Automatic extraction method of laser point cloud of overhead line support facilities considering spatial relationships

Abstract

The present invention discloses a fully automatic extraction method for laser point clouds of contact network support facilities that takes spatial relationships into account, including: S1, obtaining key trajectory points through GNSS trajectory point thinning processing, and layering and blocking the original point cloud data accordingly to achieve the division of the pillar area and support facility area in the original point cloud data; S2, obtaining the pillar center point through neighborhood search and constructing a spatial index based on its spatial relationship with the trajectory point and support facility search layer; S3, using this index as a driving force to achieve the initial extraction of support facilities, obtain support facilities containing contact lines, and introduce columnar search and parametric projection filtering to filter out contact line point clouds to achieve optimization of extraction results. The present invention can quickly locate the support facility point cloud area, and combines point cloud filtering such as column filtering and parametric projection to achieve initial extraction and structural optimization of support facilities, greatly improving extraction accuracy and better adapting to complex support facility types.

Description

Technical Field

[0001] The present invention relates to the field of contact network detection, and in particular to a method for fully automatic extraction of laser point clouds of contact network support facilities taking into account spatial relationships. Background Art

[0002] The overhead catenary system (OCS) refers to the electromechanical system that provides electrical energy to electric traction units via current collectors. It is generally composed of five parts: pillars and pillar foundations, supporting facilities, positioning devices, contact suspensions, and auxiliary power supply facilities. It is the core of electrified railways. Affected by severe natural weather such as wind, rain, snow, and dust, the overhead catenary system may experience loosening of the supporting structure, aging, and geometric changes in the position of the supporting structure. This may cause the supporting facilities to deform or even break, seriously affecting the safety of railway operations. To meet the major challenges brought about by the rapid growth of railway mileage, the design of a normalized and efficient detection method for the overhead catenary will play an important role in ensuring railway transportation safety, improving transportation efficiency, and reducing transportation energy consumption.

[0003] Currently, there are two main methods for collecting support facility data: contact and non-contact. The former uses inductive sensors installed on the pantograph slide to dynamically collect support facility data. However, this method suffers from drawbacks such as severe electromagnetic interference and inconvenient maintenance, making it difficult to meet the current needs of railway maintenance. The latter acquires measurement data based on 2D images or laser point clouds. 2D images provide rich target data, but data quality is difficult to guarantee due to limited lighting conditions. Laser point clouds, on the other hand, can efficiently and accurately capture rich spatial data about railway scenes. In recent years, vehicle-mounted mobile laser scanning systems (MLS) have developed rapidly. However, the acquired LiDAR data contains a wide range of target information, including the catenary and track. Therefore, exploring 3D LiDAR-based automatic support facility extraction methods to achieve accurate support facility point cloud extraction will be of great significance for the precise segmentation and geometric parameter detection of railway facilities.

[0004] Support facility extraction methods can be roughly divided into three categories according to their mechanisms: filtering method, geometric feature method, and segmentation method. Liu Wenqiang et al. (Correction of contact network geometric parameter detection values based on camera calibration and Kalman filtering [J]. Journal of the China Railway Society. 2014, 36(09): 28-33.) proposed Kalman filtering to filter the original image of the railway scene. Chen Guocui et al. (Contact network insulator defect detection method integrating PHOG and BOW-SURF features [J]. Small and Microcomputer Systems. 2021, 42(01): 172-177) used an improved fast guided filtering algorithm to filter the original image of the contact network insulator. The geometric feature method uses the common geometric features of objects of the same type to classify and extract them; Guo Baoqing et al. (Three-dimensional point cloud segmentation and classification recognition algorithm for railway scenes [J]. Chinese Journal of Instrumentation. 2017(09): 2103-2111) proposed a region growing extraction algorithm based on the consistency of normal direction based on the geometric feature method. By calculating and comparing the size of the normal direction angle between the seed point and the neighboring point, the extraction of railway scene objects including contact network pillars is completed; Lamas et al. (Li Xiangtai, Li Xia, Zhang Li, et al. Improving Semantic Segmentation via Decoupled Body and EdgSupervision[M]. 2020(11): 435-452.) proposed a heuristic-based semantic segmentation workflow for complex railway environments, and verified the effectiveness of the method using a 90-kilometer-long railway dataset. However, due to the characteristics of railway scenes such as large point cloud data volume and complex scene objects, the above methods are unable to quickly and effectively extract supporting facilities from the original point cloud data. In recent years, deep learning has also been applied to semantic segmentation of point cloud data to extract supporting facilities. However, these methods all implement contact network semantic segmentation based on 2D LiDAR data, and the implementation process is relatively complex and the accuracy is relatively low. Summary of the Invention

[0005] In order to solve the problems existing in the prior art, the present invention proposes a fully automatic support facility extraction algorithm that takes into account the spatial relationship between the pillar center points, key trajectory points and the support facility search layer. The spatial relationship between the pillar center points, key trajectory points and the support facility search layer is used to quickly locate the support facility point cloud area, and combined with point cloud filtering algorithms such as cylinder filtering and parametric projection. By fully combining the advantages of both, the initial extraction and result optimization of support facilities are achieved, the extraction accuracy is greatly improved, and it is better applicable to complex support facility types.

[0006] To this end, the present invention adopts the following technical solutions:

[0007] A fully automatic extraction method for supporting facilities taking into account spatial relationships comprises the following steps:

[0008] S1. Obtain key trajectory points by thinning the GNSS trajectory points, and then perform layered and block processing on the original point cloud data to achieve the division of the pillar area and supporting facility area in the original point cloud data;

[0009] S2. Obtain the pillar center point through neighborhood search, and construct a spatial index based on its spatial relationship with the key trajectory points and the support facility search layer;

[0010] S3. Using the spatial index as a driver, the supporting facilities are initially extracted to obtain the supporting facilities containing the contact lines. The contact line point cloud is filtered out by introducing columnar search and parametric projection filtering to optimize the extraction results.

[0011] The steps of layering and block processing of the original point cloud data in step S1 are as follows:

[0012] 1) Thinning GNSS trajectory points: Divide m GNSS trajectory points into several segments of distance n. After each segment of distance n, the center point of the segment is used as the key trajectory point. Segments with distances less than n / 2 are ignored. The segment calculation formula is as follows:

[0013]

[0014] Where:

[0015] f(p (i,i+n) ): The rarefaction formula at the key points i to i+n;

[0016] P i : coordinate value of the i-th trajectory point;

[0017] 2) Layered and block processing of the original point cloud: Based on the key trajectory points, a series of pillar clipping regions V1 and supporting facility clipping regions V2 with lengths, widths, and heights of 2Δx, 2Δy, z2-z1, 2Δx, 2Δy, z4-z3 are constructed respectively. After affine transformation by the rotation matrix Matrix, the original point cloud data is clipped to achieve layering and spatial blocking of the original point cloud data. The relevant formulas are as follows:

[0018]

[0019] Where:

[0020] V0: other irrelevant areas;

[0021] x i ,y i ,z i :pi Point coordinates;

[0022] Step S2 includes the following sub-steps:

[0023] 1) By setting the neighborhood range CBox with a length and width of 2Δx1 and 2Δy1 respectively, a neighborhood search is performed on the pillar search layer. The statistical density threshold is used to determine whether the point cloud in the neighborhood is a pillar point cloud. If the conditions are met, it is determined to be pillar point cloud data and its pillar center point is calculated. Otherwise, it is discarded;

[0024] 2) Perform neighborhood search again based on the initial extraction results of the pillar center points, filter out the pseudo pillar center points in the extraction results, and make each pillar center point correspond to the pillar area point cloud data one by one;

[0025] 3) By calculating the Euclidean distance between the key trajectory points and the pillar center points, and by comparing the key trajectory point index value with the closest Euclidean distance to each pillar center point, the topological relationship between the pillar center points and the key trajectory points is constructed. Then, using the pillar center points as the link, the topological relationship between the key trajectory points and the supporting facilities above the pillar center points is constructed, realizing the construction of the search layer spatial index of the pillar center points, key trajectory points and supporting facilities. The formula for constructing the spatial index is as follows:

[0026] Dis=min(dis i ),i∈R + (3)

[0027]

[0028] Where:

[0029] index j : index value of the center point of the jth pillar;

[0030] min(dis i ): The shortest distance between the center point of the pillar and the key trajectory point.

[0031] Step S3 includes the following sub-steps:

[0032] (1) First, the initial extraction area of the supporting facilities associated with the pillar center point is quickly located based on the spatial index, and the clipping range is determined based on the coordinates of the pillar center point. Then, the coordinates of the key trajectory points associated with it are quickly determined through the spatial index, and the clipping range is affine transformed. Finally, the clipping range is narrowed based on the difference in the number characteristics of the point clouds on both sides of the pillar. The pillars, loops, and ratchets are filtered out using a cylindrical filter to obtain the initial extraction results of the supporting facilities.

[0033] (2) The initial extraction results are projected into a parametric model, and the point cloud data is projected into a two-dimensional plane. The plane is then discretized at a set resolution to obtain a series of grid point cloud data. The quantitative characteristics of the point cloud data in the grid are judged to determine whether the point cloud in the grid should be filtered out. The optimized extraction result is the extracted support facility point cloud data.

[0034] In the above step (1), pillars and ratchets are important erection devices of the contact network, which have the characteristics of being distributed in the vertical direction. A column filter is proposed by combining the point cloud library radius filter and the pillar distribution characteristics. The column filter range is set according to the coordinates of the pillar center point to process the support facility search layer point cloud data, and the point cloud data determined to be non-support facilities are filtered out.

[0035] In the above step (2), the contact line is an important power supply device of the contact network. It has the characteristics of horizontal distribution and low proportion of point cloud features. The initial extraction results are optimized by combining the PCL parameterized model projection and discretization processing ideas, and the point cloud data determined to be the contact line are filtered out.

[0036] The present invention utilizes the spatial relationship between pillar center points, key trajectory points, and the support facility search layer to quickly locate the support facility point cloud area. It also combines point cloud filtering such as cylinder filtering and parametric projection, fully combining the respective advantages of both to achieve the initial extraction and structural optimization of support facilities. Compared with the existing technology, the present invention has the following beneficial effects:

[0037] (1) The present invention uses laser point cloud as the data basis for the fully automatic extraction algorithm of railway contact network support facilities. The collected data is less affected by external factors and can express rich spatial and attribute information of railway facilities;

[0038] (2) The present invention optimizes the initial extraction results of support facilities by using cylindrical point cloud filtering and PCL parametric projection filtering, achieving accurate elimination of contact lines and support facility connection facilities, improving the extraction accuracy of support facilities, and better adapting to complex support facility types;

[0039] (3) The present invention uses neighborhood search to extract the center points of pillars, clarify the spatial relationship between railway facilities, and construct a spatial index based on the topological relationship between the center points of pillars, key trajectory points and support facility search layers, so as to achieve the purpose of quickly obtaining trajectory coordinates to determine the railway extension direction vector and quickly locking the associated support facility search layer area.

[0040] (4) The present invention analyzes the spatial relationship of railway facilities and performs fully automatic extraction of supporting facilities based on this. Compared with the traditional supporting facility extraction method, the extraction efficiency is high, the extraction results are comprehensive and detailed, and it can adapt to various types of scenarios. The measurement accuracy meets the requirements of subsequent contact network detection projects such as geometric parameter measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a flow chart of the automatic extraction method of the present invention;

[0042] Figure 2a This is a schematic diagram of the processing of the case where the end of the trajectory point cloud exceeds n / 2 in the present invention;

[0043] Figure 2b This is a schematic diagram of processing the case where the end of the trajectory point cloud does not exceed n / 2 in the present invention;

[0044] Figure 3 is a schematic diagram of performing affine transformation on the cropping area in the present invention;

[0045] Figure 4 It is a schematic diagram of layering and block processing of original point cloud data in the present invention;

[0046] Figure 5a It is a schematic diagram of the neighborhood search process for the pillar search layer area;

[0047] Figure 5b Shown is a schematic diagram of the initially extracted pillar center points;

[0048] Figure 5c It is a schematic diagram of the center point of the pseudo pillar;

[0049] Figure 5d This is a schematic diagram of the result of filtering the center point of the pseudo pillar;

[0050] Figure 6 It is a schematic diagram of spherical search and cylindrical search;

[0051] Figure 7a It is the initial extraction schematic diagram;

[0052] Figure 7b is the initial extraction result;

[0053] Figure 8 It is a filtering diagram that integrates the ideas of parameterized projection and discretization processing;

[0054] Figure 9a It is the test data of multiple types of supporting facilities;

[0055] Figure 9b It is the test result data of multiple types of support facilities;

[0056] Figure 10aIt is the test data of supporting facilities for the distribution of multiple types of pillars;

[0057] Figure 10b It is the test result data of supporting facilities based on the distribution of multiple types of pillars;

[0058] Figure 11 It is the test result of demonstration application. DETAILED DESCRIPTION

[0059] This fully automated spatially-sensitive support facility extraction algorithm uses a vehicle-mounted laser scanner to perform laser scanning of the entire railway scene. By leveraging the relatively stable spatial relationships between pillar centers, key trajectory points, and the support facility search layer, the algorithm is designed and implemented using a computer program to fully automatically extract support facilities from railway scene point cloud data.

[0060] The fully automatic extraction method for supporting facilities of the present invention is described in detail below with reference to the accompanying drawings.

[0061] Figure 1 This is a flow chart of the fully automatic extraction method for support facilities of the present invention. The method mainly includes three steps: S1, obtaining key trajectory points by thinning the GPS trajectory points, and layering and blocking the original point cloud data accordingly to realize the division of the pillar area and the support facility area in the original point cloud data; S2, obtaining the pillar center point through neighborhood search, and constructing a spatial index based on its spatial relationship with the key trajectory points and the support facility search layer; S3, using the index as a drive to realize the initial extraction of support facilities, obtain support facilities containing contact lines, and introduce columnar search and parametric projection filtering to filter out the contact line point cloud to achieve the optimization of the extraction results. The details are as follows:

[0062] S1. Railway scene point cloud data contains a large amount of irrelevant point cloud data, and there are multiple GNSS track points between any two pillars, making it difficult to establish topological relationships between track points, pillars, and layered regions. To improve algorithm efficiency and clarify the spatial relationships between railway facilities in the scene, this paper uses the key track points obtained after thinning to perform layered and block processing on the original scene point cloud data. The details are as follows:

[0063] 1) GNSS trajectory point thinning processing:

[0064] In order to improve the retrieval efficiency of supporting facility point cloud data and facilitate the formation of topological relationships between GNSS trajectory points and supporting facilities, the GNSS trajectory points are thinned out according to formula (1). First, m GNSS trajectory points are divided into several segments with a distance of n. After each segment with a distance of n, the center point of the segment is used as the trajectory key point, as follows: Figure 2a As shown, the distance segment less than n / 2 is ignored, such as Figure 2bThe calculation formula is as follows:

[0065]

[0066] Where:

[0067] f(p (i,i+n) ): The thinning formula at the i-th key point;

[0068] P i : The coordinate value of the i-th trajectory point.

[0069] 2) Layering and block processing of original point cloud:

[0070] Since the extension direction of the original point cloud data is angularly different from the coordinate axis direction and there are a large number of irrelevant point cloud areas in the railway scene, the original point cloud data needs to be layered and spatially divided. The specific steps are as follows:

[0071] First, the coordinates of the key trajectory points and the coordinates of the adjacent GNSS trajectory points are used to obtain the trajectory normal vector, that is, the trajectory direction (x1-x, y1-y, 0); then the vector dot product and vector cross product are used to obtain the vertical vector of the trajectory direction and the vertical direction of the trajectory surface; finally, the trajectory normal vector and its calculated results are normalized to obtain the trigonometric function value corresponding to the angle between the trajectory direction and the coordinate axis direction, and a rotation matrix is constructed to perform affine transformation on the original point cloud data to make the trajectory direction more consistent with the coordinate axis direction. The schematic diagram of the affine transformation is shown in the figure. Figure 3 shown.

[0072] Secondly, in view of the height distribution differences between pillars and supporting facilities, a series of pillar clipping regions V1 and supporting facility clipping regions V2 with lengths, widths, and heights of 2Δx, 2Δy, z2-z1, 2Δx, 2Δy, z4-z3 are constructed based on key trajectory points. After affine transformation by the rotation matrix Matrix, the original point cloud data is clipped, and finally the original point cloud data is layered and spatially divided into blocks. The schematic diagram of the layered and block-divided original point cloud data is shown in the figure. Figure 4 The relevant formula is as follows:

[0073] Where:

[0074] V0: other irrelevant areas;

[0075] x i ,y i ,z i :p i Point coordinates.

[0076] Through the above-mentioned steps of stratification and spatial segmentation of the original point cloud data, a large amount of point cloud data in irrelevant areas is eliminated, and the regional division of the point cloud data of pillars and supporting facilities is realized, so that the spatial relationship between the pillar center points, key trajectory points and supporting facility search layers is consistent.

[0077] S2, in order to quickly find the key track point data associated with the supporting facilities, build the rotation matrix of the supporting facility clipping box, and quickly lock the associated supporting facility search layer area, it is necessary to build a spatial index for the above three types of typical railway point cloud data. The specific steps are as follows:

[0078] 1) Pillar center point extraction based on neighborhood search:

[0079] In order to facilitate the construction of spatial indexes between pillar center points, key trajectory points, and support facility search layers, and obtain the precise clipping range of support facilities, the center point of the pillar facility point cloud data should be extracted to avoid the significant impact of facilities such as compensation weights and weight limit frames on the clipping range of support facilities. The present invention performs point cloud neighborhood search processing on the pillar search layer area after spatial block processing:

[0080] By setting the neighborhood range CBox with a length and width of 2Δx1, 2Δy1 respectively, and then performing neighborhood search on the pillar search layer, the processing diagram is shown as follows Figure 5a As shown, the statistical density threshold is used to determine whether the point cloud in the neighborhood is a pillar point cloud. If the conditions are met, it is determined to be pillar point cloud data and the pillar center point is calculated. Otherwise, it is discarded. The extraction result is shown in Figure 5b shown.

[0081] 2) After obtaining a series of pillar center point cloud data, due to the existence of local overlapping areas and pillar attachments in the pillar search layer area, pseudo pillar center points appear, such as Figure 5c In order to avoid the influence of pseudo-pillar center points on the extraction of supporting facilities, the present invention uses a statistical filter to filter out pseudo-pillar center points within the neighborhood. The filtering results are shown in Figure 5d shown.

[0082] 3) Constructing a spatial index based on the topological relationship between railway facilities:

[0083] In the railway scene, there is a consistency in the spatial relationship between the pillar center point, key trajectory point and support facility search layer. In order to improve the efficiency of the algorithm in extracting support facilities, the present invention constructs a spatial index based on the topological relationship between railway facilities, so as to achieve the purpose of inversely calculating the coordinates of the trajectory point below and the support facility search layer area based on the coordinates of the pillar center point. The specific steps are as follows: First, by traversing the key trajectory points, the Euclidean distance between each point and the pillar center point is obtained. i ; Then get the index value of the key trajectory point closest to the center point of the pillar by comparisonj , construct the topological relationship between the pillar center point and the key trajectory point; finally, using this point as a link, construct the topological relationship between the pillar center point and the supporting facility search layer above the key trajectory point, thereby realizing the construction of the spatial relationship between the pillar center point, key trajectory point and supporting facility search layer, and obtaining the spatial index. The specific formula is as follows:

[0084] Dis=min(dis i ),i∈R + (3)

[0085]

[0086] Where:

[0087] index j : index value of the center point of the jth pillar;

[0088] min(dis i ): The shortest distance between the center point of the pillar and the key trajectory point.

[0089] S3, due to the distribution uncertainty of railway support facilities and the existence of pillars, contact lines and other facilities directly connected to the support facilities, it is impossible to directly and accurately extract the support facilities based on the existing data. Therefore, the present invention proposes two steps of initial extraction and result optimization to achieve accurate extraction of support facilities. As important erection devices of the contact network, columnar accessories such as contact network pillars and ratchets have the characteristics of being distributed in the vertical direction. Therefore, the present invention combines the point cloud library (PCL) radius filter and the pillar distribution characteristics to propose a column filter, and sets the column filter range according to the coordinates of the pillar center point to process the support facility search layer point cloud data. The specific steps are as follows:

[0090] (1) Support facility rotation and initial extraction:

[0091] The present invention quickly locates the initial extraction area of the supporting facilities associated with the center point of the pillar based on the spatial index, such as Figure 7a As shown in the figure, the cropping range is determined based on the coordinates of the pillar center point, and the coordinates of the key trajectory points associated with it are quickly determined through the index. The obtained cropping range is transformed into three-dimensional coordinates to ensure that the extension direction of the cropping area is consistent with the forward direction of the trajectory. Finally, the cropping range is narrowed according to the difference in the number characteristics of the point clouds on both sides of the pillar. The pillars, loops, ratchets and other auxiliary facilities are filtered out using the column filter to obtain the initial extraction results of the support facilities. The column filter processing diagram is shown in the figure. Figure 6 The processing results are shown as Figure 7b shown.

[0092] (2) Optimization of extraction results by integrating parameterized projection and discretization ideas:

[0093] The present invention proposes a filtering method that combines parameterized model projection with discretization processing to process linear attachments, thereby optimizing the initial extraction results of supporting facilities. Figure 8 As shown in the figure, the initial extraction results still contain irrelevant point cloud data, such as contact line data. To further ensure the accuracy of the extraction results, the present invention performs PCL parametric model projection point cloud filtering on the discretized point cloud data based on the quantitative characteristics and spatial distribution of the contact line point cloud data, filtering out the contact line point cloud. Specifically, the railway point cloud data is first projected onto a planar parametric model; then, the original cropping box is discretized at a set resolution to obtain a series of grids; finally, a threshold is used to determine whether the number of point clouds within the grid should be filtered out.

[0094] Example 1

[0095] Experimental Overview:

[0096] Experimental data was collected using a light railway mobile measurement and scanning system. This system primarily consists of an on-track measurement vehicle and a high-precision Z+F Profile 9012 laser scanner. This device has a line scanning frequency of 50Hz-200Hz, a maximum scanning rate of 1.016 million points per second, an angular resolution of 0.0088°, and an angular accuracy of 0.02° RMS. It can measure distances up to 119 meters with a linear error of less than 1mm. The vehicle travels at a speed of 20-25km / h. The test scenario was selected from a portion of the 170km high-speed railway from Nantong West to Yancheng.

[0097] Extraction of multiple types of supporting facilities:

[0098] Support facility extraction will show different effects due to different types of support facilities, which in turn affects the applicability of the algorithm function. Therefore, testing different types of support facilities in support facility extraction is a key step in verifying the applicability of the algorithm. In order to evaluate the impact of different types of support facilities on the support facility extraction function, this paper selects six types of support facilities for testing. The test data is as follows: Figure 9a The test results are shown in Figure 9b shown.

[0099] Extraction of supporting facilities based on the distribution of multiple types of pillars:

[0100] The extraction of supporting facilities will show different effects due to different pillar distribution characteristics, which in turn affects the extraction accuracy of supporting facilities. In order to evaluate the impact of multiple types of pillar distribution on the extraction accuracy of supporting facilities, the present invention tests three pillar distribution situations: adjacent pillar distribution, symmetrical pillar distribution and asymmetric pillar distribution. The test data are as follows: Figure 10a The test results are shown in Figure 10b shown.

[0101] Demonstration application test:

[0102] When testing real railway facilities, the railway line is often divided into several railway unit data for testing. In order to further verify the applicability of the algorithm of the present invention to multi-contact network large-scale processing, the present invention is tested on the 10-kilometer high-speed railway scene between Nantong West and Yancheng. The test results are as follows: Figure 11 shown.

[0103] The above extraction process demonstrates that the railway vehicle-mounted 3D imaging lidar system can efficiently acquire rich target measurement data, providing a data foundation for support facility extraction. Fully automated support facility extraction takes into account the spatial relationships between pillar centers, key trajectory points, and the support facility search layer. The extraction time for a single support facility is approximately one minute, with MIoU values exceeding 93% and Dice coefficients exceeding 94%. This system can accurately extract support facilities from 10 kilometers of real railway scene data. Its extraction process is highly accurate, widely applicable, and fully automated. Furthermore, it considers support facility point cloud data from multiple scenarios and types, demonstrating strong robustness and application value.

Claims

1. A method for fully automatic extraction of supporting facilities taking into account spatial relationships, comprising the following steps: S1. Obtain key trajectory points by thinning the GNSS trajectory points, and then perform layered and block processing on the original point cloud data to achieve the division of the pillar area and supporting facility area in the original point cloud data; S2. Obtain the pillar center point through neighborhood search, and construct a spatial index based on its spatial relationship with the key trajectory points and the support facility search layer; S3. Using the spatial index as a driving force, the supporting facilities are initially extracted to obtain the supporting facilities containing the contact line, and the contact line point cloud is filtered out by introducing a columnar search and a parameterized projection filter to optimize the extraction results. Wherein, step S2 includes the following sub-steps: 1) By setting the neighborhood range CBox with a length and width of 2Δx1 and 2Δy1 respectively, a neighborhood search is performed on the pillar search layer. The statistical density threshold is used to determine whether the point cloud in the neighborhood is a pillar point cloud. If the conditions are met, it is determined to be pillar point cloud data and its pillar center point is calculated. Otherwise, it is discarded; 2) Perform neighborhood search again based on the initial extraction results of the pillar center points, filter out the pseudo pillar center points in the extraction results, and make each pillar center point correspond to the pillar area point cloud data one by one; 3) By calculating the Euclidean distance between the key trajectory points and the pillar center points, and by comparing the key trajectory point index value with the closest Euclidean distance to each pillar center point, the topological relationship between the pillar center points and the key trajectory points is constructed. Then, using the pillar center points as the link, the topological relationship between the key trajectory points and the supporting facilities above the pillar center points is constructed, realizing the construction of the search layer spatial index of the pillar center points, key trajectory points and supporting facilities. The formula for constructing the spatial index is as follows: Haze=min(haze i ),i∈R + Where: index j : index value of the center point of the jth pillar; min(dis i ): The shortest distance between the center point of the pillar and the key trajectory point.

2. The fully automatic extraction method for supporting facilities according to claim 1, characterized in that: The steps for layering and block processing of the original point cloud data described in step S1 are as follows: 1) Thinning GNSS trajectory points: Divide m GNSS trajectory points into several segments of distance n. After each segment of distance n, the center point of the segment is used as the key trajectory point. Segments with distances less than n / 2 are ignored. The segment calculation formula is as follows: Where: f(p (i,i+n) ): The rarefaction formula at the key points i to i+n; P i : coordinate value of the i-th trajectory point; 2) Layered and block processing of the original point cloud: Based on the key trajectory points, a series of pillar clipping regions V1 and supporting facility clipping regions V2 with lengths, widths, and heights of 2Δx, 2Δy, z2-z1, 2Δx, 2Δy, z4-z3 are constructed respectively. After affine transformation by the rotation matrix Matrix, the original point cloud data is clipped to achieve layering and spatial blocking of the original point cloud data. The relevant formulas are as follows: Where: V0: other irrelevant areas; x i ,y i ,z i :p i Point coordinates.

3. The fully automatic extraction method for supporting facilities according to claim 1, characterized in that: Step S3 includes the following sub-steps: (1) First, the initial extraction area of the supporting facilities associated with the pillar center point is quickly located based on the spatial index, and the clipping range is determined based on the coordinates of the pillar center point. Then, the coordinates of the key trajectory points associated with it are quickly determined through the spatial index, and the clipping range is affine transformed. Finally, the clipping range is narrowed based on the difference in the number characteristics of the point clouds on both sides of the pillar. The pillars, loops, and ratchets are filtered out using a cylindrical filter to obtain the initial extraction results of the supporting facilities. (2) The initial extraction results are projected into a parametric model, and the point cloud data is projected into a two-dimensional plane. The plane is then discretized at a set resolution to obtain a series of grid point cloud data. The quantitative characteristics of the point cloud data in the grid are judged to determine whether the point cloud in the grid should be filtered out. The optimized extraction result is the extracted support facility point cloud data.

4. The fully automatic extraction method for supporting facilities according to claim 3, characterized in that: In step (1), pillars and ratchets are important erection devices of the contact network, which have the characteristics of being distributed in the vertical direction. A column filter is proposed by combining the point cloud library radius filter and the pillar distribution characteristics. The column filter range is set according to the coordinates of the pillar center point to process the support facility search layer point cloud data, and the point cloud data determined to be non-support facilities are filtered out.

5. The fully automatic extraction method for supporting facilities according to claim 3, characterized in that: In step (2), the contact line is an important power supply device of the contact network. It has the characteristics of horizontal distribution and low proportion of point cloud features. The initial extraction results are optimized by combining the PCL parameterized model projection and discretization processing ideas, and the point cloud data determined to be the contact line are filtered out.

Images