Method for obtaining point cloud data of embankments
The method addresses the challenge of distinguishing obstacles from the ground surface in embankment point cloud data by using light intensity frequency distribution and interpolation, achieving precise and efficient ground surface data acquisition.
Patent Information
- Application Number
- JP2025078214
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-05-08
- Publication Date
- 2026-11-19
AI Technical Summary
Existing methods fail to accurately distinguish between obstacles and the ground surface in point cloud data of embankments, particularly in construction management and completion management of railway sites, where high precision is required, such as within plus or minus 15 mm.
A method involving data acquisition, mesh division, mesh extraction, and output steps to remove the influence of obstacles by determining the variation in light intensity frequency distribution and extracting meshes with a predetermined value, followed by interpolating and outputting point cloud data corresponding to the ground surface.
Accurately removes the influence of obstacles and obtains detailed point cloud data of the ground surface, enhancing precision and operational efficiency in construction management.
Smart Images

Figure 2026182194000001_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a method for acquiring point cloud data corresponding to the ground surface of embankment, and particularly relates to a method for removing the influence of obstacles on the ground surface of embankment and acquiring point cloud data corresponding to the ground surface.
Background Art
[0002] Point cloud data used for construction management and finished shape management in civil engineering work is generally acquired by a laser scanner or the like. Here, in order to obtain only the point cloud data corresponding to the ground surface, processing is performed to remove unintended ground objects from the acquired point cloud data.
[0003] For example, in Patent Document 1, as a method for removing unnecessary objects from acquired point cloud data, a point cloud data processing method is disclosed, which acquires ground point cloud data, image data, and position information from a flying object such as a drone, and separates objects and the ground surface from color information (RGB values) and elevation information. It is stated that when the contour from the color information obtained from the image data, the contour from the point cloud data, and the contour line from the elevation information obtained from the position information match, this is determined to be an "object", and image data from which this "object" is removed as an unnecessary object is generated.
Prior Art Documents
Patent Documents
[0004]
Patent Document 1
Disclosure of the Invention
Problems to be Solved by the Invention
[0005] The use of point cloud data is being considered in construction management and completion management of embankments at various railway construction and maintenance sites. In this case, it is necessary to reliably remove obstacles such as workers and heavy machinery from the acquired point cloud data and obtain only the point cloud of the ground surface (top surface). On the other hand, for example, in the completion management value of the subgrade height of a concrete roadbed, a high precision of within plus or minus 15 mm is required. In such cases, it is not possible to distinguish between obstacles and the ground surface by only the difference in height, remove the influence of obstacles, and accurately obtain point cloud data of only the ground surface.
[0006] The present invention has been made in view of the above circumstances, and its purpose is to provide a method for acquiring point cloud data corresponding to the ground surface of an embankment by removing the influence of obstacles (objects) on the ground surface. [Means for solving the problem]
[0007] The present invention provides a method for acquiring point cloud data corresponding to the ground surface of an embankment, comprising: a data acquisition step of obtaining processed point cloud data corresponding to reflected light on the surface including the embankment and objects located thereon; a mesh division step of dividing a horizontal plane including the processing target area on the embankment into meshes; a mesh extraction step of extracting target meshes from the meshes, excluding excluded meshes corresponding to the positions of the objects; and an output step of extracting and outputting point cloud data of point clouds on the target meshes from the processed point cloud data, wherein the mesh extraction step includes a step of determining the variation in the frequency distribution of light intensity for the point clouds on each mesh and extracting the excluded meshes having the variation of a predetermined value or more.
[0008] According to these characteristics, the influence of obstacles (objects) on the ground surface of the embankment can be removed, and point cloud data corresponding to that ground surface can be obtained.
[0009] In the invention described above, the frequency distribution may be characterized by obtaining the number of points in the point group for each class of light intensity and calculating it accordingly. With this characteristic, the presence of obstacles on the ground surface of the embankment can be easily detected.
[0010] In the invention described above, the output step may be characterized by including a step of obtaining the number of points in the point cloud for each elevation class and extracting the elevation position of the ground surface from the mode. With this feature, the position of the ground surface of the embankment can be easily detected.
[0011] In the invention described above, the output step may be characterized by including a step of interpolating and outputting point cloud data corresponding to the point clouds on the excluded mesh. Alternatively, the output step may be characterized by including a step of interpolating between point clouds that are on the excluded mesh and near the location of the ground surface. According to this feature, the influence of obstacles on the ground surface of the embankment can be removed and more detailed point cloud data corresponding to the ground surface can be obtained. [Brief explanation of the drawing]
[0012] [Figure 1] This is a diagram of a system configuration including a point cloud data acquisition method according to the present invention. [Figure 2] (a) A flowchart of the point cloud data acquisition method, and (b) A flowchart of the output step. [Figure 3] (a) A perspective view of an embankment, etc., and (b) Point cloud data of the embankment, etc. [Figure 4] Figure 2(a) shows the area of the embankment, the location for specifying the area of the embankment, and the top surface data of the embankment during the extraction step of the area to be processed. [Figure 5] The images show (a) a top view of the embankment, (b) a side view of the embankment, and (c) a diagram showing the side of the embankment divided into regions, all taken during the mesh division step. [Figure 6]The mesh extraction step shows (a) a diagram indicating the addresses of the top surface of the embankment, (b) the frequency distribution of the reflectance intensity of mesh (1,1), (c) the frequency distribution of the reflectance intensity of mesh (4,1), (d) a top view of the embankment, and (e) a side view of the embankment. [Figure 7] This figure shows (a) a side view of the embankment and (b) the area of the embankment near the ground surface in the output step. [Figure 8] The output steps include (a) a top view of the embankment, (b) a top view of the embankment showing the removed data, and (c) a top view of the embankment showing the interpolated data. [Modes for carrying out the invention]
[0013] Below, a method for acquiring point cloud data as one embodiment of the present invention will be described with reference to Figures 1 to 8.
[0014] As shown in Figure 1, a point cloud data acquisition system 1 for an embankment 50 according to one embodiment of the present invention comprises an unmanned aerial vehicle (UAV) 10, a server 20 that processes information from the UAV 10, and a personal computer 30 that receives information from the server 20. These components are connected via a predetermined communication line or communication network 40.
[0015] The UAV (Unmanned Aerial Vehicle) 10 is a flyable device such as a drone or aircraft equipped with a laser scanner 11. The laser scanner 11 is a sensor device that uses pulsed lasers, such as LiDAR (Light Detection and Ranging), to irradiate a wide area and perform measurements. The laser scanner 11 measures the distance to an object by measuring the time difference between irradiating with pulsed laser light and detecting the reflected light, and obtains the distances to numerous reflection points. The laser scanner 11 also obtains the light intensity, which is the strength of the reflected light at the surface of an object, as the reflection intensity. For example, it obtains data as the reflection intensity at a certain point coordinate (X1, Y1, Z1). In this way, the UAV 10 obtains information about the surrounding space as point cloud data.
[0016] The server 20 is a device that acquires point cloud data obtained by the UAV 10 via a communication network 40 such as the Internet, performs calculations, and outputs the result to the personal computer 30.
[0017] The personal computer 30 acquires the point cloud data processed by the server 20 via the communication network 40 and displays it. A user can issue instructions to the UAV 10 using the personal computer 30. Note that the processing of point cloud data is not limited to being performed on the server 20, and may be performed on the personal computer 30. The personal computer 30 equipped with the processing functions executed by the server 20 may directly communicate with the UAV 10.
[0018] In the following, the algorithm of the point cloud data acquisition method will be described with reference to FIG. 2 and FIGS. 3 to 8.
[0019] FIG. 2(a) is a flowchart of the entire point cloud data acquisition method. As shown therein, the server 20 includes a program that processes a point cloud data acquisition step (S10), a processing target region extraction step (S11), a mesh division step (S12), a mesh extraction step (S13), and an output step (S14) in this order.
[0020] In the point cloud data acquisition step (S10), to-be-processed point cloud data corresponding to reflected light from the surface including the embankment 50 and the object 60 positioned thereon is obtained.
[0021] FIG. 3(a) shows a perspective view of the embankment 50 piled up on the ground surface at various railway construction and maintenance sites. The embankment 50 serves as a base when soil is piled up on land for laying railways or the like to form a flat ground surface. On the embankment 50, there exist objects 60 which are obstacles such as workers and heavy machinery. The UAV 10 defines an arbitrary reference point (not shown) at the construction site, and creates point cloud data based on the relative positional relationship with the reference point, and acquires point cloud data including the embankment 50 to be measured, and the objects 60 such as people and heavy machinery present on the surface of the embankment 50.
[0022] Figure 3(b) shows a conceptual diagram of the point cloud data acquired by the UAV 10. The UAV 10 transmits the measured point cloud data to the server 20 via the communication network 40. The server 20 acquires the point cloud data to be processed, which is the reflected light measured on the surface of the embankment 50 and the object 60 located on it, acquired by the UAV 10. The method of instructing the UAV 10 to fly is not limited to information entered in advance by the user via the PC 30 based on the area to be measured, but may also be dynamically provided by a remote controller (not shown). Furthermore, the UAV 10 is not limited to flying directly over the embankment 50, but may also acquire the point cloud of the surface of the embankment 50 from diagonally above the embankment 50 and perform coordinate transformation.
[0023] Furthermore, the laser scanner 11 is not limited to LiDAR; it may also acquire the shape of the ground surface using millimeter-wave radar, a two-dimensional laser scanner, or a three-dimensional laser scanner. The laser scanner 11 is not limited to being mounted on the UAV 10; it may be fixed in a predetermined location on the ground, or it may be configured to acquire data in three dimensions by moving the laser scanner 11 mounted in a predetermined location. The combination of sensors is arbitrary, and multiple types of sensors may be used.
[0024] Furthermore, the number of UAV10s is not limited to one; there may be multiple. The UAV10s may acquire positional information using GPS (Global Positioning System) or GNSS (Global Navigation Satellite System) data from satellites and use it for measurement. In this case, the UAV10s may acquire positional information in XYZ coordinates, which is the world geodetic system for the embankment 50, with the center of gravity of the Earth as the origin, the direction of the intersection of the Greenwich Meridian and the equator as the X-axis, the direction of 90 degrees east longitude as the Y-axis, and the direction of the North Pole as the Z-axis, and reflect this in the reference coordinate system for the embankment.
[0025] In the step of extracting the processing target area (S11), the processing target area is extracted from the point cloud data acquired in the point cloud data acquisition step (S10).
[0026] Figure 4(a) is a top view showing the data of the embankment 50 measured by UAV10 in a line diagram. Here, the point cloud is omitted, and the position of object 60 is schematically represented. Unless otherwise specified, the same applies to other line diagrams.
[0027] In the coordinate system shown in Figure 4(a), the longitudinal direction of the embankment 50 is the X-axis, the axis perpendicular to the X-axis is the Y-axis, and the axis perpendicular to the XY plane and indicating the height direction is the Z-axis. The user specifies a rectangular area consisting of points a, b, c, and d on the horizontal plane of the embankment 50 via the PC 30 as the processing area.
[0028] Server 20 extracts the point cloud data within the region of the rectangular prism ABCD-A'B'C'D', which includes points a, b, c, and d within the specified region, from the acquired point cloud data to be processed. Point A has the same X and Y coordinates as point a, point B as point b, point C as point c, and point D as point d, and is located above each of these points in terms of their respective Z coordinates. Point A' has the same X and Y coordinates as point a, point B' as point b, point C' as point c, and point D' as point d, and is located below each of these points in terms of their respective Z coordinates. The faces abcd, ABCD, and A'B'C'D' are parallel to each other. The edges constituting the rectangular prism can be of any length as long as they can complement the embankment 50 described later.
[0029] When operating via the personal computer 30, the operation can be performed using either the perspective view of the embankment 50 as shown in Figure 4(a) or the top view of the embankment 50 as shown in Figure 4(b). Figure 4(c) is a schematic diagram of the data obtained by extracting the specified point cloud in this manner.
[0030] In the mesh division step (step S12), the horizontal plane containing the processing area on the embankment 50 is divided into a mesh. In other words, the point cloud data within the rectangular parallelepiped ABCD-A'B'C'D' region containing the processing area on the embankment 50 is divided into small regions on the XY plane and along the Z axis.
[0031] Figure 5(a) is a diagram showing the embankment 50 in Figure 4(a) in the XY plane. Figure 5(b) is a diagram showing the embankment 50 in Figure 4(a) in the XZ plane.
[0032] As shown in Figure 5(a), the server 20 uses the acquired point cloud data to divide the horizontal plane including the processing area on the embankment 50 into a mesh of predetermined lengths along the X and Y axes. In this embodiment, the X-axis is divided into 11 sections and the Y-axis into 6 sections. The length of the division of the horizontal plane including the processing area on the embankment 50 should be a suitable length that allows people and objects to be represented by one or more suitable meshes and that complements the embankment 50, which will be described later.
[0033] As shown in Figure 5(c), the server 20 then uses the acquired point cloud data to divide the point cloud data within the rectangular parallelepiped ABCD-A'B'C'D' region, which includes the processing area on the embankment 50, into a mesh-like structure along the X-axis in the XZ plane with predetermined lengths. In this embodiment, as described above, it is divided into 11 sections along the X-axis. This allows the point cloud data within the rectangular parallelepiped ABCD-A'B'C'D' region, which includes the processing area on the embankment 50, into a total of 66 small rectangular parallelepiped columns, divided into 11 sections along the X-axis and 6 sections along the Y-axis. Note that the number of divisions for the point cloud data within the rectangular parallelepiped ABCD-A'B'C'D' region is not limited to this; any number that can complement the embankment 50, as described later, is acceptable.
[0034] In the mesh extraction step (S13), the meshes to be excluded are extracted.
[0035] Figure 6(a) shows the rectangular prism ABCD-A'B'C'D' from Figure 4(a) represented in the XY plane. The top left is designated as mesh (1,1), the one to its right as mesh (2,1), and so on, with addresses assigned sequentially up to the last mesh in the X-axis direction, starting with mesh (11,1). Addresses are assigned in the same way up to mesh (11,6) in the Y-axis direction.
[0036] Figure 6(b) is a schematic diagram showing the frequency distribution of reflectance intensities for the point cloud within the region of mesh address (1,1). Here, the point cloud is not explicitly described, and the graph schematically represents the collection of point clouds. Unless otherwise specified, the same applies to other diagrams. From the 66 small rectangular regions mentioned above, the point cloud data for the sub-region at mesh address (1,1) is extracted. A graph is created for the extracted point cloud data, with the reflectance intensity at each point on the horizontal axis and the number of points in the point cloud at that reflectance on the vertical axis. P is the peak of the reflectance intensity. The point cloud data for the sub-region at mesh address (1,1) has many data points with reflectance intensities P, indicating that data points with similar reflectance intensities to P are clustered together. Furthermore, the variability of the point cloud data is low, meaning the degree of dispersion is small.
[0037] Figure 6(c) is a schematic diagram showing the frequency distribution of reflectance intensities for the point cloud of the sub-region at mesh address (4,1). Point cloud data for the sub-region at mesh address (4,1) is extracted from the 66 small rectangular regions mentioned above, and a graph is created with the reflectance intensity at each point on the horizontal axis and the number of points in the point cloud within the mesh at that reflectance on the vertical axis. Here, P' is the peak of the reflectance intensity. It can also be seen that the variability of the point cloud data is large, that is, the degree of dispersion is large. It can be seen that the point cloud data for the sub-region at mesh address (4,1) includes not only data around reflectance intensity P', but also data at reflectances outside of reflectance P'.
[0038] Server 20 determines that there is an obstacle such as an object 60 in each mesh if the variation in the point cloud data exceeds a predetermined value. This is because if the variation in the point cloud data exceeds a predetermined value, it means that not only the soil color on the surface of the embankment 50 is included, but also the color of heavy machinery, etc., and therefore an obstacle is considered to exist.
[0039] Figures 6(d) and 6(e) are schematic diagrams showing the determination of whether or not obstacles are present in each mesh.
[0040] Figure 6(d) is a representation of the rectangular parallelepiped ABCD-A'B'C'D' from Figure 6(a) in the XY plane.
[0041] Server 20 determines whether the dispersion of point cloud data for each mesh, from the top-left mesh (1,1) to the bottom-right mesh (11,6), is greater than or equal to a predetermined value. If the dispersion of point cloud data is greater than or equal to the predetermined value, that mesh is determined to have an obstacle such as object 60. If the dispersion of point cloud data is less than the predetermined value, that mesh is determined to not have an obstacle such as object 60. Meshes with obstacles are designated as excluded meshes (E), and meshes without obstacles are designated as target meshes (F).
[0042] In Figures 6(d) and 6(e), the target mesh (F) is the area where all four sides are displayed with thick lines. The excluded mesh (E) is the area that is not included in the target mesh, where all four sides are not displayed with thick lines, and at least one side is displayed with a dotted line.
[0043] Server 20 classifies a total of 66 meshes into excluded meshes (E) and target meshes (F), and extracts 17 target meshes (F) to be processed. The number of meshes extracted varies depending on the area of object 60 and the number of meshes to be divided. Note that in Figures 6(a) and (b), some symbols for excluded meshes (E) and target meshes (F) are omitted in the illustration.
[0044] The threshold for the dispersion of the point cloud data is set to a value that allows for the extraction of meshes containing obstacles such as object 60. In this way, in the mesh extraction step (S13), the variability of the frequency distribution of the light intensity, or reflectance, for the point cloud on each mesh is determined, and meshes with a variability exceeding a predetermined value are extracted as excluded meshes (E). The frequency distribution may also be calculated by obtaining the number of points in the point cloud for each class of light intensity, or reflectance. Specifically, depending on the strength of the reflectance, points with a certain predetermined range of reflectance may be classified as class 1, points with a different predetermined range of reflectance as class 2, and so on, and a table may be created. In other words, the frequency distribution may be calculated using the reflectance values directly or by separating them by class. The number of classes should be selected to a suitable value that can separate obstacles such as object 60 and embankment 50.
[0045] The output step extracts and outputs point cloud data of points on the target mesh from the processed point cloud data (step S14). In other words, from the processed point cloud data acquired by the UAV10, which includes the embankment 50 and the object 60 located on it, data of a small region including the target mesh (F) where it was determined that there are no obstacles such as object 60 is extracted and output.
[0046] The output step, as shown in Figure 2(b), includes a step (step S140) in which the number of points in the point cloud is obtained for each elevation class, and the elevation position of the ground surface is extracted from the mode.
[0047] The diagram and graph in Figure 7(a) show the rectangular prism ABCD-A'B'C'D' from Figure 4(a) represented on the XZ plane (left) and its associated graph (right). Point cloud data is extracted from the target mesh (F) which is determined to be free of obstacles such as object 60, and a frequency distribution is created with the class on the vertical axis and the number of points on the horizontal axis, as shown in the graph.
[0048] The graph in Figure 7(a) is a table showing the number of points for each numerical value that divides the height in the Z-axis direction, which is elevation, into classes. Specifically, depending on the elevation, points in a certain predetermined range are classified as Class 1, points in another predetermined range are classified as Class 2, and so on, and a table is created. The range that constitutes the class is smaller than the length between H1 and H2, and is a suitable value that allows for the extraction of the ground surface. By using elevation classes, it is possible to determine whether or not there is an obstacle with greater accuracy. In the graph, the mode is P''. Here, from the point cloud data of the target mesh (F), data within the range of ±1σ for P'' is extracted.
[0049] As shown in Figure 7(a), the minimum Z coordinate of the extracted data is H1, and the maximum Z coordinate of the extracted data is H2. H1 is located a few centimeters below obstacles such as object 60 in surface BB'C'C. H2 is located a few centimeters above the top surface of embankment 50 in surface BB'C'C. In step S14, only the data with Z coordinates between H1 and H2 are extracted from the target mesh (F) where it was determined that there are no obstacles such as object 60.
[0050] As shown in Figure 7(b), point cloud data (F') of the region near the ground surface enclosed by the thick dotted line is extracted. In this way, the ground surface can be extracted from the point cloud remaining after step S14 based on the elevation frequency distribution. Note that the range from which data is extracted is not limited to the ±1σ range; any range that allows for suitable extraction of the ground surface is acceptable.
[0051] The output step may include a step (step S141) for interpolating and outputting point cloud data corresponding to point clouds located on the excluded mesh (E). The output step may also include a step (step S142) for interpolating between point clouds located on the excluded mesh (E) and near the ground surface.
[0052] Figure 8(a) shows the rectangular parallelepiped ABCD-A'B'C'D' from Figure 4(a) represented on the XY plane. On face abcd, the excluded mesh (E), which contains obstacles such as object 60, is the region enclosed by the thick line.
[0053] Figure 8(b) is a schematic top view of the point cloud data (F') of the area near the ground surface, after the point cloud data of the excluded mesh (E) on surface abcd has been removed. Note that the frame shown with a thick line is included for explanatory purposes and is not included in the point cloud data. The same applies to other diagrams unless otherwise specified.
[0054] Figure 8(c) is a schematic diagram drawn on plane abcd, where data (E') is output and plotted using surrounding point cloud data at the location of the excluded mesh (E) from which point cloud data has been excluded. To calculate the interpolated data (E'), for example, the average or weighted average of the elevation, which is the Z coordinate of the point cloud data (F') in the area near the ground surface, is obtained. The obtained average value is used as the Z coordinate of each point in the excluded mesh (E). This allows point cloud data above the excluded mesh (E) to be interpolated using point cloud data from the area near the ground surface. This makes it possible to remove point clouds corresponding to obstacles such as object 60, and by interpolating the removed point clouds, the accuracy of obstacle removal is improved.
[0055] In this way, data containing obstacles such as object 60 is removed from the point cloud data including the processing area, and the point cloud of only the ground surface of the embankment 50 is extracted and interpolated. This eliminates the need to interrupt work within the measurement range when acquiring the point cloud, thereby improving applicability to the work site and increasing operational efficiency. Furthermore, by removing the point cloud of obstacles at the boundary with the ground surface and interpolating it using surrounding data, the influence of obstacles (objects) on the ground surface of the embankment can be removed, and point cloud data corresponding to that ground surface can be obtained.
[0056] According to the point cloud data acquisition method described above, the influence of obstacles (objects) on the ground surface of the embankment can be removed, and point cloud data corresponding to that ground surface can be accurately acquired.
[0057] Although embodiments and modifications based thereon have been described above, the present invention is not necessarily limited thereto, and those skilled in the art will be able to find various alternative embodiments and modifications without departing from the spirit of the invention or the scope of the attached claims. [Explanation of symbols]
[0058] 1. Point cloud data acquisition system 10 UAV 20 servers 30 PCs 40 Communication Networks 50 Embankment 60 Object
Claims
1. A method for obtaining point cloud data corresponding to the ground surface of an embankment, A data acquisition step to obtain processed point cloud data corresponding to reflected light on the surface including the embankment and objects located thereon, A mesh division step of dividing the horizontal plane including the area to be processed on the embankment into a mesh, A mesh extraction step of extracting target meshes from the mesh, excluding excluded meshes corresponding to the position of the object, The output step includes extracting and outputting point cloud data of point clouds on the target mesh from the processed point cloud data, A method for acquiring point cloud data of an embankment, characterized in that the mesh extraction step includes a step of determining the variation in the frequency distribution of light intensity for each point cloud above the mesh, and extracting the excluded meshes having the variation of a predetermined value or more.
2. The method for acquiring point cloud data of an embankment according to claim 1, characterized in that the frequency distribution is calculated by obtaining the number of points in the point cloud for each class of light intensity.
3. The method for acquiring point cloud data of an embankment according to claim 1 or 2, characterized in that the output step includes obtaining the number of points in the point cloud for each elevation class and extracting the elevation position of the ground surface from the mode.
4. The method for acquiring point cloud data of an embankment according to claim 3, characterized in that the output step includes a step of interpolating and outputting point cloud data corresponding to the point clouds on the excluded mesh.
5. The method for acquiring point cloud data of an embankment according to claim 4, characterized in that the output step includes a step of interpolating between point clouds that are located on the excluded mesh and near the location of the ground surface.
Citation Information
Patent Citations
Unnecessary object removal system, unnecessary object removal method and program
WO2018042551A1
