Method and device for recognizing and segmenting roadway support components facing three-dimensional point cloud data

Abstract

The application provides a kind of roadway support component identification and segmentation method and device for three-dimensional point cloud data, method includes: obtaining the original point cloud data of target scene area, and based on the preset block rule, obtain the multiple block results corresponding to the original point cloud data under multiple grids;Wherein, a grid corresponds to a block result;Original point cloud data and multiple block results are input into pre-trained deep learning model, and the multiple category prediction labels corresponding to each point in original point cloud data are obtained;Multiple category prediction labels corresponding to each point are input into pre-trained decision tree model, and the target label result of each point is obtained, and the support component position in original point cloud data is determined based on the target label result of each point.The method provided by the application determines the final label result of each point cloud block in multiple category prediction labels through decision tree model, realizes the quick automatic support component identification and separation.

Description

Technical Field

[0001] This invention relates to the field of tunnel point cloud processing technology, and in particular to a method and apparatus for identifying and segmenting tunnel support components based on three-dimensional point cloud data. Background Technology

[0002] 3D laser scanning, as a high-precision, rapid, and non-destructive testing technology, has been widely applied in the field of tunnel support. Currently, laser scanning can achieve a single scan of the entire tunnel cross-section, using the acquired massive point cloud data to construct a digital model for further numerical analysis and cross-sectional deformation inversion. Anchor bolt support is one of the most common reinforcement methods for tunnels, with at least one anchor bolt per square meter of tunnel cross-section. In addition, the tunnel surface has numerous auxiliary support components such as metal mesh and steel strips. Constructing numerical models and inverting cross-sectional deformation based on point cloud data often requires removing the point cloud containing the support plate and other components. However, the massive amount of point cloud data makes manual removal of the support plate difficult. Furthermore, the deformation patterns of the tunnel where the support plate is located differ significantly from those in areas without support plates.

[0003] How to effectively identify the support components in the roadway in order to analyze the deformation law of the surrounding rock and achieve roadway stability is a technical problem that needs to be solved. Summary of the Invention

[0004] This invention provides a method and apparatus for identifying and segmenting roadway support components based on three-dimensional point cloud data, in order to overcome the deficiencies in the prior art.

[0005] This invention provides a method for identifying and segmenting tunnel support components based on three-dimensional point cloud data, including:

[0006] The original point cloud data of the target scene area is acquired, and multiple block results corresponding to the original point cloud data under various grids are obtained based on preset block rules; wherein, one grid corresponds to one block result.

[0007] The original point cloud data and the multiple block results are input into a pre-trained deep learning model to obtain multiple category prediction labels corresponding to each point in the original point cloud data.

[0008] The predicted labels of multiple categories corresponding to each point are input into a pre-trained decision tree model to obtain the target label result for each point, and the location of the support component in the original point cloud data is determined based on the target label result for each point.

[0009] According to the present invention, a method for identifying and segmenting tunnel support components based on three-dimensional point cloud data includes obtaining multiple segmentation results of the original point cloud data under various grids based on preset segmentation rules, comprising:

[0010] A first grid covering the target scene area is created to obtain the first block result of the original point cloud data under the first grid; wherein, the first grid is composed of multiple squares with a side length of a first preset size;

[0011] The first grid is moved according to a preset moving rule to obtain the target grid and the target block result of the original point cloud data under the target grid.

[0012] According to the present invention, a method for identifying and segmenting tunnel support components based on three-dimensional point cloud data includes the following steps: moving the first grid according to a preset movement rule to obtain a target grid and target block results of the original point cloud data under the target grid.

[0013] The first grid is moved horizontally according to a second preset size to obtain a second grid and a second block result of the original point cloud data under the second grid; wherein, the second preset size is the maximum size of the tray, and the first preset size is twice the second preset size;

[0014] The first grid is moved vertically according to the second preset size to obtain the third grid and the third block result of the original point cloud data under the third grid;

[0015] The first grid is moved horizontally and vertically according to the second preset size to obtain the fourth grid and the fourth block result of the original point cloud data under the fourth grid.

[0016] According to the present invention, a method for identifying and segmenting roadway support components based on three-dimensional point cloud data, the method further includes, before inputting the original point cloud data and the multiple segmentation results into a pre-trained deep learning model:

[0017] The original point cloud data is preprocessed using a normalization preprocessing head; wherein, the normalization preprocessing head is used to normalize the original point cloud data.

[0018] According to the present invention, a method for identifying and segmenting roadway support components based on three-dimensional point cloud data is provided, wherein the normalization preprocessing head includes a translation layer, a principal component analysis layer, and an L1 norm normalization layer.

[0019] The preprocessing operation performed on the raw point cloud data through the normalization preprocessing head includes:

[0020] The original point cloud data is input into the translation layer to obtain the first point cloud data;

[0021] The first point cloud data is input into the principal component analysis layer to obtain the second point cloud data and the corresponding first rotation matrix;

[0022] The second point cloud data is input into the L1 norm normalization layer to obtain the third point cloud data and the corresponding second rotation matrix.

[0023] The step of inputting the original point cloud data and the multiple block results into a pre-trained deep learning model to obtain multiple category prediction labels corresponding to each point in the original point cloud data includes:

[0024] The third point cloud data and the four block results are input into a pre-trained deep learning model to obtain four category prediction labels for each point and a fourth point cloud data containing category prediction labels after segmentation.

[0025] According to the present invention, a method for identifying and segmenting tunnel support components based on three-dimensional point cloud data, after obtaining four category prediction labels corresponding to each point and a fourth point cloud data containing the category prediction labels, the method further includes:

[0026] Based on the first rotation matrix, the second rotation matrix, and the fourth point cloud data, an inverse transformation is performed to recover the original point cloud data containing the category prediction label.

[0027] According to the present invention, a method for identifying and segmenting roadway support components based on three-dimensional point cloud data includes inputting multiple category prediction labels corresponding to each point into a pre-trained decision tree model to obtain the target label result for each point, and determining the location of the support component in the original point cloud data based on the target label result for each point, comprising:

[0028] The four category prediction labels corresponding to each point are input into the pre-trained decision tree model to obtain the target label result for each point, and the location of the support component in the original point cloud data is determined based on the target label result for each point.

[0029] The present invention also provides a device for identifying and segmenting roadway support components based on three-dimensional point cloud data, comprising:

[0030] The segmentation module is used to acquire the original point cloud data of the target scene area and obtain multiple segmentation results of the original point cloud data under various grids based on preset segmentation rules; wherein, one grid corresponds to one segmentation result.

[0031] The prediction module is used to input the original point cloud data and the multiple block results into a pre-trained deep learning model to obtain multiple category prediction labels corresponding to each point in the original point cloud data.

[0032] The determination module is used to input multiple category prediction labels corresponding to each point into a pre-trained decision tree model to obtain the target label result for each point, and determine the location of the support component in the original point cloud data based on the target label result for each point.

[0033] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the tunnel support component identification and segmentation method for three-dimensional point cloud data as described above.

[0034] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for identifying and segmenting roadway support components based on three-dimensional point cloud data as described above.

[0035] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the method for identifying and segmenting roadway support components based on three-dimensional point cloud data as described above.

[0036] This invention provides a method and apparatus for identifying and segmenting roadway support components based on 3D point cloud data. The method involves acquiring original point cloud data of a target scene area and obtaining multiple segmentation results for the original point cloud data under various grid types based on preset segmentation rules, where one grid type corresponds to one segmentation result. The original point cloud data and multiple segmentation results are input into a pre-trained deep learning model to obtain multiple predicted category labels for each point in the original point cloud data. These predicted category labels are then input into a pre-trained decision tree model to obtain the target label result for each point. Based on the target label result for each point, the location of the support component in the original point cloud data is determined. Therefore, this invention segments the original point cloud data using preset segmentation rules, reducing the point cloud size while ensuring that each support component is completely contained within at least one point cloud block, facilitating subsequent fine segmentation. The final label result for each point cloud block is determined by the decision tree model from multiple predicted category labels, achieving rapid and automatic identification and separation of support components. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1This is one of the flowcharts illustrating the method for identifying and segmenting roadway support components based on three-dimensional point cloud data provided by the present invention.

[0039] Figure 2 This is a partial schematic diagram of the original point cloud provided by the present invention;

[0040] Figure 3 This is an example diagram of the preset block division rules in the method for identifying and segmenting roadway support components based on three-dimensional point cloud data provided by the present invention;

[0041] Figure 4 This is an example diagram of block division based on preset block division rules provided by the present invention;

[0042] Figure 5 This is a schematic diagram of the segmentation process provided by the present invention, which uses a normalized preprocessing head and a deep learning model.

[0043] Figure 6 This is a schematic diagram of the identification results of the method for identifying and segmenting roadway support components based on three-dimensional point cloud data provided by the present invention;

[0044] Figure 7 This is a schematic diagram of the identification results of the tunnel support component identification and segmentation method for three-dimensional point cloud data provided by the present invention in a real-world scenario;

[0045] Figure 8 This is a schematic diagram of the structure of the tunnel support component identification and segmentation device for three-dimensional point cloud data provided by the present invention;

[0046] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0048] The following is combined Figures 1-9 This invention describes a method and apparatus for identifying and segmenting roadway support components based on three-dimensional point cloud data.

[0049] It should be noted that 3D laser scanning of underground tunnels generates billions of points in 3D point cloud data of the tunnel cross-section, which includes numerous components such as anchor bolts, anchor cable trays, metal mesh, and steel strips. Existing methods for removing the point cloud data containing these components have significant limitations. Therefore, this invention provides a method for identifying and segmenting tunnel support components based on 3D point cloud data to address the aforementioned problems.

[0050] Figure 1 This is one of the flowcharts illustrating the method for identifying and segmenting roadway support components based on 3D point cloud data provided in this embodiment. Figure 1 As shown in this embodiment, the method for identifying and segmenting roadway support components based on three-dimensional point cloud data includes:

[0051] Step 100: Obtain the original point cloud data of the target scene area, and obtain multiple block results corresponding to the original point cloud data under various grids based on preset block rules; wherein, one grid corresponds to one block result.

[0052] Figure 2 This is a partial schematic diagram of the original point cloud provided in this embodiment. See also... Figure 2 The target scene area can be, for example, a long tunnel, with at least one anchor bolt per square meter of tunnel cross-section. Furthermore, the tunnel surface has numerous auxiliary support components such as metal mesh and steel strips. The following explanation uses the identification of a pallet within the tunnel as an example, but the target object can also be metal mesh, steel strips, etc. This embodiment does not impose any particular limitation on this. It should be noted that long tunnels are often hundreds of meters long, and the amount of point cloud data acquired from multiple high-precision 3D scans often exceeds several billion. Such a large amount of data poses a significant challenge to point cloud processing and segmentation. Therefore, this embodiment first needs to reduce the original point cloud data. However, due to the working principle of the acquisition equipment and environmental factors, the density of point cloud data is often uneven; and since scanning can only acquire surface information, spatial and geometric features of the point cloud often suffer information loss. Relying solely on the XYZ coordinate data of the point cloud itself is insufficient as sufficient features for classification. Therefore, it is essential to ensure that the point cloud containing each pallet is completely contained by at least one point cloud block.

[0053] Step 100 specifically includes:

[0054] A first grid covering the target scene area is created to obtain the first block result of the original point cloud data under the first grid; wherein, the first grid is composed of multiple squares with a side length of a first preset size;

[0055] The first grid is moved according to a preset moving rule to obtain the target grid and the target block result of the original point cloud data under the target grid.

[0056] In one embodiment, the maximum size of the aisle tray is defined as X (i.e., the second preset size). First, a 2X*2X first mesh-1 is laid at intervals of 2X (i.e., the first preset size) units throughout the entire target scene area to cover the entire area.

[0057] Further, the first grid is moved according to a preset movement rule to obtain the target grid and the target block result of the original point cloud data under the target grid, specifically including:

[0058] The first grid is moved horizontally according to a second preset size to obtain a second grid and a second block result of the original point cloud data under the second grid; wherein, the second preset size is the maximum size of the tray, and the first preset size is twice the second preset size;

[0059] The first grid is moved vertically according to the second preset size to obtain the third grid and the third block result of the original point cloud data under the third grid;

[0060] The first grid is moved horizontally and vertically according to the second preset size to obtain the fourth grid and the fourth block result of the original point cloud data under the fourth grid.

[0061] In one embodiment, the first grid is moved horizontally to the left / right by X units to form the second grid mesh-2; the first grid is moved vertically upward / downward by X units to form the third grid mesh-3; and the first grid is simultaneously moved horizontally to the upper left / lower right by X units to form the fourth grid mesh-4. This results in all trays being covered by at least one grid.

[0062] Specifically, Figure 3 This is an example diagram of the preset block division rules in the method for identifying and segmenting roadway support components based on 3D point cloud data provided in this embodiment, such as... Figure 3 As shown, squares 1-4 represent the typical positions of the trays after the first coverage. Tray 1 is already completely contained within mesh-1, so only trays 2, 3, and 4 need to be considered. Tray 2 only needs to be completely covered by moving the first grid up / down X units (mesh-3); tray 4 only needs to be completely covered by moving the first grid left / right X units (mesh-2); and tray 3 needs to be completely covered by moving the first grid simultaneously to the upper left / lower right X units (mesh-4).

[0063] Figure 4 This is an example diagram of block division based on preset block division rules provided in this embodiment, such as... Figure 4 As shown, the preset segmentation rule proposed in this embodiment reduces the size of the point cloud while ensuring that each tray is completely contained by at least one point cloud block, which facilitates fine segmentation in the next stage.

[0064] Step 200: Input the original point cloud data and the multiple block results into a pre-trained deep learning model to obtain multiple category prediction labels corresponding to each point in the original point cloud data.

[0065] It's important to note that a key characteristic of 3D point clouds is their disordered nature. Ensuring consistent network output across different input sequences is crucial for the algorithm. PointNet addresses this disorder issue using symmetric functions (maximum pooling layers) and is widely used in point cloud recognition and segmentation. However, to address the rotation invariance of point clouds—that is, ensuring consistent network output for the same input after different rotations—existing PointNet's T-Net transformation matrix method still has limitations.

[0066] In this embodiment, before inputting the original point cloud data and the multiple block results into the pre-trained deep learning model, the method further includes:

[0067] The original point cloud data is preprocessed using a normalization preprocessing head; wherein the normalization preprocessing head is used to normalize the original point cloud data.

[0068] Specifically, the normalization preprocessing head includes a translation layer, a principal component analysis layer, and an L1 norm normalization layer to achieve more accurate segmentation results.

[0069] The preprocessing operation performed on the raw point cloud data through the normalization preprocessing head includes:

[0070] The original point cloud data is input into the translation layer to obtain the first point cloud data;

[0071] The first point cloud data is input into the principal component analysis layer to obtain the second point cloud data and the corresponding first rotation matrix;

[0072] The second point cloud data is input into the L1 norm normalization layer to obtain the third point cloud data and the corresponding second rotation matrix.

[0073] Figure 5 This is a schematic diagram of the segmentation process using a normalized preprocessing head and a deep learning model provided in this embodiment, as shown below. Figure 5 As shown, in one embodiment, the original point cloud data is first translated:

[0074]

[0075] Among them, O i The original point cloud data is in the form of N*3. The original point cloud centroid coordinates, i.e., O i The mean, O1 is the first point cloud data.

[0076] Furthermore, the first point cloud data is input into the principal component analysis layer PCAnorm, the transformed point cloud is normalized based on PCA, and the corresponding rotation matrix is ​​solved:

[0077] O2 = M1 * O1

[0078] Where O2 is the second point cloud data after PCA normalization, and M1 is the first rotation matrix, which is in the form of 3*3.

[0079] Furthermore, the second point cloud data is input into the L1-norm L1 norm normalization layer. The transformed point cloud is normalized based on the L1-norm layer, and the corresponding rotation matrix is ​​solved.

[0080] O f =M2*O2

[0081] Among them, O f M1 represents the third point cloud data after L1-norm normalization, and M2 is the second rotation matrix, which is in the form of 3*3.

[0082] The step involves inputting the original point cloud data and the multiple block results into a pre-trained deep learning model to obtain multiple category prediction labels corresponding to each point in the original point cloud data. Specifically, this includes:

[0083] The third point cloud data and the four block results are input into a pre-trained deep learning model to obtain four category prediction labels for each point and a fourth point cloud data containing category prediction labels after segmentation.

[0084] It should be noted that after normalization, the input point cloud is adjusted to the same pose, and the PointNet deep learning model is then used for object recognition and segmentation. See also... Figure 5 The specific steps are as follows:

[0085] 1) Multilayer Perceptron (MLP): A multilayer perceptron (MLP) is used to extract local features for each point. An MLP is a fully connected neural network that learns non-linear feature representations of points. In PointNet, the output of the MLP is an N*64 feature vector for each point.

[0086] 2) Feature Transformation (T-Net): Spatially aligns the input feature vector N*64 to make the network invariant to geometric transformations of the input data.

[0087] 3) Multilayer Perceptron (MLP): Further expands the size of the feature vector of each point, so that the feature vector of each point increases to N*1024.

[0088] 4) Max Pooling: To obtain a global feature representation of the point cloud, max pooling is used to aggregate the feature vectors of all points. Max pooling is a symmetric function that selects the maximum value in each dimension, thereby creating a fixed-size global feature vector that remains unchanged regardless of the order of the input points.

[0089] 5) Classification and Segmentation: For classification tasks, the global feature vector is input into a final MLP, which outputs a class prediction label. For segmentation tasks, the global feature vector is copied and concatenated with the local features of each point, and then passed through another MLP to output a class prediction label for each point.

[0090] Furthermore, after obtaining the four category prediction labels corresponding to each point and the fourth point cloud data containing the category prediction labels, the method further includes:

[0091] Based on the first rotation matrix, the second rotation matrix, and the fourth point cloud data, an inverse transformation is performed to recover the original point cloud data containing the category prediction label.

[0092] In one embodiment, the original point cloud data containing the predicted category labels can be obtained through an inverse transformation recovery operation:

[0093]

[0094] Among them, O f ′ represents the fourth point cloud data containing category prediction label data after Pointnet segmentation, O i ′ represents the original point cloud data after restoration, including category prediction label data.

[0095] It should be noted that this embodiment obtains prediction results under four background grids based on the improved PointNet++ point cloud segmentation method. Unlike existing methods, the algorithm used in this embodiment adds a point cloud normalization preprocessing head before PointNet++ to achieve more accurate segmentation results.

[0096] Step 300: Input the multiple category prediction labels corresponding to each point into the pre-trained decision tree model to obtain the target label result for each point, and determine the location of the support component in the original point cloud data based on the target label result for each point.

[0097] Step 300 specifically includes:

[0098] The four category prediction labels corresponding to each point are input into the pre-trained decision tree model to obtain the target label result for each point, and the location of the support component in the original point cloud data is determined based on the target label result for each point.

[0099] Specifically, Figure 6 This is a schematic diagram illustrating the recognition results of the tunnel support component identification and segmentation method for three-dimensional point cloud data provided in this embodiment. Figure 7 This is a schematic diagram illustrating the recognition results of the tunnel support component identification and segmentation method based on 3D point cloud data provided in this embodiment in a real-world scenario, combined with... Figure 6 and Figure 7 As explained in the above embodiments, since a total of four background grids are generated, the model will output four category prediction labels for each point in the original point cloud data. Therefore, in order to form the final segmentation result, it is necessary to vote based on the four generated category prediction labels to determine the final label result for each point.

[0100] This embodiment uses a decision tree model to implement this process. A decision tree is a tree structure. The prediction result of each background grid represents a feature attribute. The output value of this feature attribute is obtained through branches, and each leaf node stores a category (i.e., surrounding rock, pallet, metal mesh, etc.). The process of making a decision using a decision tree starts from the root node, tests the corresponding feature attribute of the item to be classified, and selects the output branch according to its value, until the leaf node is reached. The category stored in the leaf node is taken as the decision result.

[0101] This embodiment uses a decision tree model to summarize the prediction results of four types of grids and form the final point cloud segmentation result. It can realize fast and automatic pallet recognition and separation, and can be extended to the separation of other components in the roadway. Ultimately, it provides a basis for building a roadway numerical model, which facilitates the inversion of the full cross-sectional deformation of the roadway and the prevention of roadway stability disasters.

[0102] The above describes the steps of the method for identifying and segmenting roadway support components based on 3D point cloud data provided in this embodiment. As can be seen from the above description, the method for identifying and segmenting roadway support components based on 3D point cloud data provided in this embodiment acquires the original point cloud data of the target scene area and obtains multiple segmentation results corresponding to the original point cloud data under various grids based on preset segmentation rules, where one grid corresponds to one segmentation result; the original point cloud data and multiple segmentation results are input into a pre-trained deep learning model to obtain multiple category prediction labels corresponding to each point in the original point cloud data; the multiple category prediction labels corresponding to each point are input into a pre-trained decision tree model to obtain the target label result for each point, and the location of the support component in the original point cloud data is determined based on the target label result for each point. Therefore, this invention segments the original point cloud data into blocks using preset segmentation rules, reducing the point cloud size while ensuring that each support component is completely contained by at least one point cloud block, facilitating subsequent fine segmentation; the final label result of each point cloud block is determined by the decision tree model among multiple category prediction labels, achieving rapid and automatic identification and separation of support components.

[0103] The following describes the tunnel support component identification and segmentation device for three-dimensional point cloud data provided by the present invention. The tunnel support component identification and segmentation device for three-dimensional point cloud data described below can be referred to in correspondence with the tunnel support component identification and segmentation method for three-dimensional point cloud data described above.

[0104] Figure 8 This is a schematic diagram of the structure of the tunnel support component identification and segmentation device for three-dimensional point cloud data provided in this embodiment, as shown below. Figure 8 As shown, the tunnel support component identification and segmentation device for three-dimensional point cloud data provided in this embodiment includes:

[0105] The segmentation module 801 is used to acquire the original point cloud data of the target scene area and obtain multiple segmentation results of the original point cloud data under various grids based on preset segmentation rules; wherein, one grid corresponds to one segmentation result.

[0106] The prediction module 802 is used to input the original point cloud data and the multiple block results into a pre-trained deep learning model to obtain multiple category prediction labels corresponding to each point in the original point cloud data.

[0107] The determination module 803 is used to input multiple category prediction labels corresponding to each point into a pre-trained decision tree model to obtain the target label result for each point, and determine the position of the support component in the original point cloud data based on the target label result for each point.

[0108] The tunnel support component identification and segmentation device for 3D point cloud data provided in this embodiment acquires the original point cloud data of the target scene area and obtains multiple segmentation results corresponding to the original point cloud data under various grids based on preset segmentation rules, wherein one grid corresponds to one segmentation result. The original point cloud data and multiple segmentation results are input into a pre-trained deep learning model to obtain multiple category prediction labels corresponding to each point in the original point cloud data. The multiple category prediction labels corresponding to each point are input into a pre-trained decision tree model to obtain the target label result for each point, and the location of the support component in the original point cloud data is determined based on the target label result for each point. Thus, this invention segments the original point cloud data into blocks by preset segmentation rules, reducing the point cloud size, while ensuring that each support component is completely contained by at least one point cloud block, facilitating subsequent fine segmentation; the final label result of each point cloud block is determined by the decision tree model from multiple category prediction labels, achieving rapid and automatic identification and separation of support components.

[0109] Based on the above embodiments, in this embodiment, the segmentation module 801 is specifically used for:

[0110] A first grid covering the target scene area is created to obtain the first block result of the original point cloud data under the first grid; wherein, the first grid is composed of multiple squares with a side length of a first preset size;

[0111] The first grid is moved according to a preset moving rule to obtain the target grid and the target block result of the original point cloud data under the target grid.

[0112] Based on the above embodiments, in this embodiment, the segmentation module 801 is specifically used for:

[0113] The first grid is moved horizontally according to a second preset size to obtain a second grid and a second block result of the original point cloud data under the second grid; wherein, the second preset size is the maximum size of the tray, and the first preset size is twice the second preset size;

[0114] The first grid is moved vertically according to the second preset size to obtain the third grid and the third block result of the original point cloud data under the third grid;

[0115] The first grid is moved horizontally and vertically according to the second preset size to obtain the fourth grid and the fourth block result of the original point cloud data under the fourth grid.

[0116] Based on the above embodiments, in this embodiment, the device further includes a normalization module, specifically used for:

[0117] Before inputting the raw point cloud data and the multiple block results into the pre-trained deep learning model, a preprocessing operation is performed on the raw point cloud data through a normalization preprocessing head; wherein, the normalization preprocessing head is used to perform normalization processing on the raw point cloud data.

[0118] Based on the above embodiments, in this embodiment, the normalization preprocessing head includes a translation layer, a principal component analysis layer, and an L1 norm normalization layer;

[0119] The normalization module is specifically used for:

[0120] The original point cloud data is input into the translation layer to obtain the first point cloud data;

[0121] The first point cloud data is input into the principal component analysis layer to obtain the second point cloud data and the corresponding first rotation matrix;

[0122] The second point cloud data is input into the L1 norm normalization layer to obtain the third point cloud data and the corresponding second rotation matrix.

[0123] The prediction module 802 is specifically used for:

[0124] The third point cloud data and the four block results are input into a pre-trained deep learning model to obtain four category prediction labels for each point and a fourth point cloud data containing category prediction labels after segmentation.

[0125] Based on the above embodiments, in this embodiment, the device further includes a recovery module, specifically used for:

[0126] After obtaining the four category prediction labels corresponding to each point and the fourth point cloud data containing the category prediction labels, an inverse transformation is performed based on the first rotation matrix, the second rotation matrix, and the fourth point cloud data to recover the original point cloud data containing the category prediction labels.

[0127] Based on the above embodiments, in this embodiment, the determining module 803 is specifically used for:

[0128] The four category prediction labels corresponding to each point are input into the pre-trained decision tree model to obtain the target label result for each point, and the location of the support component in the original point cloud data is determined based on the target label result for each point.

[0129] Figure 9 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 9As shown, the electronic device may include: a processor 910, a communication interface 920, a memory 930, and a communication bus 940, wherein the processor 910, the communication interface 920, and the memory 930 communicate with each other through the communication bus 940. The processor 910 can call logical instructions in the memory 930 to execute a method for identifying and segmenting roadway support components based on three-dimensional point cloud data. This method includes:

[0130] The original point cloud data of the target scene area is acquired, and multiple block results corresponding to the original point cloud data under various grids are obtained based on preset block rules; wherein, one grid corresponds to one block result.

[0131] The original point cloud data and the multiple block results are input into a pre-trained deep learning model to obtain multiple category prediction labels corresponding to each point in the original point cloud data.

[0132] The predicted labels of multiple categories corresponding to each point are input into a pre-trained decision tree model to obtain the target label result for each point, and the location of the support component in the original point cloud data is determined based on the target label result for each point.

[0133] Furthermore, the logical instructions in the aforementioned memory 930 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0134] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the tunnel support component identification and segmentation method for three-dimensional point cloud data provided by the above methods. The method includes:

[0135] The original point cloud data of the target scene area is acquired, and multiple block results corresponding to the original point cloud data under various grids are obtained based on preset block rules; wherein, one grid corresponds to one block result.

[0136] The original point cloud data and the multiple block results are input into a pre-trained deep learning model to obtain multiple category prediction labels corresponding to each point in the original point cloud data.

[0137] The predicted labels of multiple categories corresponding to each point are input into a pre-trained decision tree model to obtain the target label result for each point, and the location of the support component in the original point cloud data is determined based on the target label result for each point.

[0138] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for identifying and segmenting roadway support components based on three-dimensional point cloud data provided by the methods described above, the method comprising:

[0139] The original point cloud data of the target scene area is acquired, and multiple block results corresponding to the original point cloud data under various grids are obtained based on preset block rules; wherein, one grid corresponds to one block result.

[0140] The original point cloud data and the multiple block results are input into a pre-trained deep learning model to obtain multiple category prediction labels corresponding to each point in the original point cloud data.

[0141] The predicted labels of multiple categories corresponding to each point are input into a pre-trained decision tree model to obtain the target label result for each point, and the location of the support component in the original point cloud data is determined based on the target label result for each point.

[0142] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0143] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying and segmenting roadway support components based on 3D point cloud data, characterized in that, include: The original point cloud data of the target scene area is acquired, and multiple block results corresponding to the original point cloud data under various grids are obtained based on preset block rules; wherein, one grid corresponds to one block result. The original point cloud data and the multiple block results are input into a pre-trained deep learning model to obtain multiple category prediction labels corresponding to each point in the original point cloud data. The multiple category prediction labels corresponding to each point are input into a pre-trained decision tree model to obtain the target label result for each point, and the location of the support component in the original point cloud data is determined based on the target label result for each point. The process of obtaining multiple block results of the original point cloud data under various grids based on preset block rules includes: A first grid covering the target scene area is created to obtain the first block result of the original point cloud data under the first grid; wherein, the first grid is composed of multiple squares with a side length of a first preset size; The first grid is moved according to a preset moving rule to obtain the target grid and the target block result of the original point cloud data under the target grid; The step of moving the first grid according to a preset moving rule to obtain the target grid and the target block result of the original point cloud data under the target grid includes: The first grid is moved horizontally according to a second preset size to obtain a second grid and a second block result of the original point cloud data under the second grid; wherein, the second preset size is the maximum size of the tray, and the first preset size is twice the second preset size; The first grid is moved vertically according to the second preset size to obtain the third grid and the third block result of the original point cloud data under the third grid; The first grid is moved horizontally and vertically according to the second preset size to obtain the fourth grid and the fourth block result of the original point cloud data under the fourth grid.

2. The method for identifying and segmenting roadway support components based on three-dimensional point cloud data according to claim 1, characterized in that, Before inputting the raw point cloud data and the multiple block results into the pre-trained deep learning model, the method further includes: The original point cloud data is preprocessed using a normalization preprocessing head; wherein, the normalization preprocessing head is used to normalize the original point cloud data.

3. The method for identifying and segmenting tunnel support components based on three-dimensional point cloud data according to claim 2, characterized in that, The normalization preprocessing head includes a translation layer, a principal component analysis layer, and an L1 norm normalization layer; The preprocessing operation performed on the raw point cloud data through the normalization preprocessing head includes: The original point cloud data is input into the translation layer to obtain the first point cloud data; The first point cloud data is input into the principal component analysis layer to obtain the second point cloud data and the corresponding first rotation matrix; The second point cloud data is input into the L1 norm normalization layer to obtain the third point cloud data and the corresponding second rotation matrix. The step of inputting the original point cloud data and the multiple block results into a pre-trained deep learning model to obtain multiple category prediction labels corresponding to each point in the original point cloud data includes: The third point cloud data and the four block results are input into a pre-trained deep learning model to obtain four category prediction labels for each point and a fourth point cloud data containing category prediction labels after segmentation.

4. The method for identifying and segmenting roadway support components based on three-dimensional point cloud data according to claim 3, characterized in that, After obtaining the four category prediction labels corresponding to each point and the fourth point cloud data containing the category prediction labels, the method further includes: Based on the first rotation matrix, the second rotation matrix, and the fourth point cloud data, an inverse transformation is performed to recover the original point cloud data containing the category prediction label.

5. The method for identifying and segmenting tunnel support components based on three-dimensional point cloud data according to claim 1, characterized in that, The step of inputting multiple category prediction labels corresponding to each point into a pre-trained decision tree model to obtain the target label result for each point, and determining the location of the support component in the original point cloud data based on the target label result for each point, includes: The four category prediction labels corresponding to each point are input into the pre-trained decision tree model to obtain the target label result for each point, and the location of the support component in the original point cloud data is determined based on the target label result for each point.

6. A device for identifying and segmenting tunnel support components based on three-dimensional point cloud data, characterized in that, include: The segmentation module is used to acquire the original point cloud data of the target scene area and obtain multiple segmentation results of the original point cloud data under various grids based on preset segmentation rules; wherein, one grid corresponds to one segmentation result. The prediction module is used to input the original point cloud data and the multiple block results into a pre-trained deep learning model to obtain multiple category prediction labels corresponding to each point in the original point cloud data. The determination module is used to input multiple category prediction labels corresponding to each point into a pre-trained decision tree model to obtain the target label result for each point, and determine the location of the support component in the original point cloud data based on the target label result for each point. The segmentation module is specifically used for: A first grid covering the target scene area is created to obtain the first block result of the original point cloud data under the first grid; wherein, the first grid is composed of multiple squares with a side length of a first preset size; The first grid is moved according to a preset moving rule to obtain the target grid and the target block result of the original point cloud data under the target grid; The segmentation module is further used for: The first grid is moved horizontally according to a second preset size to obtain a second grid and a second block result of the original point cloud data under the second grid; wherein, the second preset size is the maximum size of the tray, and the first preset size is twice the second preset size; The first grid is moved vertically according to the second preset size to obtain the third grid and the third block result of the original point cloud data under the third grid; The first grid is moved horizontally and vertically according to the second preset size to obtain the fourth grid and the fourth block result of the original point cloud data under the fourth grid.

7. An electronic 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, it implements the method for identifying and segmenting roadway support components based on three-dimensional point cloud data as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for identifying and segmenting roadway support components based on three-dimensional point cloud data as described in any one of claims 1 to 5.

Images