A tunnel over-under excavation intelligent detection method based on three-dimensional laser scanning technology

Abstract

A tunnel over-under excavation intelligent detection method based on three-dimensional laser scanning technology relates to the field of tunnel engineering construction. Input the tunnel three-dimensional laser scanning point cloud data set collected on site, the algorithm automatically performs fine-grained processing and training learning, automatically extracts data features, obtains classification model parameters, and serves as a fixed value of the machine learning algorithm model. Each three-dimensional point cloud model is divided into point cloud slices, the point cloud label information is compared, the same feature domain point cloud is merged, the manifold is introduced as a constraint condition, and the tunnel section over-under excavation three-dimensional laser scanning model is reconstructed by means of the surface construction method. The present application intelligently processes massive point cloud data, so that the constructed tunnel surface as a whole satisfies the Euclidean space characteristics, guarantees the local characteristics of the tunnel section model, more accurately reflects the real space form of the tunnel, realizes the accurate discrimination of over-under excavation in the tunnel section construction process, and provides strong technical support for the safe construction of the tunnel.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering construction, specifically to an intelligent detection method for tunnel over-excavation and under-excavation based on three-dimensional laser scanning technology. Background Technology

[0002] Over the past 20 years, my country has achieved remarkable results in tunnel engineering construction. Tunnel engineering is characterized by complex geological conditions, numerous adverse geological formations, tight schedules, and heavy workloads. Due to limitations imposed by the surrounding rock geology and construction technology, over-excavation or under-excavation of tunnels is unavoidable, resulting in poor excavation quality on site. This not only significantly impacts the quality, safety, and progress of tunnel construction but also affects the project's economic benefits.

[0003] Excessive over-excavation in tunnels is a major cause of wasted construction costs and concrete consumption. Traditional methods for detecting over-excavation and under-excavation and calculating concrete volume are mostly based on total stations or cross-section instruments. These methods measure the excavation profile through single-point sampling, calculate the average over-excavation value, and then back-calculate the volume. However, these traditional "point-based" surveying methods are inefficient and may suffer from insufficient data representativeness. Three-dimensional laser scanning technology offers advantages such as high scanning speed, high scanning frequency, high point cloud density, and high modeling accuracy. However, 3D laser scanning data is characterized by its disorder, rotation-invariant nature, and unstructured nature, resulting in large data volume, high noise levels, and high density. While this provides convenience for model reconstruction and modification, it also presents new challenges.

[0004] Tunnels are complex geometric bodies with a three-dimensional spatial structure, and their contour surfaces are uneven after construction and excavation. In current methods for reconstructing tunnel over- and under-excavation surfaces, point cloud segmentation suffers from the inability to directly determine the segmentation depth and breadth, the constructed tunnel surface model does not satisfy Euclidean space properties, local feature information of the tunnel cross-section model is lost, and the three-dimensional laser scanning tunnel model cannot accurately reflect the true shape of the tunnel contour in various directions. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide an intelligent detection method for tunnel over-excavation and under-excavation that combines 3D laser scanning technology with machine learning. This method effectively solves the problem of uncertain depth and breadth of 3D point cloud segmentation, intelligently processes massive point cloud data, ensures that the constructed tunnel surface as a whole satisfies Euclidean space characteristics, and protects the local features of the tunnel cross-section model. It more accurately reflects the true spatial morphology of the tunnel, enabling precise identification of over-excavation and under-excavation during tunnel cross-section construction, and providing strong technical support for safe tunnel construction.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A smart detection method for tunnel over-excavation and under-excavation based on three-dimensional laser scanning technology includes the following steps:

[0008] Step 1: Collect point cloud data of the entire tunnel using a 3D laser scanner, establish a laser scanning image model of the tunnel cross section, and perform preliminary filtering and noise reduction processing;

[0009] Step 2: By optimizing the loss function in the machine learning method PointNet++, a point cloud segmentation algorithm with adaptive classification discrimination is proposed;

[0010] Step 3: Input the 3D laser scanning point cloud dataset of the tunnel collected on site. The algorithm will automatically perform fine-grained processing and training, and automatically extract data features.

[0011] Step 4: Obtain the classification model parameters and use them as fixed values ​​for the machine learning algorithm model. Divide each 3D point cloud model into point cloud patches, compare the point cloud label information, and merge the neighborhood point clouds with the same features.

[0012] Step 5: Introduce manifolds as constraints and reconstruct the three-dimensional laser scanning model of the tunnel cross-section (over-excavation and under-excavation) using surface construction methods;

[0013] Step 6: Output data and images of tunnel over-excavation and under-excavation.

[0014] Step 1 includes:

[0015] Step 1.1: Using a high-speed laser scanning measurement method, 100,000 points per second with a ranging error of 2cm are obtained quickly and over a large area to acquire the three-dimensional coordinates of the tunnel surface, forming a point cloud dataset.

[0016] Step 1.2: Perform preliminary filtering on the collected raw point cloud data, using statistical filtering or radius filtering methods to remove discrete points that are too far from the main point cloud group;

[0017] Step 1.3: Filter out non-target point clouds with abnormally low or high reflection intensity using laser reflection intensity information;

[0018] Step 1.4: If multi-station scanning is required to cover the entire tunnel, high reflectivity targets or feature points are deployed in the tunnel as common control points; using control points or feature-based point cloud registration algorithms, the point cloud data scanned by each station is accurately converted to a unified tunnel global coordinate system.

[0019] Step 1.5: Convert the registered point cloud data to the tunnel design coordinate system or construction survey coordinate system to ensure that the point cloud location information is comparable to the design model and construction layout data. This step is completed using the known control point coordinates.

[0020] Step 1.6: The point cloud dataset processed in steps 1.2, 1.3, 1.4, and 1.5 is initially organized according to mileage or cross-section location to form a tunnel 3D laser point cloud model with a unified coordinate system and preliminary noise suppression, which can be processed by subsequent algorithms.

[0021] In step 2, the loss function is: The proposed point cloud segmentation algorithm with adaptive classification and discrimination is based on the PointNet++ algorithm. It extracts point cloud features step by step from a multi-level structure through the principle of hierarchical sampling. The sampling method is multi-scale sampling or multi-resolution sampling.

[0022] In step 3, the tunnel 3D laser scanning point cloud dataset collected on site is packaged into n×3 data groups and input into the proposed point cloud segmentation algorithm with adaptive classification and discrimination. The algorithm automatically performs fine-grained processing and training. During training, the point cloud is automatically segmented and classified according to the classification effect required, and data features are automatically extracted to obtain model parameters.

[0023] Step 4 includes:

[0024] Step 4.1: After training with the machine learning algorithm, obtain the classifier model parameters and use them as fixed values ​​for the machine learning algorithm model to establish a machine learning model for point cloud segmentation and classification.

[0025] Step 4.2: According to the point cloud segmentation algorithm with adaptive classification discrimination, each 3D point cloud model is segmented into point cloud patches with topological features and labeled with label information. These point cloud patches have appropriate widths and can be described using the linear geometry of basis functions.

[0026] Step 4.3: According to the point cloud segmentation algorithm with adaptive classification discrimination, the segmented point clouds are classified according to a certain topological feature including concavity and curvature. The point cloud label information is compared and the neighborhood point clouds with the same feature are merged.

[0027] Step 5 includes: in Euclidean space, the coordinates of three-dimensional points x, y, and z are arbitrary r-dimensional vectors; the manifold constraints satisfy the mathematical condition of differentiability; the manifold surface has good smoothness; and the mathematical expression is symmetric, bilinear, and positive definite; the expression for the manifold constraints is: The mapping relationship between point clouds and surfaces is established by constructing a surface pool. The surface construction method function is as follows: Where S is the complete surface of the 3D model of the tunnel outline. For a surface represented by a known function, For surfaces that are repaired using other methods.

[0028] Step 6 includes: obtaining the coordinates, mileage station number, and over- or under-excavation values ​​of any point in the three-dimensional laser scanning model of the tunnel cross-section, quickly indicating the under-excavation location of the tunnel, completing the determination of the under-excavation location on site, generating an over- or under-excavation volume calculation table, and generating an over- or under-excavation image.

[0029] The beneficial effects of this invention are:

[0030] ① This paper addresses the challenge of uncertain depth and breadth in point cloud segmentation. By optimizing the loss function of PointNet++, an adaptive classification mechanism is proposed, overcoming the limitation of traditional point cloud segmentation that relies on manually setting the segmentation level. The algorithm automatically identifies local geometric features of the point cloud (such as curvature and concavity / convexity), dynamically adjusts the segmentation granularity and merging strategy, significantly improving its adaptability to complex tunnel surfaces and avoiding over-segmentation or under-segmentation.

[0031] ② Achieve efficient and intelligent processing of massive point clouds. Combining multi-scale sampling technology with fine-grained automatic feature extraction, it can process hundreds of thousands of point cloud data per second. Compared with traditional total station single-point sampling detection which only covers 5%-10% of the cross-section, this invention achieves 100% full cross-section coverage, improving detection efficiency by more than 10 times, and is not limited by tunnel length, making it suitable for full-cycle construction monitoring of long tunnels.

[0032] ③ Ensure the mathematical completeness of the tunnel surface model. Manifold constraints are introduced to ensure that the reconstructed surface strictly conforms to the characteristics of Euclidean space. The surface maintains geometric invariance under arbitrary coordinate system transformations, and the millimeter-level concave and convex deformation details of the over-excavated / under-excavated areas are accurately preserved, avoiding the feature smoothing loss problem in existing methods.

[0033] ④ Improved accuracy in identifying over-excavation and under-excavation areas drives intelligent management and control of tunnel construction. Three-dimensional laser scanning technology enables real-time dynamic monitoring, rapid full-section scanning and analysis of the tunnel, accurate identification of over-excavation areas, and automatic early warning for under-excavation areas, avoiding the risk of collapse caused by insufficient gaps in the support structure. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the detection method of the present invention;

[0035] Figure 2 This is a schematic diagram illustrating the principle of the point cloud segmentation algorithm with adaptive classification discrimination in an embodiment of the present invention;

[0036] Figure 3 This is a schematic diagram illustrating the PointNet feature extraction principle in an embodiment of the present invention;

[0037] Figure 4 This is a schematic diagram of the point cloud segmentation algorithm process for adaptive classification and discrimination in this invention. Detailed Implementation

[0038] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0039] Various aspects of the invention are described in this disclosure with reference to the accompanying drawings, which illustrate numerous illustrative embodiments. The embodiments of this disclosure are not necessarily intended to encompass all aspects of the invention. It should be understood that the various concepts and embodiments described above, as well as those described in more detail below, can be implemented in any of many ways, because the concepts and embodiments disclosed herein are not limited to any particular implementation. Furthermore, some aspects of the invention disclosed may be used alone or in any suitable combination with other aspects of the invention disclosed.

[0040] Example

[0041] See Figure 1 As shown, this embodiment of the invention provides an intelligent detection method for tunnel over-excavation and under-excavation based on three-dimensional laser scanning technology, including:

[0042] Step 1: Collect point cloud data of the entire tunnel using a 3D laser scanner to establish a laser scan image model of the tunnel cross-section, and perform preliminary filtering and noise reduction processing. Specifically, this includes:

[0043] Step 1.1: Using a high-speed laser scanning measurement method, 100,000 points per second with a ranging error of 2cm are obtained quickly and over a large area to acquire the three-dimensional coordinates of the tunnel surface, forming a point cloud dataset.

[0044] Step 1.2: Perform preliminary filtering on the collected raw point cloud data, using statistical filtering or radius filtering methods to remove discrete points that are too far from the main point cloud group;

[0045] Step 1.3: Use the laser reflection intensity information to filter out non-target point clouds with abnormally low or high reflection intensity.

[0046] Step 1.4: If multi-station scanning is required to cover the entire tunnel, deploy high-reflectivity targets or feature points within the tunnel as common control points. Use control points or feature-based point cloud registration algorithms to accurately transform the point cloud data scanned by each station into a unified global tunnel coordinate system.

[0047] Step 1.5: Convert the registered point cloud data to the tunnel design coordinate system or construction survey coordinate system to ensure that the point cloud location information is comparable to the design model and construction layout data. This step is completed using known control point coordinates.

[0048] Step 1.6: The point cloud dataset processed in steps 1.2, 1.3, 1.4, and 1.5 is initially organized according to mileage or cross-section location to form a tunnel 3D laser point cloud model with a unified coordinate system and preliminary noise suppression, which can be processed by subsequent algorithms.

[0049] Step 2: By optimizing the loss function in the PointNet++ machine learning method, a point cloud segmentation algorithm with adaptive classification is proposed. The loss function is... , combined Figure 2 As shown in Figure 4, in this step, a point cloud segmentation algorithm with adaptive classification and discrimination is proposed. The specific process is shown in Figure 4. The algorithm is based on the PointNet++ algorithm. Through the principle of hierarchical sampling, it gradually extracts point cloud features from multi-level (multi-layer) structures. The sampling method is multi-scale sampling or multi-resolution sampling.

[0050] Step 3: Input the 3D laser scan point cloud dataset of the tunnel collected on-site. The algorithm automatically performs fine-grained processing and training, and automatically extracts data features. Combined with... Figure 3 As shown, the input is a 3D laser scanning point cloud dataset of the tunnel collected on-site. The algorithm automatically performs fine-grained processing and training, and automatically extracts data features.

[0051] Step 4: Obtain the classification model parameters and use them as fixed values ​​for the machine learning algorithm model. Segment each 3D point cloud model into point cloud patches, compare the point cloud label information, and merge neighborhood point clouds with the same features. Combined with... Figure 1 As shown, it specifically includes:

[0052] Step 4.1: After training with the machine learning algorithm, obtain the classifier model parameters and use them as fixed values ​​for the machine learning algorithm model to establish a machine learning model for point cloud segmentation and classification.

[0053] Step 4.2: According to the point cloud segmentation algorithm with adaptive classification discrimination, each 3D point cloud model is segmented into point cloud patches with topological features and labeled with label information. These point cloud patches have appropriate widths and can be described using the linear geometry of basis functions.

[0054] Step 4.3: According to the point cloud segmentation algorithm with adaptive classification discrimination, the segmented point clouds are classified according to a certain topological feature, such as concavity / convexity, curvature, etc. The point cloud label information is compared and the neighborhood point clouds with the same feature are merged.

[0055] Step 5: Introduce manifolds as constraints and reconstruct the 3D laser scanning model of the tunnel cross-section (over-excavation / under-excavation) using surface construction methods. In this step, manifolds are introduced as constraints, and the 3D laser scanning model of the tunnel cross-section (over-excavation / under-excavation) is reconstructed using surface construction methods. In Euclidean space, the coordinates x, y, and z of the three-dimensional points are arbitrary r-dimensional vectors. The manifold constraints satisfy the mathematical condition of differentiability, the manifold surface has good smoothness, and the mathematical expression has symmetry, bilinearity, and positive definiteness. The expression for the manifold constraints is: The mapping relationship between point clouds and surfaces is established by constructing a surface pool. The surface construction method function is as follows: Where S is the complete surface of the 3D model of the tunnel outline. For a surface represented by a known function, For surfaces that are repaired using other methods.

[0056] Step 6: Output of over-excavation and under-excavation data and images of the tunnel. Obtain the coordinates, mileage markers, and over-excavation / under-excavation values ​​of any point in the 3D laser scanning model of the tunnel cross-section. Quickly indicate the location of under-excavation, complete the on-site determination of the under-excavation location, generate a calculation table for over-excavation / under-excavation volume, and generate over-excavation / under-excavation images.

[0057] This invention is not limited to the embodiments described above. Those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications are also considered within the scope of protection of this invention. Contents not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A method for intelligent detection of tunnel over-excavation and under-excavation based on three-dimensional laser scanning technology, characterized in that, Includes the following steps: Step 1: Collect point cloud data of the entire tunnel using a 3D laser scanner, establish a laser scanning image model of the tunnel cross section, and perform preliminary filtering and noise reduction processing; Step 2: Optimize the loss function in the PointNet++ machine learning method A point cloud segmentation algorithm with adaptive classification discrimination is proposed; Step 3: Input the 3D laser scanning point cloud dataset of the tunnel collected on site. The algorithm will automatically perform fine-grained processing and training, and automatically extract data features. Step 4: Obtain the classification model parameters and use them as fixed values ​​for the machine learning algorithm model. Segment each 3D point cloud model into point cloud patches, compare the point cloud label information, and merge neighborhood point clouds with the same features; specifically including: Step 4.1: After training with the machine learning algorithm, obtain the classifier model parameters and use them as fixed values ​​for the machine learning algorithm model to establish a machine learning model for point cloud segmentation and classification. Step 4.2: According to the point cloud segmentation algorithm with adaptive classification discrimination, each 3D point cloud model is segmented into point cloud patches with topological features and labeled with label information. These point cloud patches have appropriate widths and can be described using the linear geometry of basis functions. Step 4.3: According to the point cloud segmentation algorithm with adaptive classification discrimination, the segmented point cloud is classified according to a certain topological feature including concavity and curvature, the point cloud label information is compared, and the neighborhood point clouds with the same feature are merged. Step 5: Introduce manifolds as constraints and reconstruct the three-dimensional laser scanning model of the tunnel cross-section (over-excavation and under-excavation) using surface construction methods; Step 6: Output data and images of tunnel over-excavation and under-excavation.

2. The intelligent detection method for tunnel over-excavation and under-excavation based on three-dimensional laser scanning technology according to claim 1, characterized in that, Step 1 includes: Step 1.1: Using a high-speed laser scanning measurement method, 100,000 points per second with a ranging error of 2cm are obtained quickly and over a large area to acquire the three-dimensional coordinates of the tunnel surface, forming a point cloud dataset. Step 1.2: Perform preliminary filtering on the collected raw point cloud data, using statistical filtering or radius filtering methods to remove discrete points that are too far from the main point cloud group; Step 1.3: Filter out non-target point clouds with abnormally low or high reflection intensity using laser reflection intensity information; Step 1.4: If multi-station scanning is required to cover the entire tunnel, high reflectivity targets or feature points are deployed in the tunnel as common control points; using control points or feature-based point cloud registration algorithms, the point cloud data scanned by each station is accurately converted to a unified tunnel global coordinate system. Step 1.5: Convert the registered point cloud data to the tunnel design coordinate system or construction survey coordinate system to ensure that the point cloud location information is comparable to the design model and construction layout data. This step is completed using the known control point coordinates. Step 1.6: The point cloud dataset processed in steps 1.2, 1.3, 1.4, and 1.5 is initially organized according to mileage or cross-section location to form a tunnel 3D laser point cloud model with a unified coordinate system and preliminary noise suppression, which can be processed by subsequent algorithms.

3. The intelligent detection method for tunnel over-excavation and under-excavation based on three-dimensional laser scanning technology according to claim 1, characterized in that, In step 2, point cloud features are extracted step by step from the multi-level structure using the principle of hierarchical sampling. The sampling method is multi-scale sampling or multi-resolution sampling.

4. The intelligent detection method for tunnel over-excavation and under-excavation based on three-dimensional laser scanning technology according to claim 1, characterized in that, In step 3, the tunnel 3D laser scanning point cloud dataset collected on site is packaged into n×3 data groups and input into the proposed point cloud segmentation algorithm with adaptive classification and discrimination. The algorithm automatically performs fine-grained processing and training. During training, the point cloud is automatically segmented and classified according to the classification effect required, and data features are automatically extracted to obtain model parameters.

5. The intelligent detection method for tunnel over-excavation and under-excavation based on three-dimensional laser scanning technology according to claim 1, characterized in that, Step 5 includes: in Euclidean space, the coordinates of three-dimensional points x, y, and z are arbitrary r-dimensional vectors; the manifold constraints satisfy the mathematical condition of differentiability; the manifold surface has good smoothness; and the mathematical expression is symmetric, bilinear, and positive definite; the expression for the manifold constraints is: The mapping relationship between point clouds and surfaces is established by constructing a surface pool. The surface construction method function is as follows: Where S is the complete surface of the 3D model of the tunnel outline. For a surface represented by a known function, For surfaces that are repaired using other methods.

6. The intelligent detection method for tunnel over-excavation and under-excavation based on three-dimensional laser scanning technology according to claim 1, characterized in that, Step 6 includes: obtaining the coordinates, mileage station number, and over- or under-excavation values ​​of any point in the three-dimensional laser scanning model of the tunnel cross-section, quickly indicating the under-excavation location of the tunnel, completing the determination of the under-excavation location on site, generating a calculation table of over- or under-excavation volume, and generating an over- or under-excavation image.

Images