Transformer substation modeling method based on three-dimensional scanning
阅读说明：本技术 一种基于三维扫描的变电站建模方法 (Transformer substation modeling method based on three-dimensional scanning ) 是由 孙蓉 郑家滨 崔林 唐锦 邵剑 赵新东 张力 卢茜 于 2020-06-02 设计创作，主要内容包括：本发明涉及3D建模领域,特别是一种基于三维扫描的变电站建模方法,过程如下：利用三维扫描仪扫描得到变电站的点云数据后,首先进行数据的预处理,即自动去除变电站地面和设备上的电力线的点云数据,然后进行点云聚类,将设备的点云数据独立化,接着通过数据拼接、数据去噪、数据简化和特征提取建立点云模型,在此基础上,通过采用基于二维平面包围盒拓扑结构的重复结构检测的算法,分别电站场景重复结构模式,最后,采用基于组件匹配的建模方法对最小重复单元进行建模拟合,并与重复结构模式相结合,从而得到变电站模型,使用本发明方法建模时,点云数据的处理速度快,精度高,有效的提高了电力系统虚拟现实建模的快速性和有效性。(The invention relates to the field of 3D modeling, in particular to a transformer substation modeling method based on three-dimensional scanning, which comprises the following steps: after the point cloud data of the transformer substation is obtained by scanning with the three-dimensional scanner, the data is preprocessed firstly, namely, automatically removing the point cloud data of the power lines on the ground of the transformer substation and the equipment, then carrying out point cloud clustering, and making the point cloud data of the equipment independent, then establishing a point cloud model through data splicing, data denoising, data simplification and feature extraction, on the basis, by adopting an algorithm of repetitive structure detection based on a two-dimensional plane bounding box topological structure, power station scene repetitive structure modes are respectively determined, and finally, modeling and fitting the minimum repeating unit by adopting a modeling method based on component matching, combining with a repeating structure mode, when the method is used for modeling, the processing speed of the point cloud data is high, the precision is high, and the rapidity and the effectiveness of virtual reality modeling of the power system are effectively improved.)
1. A transformer substation modeling method based on three-dimensional scanning is characterized by comprising the following steps:
s10: the method comprises the steps that a plurality of measuring stations and reflection reference points are arranged on a transformer substation, and the measuring stations are scanned by using a three-dimensional laser scanner to obtain geometric position information and point cloud data of the transformer substation;
s20: performing point cloud segmentation and point cloud clustering on the obtained point cloud data, wherein the point cloud segmentation comprises ground segmentation and power line segmentation, the ground segmentation is the point cloud data from which the ground of the transformer substation is removed, the power line segmentation is the extraction by using height filtering and space density filtering, and the point cloud clustering is the blocking operation on the whole point cloud data to obtain the point cloud data of each equipment monomer;
s30: sequentially carrying out data splicing, data denoising, data simplification and feature extraction on the point cloud data of each block to obtain a point cloud model of the point cloud data of each block;
s40: extracting minimum repeating units from all point cloud models, and then matching the models to obtain a three-dimensional model of each minimum repeating unit, wherein the extraction of the minimum repeating units adopts a point cloud repeating structure detection algorithm based on a two-dimensional plane bounding box topological structure, and the matching of the models is to perform shape retrieval and template fitting on the point cloud models of the minimum repeating units and a model in a template library;
s50: and (4) reconstructing the three-dimensional model of the minimum repeating unit according to the detected repeating mode, so as to quickly obtain the model of the whole transformer substation.
2. A modeling method in accordance with claim 1, wherein: the point cloud data between two adjacent measuring stations has 30% -40% of overlapping degree, and the acquisition precision of the point cloud data is as follows: the coordinate linearity error of the point cloud data is not more than +/-1 mm, the registration error of the point cloud data is not more than +/-2 mm, and the three-dimensional coordinate of the point cloud data is accurate to 1 mm.
3. A modeling method in accordance with claim 1, wherein: the data splicing comprises the following steps:
(1) uniformly selecting 3 or more homonymous points in the overlapped area of the point cloud data of two adjacent sites;
(2) the overlapping areas are stitched using the ICP algorithm.
4. A modeling method in accordance with claim 3, wherein: the data denoising comprises the steps of judging noise points and eliminating the noise points, wherein the judging of the noise points adopts one or more of a direct inspection method, a curve inspection method, a chord height difference method and a value limiting method;
the direct inspection method comprises the following steps: directly observing point cloud data, and judging points which are isolated or far away from the point set as noise points;
the curve checking method comprises the following steps:
(1) setting a threshold value epsilon;
(2) fitting a curve passing through head and tail data points of the section by a least square method;
(3) selecting data points as UiCalculating the data point UiEuclidean distance to curve ei；
(4) If eiIf | is greater than or equal to ε, then UiThe points are noise points;
the string height difference method comprises the following steps:
(1) a set threshold value xi;
(2) select data point QiConnecting data points QiObtaining a chord from the front point and the rear point;
(3) calculating QiDistance f to the chordi；
(4) If fiIf | is not less than xi, then QiThe points are noise points;
the value limiting method comprises the following steps:
(1) minimum and maximum distance values between data points in the point cloud data, and a limit value D is set accordinglyminAnd Dmax；
(2) Selecting the distance O between any two data points of the point cloud datai；
(3) If O is presenti>DmaxThen O isi＝DmaxIf O is presenti<Dmin，Oi＝Dmin。
5. The modeling method of claim 4, wherein: the data reduction comprises the following steps:
(1) any data point Y in the cloud data is taken, and N data points closest to the data point Y are obtained through a kd-tree algorithm;
(2) fitting a least square surface through the N points with the minimum distance;
(3) setting a threshold value D0Calculating the actual distance D from other data points in the point cloud data to the least square surfacei；
(4) Determining the actual distance DiWhether or not it is greater than a threshold value D0If yes, removing the data point and repeating the step, and if not, finishing the step.
6. A modeling method in accordance with claim 1, wherein: the extraction of the point cloud data features comprises the following steps:
(1) calculating the main shaft direction of each part of the point cloud data;
(2) acquiring a section contour line of each part of the point cloud data from the main shaft direction;
(3) acquiring characteristic points of the cross section by the cross section contour line;
(4) and fitting and outputting a characteristic curve by using the characteristic line points.
7. The modeling method of claim 6, wherein: the main shaft direction acquisition comprises the following steps:
(1) selecting any point T (x, y, z) in the point cloud data, acquiring M data points closest to the point T (x, y, z), and establishing a least square surface of the data point T by using the M data points;
(2) solving a normal vector of a least square surface;
(3) taking a point domain H near the T point, wherein the point domain H contains K data points, and performing the steps (1) - (2) on all points in the point domain H to obtain the normal vector of the least square surface where all the points are located;
(5) arbitrarily taking two vector normal vectors, and calculating the vector product of the two vector normal vectors to obtain a principal axis vector l1、l2....lRAnd calculating the average value l of all the vector productsFlat plateWherein
(6) Setting an included angle threshold alpha, traversing all the principal axis vectors, and selecting any vector liJudging the principal axis vector liAnd lFlat plateIf the included angle is greater than the included angle threshold value alpha, removing the selected principal axis vector liIf not, retaining the selected main axis vector;
(7) taking the average value l of all the remaining principal axis vectorsiObtaining the vector l of the main axis direction of the part of point cloudT1；
(8) And repeating the steps to obtain vectors of other main shaft directions of the point cloud data.
8. The modeling method of claim 7, wherein: the acquisition of the section contour line comprises the following steps:
(1) taking a point set W near a point T, and setting a threshold value V;
(2) take any point E (x) in the set of pointse，ye，ze) Calculating the principal axis vector l of the point cloud data of the part where the vector ET and the point T are locatedT1The dot product of (a) is the distance from the point E to the cross section
(3) Traversing all data points in the point set W, repeating the step (2), and obtaining the distance from all points in the point set W to the section;
(4) judging whether the distance is smaller than a threshold value V, if so, selecting the data point, and if not, discarding;
(5) and (4) reading the points screened in the step (3), and fitting to obtain a corresponding section contour line.
9. The modeling method of claim 8, wherein: the acquisition of the characteristic points of the section comprises the following steps:
(1) randomly taking three data points from a point cloud data set S in the obtained section contour line, establishing a plane through the three points, and solving a normal vector of the plane;
(2) setting the data point set S of the section to contain X data points, continuously extracting other data points in the point cloud data set S of the section, and repeating the operation of the first step until all the data points are extracted;
(3) carrying out mean value operation on the obtained normal vectors of the plurality of planes to obtain an equation of the plane where the section is located;
(4) taking any two non-parallel unit vectors a (i) in the cross-section planea，ja，ka)，b(ib，jb，kb) Calculating the projection lengths of all data points in the cross-section point cloud data set S projected on two unit vectors, and respectively summing to obtain daAnd db，daIs the sum of the projected distances of all data points in the unit vector a, dbIs the sum of the projected distances of all data points in the unit vector b;
(5) and (3) calculating the coordinates of the centroid Z of the section on two vectors, wherein the calculation formula is as follows:
U＝da/X，V＝dbthe coordinate value of the cross section centroid Z on the unit vector a is U, and the coordinate value of the cross section centroid Z on the unit vector b is V;
(6) and (3) calculating the coordinate of the section centroid Z (p, q, r) in an xyz coordinate system, wherein the formula is as follows:
p＝U*ia+V*ib，
q＝U*ja+V*jb，
r＝U*ka+V*kb。
10. a modeling method in accordance with claim 1, wherein: the steps of carrying out shape retrieval and template fitting on the point cloud model and the model in the template library are as follows:
(1) respectively counting the poor gray scale of each data point and the template library in the detection point cloud model, and using a set D ═ D1，d2，d3，...，dnStore, diIndicating that the i-th pixel and the kernel pixel have poor gray levels;
(2) sorting the elements of the set D from small to large, and then splitting D into a head class and a tail class which are expressed as D1＝{d1，d2，d3，...，dqAnd D2＝{dq+1，dq+2，dq+3，...，dnAnd calculating the inter-class variance of two gray level difference classesThe formula is as follows:
wherein the content of the first and second substances,
(3) when the inter-class variance takes a maximum value, i.e.When the maximum value is obtained, the optimal gray level difference threshold value is judged to be obtained, namely, the optimal gray level threshold value q is obtained*When, satisfy the following formula:
(4) obtaining point cloud dataBest gray segmentation threshold q for poor gray of single point and template library*Then, the point cloud data is divided into two types, and when the gray value is poorer than q*If so, the point cloud data is considered to be optimally matched with the template base, and the part of point cloud data is reserved; when the gray value is worse than q*And discarding the part of point cloud data to finally obtain the point cloud set with the optimal matching degree with the template library.
Technical Field
The invention relates to the field of 3D modeling, in particular to a transformer substation modeling method based on three-dimensional scanning.
Background
In recent years, with the development of smart devices and digitalization concepts, applications such as virtual reality technology, smart cities, and digital factories are actively advancing social operations and progress. The accurate three-dimensional digital model provides a vivid expression mode for real scenes, equipment facilities and surrounding environments. Traditional modeling techniques create object surface shapes using manual markers by using specialized software, and the enormous amount of data results in inefficiency and labor intensity. With the development of science and technology, digital photogrammetry technology is widely applied to scene reconstruction
The three-dimensional laser scanning technology is a measurement technology for instantaneously measuring a spatial three-dimensional coordinate value by using a laser ranging principle (including pulse laser and phase laser) through a high-speed laser transmitter, and enables an acquisition method, service capability and level of mapping data, a data processing method and the like to enter a new development stage. The three-dimensional laser scanning is a data acquisition mode based on a surface, and data measured each time not only contains information of X, Y and Z points, but also comprises R, G and B color information and information of object reflectivity, so that comprehensive information can give people a feeling that an object really reproduces in a computer.
Under the background of intelligent power grid construction, the data visualization, integration, intellectualization and auxiliary analysis capability represent important characteristics and development directions of a novel management system. With the continuous development of the intelligent equipment concept and the continuous improvement of the intelligent power grid theory, the virtual reality technology is deeply applied to various aspects of power grid operation, and at present, some obstacles still exist in the establishment of a three-dimensional power station model, such as large workload, low accuracy and long data processing period. Therefore, how to realize the overall modeling of the substation in a fast and accurate manner is still an interval task
Disclosure of Invention
The invention aims to provide a substation modeling method based on three-dimensional scanning, and aims to solve the problem of how to realize overall modeling of a substation quickly and accurately.
In order to achieve the technical purpose and achieve the technical effect, the invention discloses a transformer substation modeling method based on three-dimensional scanning, which comprises the following steps:
s10: the method comprises the steps that a plurality of measuring stations and reflection reference points are arranged on a transformer substation entity, and the measuring stations are scanned by using a three-dimensional laser scanner to obtain geometric position information and point cloud data of the transformer substation;
s20: performing point cloud segmentation and point cloud clustering on the obtained point cloud data, wherein the point cloud segmentation comprises ground segmentation and power line segmentation, the ground segmentation is the point cloud data from which the ground of the transformer substation is removed, the power line segmentation is the extraction by using height filtering and space density filtering, and the point cloud clustering is the blocking operation on the whole point cloud data to obtain the point cloud data of each equipment monomer;
s30: sequentially carrying out data splicing, data denoising, data simplification and feature extraction on the point cloud data of each block to obtain a point cloud model of the point cloud data of each block;
s40: extracting minimum repeating units from all point cloud models, and then matching the models to obtain a three-dimensional model of each minimum repeating unit, wherein the extraction of the minimum repeating units adopts a point cloud repeating structure detection algorithm based on a two-dimensional plane bounding box topological structure, and the matching of the models is to perform shape retrieval and template fitting on the point cloud models of the minimum repeating units and a model in a template library;
s50: and (4) reconstructing the three-dimensional model of the minimum repeating unit according to the detected repeating mode, so as to quickly obtain the model of the whole transformer substation.
Further, the point cloud data between two adjacent measurement stations has an overlap degree of 30% -40%, and the acquisition precision of the point cloud data is as follows: the coordinate linearity error of the point cloud data is not more than +/-1 mm, the registration error of the point cloud data is not more than +/-2 mm, and the three-dimensional coordinate of the point cloud data is accurate to 1 mm.
Further, the data splicing comprises the following steps:
(1) uniformly selecting 3 or more homonymous points in the overlapped area of the point cloud data of two adjacent sites;
(2) the overlapping areas are stitched using the ICP algorithm.
Further, the data denoising comprises a noise point judgment step and a noise point elimination step, wherein the noise point judgment step adopts one or more of a direct inspection method, a curve inspection method, a chord height difference method and a value limiting method;
the direct inspection method comprises the following steps: directly observing point cloud data, and judging points which are isolated or far away from the point set as noise points;
the curve checking method comprises the following steps:
(1) setting a threshold value epsilon;
(2) fitting a curve passing through head and tail data points of the section by a least square method;
(3) selecting data points as UiCalculating the data point UiEuclidean distance to curve ei；
(4) If ei||Greater than or equal to epsilon, then UiThe points are noise points;
the string height difference method comprises the following steps:
(1) a set threshold value xi;
(2) select data point QiConnecting data points QiObtaining a chord from the front point and the rear point;
(3) calculating QiDistance f to the chordi；
(4) If fiIf | is not less than xi, then QiThe points are noise points;
the value limiting method comprises the following steps:
(1) minimum and maximum distance values between data points in the point cloud data, and a limit value D is set accordinglyminAnd Dmax；
(2) Selecting the distance O between any two data points of the point cloud datai；
(3) If O is presenti>DmaxThen O isi＝DmaxIf O is presenti<Dmin，Oi＝Dmin。
Further, the data reduction comprises the following steps:
(1) any data point Y in the cloud data is taken, and N data points closest to the data point Y are obtained through a kd-tree algorithm;
(2) fitting a least square surface through the N points with the minimum distance;
(3) setting a threshold value D0Calculating the actual distance D from other data points in the point cloud data to the least square surfacei；
(4) Determining the actual distance DiWhether or not it is greater than a threshold value D0If yes, removing the data point and repeating the step, and if not, finishing the step.
Further, the extraction of the point cloud data features comprises the following steps:
(1) calculating the main shaft direction of each part of the point cloud data;
(2) acquiring a section contour line of each part of the point cloud data from the main shaft direction;
(3) acquiring characteristic points of the cross section by the cross section contour line;
(4) and fitting and outputting a characteristic curve by using the characteristic line points.
Further, the obtaining of the main shaft direction comprises the following steps:
(1) selecting any point T (x, y, z) in the point cloud data, acquiring M data points closest to the point T (x, y, z), and establishing a least square surface of the data point T by using the M data points;
(2) solving a normal vector of a least square surface;
(3) taking a point domain H near the T point, wherein the point domain H contains K data points, and performing the steps (1) - (2) on all points in the point domain H to obtain the normal vector of the least square surface where all the points are located;
(5) arbitrarily taking two vector normal vectors, and calculating the vector product of the two vector normal vectors to obtain a principal axis vector l1、l2....lRAnd calculating the average value l of all the vector productsFlat plateWherein
(6) Setting an included angle threshold alpha, traversing all the principal axis vectors, and selecting any vector liJudging the principal axis vector liAnd lFlat plateIf the included angle is greater than the included angle threshold value alpha, removing the selected principal axis vector liIf not, retaining the selected main axis vector;
(7) taking the average value l of all the remaining principal axis vectorsiObtaining the vector l of the main axis direction of the part of point cloudT1；
(8) And repeating the steps to obtain vectors of other main shaft directions of the point cloud data.
Further, the obtaining of the cross-section contour line comprises the following steps:
(1) taking a point set W near a point T, and setting a threshold value V;
(2) take any point E (x) in the set of pointse，ye，ze) Calculating the principal axis vector l of the point cloud data of the part where the vector ET and the point T are locatedT1Is multiplied by the point ofMultiplication result is distance from point E to cross section
(3) Traversing all data points in the point set W, repeating the step (2), and obtaining the distance from all points in the point set W to the section;
(4) judging whether the distance is smaller than a threshold value V, if so, selecting the data point, and if not, discarding;
(5) and (4) reading the points screened in the step (3), and fitting to obtain a corresponding section contour line.
Further, the obtaining of the characteristic points of the cross section comprises the following steps:
(1) randomly taking three data points from a point cloud data set S in the obtained section contour line, establishing a plane through the three points, and solving a normal vector of the plane;
(2) setting that the point cloud data set S of the cross section contains X data points, continuously extracting other data points in the point cloud data set S of the cross section, and repeating the operation of the first step until all the data points are extracted;
(3) carrying out mean value operation on the obtained normal vectors of the plurality of planes to obtain an equation of the plane where the section is located;
(4) taking any two non-parallel unit vectors a (i) in the cross-section planea，ja，ka)，b(ib，jb，kb) Calculating the projection lengths of all data points in the cross-section point cloud data set S projected on two unit vectors, and respectively summing to obtain daAnd db，daIs the sum of the projected distances of all data points in the unit vector a, dbIs the sum of the projected distances of all data points in the unit vector b;
(5) and (3) calculating the coordinates of the centroid Z of the section on two vectors, wherein the calculation formula is as follows:
U＝da/X，V＝dbthe coordinate value of the cross section centroid Z on the unit vector a is U, and the coordinate value of the cross section centroid Z on the unit vector b is V;
(6) and (3) calculating the coordinate of the section centroid Z (p, q, r) in an xyz coordinate system, wherein the formula is as follows:
p＝U*ia+V*ib，
q＝U*ja+V*jb，
r＝U*ka+V*kb。
further, the steps of carrying out shape retrieval and template fitting on the point cloud model and the model in the template library are as follows:
(1) respectively counting the gray scale of each data point and the template library in the detection point cloud model, and using the set
D＝{d1，d2，d3，...，dnStore, diIndicating that the i-th pixel and the kernel pixel have poor gray levels;
(2) sorting the elements of the set D from small to large, and then splitting D into a head class and a tail class which are expressed as D1＝{d1，d2，d3，...，dqAnd D2＝{dq+1，dq+2，dq+3，...，dnAnd calculating the inter-class variance of two gray level difference classesThe formula is as follows:
wherein the content of the first and second substances,
(3) when the inter-class variance takes the maximum value, the separability of two gray difference classes is the strongest, namely the inter-class varianceWhen the maximum value is obtained, the optimal gray level difference threshold value is judged to be obtained, namely, the optimal gray level threshold value q is obtained*When, satisfy the following formula:
(4) obtaining point cloud dataBest gray segmentation threshold q for poor gray of middle single point and template library*Then, the point cloud data is divided into two types, and when the gray value is poorer than q*If so, the point cloud data is considered to be optimally matched with the template base, and the part of point cloud data is reserved; when the gray value is worse than q*And discarding the part of point cloud data to finally obtain the point cloud set with the optimal matching degree with the template library.
In step S10, the surface texture and the name plate of the single equipment in the power station are photographed by using an oblique photography technique, and the texture and the name plate information of the single equipment are mapped onto the model in step S50.
The invention has the following beneficial effects:
1. the three-dimensional laser scanning measurement technology has the characteristics of high scanning speed, strong real-time performance, high precision, strong initiative, full digital characteristics and the like, can directly obtain a high-precision three-dimensional digital model through simple processing of later software, does not need time-consuming and labor-consuming data processing, and can greatly reduce the cost.
2. When the method is used for modeling, the point cloud data is high in processing speed and precision, and the rapidity and effectiveness of virtual reality modeling of the power system are effectively improved.
3. The accurate three-dimensional model structure obtained by the method can be used for constructing high-precision virtual reality model resources, and can be used for analyzing and calculating the three-dimensional characteristics of different power equipment to realize the parameterized organization and management of the virtual reality model resources.
Drawings
Fig. 1 shows 7 two-dimensional point data points.
FIG. 2 is an initial segmentation of two-dimensional data points.
Fig. 3 is a topological relationship network between all points.
FIG. 4 is a kd-Tree distribution between data points.
FIG. 5 is a diagram that names data points on a kd-Tree distribution.
Fig. 6 shows the distance between the point P and other points.
In FIG. 7, (a) the normal vector distribution of the section of the circular tube type steel, and (b) the normal vector distribution of the section of the rectangular tube.
FIG. 8 is a schematic drawing of a cross-sectional contour line with dots.
FIG. 9 is a schematic diagram of cross-sectional centroid calculations.
FIG. 10 is a flow chart of the method of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments.
The invention provides a transformer substation modeling method based on three-dimensional scanning, which comprises the following steps:
s10, collecting data;
exploring a transformer substation site: the method is characterized in that the three-dimensional laser scanner has great influence on the arrangement of scanning stations due to different measuring distances and measuring speeds, when the measuring stations are arranged, the complementarity among the stations and the visibility among adjacent stations need to be considered, the coverage of the stations can cover the whole transformer substation, the point cloud data between two adjacent measuring stations has 30% -40% of overlapping degree, the acquisition precision of the scanner is controlled, and the precision is that the coordinate linear error does not exceed +/-1 mm; the error of point cloud registration is not more than +/-2 mm; the three-dimensional coordinate unit is meter and is accurate to three positions after decimal points, when the data of the transformer substation is acquired, the three-dimensional laser scanner can be controlled through the control interface, the three-dimensional laser scanner is controlled to scan the entity and the reflection reference point of the transformer substation, and finally the geometric position information and the point cloud data of the transformer substation are obtained
The method comprises the steps of utilizing an oblique photography technology to photograph substation equipment, obtaining appearance textures of the substation equipment, and constructing an equipment texture library so as to facilitate later modeling processing and utilization and ensure that a photographed equipment nameplate is clear and visible.
S20 point cloud segmentation and point cloud clustering
And performing point cloud segmentation and point cloud clustering on the obtained point cloud data, wherein the point cloud segmentation comprises ground segmentation and power line segmentation, the ground segmentation is the point cloud data with the ground of the transformer substation removed, the power line segmentation is the extraction by using height filtering and space density filtering, and the point cloud clustering is the blocking operation on the integral point cloud data to obtain the point cloud data of each equipment monomer.
S30 point cloud data processing of each equipment monomer;
and sequentially carrying out data splicing, data denoising, data simplification and feature extraction on the point cloud data of each block to obtain a point cloud model of the point cloud data of each block.
S40 fast modeling
Extracting minimum repeating units from point cloud models of all equipment monomers, then matching the models to obtain a three-dimensional model of each minimum repeating unit, wherein the extraction of the minimum repeating units adopts a point cloud repeating structure detection algorithm based on a two-dimensional plane bounding box topological structure, the extracted point cloud models of the minimum repeating units are subjected to shape retrieval on a CAD model in a template library, and the retrieved models and the point cloud models are subjected to template fitting;
s50: and (4) reconstructing the three-dimensional model of the minimum repeating unit according to the detected repeating mode, so as to quickly obtain the model of the whole transformer substation.
The data splicing method of the point cloud data adopts indirect splicing-splicing of a point set and a point set.
The point set to point set splicing does not need a target to splice because very accurate homonymous points are not needed, but the 30% -40% overlap degree of point cloud data between two adjacent measurement stations must be ensured, and the method comprises the following steps:
(1) 3 or more homonymous points need to be uniformly selected in the overlapped area during splicing. And calculating conversion parameters. The method comprises the following steps that when the homonymous points are selected, manual operation causes that the accuracy of the selected homonymous points is not enough, so that rough splicing is selected only preliminarily;
(2) on the basis, an Iterative Close Point algorithm (ICP) is performed. The ICP algorithm is iterative to compute a set of neighboring nearest points. The conversion model error is adjusted and evaluated. An algorithm that minimizes the error. The ICP algorithm is highly accurate and the scanning process is convenient because no reflector is required.
Due to the fact that the area of a transformer substation is generally large and the characteristics of a laser scanner, the existing three-dimensional laser scanner cannot acquire complete point cloud data of the transformer substation through one-time scanning. Therefore, the point cloud data of the transformer substation can be rapidly acquired by adopting a multi-station scanning and splicing technology. Accurate splicing of multi-station scanning data is realized based on a point cloud data splicing algorithm, so that complete three-dimensional point cloud data of the transformer substation is obtained
The method for judging the noise point of the point cloud data comprises a direct inspection method, a curve inspection method, a chord height difference method and a value limiting method
The direct examination method comprises the following steps: the point cloud data is directly observed, points which are isolated or far away from the point set are judged to be noise points, and points which are different from the point cloud of the measured object obtained by scanning and appear on a screen are found out by carrying out operations such as amplification, reduction, rotation and the like on the point cloud data, wherein the points are generally isolated or far away from the point set. The primary inspection is carried out in this way to remove more obvious noise points;
the curve checking method comprises the following steps:
(1) setting a threshold value epsilon;
(2) fitting a curve passing through head and tail data points of the section by a least square method;
(3) selecting data points as UiCalculating the data point UiEuclidean distance to curve ei；
(4) If eiIf | is greater than or equal to ε, then UiThe points are noise points;
the string height difference method comprises the following steps:
(1) a set threshold value xi;
(2) select data point QiConnecting data points QiObtaining a chord from the front point and the rear point;
(3) calculating QiDistance f to the chordi；
(4) If fiIf | is not less than xi, then QiThe points are noise points.
The value limiting method comprises the following steps:
(1) data in point cloud dataMinimum and maximum distance values between points and setting a limit value D therefromminAnd Dmax；
(2) Selecting the distance O between any two data points of the point cloud datai；
(3) If O is presenti>DmaxThen O isi＝DmaxIf O is presenti<Dmin，Oi＝Dmin。
During the scanning process, the scanned data may not truly represent the object to be studied due to external interference. Because the laser is used for scanning, the instrument has certain errors because the laser beam has certain dispersion, and this makes the machine possibly receive reflected beams reflected by objects which are not measured, and when the reflected beams are reflected by other objects, the reflected beams are noise to us. The noise points not only make the number of the scanned point clouds large, which occupies a lot of storage space, but also have bad influence on the modeling precision and speed.
In the scanning process, after the surrounding environment is observed, the main sources of the noise points are determined to be surrounding personnel, the flow of vehicles and tree greening. So doing more preparation in selecting the measurement location at the early stage reduces these efforts at the later stage. These noisy points need to be processed prior to modeling in order to avoid that they adversely affect the modeling.
The method for simplifying the point cloud data in the method comprises the following steps
(1) Taking any data point Y in the cloud data, and obtaining N data points closest to the data point Y through a kd-tree algorithm, wherein the value range of N is usually from several to dozens;
(2) fitting a least square surface through the N points with the minimum distance;
(3) setting a threshold value D0Calculating the actual distance D from other data points in the point cloud data to the least square surfacei；
(4) Determining the actual distance DiWhether or not it is greater than a threshold value D0If yes, removing the data point and repeating the step, and if not, finishing the step.
The simplified process of this data is now simulated with a set of data, here shown as two-dimensional data, (the dividing line between a point and a point needs to be replaced by a dividing plane).
(1) Building kd-Tree
The kd-tree is actually a binary tree, but each node of the kd-tree represents a point domain, the process of establishing the kd-tree is actually a dividing point, and the topological relation between the points is established, so that the points are ordered. Let the data points be (1, 5), (7, 9), (2, 3), (8, 6), (4, 1), (3, 8), (6, 4), 7 points as shown in fig. 1, and now a connection between them is to be established.
The first step is as follows: the variance of the data points in each dimension is calculated, and the dimension with the largest variance is obtained and is used as the basis for determining the direction of the subsequent expansion. (note: the variance in the dimension is the largest, which means that the distribution of all points in the dimension is the most sparse, so that the binary tree distribution of the point cloud data obtained after region division is performed according to the direction is more dispersed and is more beneficial to observation.) in the sample data, the variance in the Y-axis direction is found to be larger through calculation, so that the direction perpendicular to the Y-axis is taken as the first dividing line.
The second step is that: all points on the Y axis are sorted from small to large, and then a midpoint value is obtained (if the points are even numbers, the middle two points can be selected), so far, the midpoint in the Y direction can be obtained, and the midpoint in the sample data is (1, 5). Therefore, a dividing line perpendicular to the Y axis is made for the passing points (1, 5), and the data group division of the first step is realized. As shown in fig. 2
The third step: the intermediate points selected in the second step are sorted according to the size of the Y-axis data, and then the data group may be divided into two blocks, one block being a point (4, 1), (2, 3), (6, 4) below Y5, and the other block being a point (8, 6), (3, 8), (7, 9) above Y5.
The fourth step: and (3) sequentially repeating the work from the first step to the third step aiming at the upper and lower point sets, continuously increasing the segmentation lines, performing deeper area segmentation on the data group, and finally dividing the data group into the condition that each point set only contains one point. Note that subsequent split line directions should be determined in sequential cycles along the dimension order (in this case, Y-axis, X-axis, and so on, until all splits are performed, and further, subsequent sorted queries should be performed in this order). And until the kd-tree is established, each point corresponds to a point set, and the establishment of the topological relation network among all the points is completed. As shown in fig. 3.
The process of establishing a kd-Tree by a person is shown in the grid diagram. The process of building a kd-tree is actually a process of building a binary tree. The first point obtained in the first step and the second step can be regarded as a root node and a first-order father node of the binary tree, and then the data is divided into a left subset and a right subset. The nodes obtained by the two subsets when the previous steps are repeated are the child nodes of the upper-level tree and the parent nodes of the later-level tree. And (4) subdividing until all the subset points become leaf nodes (or the tree can not be subdivided), and completing tree building. As shown in fig. 4
(2) Obtaining the nearest N points
After the kd-tree is established, the kd-tree can be used for further point distance acquisition. In the former algorithm, the coordinates of all points are used for directly calculating the Euclidean distance between a certain point and other points, so that the calculation amount and the time complexity are greatly increased, and the time for searching the closest point can be effectively shortened by establishing the relationship between the points after the kd-tree.
Arbitrarily, a point is designated as the point being sought, such as point P (1, 6). And points a (1, 5), B (7, 9), C (4, 1), D (3, 8), E (8, 6), F (2, 3), and G (6, 4) are set. As shown in fig. 5
The first step is as follows: inquiring the A point (1, 5) of the root node, and P in the Y directiony＝6>AyAnd 5, entering the space on the right side of the point A to perform the next inquiry.
The second step is that: inquiring sub-node B point (7, 9), X direction: Px＝1<BxAnd 7, entering the space on the left side of the point B for the next query.
The third step: the resulting leaf node D (3, 8) may be the closest point to point P. The distance D between the point P and the point D is calculated to be 2.828.
The fourth step: backtracking the previously queried points A and B and calculating the distance between points P and them, the distance d between points A and P11, the distance between the points P and B is d26.708. Discovery d2>d, so all points on the other side of the query B point do not need to be backtracked. d1<D, namely the point D is certainly not the closest point, and the other side of the point A needs to be entered for sequentially inquiring. As shown in fig. 6
Calculating the distance d from the point P to the first parting line on the right side of the point A (the parting line of the point C)3＝3>d, so there is no need to query all points to the right of point C. Then, the distance d from the point P to the left side of the point C (the division line of the point F) is calculated4＝3>d, so it is not necessary to query all points to the right of point F, and the just operation is repeated until the leaf node (e.g., point F in this sample) is reached again, i.e., all backtracking is completed, and it can be seen from the calculated distance result that the closest point is point a (if there is no point with a distance smaller than d in the backtracking query of the fourth step, the previous operations are not needed). The N value set at the initial time can be obtained by repeating the above procedure. It should be noted that, for different graphs and point cloud data, the value of N may have a large difference, and needs to be determined according to actual conditions, and the number of N is generally several to several tens.
(3) Establishing an optimized least squares surface
By the above operation, we can obtain the N nearest points, and the content of this operation is to use the points obtained previously and then fit a nearest plane. The method comprises the following steps:
the first step is as follows: let the parametric equation for the spatial plane S be:
ax + By + Cz + D ═ 0 formula (1-1),
the four letters A, B, C and D in the formula (1-1) represent four different parameters of the spatial plane.
The second step is that: and (3) randomly taking out three points from the N points, substituting the three points into the parameter equation (1-1) of the S space plane, and solving a multivariate linear equation set to obtain the relation of unknown parameters A, B, C and D.
The third step: and then, utilizing a distance formula from the space point to the space plane:
wherein, the parameters A, B, C and D in the formula (1-2) are the parameter values of the parameter equation of the solved space plane, and x, y and z are the coordinate values of a certain point in the space to be calculated.
The fourth step: calculating the distance d of each point in the set N to the plane SiThen for these distance values diPerforming a summation operation, namely:
wherein L in the formulae (1-3) represents the sum of all distances, diRepresenting the distance of each point to the plane S and N representing how many distance values there are in total.
The fifth step: since the previously created surface S is a randomly chosen point, according to the calculation formula:
where N in formulas (1-4) represents how many neighboring values are extracted in total and B represents the number of possibilities to create different planes S.
As can be seen from the formulae (1-4), there are different spatial planes S in total1，S2，……SBOne B plane. Therefore, there will also be L accordingly1，L2，……LBAnd a total of B distance sums.
And a sixth step: then mixing L1Store L1And L2Comparing the sizes, if it is L1<L2Then continue to compare L1And L3The size of (2) is calculated sequentially backwards. If L appearsiValue ratio L of1Smaller, then with just LiReplace the preceding L1And new L1And storing the data. Then, L is followed1And LiTo LBAnd (4) the process of comparing subsequent values. Until the comparison is completed, we will obtain the minimum distance sum value LminThen L isminCorresponding surface SminI.e. the closest plane, i.e. we optimize the least squares.
(4) Setting a threshold value, removing outliers
Based on the fact that an optimized least square surface has been established in the previous step, the actual distances D from other points to the least square surface are now calculatedi. At the same time, a suitable threshold value D estimated according to actual conditions is set0Comparing the two, if Di<D0If the point is not outlier or noise, the change point is determined to be not outlier or noise; if it is Di>D0The change point is removed as a noise value. It should be noted that the threshold value cannot be set too large, otherwise the simplification and denoising effects are not obvious; if the threshold is set too small, the removal of the point is excessive, and part of the important data may be lost, thereby distorting the point cloud.
The method for simplifying the point cloud data may also be the following method
One such method is to simplify for a given number of data points. The algorithm deletes a part of points according to a certain priority, and the distance between the points is taken as a measurement. The simple idea of the algorithm is that the number of a data point is given as a threshold, and when the number of the existing points is larger than the threshold, one of the two points closest to the existing point is deleted. Each step is a loop such that the calculation stops until the number of current points is less than a given threshold.
Another simplification by distance between given data points is a simpler criterion. The main idea is that, given a threshold value of a distance, only the distance between points needs to be simply compared with the threshold value, and edge nodes with the distance smaller than the threshold value are deleted, and those with the distance larger than the threshold value are retained. The algorithm only needs to traverse all point cloud data once, and is simple to implement and high in efficiency.
The two simplifying methods are not suitable for simplifying the position with larger curvature, the method of the invention simplifies the position with higher accuracy according to the normal direction, and when a point is deleted, an error is generated in the normal direction of the curved surface at the point. We can evaluate if it can be deleted by calculating the error magnitude.
The method disclosed by the invention is based on the kd-tree algorithm, a least square fitting method is fused, and the method is provided after autonomous analysis research and algorithm improvement, so that the actual point cloud data of the industrial steel structure can be simplified, and the step can also be used for denoising the data.
The feature extraction of the point cloud data comprises the following steps:
(1) spindle direction acquisition
The object processed by the feature extraction algorithm is an industrial steel structure, and the surfaces of most steel structures have certain regularity, so that the feature can be utilized when a normal vector is established. For example, the shapes of the round tube type steel structure and the rectangular steel structure are almost plane structures, so the normal vector is always perpendicular to a tiny plane near the point. For example, in a circular tube steel structure, a certain point on the surface is selected, the normal vector of the certain point is the direction which passes through the center of a circle and is perpendicular to the axial lead of the circular tube, and meanwhile, the plane which is fitted by a plurality of points which are closest to the certain point is the shape and the position which are similar to the longitudinal section of a cylinder. So it is summarized that, for the characteristics of the processed steel structure, the required normal vector and the least square surface of the point are vertical, as shown in fig. 7, the specific process is as follows:
the first step is as follows: m points with the nearest distance to a certain point T (x, y, z) are obtained, and a least square surface of the point T is established by using the M points. The algorithm part of the step can directly call the algorithm of the previous data simplification part, and the algorithm is largeThe process is as follows: firstly, arbitrarily taking a certain point T, and then acquiring a plurality of points with the nearest distance through a binary tree (note that the point is taken to approximately obtain the normal vector of the point, so that the point does not need to be taken greatly when the N points with the nearest distance are taken, and the accuracy is not needed); finally, a least square surface S is established by the N points by using the method described aboveTThe equation is:
ax + By + Cz + D ═ 0 formula (2-1)
Wherein A, B, C, D in the formula (2-1) represent several parameters of the plane, and x, y, z represent coordinates of points on the plane. After the least squares are found, the four parameters A, B, C, D should have been found.
The second step is that: finding the least square surface STThe normal vector of (1). Since the preceding plane equation is Ax + By + Cz + D equal to 0, the normal vector of the plane can be directly obtained, and the direction is the vector r1The direction in which (a, B, C) points.
The third step: solving the normal vector of the point T: from the previous analysis, the normal vector of the T point and the least square surface S of the T point can be knownTIs in the same direction, i.e. vector r1The direction in which (a, B, C) points.
The fourth step: taking a larger point domain H near the T point, wherein the point domain H contains K data points, then carrying out the first step to the third step on all the points in the point domain H, and solving all normal vector vectors, namely r1，r2，……rKThe number of the cells is K.
The fifth step: and calculating the rough direction of the main shaft direction. For K normal vectors, arbitrarily take two vectors r among themiAnd rjThe inner product of the two vectors is obtained, i.e. the cross product operation is performed to obtain the vector l1. From the nature of cross multiplication: vector l1Direction and vector r ofiAnd rjAre all perpendicular, vector l1Is the direction of the main axis. Moreover, since the vector riAnd rjIs obtained randomly according to the calculation formula:
wherein K in the formula (2-2) represents how many normal vectors are extracted in total, and R represents establishing different main axis direction vectors l1The number of possibilities of (a).
So that R principal axis vectors l are obtained1，l2……lRThen, the average value/average of the R direction vectors can be found as a rough estimate of the principal axis direction.
And a sixth step: and (5) calculating the accurate main shaft direction. Calculating lFlat plateWith other R principal axis vectors l1，l2……lRSetting an included angle threshold alpha, if the included angle between the two is larger than the threshold alpha, removing the main shaft direction vector, if the included angle does not exceed the threshold alpha, reserving the included angle, and finally carrying out average operation on all reserved main shaft direction vectors to obtain the main shaft direction vector l of the part of point cloudT1。
And a sixth step: the overall main shaft direction is obtained, the main shaft direction of a certain section of the pipeline is obtained from the first step to the fifth step, and the main shaft direction of the rest part needs to be obtained considering that the pipeline and the industrial steel are possibly long in length. From the point T of the third step, along the main axis direction, the next point T can be obtained2And a certain distance value is arranged between the two points. Then for point T2Performing the algorithm from the first step to the fifth step, and finally obtaining other main axis direction vectors l in sequenceT1，lT2，lT3……。
(2) Obtaining a cross-sectional profile
The main axis direction vector of the integral steel structure is obtained in the previous step, and the section contour line is obtained by using the main axis direction. Because it includes the main axis direction vector lT1，lT2，lT3… …, so that there will be correspondingly many cross-sectional profiles, now given one of the principal axis directions lT1To (p, q, r), the interface contour of the neighborhood is determined. From the foregoing algorithm, the principal axis direction is foundlT1The basic point used is point T, and the coordinates (x, y, z) of point T are known, so the cross-sectional profile of point T is traversed by the following steps:
the first step is as follows: taking other points in the vicinity of point T, e.g. point E (x)e，ye，ze) Point F (x)f，yf，zf) Then, from the coordinates of point T, E, F, the vector directions of the straight line ET and the straight line FT, the length of the line segment, and the expression of the straight line can be obtained.
The second step is that: by dividing the vector lT1Dot-multiplied with the vector ET, and the result modulo is given as the vector lT1Is modulo by vector ET at vector lT1Length of projection in direction. The length of the line segment EE' is shown in fig. 8 below. The amount of EE' can be used to measure the distance from point E to the desired cross-sectional profile near point T;
the third step: and repeating the second step to calculate the distances from all the points in a larger area nearby to the contour surface. Then setting a threshold value d, if the distance between the point and the surface is less than the threshold value d, as the length of FF' in the following figure 8, reserving F and storing a new address, and calculating the point F as a point on the section contour; if the point-surface distance is greater than the threshold d, as shown in the length of EE' in FIG. 8, skipping the process of storing a new address of the point E, and entering the screening of the next point until the screening of the points in the neighborhood is finished;
the fourth step: reading the points screened in the third step to form a section contour thin point cloud film, and fitting the points to obtain a corresponding section contour curve, such as the section of a circular pipeline is circular, and the characteristic points of the rectangular (3) section of rectangular steel are obtained
If the circular tube structure is adopted, the feature points needing to be extracted are certainly the circle center; if the steel is rectangular steel, the characteristic point to be extracted is the circle center of the circumscribed circle of the steel; correspondingly, for the H-shaped steel, the center of gravity and the center of the pattern need to be certain, and all the patterns which are summarized can be carried out by using a unified algorithm, namely extracting the center of mass of the thin-point cloud film, and the steps are as follows:
the first step is as follows: and (3) randomly removing three points from the obtained cross-section point cloud data set S, establishing a plane through the three points, and then calculating a space plane normal vector of the plane. The method of establishing the plane and the method of obtaining the normal vector of the plane are as described above.
The second step is that: and continuously extracting other groups of data in the section data set S, and repeating the operation of the first step until all the groups of data are extracted.
The third step: and (2) carrying out averaging operation on the plurality of normal vectors according to the obtained normal vectors to obtain an equation of the whole plane where the final point cloud is located, (the step can be carried out by firstly obtaining an average value, then setting a threshold value, comparing the included angle between each normal vector and the uniform normal vector, removing a larger included angle value, and finally obtaining the average value of the remaining normal vectors, so that the calculated plane is more accurate, and the error is smaller). Let this found plane be F.
The fourth step: arbitrarily take a unit vector a (i) in the surface Fa，ja，ka) Then another unit vector b (i) not parallel to the unit vector a is takenb，jb，kb) If X points are included in the slice point cloud data set S, the length values d of the X points projected to the vector a and the vector b are calculated respectively1a，d1b，d2a，d2b，d3a，d3b……dXa，dXb。
The method of calculating the point-to-unit vector distance is as follows: setting the required point as H, and taking a point G on the unit vector a (directly taking the actual coordinate origin in the algorithm) by formula
In the formula (2-3)Is a unit vector, the value after taking the modulus is 1, so the final result of the above formula is a vectorIn the vectorProjection in direction, i.e. vector of point HThe length values of (a) are shown in figure 9,
the fifth step: calculating daAnd dbThe value of (c). The weight of each point is set to be 1, so the distance calculation process is a summation formula
Wherein X in equations (2-4) and (2-5) represents the number of points included in common, i.e., the number d of distance values to the unit vectoria、dibRepresenting the distance of each point to the unit vector, da、dbRepresenting the sum of the distances of all points to two unit vectors.
The fifth step: the coordinates of the center of mass on the two unit vectors are calculated. Since the weight of each point is set to 1, the total mass is X, and the result can be calculated
U＝da/X (2-6)，
V＝db/X (2-7)，
Wherein, U is the coordinate value of the center of mass on the unit vector a, and V is the coordinate value of the center of mass on the unit vector b;
and a sixth step: from the above two equations, the coordinates (U, V) of the centroid under the new vector basis can be found, so the coordinate position R (p, q, R) of the centroid within the original xyz coordinate system can be found, so the coordinate calculation equation of the centroid R is as follows:
p＝U*ia+V*ib (2-8)，
q＝U*ja+V*jb (2-9)，
r＝U*ka+V*kb (2-10)，
the algorithm can also directly take x (1, 0, 0), y (0, 1, 0), z (0, 0, 1) and three vector bases, calculate the mass center on the three vectors, and finally obtain the central point of the section, namely the required characteristic point.
(4) Characteristic line forming and outputting
The method for fitting the characteristic line by the characteristic points is various, and the common method, such as using MATLAB and opengl, can realize the function of curve fitting, but considering that in practical application, an entity model may be further acquired or the characteristic curve may be further processed, so that the principle of twice development of the CATIA is used to fit the required characteristic line in the CATIA. The general idea is to utilize the CAA to carry out secondary development on the CATIA, and add a function of curve fitting to the CATIA.
In the modeling process of the transformer substation, firstly, preprocessing data is carried out, namely, point cloud data of power lines on the ground of the transformer substation and equipment are automatically removed, then, a region growing algorithm is utilized to enable the point cloud data of the equipment to be independent, then, a point cloud model is established through data splicing, data denoising, data simplification and feature extraction, on the basis, a power station scene repetitive structure mode is respectively set through an algorithm of repetitive structure detection based on a two-dimensional plane bounding box topological structure, finally, a modeling method based on component matching is adopted to model a minimum repetitive unit (including shape retrieval and template fitting), and the minimum repetitive unit is combined with the repetitive structure mode, so that a transformer substation model is obtained, and the steps of shape retrieval and template fitting of the point cloud model and a model in a template library are as follows:
(1) respectively counting the gray scale of each data point and the template library in the detection point cloud model, and using the set
D＝{d1，d2，d3，...，dnStore, diIndicating that the i-th pixel and the kernel pixel have poor gray levels;
(2) the elements of the set D are sorted from small to large, and then D is split into a head class and a tail class which are expressed asD1＝{d1，d2，d3，...，dqAnd D2＝{dq+1，dq+2，dq+3，...，dnAnd calculating the inter-class variance of two gray level difference classesThe formula is as follows:
wherein the content of the first and second substances,
(3) when the inter-class variance takes the maximum value, the separability of the two gray difference classes is the strongest, namely the inter-class varianceWhen the maximum value is obtained, the optimal gray level difference threshold value is judged to be obtained, namely, the optimal gray level threshold value q is obtained*When, satisfy the following formula:
(4) obtaining the best gray segmentation threshold q of the gray difference between the single point in the point cloud data and the template library*Then, the point cloud data is divided into two types, and when the gray value is poorer than q*If so, the point cloud data is considered to be optimally matched with the template base, and the part of point cloud data is reserved; when the gray value is worse than q*And discarding the part of point cloud data to finally obtain the point cloud set with the optimal matching degree with the template library.
The specific flow of the method is shown in fig. 10.
Three-dimensional data acquisition implementation requirements of main building and main components of transformer substation
(1) Three-dimensional data acquisition of main building of transformer substation
1. And recording and sorting related information, and recording the name, the voltage grade, the property of the transformer substation and three-dimensional data information of the transformer substation in detail.
2. And accurately acquiring three-dimensional point cloud data and acquiring the three-dimensional point cloud data of the transformer substation building.
3. The precision is that the linear error of the coordinate does not exceed +/-1 millimeter; the error of point cloud registration is not more than +/-2 mm; the three-dimensional coordinate unit is meter (m) and is accurate to three bits after decimal point.
4. Shoot high definition photo
And (3) taking two pictures of the periphery of the transformer substation: a transformer substation enclosure photo; a gate photo can clearly see the name of the transformer station. The format of the photo is jpg, and the pixel is recommended to be not less than 1024 × 768 px.
(2) Three-dimensional data acquisition of main components of transformer substation
The data acquisition contents of the station transformer mainly comprise space position coordinates, equipment appearance photo data and the like, and the specific contents comprise
1. Acquiring the position of the power converter equipment for the station, and acquiring three-dimensional point cloud data of the power converter equipment for the station;
2. and recording and sorting related information, recording information such as the number, material, height, loop name, wiring mode and the like of equipment in detail, and recording equipment information such as a disconnecting link, a switch, a drop, a collection terminal and the like of equipment on the power converter.
3. Controlling acquisition accuracy
The precision is that the linear error of the coordinate does not exceed +/-1 millimeter; the error of point cloud registration is not more than +/-2 mm; the three-dimensional coordinate unit is meter (m) and is accurate to three bits after decimal point.
4. Shoot high definition photo
Device photo, taking three photos: a device overview picture; the equipment nameplate photo can clearly see the equipment nameplate (if the condition that the equipment body is suspended and dispersed with the nameplate is existed, a plurality of nameplate photos can be shot according to the actual requirement); a photograph showing the geometry of the device clearly. The plate and geometric duplication is recorded and photographed in the format of jpg, pixel suggestions no less than 1024 × 768 px.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.
- 上一篇：一种医用注射器针头装配设备
- 下一篇：三维人脸重建方法、装置、电子设备及存储介质