Railway track line automatic extraction method and system based on laser point cloud

Abstract

The application discloses a railway track line automatic extraction method and system based on laser point clouds, and is characterized in that the method is based on quasi-dense laser point cloud data for discrete sampling data of a track, and utilizes a directional feature window iterative optimization tracking mode to first highlight effective data information of the track, remove nearby noise and irrelevant points, then optimize and fit local track lines through the window, correct a growth direction, iteratively realize extraction of the track lines, and finally extract complete track line elements through left and right track lines. Compared with existing non-contact track line extraction technologies, the method has the characteristics of high performance-price ratio, directly utilizes aerial discrete point cloud data, reduces requirements on point cloud density, and meets requirements of railway engineering in the digital engineering era on efficient and high-quality extraction of existing line elements.

Description

Technical Field

[0001] This invention relates to the field of surveying and mapping technology, specifically to a method and system for automatic extraction of railway tracks based on laser point clouds. Background Technology

[0002] Railway track resurveying and information reconstruction are crucial processes in railway engineering. The reconstructed track information carries a wealth of full-lifecycle geographic information for the railway project, serving as indispensable foundational geographic information for surveying, design, and operation. Traditional railway track extraction techniques still rely heavily on extensive field measurements using RTK (Real-time Kinematic) and total stations, followed by automated fitting and linking processes through office software.

[0003] With advancements in computer-aided technology and aerial remote sensing, aerial non-contact measurement technology has been successfully applied, giving rise to indoor work methods based on low-altitude oblique imagery and laser point clouds. This provides a feasible way to eliminate the impact of work time windows and safety windows on track measurement, thus replacing a large amount of manual fieldwork and shifting more fieldwork to indoor screens. However, it still relies heavily on interactive operations. As the railway industry enters a period of rapid development and network reinforcement, many existing railway lines require regular resurveying and maintenance. Existing methods for reconstructing and extracting existing track lines are clearly insufficient to meet the demands of modern railway construction, facing problems such as excessively long construction periods, low efficiency, high costs, uneven quality, and low cost-effectiveness. This hinders the sustainable growth of subsequent industries related to railway engineering and fails to efficiently guarantee the quality of railway operation, maintenance, and renovation processes. Summary of the Invention

[0004] This application provides a method and system for automatic railway track extraction based on laser point clouds, in order to solve the problem that existing non-contact track extraction technologies rely on a large number of interactive operations and cannot meet the needs of modern railway construction.

[0005] According to a first aspect, one embodiment provides a method for automatic extraction of railway track lines based on laser point clouds, the method comprising:

[0006] The system utilizes airborne drones to collect dense laser point cloud data along railway tracks, preprocesses the collected point cloud data, and crops out the point cloud data for the target track area.

[0007] The starting position of the track line is specified through interactive operation, and an initial track line vector edge is constructed based on the starting position. The direction of the track line vector edge is consistent with the track line mileage direction.

[0008] Establish a long cube window on the edge of the current trajectory vector, acquire the point cloud data within the long cube window, calculate the normal vector and curvature information of the point cloud within the long cube window, and remove noise and irrelevant points based on the calculated point cloud normal vector and curvature information.

[0009] Extract the plane and elevation information within the long cube window after noise and irrelevant points have been removed. Optimize and update the current trajectory vector edges based on the plane and elevation information to correct the direction of trajectory growth, and calculate the local three-dimensional trajectory segments within the current long cube window.

[0010] Based on the optimized current trajectory vector edge, iteratively predict the next trajectory vector edge, or if the iteration requirements are not met, construct the next trajectory vector edge through interactive operation, obtain the local 3D trajectory segment within the cube window of the next trajectory vector edge length, until the trajectory extraction ends;

[0011] The local 3D trajectory segments extracted from the cube window of each trajectory vector side length are sequentially connected and Gaussian smoothed to obtain the entire trajectory extraction result.

[0012] Furthermore, dense laser point cloud data along the railway track is collected using airborne drones. The collected point cloud data is preprocessed, and the point cloud data for the target track area is cropped. Specifically, this includes:

[0013] The acquisition process utilizes an external attitude perception and control system that eliminates or minimizes the need for image control, ensuring that the absolute accuracy of the acquired laser point cloud meets the accuracy requirements for track re-measurement on existing lines, and that the point cloud density is better than 1 / 2 of the width of the track rail surface.

[0014] The collected point cloud data undergoes preprocessing, including data quality checking, storage processing, and data filtering. This preprocessing ensures the quality of the point cloud data and the effectiveness of subsequent trajectory extraction.

[0015] If there are multiple tracks or redundant track intervals in the survey area, the point cloud data that needs to be processed is extracted through spatial clipping. At the same time, the starting direction of the track point cloud data is identified to guide and constrain the starting direction of the track extraction in the later stage.

[0016] Furthermore, a long cube window is established along the current trajectory vector edge, and the point cloud data within the long cube window is acquired, specifically including:

[0017] Based on the track type extracted, construct a long cube window with a width of 2w, where w is the width of the upper surface of the track rail. The length len of the long cube window depends on the maximum slope α of the track and the track height h. The geometric relationship must satisfy: len×α≤1 / 2h.

[0018] Orientation of the rectangular cube window: The generatrix of the rectangular cube window overlaps with the edge of the orbital vector, the upper and lower surfaces of the rectangular cube window are parallel to the orbital plane, and the vertical axis is perpendicular to the orbital line to be extracted.

[0019] The orientation parameters of the rectangular cube window are corrected using the extracted laser point cloud, and a local coordinate system (LOC) is constructed. The length, width, and height of the rectangular cube window correspond to the x, y, and z axes of the LOC, respectively. The center point of the rectangular cube window is the origin of the LOC, and the LOC basis vectors are used to construct the local coordinate system. With the basis vectors of the real external coordinate system (e) x ,e y ,e z The correspondence between the local coordinate system and the external coordinate system is calculated to determine the transformation relationship between the local coordinate system and the true external coordinate system.

[0020] Furthermore, the normal vector and curvature information of the point cloud within the rectangular window are calculated. Based on the calculated point cloud normal vector and curvature information, noise and irrelevant points are removed, specifically including:

[0021] Calculation of normal vector:

[0022] Calculate the spatial normal vector of the current point p in the point cloud. It is through each point and its K neighboring points q i Constructed spatial plane The determined solution is the normal vector of the space plane. The process is shown in the following formula, where the known quantities are the current point p and its K neighboring points q. i :

[0023]

[0024] Solving the above optimization function is equivalent to solving the matrix equation BX. n =0, B is the coefficient of the above equation, X n for The coefficients B are in matrix form. After the coefficients are centroided, principal component analysis (PCA) is used to find the right null space vector corresponding to the smallest eigenvalue of the coefficient matrix.

[0025] Curvature calculation:

[0026] Based on the spatial normal vector of the current point p in the point cloud and the K neighboring points q of the current point p i Space normal vector Calculate the spatial curvature of the current point p, and the normal vector between the neighboring points and the current point. The included angle is α, and the normal vector of the neighboring point is... with the current point normal vector If the angle between the points is β, then the current point p relative to its neighboring point qi The formula for calculating the spatial curvature is as follows:

[0027]

[0028] The formula for calculating the spatial curvature value k of the current point p is as follows:

[0029] k = avrage|k i |

[0030] Among them, the neighboring points q of the current point p are selected. i When selecting, a directional operation window is used. The directional operation window is a spatial ellipsoid, with the current point as the center point of the window. The semi-major axis of the spatial ellipsoid of the directional operation window overlaps with the trajectory.

[0031] Furthermore, the normal vector and curvature information of the point cloud within the rectangular window are calculated. Based on the calculated point cloud normal vector and curvature information, noise and irrelevant points are removed, specifically including:

[0032] Ideally, the normal vector at the midpoint of the track surface is always vertically upward and the curvature information is continuous. However, the normal vector at the horseshoe-shaped edge of the track surface will change abruptly and the curvature will be discontinuous.

[0033] Based on the elevation from large to small, extract the points where the normal vector and curvature change abruptly and calculate the average height ha of the points. To prevent the influence of noise and scanning accuracy, the average height ha is reduced by a point cloud resolution rs, i.e., ha = ha - rs. Points with heights below ha are removed. The removed points are basically those below the horseshoe edge of the track surface.

[0034] Furthermore, the planar and elevation information within the rectangular window after noise and irrelevant point removal is extracted, specifically including:

[0035] Statistical information enhancement is performed on points above height ha in the current working area. Points are sorted in ascending order of elevation. The height value hz at the 97th largest elevation value is extracted, and points with elevation greater than hz are removed, thereby avoiding the influence of the maximum noise fluctuation.

[0036] Using points within the current working area with a height range of [hz ± resolution], the plane line segment and elevation information hl of the points within the working area are obtained by least squares fitting.

[0037] Furthermore, the actual 3D trajectory segment within the current rectangular window is calculated, specifically including:

[0038] By utilizing the relationship between the local coordinate system (LOC) and the real external coordinate system, the real 3D trajectory segment (line3D) is calculated.

[0039] Furthermore, the next trajectory vector edge is predicted iteratively based on the optimized current trajectory vector edge, or the next trajectory vector edge is constructed through interactive operations when the iteration requirements are not met. Specifically, this includes:

[0040] If the number of points in the long cube window constructed by the current trajectory vector edge is insufficient or the point data is not aggregated, the iteration is terminated. At the same time, the interaction determines whether to continue. If to continue, the next trajectory vector edge is constructed through the interaction. Otherwise, the point cloud trajectory extraction process ends.

[0041] Point data non-aggregation typically occurs at track bifurcation points, which can be identified using a horizontal scaling method for elongated cube windows. The specific steps for identifying non-aggregation include: calculating the range of height values ​​[hmax, hmin] distributed along the elevation centerline within the current elongated cube window; scaling the elongated cube window horizontally within this range; and scaling and aggregating nearby points. The scaling and aggregating distance is controlled by a parameter three times the point cloud resolution rs. If the elongated cube window shows a scaling of more than 30%, it indicates an increased probability of other track information near the track line, requiring manual intervention to determine whether to continue iterating and extracting track information.

[0042] Furthermore, the method also includes:

[0043] The left and right track information is automatically extracted, and then the spatial centerline information of the railway is calculated through spatial vertical equidistant constraints. Finally, the mileage information is attached to obtain the final track line extraction result.

[0044] According to a second aspect, one embodiment provides an automatic railway track extraction system based on laser point clouds, the system comprising:

[0045] The point cloud acquisition and processing module is used to collect dense laser point cloud data along the railway track using an airborne drone, preprocess the collected point cloud data, and crop out the point cloud data of the target track area.

[0046] The track vector edge construction module is used to specify the starting position of the track line in the track through interactive operation, and to construct an initial track line vector edge based on the starting position. The direction of the track line vector edge is consistent with the track line mileage direction.

[0047] The effective data extraction module is used to establish a long cube window on the current trajectory vector edge, obtain the point cloud data within the long cube window, calculate the normal vector and curvature information of the point cloud within the long cube window, and remove noise and irrelevant points based on the calculated point cloud normal vector and curvature information.

[0048] The local trajectory extraction module is used to extract the plane and elevation information within the long cube window after noise and irrelevant points have been removed. Based on the plane and elevation information, the current trajectory vector edges are optimized and updated to correct the direction of trajectory growth, and the local three-dimensional trajectory segments within the current long cube window are calculated.

[0049] The iterative extraction module is used to iteratively predict the next trajectory vector edge based on the optimized current trajectory vector edge, or to construct the next trajectory vector edge through interactive operation when the iterative requirements are not met, and to obtain the local 3D trajectory segment within the cube window of the next trajectory vector edge length, until the trajectory extraction ends.

[0050] The local trajectory merging module is used to sequentially connect the local 3D trajectory segments extracted within the cube window of each trajectory vector side length and then perform Gaussian smoothing to obtain the entire trajectory extraction result.

[0051] This application provides a method and system for automatic extraction of railway track lines based on laser point clouds, which has the following features:

[0052] Beneficial effects:

[0053] (1) This invention overcomes the insufficient resolution in trajectory extraction of airborne laser point clouds by introducing a long cubic window mechanism. By using the longitudinal length, more laser point cloud observation data can be captured, thereby accurately determining the precise three-dimensional coordinates of the center point within the long cubic window. This greatly reduces the density requirements of airborne laser scanning and unlocks the potential advantages of airborne office operations.

[0054] (2) The present invention uses a local coordinate system to offset the influence of the trajectory slope on the extraction accuracy, and restores the flatness of the local point cloud under the local coordinate system, thereby extending the length of the long three-dimensional window and the dynamic movement step size.

[0055] (3) This invention is highly automated and has a great efficiency advantage compared with existing interactive operation methods. Attached Figure Description

[0056] Figure 1 A flowchart illustrating an automatic railway track extraction method based on laser point clouds, provided as an embodiment of the present invention;

[0057] Figure 2 A flowchart illustrating the specific implementation of an automatic railway track extraction method based on laser point clouds, as provided in one embodiment of the present invention;

[0058] Figure 3 A schematic diagram of the elongated cubic window structure in an automatic railway track extraction method based on laser point clouds provided in one embodiment of the present invention;

[0059] Figure 4A schematic diagram of the distribution of laser-scanned track point cloud in an automatic railway track extraction method based on laser point cloud provided in one embodiment of the present invention;

[0060] Figure 5 A schematic diagram illustrating the calculation of spatial curvature of a point cloud in an automatic railway track extraction method based on laser point clouds, provided as an embodiment of the present invention;

[0061] Figure 6 A schematic diagram of the point cloud directionality calculation window in an automatic railway track extraction method based on laser point clouds provided in an embodiment of the present invention;

[0062] Figure 7 The image shows the effect of railway track extraction in an automatic railway track extraction method based on laser point cloud, provided in one embodiment of the present invention.

[0063] Figure 8 This is a schematic diagram of the logical structure of an automatic railway track extraction system based on laser point clouds, provided as an embodiment of the present invention. Detailed Implementation

[0064] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of this application. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to this application are not shown or described in the specification. This is to avoid obscuring the core parts of this application with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0065] Furthermore, the features, operations, or characteristics described in the specification can be combined in any suitable manner to form various embodiments. At the same time, the steps or actions in the method description can be rearranged or adjusted in a manner obvious to those skilled in the art. Therefore, the various orders in the specification and drawings are only for the clear description of a particular embodiment and do not imply a necessary order, unless otherwise stated that a particular order must be followed.

[0066] The first embodiment of this invention provides an automatic railway track extraction method based on laser point clouds, aiming to improve the efficiency and accuracy of non-contact track extraction. Addressing the insufficient density of existing airborne laser scanning methods, a track extraction accuracy compensation method is proposed. This method compensates for the insufficient resolution of airborne laser point clouds in track extraction, significantly reducing the density requirements of airborne laser scanning and unlocking the potential advantages of airborne office operations. To address the limitation of the automation process step size by the track slope, a local coordinate system is proposed to offset the impact of track slope on extraction accuracy. Ultimately, a highly automated, high-quality track extraction method based on airborne laser point clouds is achieved. The following section combines... Figure 1 and Figure 2 Please provide a detailed explanation.

[0067] like Figure 1 As shown, in step S100, dense laser point cloud data along the railway track is collected using an airborne UAV, the collected point cloud data is preprocessed, and the point cloud data of the target track area is cropped out.

[0068] The above steps specifically include:

[0069] S110. Utilize airborne UAVs to collect dense point clouds along the railway track. The collection process can employ a minimal or no external attitude perception and control system to ensure that the absolute accuracy of the collected laser point clouds meets the accuracy requirements for track re-measurement on existing lines, and the point cloud density must be greater than 1 / 2 the width of the rail surface. The point cloud files are saved in the universal LAS storage format.

[0070] S120. Point cloud preprocessing includes data quality checking, storage processing, and data filtering. Preprocessing ensures the quality of the point cloud data and the effectiveness of subsequent trajectory extraction. Specific details are as follows:

[0071] S121. Firstly, during the point cloud acquisition process, due to environmental influences, the attitude perception control (POS) information may fail, resulting in a loss of absolute accuracy in local point cloud information. Therefore, it is necessary to utilize the smoothing characteristics of the POS information of the aerial laser scanning acquisition station, use the effective POS information of nearby stations to fit and fill the lost and failed station POS information, and recalculate and export the fused laser point cloud data.

[0072] S122. Secondly, since the point cloud data collected often spans tens or hundreds of kilometers, the massive amount of point cloud data is not conducive to the execution of any spatial analysis or calculation algorithm. Therefore, it is necessary to store and process this massive point cloud. By using the octree technology outside the core, a spatial distribution storage file for the massive point cloud is constructed while taking into account limited computing resources. The massive point cloud is divided and stored in smaller octree leaf nodes. Based on the excellent spatial retrieval capability of the octree, efficient data access capability is provided for subsequent point cloud operation calculations.

[0073] S123. The point cloud data contains a large number of flying points and noise points, so the point cloud data must be filtered.

[0074] S124. In order to further reduce the amount of data involved in the point cloud, the characteristic that laser point clouds have obvious differences in reflection intensity to different scanning media can be used to roughly extract the point cloud near the rail for subsequent calculation steps.

[0075] S130. Obtain the current working area of ​​the point cloud. This mainly reduces the involvement of useless point clouds and speeds up subsequent point cloud processing. Since there may be multiple tracks and redundant track intervals in the survey area, spatial clipping is used to extract the point cloud data that needs to be processed. At the same time, the starting direction of the track point cloud data is identified to guide and constrain the starting direction of subsequent track extraction.

[0076] like Figure 1 As shown, in step S200, the starting position of the track line is specified through interactive operation, and an initial track line vector edge V is constructed based on the starting position. The direction of the track line vector edge is consistent with the track line mileage direction.

[0077] In this embodiment, the starting position of the track is interactively specified to constrain the position and growth direction of subsequent elongated cubic windows. The starting point must be selected on the track surface, and the direction of the starting point must be consistent with the direction of the track. Specific details include:

[0078] S210. Use interactive method to select a pair of points on the surface of the track to form the track vector edge. The direction of the track vector edge must be consistent with the mileage direction. Otherwise, the track inversion and flipping operation needs to be performed after the track is extracted. In addition, the track vector edge should be as close as possible to the track surface and the direction should be accurate and consistent with the track direction.

[0079] S220. The dynamic advancement of the trajectory vector edge is implemented throughout the entire trajectory extraction process. Only the initial trajectory vector edge needs to be specified interactively. The remaining calculation steps automatically generate new trajectory vector edges.

[0080] The specific function of S230 and the track vector edge is mainly to select potential track surface laser scanning points in a directional manner, so that the complex track extraction is decomposed into several short interval windows of operation.

[0081] like Figure 1 As shown, in step S300, a long cube window is established on the edge of the current trajectory vector, the point cloud data within the long cube window is obtained, the normal vector and curvature information of the point cloud within the long cube window are calculated, and noise and irrelevant points are removed based on the calculated point cloud normal vector and curvature information.

[0082] The above steps specifically include:

[0083] S310. Obtain the point cloud within the cube window containing the current trajectory vector. Simultaneously, construct the initial position and shape of the cube window. Correct its orientation parameters using the point cloud within the cube window to facilitate accurate extraction of the local laser point cloud data to be processed. Furthermore, to eliminate the influence of track slope on elevation extraction accuracy, all subsequent point cloud computing steps are performed within the local coordinate system (LOC) constructed using the cube window. The length, width, and height of the cube window correspond to the x, y, and z axes, respectively, and the center point of the cube window is the origin of the LOC. Details are as follows:

[0084] S311. Based on the track type extracted according to the requirements (such as the width w and height h of the upper surface of the track rail), construct a long cubic window. The width of the long cubic window is 2w, and the length len of the long cubic window depends on the maximum slope α of the track and the track height h. The geometric relationship must satisfy: len×α≤1 / 2h.

[0085] S312. Orientation of the elongated cubic window. The generatrix of the elongated cubic window overlaps with the edge of the orbital vector. The top and bottom surfaces of the elongated cubic window are parallel to the orbital plane, and its vertical axis is perpendicular to the orbital line to be extracted. Figure 3 As shown.

[0086] S313. The laser points falling within the elongated cubic window are quickly extracted using the constructed octree data storage structure. Simultaneously, the orientation parameters of the elongated cubic window are corrected using the extracted laser point cloud, and a local coordinate system (LOC) is constructed. The LOC basis vectors are then used to... With the basis vectors of the real external coordinate system (e) x ,e y ,e z The correspondence between the local coordinate system and the external coordinate system is calculated to determine the transformation relationship between the local coordinate system and the true external coordinate system.

[0087] S320. Calculate the normal vector and curvature information of the point cloud within the long cubic window. Simultaneously, construct a directional computation window to improve the reliability of the normal vector and curvature information, which is used for subsequent removal of non-track points. Figure 4 As shown, only by extracting effective laser points can the trajectory information be accurately obtained. Furthermore, by constructing a directional computation window to constrain the calculation of the normal vector and curvature, the geometric characteristics of the track surface information are enhanced.

[0088] S321. Calculation of the normal vector:

[0089] Calculate the spatial normal vector of the current point p in the point cloud. It is through each point and its K neighboring points q i Constructed spatial plane The determined solution is the normal vector of the space plane. The process is shown in the following formula, where the known quantities are the current point p and its K neighboring points q. i :

[0090]

[0091] Solving the above optimization function is equivalent to solving the matrix equation BX. n =0, B is the coefficient of the above equation, X n for The coefficients B are in matrix form. After the coefficients are centroided, principal component analysis (PCA) is used to find the right null vector corresponding to the smallest eigenvalue of the coefficient matrix.

[0092] S322, Calculation of curvature:

[0093] Based on the spatial normal vector of the current point p in the point cloud and the K neighboring points q of the current point p i Space normal vector Calculate the spatial curvature of the current point p, such as Figure 5 As shown, the normal vectors of neighboring points and the current point are... The included angle is α, and the normal vector of the neighboring point is... with the current point normal vector If the angle between the points is β, then the current point p relative to its neighboring point q i The formula for calculating the spatial curvature is as follows:

[0094]

[0095] The formula for calculating the spatial curvature value k of the current point p is as follows:

[0096] k = avrage|k i |

[0097] S323. Constructing a directional calculation window. To suppress the influence of lateral scattered points on the track surface on the calculation of curvature and normal vector, a directional calculation window is used when selecting the neighborhood points of the current point. This window is a spatial ellipsoid, with the current point as its center. The semi-major axis of the spatial ellipsoid of the directional calculation window overlaps with the track alignment, as shown below. Figure 6 As shown. The advantage of the directional operation window is that it can greatly enhance the geometric properties of the orbital points, making it easier to remove non-orbital points later.

[0098] S330. Remove non-rail surface laser scan points. Based on the previous steps, calculate the normal vector and curvature of all points within the cubic window of the rail vector side length. Ideally, the normal vector of the midpoint of the rail surface is always vertically upward, and the curvature information is relatively continuous; however, the normal vector of points at the horseshoe edge of the rail surface will change abruptly, and the curvature will be discontinuous; the normal vector and curvature of points at other locations have no regularity. Based on this pattern, calculate the height of the horseshoe edge of the rail surface, extract the rail surface laser scan, and remove non-rail surface laser scan points.

[0099] S331. Based on the elevation from large to small, extract the points where the normal vector and curvature change abruptly and calculate the average height ha of the points. In order to prevent the influence of noise and scanning accuracy, the average height ha is reduced by a point cloud resolution rs, that is: ha = ha - rs.

[0100] S332. Remove points with a height below ha. The removed points are basically those below the horseshoe-shaped edge of the track surface.

[0101] like Figure 1 As shown, in step S400, the plane and elevation information within the long cube window after noise and irrelevant points are removed is extracted. The current trajectory vector edge is optimized and updated based on the plane and elevation information to correct the direction of trajectory growth, and the local three-dimensional trajectory segment within the current long cube window is calculated.

[0102] The above steps specifically include:

[0103] S410 Extract the plane and elevation information within the cubic window (current working area) of the trajectory vector side length.

[0104] S411. Statistical information enhancement is performed on points above height ha in the current working area. The points are sorted in ascending order of elevation. The height value hz at the 97th highest elevation value is extracted, and points with elevation greater than hz are removed to avoid the influence of maximum noise fluctuation.

[0105] S412. Using points within the current working area with a height range of [hz ± resolution], use least squares to fit the plane line segments and elevation information hl of the points within the working area.

[0106] S420. Update the trajectory vector edges. Since the current trajectory vector edges are initialized edges and deviate from the actual situation, update the current trajectory vector edges V using the planar line segment (line) and elevation information (hl) calculated in the previous step. Since the planar line segment (line) and elevation information (hl) are in the local coordinate system (LOC), and the required 3D trajectory is data in the real coordinate system, it is also necessary to use the relationship between the local coordinate system (LOC) and the real external coordinate system to calculate the real 3D trajectory segment (line3D).

[0107] like Figure 1 As shown, in step S500, the next trajectory vector edge is predicted iteratively based on the optimized current trajectory vector edge, or the next trajectory vector edge is constructed through interactive operation when the iteration requirements are not met, and the local three-dimensional trajectory segment within the cube window of the next trajectory vector edge length is obtained until the trajectory extraction ends.

[0108] Iteratively predict the next trajectory vector edge: Predict the next trajectory vector using the optimized trajectory vector edge from the previous step, and transform the next trajectory vector into the current trajectory vector. If the data iteration requirements are met, iteratively execute steps S300 to S400; otherwise, restart or stop the iteration through interactive operation according to step S200.

[0109] The data iteration requirements are not met. If the number of points in the long cubic window constructed from the current track vector edges is insufficient, or if the point data is not aggregated, the iteration is terminated. An interactive check is then performed to determine whether to continue. If to continue, step S200 is re-executed; otherwise, the point cloud track extraction process ends. Point data non-aggregation typically occurs at track bifurcation points (turnouts), which can be identified by the horizontal growth method of the long cubic window.

[0110] Point data aggregation is not used for discrimination. The height range [hmax, hmin] of the current long cubic window distributed along the elevation centerline is calculated. Within the height range of [hmax, hmin], the long cubic window is horizontally grown left and right, and nearby points are aggregated during the growth. The distance control parameter for growth aggregation is 3 times the point cloud resolution rs. If the long cubic window shows more than 30% growth points, it indicates that the probability of other track information near the track line has increased. It is necessary to manually determine whether to continue iterating to extract track information through interactive methods.

[0111] like Figure 1 As shown, in step S600, the local three-dimensional trajectory segments extracted within the cube window of each trajectory vector side length are sequentially connected and Gaussian smoothed to obtain the extraction result of the entire trajectory.

[0112] Extracting track geometry and deriving the track line: Connecting the 3D points (line3D) extracted from each long cubic window forms the track line. To ensure the smoothness of the track line information, Gaussian smoothing can be applied to adjacent 3D line segments to improve the final smoothness of the track line. Similarly, this method can be used to extract the left and right track line information simultaneously, then calculate the railway spatial centerline information through spatial vertical equidistant constraints, and finally append the mileage information.

[0113] Using the method of this invention, trajectory information was extracted from an airborne laser point cloud. The original point cloud data was 11.5 GB in size, with a scanning resolution of 2 cm. The extracted trajectory information is as follows: Figure 7 As shown, the left and right lines and the center line of the track can be automatically extracted. The extraction process is fully automated, which proves and highlights the advantages of the method of the present invention.

[0114] This invention fully incorporates the advantages and disadvantages of existing technologies. After analyzing the characteristics and engineering requirements of aerial laser scanning, it proposes an automatic track extraction technology based on laser point cloud data. This technology uses discrete sampling data of the track from quasi-dense laser point cloud data and employs an iterative optimization tracking method with directional feature windows. First, it highlights the effective track data information and eliminates nearby noise and irrelevant points. Then, it optimizes and fits local track lines through windowing, corrects the growth direction, and iteratively extracts the track lines. Finally, it extracts the complete track line elements from the left and right track lines. Compared to existing non-contact track extraction technologies, this invention offers high cost-effectiveness. By directly utilizing aerial discrete point cloud data, it reduces the requirement for point cloud density and meets the requirements of efficient and high-quality extraction of existing line elements in railway engineering during the era of digital and intelligent engineering.

[0115] Corresponding to the above-disclosed method for automatic railway track extraction based on laser point clouds, this invention also discloses an automatic railway track extraction system based on laser point clouds, such as... Figure 8 As shown, it specifically includes:

[0116] The point cloud acquisition and processing module is used to collect dense laser point cloud data along the railway track using an airborne drone, preprocess the collected point cloud data, and crop out the point cloud data of the target track area.

[0117] The track vector edge construction module is used to specify the starting position of the track line in the track through interactive operation, and to construct an initial track line vector edge based on the starting position. The direction of the track line vector edge is consistent with the track line mileage direction.

[0118] The effective data extraction module is used to establish a long cube window on the current trajectory vector edge, obtain the point cloud data within the long cube window, calculate the normal vector and curvature information of the point cloud within the long cube window, and remove noise and irrelevant points based on the calculated point cloud normal vector and curvature information.

[0119] The local trajectory extraction module is used to extract the plane and elevation information within the long cube window after noise and irrelevant points have been removed. Based on the plane and elevation information, the current trajectory vector edges are optimized and updated to correct the direction of trajectory growth, and the local three-dimensional trajectory segments within the current long cube window are calculated.

[0120] The iterative extraction module is used to iteratively predict the next trajectory vector edge based on the optimized current trajectory vector edge, or to construct the next trajectory vector edge through interactive operation when the iterative requirements are not met, and to obtain the local 3D trajectory segment within the cube window of the next trajectory vector edge length, until the trajectory extraction ends.

[0121] The local trajectory merging module is used to sequentially connect the local 3D trajectory segments extracted within the cube window of each trajectory vector side length and then perform Gaussian smoothing to obtain the entire trajectory extraction result.

[0122] It should be noted that for a detailed description of the automatic railway track extraction system based on laser point cloud provided in the embodiments of the present invention, please refer to the relevant description of the automatic railway track extraction method based on laser point cloud provided in the embodiments of this application, which will not be repeated here.

[0123] In addition, embodiments of the present invention also provide an electronic device, the device comprising: a processor and a memory; the memory being used to store one or more program instructions; the processor being used to execute one or more program instructions to perform the steps of the automatic railway track extraction method based on laser point clouds as described in any of the preceding embodiments.

[0124] It should be noted that for a detailed description of the electronic device provided in the embodiments of the present invention, please refer to the relevant description of the automatic railway track extraction method based on laser point cloud provided in the embodiments of this application, which will not be repeated here.

[0125] In addition, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the automatic railway track extraction method based on laser point clouds as described in any of the preceding claims.

[0126] It should be noted that for a detailed description of the computer-readable storage medium provided in the embodiments of the present invention, please refer to the relevant description of the automatic railway track extraction method based on laser point cloud provided in the embodiments of this application, which will not be repeated here.

[0127] Those skilled in the art will understand that all or part of the functions of the various methods in the above embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, which may include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to achieve the above functions. For example, the program can be stored in the memory of a device, and when the program in the memory is executed by the processor, all or part of the above functions can be achieved. In addition, when all or part of the functions in the above embodiments are implemented by computer programs, the program can also be stored in a server, another computer, disk, optical disk, flash drive, or external hard drive, etc., and can be downloaded or copied to the memory of a local device, or the system of the local device can be updated. When the program in the memory is executed by the processor, all or part of the functions in the above embodiments can be achieved.

[0128] The above examples illustrate the present invention only to aid in understanding it and are not intended to limit the scope of the invention. Those skilled in the art can make various simple deductions, modifications, or substitutions based on the ideas of this invention.

Claims

1. A method for automatic extraction of railway track lines based on laser point clouds, characterized in that, The method includes: The system utilizes airborne drones to collect dense laser point cloud data along railway tracks, preprocesses the collected point cloud data, and crops out the point cloud data for the target track area. The starting position of the track line is specified through interactive operation, and an initial track line vector edge is constructed based on the starting position. The direction of the initial track line vector edge is consistent with the track line mileage direction. Establish a long cube window on the edge of the current trajectory vector, acquire the point cloud data within the long cube window, calculate the normal vector and curvature information of the point cloud within the long cube window, and remove noise and irrelevant points based on the calculated point cloud normal vector and curvature information. Extract the plane and elevation information within the long cube window after noise and irrelevant points have been removed. Optimize and update the current trajectory vector edges based on the plane and elevation information to correct the direction of trajectory growth, and calculate the local three-dimensional trajectory segments within the current long cube window. Based on the optimized current trajectory vector edge, iteratively predict the next trajectory vector edge, or if the iteration requirements are not met, construct the next trajectory vector edge through interactive operation, obtain the local 3D trajectory segment within the cube window of the next trajectory vector edge length, until the trajectory extraction ends; The local 3D trajectory segments extracted from the cube window of each trajectory vector side length are sequentially connected and Gaussian smoothed to obtain the entire trajectory extraction result.

2. The method for automatic extraction of railway track lines based on laser point clouds as described in claim 1, characterized in that, Dense laser point cloud data along railway tracks is collected using airborne drones. The collected point cloud data is preprocessed, and point cloud data for the target track area is cropped. Specifically, this includes: The acquisition process utilizes an external attitude perception and control system that requires minimal or no image control, ensuring that the absolute geographic accuracy of the acquired laser point cloud meets the accuracy requirements for track resurvey on existing lines, and that the point cloud density is better than 1 / 2 of the width of the track rail surface. The collected point cloud data undergoes preprocessing, including data quality checking, storage processing, and data filtering. This preprocessing ensures the quality of the point cloud data and the effectiveness of subsequent trajectory extraction. If there are multiple tracks or redundant track intervals in the survey area, the point cloud data that needs to be processed is extracted through spatial clipping. At the same time, the starting direction of the track point cloud data is identified to guide and constrain the starting direction of the track extraction in the later stage.

3. The method for automatic extraction of railway track based on laser point clouds as described in claim 1, characterized in that, Create a long cube window along the current trajectory vector edge, and acquire the point cloud data within the long cube window, specifically including: Based on the track type extracted, construct a long cube window with a width of 2w, where w is the width of the upper surface of the track rail. The length len of the long cube window depends on the maximum slope α of the track and the track height h. The geometric relationship must satisfy: len×α≤1 / 2h. Orientation of the rectangular cube window: The generatrix of the rectangular cube window overlaps with the edge of the orbital vector, the upper and lower surfaces of the rectangular cube window are parallel to the orbital plane, and the vertical axis is perpendicular to the orbital line to be extracted. The orientation parameters of the rectangular cube window are corrected using the extracted laser point cloud, and a local coordinate system (LOC) is constructed. The length, width, and height of the rectangular cube window correspond to the x, y, and z axes of the LOC, respectively. The center point of the rectangular cube window is the origin of the LOC, and the LOC basis vectors are used to construct the local coordinate system. basis vectors of the real external coordinate system The correspondence between them is used to calculate the transformation relationship between the local coordinate system and the real external coordinate system.

4. The method for automatic extraction of railway track based on laser point clouds as described in claim 1, characterized in that, Calculate the normal vector and curvature information of the point cloud within the rectangular window. Based on the calculated point cloud normal vector and curvature information, remove noise and irrelevant points, specifically including: Ideally, the normal vector at the midpoint of the track surface is always vertically upward and the curvature information is continuous. However, the normal vector at the horseshoe-shaped edge of the track surface will change abruptly and the curvature will be discontinuous. Based on the elevation from large to small, extract the points where the normal vector and curvature change abruptly and calculate the average height ha of the points. To prevent the influence of noise and scanning accuracy, the average height ha is lowered by a point cloud resolution rs. Points with heights below ha are removed. The removed points are all below the horseshoe edge of the track surface.

5. The method for automatic extraction of railway track lines based on laser point clouds as described in claim 4, characterized in that, Extract the planar and elevation information within the rectangular window after noise and irrelevant point removal, specifically including: Statistical information enhancement is performed on points above height ha in the current working area. Points are sorted in ascending order of elevation. The height value hz at the 97th percentile of the elevation value is extracted. Points with elevation greater than hz are removed to avoid the influence of the maximum noise fluctuation. Using points within the current working area with a height range of [hz±rs], the plane line segment and elevation information hl of the points within the working area are obtained by least squares fitting.

6. The method for automatic extraction of railway track based on laser point clouds as described in claim 1, characterized in that, Calculate the actual 3D trajectory segment within the current rectangular cube window, specifically including: By utilizing the relationship between the local coordinate system (LOC) and the real external coordinate system, the real 3D trajectory segment (line3D) is calculated.

7. The method for automatic extraction of railway track based on laser point clouds as described in claim 1, characterized in that, The next trajectory vector edge is predicted iteratively based on the optimized current trajectory vector edge, or the next trajectory vector edge is constructed through interactive operations when the iteration requirements are not met. Specifically, this includes: If the number of points in the long cube window constructed by the current trajectory vector edge is insufficient or the point data is not aggregated, the iteration is terminated. At the same time, the interaction determines whether to continue. If to continue, the next trajectory vector edge is constructed through the interaction. Otherwise, the point cloud trajectory extraction process ends. Point data non-aggregation occurs at track bifurcation points, which is identified by horizontally growing a rectangular window. The specific steps for identifying non-aggregation include: calculating the height range [hmax, hmin] of the points distributed along the elevation centerline within the current rectangular window; horizontally growing the rectangular window within this range; aggregating nearby points; and controlling the aggregating distance to be three times the point cloud resolution rs. If the rectangular window shows an increase of more than 30% in points, it indicates an increased probability of other track information near the track line, requiring manual intervention to determine whether to continue iterating and extracting track information.

8. The method for automatic extraction of railway track based on laser point clouds as described in claim 1, characterized in that, The method further includes: The left and right track information is automatically extracted, and then the spatial centerline information of the railway is calculated through spatial vertical equidistant constraints. Finally, the mileage information is attached to obtain the final track line extraction result.

9. An automatic railway track extraction system based on laser point clouds, characterized in that, The system includes: The point cloud acquisition and processing module is used to collect dense laser point cloud data along the railway track using an airborne drone, preprocess the collected point cloud data, and crop out the point cloud data of the target track area. The track vector edge construction module is used to specify the starting position of the track line in the track through interactive operation, and to construct an initial track line vector edge based on the starting position. The direction of the initial track line vector edge is consistent with the track line mileage direction. The effective data extraction module is used to establish a long cube window on the current trajectory vector edge, obtain the point cloud data within the long cube window, calculate the normal vector and curvature information of the point cloud within the long cube window, and remove noise and irrelevant points based on the calculated point cloud normal vector and curvature information. The local trajectory extraction module is used to extract the plane and elevation information within the long cube window after noise and irrelevant points have been removed. Based on the plane and elevation information, the current trajectory vector edges are optimized and updated to correct the direction of trajectory growth, and the local three-dimensional trajectory segments within the current long cube window are calculated. The iterative extraction module is used to iteratively predict the next trajectory vector edge based on the optimized current trajectory vector edge, or to construct the next trajectory vector edge through interactive operation when the iterative requirements are not met, and to obtain the local 3D trajectory segment within the cube window of the next trajectory vector edge length, until the trajectory extraction ends. The local trajectory merging module is used to sequentially connect the local 3D trajectory segments extracted within the cube window of each trajectory vector side length and then perform Gaussian smoothing to obtain the entire trajectory extraction result.

Images