feature set is brought into formula (2) for feature extraction to obtain a second feature point set.

[0028] In some embodiments of the present invention, after step S303, the method further includes: Determine whether there are domain points in the local feature set that have features that are different from those of the center point; If so, the features of the domain points in the local feature set are corrected based on the BALR module and the center point to obtain the target local feature set.

[0029] In a specific embodiment of the present invention, after obtaining the local feature set, the local feature set can be judged to determine whether there is a domain point in the local feature set whose features are different from those of the center point; if so, step S304 can be performed; if not, it means that the difference between the features of the domain point and the features of the center point is relatively small, and step S304 is also performed, but the result after correction is not much different from that before correction, which is close to no correction.

[0030] In some embodiments of the present invention, the neighborhood points in the local feature set include neighborhood point features and neighborhood point spatial coordinates; the center point includes center point features and center point spatial coordinates; Figure 4 As shown, step S304 includes: S401, calculating the neighborhood point features and the center point features to obtain feature differences; S402, calculating the spatial coordinates of the neighborhood points and the spatial coordinates of the center point to obtain a coordinate difference; S403, respectively encode the feature difference and the coordinate difference to obtain a semantic weight and a spatial weight; S404: Modify the domain points in the local feature set according to the semantic weight and the spatial weight to obtain a target local feature set.

[0031] In a specific embodiment of the present invention, for a local feature set For each neighborhood point in, the BALR module first calculates the neighborhood point features With center point feature The characteristic difference And the spatial coordinates of the neighborhood points With center point The spatial coordinates of the center point The coordinate difference of , respectively, for the characteristic difference and coordinate difference Encode to get the semantic weight of each neighborhood point and spatial weight , as shown in formulas (3) and (4): (3) (4) In the formula, As a nonlinear mapping function The values ​​are mapped to the range 0 to 1. is the convolution function, is batch normalization, represents a non-linear activation function. , The values ​​of are all between 0 and 1. The larger the characteristic difference, The closer it is to 1, the closer the distance is. The closer to 1.

[0032] Then calculate the corrected neighborhood point features , as shown in formula (5): (5) In the formula, represents matrix multiplication, , update the center point S i The domain set is obtained to obtain the target local feature set .

[0033] In some embodiments of the present invention, step S305 includes: Extract features from the target local feature set to obtain the domain clustering features centered on the center point; According to the domain aggregation features, the corrected features output by the BALR module are obtained; The corrected features are concatenated with the set of spatial coordinates of neighborhood points in the local feature set to obtain a second feature point set.

[0034] In a specific embodiment of the present invention, the corrected feature in the target local feature set can be Substitute into formula (2) to extract domain features and get Clustering features in the center Then the BALR module can output the corrected features based on the domain aggregation features. , and then correct the feature By splicing with the set of spatial coordinates of the neighborhood points in the local feature set, the second feature point set output by the second layer can be obtained. , where the domain aggregation feature is shown in formula (6): (6) In some embodiments of the present invention, Figure 5 As shown, step S104 includes: S501, performing feature comparison on the second feature point set based on a cross-layer cross attention network to obtain feature connections; S502, determining the weight between the second feature point set and the representative point set according to the feature connection; S503: Obtain a fourth feature point set according to feature connections and weights.

[0035] In some embodiments of the present invention, feature comparison is performed on the second feature point set based on a cross-layer cross attention network to obtain feature connections, including: The second feature point set is selected based on the farthest point sampling method to obtain a representative point set; The second feature point set and the representative point set are processed based on a cross-layer cross attention network to obtain feature connections.

[0036] In a specific embodiment of the present invention, in order to expand the receptive field and strengthen the information exchange between all local areas, the BALR-NET model incorporates a cross-layer cross attention network into the BALR module of the third layer. The network establishes a connection between the fourth-layer representative point features and the third-layer features, and then embeds the fourth-layer representative point features in the third-layer local domain center point features, so that when the fourth layer performs local feature extraction, it not only gathers the point features of the domain, but also obtains the features of other domains, thereby strengthening the information exchange between local areas. Assume that the representative point set of the third layer after sampling is , the point set of the fourth layer input , the representative point set after sampling is The specific calculation process of the cross-layer attention network is as follows.

[0037] First, on the third floor Use the farthest point sampling to select the representative point set , , represents a point set In Euclidean space, the fourth level represents the point set Then, establish the point set With the connection of the third layer points, we get the characteristic connection ( , , ). , , The calculation of is shown in formula (7): (7) in, represents a linear layer, express The semantic features of express semantic features.

[0038] Then the weight calculation between the representative points of the fourth layer and the second feature point set of the third layer is shown in formula (8): (8) In the formula, represents the summation function, represents matrix multiplication, .

[0039] Then, based on the feature connections and weights, the established representative point set can be calculated. The fourth feature point set after being connected with the features of the second feature point set of the third layer is as shown in formula (9): (9) In the formula, represents the convolution operation, represents addition, It is the fifth feature output by the cross-layer criss-cross attention network.

[0040] Finally, the fourth feature point set is mixed with the third feature point set of the third layer BALR module to obtain the output feature of the third layer CLCA-BALR module , splicing And the coordinates of the third layer representative points are obtained to obtain the fifth feature point set of the third layer encoding output , as shown in formula (10): (10) In the formula, represents the convolution operation, Represents a splicing operation, Output of BALR module.

[0041] The embodiment of the present invention proposes a novel point cloud segmentation model (BALR-NET), which shows excellent performance in semantic segmentation and component segmentation tasks, especially in processing edge segmentation between objects, and effectively improves the clarity of edge segmentation. In order to solve the problem of fuzzy edge segmentation between objects in semantic prediction, an edge-aware mechanism based on semantic distance and Euclidean distance between neighborhood points is innovatively introduced, and an edge-aware local feature expression module (BALR) is proposed to enhance the feature representation of edge points. In order to solve the problem of feature isolation in the feature encoding process of the traditional abstract aggregation module (SA), a cross-layer cross-attention network (CLCA) is designed with the representative points of the next layer as the query. The network breaks the isolation of features by establishing long-range dependencies between the upper and lower layers.

[0042] In order to better implement the three-dimensional point cloud segmentation method in the embodiment of the present invention, based on the three-dimensional point cloud segmentation method, the embodiment of the present invention also provides a three-dimensional point cloud segmentation device, such as Figure 6 As shown, the three-dimensional point cloud segmentation device 600 includes: A data acquisition module 601 is used to acquire three-dimensional point cloud data; The point set sampling module 602 is used to extract features from the three-dimensional point cloud data to obtain a first feature point set; A first feature extraction module 603, used to modify the features in the first feature point set to obtain a second feature point set; A second feature extraction module 604 is used to perform feature correction extraction and feature comparison extraction on the second feature point set to obtain a third feature point set and a fourth feature point set; The segmentation result determination module 605 is used to obtain a three-dimensional point cloud segmentation result according to the fifth feature point set.

[0043] The three-dimensional point cloud segmentation device 600 provided in the above embodiment can implement the technical solution described in the above three-dimensional point cloud segmentation method embodiment. The specific implementation principles of the above modules or units can refer to the corresponding contents in the above three-dimensional point cloud segmentation method embodiment, which will not be repeated here.

[0044] like Figure 7 As shown, the present invention also provides an electronic device 700. The electronic device 700 includes a processor 701, a memory 702 and a display 703. Figure 7 Only some components of the electronic device 700 are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0045] In some embodiments, the memory 702 may be an internal storage unit of the electronic device 700, such as a hard disk or memory of the electronic device 700. In other embodiments, the memory 702 may also be an external storage device of the electronic device 700, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc., equipped on the electronic device 700.

[0046] Furthermore, the memory 702 may include both an internal storage unit of the electronic device 700 and an external storage device. The memory 702 is used to store application software installed in the electronic device 700 and various data.

[0047] In some embodiments, the processor 701 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run program codes or process data stored in the memory 702, such as the three-dimensional point cloud segmentation method of the present invention.

[0048] In some embodiments, the display 703 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 703 is used to display information of the electronic device 700 and to display a visual user interface. The components 701-703 of the electronic device 700 communicate with each other via a system bus.

[0049] In some embodiments of the present invention, when the processor 701 executes the 3D point cloud segmentation program in the memory 702, the following steps may be implemented: Obtain 3D point cloud data; Perform feature extraction on the three-dimensional point cloud data to obtain a first feature point set; Modify the features in the first feature point set to obtain a second feature point set; Performing feature correction extraction and feature comparison extraction on the second feature point set respectively to obtain a third feature point set and a fourth feature point set; According to the fifth feature point set, a three-dimensional point cloud segmentation result is obtained.

[0050] It should be understood that: when the processor 701 executes the three-dimensional point cloud segmentation program in the memory 702, in addition to the above functions, other functions can also be implemented. For details, please refer to the description of the corresponding method embodiment above.

[0051] Furthermore, the embodiment of the present invention does not specifically limit the type of the electronic device 700 mentioned, and the electronic device 700 may be a portable electronic device such as a mobile phone, a tablet computer, a personal digital assistant (PDA), a wearable device, a laptop computer, etc. Exemplary embodiments of portable electronic devices include but are not limited to portable electronic devices equipped with IOS, Android, Microsoft or other operating systems. The above-mentioned portable electronic devices may also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, the electronic device 700 may not be a portable electronic device, but a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0052] Accordingly, an embodiment of the present application also provides a computer-readable storage medium, which is used to store computer-readable programs or instructions. When the program or instructions are executed by a processor, the three-dimensional point cloud segmentation method steps or functions provided in the above-mentioned method embodiments can be implemented.

[0053] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the computer program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0054] The three-dimensional point cloud segmentation method and device provided by the present invention are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A three-dimensional point cloud segmentation method, characterized in that: include: Obtain 3D point cloud data; Performing feature extraction on the three-dimensional point cloud data to obtain a first feature point set; Modifying the features in the first feature point set to obtain a second feature point set; performing feature correction extraction and feature comparison extraction on the second feature point set respectively to obtain a third feature point set and a fourth feature point set, and concatenating the third feature point set and the fourth feature point set to obtain a fifth feature point set; A three-dimensional point cloud segmentation result is obtained according to the fifth feature point set.

2. The three-dimensional point cloud segmentation method according to claim 1, characterized in that: The modifying the features in the first feature point set to obtain a second feature point set includes: Sampling the first feature point set to obtain a center point set; Grouping the first feature point set with the center point in the center point set as the center to obtain a neighborhood; According to the features of the domain points in the neighborhood, a local feature set is obtained; Based on the BALR module and the center point, the features of the domain points in the local feature set are corrected to obtain a target local feature set; Feature extraction is performed on the target local feature set to obtain a second feature point set.

3. The three-dimensional point cloud segmentation method according to claim 2, characterized in that: After obtaining the local feature set according to the features of the domain points in the neighborhood, the method further includes: Determine whether there is a feature of a domain point in the local feature set that is different from a feature of the center point; If so, the features of the domain points in the local feature set are corrected based on the BALR module and the center point to obtain the target local feature set.

4. The three-dimensional point cloud segmentation method according to claim 2, characterized in that: The domain points in the local feature set include neighborhood point features and neighborhood point spatial coordinates; the center point includes center point features and center point spatial coordinates; the features of the domain points in the local feature set are corrected based on the BALR module and the center point to obtain a target local feature set, including: Calculating the neighborhood point features and the center point features to obtain feature differences; Calculating the spatial coordinates of the neighborhood point and the spatial coordinates of the center point to obtain a coordinate difference; Encoding the feature difference and the coordinate difference respectively to obtain a semantic weight and a spatial weight; The domain points in the local feature set are modified according to the semantic weight and the spatial weight to obtain a target local feature set.

5. The three-dimensional point cloud segmentation method according to claim 1, characterized in that: Performing feature comparison and extraction on the second feature point set to obtain a fourth feature point set, including: Performing feature comparison on the second feature point set based on a cross-layer cross attention network to obtain feature connections; Determining the weight between the second feature point set and the representative point set according to the feature connection; A fourth feature point set is obtained according to the feature connection and the weight.

6. The three-dimensional point cloud segmentation method according to claim 5, characterized in that: The performing feature comparison on the second feature point set based on the cross-layer cross attention network to obtain feature connections includes: Selecting the second feature point set based on the farthest point sampling method to obtain a representative point set; The second feature point set and the representative point set are processed based on a cross-layer cross attention network to obtain feature connections.

7. The three-dimensional point cloud segmentation method according to claim 2, characterized in that: The step of extracting features from the target local feature set to obtain a second feature point set includes: Extracting features from the target local feature set to obtain a domain clustering feature centered on the center point; Obtaining, according to the domain aggregation features, the corrected features output by the BALR module; The modified features are concatenated with a set of neighborhood point spatial coordinates in the local feature set to obtain a second feature point set.

8. A three-dimensional point cloud segmentation device, characterized in that: include: A data acquisition module, used to acquire three-dimensional point cloud data; A point set sampling module, used for performing feature extraction on the three-dimensional point cloud data to obtain a first feature point set; A first feature extraction module, used for correcting the features in the first feature point set to obtain a second feature point set; A second feature extraction module, used to perform feature correction extraction and feature comparison extraction on the second feature point set to obtain a third feature point set and a fourth feature point set; The segmentation result determination module is used to obtain a three-dimensional point cloud segmentation result according to the fifth feature point set.

9. An electronic device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein when the computer program is executed by the processor, the steps of the three-dimensional point cloud segmentation method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the three-dimensional point cloud segmentation method according to any one of claims 1 to 7 are implemented.

Images