CN113066162A - Urban environment rapid modeling method for electromagnetic calculation - Google Patents
Urban environment rapid modeling method for electromagnetic calculation Download PDFInfo
- Publication number
- CN113066162A CN113066162A CN202110269641.7A CN202110269641A CN113066162A CN 113066162 A CN113066162 A CN 113066162A CN 202110269641 A CN202110269641 A CN 202110269641A CN 113066162 A CN113066162 A CN 113066162A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- modeling
- primitive
- electromagnetic
- geometric primitive
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/005—Tree description, e.g. octree, quadtree
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/10—Constructive solid geometry [CSG] using solid primitives, e.g. cylinders, cubes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/20—Finite element generation, e.g. wire-frame surface description, tesselation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Computer Graphics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Processing Or Creating Images (AREA)
Abstract
The invention discloses a method for quickly modeling an urban environment for electromagnetic calculation, which comprises the following steps: step A: scanning a modeling area by using a laser radar and an optical camera, and establishing a dense point cloud picture containing electromagnetic material information in the modeling area according to the scanned laser point cloud picture and an optical image; and B: dividing the point cloud in the dense point cloud picture according to the corresponding category of the electromagnetic material information of each point in the dense point cloud to obtain an entity formed by the point cloud; and C: and identifying the characteristics of each object according to the geometric structure of each entity, fitting the characteristics of each object to obtain a reconstruction model of the entity, and further establishing a three-dimensional model of the city. By applying the method, the three-dimensional model which can be suitable for electromagnetic calculation can be established.
Description
Technical Field
The invention relates to the technical field of electromagnetic environment simulation, in particular to a method for quickly modeling an urban environment for electromagnetic calculation.
Background
The urban electromagnetic radiation environment is an important environmental factor, and in recent 20 years, the development and application of electromagnetic technology are increasingly wide, the influence on the surrounding environment, human health and the like is increasingly serious, and the detection of the electromagnetic environment by environmental protection management departments, scientific research departments and the like is highly regarded. For urban electromagnetic environment monitoring, obtaining a high-precision three-dimensional image of an urban area to be monitored is a necessary precondition.
In the prior art, a method for acquiring a high-precision three-dimensional image of an urban area mainly comprises the following steps: satellite mapping and airborne laser radar mapping. However, the traditional satellite mapping image technology has the defects of insufficient data acquisition capability, poor current situation and slow return visit. Compared with satellite mapping, the traditional large airplane is more flexible in data acquisition and higher in image quality, but airplane leasing, airport management and airspace management processes are too complex, and the requirement on cloud layers is relatively high. The environment information is difficult to be rapidly, timely and omnidirectionally acquired only by satellite mapping and man-machine. In order to overcome the defects, at present, a laser radar system based on an unmanned aerial vehicle platform carries out omnidirectional scanning on a region to be measured by integrating a high-frequency laser scanner, a global positioning system and an inertia measurement unit installed on an airplane, can accurately and high-density collect data from the surface of the earth, and can quickly acquire accurate three-dimensional coordinates of a large-area measurement region. The system has the characteristics of high automation degree, small influence of weather, high precision and the like. Laser pulses emitted by the airborne laser radar sensor can partially penetrate through the forest for shielding, and high-precision three-dimensional surface topography data can be directly obtained.
However, the building three-dimensional model established in the prior art only includes the shape data of the building, and the propagation of the electromagnetic wave depends on factors such as the shape of the building, the material of the building and the like, so that the existing building three-dimensional model can only be used in the fields of unmanned driving and navigation, and cannot be directly applied to the simulation of the urban electromagnetic environment. Therefore, how to build a three-dimensional model suitable for electromagnetic calculation is an urgent technical problem to be solved.
Disclosure of Invention
The invention aims to solve the technical problem of how to build a three-dimensional model suitable for electromagnetic calculation.
The invention solves the technical problems through the following technical scheme:
the invention provides a city environment rapid modeling method for electromagnetic calculation, which comprises the following steps:
a: scanning a modeling area by using a laser radar and an optical camera, and establishing a dense point cloud picture containing electromagnetic material information in the modeling area according to the scanned laser point cloud picture and an optical image;
b: according to the category corresponding to the electromagnetic material information of each point in the dense point cloud, the point cloud in the dense point cloud picture is divided to obtain an entity formed by the point cloud, wherein the electromagnetic material information comprises: plants, glass, concrete, ground;
c: identifying the characteristics of each object aiming at the geometric structure of each entity, fitting the characteristics of each object to obtain a reconstruction model of the entity, and further establishing a three-dimensional model of the city, wherein the characteristics of each object comprise: the object features include: wall, door, roof, protrusion and window, and the geometry includes: size, position, orientation, and topology.
Optionally, step a includes:
scanning the modeling area by using a laser radar to obtain a laser point cloud picture of the modeling area, acquiring electromagnetic material information corresponding to each point in the laser point cloud picture, and adding the electromagnetic material information into a label of the point;
scanning the modeling area by using a full-band multi-view optical camera to obtain an optical image; identifying object features contained in the optical image by using an algorithm of a motion acquisition structure, and constructing a sparse point cloud of the optical image based on the object features;
calculating a depth point cloud map of a modeled region from the optical image;
according to the optical image, segmenting the optical image of the modeling area by utilizing a pre-trained electromagnetic medium recognition neural network model to obtain electromagnetic material information corresponding to an object contained in the optical image, and calibrating the corresponding area in the optical image by utilizing the electromagnetic material information;
fusing the depth point cloud picture, the sparse point cloud picture and the calibrated laser point cloud picture, and adding a label to each point according to the coordinate of each point in the fused point cloud picture to obtain a labeled dense point cloud picture, wherein the label comprises: electromagnetic material information of the point, and object characteristic information of the point.
Optionally, the segmenting the optical image of the modeling area by using the pre-trained electromagnetic medium recognition neural network model includes:
the method comprises the steps of training a pre-built neural network model by using an optical image of an object marked with electromagnetic material information as a sample to obtain a converged neural network model, and then segmenting the optical image of a modeling area by using the neural network model.
Optionally, the scanning the modeling area by using the laser radar to obtain the laser point cloud chart of the modeling area includes:
the modeling area is scanned by using the laser radar, and based on the three-dimensional coordinates of the laser radar, the modeling area is modeled by using a formula,
Xpcoordinates on the x-axis for points in the point cloud; xGCoordinates of the laser radar on an x axis; y ispCoordinates on the y-axis for points in the point cloud; y isGCoordinates of the laser radar on the y axis are obtained; s is a laser radar point-to-point vector; theta is an included angle between the pixel corresponding to the laser ranging point P and the middle pixel in the scanning period; b is cos omega sin alpha cos kappa +sin κ sin ω; aerial attitude parameters of laser radar such as roll angle, pitch angle and yaw angle
Obtaining photogrammetric attitude angles (alpha, omega, kappa) through coordinate system transformation; zpHeight of a point in the point cloud; zGIs the height of the lidar.
Optionally, the method further comprises, before step B,
and processing the dense point cloud based on a statistical outlier removing algorithm to obtain the processed dense point cloud.
Optionally, the segmenting the point cloud in the dense point cloud picture in the step B includes:
the point cloud segmentation algorithm based on region growing segments the segmented building point cloud into point clouds of a plurality of entities, wherein the entities comprise: relief topography, level ground, simple building, complicated building, trees.
Optionally, the step C of fitting a reconstruction model of the entity by using the characteristics of each object, and further establishing the three-dimensional model of the city includes:
judging whether the entity belongs to undulating terrain, flat ground, simple buildings, complex buildings and trees;
building modeling is carried out based on a global fitting algorithm under the condition that the entity belongs to a simple building;
under the condition that the entity belongs to a tree, performing data modeling by using a tree skeleton model constructed in advance;
under the condition that the entity belongs to a complex building or an undulating terrain, a Poisson reconstruction algorithm is utilized for modeling;
and under the condition that the entity belongs to a flat ground, modeling by adopting a least square fitting plane method.
Optionally, the building modeling based on the global fitting algorithm includes:
establishing constraints on the characteristics of the building model object, wherein the constraints comprise: area, position, orientation, topology;
by usingThe random sampling consistency algorithm divides the entity to obtain a geometric primitive set { chi } composed of local geometric primitivesi}；
According to the geometric primitive set { χiThe parallel and orthogonal relations between every two geometric primitives contained in the method obtain the combined relation between the geometric primitives, and the maximum non-conflict set is selected from the set formed by the combined relation between the geometric primitives
Aligning each geometric primitive according to the combination relation in the maximum non-conflict set by utilizing a nonlinear optimization algorithm with constraints;
for aligned maximum non-conflicting set
Establishing a plurality of geometric primitive pairs according to the same angle, and establishing a relational graph by taking the geometric primitive pairs as vertexes;
calculating the space distance between every two geometric primitive pairs in the relational graph, and deleting the geometric primitive pairs of which the space distances to other geometric primitive pairs are larger than a set distance;
for each geometric primitive pair, connecting the geometric primitive pair with other geometric primitive pairs by using edges, wherein the geometric primitive pair and the other geometric primitive pairs have similar angles; at the same time, by using the formula,
a constraint is added to the newly added edge, wherein,
sca confidence score for an edge;
the included angle between the direction vector of the geometric primitive i and the direction vector of the geometric primitive j is shown;
the included angle between the direction vector of the geometric primitive k and the direction vector of the geometric primitive l is shown; gcIs edgedConstraining; n isiIs a geometric primitive i; n isjIs a geometric primitive j; n iskIs a geometric primitive k; n islIs a geometric primitive l;
taking the newly added edge set as an initial candidate set, and extracting an isogonal relation set consisting of edges with isogonal relation in the initial candidate set;
by means of the formula (I) and (II),
the set of angular relationships is optimized, wherein,
min is a minimum evaluation symbol; sigma is a summation symbol; ed(Pi,χi) As a set of points PiAnd primitive xiThe error between; w is apd2(p,χi) Is the accumulated error of the data; w is apIs the weight of the geometric primitive; d2(p,χi) Distance of p points to the geometric primitive;
judging whether each side in the optimized set is the same as the side before optimization;
if yes, deleting the primitive pair corresponding to the edge with the lowest confidence score from the same edges before and after the same-angle relation set optimization;
if not, returning to the step of executing the optimization of the corresponding angle relation set until the optimization times reach the set times;
acquiring a set of primitive pairs corresponding to the edge with the lowest confidence score, and aiming at each primitive pair in the set, utilizing a formula,
computing a confidence score for the primitive pair and a constraint, wherein,
spis the confidence score of the primitive pair; p is a radical ofiPoints on each axis of the geometric primitive i; p is a radical ofjPoints on each axis of the geometric primitive j; | | is a modulo symbol; (p)i-pj)TIs the transposition of the vector; gpConstraint of the primitive pair;
carrying out coaxial alignment processing on the primitive pairs in the maximum relation subset according to the direction of a normal vector between each primitive pair in the maximum relation subset, and sorting the primitive pairs in the optimized isometric relation set according to the confidence scores of the primitive pairs to obtain a corresponding maximum relation subset; and obtaining a three-dimensional model of the building according to the relationship between the entities corresponding to the maximum relationship subset.
Optionally, in a case that the entity belongs to a complex building, performing data modeling by using a poisson reconstruction algorithm includes:
taking the point cloud data corresponding to the fitted building entity as input, and determining an indication function of the point cloud data according to the vector direction of each point cloud;
after smooth filtering is carried out on the indicating function, a gradient field of the indicating function is calculated, and the gradient field is discretized by establishing an implicit function of an octree for adjusting the depth of a grid according to the density of point cloud data;
carrying out segmentation sampling on the discretized gradient field, and calculating a vector field of a sampling point by using a cubic linear interpolation method;
the poisson equation is solved and the solution,
a scalar indicator function of the point cloud data is obtained, wherein,
Δ is laplace operator; x is an indicator function; v is a vector differential operator;
a space vector of the point cloud data;
based on a scalar indication function, extracting an isosurface by adopting a mobile cube algorithm, and establishing a three-dimensional model of the building based on the isosurface.
Compared with the prior art, the invention has the following advantages:
by applying the embodiment of the invention, the optical image acquisition cost is lower, the acquisition is more convenient, so that the urban three-dimensional model can be established more quickly, and meanwhile, when the urban three-dimensional model is established, the multi-source point cloud data combining the optical image matching and the laser scanning is selected to carry out the matching fusion mode to obtain the urban three-dimensional point cloud data, so that the established urban three-dimensional model has high precision.
Drawings
FIG. 1 is a schematic flow chart of a method for rapidly modeling an urban environment for electromagnetic computing according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a method for quickly modeling an urban environment for electromagnetic computing according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a process for obtaining a dense point cloud according to the present invention;
FIG. 4 is a schematic diagram of a dense point cloud of a modeled region obtained by an embodiment of the present invention;
FIG. 5 is a simplified three-dimensional model of a building reconstructed according to an embodiment of the present invention;
fig. 6 is a three-dimensional model of a complex building reconstructed according to an embodiment of the present invention.
Detailed Description
The following examples are given for the detailed implementation and specific operation of the present invention, but the scope of the present invention is not limited to the following examples.
Fig. 1 is a schematic flow chart of a method for fast modeling an urban environment for electromagnetic computing according to an embodiment of the present invention, as shown in fig. 1, the method includes:
s101: and scanning the modeling area by using a laser radar and an optical camera, and establishing a dense point cloud picture containing electromagnetic material information in the modeling area according to the scanned laser point cloud picture and the optical image.
Fig. 2 is a schematic diagram of a method for quickly modeling an urban environment for electromagnetic computing according to an embodiment of the present invention, as shown in fig. 2,
firstly, an airborne laser radar is used for scanning a modeling area, and the laser pulse emitted by the laser radar and partially penetrates through trees to be shielded, so that a high-precision image of the earth surface or a building can be obtained; meanwhile, the airborne inertial measurement unit and the global positioning unit can obtain the accurate coordinates and the motion state of the unmanned aerial vehicle, and further can obtain laser point cloud pictures of all objects in a modeling area, such as trees, the ground, buildings and the like. In the embodiment of the invention, the scanning mode of the laser radar is different from the omnidirectional scanning in the prior art for acquiring the point cloud images of all surfaces of the entity, and in the embodiment of the invention, a unidirectional scanning mode or a reciprocating scanning mode is used for acquiring only the point cloud images of one surface or two surfaces and other parts of surfaces of the entity without acquiring all the laser point cloud images of all the surfaces.
A Laser Detection and Ranging (LiDAR) system is an active earth observation system. The system integrates a laser ranging technology, a computer technology, an inertia measurement unit and a GPS differential positioning technology. When the airborne laser radar carries out aerial photography, the three-dimensional accurate position (X) of the laser radar in the air is determined by a differential global positioning systemG，YG，ZG) The inertial measurement unit measures the pitch angle, roll angle and course angle of the aircraft
The laser pulse emitted by laser radar can directly measure the topographic relief condition, and the projection center of laser radar can be projected on the unknown point P (X) on the groundG，YG，ZG) The vector S of (a) can be accurately measured, and therefore, the modeling area can be scanned using the lidar, based on the three-dimensional coordinates of the lidar, using a formula,
Xpcoordinates on the x-axis for points in the point cloud; xGCoordinates of the laser radar on an x axis; y ispCoordinates on the y-axis for points in the point cloud; y isGCoordinates of the laser radar on the y axis are obtained; s is a laser radar point-to-point vector; theta is an included angle between the pixel corresponding to the laser ranging point P and the middle pixel in the scanning period;
b ═ cos ω · sin α · cos κ + sin κ · sin ω; aerial attitude parameters of laser radar such as roll angle, pitch angle and yaw angle
Obtaining photogrammetric attitude angles (alpha, omega, kappa) through coordinate system transformation; zpHeight of a point in the point cloud; zGIs the height of the lidar.
Then, the pre-trained neural network model is used for obtaining electromagnetic material information corresponding to each point in the laser cloud point diagram, and then the electromagnetic material information is added into the label of the point, so that the laser cloud point diagram containing the electromagnetic material information can be obtained, wherein the electromagnetic material information is glass, a wall body, the ground, trees and the like. It is understood that the point cloud electromagnetic material information in the laser point cloud image of the similar area to the modeling area or the tenth or hundredth area in the modeling area can be marked manually, and then the neural network model can be trained by using the manually marked laser point cloud image as a sample.
Then, respectively carrying out semantic identification operations of reconstructing the sparse point cloud, the depth map corresponding to the optical image and the optical image:
shooting a full-frequency-band optical image of a modeling area by using a five-eye lens carried on an airborne camera or a cluster; identifying object features contained in the optical image by using an algorithm of a motion acquisition structure, and constructing a sparse point cloud of the optical image by using the prior art based on the object features, wherein the object features comprise: corner points, walls, doors, roofs, protrusions, and windows. The process is the prior art, and comprises the following specific processes: extracting metadata of the image, extracting key features of the image (such as corner points of a building target), matching and tracking the features, and reconstructing the sparse point cloud.
It should be noted that, the process of establishing a three-dimensional point cloud image based on an optical image captured by a multi-view lens is the prior art, and the embodiment of the present invention is not described herein again.
According to the optical image, the optical image of the modeling area is segmented by utilizing a pre-trained electromagnetic medium recognition neural network model, electromagnetic material information corresponding to objects, such as windows, walls, roofs, floors and the like, contained in the optical image is obtained, and the corresponding area in the optical image is calibrated by utilizing the electromagnetic material information.
In practical application, optical images in some urban areas can be shot in advance, objects of different electromagnetic materials contained in the images are respectively marked to obtain training samples, the samples are used for training a pre-built neural network model (using a Deeplab structure) until the neural network model converges to obtain a trained model, and then the neural network model is used for carrying out semantic segmentation on the optical images in the modeling area.
Then, the depth point cloud image, the sparse point cloud image and the calibrated laser point cloud image are fused, and the fusion process comprises the following specific steps: the method is a process of calculating the precise mapping of the space geometric relationship among different point cloud sets, solving coordinate conversion parameters and carrying out rigid body transformation on a data set to be converted. The embodiment of the invention uses an Iterative Closest Point (ICP) algorithm based on a local feature matching initial value. The ICP algorithm is an iterative algorithm for realizing point cloud fusion by global registration, and in order to further improve the precision of the point cloud fusion, a point cloud rotation matrix constructed by seven parameters or quaternions is obtained by using point-line and other local feature identification and matching to realize multi-point cloud fusion. And adding corresponding material information as the content of a label to each point in the fused point cloud picture according to the coordinates of each point in the fused point cloud picture to obtain a dense point cloud picture added with the label, wherein the label comprises: electromagnetic material information of the point, and object characteristic information of the point.
In the process of fusing the laser point cloud picture and the optical image, the depth information of each pixel point is calculated for each unmanned aerial vehicle image based on the estimated acquisition point position by using the multi-view geometry principle, the two-dimensional optical image acquired by the unmanned aerial vehicle can be converted into a three-dimensional space by using the depth information, and then an image loaded with electromagnetic material information (electromagnetic material coefficient) is introduced into the reconstruction. Further, can install five-eye optical camera and laser radar on same unmanned aerial vehicle, and then can utilize the unmanned aerial vehicle position that laser radar measured to carry out the generation of degree of depth point cloud picture. The embodiment of the invention takes the depth information in the dense point cloud as a medium, converts the pixels of the optical picture of the unmanned aerial vehicle loaded with the electromagnetic material information into a three-dimensional space by using the idea of depth information sharing, and loads the electromagnetic material information into the point cloud to obtain the point cloud containing the electromagnetic material information, so that each point in the dense point cloud picture obtained by the method comprises a piece of label information besides the position and color information, and the label information comprises the corresponding electromagnetic material information.
Fig. 4 is a schematic diagram of the dense point cloud of the modeling area obtained in the embodiment of the present invention, and as shown in fig. 4, the above-described processing is performed on each image, so that the dense 3d point cloud of the entire area based on the optical image reconstruction can be finally obtained, and then the laser point cloud data obtained by the unmanned aerial vehicle cluster is obtained and processed.
The points in the dense point cloud chart are discrete and irregularly distributed in a three-dimensional space, wherein the points comprise points on complex ground surfaces, trees and other ground objects, and the application of generating a three-dimensional model of an urban area and ground point data is difficult. In order to eliminate noise points in the data and further reduce the influence of the noise points on the modeling of the three-dimensional model, the noise point data needs to be removed by performing an outlier processing on the acquired dense point cloud before the step S102 is performed. The processing algorithm needs to select a proper classification threshold according to different terrain conditions, but it is not easy to select a parameter capable of removing the ground features with different sizes for a complex ground surface, so that it is very necessary to individually set the corresponding parameter according to specific ground features and terrain conditions. However, it is not practical if different parameters are artificially set for specific land features and landforms, respectively, and in the embodiment of the present invention, typical characteristics of an urban environment and timeliness of data processing requirements are comprehensively considered, and an outlier removing method based on statistics is selected to process point clouds. The average distance between each point and adjacent points in the point cloud plus the deviation tolerance predefined by the user is used as a threshold, namely, if the number of adjacent points contained in a sphere of which the radius of a certain point in the dense point cloud is equal to the average distance plus the tolerance is less than a defined threshold, the point is considered as an outlier and is removed. According to the embodiment of the invention, under the condition of not depending on ground objects and landforms too much, simpler parameters can be set, and ground points in mass point cloud data can be automatically extracted, so that the removal efficiency of noise points is improved.
S102: according to the category corresponding to the electromagnetic material information of each point in the dense point cloud, the point cloud in the dense point cloud picture is divided to obtain an entity formed by the point cloud, wherein the electromagnetic material information comprises: plants, glass, concrete, ground.
In the process of segmenting dense point clouds in urban environments of modeling areas to obtain point clouds of each object in the dense point clouds, useless information such as vehicles and the like exists near the point clouds of the buildings due to the fact that environments of the buildings in the urban environments are complex, and point cloud distribution corresponding to the useless information is often located near other electromagnetic media, and therefore the difficulty in removing the useless information is large. In practical application, a point cloud segmentation mode based on a voxel can be used for directly segmenting the point cloud and a point cloud segmentation mode based on a multi-vision picture.
However, the inventors found that the voxel-based method is too resource-consuming, and the way of directly segmenting the point cloud is simpler than the voxel-based segmentation method, but still consumes a large amount of computing resources. Although the reconstruction method based on multi-view reconstruction is simple in calculation, the disadvantages are also obvious: in the process of projecting the 3d structure to 2d, relatively serious structure information loss exists, and in the process of projecting the 3d structure to 2d information, the selection of the projection position, namely the position of the virtual camera, has a great influence on the reconstruction result, so that the consistency of the result based on multi-view reconstruction is poor.
In order to solve the above problems, in the embodiments of the present invention, depth information in dense point cloud is used as a medium, and the idea of depth information sharing is used to transform the pixels of the optical picture of the unmanned aerial vehicle loaded with electromagnetic material information into a three-dimensional space, so that the electromagnetic material information can be loaded into the point cloud to obtain the point cloud containing the electromagnetic material information. Each point in the point cloud comprises a piece of label information besides position and color information, and the label information comprises the electromagnetic material information corresponding to the point, so that the point cloud segmentation can be performed by using the label information. Finally, point cloud images of single objects are obtained by carrying out point cloud segmentation on point clouds of the same category based on a region growing algorithm according to the known distribution rules of buildings, trees, the ground, lakes and the like in cities.
S103: identifying the characteristics of each object aiming at the geometric structure of each entity, fitting the characteristics of each object to obtain a reconstruction model of the entity, and further establishing a three-dimensional model of the city, wherein the characteristics of each object comprise: wall, window, roof, trees, earth's surface, just the geometry includes: size, position, orientation, and topology.
Firstly, judging which entity belongs to a simple building, a complex building, a tree, or an undulating terrain, a flat ground (under the condition that the entity belongs to the flat ground, a least square fitting plane method is adopted for data modeling, the method is the prior art, and the invention is not described); the point cloud semantic segmentation can be carried out by using a pre-trained deep learning model, the point cloud is obtained through reconstruction, each point of the point cloud contains semantic category information, and then entities are classified according to the category information.
And under the condition that the entity belongs to a simple building, building modeling is carried out based on a global fitting algorithm.
And under the condition that the entity belongs to the tree, performing data modeling by using a tree skeleton model constructed in advance. The tree skeleton is the basis for building a three-dimensional tree model, and the basic idea is to extract an initial skeleton of a tree from ground laser point cloud, further optimize the initial skeleton and finally realize the extraction of the tree skeleton. Extracting 'similar trunk points' of the trees from the preprocessed original tree point cloud according to the difference of the laser reflection intensity values, organizing the 'similar trunk points' and the rest 'non-trunk points' by adopting a 'minimum spanning tree' algorithm in a graph theory to form an initial skeleton of the trees, and finally further optimizing the initial skeleton of the trees through 'point density adjustment' and 'branch smoothing' to obtain the more accurate and reasonably distributed skeleton of the trees.
And when the number of planes contained in the entity is more than a set number, such as 100, and the building is judged to belong to a complex building, building modeling is carried out by using a Poisson reconstruction algorithm.
Firstly, three-dimensional modeling is carried out on a simple building, and because the point cloud is obtained in the step S102, the difficulty of directly carrying out electromagnetic calculation is high. Therefore, surface reconstruction needs to be performed on the monomer point cloud according to the electromagnetic material information, and then electromagnetic calculation is performed on the basis of the reconstructed surface so as to reduce the calculation complexity.
The surface reconstruction method commonly used in the prior art is a method of dironi triangulation or von neumoniae mapping. The reconstruction result can basically reflect the point cloud position and has higher precision, the general surface reconstruction mode has wide application and can be suitable for the rapid modeling of almost all point clouds, however, the quantity of triangular surfaces contained in a three-dimensional model reconstructed by the method is very large, usually tens of thousands or even tens of thousands, and if the three-dimensional model is used as the direct input of a ray tracing model, the time complexity of calculation is very high. At least hundreds of buildings are often distributed inside urban areas, and even if calculations can be made, the time consuming costs are very high.
In order to solve the problems, the inventor finds that the buildings in the modeling area belong to objects with regular structures, and particularly the buildings have larger occupation ratio in the urban environment modeling process; therefore, the building model can be reconstructed by adopting a building facade reconstruction algorithm based on the prior knowledge. The building facade comprises important object characteristics such as electromagnetic reflecting surfaces such as wall surfaces, roofs and protrusions and windows, each element has geometrical structural constraint which is different from other object characteristics, such as one or a combination of area, position, direction and topological structure:
size: by area size, the wall surface is easily separated from other features, and some small noise partitions are easily filtered out.
Position: specific features may only be present at specific locations, such as windows and doors on walls and protrusions, while roofs are always above walls.
The direction is as follows: the wall surface is mostly vertical to the ground, the roof is not vertical to the ground, and the wall surface can only be horizontal or inclined: it is also possible to add a floor as an auxiliary feature for better identification of the main wall surface.
Topological structure: the topology between features is often an important cue, such as a wall always intersecting the ceiling, and a roof always intersecting a wall.
Dot density: since the laser beam is reflected to obtain the three-dimensional coordinates of the measured object, the characteristics can be distinguished by the point density.
The object identification method based on the electromagnetic medium semantic segmentation can add the category to each entity in the point cloud, but in order to more accurately identify the object characteristics of the vertical face, each category of semantic characteristics has unique attributes, and the semantic characteristics can be automatically, efficiently and accurately extracted by regularizing the attributes into specific feature constraints.
Since artificial engineering elements comprise basic geometrical structures such as planes, cylinders, spheres and cones. The method comprises the steps of utilizing a random sampling consistency RANSAC algorithm to segment point clouds of entities to obtain elements of geometric structures contained in the point clouds, wherein the elements contained in all the entities form a geometric element set { chi }i}. Then, global constraints such as direction relation, displacement relation, equality relation and the like contained in the geometric primitive set are extracted step by step so as to improve the accuracy of the reconstructed model and the universality of the algorithm. In the process of model reconstruction, the relationship between three types of primitives is mainly concerned, which is also called as alignment relationship: directional relationships, such as parallel and orthogonal, permutation relationships, such as co-planar or coaxial presence between elements; there is an equality relationship between the primitives. Based on the definition of the alignment relationship, the alignment global direction relationship is obtained by optimizing the model parameters from the global consideration, and the alignment global direction relationship can be accurately matched with the original data. Considering the sensitivity of different types of relationships to noise,and adopting a step-by-step extraction strategy, namely, preferentially processing the easiest relationship each time. In the embodiment of the invention, the relationship between the element pairs can be well recovered by establishing the relationship graph and the constraint of the data item for the global constraint relationship.
The appearance of the urban artificial building is relatively regular, and objects such as wall surfaces, ground surfaces, roofs and the like are basically in parallel or orthogonal relation. According to the geometric primitive set { χiThe parallel and orthogonal relations between every two geometric primitives contained in the structure are obtained to obtain the combined relation between the geometric primitives, namely a plurality of candidate relation sets C0(ii) a From C0Extracting the largest non-conflict set
Associating primitives with relationships using a constrained nonlinear optimization algorithm
And data alignment.
Then, alignment of the same direction primitives is performed: in the building of the urban scene, a plurality of regular structures exist, and the regularity often causes equal angles to exist between the elements, and the elements with the same angles form element pairs. While any angular relationship necessarily contains a pair of primitives, i.e., four primitives. Therefore, the relationship graph G is established by taking the geometric primitive pairs as the verticese。
Then, the relation graph GeEach vertex of (a) represents a primitive pair out of order. In noisy data, spatially distant primitives can produce erroneous relationships. Thus, if the primitive pair { χ }i,χjIf the distance between them exceeds a certain distance, the corresponding node, i.e. vertex, will be deleted. After the above processing, the relationship diagram GeContains only O (m) vertices.
For each geometric primitive pair, connecting the geometric primitive pair with other geometric primitive pairs by using edges, wherein the geometric primitive pair and the other geometric primitive pairs have similar angles; at the same time, by using the formula,
a constraint is added to the newly added edge, wherein,
sca confidence score for an edge;
the included angle between the direction vector of the geometric primitive i and the direction vector of the geometric primitive j is shown;
the included angle between the direction vector of the geometric primitive k and the direction vector of the geometric primitive l is shown; gcA constraint that is an edge; n isiIs a geometric primitive i; n isjIs a geometric primitive j; n iskIs a geometric primitive k; n islIs the geometric primitive/.
According to the above processing rule, the obtained initial candidate set is Ce＝{c1,c2,...}。
Then, alignment of the parallel and orthogonal primitives is performed: similar to the processing method of the direction alignment relationship, the edge C epsilon C is processed step by step according to the order of the scores of the edges from high to lowe. And extracting the primitive pair relation with the equal angle relation to obtain a set of primitive pair relation. Meanwhile, the isometric relationship is transitive, so that the relationship diagram GeIt cannot be a circular graph, and such a process significantly reduces the number of edges in the graph. Furthermore, the extracted equiangular relationship set still has a possibly conflicting relationship, and the conflicting relationship needs to be detected by means of interior point nonlinear programming.
Then the formula is utilized to obtain the final product,
the set of angular relationships is optimized, wherein,
min is a minimum evaluation symbol; sigma is a summation symbol; ed(Pi,χi) As a set of points PiAnd primitive xiThe error between; w is apd2(p,χi) Is the accumulated error of the data; w is apIs the weight of the geometric primitive; d2(p,χi) Is composed ofDistance of p points to the geometric primitive. And if the optimization result is not changed, deleting the primitive pair relationship with the lowest score from the optimized pair relationship set. And then optimizing the relation of the rest primitives again, wherein generally, only two times of optimization are needed to obtain a relatively ideal effect.
Then, alignment of the primitives of the permutation relationship is performed: most of urban buildings have coplanar and coaxial parts, and after the directional alignment relationship is processed, the aligned directional relationship is kept unchanged, and then the replacement relationship is processed and aligned. Since the algorithm has already performed a direction-aligned process on the data, if two primitives χiHexix-jIs parallel, then the pair of primitives is likely to be coaxial. By means of the interaxial distance between elements, the fraction s is obtainedpAnd a set C of aligned relationships of the primitives carrying out the permutation relationshippConstraint g of the maximum subset of relationships ofp：
spis the confidence score of the primitive pair; p is a radical ofiPoints on each axis of the geometric primitive i; p is a radical ofjPoints on each axis of the geometric primitive j; | | is a modulo symbol; (p)i-pj)TIs the transposition of the vector; gpIs the constraint of the primitive pair. Coplanar pairs of elements are typically spaced apart by some distance. All cylinders and cones need not be reprocessed due to the extraction via the coaxial relationship. Therefore, the extraction of the coplanar relationship is only to detect the coplanar relationship between the two planar models. For primitive pair χiHexix-jIf their normal vectors are in the same direction, then there is di＝dj(ii) a If the two are reversed then di＝-dj。
And then arranging the extracted relations according to the descending order of confidence scores, and extracting the maximum relation subset of the ordered set
And the alignment relationship thereof comprises the three alignment relationships. And finally, carrying out minimum data item error processing to obtain the aligned primitive.
Fig. 5 is a reconstructed simple building three-dimensional model according to an embodiment of the present invention, and as shown in fig. 5, the fitted object features are aligned according to the aligned primitives, so as to obtain a building three-dimensional model.
The building facade (electromagnetic reflecting surface) is fitted using an algorithm of global fitting: given an input point cloud set, the algorithm simultaneously performs local primitive matching and global primitive relationship extraction. The algorithm can finally extract and reconstruct the reflecting surfaces of the buildings in the city, each reflecting surface comprises the electromagnetic material coefficient information of the reflecting surface, and the information is obtained by the electromagnetic medium semantic segmentation based on deep learning. The global fitting algorithm in the embodiment of the invention is the existing GlobFit algorithm.
The three-dimensional reconstruction of the building needs to identify and extract basic geometric structures such as points, lines or surfaces and the like in point cloud data, and the algorithm used by the invention has the following advantages: first, optical image reconstruction and laser point clouds cannot scan a complete building due to the presence of trees or vehicle obscuration. Through the priori semantic knowledge, the defect can be made up, so that the sealing performance of a reconstruction result is ensured; second, depending on the type of semantics, the building model may be given more electromagnetic material information, such as ground, owner, target name or code number, etc., which helps to build a more refined electromagnetic map.
In addition, the embodiment of the invention can directly use the building model and the electromagnetic material information thereof to rapidly acquire the electromagnetic situation of the city, and further can be used for electromagnetic calculation. Furthermore, the influence of absorbers and diffuse scattering objects propagated by electromagnetic waves can be considered in electromagnetic calculation so as to obtain more accurate electromagnetic distribution.
When the fitted building is a complex building or an undulating terrain, the three-dimensional model can be reconstructed by using a poisson reconstruction method, and the reconstruction process of the corresponding three-dimensional model is described below by taking the complex building as an example:
the method of poisson reconstruction is selected for the complex building reconstruction of urban areas. The principle of the method is to convert the surface reconstruction problem of the point cloud into a solving process of a Poisson equation in a space. The poisson equation is used for solving the tone mapping problem of the image in the strong non-static state range at first and also solving the seamless editing problem of the image in the connection area. The typical advantage of poisson's equation is to solve the problem of large scale integrity, which is well-suited for complex architectural point clouds.
The poisson equation under the three-dimensional rectangular coordinate system is as follows:
and taking the point cloud data of the urban building as a set S, sampling to obtain a subset S belonging to S, and obtaining an object edge S. The gradient index function is a zero vector field that is nearly ubiquitous, except near the surface, where it is equal to the surface normal. Therefore, the subset of sampling points can be regarded as an index function of a gradient model of the sample, and the problem of calculating the index function is reduced to an inverse gradient operator, i.e. determining the gradient of a scalar function X that most closely approximates the standard vector field. This gradient determines the vector field in dependence on the sample
However, the gradient and the vector field are not completely fitted, and approximate calculation is needed
Adding a non-convergence operator to convert the variation problem into a Poisson problem:
obtaining a vector field
Then, the Poisson equation can be solved to obtain a scalar quantityIndicating function X, but not an exact solution, because of the vector field
Usually, the product is not obtained, and a poisson equation needs to be constructed by using a divergence operator to find the optimal approximate least square estimation solution.
Therefore, point cloud data corresponding to the fitted building entity are used as input, and an indicating function of the point cloud data is determined according to the vector direction of each point cloud;
since the indicator function X is a piecewise function, calculating the gradient directly yields infinite values, and, using a formula,
after smooth filtering of the indicative function, calculating a gradient field of the indicative function and discretizing the gradient field by establishing an implicit function of an octree adjusting the depth of the mesh according to the density of the point cloud data, wherein,
is an object, M is a surface edge, XMIs an object
Is used to indicate the function of (a),
is a function of a smoothing filter that is,
is the inner normal direction of point p. (ii) a
And (3) carrying out segmentation sampling on the discretized gradient field, and calculating a vector field of a sampling point by using a cubic linear interpolation method: dividing discrete points into small curved patches
Facet ρSIs approximated by the following equation:
The poisson equation is solved and the solution,
a scalar indicator function of the point cloud data is obtained, wherein,
Δ is laplace operator; x is an indicator function; v is a vector differential operator;
a space vector of the point cloud data;
based on a scalar indication function, extracting an isosurface by adopting a mobile cube algorithm, and establishing a three-dimensional model of the building based on the isosurface, wherein the established three-dimensional model is shown in FIG. 6.
According to the embodiment of the invention, the dense point cloud result of the whole urban environment is segmented according to the self-adaptive surface reconstruction requirement of the electromagnetic medium, and each ground object is obtained according to the category (buildings, water surface, ground, trees and the like) of the geographic elements; and then, carrying out surface reconstruction on the electromagnetic media of different types to obtain a three-dimensional model for establishing a three-dimensional reconstructed model of a specific region, and further drawing an electromagnetic situation distribution map of the region by using an electromagnetic calculation method.
In addition, according to the characteristics and the timeliness requirements of the urban environment, the multi-source point cloud data fusion mode combining optical image matching and laser scanning is selected to obtain the urban three-dimensional point cloud data.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (9)
1. A method for rapidly modeling an urban environment for electromagnetic computing, the method comprising the steps of:
a: scanning a modeling area by using a laser radar and an optical camera, and establishing a dense point cloud picture containing electromagnetic material information in the modeling area according to the scanned laser point cloud picture and an optical image;
b: according to the category corresponding to the electromagnetic material information of each point in the dense point cloud, the point cloud in the dense point cloud picture is divided to obtain an entity formed by the point cloud, wherein the electromagnetic material information comprises: plants, glass, concrete, ground;
c: identifying the characteristics of each object aiming at the geometric structure of each entity, fitting the characteristics of each object to obtain a reconstruction model of the entity, and further establishing a three-dimensional model of the city, wherein the characteristics of each object comprise: wall, door, roof, protrusion and window, and the geometry includes: size, position, orientation, and topology.
2. The urban environment rapid modeling method for electromagnetic computing according to claim 1, characterized in that: the specific implementation of the step A comprises the following steps:
scanning the modeling area by using a laser radar to obtain a laser point cloud picture of the modeling area, acquiring electromagnetic material information corresponding to each point in the laser point cloud picture, and adding the electromagnetic material information into a label of the point;
scanning the modeling area by using a full-band multi-view optical camera to obtain an optical image; identifying object features contained in the optical image by using an algorithm of a motion acquisition structure, and constructing a sparse point cloud of the optical image based on the object features;
calculating a depth point cloud map of a modeled region from the optical image;
according to the optical image, segmenting the optical image of the modeling area by utilizing a pre-trained electromagnetic medium recognition neural network model to obtain electromagnetic material information corresponding to an object contained in the optical image, and calibrating the corresponding area in the optical image by utilizing the electromagnetic material information;
fusing the depth point cloud picture, the sparse point cloud picture and the calibrated laser point cloud picture, and adding a label to each point according to the coordinate of each point in the fused point cloud picture to obtain a labeled dense point cloud picture, wherein the label comprises: electromagnetic material information of the point, and object characteristic information of the point.
3. The urban environment rapid modeling method for electromagnetic computing according to claim 2, characterized in that: the method for segmenting the optical image of the modeling area by utilizing the pre-trained electromagnetic medium recognition neural network model comprises the following steps:
the method comprises the steps of training a pre-built neural network model by using an optical image of an object marked with electromagnetic material information as a sample to obtain a converged neural network model, and then segmenting the optical image of a modeling area by using the neural network model.
4. The urban environment rapid modeling method for electromagnetic computing according to claim 2, characterized in that: the method for scanning the modeling area by using the laser radar to obtain the laser point cloud picture of the modeling area comprises the following steps:
the modeling area is scanned by using the laser radar, and based on the three-dimensional coordinates of the laser radar, the modeling area is modeled by using a formula,
Xpcoordinates on the x-axis for points in the point cloud; xGCoordinates of the laser radar on an x axis; y ispCoordinates on the y-axis for points in the point cloud; y isGCoordinates of the laser radar on the y axis are obtained; s is lidarA vector of points; theta is an included angle between the pixel corresponding to the laser ranging point P and the middle pixel in the scanning period; b ═ cos ω · sin α · cos κ + sin κ · sin ω; aerial attitude parameters of laser radar such as roll angle, pitch angle and yaw angle
Obtaining photogrammetric attitude angles (alpha, omega, kappa) through coordinate system transformation; zpHeight of a point in the point cloud; zGIs the height of the lidar.
5. The urban environment rapid modeling method for electromagnetic computing according to claim 1, characterized in that: and before the step B, processing the dense point cloud based on a statistical outlier removing algorithm to obtain the processed dense point cloud.
6. The urban environment rapid modeling method for electromagnetic computing according to claim 1, characterized in that: in the step B, segmenting the point cloud in the dense point cloud picture, including:
the point cloud segmentation algorithm based on region growing segments the segmented building point cloud into point clouds of a plurality of entities, wherein the entities comprise: relief topography, flat ground topography, simple construction, complex construction, trees.
7. The urban environment rapid modeling method for electromagnetic computing according to claim 1, characterized in that: and C, fitting a reconstruction model of the entity by using the characteristics of each object to further establish a three-dimensional model of the city, wherein the method comprises the following steps:
judging whether the entity belongs to undulating terrain, flat ground, simple buildings, complex buildings and trees;
building modeling is carried out based on a global fitting algorithm under the condition that the entity belongs to a simple building;
under the condition that the entity belongs to a tree, performing data modeling by using a tree skeleton model constructed in advance;
under the condition that the entity belongs to a complex building or an undulating terrain, carrying out data modeling by utilizing a Poisson reconstruction algorithm;
and under the condition that the entity belongs to a flat ground, performing data modeling by adopting a least square fitting plane method.
8. The method for rapidly modeling an urban environment for electromagnetic computing according to claim 7, wherein: the building modeling based on the global fitting algorithm comprises the following steps:
establishing constraints on the characteristics of the building model object, wherein the constraints comprise: area, position, orientation, topology;
the entity is divided by utilizing a random sampling consistency algorithm to obtain a geometric primitive set { chi } composed of local geometric primitivesi}；
According to the geometric primitive set { χiThe parallel and orthogonal relations between every two geometric primitives contained in the method obtain the combined relation between the geometric primitives, and the maximum non-conflict set is selected from the set formed by the combined relation between the geometric primitives
Aligning each geometric primitive according to the combination relation in the maximum non-conflict set by utilizing a nonlinear optimization algorithm with constraints;
for aligned maximum non-conflicting set
Establishing a plurality of geometric primitive pairs according to the same angle, and establishing a relational graph by taking the geometric primitive pairs as vertexes;
calculating the space distance between every two geometric primitive pairs in the relational graph, and deleting the geometric primitive pairs of which the space distances to other geometric primitive pairs are larger than a set distance;
for each geometric primitive pair, connecting the geometric primitive pair with other geometric primitive pairs by using edges, wherein the geometric primitive pair and the other geometric primitive pairs have similar angles(ii) a At the same time, by using the formula,
a constraint is added to the newly added edge, wherein,
sca confidence score for an edge;
the included angle between the direction vector of the geometric primitive i and the direction vector of the geometric primitive j is shown;
the included angle between the direction vector of the geometric primitive k and the direction vector of the geometric primitive l is shown; gcA constraint that is an edge; n isiIs a geometric primitive i; n isjIs a geometric primitive j; n iskIs a geometric primitive k; n islIs a geometric primitive l;
taking the newly added edge set as an initial candidate set, and extracting an isogonal relation set consisting of edges with isogonal relation in the initial candidate set;
by means of the formula (I) and (II),
the set of angular relationships is optimized, wherein,
min is a minimum evaluation symbol; sigma is a summation symbol; ed(Pi,χi) As a set of points PiAnd primitive xiThe error between; w is apd2(p,χi) Is the accumulated error of the data; w is apIs the weight of the geometric primitive; d2(p,χi) Distance of p points to the geometric primitive;
judging whether each side in the optimized set is the same as the side before optimization;
if yes, deleting the primitive pair corresponding to the edge with the lowest confidence score from the same edges before and after the same-angle relation set optimization;
if not, returning to the step of executing the optimization of the corresponding angle relation set until the optimization times reach the set times;
acquiring a set of primitive pairs corresponding to the edge with the lowest confidence score, and aiming at each primitive pair in the set, utilizing a formula,
computing a confidence score for the primitive pair and a constraint, wherein,
spis the confidence score of the primitive pair; p is a radical ofiPoints on each axis of the geometric primitive i; p is a radical ofjPoints on each axis of the geometric primitive j; | | is a modulo symbol; (p)i-pj)TIs the transposition of the vector; gpConstraint of the primitive pair;
carrying out coaxial alignment processing on the primitive pairs in the maximum relation subset according to the direction of a normal vector between each primitive pair in the maximum relation subset, and sorting the primitive pairs in the optimized isometric relation set according to the confidence scores of the primitive pairs to obtain a corresponding maximum relation subset; and obtaining a three-dimensional model of the building according to the relationship between the entities corresponding to the maximum relationship subset.
9. The method for rapidly modeling an urban environment for electromagnetic computing according to claim 7, wherein: under the condition that the entity belongs to a complex building, data modeling is carried out by utilizing a Poisson reconstruction algorithm, and the method comprises the following steps:
taking the point cloud data corresponding to the fitted building entity as input, and determining an indication function of the point cloud data according to the vector direction of each point cloud;
after smooth filtering is carried out on the indicating function, a gradient field of the indicating function is calculated, and the gradient field is discretized by establishing an implicit function of an octree for adjusting the depth of a grid according to the density of point cloud data;
carrying out segmentation sampling on the discretized gradient field, and calculating a vector field of a sampling point by using a cubic linear interpolation method;
the poisson equation is solved and the solution,
a scalar indicator function of the point cloud data is obtained, wherein,
Δ is laplace operator; x is an indicator function;
is a vector differential operator;
a space vector of the point cloud data;
based on a scalar indication function, extracting an isosurface by adopting a mobile cube algorithm, and establishing a three-dimensional model of the building based on the isosurface.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110269641.7A CN113066162B (en) | 2021-03-12 | 2021-03-12 | Urban environment rapid modeling method for electromagnetic calculation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110269641.7A CN113066162B (en) | 2021-03-12 | 2021-03-12 | Urban environment rapid modeling method for electromagnetic calculation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113066162A true CN113066162A (en) | 2021-07-02 |
| CN113066162B CN113066162B (en) | 2022-04-29 |
Family
ID=76560187
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110269641.7A Active CN113066162B (en) | 2021-03-12 | 2021-03-12 | Urban environment rapid modeling method for electromagnetic calculation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113066162B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113532424A (en) * | 2021-08-10 | 2021-10-22 | 广东师大维智信息科技有限公司 | Integrated equipment for acquiring multidimensional information and cooperative measurement method |
| CN113644997A (en) * | 2021-10-14 | 2021-11-12 | 中国民用航空总局第二研究所 | Electromagnetic environment detection method, device and system |
| CN114167505A (en) * | 2021-11-22 | 2022-03-11 | 武汉大学 | Low ionosphere D layer detection system and method based on Loranc very low frequency signal |
| CN114332348A (en) * | 2021-11-16 | 2022-04-12 | 西南交通大学 | Three-dimensional reconstruction method for track integrating laser radar and image data |
| CN114722938A (en) * | 2022-04-06 | 2022-07-08 | 东南大学 | Voronoi density classification reverse reconstruction method for concrete beam point cloud |
| CN115937454A (en) * | 2023-01-09 | 2023-04-07 | 航天宏图信息技术股份有限公司 | Method and device for automatically placing tree models in large-scale urban scene |
| JP7412514B1 (en) | 2022-09-28 | 2024-01-12 | 財団法人車輌研究測試中心 | Cabin monitoring method and cabin monitoring system that implements the above cabin monitoring method |
| CN117523124A (en) * | 2023-11-13 | 2024-02-06 | 西安科技大学 | Large-drainage-basin terrain change slicing method based on unmanned aerial vehicle laser radar |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2133829A1 (en) * | 2008-06-10 | 2009-12-16 | Integrative Biocomputing S.a.r.l. | Simulation of complex systems |
| WO2011070927A1 (en) * | 2009-12-11 | 2011-06-16 | 株式会社トプコン | Point group data processing device, point group data processing method, and point group data processing program |
| CN105046735A (en) * | 2015-07-13 | 2015-11-11 | 长沙理工大学 | Three-dimensional surface texture synthesis method based on primitive element distribution |
| WO2015171355A1 (en) * | 2014-05-08 | 2015-11-12 | Tandent Vision Science, Inc. | A method for identifying color-based vectors for analysis of frames of a video |
| CN106600686A (en) * | 2016-12-06 | 2017-04-26 | 西安电子科技大学 | Three-dimensional point cloud reconstruction method based on multiple uncalibrated images |
| WO2017222558A1 (en) * | 2016-06-24 | 2017-12-28 | Isee, Inc. | Laser-enhanced visual simultaneous localization and mapping (slam) for mobile devices |
| CN107610084A (en) * | 2017-09-30 | 2018-01-19 | 驭势科技（北京）有限公司 | A kind of method and apparatus that information fusion is carried out to depth image and laser spots cloud atlas |
| US20190147245A1 (en) * | 2017-11-14 | 2019-05-16 | Nuro, Inc. | Three-dimensional object detection for autonomous robotic systems using image proposals |
| US20190172249A1 (en) * | 2017-12-01 | 2019-06-06 | Analytical Graphics, Inc. | Systems and Methods for Real-Time Large-Scale Point Cloud Surface Reconstruction |
| CN112001272A (en) * | 2020-08-06 | 2020-11-27 | 苏州富洁智能科技有限公司 | Laser radar environment sensing method and system based on deep learning |
| CN112132972A (en) * | 2020-09-29 | 2020-12-25 | 凌美芯（北京）科技有限责任公司 | Three-dimensional reconstruction method and system for fusing laser and image data |
| CN112418245A (en) * | 2020-11-04 | 2021-02-26 | 武汉大学 | Electromagnetic emission point positioning method based on urban environment physical model |
-
2021
- 2021-03-12 CN CN202110269641.7A patent/CN113066162B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2133829A1 (en) * | 2008-06-10 | 2009-12-16 | Integrative Biocomputing S.a.r.l. | Simulation of complex systems |
| WO2011070927A1 (en) * | 2009-12-11 | 2011-06-16 | 株式会社トプコン | Point group data processing device, point group data processing method, and point group data processing program |
| WO2015171355A1 (en) * | 2014-05-08 | 2015-11-12 | Tandent Vision Science, Inc. | A method for identifying color-based vectors for analysis of frames of a video |
| CN105046735A (en) * | 2015-07-13 | 2015-11-11 | 长沙理工大学 | Three-dimensional surface texture synthesis method based on primitive element distribution |
| WO2017222558A1 (en) * | 2016-06-24 | 2017-12-28 | Isee, Inc. | Laser-enhanced visual simultaneous localization and mapping (slam) for mobile devices |
| CN106600686A (en) * | 2016-12-06 | 2017-04-26 | 西安电子科技大学 | Three-dimensional point cloud reconstruction method based on multiple uncalibrated images |
| CN107610084A (en) * | 2017-09-30 | 2018-01-19 | 驭势科技（北京）有限公司 | A kind of method and apparatus that information fusion is carried out to depth image and laser spots cloud atlas |
| US20190147245A1 (en) * | 2017-11-14 | 2019-05-16 | Nuro, Inc. | Three-dimensional object detection for autonomous robotic systems using image proposals |
| US20190172249A1 (en) * | 2017-12-01 | 2019-06-06 | Analytical Graphics, Inc. | Systems and Methods for Real-Time Large-Scale Point Cloud Surface Reconstruction |
| CN112001272A (en) * | 2020-08-06 | 2020-11-27 | 苏州富洁智能科技有限公司 | Laser radar environment sensing method and system based on deep learning |
| CN112132972A (en) * | 2020-09-29 | 2020-12-25 | 凌美芯（北京）科技有限责任公司 | Three-dimensional reconstruction method and system for fusing laser and image data |
| CN112418245A (en) * | 2020-11-04 | 2021-02-26 | 武汉大学 | Electromagnetic emission point positioning method based on urban environment physical model |
Non-Patent Citations (3)
| Title |
|---|
| CHENG L等: "Semi-automatic registration of airborne and terrestrial laser scanning data using building corner matching with boundaries as reliability check", 《REMOTE SENSING》 * |
| 曾俊飞等: "面向三维重建的自适应列文伯格-马夸尔特点云配准方法", 《计算机科学》 * |
| 李林泽等: "基于地面激光雷达点云的层析合成孔径雷达建筑物三维重构精度评定方法", 《北京测绘》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113532424A (en) * | 2021-08-10 | 2021-10-22 | 广东师大维智信息科技有限公司 | Integrated equipment for acquiring multidimensional information and cooperative measurement method |
| CN113532424B (en) * | 2021-08-10 | 2024-02-20 | 广东师大维智信息科技有限公司 | Integrated equipment for acquiring multidimensional information and cooperative measurement method |
| CN113644997A (en) * | 2021-10-14 | 2021-11-12 | 中国民用航空总局第二研究所 | Electromagnetic environment detection method, device and system |
| CN114332348A (en) * | 2021-11-16 | 2022-04-12 | 西南交通大学 | Three-dimensional reconstruction method for track integrating laser radar and image data |
| CN114332348B (en) * | 2021-11-16 | 2023-06-02 | 西南交通大学 | Track three-dimensional reconstruction method integrating laser radar and image data |
| CN114167505A (en) * | 2021-11-22 | 2022-03-11 | 武汉大学 | Low ionosphere D layer detection system and method based on Loranc very low frequency signal |
| CN114167505B (en) * | 2021-11-22 | 2023-10-27 | 武汉大学 | Low ionosphere D layer detection system and method based on Loran very low frequency signal |
| CN114722938A (en) * | 2022-04-06 | 2022-07-08 | 东南大学 | Voronoi density classification reverse reconstruction method for concrete beam point cloud |
| JP7412514B1 (en) | 2022-09-28 | 2024-01-12 | 財団法人車輌研究測試中心 | Cabin monitoring method and cabin monitoring system that implements the above cabin monitoring method |
| CN115937454A (en) * | 2023-01-09 | 2023-04-07 | 航天宏图信息技术股份有限公司 | Method and device for automatically placing tree models in large-scale urban scene |
| CN117523124A (en) * | 2023-11-13 | 2024-02-06 | 西安科技大学 | Large-drainage-basin terrain change slicing method based on unmanned aerial vehicle laser radar |
| CN117523124B (en) * | 2023-11-13 | 2024-04-26 | 西安科技大学 | Large-drainage-basin terrain change slicing method based on unmanned aerial vehicle laser radar |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113066162B (en) | 2022-04-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113066162B (en) | Urban environment rapid modeling method for electromagnetic calculation | |
| CN110570428B (en) | Method and system for dividing building roof sheet from large-scale image dense matching point cloud | |
| Lohani et al. | Airborne LiDAR technology: A review of data collection and processing systems | |
| Biosca et al. | Unsupervised robust planar segmentation of terrestrial laser scanner point clouds based on fuzzy clustering methods | |
| Gross et al. | Extraction of lines from laser point clouds | |
| Romero-Jarén et al. | Automatic segmentation and classification of BIM elements from point clouds | |
| CN106127857B (en) | The on-board LiDAR data modeling method of integrated data driving and model-driven | |
| WO2018061010A1 (en) | Point cloud transforming in large-scale urban modelling | |
| US20230024326A1 (en) | Using maps comprising covariances in multi-resolution voxels | |
| Previtali et al. | A flexible methodology for outdoor/indoor building reconstruction from occluded point clouds | |
| WO2006027339A2 (en) | Method and system for 3d scene change detection | |
| Ma | Building model reconstruction from LiDAR data and aerial photographs | |
| Alidoost et al. | An image-based technique for 3D building reconstruction using multi-view UAV images | |
| CN112418245A (en) | Electromagnetic emission point positioning method based on urban environment physical model | |
| Özdemir et al. | A multi-purpose benchmark for photogrammetric urban 3D reconstruction in a controlled environment | |
| Zhao et al. | Completing point clouds using structural constraints for large-scale points absence in 3D building reconstruction | |
| CN113066161B (en) | Modeling method of urban radio wave propagation model | |
| Gujski et al. | Machine learning clustering for point clouds optimisation via feature analysis in Cultural Heritage | |
| Sun et al. | Window detection employing a global regularity level set from oblique unmanned aerial vehicle images and point clouds | |
| Cui et al. | A Review of Indoor Automation Modeling Based on Light Detection and Ranging Point Clouds. | |
| Wang et al. | 3D building reconstruction from LiDAR data | |
| Zhu | A pipeline of 3D scene reconstruction from point clouds | |
| Huang et al. | Ground filtering algorithm for mobile LIDAR using order and neighborhood point information | |
| Guo et al. | Research on 3D geometric modeling of urban buildings based on airborne lidar point cloud and image | |
| Khoshelham et al. | Registering point clouds of polyhedral buildings to 2D maps |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |