A tunnel lining space detection and extraction method based on multi-scale space clustering

Abstract

The application relates to a tunnel point cloud processing technical field and relates to a tunnel lining space detection and extraction method based on multi-scale space clustering, which comprises the following steps: S1, dividing the structural features of a tunnel point cloud scene; S2, extracting tunnel point clouds of a lining structure through a space structure detection mode; S3, segmenting tunnel point clouds with lining structures and projecting along a tunnel direction; S4, constructing a global clustering rule to screen the projected tunnel point clouds; and S5, constructing a local clustering rule to further screen lining point clouds. The detection and extraction method can improve the extraction precision of tunnel lining point clouds and improve the robustness, reliability and applicability of the tunnel lining extraction method.

Description

Technical Field

[0001] This invention relates to the field of tunnel point cloud processing technology, and in particular to a method for spatial detection and extraction of tunnel lining based on multi-scale spatial clustering. Background Technology

[0002] In recent years, with the rapid development of my country's road network, the number of tunnels has continued to increase. Due to their large capital investment, complex excavation environment, and high degree of concealment, tunnel construction demands high precision. Three-dimensional tunnel models, with their intuitive and high-precision advantages, provide a practical digital solution for tunnel construction and structural monitoring throughout the tunnel's life cycle. Therefore, extracting tunnel lining structure data from the three-dimensional spatial data of the tunnel scene plays a crucial role in tunnel construction and deformation monitoring. Laser point cloud data is the fundamental data source for constructing high-precision three-dimensional geographic scenes. The data production principle involves measuring instruments actively acquiring accurate three-dimensional information by emitting laser beams towards the target object. In tunnel construction surveying, three-dimensional laser scanners are characterized by high precision and efficiency, suitable for detecting information inside tunnels with insufficient lighting. However, the original data contains many noise points, such as non-tunnel lining points like auxiliary facilities, bolt holes, workers, pipes, and tracks. Removing these non-lining points and retaining the remaining lining point cloud is essential for providing a basis for tunnel structural deformation analysis. Therefore, accurate extraction of the tunnel lining is a key issue to be addressed in tunnel structural analysis. Currently, the method for extracting tunnel lining point clouds involves using the tunnel's central axis and setting the angle between the tunnel cross-section point cloud and the center point of the cross-section to extract the lining point cloud on the tunnel cross-section. If a method for extracting the tunnel lining as a whole is adopted, a cylindrical or elliptical cylindrical model is used to fit the tunnel structure as a whole. The fitting process involves setting the variance of the distance from the tunnel point cloud to the tunnel's central axis to extract the overall tunnel lining point cloud. Another method for extracting lining point clouds involves changing the original tunnel structure and using the tunnel's central axis to project and transform the tunnel point cloud onto a two-dimensional plane before extracting the tunnel lining point cloud. Considering existing lining extraction methods, the central axis extraction method is easily affected by tunnel orientation, point cloud pores, and point cloud noise, leading to deviations in the extraction accuracy of the central axis. While flat point clouds can reduce the difficulty of tunnel lining extraction, the tunnel point cloud, after projection transformation, will affect the lining extraction accuracy to some extent, especially for tunnel lining structures with uneven surfaces. At the same time, research on lining extraction methods for tunnels of different shapes has not yet formed a standardized tunnel lining extraction method. Summary of the Invention

[0003] To overcome the shortcomings in the background technology and solve the existing technical problems, this invention discloses a spatial detection and extraction method for tunnel lining based on multi-scale spatial clustering, which can improve the extraction accuracy of tunnel lining point clouds and enhance the robustness, reliability and applicability of the tunnel lining extraction method.

[0004] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: A method for spatial detection and extraction of tunnel lining based on multi-scale spatial clustering includes the following steps: S1, dividing the structural features of the tunnel point cloud scene; S2, extracting the tunnel point cloud with lining structure through spatial structure detection; S3, segmenting the tunnel point cloud with lining structure and projecting it along the tunnel direction; S4, constructing global clustering rules to filter the projected tunnel point cloud; S5, constructing local clustering rules to further filter out the lining point cloud.

[0005] Furthermore, in S1, based on the scene characteristics of the tunnel point cloud, the spatial structure corresponding to the tunnel lining is determined; if the tunnel shape belongs to a circular tunnel, an arched tunnel, or a horseshoe-shaped tunnel, the overall tunnel lining structure is identified as a cylinder, while the lining structure of a rectangular tunnel is identified as a structure spliced ​​from planar structures; the spatial structure of non-lining points is extracted by combining cylindrical, spherical, and planar spatial structures.

[0006] Furthermore, S2 includes the following steps: S21, detecting the spatial structure of the tunnel scene, the detection results include three different spatial structures: cylindrical, spherical, and planar; S22, setting the spatial structure that makes up the tunnel lining: ψ L When the tunnel shape is approximately cylindrical, the candidate point set for the cylindrical surface structure is: D ψ If no new candidate point set is generated from the sample points, the point set with the largest number of points within the cylindrical candidate point set is selected as the point cloud of the lining structure; when the tunnel shape is approximately rectangular, the candidate point set for the planar structure is: D ψ The point set with the largest number of candidate points for the planar structure is the point cloud of the lining structure: S ψ The extraction process of the lining structure can be expressed as: S23. Remove the point set of non-lining structures to obtain the preliminary tunnel lining point cloud.

[0007] Furthermore, S3 includes the following steps: S31, projecting the tunnel point cloud onto a horizontal plane and extracting the boundary lines of the tunnel point cloud using the Alpha Shapes algorithm; S32, segmenting the rough lining point cloud at equal intervals along the boundary lines in the tunnel direction to obtain tunnel lining point cloud slices; S33, projecting the rough lining point cloud slices onto a two-dimensional plane perpendicular to the tunnel direction, and using the projected tunnel point cloud for multi-scale tunnel point cloud clustering and screening.

[0008] Furthermore, S4 includes the following steps: S41, randomly select a sample point from the projected point cloud, determine whether the number of points adjacent to the sample point within its radius ε meets the threshold N; if it exceeds the threshold, the point is identified as a core point F, otherwise it is a density reachable point D; S42, traverse the density reachable points of the core point within the threshold range, and generate clustering results C with different point cloud densities by summarizing the density reachable points. The clustering result can be expressed by the above formula, where i is the type of clustering result and R is the point set represented by a certain clustering result; S43, calculate the bounding box of the clustering result and abstract different clustering results as the center point of the bounding box; the center point corresponding to the tunnel lining cluster is called the external point, and the center point corresponding to the non-lining point cluster is called the internal point; S44, by setting the azimuth angle of the cluster center, eliminate the spatial clustering formed by internal noise.

[0009] Furthermore, S5 includes the following steps: S51, expanding the globally filtered cluster points from the original two-dimensional space to a three-dimensional space; S52, calculating the three-dimensional bounding box corresponding to the cluster points after spatial expansion; S53, dividing the three-dimensional space of the bounding box at equal intervals along the coordinate axis direction. If the spatial segmentation yields an independent rectangular space corresponding to the noise point, the loop ends; otherwise, spatial segmentation continues until the three-dimensional space where the independent noise point is located is obtained; S54, deleting the three-dimensional spatial grid where the noise point is located to obtain the lining cluster points that do not contain noise points; and outputting the final lining extraction result.

[0010] By employing the technical solution described above, the present invention has the following beneficial effects: The tunnel lining spatial detection and extraction method based on multi-scale spatial clustering disclosed in this invention fills the gap of no standardized method for tunnel lining extraction in different tunnel shapes and construction scenarios. This extraction method can serve the tunnel structure monitoring in different construction phases. The tunnel lining spatial detection and extraction method based on multi-scale spatial clustering disclosed in this invention can extract as many tunnel lining points as possible by taking into account the spatial clustering characteristics of tunnel points in the global and local contexts, avoiding damage to the original data features, and improving the extraction accuracy of tunnel lining point cloud. The tunnel lining spatial detection and extraction method disclosed in this invention, based on multi-scale spatial clustering, does not rely on the tunnel's central axis during the extraction process. Therefore, it can avoid lining extraction errors caused by inaccurate central axes, further improving the robustness and reliability of the tunnel lining extraction method. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a schematic diagram of the tunnel scene structure proposed in this invention; Figure 3 This is a schematic diagram of the multi-scale clustering screening proposed in this invention; Figure 4 This is a diagram showing the extraction results of the point cloud of the tunnel lining according to the present invention. Detailed Implementation

[0012] The technical solution of the present invention will now be described with reference to the accompanying drawings in the embodiments of the present invention: Combined with appendix Figure 1-4 The aforementioned method for spatial detection and extraction of tunnel lining based on multi-scale spatial clustering includes the following steps: Step 1: Determine the structural features of the tunnel point cloud scene; Based on the scene characteristics of the tunnel point cloud, the spatial structure corresponding to the tunnel lining is determined. If the tunnel shape is a circular tunnel, an arched tunnel, or a horseshoe-shaped tunnel, the overall tunnel lining structure is identified as a cylinder, while the lining structure of a rectangular tunnel is identified as a structure spliced ​​together from planar structures. The spatial structure of non-lining points is extracted by combining cylindrical, spherical, and planar spatial structures.

[0013] To verify the applicability of this method, the specific implementation process involved acquiring point cloud data of five different tunnel shapes using ground-based laser scanning equipment and a mobile laser scanning system: circular tunnels, arched tunnels, horseshoe-shaped tunnels, rectangular tunnels, and irregular tunnels. The tunnel scenarios mainly included two scenarios: the completion phase and the excavation phase. The completion phase scenario included tunnel lining, tunnel auxiliary facilities, bolt holes, workers, pipelines, and tracks (non-lining points). The excavation phase scenario included tunnel lining and features present in the completion phase, as well as building materials such as exposed rebar. Furthermore, specialized equipment used during tunnel construction, such as lining trolleys, were also characteristic elements in the tunnel excavation scenario. To accurately extract lining points from the point cloud scenario, structural analysis of the tunnel scene's components was required. If the tunnel shape was circular, arched, or horseshoe-shaped, the overall tunnel lining structure was considered a cylinder; the lining of a rectangular tunnel was considered a structure composed of planar structures joined together. Figure 2 As shown, in addition to the tunnel lining, the tunnel point cloud scene also includes non-tunnel lining points such as auxiliary facilities, bolt holes, workers, pipes, tracks, building materials and construction equipment. Since the spatial structure of non-lining points is quite diverse, mainstream spatial structures such as cylindrical surfaces, spherical surfaces and planes are used for extraction.

[0014] Step 2: Extract the tunnel point cloud of the lining structure using spatial structure detection methods; The specific steps are as follows: First, the spatial structure of the tunnel scene is detected, and the detection results include three different spatial structures: cylindrical, spherical, and planar. Then, the spatial structure that makes up the tunnel lining is set: ψ LWhen the tunnel shape is approximately cylindrical, the candidate point set for the cylindrical surface structure is: D ψ If no new candidate point set is generated from the sample points, the point set with the largest number of points within the cylindrical candidate point set is selected as the point cloud of the lining structure; when the tunnel shape is approximately rectangular, the candidate point set for the planar structure is: D ψ The point set with the largest number of candidate points for the planar structure is the point cloud of the lining structure: S ψ The extraction process of the lining structure can be expressed as: Finally, the point set of non-lining structures is removed to obtain the preliminary tunnel lining point cloud.

[0015] During implementation, the predefined parameters for spatial structure detection include: the normal deviation angle α at the point where the detected structure is located; the minimum point size n of the defined shape; the probability value p that there is no better candidate among all samples; the minimum point size of the shape is set to 50; for different tunnel shapes, the sampling resolution of the distance is usually set to a ratio of three. Through multiple sets of comparative experiments, the range of the normal deviation angle is set to 20° to 30°, and the extraction results of the tunnel lining point cloud are the best. First, the tunnel space exploration process is as follows: Figure 1 As shown, three spatial detection structures (cylindrical, spherical, and planar) are used to detect the spatial structure of point clouds of tunnels of different shapes in different scenarios. At the same time, the spatial detection structure type of the tunnel lining is determined by combining the spatial structural characteristics of the tunnel lining. Then, the type of spatial detection structure to which the point set belongs is determined by calculating the normal vector corresponding to the set minimum number of point clouds. Next, the point set of spatial exploration is selected based on the normal deviation angle to screen out candidate structures that meet the conditions; Finally, candidate structures that meet the requirements of tunnel lining are selected, and the candidate structure with the most point cloud data is found among these candidate structures. The candidate structure with the most point cloud data is output to obtain the coarse point cloud extraction result of the lining.

[0016] Step 3: Segment the point cloud of the tunnel with lining structure and project it along the tunnel direction; The specific steps are as follows: First, as... Figure 3 As shown, the tunnel point cloud is projected onto a horizontal plane, and the boundary lines of the tunnel point cloud are extracted using the Alpha Shapes algorithm. Then, along the boundary lines in the tunnel direction, the rough lining point cloud is segmented at equal intervals to obtain tunnel lining point cloud slices. Then, the rough lining point cloud slices are projected onto a two-dimensional plane perpendicular to the tunnel direction, and the projected tunnel point cloud is used for multi-scale tunnel point cloud clustering and screening.

[0017] Step 4: Construct global clustering rules to filter the projected tunnel point cloud; The specific steps are as follows: First, randomly select a sample point from the projected point cloud, determine its radius ε, and check if the number of points adjacent to the sample point meets the threshold N. If it exceeds the threshold, the point is identified as a core point F; otherwise, it is a density-reachable point D. Next, iterate through the density-reachable points of the core point within the threshold range, and generate clustering results C with different point cloud densities by summarizing the density-reachable points. The clustering results can be expressed by the above formula, where i is the type of clustering result and R is the point set represented by a certain clustering result. Then, the bounding box of the clustering results is calculated, and different clustering results are abstracted as the center point of the bounding box. The center point corresponding to the tunnel lining cluster is called the external point, and the center point corresponding to the non-lining point cluster is called the internal point. Finally, by setting the azimuth angle of the cluster center, the spatial clustering formed by internal noise is eliminated.

[0018] During implementation, the input parameters of the clustering screening module include the segmentation distance, minimum value, and search radius. The clustering screening module will find the core points and cluster points from the high-density area. The minimum value parameter is set to 50. Since the spatial accuracy of the point cloud data is high, in order to better extract the supporting cluster points, the example selected 0.04 meters as the search radius between adjacent sample points.

[0019] Step 5: Construct local clustering rules to further filter out the lining point cloud; The specific steps are as follows: First, the cluster points after global filtering are expanded from the original two-dimensional space to three-dimensional space; then, the bounding box of the three-dimensional space corresponding to the expanded cluster points is calculated; next, the three-dimensional space of the bounding box is divided at equal intervals along the coordinate axis directions. In this embodiment, the bounding box is divided into equal-interval two parts along the three coordinate system directions. If the spatial segmentation yields an independent rectangular space corresponding to the noise point, the loop ends; otherwise, the spatial segmentation continues until the three-dimensional space where the independent noise point is located is obtained; finally, the three-dimensional space grid where the noise point is located is deleted to obtain the lining cluster points that do not contain noise points; the final lining extraction result is output. The lining extraction result of this embodiment is as follows: Figure 4 As shown, the tunnel lining points were largely preserved.

[0020] The parts of this invention not described in detail are prior art. It will be apparent to those skilled in the art that this invention is not limited to the details of the above exemplary embodiments, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the above embodiments should be regarded as exemplary and non-limiting in all respects. The scope of this invention is defined by the appended claims rather than the foregoing description. Therefore, it is intended to include all changes that fall within the meaning and scope of the equivalents of the claims within this invention, and no reference numerals in the claims should be regarded as limiting the content of the claims.

Claims

1. A spatial detection and extraction method for tunnel lining based on multi-scale spatial clustering, characterized in that: it includes... The following steps: S1. Defining the structural features of the tunnel point cloud scene; S2. Extract the tunnel point cloud of the lining structure through spatial structure detection; S3. Segment the point cloud of tunnels with lining structures and project it along the tunnel direction; S4. Construct a global clustering rule to filter the projected tunnel point cloud; Includes the following steps: S41. Randomly select a sample point from the projected point cloud, determine its radius ε, and check whether the number of points adjacent to the sample point meets the threshold N; if it exceeds the threshold, the point is identified as the core point F, otherwise it is the density reachable point D. S42. Traverse the density reachable points of the core points within the threshold range, and generate clustering results C with different point cloud densities by summarizing the density reachable points; The clustering result can be expressed by the above formula, where i is the type of clustering result and R is the set of points represented by a certain clustering result; S43. Calculate the bounding box of the clustering results and abstract different clustering results as the center point of the bounding box; the center point corresponding to the tunnel lining cluster is called the outer point, and the center point corresponding to the non-lining point cluster is called the inner point. S44. By setting the azimuth angle of the cluster center, spatial clustering caused by internal noise is eliminated; S5. Construct local clustering rules to further filter out the lining point cloud; Includes the following steps: S51. Extend the cluster points after global filtering from the original two-dimensional space to three-dimensional space; S52. Calculate the 3D bounding box of the cluster points after spatial expansion; S53. Divide the three-dimensional space of the bounding box at equal intervals along the coordinate axis. If the spatial segmentation yields an independent rectangular space corresponding to the noise point, the loop ends; otherwise, the spatial segmentation continues until the three-dimensional space where the independent noise point is located is obtained. S54. Delete the 3D spatial grid where the noise points are located to obtain the lining cluster points that do not contain noise points; output the final lining extraction result.

2. The method for spatial detection and extraction of tunnel lining based on multi-scale spatial clustering according to claim 1, characterized in that: in In S1, the spatial structure corresponding to the tunnel lining is determined based on the scene characteristics of the tunnel point cloud. If the tunnel shape is a circular tunnel, an arched tunnel, or a horseshoe-shaped tunnel, the overall tunnel lining structure is identified as a cylinder, while the lining structure of a rectangular tunnel is identified as a structure spliced ​​together from planar structures; the spatial structure of non-lining points is extracted by combining cylindrical, spherical, and planar spatial structures.

3. The tunnel lining spatial detection and extraction method based on multi-scale spatial clustering according to claim 2, characterized in that: S2 includes the following steps: S21. Detect the spatial structure of the tunnel scene. The detection results include three different spatial structures: cylindrical, spherical, and planar. S22. Setting up the spatial structure that makes up the tunnel lining: ψ L When the tunnel shape is approximately cylindrical, the candidate point set for the cylindrical surface structure is: D ψ If no new candidate point set is generated from the sample points, the point set with the largest number of points in the cylindrical candidate point set is selected as the lining structure point cloud. When the tunnel shape is approximately rectangular, the candidate point set for the planar structure is: D ψ The point set with the largest number of candidate points for the planar structure is the point cloud of the lining structure: S ψ The extraction process of the lining structure can be expressed as: S23. Remove the point set of non-lining structures to obtain the preliminary tunnel lining point cloud.

4. The tunnel lining spatial detection and extraction method based on multi-scale spatial clustering according to claim 3, characterized in that: S3 includes the following steps: S31. Project the tunnel point cloud onto the horizontal plane and use the Alpha Shapes algorithm to extract the boundary lines of the tunnel point cloud; S32. Along the boundary line of the tunnel direction, the rough lining point cloud is divided at equal intervals to obtain tunnel lining point cloud slices. S33. The rough lining point cloud slices are projected onto a two-dimensional plane perpendicular to the tunnel direction. The projected tunnel point cloud is then used for multi-scale tunnel point cloud clustering and screening.

Images