CN107610223B - Three-dimensional reconstruction method for power tower based on LiDAR point cloud - Google Patents
Three-dimensional reconstruction method for power tower based on LiDAR point cloud Download PDFInfo
- Publication number
- CN107610223B CN107610223B CN201710853206.2A CN201710853206A CN107610223B CN 107610223 B CN107610223 B CN 107610223B CN 201710853206 A CN201710853206 A CN 201710853206A CN 107610223 B CN107610223 B CN 107610223B
- Authority
- CN
- China
- Prior art keywords
- tower
- point
- model
- point cloud
- tower head
- 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.)
- Active
Links
Images
Abstract
The invention belongs to the technical field of laser radar point cloud data information extraction, and relates to a power tower three-dimensional reconstruction method based on LiDAR point cloud. The method comprises the following steps: s1, decomposing the power tower, namely decomposing the power tower into a tower body and a tower head by performing statistical analysis on density and width; s2, reconstructing a tower body based on data drive, extracting and segmenting angle points, and performing three-dimensional straight line fitting on four main contour lines of the tower body based on an RANSAC algorithm; s3, driving and reconstructing a tower head based on a model, predefining a model base containing a basic type of the tower head, then adopting a shape context algorithm to identify the basic type of the tower head, and then combining a Metropolis-Hastings algorithm and simulated annealing to estimate the optimal parameters of the tower head model; and S4, combining the positions and the directions of the results obtained in the steps S2 and S3 to obtain a complete three-dimensional power tower model. The method can effectively and accurately reconstruct the power tower and meet the requirements of three-dimensional visualization and digitization of the power transmission line.
Description
Technical Field
The invention belongs to the technical field of laser radar point cloud data information extraction, and particularly relates to a three-dimensional reconstruction method of an electric power tower based on LiDAR point cloud.
Background
The reconstruction of the power tower as an important infrastructure of the high-voltage transmission line attracts more and more attention in the aspect of three-dimensional visualization of the transmission line. In the past years, the most common method for modeling the power tower is mainly manual modeling by CAD or 3dmax, which consumes a lot of manpower and material resources and has low reconstruction precision. The power grid management department urgently needs an automatic, high-efficiency and high-precision power tower reconstruction method to meet the visual and digital requirements of modern power transmission lines. With the rapid development of airborne LiDAR technology, high density, precise 3D point clouds provide an effective solution to this need.
At present, a lot of researchers do a lot of research on reconstruction of a three-dimensional target based on LiDAR point cloud, and great progress is made in reconstruction of natural ground features and artificial ground features. However, due to the complexity and variety of power tower structures, research on power tower reconstruction is still quite rare. The method comprises the steps of firstly decomposing the power tower into a tower head part, a tower body part and a tower foot part, then reconstructing the tower body part by using four main planes, and classifying the tower head part by using an SVM classifier. The method only decomposes the power tower according to the density characteristics, is lack of applicability, and often generates wrong results when other common types of power towers (such as a cat-head tower, a wine glass tower and a hexagonal tower) are decomposed; in addition, because the method introduces manual establishment of the tower head model, the automation degree is reduced, and the practicability is lacked. Guo Bo et al propose a method for reconstructing a power tower based on random geometry, which combines RJMCMC sampler with simulated annealing to automatically solve the problems of power tower type and model parameters, but the method is not efficient, needs to consume a large number of iteration times to judge the power tower type, and does not consider the geometric relationship among parameters in the process of estimating the power tower model parameters, resulting in that a large number of redundant parameters need to be estimated.
At present, the method for building reconstruction is mainly divided into the following two strategies: data-driven based policies and model-driven based policies. The data-driven strategy adopts a top-down reconstruction method. Most building reconstruction processes generally involve two key steps: extracting the edge of the roof of the building and reconstructing the topological relation. The method for extracting the edge of the complicated roof structure or the high-density point cloud data generally adopts a method based on patch segmentation, such as edge or region growing, three-dimensional RANSAC algorithm, classification or feature clustering and the like. The data-driven based reconstruction results are not affected by the integrity of the model base, and any shape can be reconstructed theoretically. When the data is complete, the data-driven method provides accurate description for a simple target; however, this method may generate reconstruction errors or even fail when data is missing or contains more noise.
Model-driven based strategies employ a bottom-up reconstruction method, which is based on a predefined model library. The method mainly comprises two key steps: and carrying out optimal matching and optimal solution of corresponding model parameters with the models in the model library. In many model-driven based reconstruction methods, the most basic assumption is that a building is a collection of many walls. Many methods such as RJMCMC are introduced to solve the model parameter problem and show great potential for development. The method based on model driving has robustness on the quality of data and is suitable for large scenes, and due to the fact that the topological relation is defined in the model in advance, the method can give play to the superiority of the method when the density of the point cloud is small and the correct topological relation is guaranteed. However, the reconstruction result of the method is influenced by the model library, and the method is low in efficiency when the number of model parameters to be solved is large.
Due to the complexity and the diversity of shapes of the target structures, it is difficult to meet the requirements of modern reconstructions with only a single method. Therefore, hybrid driving strategies combining data driving with model driving have been proposed in recent years. The method introduces construction rules (such as parallelism, coplanarity and symmetry) and the like to optimize the model in the target reconstruction stage. Different from a single strategy, the hybrid driving strategy integrates the advantages of the two methods, so that the flexibility of data driving is reserved, and the model driving robustness is also realized.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a power tower three-dimensional reconstruction method based on LiDAR point cloud, which can effectively and accurately reconstruct a power tower and meet the requirements of three-dimensional visualization and digitization of a power transmission line.
In order to solve the problems, the technical scheme provided by the invention is as follows: a three-dimensional reconstruction method of a power tower based on LiDAR point cloud comprises the following steps:
s1, decomposing an electric power tower, acquiring LiDAR point cloud data of the electric power tower, and decomposing the electric power tower into a tower body and a tower head by performing statistical analysis on density and width of the point cloud of the electric power tower;
s2, reconstructing a tower body based on data drive, extracting and segmenting angle points, and performing three-dimensional straight line fitting on four main contour lines of the tower body based on an RANSAC algorithm;
s3, driving and reconstructing a tower head based on a model, predefining a model base containing a basic type of the tower head, then adopting a shape context algorithm to identify the basic type of the tower head, and then combining a Metropolis-Hastings algorithm and simulated annealing to estimate the optimal parameters of the tower head model;
and S4, combining the positions and the directions of the results obtained in the steps S2 and S3 to obtain a complete three-dimensional power tower model.
In the invention, considering the structural characteristics of the power tower, the power tower is firstly decomposed into a tower body and a tower head by carrying out statistical analysis on the density and the width; then, three-dimensional reconstruction is carried out on the tower body based on data driving, and three-dimensional straight line fitting is carried out on four main contour lines of the tower body based on RANSAC algorithm; the method comprises the steps of conducting three-dimensional reconstruction on a tower head based on model driving, predefining a parameterized tower head model base, identifying the type of the tower head of a tower head point cloud based on the model base by adopting a shape context algorithm, estimating parameters of a corresponding tower head model by adopting a method of combining a Metropolis-Hastings sampler and simulated annealing, and finally combining the positions and the directions of the two parts to obtain a complete three-dimensional power tower model.
The electric power tower is an artificial ground object with a certain building rule, and the tower body of the electric power tower is simple in structure, single in type, multiple in types of tower heads and complex in structure. In view of the structural features of the power tower, the power tower can be broken down into two parts: a tower body and a tower head; although there are many types of power towers, the profiles include two distinct features: (1) the local density is maximum, and (2) the local width is minimum. The density is defined as the point cloud number of the tower section, and the width is defined as the maximum distance from each section to the center of the section; therefore, the invention adopts a statistical analysis method to position the characteristic surface of the power tower. In order to obtain the statistical histogram of the density and the width, the power tower is divided into a plurality of layers, and the layer with the local maximum density and the local minimum width is selected as the position of the feature plane by counting the density and the width of each layer.
Further, the statistical analysis of the density and width of the power tower point cloud in the step S1 includes the following steps:
s101, dividing the power tower point cloud into a plurality of layers according to equal intervals delta h;
s102, calculating the density and width of each layer, and establishing a density histogram and a width histogram by taking the density and width as X axes and the number of layers as Y axes respectively;
s103, defining a moving window which takes the current layer number as the center and is 2L x1 in size, respectively searching a density maximum value and a width minimum value in the density and width histograms, and marking the layers where the maximum value and the minimum value are located;
s104, selecting the first layer number which simultaneously meets the local maximum density of the layer and the local minimum width in the range of the upper layer and the lower layer from bottom to top as the position of the characteristic surface of the power tower, and decomposing by taking the characteristic surface as a boundary line, wherein the part above the characteristic surface is decomposed into a tower head, and the part below the characteristic surface is decomposed into a tower body.
A method based on feature point extraction and segmentation is a commonly used method in data-driven strategies. In order to improve the accuracy of segmentation, the invention only requires to extract angular points; in order to correctly extract the corner points of the tower body, the tower body is first divided into a plurality of layers at equal intervals according to the elevation. Each layer can be viewed as a rectangle with the same center. For each layer, the contour points are extracted by adopting a convex hull algorithm, and not all the contour points are located at the four corners of the rectangle, so that the extracted contour points need to be simplified based on a pipeline algorithm.
Further, the step of extracting and dividing the corner points in the step S2 includes the following steps:
s2011, calculating an angle between the current point and two adjacent points, and if the angle is larger than a preset threshold df; keeping the current point, otherwise, removing;
s2012, simplifying through a pipeline algorithm, and deleting most of non-angular points;
s2013, dividing each simplified corner point of each layer into subsets in four directions according to the position of the simplified corner point at the center of the minimum circumscribed rectangle.
And after the processing of the steps, obtaining a subset of the four tower corner points. Because not all the simplified corner points in each subset are at four corners, the method adopts a RANSAC-based three-dimensional line fitting algorithm to fit each corner point subset.
Further, the RANSAC-based contour line fitting in the step S2 includes the steps of:
s2021, randomly selecting two points P from the subset obtained in the step S11(x1,y1,z1) And P2(x2,y2,z2) Calculating a three-dimensional linear equation L of the three-dimensional linear equation;
s2022, calculating the distance from each point in the subset to the straight line L, if the distance from the current point to the straight line is smaller than a distance threshold Td, determining the point to be a valid point, and if not, defining the point to be an invalid point; and counting the proportion inner _ ratio of the effective points, wherein the inner _ ratio is the proportion of the effective points to the total number of the points;
s2023, repeating random sampling until P is less than a preset confidence probability TP or the sampling times M are greater than a preset maximum sampling time TN, and selecting a group of parameters with the minimum fitting residual as parameters of the subset three-dimensional linear equation; wherein P is the probability of sampling the subset;
and S2024, after the four subsets are all fitted, adding symmetry and coplanarity limiting conditions to optimize the model.
Preferably, said predetermined threshold df is 30 °. In the invention, only the corner points of four corners are needed, so that df is set to be 30 degrees; the larger the df setting, the fewer the contour points, but Tf should be less than 90 °.
Further, the calculation method of the three-dimensional linear equation L is as follows:
further, the probability P of the sampling subset satisfies the following relationship:
P＝1-(1-inner_ratiom)M，
where m is the minimum number of parameters required to construct a criterion.
Further, the step of identifying the basic type of the tower head in S3 includes the following steps:
s3011, converting the tower head point cloud to be identified and all tower head models in a model base into binary images;
s3012, extracting outer contours of all images, and sampling edges of the contours to obtain n sampling point sets P ═ { P ═ P }1，p2，p3，.....pn}；
S3013, calculating the shape context h of each sampling point according to a shape context algorithmi(k) The calculation formula is as formula (1), K is {1,2,. K }, and K is M × N;
hi(k)＝{q≠pi:(q-pi∈bin(k)}， (1)
s3014, calculating a shadow to be matchedThe matching cost between each point of the shape histogram of the image and each point of the shape histogram of each model image is calculated according to the formula (2), wherein h isi(k) Point P being a target PiThe shape histogram of (1); h isj(k) Point Q being target QiIs used to generate a histogram of the shape of (1),
s3015, performing point matching operation based on the cost matrix C obtained through calculation, and selecting a model image with the minimum shape distance as an optimal matching result, wherein the model is a tower head point cloud corresponding model.
Further, the parameter optimization in step S3 includes the following steps:
s3021 Gibbs energy definition Gibbs energy is expressed as (3):
extracting key points of the model and the point cloud by using an alpha shape algorithm, and extracting the distance um(xi) The distance u from each model key point to the nearest tower head point cloud key pointp(xi) The distance from each tower head point cloud key point to the nearest model key point, wherein,
wherein the content of the first and second substances,
degree of similarity u (x) average distance u from model to point cloudm(xi) And the average distance u from the point cloud to the modelp(xi) And u (x) aum(x)+bup(x) A + b is 1, and a and b are two distances, respectivelyThe weight of the separation; n is the number of key points of the model, and m is the number of key points of the tower head point cloud;
s3022 Metropolis-Hastings sampling and simulated annealing, first, according to the proposed distribution q (x)*| X) one candidate parameter value X giving the current parameter value X in the search space X*Then, the acceptance rate A (x, x) of the candidate parameter values is calculated according to the calculation formula (6)*) (ii) a If the acceptance rate A (x, x)*) If the value is larger than the preset acceptance threshold value Ta, the candidate value x is determined*Replacing the current value x, otherwise, keeping the current value x;
further, the acceptance threshold Ta is 0.9+0.1 × random (0,1), and random (0,1) is a result of random sampling performed in a range of 0 to 1.
Metropolis-Hastings (MH) sampling, which is the most popular MCMC sampling method, is widely used to estimate approximations of model parameters. The core idea of MH sampling is to approximate a complex combinatorial problem with a simple problem by statistical sampling. Two distributions are involved in MH sampling: target distribution p (x) and proposed distribution q (x)*| x). First, according to the proposed distribution q (x)*| X) one candidate parameter value X giving the current parameter value X in the search space X*. Then, the acceptance rate A (x, x) of the candidate parameter values is calculated according to the equation (10)*) (ii) a If the acceptance rate A (x, x)*) If the value is larger than the preset acceptance threshold value Ta, the candidate value x is determined*Instead of the current value x, otherwise, the current value x is retained. By means of statistical resampling, the candidate value x*And finally tends to be stable. In order to prevent the sampling from falling into the local optimum, the threshold Ta is 0.9+0.1 × random (0,1), and random (0,1) is a random sampling result in the range of 0 to 1. Although the MH sampling algorithm principle is quite simple, its implementation is difficult. This is mainly due to the choice of the objective function and the recommendation function. Different selection results may result in different sampling results. The invention selects Gaussian distribution as the suggestion distribution and Gibbs energy as the target distribution. Due to the symmetry of the proposed distribution, q (x)*|x)＝q(x|x*) Therefore, the acceptance rate can be simplified as the formula:
MH sampling is less efficient because random sampling is rarely near the true value. It takes more time to search for non-interest areas. To make sampling more efficient, simulated annealing is used in the present invention. The simulated annealing starts from a higher initial temperature, an annealing strategy is established, the target distribution gradually tends to be stable along with the reduction of the temperature, and finally the global optimum is reached with a higher acceptance rate. In simulated annealing, the selection of annealing strategy is an important influencing factor, and the optimal parameters of the model are estimated by simulating a heterogeneous Markov chain. Unlike MH sampling, target distribution
Instead of p (x). Wherein Ti is the temperature at the iteration number i and satisfies limi→∞Ti＝0。
Compared with the prior art, the beneficial effects are: the invention provides a three-dimensional reconstruction method of an electric power tower based on LiDAR point cloud, which fully considers the construction characteristics of the electric power tower and adopts corresponding strategies for different electric power tower parts; in the process of tower head reconstruction, geometric relations among tower body reconstruction information, original point cloud information and parameters are fully utilized to reduce the number of the parameters and the parameter search space as much as possible; the method can effectively and accurately reconstruct the power tower and meet the requirements of three-dimensional visualization and digitization of the power transmission line.
Drawings
FIG. 1 is a schematic diagram of a power tower according to the present invention.
Fig. 2 is a schematic diagram of layering in corner extraction and segmentation according to the present invention.
Fig. 3 is a schematic diagram of corner extraction in the corner extraction and segmentation of the present invention.
Fig. 4 is a simplified schematic diagram of a corner point in the corner point extraction and segmentation of the present invention.
Fig. 5 is a schematic diagram of corner segmentation in the corner extraction and segmentation of the present invention.
Fig. 6 is a schematic diagram of a subset of corner points in the tower reconstruction according to the present invention.
Fig. 7 is a schematic diagram of a corner fitting result in tower reconstruction according to the present invention.
Fig. 8 is a schematic diagram of a corner-point tower optimization result in tower reconstruction according to the present invention.
Fig. 9 is a schematic top view of a corner point optimization result in tower reconstruction according to the present invention.
FIG. 10 is a schematic structural diagram of a parameterized tower head Model library Model1 according to the present invention.
FIG. 11 is a schematic diagram of a parameterized tower head Model library Model 2 according to the present invention.
Fig. 12 is a schematic structural diagram of a parameterized tower head Model library Model 3 according to the present invention.
FIG. 13 is a schematic diagram of a parameterized tower head Model library Model 4 according to the present invention.
FIG. 14 is a schematic diagram of point cloud contour sampling points in the shape context algorithm of the present invention.
FIG. 15 is a schematic diagram of shape context algorithm model contour sampling points according to the present invention.
FIG. 16 is a diagram illustrating a shape context of the shape context algorithm of the present invention.
Wherein, 1 is a tower head, and 2 is a tower body.
Detailed Description
The drawings are for illustrative purposes only and are not to be construed as limiting the patent; for the purpose of better illustrating the embodiments, certain features of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product; it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted. The positional relationships depicted in the drawings are for illustrative purposes only and are not to be construed as limiting the present patent.
A three-dimensional reconstruction method of a power tower based on LiDAR point cloud comprises the following steps:
the method comprises the following steps: and decomposing the power tower, acquiring LiDAR point cloud data of the power tower, and decomposing the power tower into a tower body and a tower head by performing statistical analysis on the density and the width of the point cloud of the power tower.
The electric power tower is the artificial ground thing that has certain construction rule, and its tower body simple structure, type are single, and the tower head is many and the structure is complicated, considers the structural feature of electric power tower, and the electric power tower can be decomposed into two parts: a tower body and a tower head, as shown in figure 1. In order to clearly distinguish the tower head from the tower body, the lower chord of the lower cross arm or the position where the tower section changes sharply is defined as a characteristic surface in the invention, as shown in fig. 1a and 1b, the part above the characteristic surface is defined as the tower head, and the part below the characteristic surface is defined as the tower body. Although there are many types of power towers, the profiles include two distinct features: (1) local density maximum (2) local width minimum. The density is defined as the point cloud number of the tower section, and the width is defined as the maximum distance from each section to the center of the section; accordingly, statistical analysis methods are employed herein for power towers to locate the facets. In order to obtain the statistical histogram of the density and the width, the power tower is divided into a plurality of layers, and the layer with the local maximum density and the local minimum width is selected as the position of the feature plane by counting the density and the width of each layer.
The method comprises the following specific steps:
1. and dividing the power tower point cloud into a plurality of layers at equal intervals delta h according to the elevation. The smaller the value of the delta h is set, the higher the positioning precision of the characteristic surface is, generally delta h/2, but the point of each layer is ensured. In the present invention,. DELTA.h is 0.5 m;
2. calculating the density and width of each layer, and establishing a density histogram and a width histogram by taking the density and width as X axes and the number of layers as Y axes respectively;
3. a moving window of size 2L x1 centered on the current level is defined to find the density maxima and width minima in the density and width histograms, respectively. If the density of the current layer is maximum within 2L 1, the layer is marked. If the width of the current layer is the smallest in the range of 2L x1, the layer is marked. The size 2L of the window is related to the height of the power tower, and L is set to be smaller than the minimum distance between two adjacent cross sections. In the invention, L is 3 m;
4. and selecting the first layer number which simultaneously meets the local maximum density of the layer and has the local minimum width in the range of the upper layer and the lower layer from bottom to top as the position of the feature plane. The decomposition is carried out by taking the characteristic surface as a boundary line, the part above the characteristic surface is decomposed into a tower head, and the part below the characteristic surface is decomposed into a tower body.
Step two: and reconstructing a tower body based on data drive, extracting and segmenting the corner points, and performing three-dimensional straight line fitting on four main contour lines of the tower body based on an RANSAC algorithm.
Firstly, extracting and dividing angular points; as shown in fig. 2 and 3, not all contour points are located at the four corners of the rectangle, and therefore, simplification of the extracted contour points based on the pipeline algorithm is required. And calculating the angle between the current point and the adjacent two points, if the angle is larger than a preset threshold value df, keeping the current point, and otherwise, rejecting. Since only the corner points of the four corners are required in this document, df is set to 30 °. When Tf is set larger, there are fewer contour points, but df should be smaller than 90 °. As shown in fig. 4, most of the non-corner points can be removed by simplification through the pipeline algorithm. Finally, dividing each simplified corner point of each layer according to the position of the simplified corner point at the center of the minimum circumscribed rectangle, and dividing the corner points into subsets in four directions, wherein the final result is shown in fig. 5. Each subset is approximately three-dimensional straight lines in shape.
Secondly, performing contour line fitting based on RANSAC; and after the processing of the steps, obtaining a subset of the four tower corner points. Since not all the simplified corner points in each subset are at the four corners, the RANSAC-based three-dimensional line fitting algorithm is used herein to fit each subset of corner points.
As shown in FIG. 6, to obtain the three-dimensional linear equation for each subset, two points P are randomly selected first1(x1,y1,z1) And P2(x2,y2,z2) And calculating a three-dimensional linear equation L where the three-dimensional linear equation is located. Wherein, a three-dimensional linear equation L of the two points is calculated according to the formula (1); the calculation formula is as follows:
then, for each point in the subset, its distance to the straight line L is calculated. If the distance from the current point to the straight line is less than the distance threshold Td, it is a valid point, otherwise, it is defined as an invalid point. The distance threshold Td is a noise point tolerance level, and the smaller the Td is set, the fewer noise points are involved in calculation, and the Td is set to 0.1 meter in the present invention. The statistical significance accounts for the proportion of the total number of points of the subset, inner _ ratio. And repeating the random sampling until the condition that the P is less than the preset confidence probability TP or the sampling times M are more than the preset maximum sampling times TN is met, and selecting a group of parameters with the minimum fitting residual as the parameters of the subset three-dimensional linear equation, as shown in FIG. 7.
Wherein, under a certain confidence probability TP, the minimum sampling number M of the basic subset in the RANSAC algorithm and the probability P of at least one correct sampling subset satisfy the following relation:
P＝1-(1-inner_ratiom)M，
in the above formula, inner _ ratio is the ratio of the effective points to the total number of points, and m is the minimum number of parameters required for constructing a criterion. The invention sets the confidence probability to be 0.98 and the maximum sampling times to be 30. After the four subsets are fitted independently, the tower body reconstruction result is optimized by using the construction specifications (such as symmetry, coplanarity, parallelism and the like), so that the reconstruction result conforms to the actual structural characteristics, as shown in fig. 8. In addition, the tower body reconstruction result can be used as auxiliary information in the tower head reconstruction process. After determining the location of the feature plane, as shown in FIG. 9, the feature plane length tdx and width tdy are calculated and used as known parameters for tower head reconstruction.
Step three: and reconstructing the tower head based on model driving, predefining a model base containing the basic type of the tower head, then adopting a shape context algorithm to identify the basic type of the tower head, and then combining a Metropolis-Hastings algorithm and simulated annealing to estimate the optimal parameters of the tower head model.
First, a parameterized tower head model library is established.
The tower head has complex structure and various types, and the reconstruction of the tower head is difficult to realize by adopting a data driving strategy. Because the tower head has a certain construction rule and only the main material part needs to be reconstructed, the tower head is reconstructed by adopting a method based on model driving. The model-driven strategy is based on a predefined model library. Referring to the basic type of the high voltage power tower in China, a three-dimensional parameterized model library is first defined. The integrity of the model library directly affects the results of the reconstruction. The model driving method of the invention can only reconstruct the tower head point cloud of which the tower head type is defined in the model base. If the tower head model type is not defined, the type needs to be defined in a model library firstly, and then reconstruction is carried out.
As shown in fig. 10 to 13, in the present invention, the tower head model library includes four power tower types widely used in the chinese 220-and 500-kV high voltage transmission line. For simplicity, the tower head model is represented by key points. In order to obtain the coordinates of the key points, the tower head parameters are mainly composed of feature height and feature length. Due to the complexity of the tower head, the tower head model contains many parameters, resulting in expensive computation costs to estimate the optimal parameters of the model. In order to reduce the number of model parameters and the parameter search space, the invention fully utilizes the geometric relationship among the tower body reconstruction result, the original point cloud information and the parameters. Four types of tower head parameters are shown in table 1. The unknown parameters are tower head parameters to be estimated, and the known parameters are parameters which can be deduced according to information such as geometric relations among the parameters and the like, and do not participate in parameter estimation.
TABLE 1 Tower head model parameters
As can be seen from table 1, the tower head model contains four basic parameters: tdx, tdy, Height, Length. tdx and tdy are the length and width of the tip face, respectively, and can be derived from the tower reconstruction results. Height and Length are the tower head size, respectively, and can be derived from the Length and width of the original tower head point cloud. Other parameters may be derived from geometric relationships with other parameters. Take Model1 as an example, H3、L3Can be obtained from the formula (2) and the formula (3), respectively.
H3＝Height-H1-H2； (2)
In addition, the parameter search space can be greatly reduced according to the geometric relationship. Thus, the efficiency of tower body reconstruction can be greatly improved.
Secondly, the tower head type is identified.
The tower heads of different types have quite large differences in shape, so that the type of the tower head can be identified according to the shape of the tower head. In the invention, the tower head point cloud and the tower head model are firstly converted into binary images, and then the matching of the tower head point cloud image and the corresponding model image is realized by adopting a shape context algorithm.
The shape context is a feature descriptor that measures shape similarity and finds point correspondences. It is widely used in digital recognition, similarity-based retrieval, trademark retrieval, three-dimensional object recognition and the like. As shown in fig. 14 and 15, the most basic idea of the shape context algorithm is to select n sampling points on the contour and then calculate the shape context of each point. Wherein the shape context of each point is defined as a histogram of the remaining n-1 points versus the relative coordinates of the current point, with the current point piN concentric circles are established at intervals of logarithmic distance in a local area with the circle center as the center and the radius as the radius. This region was divided equally in the circumferential direction M to form a target-like template as shown in FIG. 16. The relative position of the vector from the current point to other points is simplified into the distribution number of points in each sector on the template. Histogram h of statistical distribution of these pointsi(k) Referred to as point piThe shape context of (1). And finally, selecting the type of the model image with the minimum cost as the type of the tower head point cloud by calculating the shape distance between each pair of the image to be matched and each pair of the model image.
The method comprises the following specific steps:
1. converting the tower head point cloud to be identified and all tower head models in the model base into binary images;
2. extracting the outer contour of all images, and sampling the edge of the outer contour to obtain n sampling point sets P ═ P1,p2,p3,...pn}；
3. Calculating the shape context h of each sampling point according to a shape context algorithmi(k) Formula (4), K ═ {1, 2.. K }, K ═ M × N;
hi(k)＝{q≠pi:(q-pi∈bin(k)}， (4)
4. calculating the matching cost between each pair of points of the shape histogram of the image to be matched and the shape histogram of each model image, wherein the formula is shown as the formula (5), wherein h isi(k) Point P being a target PiThe shape histogram of (1); h isj(k) Point Q being target QiThe shape histogram of (1);
5. and performing point matching operation based on the cost matrix C obtained by calculation, and selecting a model image with the minimum shape distance as an optimal matching result, wherein the model is a tower head point cloud corresponding model.
Finally, the invention converts the tower head reconstruction process into a Gibbs energy optimization solution, the Gibbs energy is defined as the similarity between the tower head point cloud and the model, and the Metropolis-Hastings sampler and the simulated annealing algorithm are used for estimating the optimization parameters of the model.
Step four: and combining according to the positions and the directions in the results obtained in the first step, the second step and the third step to obtain a complete three-dimensional power tower model.
It should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.
Claims (8)
1. A three-dimensional reconstruction method of a power tower based on LiDAR point cloud is characterized by comprising the following steps:
s1, decomposing an electric power tower, acquiring LiDAR point cloud data of the electric power tower, and decomposing the electric power tower into a tower body and a tower head by performing statistical analysis on density and width of the point cloud of the electric power tower;
s2, reconstructing a tower body based on data drive, extracting and segmenting angle points, and performing three-dimensional straight line fitting on four main contour lines of the tower body based on an RANSAC algorithm;
s3, driving and reconstructing a tower head based on a model, predefining a model base containing a basic type of the tower head, then adopting a shape context algorithm to identify the basic type of the tower head, and then combining a Metropolis-Hastings algorithm and simulated annealing to estimate the optimal parameters of the tower head model;
s4, combining according to the positions and directions in the results obtained in the S2 and S3 steps to obtain a complete three-dimensional power tower model; wherein the content of the first and second substances,
the step of identifying the basic type of the tower head in the step of S3 includes the following steps:
s3011, converting the tower head point cloud to be identified and all tower head models in a model base into binary images;
s3012, extracting outer contours of all images, and sampling edges of the contours to obtain n sampling point sets P ═ { P ═ P }1，p2，p3，.....pn}；
S3013, calculating the shape context h of each sampling point according to a shape context algorithmi(k) The calculation formula is as formula (1), K is {1,2,. K }, and K is M × N; m represents the sampling times;
hi(k)＝{q≠pi:(q-pi∈bin(k)}， (1)
s3014, calculating the matching cost between each point of the shape histogram of the image to be matched and each point of the shape histogram of each model image, wherein the calculation formula is shown as a formula (2), and hi (k) is the shape histogram of the point pi of the target P; hj (k) is the shape histogram of the point qi of the target Q,
s3015, performing point matching operation based on the cost matrix C obtained through calculation, and selecting a model image with the minimum shape distance as an optimal matching result, wherein the model is a corresponding model of the tower head point cloud;
the parameter optimization in the step S3 includes the following steps:
s3021 Gibbs energy definition Gibbs energy is expressed as:
extracting key points of the model and the point cloud by using an alpha shape algorithm, and extracting the distance um(xi) The distance u from each model key point to the nearest tower head point cloud key pointp(xi) And (3) the distance from each tower head point cloud key point to the nearest model key point, wherein:
wherein the content of the first and second substances,
degree of similarity u (x) average distance u from model to point cloudm(xi) And the average distance u from the point cloud to the modelp(xi) And u (x) aum(x)+bup(x) A + b is 1, and a and b are weights of two distances respectively; n is the number of key points of the model, and m is the number of key points of the tower head point cloud;
s3022 Metropolis-Hastings sampling and simulated annealing, first, according to the proposed distribution q (x)*| X) one candidate parameter value X giving the current parameter value X in the search space X*Then, the acceptance rate of the candidate parameter values is calculated according to the calculation formula (6)A(x,x*) (ii) a If the acceptance rate A (x, x)*) If the value is larger than the preset acceptance threshold value Ta, the candidate value x is determined*Replacing the current value x, otherwise, keeping the current value x;
2. the method according to claim 1, wherein the step of S1 for performing statistical analysis on the density and width of the power tower point cloud comprises the steps of:
s101, dividing the power tower point cloud into a plurality of layers according to equal intervals delta h;
s102, calculating the density and width of each layer, and establishing a density histogram and a width histogram by taking the density and width as X axes and the number of layers as Y axes respectively;
s103, defining a moving window which takes the current layer number as the center and is 2L x1 in size, respectively searching a density maximum value and a width minimum value in the density and width histograms, and marking the layers where the maximum value and the minimum value are located;
s104, selecting the first layer number which simultaneously meets the local maximum density of the layer and the local minimum width in the range of the upper layer and the lower layer from bottom to top as the position of the characteristic surface of the power tower, and decomposing by taking the characteristic surface as a boundary line, wherein the part above the characteristic surface is decomposed into a tower head, and the part below the characteristic surface is decomposed into a tower body.
3. The method of claim 1, wherein the step of S2 for extracting and segmenting corners comprises the steps of:
s2011, calculating an angle between the current point and two adjacent points, and if the angle is larger than a preset threshold df; keeping the current point, otherwise, removing;
s2012, simplifying through a pipeline algorithm, and deleting most of non-angular points;
s2013, dividing each simplified corner point of each layer into subsets in four directions according to the position of the simplified corner point at the center of the minimum circumscribed rectangle.
4. The method of claim 3, wherein the RANSAC-based contour line fitting in step S2 comprises the following steps:
s2021, randomly selecting two points P1(x1, y1, z1) and P2(x2, y2, z2) from the subset obtained in the step S1, and calculating a three-dimensional linear equation L where the two points are located;
s2022, calculating the distance from each point in the subset to the straight line L, if the distance from the current point to the straight line is smaller than a distance threshold Td, determining the point to be a valid point, and if not, defining the point to be an invalid point; and counting the proportion inner _ ratio of the effective points, wherein the inner _ ratio is the proportion of the effective points to the total number of the points;
s2023, repeating random sampling until P is less than a preset confidence probability TP or the sampling times M are greater than a preset maximum sampling time TN, and selecting a group of parameters with the minimum fitting residual as parameters of the subset three-dimensional linear equation; wherein P is the probability of sampling the subset;
and S2024, after the four subsets are all fitted, adding symmetry and coplanarity limiting conditions to optimize the model.
5. The method for three-dimensional reconstruction of power towers based on LiDAR point clouds of claim 3, wherein the predetermined threshold df is 30 °.
6. The method of claim 4, wherein the three-dimensional equation L is calculated by:
7. the method according to claim 4, wherein the probabilities P of the sampled subsets satisfy the following relationship:
P＝1-(1-inner_ratiom)M，
where m is the minimum number of parameters required to construct a criterion.
8. The method according to claim 7, wherein the acceptance threshold Ta is 0.9+0.1 random (0,1), and random (0,1) is a random sampling result in the range of 0-1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710853206.2A CN107610223B (en) | 2017-09-20 | 2017-09-20 | Three-dimensional reconstruction method for power tower based on LiDAR point cloud |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710853206.2A CN107610223B (en) | 2017-09-20 | 2017-09-20 | Three-dimensional reconstruction method for power tower based on LiDAR point cloud |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107610223A CN107610223A (en) | 2018-01-19 |
| CN107610223B true CN107610223B (en) | 2020-09-25 |
Family
ID=61061715
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710853206.2A Active CN107610223B (en) | 2017-09-20 | 2017-09-20 | Three-dimensional reconstruction method for power tower based on LiDAR point cloud |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107610223B (en) |
Families Citing this family (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108830933B (en) * | 2018-06-26 | 2022-04-19 | 广东电网有限责任公司 | Method, system, medium and equipment for rebuilding tower body of electric tower |
| CN109341667B (en) * | 2018-11-29 | 2020-12-18 | 武汉汇卓航科技有限公司 | Three-dimensional power line measuring method based on region judgment selection |
| CN109872384B (en) * | 2018-12-29 | 2021-03-09 | 中国科学院遥感与数字地球研究所 | Automatic tower modeling method based on airborne LIDAR point cloud data |
| CN111739164A (en) * | 2019-03-19 | 2020-10-02 | 北京京东尚科信息技术有限公司 | Box body modeling method and device, robot picking system, electronic device and medium |
| CN110136260A (en) * | 2019-04-24 | 2019-08-16 | 广州供电局有限公司 | Electric power line pole tower and its implementation based on laser point cloud full feature model library |
| CN110766750B (en) * | 2019-09-09 | 2022-04-05 | 浙江大学 | Mechanical arm visual servo target identification method based on local contour |
| CN111126211B (en) * | 2019-12-13 | 2023-08-29 | 北京四维图新科技股份有限公司 | Label identification method and device and electronic equipment |
| CN112215867A (en) * | 2020-10-23 | 2021-01-12 | 劢微机器人科技(深圳)有限公司 | Point cloud data segmentation method, point cloud data segmentation equipment, storage medium and device |
| CN112698303A (en) * | 2020-12-23 | 2021-04-23 | 国网电力科学研究院武汉南瑞有限责任公司 | Method and system for measuring point cloud tower inclination parameters based on unmanned aerial vehicle laser radar |
| CN112884823B (en) * | 2021-01-13 | 2023-09-22 | 上海建工四建集团有限公司 | Guiding rule simulation point cloud selection and calculation method for actual measurement actual quantity application |
| CN112767352B (en) * | 2021-01-19 | 2023-07-07 | 国网电力空间技术有限公司 | Automatic refined extraction method for high-voltage line tower for airborne LiDAR power inspection |
| CN113076870A (en) * | 2021-03-31 | 2021-07-06 | 广东电网有限责任公司 | Tower identification method and system based on laser point cloud |
| CN113345089B (en) * | 2021-05-31 | 2023-06-23 | 西北农林科技大学 | Regularized modeling method based on power tower point cloud |
| CN113554595B (en) * | 2021-06-21 | 2022-11-25 | 国网山西省电力公司电力科学研究院 | Unmanned aerial vehicle laser radar point cloud tower head deformation detection device and method |
| CN114418194B (en) * | 2021-12-29 | 2022-10-11 | 广东电网有限责任公司 | Tower damage prediction method and device based on data driving and model driving |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104866840A (en) * | 2015-06-04 | 2015-08-26 | 广东中城规划设计有限公司 | Method for recognizing overhead power transmission line from airborne laser point cloud data |
| CN106157361A (en) * | 2016-05-31 | 2016-11-23 | 中国科学院遥感与数字地球研究所 | A kind of multiple fission conductor full-automatic three-dimensional method for reconstructing based on LiDAR point cloud |
| CN106683132A (en) * | 2015-11-10 | 2017-05-17 | 星际空间（天津）科技发展有限公司 | High-precision three-dimensional city modeling method |
| CN107102339A (en) * | 2017-04-28 | 2017-08-29 | 上海华测导航技术股份有限公司 | A kind of automatic generation method of above-ground route data |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104063894B (en) * | 2014-06-13 | 2017-02-22 | 中国科学院深圳先进技术研究院 | Point cloud three-dimensional model reestablishing method and system |
-
2017
- 2017-09-20 CN CN201710853206.2A patent/CN107610223B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104866840A (en) * | 2015-06-04 | 2015-08-26 | 广东中城规划设计有限公司 | Method for recognizing overhead power transmission line from airborne laser point cloud data |
| CN106683132A (en) * | 2015-11-10 | 2017-05-17 | 星际空间（天津）科技发展有限公司 | High-precision three-dimensional city modeling method |
| CN106157361A (en) * | 2016-05-31 | 2016-11-23 | 中国科学院遥感与数字地球研究所 | A kind of multiple fission conductor full-automatic three-dimensional method for reconstructing based on LiDAR point cloud |
| CN107102339A (en) * | 2017-04-28 | 2017-08-29 | 上海华测导航技术股份有限公司 | A kind of automatic generation method of above-ground route data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107610223A (en) | 2018-01-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107610223B (en) | Three-dimensional reconstruction method for power tower based on LiDAR point cloud | |
| CN110120097B (en) | Semantic modeling method for airborne point cloud of large scene | |
| CN106970375B (en) | Method for automatically extracting building information from airborne laser radar point cloud | |
| CN103413133B (en) | Automatically-extracting power line method in random laser point cloud data | |
| CN113487730B (en) | Urban three-dimensional automatic modeling method based on laser radar point cloud data | |
| CN113379898B (en) | Three-dimensional indoor scene reconstruction method based on semantic segmentation | |
| CN107330901B (en) | Skeleton-based object component decomposition method | |
| CN109299301B (en) | Three-dimensional model retrieval method based on shape distribution and curvature | |
| CN112396641A (en) | Point cloud global registration method based on congruent two-baseline matching | |
| Niemeyer et al. | Contextual classification of point clouds using a two-stage CRF | |
| Wu et al. | Building reconstruction from high-resolution multiview aerial imagery | |
| He | Automated 3D building modelling from airborne LiDAR data | |
| Li | A super voxel-based Riemannian graph for multi scale segmentation of LiDAR point clouds | |
| Baloch et al. | Object recognition through topo-geometric shape models using error-tolerant subgraph isomorphisms | |
| CN116152494A (en) | Building foot point identification segmentation method based on two-stage 3D point cloud semantic segmentation | |
| CN114463396B (en) | Point cloud registration method utilizing plane shape and topological graph voting | |
| CN114565730A (en) | Method and device for processing map vector sample and computer readable storage medium | |
| Sayed et al. | Point clouds reduction model based on 3D feature extraction | |
| CN113850254A (en) | Building vector outline simplifying method, model and model establishing method based on deep learning | |
| CN103336781A (en) | Medical image clustering method | |
| Lu et al. | A surface division method of parts mesh model for on-machine inspection | |
| Sun et al. | Research on target classification method for dense matching point cloud based on improved random forest algorithm | |
| CN117171856B (en) | Highway railway digital information modeling method based on handheld Lidar | |
| Zhao et al. | a Review of Point Cloud Segmentation of Architectural Cultural Heritage | |
| Liao et al. | Robust and Effective Airborne LIDAR Point Cloud Classification Based on Hybrid Features |
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 |