A method and device for manufacturing a bridge real scene three-dimensional point cloud dataset and a medium

Abstract

The application belongs to the field of point cloud data, and discloses a bridge real scene three-dimensional point cloud dataset production method, device and medium, the method comprises the following steps: obtaining dataset production demand; when the dataset production demand is the point cloud dataset production demand of the bridge component, obtaining a plurality of bridge components in the bridge model; extracting the component parameters of each bridge component, and constructing the point cloud dataset of the bridge component based on the component parameters of each bridge component and the parameters of the bridge model; and / or when the dataset production demand is the point cloud dataset production demand of the bridge model, obtaining the entity object corresponding to each bridge component in the bridge model; performing triangulation and laser simulation scanning on the surface of the entity object to obtain the point cloud space information of the entity object, and constructing the point cloud dataset of the bridge model according to the point cloud space information of the entity object. The dataset production method can efficiently generate the dataset meeting the deep learning demand.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of point cloud data, and particularly relates to a bridge real scene three-dimensional point cloud dataset production method and device and medium. BACKGROUND

[0002] Bridge main component identification and three-dimensional scene construction are important contents of bridge construction process and later dynamic detection, and a deep learning method needs to perform sequence real scene three-dimensional observation in processing such problems. However, bridge structures are different, the construction period is long, and the sample quantity is seriously insufficient.

[0003] Existing small amount of simulation data acquisition methods mainly rely on surface sampling methods, cannot simulate the real laser scanning process, cannot fully estimate the influence of scanning intervals, occlusions and other factors, and the obtained results are greatly different from real data. SUMMARY

[0004] The application aims to provide a bridge real scene three-dimensional point cloud dataset production method and device and medium, to solve the problem that existing small amount of simulation data acquisition methods mainly rely on surface sampling methods, cannot simulate the real laser scanning process, cannot fully estimate the influence of scanning intervals, occlusions and other factors, and the obtained results are greatly different from real data.

[0005] In order to achieve the above-mentioned purpose, the application adopts the following technical scheme:

[0006] In the first aspect, the application provides a bridge real scene three-dimensional point cloud dataset production method, which comprises:

[0007] Obtaining dataset production requirements, wherein the dataset production requirements comprise bridge model point cloud dataset production requirements and / or bridge component point cloud dataset production requirements;

[0008] When the dataset production requirements are the bridge component point cloud dataset production requirements, obtaining a plurality of bridge components in the bridge model; extracting component parameters of each bridge component, constructing a bridge component point cloud dataset based on the component parameters of each bridge component and the parameters of the bridge model; and / or

[0009] When the dataset production requirements are the bridge model point cloud dataset production requirements, obtaining entity objects corresponding to each bridge component in the bridge model; performing triangulation and laser simulation scanning on the surface of the entity objects to obtain point cloud space information of the entity objects, and constructing a bridge model point cloud dataset according to the point cloud space information of the entity objects.

[0010] Preferably, constructing a bridge component point cloud dataset based on the component parameters of each bridge component and the parameters of the bridge model comprises:

[0011] The component parameters of each bridge component are automatically corrected based on the proportion of each bridge component, to obtain the corrected component parameters of each bridge component;

[0012] The bridge model corresponding to the corrected component parameters of each bridge component is converted into a first mesh model, and the first mesh model includes a plurality of triangular faces;

[0013] Based on the uniform sampling method, the sampling points of each triangular face are determined;

[0014] Based on the sampling points of each triangular face and the corrected component parameters of each bridge component, a point cloud dataset of each bridge component is constructed.

[0015] Preferably, the uniform sampling method is:

[0016] The area of each triangular face of the first mesh model and the total area of all triangular faces of the first mesh model are calculated;

[0017] According to the proportion of the area of each triangular face of the first mesh model in the total area of all triangular faces of the first mesh model, the number of sampling points to be generated on each triangular face of the first mesh model is determined;

[0018] A barycentric coordinate system of each triangular face of the first mesh model is constructed;

[0019] In the barycentric coordinate system of each triangular face of the first mesh model, according to the vertices of the triangular face and the number of sampling points to be generated on the triangular face, uniformly distributed sampling points are generated.

[0020] Preferably, before constructing the point cloud dataset of each bridge component based on the sampling points of each triangular face and the corrected component parameters of each bridge component, the method further comprises adding random noise to the sampling points of each triangular face, and the random noise includes Gaussian noise, random block noise and / or random block removal.

[0021] Preferably, when the random noise is random block noise and / or random block removal, the random block is a cubic block and / or a spherical block, wherein when the random block is a spherical block, spherical block noise and / or spherical block removal are added to the sampling points of each triangular face based on a spherical space constraint; when the random block is a cubic block, cubic block noise and / or cubic block removal are added to the sampling points of each triangular face based on a cubic space constraint.

[0022] Preferably, the spherical space constraint is that the sampling points to be added with noise or removed are uniformly distributed within a spherical radius range;

[0023] The cubic space constraint is that the sampling points to be added with noise or removed are located within the range of a cubic block and are distributed according to a preset distribution rule.

[0024] Preferably, the surface of the solid object is triangulated and laser simulation scanning is performed to obtain the point cloud spatial information of the solid object, including:

[0025] The surface of the solid object is read, the surface of the solid object is triangulated to obtain a plurality of triangular patches, and the labels of the triangular patches are constructed based on the corresponding bridge component of the solid object;

[0026] The bridge model is converted into a second mesh model, the second mesh model is loaded, and the second mesh model is laser simulation scanned by using the simulated unmanned aerial vehicle;

[0027] The laser emission direction of the simulated unmanned aerial vehicle, the position of the unmanned aerial vehicle, the time interval of emission and reception of the laser signal, and the scanning time stamp in the laser simulation scanning process are obtained;

[0028] The point cloud coordinates of the triangular patches are determined according to the laser emission direction of the simulated unmanned aerial vehicle, the position of the unmanned aerial vehicle, and the time interval of emission and reception of the laser signal in the laser simulation scanning process, and the point cloud coordinates of the triangular patches, the labels of the triangular patches, and the point cloud coordinates of the triangular patches are used as the point cloud spatial information of the solid object.

[0029] Preferably, the laser simulation scanning of the second mesh model by using the simulated unmanned aerial vehicle includes:

[0030] The model center line is constructed based on the deck of the bridge model, and the scanning route is generated from the model center line;

[0031] The simulated unmanned aerial vehicle flies and scans according to the scanning route, and emits scanning laser in a preset laser emission range and a preset laser emission direction to perform laser simulation scanning on the second mesh model.

[0032] In a second aspect, the present application provides a bridge real scene three-dimensional point cloud dataset production device for realizing the bridge real scene three-dimensional point cloud dataset production method, and the device comprises:

[0033] An acquisition module is configured to acquire dataset production requirements, wherein the dataset production requirements include point cloud dataset production requirements of a bridge model and / or point cloud dataset production requirements of a bridge component;

[0034] A production module is configured to, when the dataset production requirements are the point cloud dataset production requirements of the bridge component, acquire a plurality of bridge components in the bridge model, extract component parameters of each bridge component, and construct the point cloud dataset of the bridge component based on the component parameters of each bridge component and the parameters of the bridge model; and / or

[0035] When the dataset making demand is the point cloud dataset making demand of the bridge model, the entity object corresponding to each bridge component in the bridge model is acquired; the surface of the entity object is triangulated and laser simulated scanning is performed to obtain the point cloud space information of the entity object, and the point cloud dataset of the bridge model is constructed according to the point cloud space information of the entity object.

[0036] In a third aspect, the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned bridge real scene three-dimensional point cloud dataset making method when executing the computer program.

[0037] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program, wherein the program is executable on a processor to implement the above-mentioned bridge real scene three-dimensional point cloud dataset making method.

[0038] Beneficial effects:

[0039] 1. According to the demand, two different point cloud datasets can be made, that is, the point cloud dataset of each bridge component is constructed based on the component parameters of each bridge component and the parameters of the bridge model by extracting the component parameters of each bridge component; meanwhile, the entity object corresponding to each bridge component in the bridge model is acquired; the surface of the entity object is triangulated and laser simulated scanning is performed to obtain the point cloud space information of the entity object, and the point cloud dataset of the bridge model is constructed according to the point cloud space information of the entity object; therefore, in the training process, different point cloud datasets can be selected according to different training tasks, for example, the point cloud dataset of the bridge component is selected for the classification task, and the point cloud dataset of the bridge model is selected for the segmentation task;

[0040] 2. The dataset making method of the present application can efficiently generate a dataset meeting the deep learning demand. BRIEF DESCRIPTION OF DRAWINGS

[0041] The accompanying drawings are included to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used together with the following detailed description to explain the embodiments of the present application, but do not constitute a limitation on the embodiments of the present application. In the drawings:

[0042] Figure 1 is a flow chart of the bridge real scene three-dimensional point cloud dataset making method provided by an embodiment of the present application;

[0043] Figure 2 is a block diagram of the bridge real scene three-dimensional point cloud dataset making device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the present application will be briefly introduced below in connection with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the drawings only constitutes some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings. It should be noted that the description of these embodiments is used to help understand the present application, but does not constitute a limitation on the present application.

[0045] Embodiment one

[0046] Figure 1 A flowchart of a method for manufacturing a bridge real scene three-dimensional point cloud data set provided by an embodiment of the present application is shown in FIG. 1, and the embodiment provides a method for manufacturing a bridge real scene three-dimensional point cloud data set, which comprises the following steps. Figure 1

[0047] Step S10: Obtain data set manufacturing requirements, which include point cloud data set manufacturing requirements of a bridge model and / or point cloud data set manufacturing requirements of a bridge component; in this embodiment, the point cloud data set of the bridge model is mainly used for a segmentation task, and the purpose of the segmentation task is to accurately separate and label the boundaries and areas of each part of the bridge, so the segmentation task requires a complete bridge model and its accurate geometric information to ensure that the model can identify and segment each component; the point cloud data set of the bridge component is mainly used for a classification task, and the classification task usually only needs the data set to contain images or models of a single bridge component, and this kind of task mainly focuses on feature recognition or parameter prediction of a specific component, so the data set manufacturing can focus on a local detail of the bridge;

[0048] Step S20: When the data set manufacturing requirements are the point cloud data set manufacturing requirements of the bridge component, obtain a plurality of bridge components in the bridge model; extract component parameters of each bridge component, and construct the point cloud data set of the bridge component based on the component parameters of each bridge component and the parameters of the bridge model; and / or

[0049] When the data set manufacturing requirements are the point cloud data set manufacturing requirements of the bridge model, obtain entity objects corresponding to each bridge component in the bridge model; perform triangulation and laser simulation scanning on the surfaces of the entity objects to obtain point cloud spatial information of the entity objects, and construct the point cloud data set of the bridge model according to the point cloud spatial information of the entity objects.

[0050] ​The application can manufacture two different point cloud data sets according to requirements, that is, constructing a point cloud data set of a bridge component based on the component parameters of each bridge component and the parameters of the bridge model by extracting the component parameters of each bridge component; meanwhile, the entity object corresponding to each bridge component in the bridge model is obtained; the surface of the entity object is triangulated and laser simulated scanning is performed to obtain the point cloud spatial information of the entity object, and a point cloud data set of the bridge model is constructed according to the point cloud spatial information of the entity object; therefore, in the training process, different point cloud data sets can be selected according to different training tasks, for example, the point cloud data set of the bridge component is selected for the classification task, and the point cloud data set of the bridge model is selected for the segmentation task; and the data set manufacturing method of the application can efficiently generate a data set meeting the requirements of deep learning.

[0051] As a further optimization of the embodiment, constructing a point cloud data set of a bridge component based on the component parameters of each bridge component and the parameters of the bridge model comprises:

[0052] Step A10: automatically correcting the component parameters based on the proportions of each bridge component to obtain the corrected component parameters of each bridge component; by automatically correcting the component parameters, a model library of different types and different parameters of a certain structure of a bridge can be batch-manufactured to meet the requirement that the sample data for deep learning is diverse enough; taking a pier as an example: a pier of a certain category is manufactured as a reference pier, the transformation range of the model parameters of the reference pier is limited to 70% to 130%, and then the pier cap height, top height, long side radius, wide side radius, pier ground width, pier ground length, pier top length, pier top width, pier body connection width, pier body connection length, and pier body height are adjusted.

[0053] In the embodiment, the bridge model adopts a BIM model (Building Information Modeling, a conventional parameterized model of a bridge), the BIM model provides detailed geometric information of the bridge and can flexibly adjust the parameterized data of the bridge structure to meet the requirements of different tasks. Through the BIM model, a data set meeting the requirements of deep learning can be efficiently generated.

[0054] In the embodiment, the specific steps of automatically correcting the component parameters are as follows: setting an interval for parameter adjustment according to the proportions of each bridge component, for example, the length of a component needs to be adjusted, and a minimum value and a maximum value can be set as the adjustment range; to ensure that the corrected component parameters of the bridge component meet the real requirements;

[0055] Then introduce time operator as the default seed value, different random numbers for each component parameter, the random number in the parameter adjustment interval; That is, using the current system time as the seed value of random number generation, to ensure that the random number sequence generated each time the program is run is different, for each parameter, using the time operator as the seed to generate random numbers within the set interval; In this embodiment, the generated random number will be a number between 0.7 and 1.3, and the component parameters are multiplied by the random number to adjust the component parameters.

[0056] Step A20: converting the bridge model corresponding to the modified component parameters of each bridge component into a first mesh model, the first mesh model comprising a plurality of triangular faces;

[0057] Wherein, the mesh model is a grid model, which is a model for approximating a three-dimensional object using a series of polygons with similar size and shape, and the first mesh model uses triangles, so the first mesh model has a plurality of triangular faces.

[0058] After converting the bridge model into the first mesh model, the first mesh model and the corresponding modified component parameters of each bridge component are automatically recorded in the txt file, and the parameters in the txt file correspond one-to-one with the saved model, and the saved model is restored to the sample model to ensure that the parameters after repeated adjustment do not distort. The sample model exists as a single category benchmark. Since the parameters are automatically adjusted by random numbers, the model may be completely distorted after hundreds or thousands of random adjustments, so a sample recovery link is needed on the basis of each automatic parameter adjustment to ensure that the model after parameter adjustment is similar to the real structure. The sample model also has multiple categories, for example, the basic structure of a bridge has piers, beams, and pile caps, and the piers have straight piers and double-curvature piers, and there are distinctions between whether they have drainage holes, so a benchmark sample is needed for each category.

[0059] Step A30: determining the sampling points of each triangular face based on the uniform sampling method;

[0060] In this embodiment, the uniform sampling method is:

[0061] Step a301: calculating the area of each triangular face of the first mesh model and the total area of all triangular faces of the first mesh model; the calculation formula of the area of each triangular face is:

[0062] (1);

[0063] In formula (1), is the area of the triangular face, A, B, and C are the three vertices of the triangle, denotes the vector the modulus of the product of the vector .

[0064] Step a302: determining the number of sampling points to be generated on each triangular face of the first mesh model according to the proportion of the area of each triangular face of the first mesh model in the total area of all triangular faces of the first mesh model; in this embodiment, the larger the area of a triangular face, the more sampling points on the triangular face.

[0065] Step a303: constructing the barycentric coordinate system of each triangular face of the first mesh model.

[0066] Step a304: generating uniformly distributed sampling points in the barycentric coordinate system of each triangular face of the first mesh model according to the vertices of the triangular face and the number of sampling points to be generated on the triangular face.

[0067] In this embodiment, for any triangular face, let the three vertices of the triangular face be , , , and the sampling point be P, and the calculation formula of the sampling point be:

[0068] ( (1- )) +( ) (2).

[0069] In formula (2), , are random numbers in the interval [0, 1], if 1, they are mapped to the inside of the triangular face, and a set of , can generate a sampling point.

[0070] As a further optimization of this embodiment, after generating the sampling points on each triangular face, the method further comprises: adding random noise to the sampling points of each triangular face, the random noise comprising: Gaussian noise, random block noise and / or random block excavation; this embodiment more realistically simulates problems in the actual acquisition process, such as tree obstruction and sampling point loss, by introducing random block noise and random block excavation.

[0071] wherein the Gaussian noise is a class of noise obeying Gaussian distribution, an offset value is added to each sampling point, and the calculation formula of the offset value is:

[0072] (3).

[0073] In formula (3), , is a parameter of distribution, respectively, the expectation and variance of Gaussian distribution, e is the natural base, x is the independent variable.

[0074] wherein, when the random block is a spherical block, the sampling points of each triangle are added with spherical block noise and / or spherical block deletion based on spherical space constraint; when the random block is a cubic block, the sampling points of each triangle are added with cubic block noise and / or cubic block deletion based on cubic space constraint.

[0075] In the embodiment, the spherical space constraint is that the sampling points added with noise or deleted are uniformly distributed within the range of spherical radius; and the cubic space constraint is that the sampling points added with noise or deleted are distributed according to a preset distribution rule within the range of a cubic block.

[0076] In the embodiment, when the random block noise and the random block deletion are introduced, the size and shape of the block region of the random noise and deletion are adjusted according to the scale of the model, so as to achieve different noise adding and deletion effects. Specifically, the random block noise and deletion mainly adopts two geometric shapes: cubic block and spherical block. In the case of adding noise, a certain number of points are added in the randomly generated block region, and in the case of deletion, the points in the region are deleted. In this way, various occlusions and deletions that may occur in the real world can be effectively simulated. The center points of these random blocks are randomly selected points, but in order to ensure that they are located in a reasonable position, they must be constrained within a certain distance range within the boundary of the model point cloud, that is, the spherical space constraint and the cubic space constraint.

[0077] The expression of the cubic space constraint is:

[0078] (4);

[0079] In formula (4), is the center point of the randomly selected noise block, assuming that the noise block is a cube, and the side length is defined as For any point in the block satisfies formula (4).

[0080] The expression of the spherical space constraint is:

[0081] (5);

[0082] In formula (5), is the center point of the randomly selected noise block, assuming that the noise block is a sphere, and the radius is defined as For any point in the block, formula (5) is satisfied.

[0083] For example, for a cubic block, all points added with noise or excavated must be located within the boundary of the block and satisfy certain distribution rules; for a spherical block, all points must be uniformly distributed within the radius range. This noise adding and data excavating method based on cubic space constraints and spherical space constraints not only increases the robustness of the model, but also better simulates the uncertainty existing in the real world.

[0084] Step A40: Based on the sampling points of each triangular face and the component parameters of the modified bridge components, a point cloud data set of each bridge component is constructed.

[0085] As a further optimization of the embodiment, the surfaces of the solid object are triangulated and laser simulated scanning is performed to obtain the point cloud spatial information of the solid object, including:

[0086] Step B10: Read the surfaces of the solid object, triangulate the surfaces of the solid object to obtain a plurality of triangular patches, and construct labels of each triangular patch based on the corresponding bridge components of the solid object;

[0087] In the embodiment, the solid object in the BIM model is obtained according to different bridge components, and the surfaces of the solid object are read to obtain the geometric information of the solid object, which is saved; then the surfaces of the solid object are triangulated, first the surfaces are initially segmented, each part of the triangular is evaluated, whether the flatness meets the condition is checked, the part that does not meet the condition is further subdivided and the above steps are repeated until all the triangulars meet the condition, then the finally segmented parts are all converted into triangular patches; and the triangular patches are named by bridge component type or self-made number to construct labels of the triangular patches.

[0088] Step B20: Convert the bridge model into a second mesh model, load the second mesh model, and use a simulated unmanned aerial vehicle to perform laser simulation scanning on the second mesh model;

[0089] In the embodiment, the simulated unmanned aerial vehicle is used to perform laser simulation scanning on the second mesh model, including:

[0090] Based on the bridge deck of the bridge model, a model center line is constructed, and a scanning route is generated based on the model center line; that is, taking the bridge deck as an object, a line similar to the model center line is drawn on the connecting line of the two end points of the object and saved separately, and then a flight route of the simulated unmanned aerial vehicle is designed based on the center line, and the flight route is the scanning route; when designing the flight route of the simulated unmanned aerial vehicle, the height from the highest point of the bridge to the bridge deck is added to design two routes parallel to the bridge and slightly higher than the bridge.

[0091] The simulation unmanned aerial vehicle flies along a scanning route and emits scanning laser light in a preset laser emission range and a preset laser emission direction to perform laser simulation scanning on the second mesh model.

[0092] After the mesh model is loaded and simulation is started, a ToF (Time of Flight) is used to calculate the distance of a scanning point, which is a point on a triangular patch. The unmanned aerial vehicle will continuously emit simulation laser signals with a range and a fixed direction from a simulation starting position, and the intersection of each geometric entity object in the scene with the laser is calculated, and the reflected light is calculated recursively to track new light. When the reflected light of the geometric entity object intersects with the unmanned aerial vehicle at the present position, it is considered that the laser ranging process is completed.

[0093] Step B30: Obtain the laser emission direction of the simulation unmanned aerial vehicle, the unmanned aerial vehicle position, the time interval of emission and reception of the laser signal, and the scanning timestamp during the laser simulation scanning process.

[0094] Step B40: Determine the point cloud coordinates of each triangular patch according to the laser emission direction of the simulation unmanned aerial vehicle, the unmanned aerial vehicle position, and the time interval of emission and reception of the laser signal during the laser simulation scanning process, and take the point cloud coordinates of each triangular patch, the label of each triangular patch, and the point cloud coordinates of each triangular patch as the point cloud spatial information of the entity object.

[0095] In this embodiment, the point cloud coordinates of each triangular patch can be calculated by the laser emission direction of the simulation unmanned aerial vehicle, the unmanned aerial vehicle position, and the time interval of emission and reception of the laser signal; wherein the point cloud coordinates calculation expression of the triangular patch is:

[0096] (6);

[0097] In formula (6), is the distance from the scanning point to the simulation unmanned aerial vehicle, is the pitch angle (vertical angle) of the laser, is the yaw angle (horizontal angle) of the laser, is the three-dimensional coordinates of the scanning point in the simulation unmanned aerial vehicle coordinate system, .

[0098] The calculation expression of the distance from the scanning point to the simulation unmanned aerial vehicle is:

[0099] (7);

[0100] (8);

[0101] In formula (7) and formula (8), c is the speed of light, a time interval for transmitting and receiving laser signals, , respectively a transmission time and a receiving time.

[0102] In the generated point cloud data, the point cloud data of a single structure can be used for the classification of bridge components, and the model parameters recorded during the output process can be used for the parameter fitting of the bridge components; the point cloud data of the whole bridge with the category label can be used for the semantic segmentation of the whole bridge.

[0103] Embodiment two

[0104] Figure 2 is a block diagram of a bridge real scene three-dimensional point cloud data set production device provided by an embodiment of the present application, as shown in Figure 2 The embodiment provides a bridge real scene three-dimensional point cloud data set production device for realizing the bridge real scene three-dimensional point cloud data set production method of embodiment one, and the device comprises:

[0105] An acquisition module is configured to acquire a data set production requirement, wherein the data set production requirement comprises a point cloud data set production requirement of a bridge model and / or a point cloud data set production requirement of a bridge component.

[0106] A production module is configured to, when the data set production requirement is the point cloud data set production requirement of the bridge component, acquire a plurality of bridge components in the bridge model, extract component parameters of each bridge component, and construct a point cloud data set of the bridge component based on the component parameters of each bridge component and parameters of the bridge model; and / or

[0107] When the data set production requirement is the point cloud data set production requirement of the bridge model, the production module is configured to acquire entity objects corresponding to each bridge component in the bridge model, perform triangulation and laser simulation scanning on surfaces of the entity objects to obtain point cloud space information of the entity objects, and construct a point cloud data set of the bridge model based on the point cloud space information of the entity objects.

[0108] The embodiment further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor realizes the bridge real scene three-dimensional point cloud data set production method of embodiment one when executing the computer program.

[0109] The embodiment further provides a computer readable storage medium having a computer program stored thereon, and the program is executable on a processor to realize the bridge real scene three-dimensional point cloud data set production method of embodiment one.

[0110] The application can manufacture two different point cloud data sets according to requirements, that is, point cloud data sets of bridge components are constructed based on component parameters of the bridge components and parameters of the bridge model by extracting the component parameters of the bridge components; meanwhile, entity objects corresponding to the bridge components in the bridge model are obtained; the surfaces of the entity objects are triangulated and laser simulated scanning is performed to obtain point cloud space information of the entity objects, and the point cloud data sets of the bridge model are constructed according to the point cloud space information of the entity objects; therefore, in the training process, different point cloud data sets can be selected according to different training tasks, for example, the point cloud data sets of the bridge components are selected for classification tasks, and the point cloud data sets of the bridge model are selected for segmentation tasks; and the data set manufacturing method of the application can efficiently generate data sets meeting the requirements of deep learning.

[0111] Those skilled in the art will understand that embodiments of the application can be provided as methods, systems, or computer program products. Therefore, the application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0112] The application is described with reference to flowcharts and / or block diagrams according to the methods, devices (systems), and computer program products of embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks Figure 1 The functions specified in a flow or multiple flows and / or blocks

[0113] The above is only an embodiment of the application and is not intended to limit the application. The application can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the application should be included in the scope of the claims of the application.

Claims

1. A method of creating a bridge real-world three-dimensional point cloud dataset, characterized by, The method comprises: obtaining dataset production requirements, the dataset production requirements comprising: point cloud dataset production requirements of a bridge model and point cloud dataset production requirements of a bridge component; when the dataset production requirements are the point cloud dataset production requirements of the bridge component, obtaining a plurality of bridge components in the bridge model; extracting component parameters of each bridge component, and constructing a point cloud dataset of the bridge component based on the component parameters of each bridge component and parameters of the bridge model; when the dataset production requirements are the point cloud dataset production requirements of the bridge model, obtaining entity objects corresponding to each bridge component in the bridge model; performing triangulation and laser simulation scanning on a surface of the entity objects to obtain point cloud spatial information of the entity objects, and constructing a point cloud dataset of the bridge model according to the point cloud spatial information of the entity objects; constructing the point cloud dataset of the bridge component based on the component parameters of each bridge component and the parameters of the bridge model comprises: automatically correcting the component parameters based on the proportions of each bridge component to obtain corrected component parameters of each bridge component; converting the bridge model corresponding to the corrected component parameters of each bridge component into a first mesh model, the first mesh model comprising a plurality of triangular faces; determining sampling points of each triangular face based on a uniform sampling method; constructing the point cloud dataset of each bridge component based on the sampling points of each triangular face and the corrected component parameters of each bridge component; performing triangulation and laser simulation scanning on the surface of the entity objects to obtain the point cloud spatial information of the entity objects comprises: reading the surface of the entity objects, performing triangulation on the surface of the entity objects to obtain a plurality of triangular surface patches, and constructing labels of each triangular surface patch based on the bridge component corresponding to the entity objects; converting the bridge model into a second mesh model, loading the second mesh model, and performing laser simulation scanning on the second mesh model by using a simulated unmanned aerial vehicle; obtaining a laser emission direction of the simulated unmanned aerial vehicle, a position of the unmanned aerial vehicle, a time interval of emission and reception of a laser signal, and a scanning time stamp in the process of laser simulation scanning; determining point cloud coordinates of each triangular surface patch according to the laser emission direction of the simulated unmanned aerial vehicle, the position of the unmanned aerial vehicle, and the time interval of emission and reception of the laser signal in the process of laser simulation scanning, and taking the point cloud coordinates of each triangular surface patch, the labels of each triangular surface patch, and the point cloud coordinates of each triangular surface patch as the point cloud spatial information of the entity objects.

2. The method of claim 1, wherein, The uniform sampling method comprises: calculating an area of each triangular face of the first mesh model and a total area of all triangular faces of the first mesh model; determining a number of sampling points to be generated on each triangular face of the first mesh model according to a proportion of the area of each triangular face of the first mesh model in the total area of all triangular faces of the first mesh model; constructing a barycentric coordinate system of each triangular face of the first mesh model; in the barycentric coordinate system of each triangular face of the first mesh model, generating uniformly distributed sampling points according to the vertices of the triangular face and the number of sampling points to be generated on the triangular face.

3. The method of claim 1, wherein, Before constructing the point cloud dataset of each bridge component based on the sampling points of each triangular face and the component parameters of the corrected bridge components, the method further comprises: adding random noise to the sampling points of each triangular face, the random noise comprising: Gaussian noise, random block noise and / or random block excavation.

4. The method of claim 3, wherein When the random noise is random block noise and / or random block excavation, the random block is a cubic block and / or a spherical block, wherein, when the random block is a spherical block, spherical block noise and / or spherical block excavation is added to the sampling points of each triangular face based on spherical space constraints; when the random block is a cubic block, cubic block noise and / or cubic block excavation is added to the sampling points of each triangular face based on cubic space constraints.

5. The method of claim 4, wherein, The spherical space constraints are that the sampling points added with noise or excavated are uniformly distributed within a spherical radius range; The cubic space constraints are that the sampling points added with noise or excavated are located within a cubic range and are distributed according to a preset distribution rule.

6. The method of claim 1, wherein, The method further comprises: performing laser simulation scanning on the second mesh model by using a simulation unmanned aerial vehicle, comprising: constructing a model center line based on the deck of the bridge model, and generating a scanning route based on the model center line; the simulation unmanned aerial vehicle flies and scans according to the scanning route, and emits scanning laser in a preset laser emission range and a preset laser emission direction to perform laser simulation scanning on the second mesh model.

7. An apparatus for producing a bridge real-world three-dimensional point cloud dataset, for implementing the method for producing a bridge real-world three-dimensional point cloud dataset according to any one of claims 1 to 6, characterized in that, The device comprises: an acquisition module configured to acquire dataset production requirements, the dataset production requirements comprising: point cloud dataset production requirements of a bridge model and point cloud dataset production requirements of a bridge component; the production module is configured to, when the dataset production requirements are the point cloud dataset production requirements of the bridge component, acquire a plurality of bridge components in the bridge model; extract component parameters of each bridge component, and construct a point cloud dataset of each bridge component based on the component parameters of each bridge component and parameters of the bridge model; when the dataset production requirements are the point cloud dataset production requirements of the bridge model, the production module is configured to acquire entity objects corresponding to each bridge component in the bridge model; perform triangulation and laser simulation scanning on the surface of the entity objects to obtain point cloud spatial information of the entity objects, and construct a point cloud dataset of the bridge model according to the point cloud spatial information of the entity objects; when the production module is configured to construct the point cloud dataset of each bridge component based on the component parameters of each bridge component and the parameters of the bridge model, the production module is specifically configured to: automatically correct the component parameters based on the proportions of the bridge components to obtain corrected component parameters of each bridge component; convert the bridge model corresponding to the corrected component parameters of each bridge component into a first mesh model, the first mesh model comprising a plurality of triangular faces; determine sampling points of each triangular face based on a uniform sampling method; construct a point cloud dataset of each bridge component based on the sampling points of each triangular face and the corrected component parameters of each bridge component; when the production module is configured to perform triangulation and laser simulation scanning on the surface of the entity objects to obtain point cloud spatial information of the entity objects, the production module is specifically configured to: read the surface of the entity objects, perform triangulation on the surface of the entity objects to obtain a plurality of triangular patches, and construct labels of each triangular patch based on the bridge component corresponding to the entity objects. The bridge model is converted into a second mesh model, the second mesh model is loaded, and the second mesh model is subjected to laser simulation scanning by using a simulation unmanned aerial vehicle; The laser emission direction of the simulation unmanned aerial vehicle, the unmanned aerial vehicle position, the time interval of emission and reception of the laser signal, and the scanning time stamp in the laser simulation scanning process are acquired; According to the laser emission direction of the simulation unmanned aerial vehicle, the unmanned aerial vehicle position, and the time interval of emission and reception of the laser signal in the laser simulation scanning process, the point cloud coordinates of each triangular facet are determined, and the point cloud coordinates of each triangular facet, the label of each triangular facet, and the point cloud coordinates of each triangular facet of the solid object are taken as the point cloud space information of the solid object.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the method for manufacturing the bridge real scene three-dimensional point cloud data set in any one of claims 1-6.

Images