Pipeline facility reconstruction and identification method based on three-dimensional laser point cloud

Abstract

The invention discloses a pipeline facility reconstruction and identification method based on a three-dimensional laser point cloud, and the method comprises the steps: obtaining to-be-reconstructed three-dimensional point cloud data, carrying out the sampling preprocessing of the three-dimensional point cloud data, and obtaining sampling point cloud data and first mapping information; combining and partitioning the sampling point cloud data to obtain a plurality of point cloud sample blocks and second mapping information, the number of point clouds in each point cloud sample block being the same, and sample points in each point cloud sample block being uniformly distributed in a point cloud space of the sampling point cloud data; according to the second mapping information, performing point cloud reconstruction on the point cloud sample block through a point cloud classification model to obtain point cloud reconstruction data; and sampling and restoring the point cloud reconstruction data according to the first mapping information to obtain a target point cloud reconstruction result. According to the method, the complexity of point cloud data processing can be effectively reduced, computing resources required by point cloud reconstruction are reduced, and the point cloud reconstruction effect and precision are improved. The invention relates to the technical field of point cloud data processing.

Description

Technical Field

[0001] The invention relates to the technical field of point cloud data processing, and in particular to a pipeline facility reconstruction and recognition method based on three-dimensional laser point cloud. Background Art

[0002] With the construction and development of modern cities, laser 3D point clouds can quickly record the latest spatial status of urban underground pipeline facilities in important tasks such as urban planning, construction and maintenance. Therefore, how to quickly identify and segment underground facility target objects from complex scenes based on the massive point cloud data scanned by LiDAR, and provide an important input source for the next step of 3D information extraction, has become an urgent issue to be solved.

[0003] At present, before reconstructing underground pipelines through point cloud data, it is necessary to perform object recognition on the point cloud data objects (i.e., data semantic segmentation). For semantic segmentation of point cloud data, traditional segmentation methods have their limitations, and it is difficult to complete the semantic segmentation task of point cloud data with huge point cloud quantities and complex scenes. Specifically, traditional point cloud data segmentation methods can be roughly divided into two types. The first is a plane-based block segmentation method, which often cannot effectively capture the overall geometric structure of the object, resulting in unsatisfactory point cloud reconstruction results; the second is a probability-based random segmentation method, which is more complicated to process point cloud data, requires more computing resources, and has low point cloud reconstruction accuracy.

[0004] Therefore, the problems existing in the prior art still need to be solved and optimized. Summary of the invention

[0005] The purpose of the present invention is to solve one of the technical problems existing in the related art to at least a certain extent.

[0006] To this end, one purpose of an embodiment of the present invention is to provide a pipeline facility reconstruction and identification method based on three-dimensional laser point cloud, wherein the method can effectively reduce the complexity of point cloud data processing, reduce the computing resources required for point cloud reconstruction, and improve the point cloud reconstruction effect and accuracy.

[0007] In order to achieve the above technical objectives, the technical solutions adopted by the embodiments of the present application include:

[0008] In a first aspect, an embodiment of the present application provides a pipeline facility reconstruction and identification method based on a three-dimensional laser point cloud, comprising:

[0009] Acquire three-dimensional point cloud data to be reconstructed, and perform sampling preprocessing on the three-dimensional point cloud data to obtain sampled point cloud data and first mapping information, wherein the first mapping information is used to characterize a point cloud mapping relationship between the three-dimensional point cloud data before sampling and the three-dimensional point cloud data after sampling;

[0010] The sampled point cloud data are combined and divided into blocks to obtain a plurality of point cloud sample blocks and second mapping information, wherein the number of point clouds in each of the point cloud sample blocks is the same, and the sample points in each of the point cloud sample blocks are evenly distributed in the point cloud space of the sampled point cloud data, and the second mapping information is used to characterize the point cloud mapping relationship between the point cloud sample blocks and the sampled point cloud data;

[0011] Based on the second mapping information, reconstructing the point cloud sample block through the trained point cloud classification model to obtain point cloud reconstruction data;

[0012] The point cloud reconstruction data is sampled and restored according to the first mapping information to obtain a target point cloud reconstruction result.

[0013] In addition, the method according to the above embodiment of the present application may also have the following additional technical features:

[0014] Furthermore, in one embodiment of the present application, the sampling preprocessing of the three-dimensional point cloud data to obtain the sampled point cloud data and the first mapping information includes:

[0015] Performing data cleaning on the three-dimensional point cloud data to obtain first point cloud data;

[0016] Performing three-dimensional grid division on the first point cloud data to obtain second point cloud data;

[0017] Grid sampling is performed on the second point cloud data to obtain the sampled point cloud data and the first mapping information.

[0018] Furthermore, in one embodiment of the present application, the sampled point cloud data is combined and divided into blocks to obtain a plurality of point cloud sample blocks and second mapping information, including:

[0019] Performing point cloud farthest point sampling on the sampled point cloud data to obtain third point cloud data;

[0020] According to the sampled point cloud data, point cloud nearest neighbor classification is performed on the third point cloud data to obtain a plurality of the point cloud sample blocks and the second mapping information.

[0021] Further, in one embodiment of the present application, the step of sampling the farthest point of the point cloud on the sampled point cloud data to obtain the third point cloud data includes:

[0022] Determining a target sampling quantity according to the sampling point cloud data;

[0023] According to the first point cloud sampling sequence, the remaining sampling point cloud data after sampling is screened for the maximum distance to obtain the distance optimum point;

[0024] According to the distance optimal point, updating the sampling points of the first point cloud sampling sequence to obtain a second point cloud sampling sequence;

[0025] Comparing the target sampling quantity with the number of sampling points in the second point cloud sampling sequence to obtain a sampling comparison result;

[0026] If the sampling comparison result is that the number of sampling points in the second point cloud sampling sequence is less than the target sampling number, the first point cloud sampling sequence is updated according to the second point cloud sampling sequence, and then the step of performing maximum distance screening on the remaining sampling point cloud data after sampling according to the first point cloud sampling sequence to obtain the distance optimum point is returned; or, if the sampling comparison result is that the number of sampling points in the second point cloud sampling sequence is equal to the target sampling number, the second point cloud sampling sequence is determined as the third point cloud data.

[0027] Further, in one embodiment of the present application, performing point cloud neighbor classification on the third point cloud data according to the sampled point cloud data to obtain a plurality of the point cloud sample blocks and the second mapping information includes:

[0028] Acquire a tree structure corresponding to the sampling point cloud data and a preset number of sample points;

[0029] According to the number of sample points and the third point cloud data, a neighbor point search is performed on the tree structure to obtain a plurality of the point cloud sample blocks and the second mapping information.

[0030] Further, in one embodiment of the present application, based on the second mapping information, performing point cloud reconstruction on the point cloud sample block through a trained point cloud classification model to obtain point cloud reconstruction data includes:

[0031] Inputting the point cloud sample block into the trained point cloud classification model to perform point cloud classification prediction to obtain a point cloud classification result;

[0032] According to the second mapping information, index averaging is performed on the point cloud classification result to obtain a reconstruction prediction result;

[0033] The category probability value corresponding to each sample point of the reconstruction prediction result is screened for the maximum value to obtain the point cloud reconstruction data.

[0034] Further, in the embodiment of the present application, the trained point cloud classification model is obtained by training through the following steps:

[0035] Get training sample blocks;

[0036] Performing data enhancement and normalization on the training sample block to obtain a target sample block;

[0037] The target sample block is input into the initialized point cloud classification model for training to obtain the trained point cloud classification model.

[0038] In a second aspect, the embodiment of the present application provides a pipeline facility reconstruction and identification system based on three-dimensional laser point cloud, including:

[0039] A first processing unit is used to obtain the three-dimensional point cloud data to be reconstructed, and perform sampling preprocessing on the three-dimensional point cloud data to obtain sampled point cloud data and first mapping information, wherein the first mapping information is used to characterize a point cloud mapping relationship between the three-dimensional point cloud data before sampling and the three-dimensional point cloud data after sampling;

[0040] A second processing unit is used to perform combined block processing on the sampled point cloud data to obtain a plurality of point cloud sample blocks and second mapping information, wherein the number of point clouds in each of the point cloud sample blocks is the same, and the sample points in each of the point cloud sample blocks are evenly distributed in the point cloud space of the sampled point cloud data, and the second mapping information is used to characterize the point cloud mapping relationship between the point cloud sample blocks and the sampled point cloud data;

[0041] A third processing unit is configured to perform point cloud reconstruction on the point cloud sample block based on the second mapping information by using a trained point cloud classification model to obtain point cloud reconstruction data;

[0042] The fourth processing unit is used to sample and restore the point cloud reconstruction data according to the first mapping information to obtain a target point cloud reconstruction result.

[0043] In a third aspect, an embodiment of the present application further provides an electronic device, including:

[0044] at least one processor;

[0045] at least one memory for storing at least one program;

[0046] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.

[0047] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, which stores a program executable by a processor, and the program executable by the processor is used to implement the above method when executed by the processor.

[0048] The advantages and benefits of the present application will be partially given in the following description, and partially become apparent from the following description, or be understood through the practice of the present application:

[0049] The embodiment of the present application discloses a pipeline facility reconstruction and identification method based on three-dimensional laser point cloud, wherein the method obtains three-dimensional point cloud data to be reconstructed, and performs sampling preprocessing on the three-dimensional point cloud data to obtain sampled point cloud data and first mapping information, wherein the first mapping information is used to characterize the point cloud mapping relationship between the three-dimensional point cloud data before sampling and the three-dimensional point cloud data after sampling; the sampled point cloud data is combined and divided into blocks to obtain a plurality of point cloud sample blocks and second mapping information, wherein the number of point clouds in each of the point cloud sample blocks is the same, and the sample points in each of the point cloud sample blocks are uniformly distributed in the point cloud space of the sampled point cloud data, and the second mapping information is used to characterize the point cloud mapping relationship between the point cloud sample block and the sampled point cloud data; based on the second mapping information, the point cloud sample block is point cloud reconstructed by a trained point cloud classification model to obtain point cloud reconstructed data; according to the first mapping information, the point cloud reconstructed data is sampled and restored to obtain a target point cloud reconstruction result. This method is based on the combined blocking of sampled point cloud data, and can obtain point cloud sample blocks in which sample points are evenly distributed in the point cloud space. The number of sample points in each point cloud sample block is the same, which can avoid the randomness problem in the selection of sampling points. The complexity of point cloud data processing is low, and the required computing resources are less, which can effectively improve the accuracy and effect of point cloud reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the embodiments of the present application or the drawings of the related technical solutions in the prior art are introduced below. It should be understood that the drawings introduced below are only for the convenience of clearly expressing some embodiments of the technical solutions of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0051] Figure 1 A schematic diagram of a process flow of a pipeline facility reconstruction and identification method based on three-dimensional laser point cloud provided in an embodiment of the present application;

[0052] Figure 2 A schematic diagram of the structural framework of a pipeline facility reconstruction and identification system based on three-dimensional laser point cloud provided in an embodiment of the present application;

[0053] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and are not to be construed as limitations on the present application. For the step numbers in the following embodiments, they are only provided for the convenience of explanation, and the order between the steps is not limited in any way, and the execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0056] At present, the reconstruction of underground pipelines through point cloud data is mainly based on deep learning. Before reconstructing the point cloud of underground pipelines, it is necessary to perform data semantic segmentation on the point cloud data objects, and then input the semantically segmented point cloud data into the model used to achieve point cloud reconstruction of underground pipelines.

[0057] The traditional point cloud data segmentation methods can be roughly divided into two types. The first is a plane-based block segmentation method, and the second is a probability-based random segmentation method. These segmentation methods have certain limitations, and it is difficult to complete the semantic segmentation task of point cloud data with huge point cloud quantities and complex scenes.

[0058] Specifically, the first plane-based block segmentation method usually divides the point cloud data into many cubes, and takes a fixed number of points in each cube. However, when the scene space is large (such as an outdoor scene space), the number of cubes input to the PointNet model will be very large, requiring more computing resources, and it is difficult to capture the overall geometric structure of objects in the scene space. It is easy to fail to capture the effective overall geometric structure of the object, which often makes the point cloud reconstruction effect unsatisfactory.

[0059] The second is a probability-based random segmentation method. Usually, in large-scale point cloud data (data scale such as millions, tens of millions, etc.), an initial sampling value is determined for each data point; then the data point with the smallest sampling value is selected as the first control point, and then the K nearest neighbor method is used to find N data points around the first control point as the model input. Then the sampling value of these N data points will increase, and the size of the increase is related to the distance from the N data points to the first control point. Then the point with the smallest sampling value is selected as the second control point, and so on. Although this method can effectively cut the point cloud data, the point cloud data processing of this method is more complicated, the required computing resources are more, and the randomness of the obtained point cloud data blocks is relatively large. It is difficult to ensure that all point cloud data are used for subsequent model (such as KPConv model) training or reconstruction (such as random problems caused by random sampling, some data points in the point cloud data are not involved in model training or reconstruction), which makes the point cloud reconstruction accuracy low.

[0060] In view of this, an embodiment of the present invention provides a pipeline facility reconstruction and identification method based on three-dimensional laser point cloud. The method is based on the combined blocking of sampled point cloud data, and can obtain point cloud sample blocks in which sample points are evenly distributed in the point cloud space. The number of sample points in each point cloud sample block is the same, which can effectively reduce the complexity of point cloud data processing and reduce the required computing resources. In addition, the method can also avoid the randomness problem in the selection of sampling points, so that each data point in the point cloud data can participate in the model training or the model's reconstruction process of the point cloud, which can effectively improve the accuracy of point cloud reconstruction.

[0061] Reference Figure 1 In an embodiment of the present application, a pipeline facility reconstruction and identification method based on three-dimensional laser point cloud includes:

[0062] Step 110: Acquire the three-dimensional point cloud data to be reconstructed, and perform sampling preprocessing on the three-dimensional point cloud data to obtain sampled point cloud data and first mapping information, wherein the first mapping information is used to characterize the point cloud mapping relationship between the three-dimensional point cloud data before sampling and the three-dimensional point cloud data after sampling;

[0063] In the embodiment of the present application, the 3D point cloud data to be reconstructed can be obtained by laser radar scanning, and specifically can be collected by using a base station laser scanner or a portable laser scanner in the ground excavation state of the underground pipe network. During the collection process, due to the influence of the construction environment, the collected 3D point cloud data often includes background points, pipeline points and pipeline accessory points (such as valves) in the excavation scene.

[0064] In some embodiments, the step 110 of performing sampling preprocessing on the three-dimensional point cloud data to obtain sampled point cloud data and first mapping information includes:

[0065] A1. Cleaning the three-dimensional point cloud data to obtain first point cloud data;

[0066] A2. Perform three-dimensional grid division on the first point cloud data to obtain second point cloud data;

[0067] A3. Perform grid sampling on the second point cloud data to obtain the sampled point cloud data and the first mapping information.

[0068] In an embodiment of the present application, for a certain n1 rows and d columns of three-dimensional point cloud data, each row of data in the three-dimensional point cloud data represents a data point in the three-dimensional point cloud data, and each column of data in the three-dimensional point cloud data represents a characteristic information of the point cloud. Specifically, step A1 may firstly perform missing value verification on the three-dimensional point cloud data, wherein the missing value verification is used to discard data points with missing characteristic information in the three-dimensional point cloud data, retain data points with complete characteristic information in the three-dimensional point cloud data, obtain the three-dimensional point cloud data after missing value verification, and record the three-dimensional point cloud data after missing value verification as the first intermediate data; then, a preset quantity threshold may be obtained, and the relationship between the quantity threshold and the number of data points of the first intermediate data may be compared, and the first intermediate data with the number of data points greater than or equal to the quantity threshold may be retained, and the first intermediate data with the number of data points greater than or equal to the quantity threshold may be recorded as the second intermediate data; then, data points with missing or abnormal data colors in the second intermediate data may be removed, which may be specifically achieved by point cloud data processing software (such as CloudCompare) or by establishing an automatic recognition program, thereby obtaining the first point cloud data with n2 rows and d columns.

[0069] It can be understood that step A2 can be based on the grid unit to downsample the first point cloud data, so as to realize the three-dimensional grid division of the first point cloud data and obtain the second point cloud data. Specifically, it can be based on the three-dimensional space of the first point cloud data, and a number of uniform grid units are determined, each grid unit includes a number of data points, and the size of the specific grid unit can be preset; step A3 specifically samples the data points in each grid unit into a grid sampling point through a statistical method, and specifically the three-dimensional coordinate information (such as x-axis coordinate information, y-axis coordinate information, y-axis coordinate information) and color channel information (such as R color information, G color information, B color information) of all data points in the grid unit can be averaged to obtain the three-dimensional coordinate information and color channel information of the grid sampling point, thereby obtaining the first mapping information and the sampling point cloud data of n3 rows and d columns, the number of rows n3 of the sampling point cloud data is less than the number of rows n2 of the first point cloud data, and the first mapping information is used to characterize the point cloud mapping relationship between the three-dimensional point cloud data before sampling and the three-dimensional point cloud data after sampling, and specifically can be the point cloud mapping relationship between multiple data points in the grid unit and the corresponding grid sampling points.

[0070] Step 120: Combining and dividing the sampled point cloud data into blocks to obtain a plurality of point cloud sample blocks and second mapping information, wherein the number of point clouds in each of the point cloud sample blocks is the same, and the sample points in each of the point cloud sample blocks are evenly distributed in the point cloud space of the sampled point cloud data, and the second mapping information is used to characterize the point cloud mapping relationship between the point cloud sample blocks and the sampled point cloud data;

[0071] In an embodiment of the present application, the combined block processing of step 120 is used to divide the sampled point cloud data of n3 rows and d columns into a number of point cloud sample blocks, each point cloud sample block includes n4 rows and d columns of point cloud data, and the second mapping information is used to characterize the point cloud mapping relationship between the point cloud sample block and the sampled point cloud data, specifically, it can be the point cloud mapping relationship between multiple data points of the point cloud sample block and a certain data point in the sampled point cloud data.

[0072] In some embodiments, the step 120 of combining and dividing the sampled point cloud data into blocks to obtain a plurality of point cloud sample blocks and second mapping information includes:

[0073] B1. Sampling the farthest point of the point cloud on the sampled point cloud data to obtain third point cloud data;

[0074] Furthermore, the step B1, sampling the farthest point of the point cloud on the sampled point cloud data to obtain third point cloud data, includes:

[0075] B11. Determine the target sampling quantity according to the sampling point cloud data;

[0076] B12. According to the first point cloud sampling sequence, the remaining sampling point cloud data after sampling is screened for maximum distance to obtain the distance optimum point;

[0077] B13. According to the optimal distance point, updating the sampling points of the first point cloud sampling sequence to obtain a second point cloud sampling sequence;

[0078] B14, comparing the target sampling quantity and the number of sampling points in the second point cloud sampling sequence to obtain a sampling comparison result;

[0079] B15. If the sampling comparison result is that the number of sampling points in the second point cloud sampling sequence is less than the target sampling number, the first point cloud sampling sequence is updated according to the second point cloud sampling sequence, and then the step of performing maximum distance screening on the remaining sampling point cloud data after sampling according to the first point cloud sampling sequence to obtain the distance optimum point is returned; or, if the sampling comparison result is that the number of sampling points in the second point cloud sampling sequence is equal to the target sampling number, the second point cloud sampling sequence is determined as the third point cloud data.

[0080] In an embodiment of the present application, step B11 can determine the target sampling number based on the number of data points of the sampling point cloud data. There are many specific determination methods. For example, the target sampling number can be determined based on the ratio between the number of data points of the three-dimensional point cloud data and the number of data points of the sampling point cloud data. This application will not go into details here.

[0081] It can be understood that, for step B12, if the number of sampling points in the first point cloud sampling sequence is equal to 1, the sampling points in the first point cloud sampling sequence can be randomly selected from the sampled point cloud data; or, if the number of sampling points in the first point cloud sampling sequence is greater than or equal to 2, the sampling points in the second point cloud sampling sequence can be the second point cloud sampling sequence in the previous sampling process. In addition, the maximum distance screening in step B12 can be to calculate the distance between all sampling points in the first point cloud sampling sequence and the remaining data points of the sampled point cloud data (i.e., the remaining sampled point cloud data after sampling). The specific distance calculation method can be Euclidean distance, etc., and then the data point corresponding to the calculated maximum distance is determined as the distance optimal point.

[0082] It should be noted that the updating of the sampling points in step B13 may be to add the optimal distance point to the first point cloud sampling sequence, thereby obtaining the second point cloud sampling sequence. Steps B14 and B15 may be to compare the size relationship between the target sampling number and the number of sampling points in the second point cloud sampling sequence. If the number of sampling points in the second point cloud sampling sequence is less than the target sampling number, the second point cloud sampling sequence of the current sampling process may be determined as the first point cloud sampling sequence of the next sampling process, and then return to execute step B12; or, if the number of sampling points in the second point cloud sampling sequence is equal to the target sampling number, it means that the second point cloud sampling sequence has sampled the required number of uniformly distributed data points, and at this time, the third point cloud data may be determined based on each sampling point in the second point cloud sampling sequence of the current sampling process.

[0083] B2. Perform point cloud nearest neighbor classification on the third point cloud data according to the sampled point cloud data to obtain a plurality of the point cloud sample blocks and the second mapping information.

[0084] Furthermore, the step B2, performing point cloud nearest neighbor classification on the third point cloud data according to the sampled point cloud data to obtain a plurality of the point cloud sample blocks and the second mapping information, includes:

[0085] B21, obtaining a tree structure corresponding to the sampling point cloud data and a preset number of sample points;

[0086] B22. Perform neighbor point search on the tree structure according to the number of sample points and the third point cloud data to obtain a plurality of point cloud sample blocks and the second mapping information.

[0087] In an embodiment of the present application, step B21 can be to construct a tree structure based on the sampling point cloud data, and the tree structure can specifically be a KD tree structure (KD tree). The preset number of sample points can be the number of nearest neighbor points corresponding to each data point in the third point cloud data, and the number of sample points is the same as the aforementioned target sampling number.

[0088] It can be understood that, for a certain data point in the third point cloud data, step B22 can be based on the coordinates of the data point and the required number of sample points, and a KD tree structure is recursively searched layer by layer to search for a number of nearest neighbor points corresponding to the data point, and the total number of the nearest neighbor points is equal to the number of sample points; then the data point and all the searched nearest neighbor points are determined as a point cloud sample block, and based on the correspondence between the data points of the point cloud sample block and the sampled point cloud data, the second mapping information is determined.

[0089] It should be noted that the sampling of the farthest point of the point cloud in step B1 is used to generate uniformly distributed control points (i.e., data points in the third point cloud data). Specifically, when processing point cloud data, random sampling directly from the original data may cause the sampling points to be unevenly distributed in space, which will cause too much data in some areas and insufficient data in other areas; while the sampling of the farthest point of the point cloud in the embodiment of the present application gradually selects the point farthest from the current set of sampled points to ensure that the new sampling points are as far away from the existing sampling points as possible, so that the control points are uniform in the coverage range of the point cloud. This uniform distribution of control points can not only represent the spatial distribution of the entire sampled point cloud data, reduce data redundancy, and improve the representativeness of sampling. In addition, this uniform distribution of control points can effectively avoid the phenomenon of dense density in some areas and sparse density in other areas, so that the data points contained in the point cloud sample block generated in step B2 are evenly distributed in the entire point cloud space. Step B2 point cloud neighbor classification is used to find the nearest neighbor points around each control point based on the control points that have been generated to form a uniform sub-data block (i.e., point cloud sample block). In addition, the embodiment of the present application, based on step B1 and step B2, can effectively improve the impact of the problem of uneven segmentation of point cloud data, and can reduce the sample differences of point cloud data while reducing the computational complexity of point cloud data processing, thereby avoiding the subsequent model from being biased towards certain local areas in the point cloud data during the training process, thereby improving the model robustness and the accuracy of point cloud reconstruction based on the model. In addition, the embodiment of the present application, based on step B1 and step B2, specifically avoids the potential deviation of random sampling (i.e., randomness problem) through the farthest point sampling in step B1, and ensures that the number of point clouds in each point cloud sample block is the same through the neighbor point search in step B2, so that each sampling point in the point cloud data can participate in the training process of the subsequent model, that is, avoids the situation where some sampling points do not participate in the model training process due to randomness problems, which can further improve the training efficiency of the point cloud classification model, as well as the point cloud reconstruction effect and accuracy.

[0090] It should be added that the embodiment of the present application can effectively avoid the potential deviation of random sampling (i.e., the randomness problem) based on the farthest point sampling in step B1.

[0091] Step 130: Based on the second mapping information, reconstruct the point cloud sample block through the trained point cloud classification model to obtain point cloud reconstruction data;

[0092] In an embodiment of the present application, step 130 may be to input the point cloud sample block into a trained point cloud classification model, output the predicted category corresponding to each point cloud sample block through the trained point cloud classification model, and then determine the point cloud reconstruction data corresponding to the sampled point cloud data based on each predicted category and the corresponding second mapping information.

[0093] In some embodiments, the step 130, based on the second mapping information, reconstructing the point cloud sample block by using a trained point cloud classification model to obtain point cloud reconstruction data, includes:

[0094] C1. Inputting the point cloud sample block into the trained point cloud classification model to perform point cloud classification prediction and obtain a point cloud classification result;

[0095] C2. According to the second mapping information, index averaging is performed on the point cloud classification result to obtain a reconstruction prediction result;

[0096] C3. Perform maximum value screening on the category probability value corresponding to each sample point of the reconstruction prediction result to obtain the point cloud reconstruction data.

[0097] In the embodiment of the present application, step C1 may be to input the obtained point cloud sample block into the trained point cloud classification model for point cloud classification prediction, and obtain the predicted category (i.e., point cloud classification result) output by the trained point cloud classification model, and the point cloud classification result generally includes multiple predicted categories and scores or probability values ​​corresponding to each predicted category. Specifically, for a point cloud sample block with n4 rows and d columns, the point cloud classification model may output a point cloud classification result with n4 rows and c columns, wherein the c column in the point cloud classification result is the total number of categories required to be classified by the point cloud classification model, and each column in the point cloud classification result is used to record the category probability of the point cloud sample block data point in the column.

[0098] It can be understood that since the sampled point cloud data will be divided into multiple point cloud sample blocks after combined block processing, there may be repeated data points between the point cloud sample blocks. After the multiple point cloud sample blocks are respectively input into the trained point cloud classification model for point cloud classification prediction, there may be data points with multiple point cloud classification predictions in the multiple point cloud classification results obtained. Therefore, step C2 can be based on the second mapping information to determine the multiple data points corresponding to each point cloud classification result in the sampled point cloud data; then, for a data point with multiple point cloud classification predictions, the multiple point cloud classification results corresponding to this data point can be averaged to obtain the overall classification result corresponding to the data point, and the other data points with multiple point cloud classification predictions are similar, which can be simply deduced; then, the overall classification results corresponding to the data points with multiple point cloud classification predictions and the point cloud classification results corresponding to the data points without multiple point cloud classification predictions are integrated to obtain the reconstructed prediction results.

[0099] It should be noted that step C3 can be to screen the maximum value of the category probability value corresponding to each sample point in the reconstructed prediction result. Specifically, for a certain data point in the reconstructed prediction result, it can be to screen the numerical values ​​between multiple prediction categories corresponding to the data point, and determine the prediction category with the largest score or probability value as the prediction category corresponding to the data point. The same is true for the remaining data points in the reconstructed prediction result, so as to obtain point cloud reconstruction data.

[0100] Exemplarily, the point cloud classification model in the embodiment of the present application can be a RandLA-Net network model, whose structure is a U-Net point cloud segmentation network, which mainly includes a coding layer, a downsampling layer, a decoding layer and an upsampling layer. The method used by the RandLA-Net network in the downsampling link is random downsampling with the fastest computing performance. In order to make up for the greater information loss caused by random downsampling, the encoding layer of the network model first divides the point cloud into different local areas, and then performs feature aggregation on the points in each local area. Then, multi-scale feature extraction is performed on the aggregated local features through a multi-layer neural network to obtain a feature representation with global perception. Finally, each data point is classified or segmented through a fully connected layer, and the corresponding point cloud classification result is output.

[0101] It should be added that, in actual applications, corresponding point cloud classification models can be constructed for specific pipeline application scenarios. Specifically, the embodiments of the present application can construct corresponding point cloud classification models for each pipeline application scenario, such as a point cloud classification model corresponding to a sewage pipeline scene (DSD-Drainage Service Department), a point cloud classification model corresponding to a drinking water pipeline scene (WSD-Water Supply Department), a point cloud classification model corresponding to a simple composite scene (DSD and WSD), and a point cloud classification model corresponding to a detailed mixed scene. The examples in this application are for illustration only and are not intended to limit the present application.

[0102] Furthermore, the trained point cloud classification model is obtained by training through the following steps:

[0103] D1, obtain training sample blocks;

[0104] D2. Performing data enhancement and normalization on the training sample block to obtain a target sample block;

[0105] D3. Input the target sample block into the initialized point cloud classification model for training to obtain the trained point cloud classification model.

[0106] In the embodiment of the present application, the training sample block in step D1 may be the aforementioned point cloud sample block, or may be a point cloud sample block obtained after the new training point cloud data is processed through the aforementioned steps 110 and 120. The data enhancement and normalization in step D2 may be to normalize the color channel information of each data point in the training sample block to the range of (0, 1), and to perform center normalization on the three-dimensional coordinate information of each data point in the training sample block, so as to obtain a normalized training sample block; then, perform random translation, random rotation and other operations on the normalized training sample block, so as to obtain a target sample block.

[0107] It is understandable that before the point cloud classification model is put into use, it needs to be trained to adjust its internal parameters so as to achieve a better classification prediction effect. Specifically, the target sample block can be input into the point cloud classification model, and the target sample block can be predicted by the model, and the output data of the model is the classification result of the target sample block. After obtaining the classification result of the target sample block output by the model, the accuracy of the model prediction can be evaluated according to the classification result of the target sample block and the label corresponding to the target sample block, so as to update the parameters of the model.

[0108] Specifically, for machine learning models, the accuracy of model prediction results can be measured by a loss function (LossFunction), which is defined on a single training data and is used to measure the prediction error of a training data. Specifically, the loss value of the training data is determined by the label of a single training data and the prediction result of the model for the training data. During actual training, a training data set has a lot of training data, so a cost function (CostFunction) is generally used to measure the overall error of the training data set. The cost function is defined on the entire training data set and is used to calculate the average value of the prediction error of all training data, which can better measure the prediction effect of the model. For general machine learning models, based on the aforementioned cost function, plus a regularization term that measures the complexity of the model, it can be used as the objective function of the training, and the loss value of the entire training data set can be calculated based on the objective function. There are many types of commonly used loss functions, such as 0-1 loss function, square loss function, absolute loss function, logarithmic loss function, cross entropy loss function, etc., which can all be used as loss functions of machine learning models, which will not be described one by one here. In the embodiment of the present application, one loss function can be selected to determine the loss value of the training, such as the cross entropy loss function. Based on the training loss value, the back propagation algorithm is used to update the model parameters, and a trained point cloud classification model can be obtained after several iterations. The specific number of iterations can be set in advance, or the training is considered complete when the test set meets the accuracy requirements.

[0109] Step 140: Sample and restore the point cloud reconstruction data according to the first mapping information to obtain a target point cloud reconstruction result.

[0110] In an embodiment of the present application, since the point cloud reconstruction data obtained in step 130 is reconstructed data corresponding to the three-dimensional point cloud data after sampling preprocessing, the point cloud reconstruction data can be upsampled based on the point cloud mapping relationship recorded in the first mapping information, thereby obtaining a target point cloud reconstruction result corresponding to the three-dimensional point cloud data to be reconstructed.

[0111] A pipeline facility reconstruction and identification system based on three-dimensional laser point cloud proposed according to an embodiment of the present application is described in detail below with reference to the accompanying drawings.

[0112] Reference Figure 2 , a pipeline facility reconstruction and identification system based on three-dimensional laser point cloud proposed in an embodiment of the present application includes:

[0113] The first processing unit 101 is used to obtain the three-dimensional point cloud data to be reconstructed, and perform sampling preprocessing on the three-dimensional point cloud data to obtain sampled point cloud data and first mapping information, wherein the first mapping information is used to characterize the point cloud mapping relationship between the three-dimensional point cloud data before sampling and the three-dimensional point cloud data after sampling;

[0114] The second processing unit 102 is used to perform combined block processing on the sampled point cloud data to obtain a plurality of point cloud sample blocks and second mapping information, wherein the number of point clouds in each of the point cloud sample blocks is the same, and the sample points in each of the point cloud sample blocks are evenly distributed in the point cloud space of the sampled point cloud data, and the second mapping information is used to characterize the point cloud mapping relationship between the point cloud sample blocks and the sampled point cloud data;

[0115] A third processing unit 103 is configured to perform point cloud reconstruction on the point cloud sample block based on the second mapping information by using a trained point cloud classification model to obtain point cloud reconstruction data;

[0116] The fourth processing unit 104 is used to perform sampling and restoration on the point cloud reconstruction data according to the first mapping information to obtain a target point cloud reconstruction result.

[0117] It can be understood that the contents of the above method embodiments are all applicable to the present system embodiments, the functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0118] Reference Figure 3 , the embodiment of the present application further provides an electronic device, including:

[0119] at least one processor 201;

[0120] At least one memory 202, used to store at least one program;

[0121] When the at least one program is executed by the at least one processor 201, the at least one processor 201 implements the above method embodiment.

[0122] Similarly, it can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0123] The embodiment of the present application further provides a computer-readable storage medium, in which a program executable by the processor 201 is stored. The program executable by the processor 201 is used to implement the above-mentioned method embodiment when executed by the processor 201.

[0124] Similarly, the contents of the above method embodiments are all applicable to the computer-readable storage medium embodiments. The functions specifically implemented by the computer-readable storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0125] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the application is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part described as a larger operation is performed independently.

[0126] In addition, although the present application is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features can be integrated into a single physical device and / or software module, or one or more functions and / or features can be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present application. More specifically, in view of the properties, functions, and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional techniques of the engineer. Therefore, those skilled in the art can implement the present application set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present application, which is determined by the full scope of the attached claims and their equivalents.

[0127] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method of the embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0128] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0129] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0130] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0131] In the above description of this specification, the description with reference to the terms "one embodiment / example", "another embodiment / example" or "certain embodiments / examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0132] Although the embodiments of the present application have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present application, and that the scope of the present application is defined by the claims and their equivalents.

[0133] The above is a specific description of the preferred implementation of the present application, but the present application is not limited to the embodiments. Technical personnel familiar with the field can make various equivalent modifications or substitutions without violating the spirit of the present application. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present application.

Claims

1. A pipeline facility reconstruction and identification method based on three-dimensional laser point cloud, characterized in that: include: Acquire three-dimensional point cloud data to be reconstructed, and perform sampling preprocessing on the three-dimensional point cloud data to obtain sampled point cloud data and first mapping information, wherein the first mapping information is used to characterize a point cloud mapping relationship between the three-dimensional point cloud data before sampling and the three-dimensional point cloud data after sampling; The sampled point cloud data are combined and divided into blocks to obtain a plurality of point cloud sample blocks and second mapping information, wherein the number of point clouds in each of the point cloud sample blocks is the same, and the sample points in each of the point cloud sample blocks are evenly distributed in the point cloud space of the sampled point cloud data, and the second mapping information is used to characterize the point cloud mapping relationship between the point cloud sample blocks and the sampled point cloud data; Based on the second mapping information, reconstructing the point cloud sample block through the trained point cloud classification model to obtain point cloud reconstruction data; The point cloud reconstruction data is sampled and restored according to the first mapping information to obtain a target point cloud reconstruction result.

2. The method according to claim 1, characterized in that The sampling and preprocessing of the three-dimensional point cloud data to obtain the sampled point cloud data and the first mapping information includes: Performing data cleaning on the three-dimensional point cloud data to obtain first point cloud data; Performing three-dimensional grid division on the first point cloud data to obtain second point cloud data; Grid sampling is performed on the second point cloud data to obtain the sampled point cloud data and the first mapping information.

3. The method according to claim 1, characterized in that The combining and block processing of the sampled point cloud data to obtain a plurality of point cloud sample blocks and second mapping information includes: Performing point cloud farthest point sampling on the sampled point cloud data to obtain third point cloud data; According to the sampled point cloud data, point cloud nearest neighbor classification is performed on the third point cloud data to obtain a plurality of the point cloud sample blocks and the second mapping information.

4. The method according to claim 3, characterized in that The step of sampling the farthest point of the point cloud on the sampled point cloud data to obtain third point cloud data includes: Determining a target sampling quantity according to the sampling point cloud data; According to the first point cloud sampling sequence, the remaining sampling point cloud data after sampling is screened for the maximum distance to obtain the distance optimum point; According to the distance optimal point, updating the sampling points of the first point cloud sampling sequence to obtain a second point cloud sampling sequence; Comparing the target sampling quantity with the number of sampling points in the second point cloud sampling sequence to obtain a sampling comparison result; If the sampling comparison result is that the number of sampling points in the second point cloud sampling sequence is less than the target sampling number, the first point cloud sampling sequence is updated according to the second point cloud sampling sequence, and then the step of performing maximum distance screening on the remaining sampling point cloud data after sampling according to the first point cloud sampling sequence to obtain the distance optimum point is returned; or, if the sampling comparison result is that the number of sampling points in the second point cloud sampling sequence is equal to the target sampling number, the second point cloud sampling sequence is determined as the third point cloud data.

5. The method according to claim 3, characterized in that: The step of performing point cloud nearest neighbor classification on the third point cloud data according to the sampled point cloud data to obtain a plurality of the point cloud sample blocks and the second mapping information includes: Acquire a tree structure corresponding to the sampling point cloud data and a preset number of sample points; According to the number of sample points and the third point cloud data, a neighbor point search is performed on the tree structure to obtain a plurality of the point cloud sample blocks and the second mapping information.

6. The method according to claim 1, characterized in that The step of reconstructing the point cloud sample block based on the second mapping information by using a trained point cloud classification model to obtain point cloud reconstruction data includes: Inputting the point cloud sample block into the trained point cloud classification model to perform point cloud classification prediction to obtain a point cloud classification result; According to the second mapping information, index averaging is performed on the point cloud classification result to obtain a reconstruction prediction result; The category probability value corresponding to each sample point of the reconstruction prediction result is screened for the maximum value to obtain the point cloud reconstruction data.

7. The method according to claim 1, characterized in that The trained point cloud classification model is obtained by training in the following steps: Get training sample blocks; Performing data enhancement and normalization on the training sample block to obtain a target sample block; The target sample block is input into the initialized point cloud classification model for training to obtain the trained point cloud classification model.

8. A pipeline facility reconstruction and identification system based on three-dimensional laser point cloud, characterized in that: include: A first processing unit is used to obtain the three-dimensional point cloud data to be reconstructed, and perform sampling preprocessing on the three-dimensional point cloud data to obtain sampled point cloud data and first mapping information, wherein the first mapping information is used to characterize a point cloud mapping relationship between the three-dimensional point cloud data before sampling and the three-dimensional point cloud data after sampling; A second processing unit is used to perform combined block processing on the sampled point cloud data to obtain a plurality of point cloud sample blocks and second mapping information, wherein the number of point clouds in each of the point cloud sample blocks is the same, and the sample points in each of the point cloud sample blocks are evenly distributed in the point cloud space of the sampled point cloud data, and the second mapping information is used to characterize the point cloud mapping relationship between the point cloud sample blocks and the sampled point cloud data; A third processing unit is configured to perform point cloud reconstruction on the point cloud sample block based on the second mapping information by using a trained point cloud classification model to obtain point cloud reconstruction data; The fourth processing unit is used to sample and restore the point cloud reconstruction data according to the first mapping information to obtain a target point cloud reconstruction result.

9. An electronic device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a program executable by a processor, characterized in that: The program executable by the processor is used to implement the method according to any one of claims 1 to 7 when executed by the processor.

Images