A three-dimensional digital profile detection system for excavating a roadway

Abstract

This invention relates to a three-dimensional digital contour detection system for tunneling roadways, belonging to the field of roadway contour measurement technology. The system includes: a data acquisition module for acquiring raw point cloud data using a three-dimensional scanning device; a data preprocessing module for preprocessing the raw point cloud data to obtain actual point cloud data, constructing design point cloud data, and aligning the actual point cloud data with the design point cloud data to obtain standard point cloud data; an over- or under-excavation area division module for dividing over- or under-excavation areas according to the design point cloud data; a plane selection module for fitting multiple random planes and selecting multiple initial candidate planes; a roadway size calculation module for classifying five actual planes and calculating the roadway dimensions based on the five actual planes; and a visualization and storage module for visually displaying the roadway contour information. This invention reduces the error in tunneling roadway contour measurement and improves the efficiency and real-time performance of roadway contour construction.

Description

Technical Field

[0001] This invention relates to the field of tunnel contour measurement technology, and in particular to a three-dimensional digital contour detection system for tunneling tunnels. Background Technology

[0002] In coal mine tunneling, over-excavation, under-excavation, and dimensional changes in the tunnel have a significant impact on project safety and tunneling efficiency. Detecting the contour of the tunnel can promptly identify over-excavation, under-excavation, and dimensional changes.

[0003] Traditional methods for detecting the contours of tunnels mainly rely on manual measurement, which suffers from low efficiency, large errors, and poor real-time performance. While some systems have emerged capable of scanning tunnels with the development of 3D laser scanning technology, most suffer from complex data processing, insufficient real-time performance, and inconvenient historical data retrieval, resulting in poor accuracy of the detected tunnel contours. Summary of the Invention

[0004] To solve the above-mentioned technical problems, this invention provides a three-dimensional digital contour detection system for tunneling roadways. The technical solution of this invention is as follows:

[0005] A three-dimensional digital contour detection system for tunnel excavation includes:

[0006] The data acquisition module is used to acquire the original point cloud data of the tunnel through a three-dimensional scanning device pre-deployed on the roof of the tunnel, and to record the acquisition time of the original point cloud data. The three-dimensional scanning device is equipped with a lidar.

[0007] The data preprocessing module is used to preprocess the raw point cloud data to obtain the actual point cloud data. It constructs the design point cloud data according to the design dimensions of the tunnel. The actual point cloud data and the design point cloud data are aligned sequentially through coarse registration and fine registration to obtain the standard point cloud data.

[0008] The over- and under-excavation area division module is used to divide the over- and under-excavation areas in the standard point cloud data according to the design point cloud data;

[0009] The plane filtering module is used to fit multiple random planes using standard point cloud data, count the number of interior points in each random plane, and filter out multiple initial candidate planes from the multiple random planes based on the number of interior points.

[0010] The tunnel size calculation module is used to classify five actual planes from all initial candidate planes to construct the tunnel profile in the lidar coordinate system, and calculate the tunnel size based on the five actual planes.

[0011] The visualization and storage module is used to combine the over-excavation and under-excavation areas and the tunnel dimensions of the tunnel to obtain the tunnel outline information, visualize the tunnel outline information, and store the acquisition time and tunnel outline information in the 3D scanning database.

[0012] Preferably, the data preprocessing module includes:

[0013] The outlier threshold calculation unit is used to calculate the average neighborhood distance of each original point in the original point cloud data, and calculate the global mean and global standard deviation of the original point cloud data based on the average neighborhood distance of each original point, and calculate the outlier threshold based on the global mean and global standard deviation.

[0014] The denoising unit removes all original points in the original point cloud data whose average neighborhood distance is greater than the outlier threshold as outliers, thus obtaining the actual point cloud data.

[0015] The feature extraction unit is used to calculate the angle features of each actual point in the actual point cloud data. The angle features of each actual point include the angle between the normal vectors, the angle between the projections, and the azimuth angle of the tangent plane. The unit divides and concatenates the angle features of each actual point to obtain the feature vector of each actual point.

[0016] Design point cloud construction unit, used to construct design point cloud data based on the design dimensions of the tunnel;

[0017] The coarse registration unit is used to obtain the design point corresponding to each actual point in the design point cloud data based on the feature vector of each actual point. Multiple coarse registration point pairs are generated based on the actual points and their corresponding design points. The rigid body transformation matrix between the actual point cloud data and the design point cloud data is calculated based on all coarse registration point pairs. Each actual point is transformed based on the rigid body transformation matrix to obtain each coarse registration point. All coarse registration points constitute the coarse registration point cloud data.

[0018] The fine registration unit is used to generate a multi-resolution pyramid based on the coarse registration point cloud data, and to iteratively update the multi-resolution pyramid to achieve fine registration and obtain standard point cloud data.

[0019] Preferably, the fine registration unit includes:

[0020] The multi-resolution pyramid generation subunit is used to downsample the coarsely registered point cloud data layer by layer to generate a multi-resolution pyramid.

[0021] The error calculation subunit is used to construct a KD tree based on the design point cloud data, query the nearest design point of each coarse registration point in the multi-resolution pyramid in the KD tree, and form a fine registration point pair with each coarse registration point in the multi-resolution pyramid and its nearest design point. The error between the coarse registration point cloud data and the design point cloud data is calculated based on the error function and all fine registration point pairs.

[0022] The fine-fit sub-unit is used to determine whether the error between the coarse-fitted point cloud data and the design point cloud data is greater than the error threshold. If the error is greater than the error threshold, the rigid body transformation matrix and the coarse-fitted point cloud data are updated, and the error between the updated coarse-fitted point cloud data and the design point cloud data is recalculated. The iteration stops when the error is less than the error threshold. The standard point cloud data is determined based on the rigid body transformation matrix and the actual point cloud data after the iteration is completed.

[0023] Preferably, the feature extraction unit calculates the actual points in the actual point cloud data. The angle between the normal vectors Projection angle and tangent plane azimuth At that time, this is achieved through formulas (1) to (3):

[0024] (1);

[0025] (2);

[0026] (3);

[0027] In formula (1), Indicates actual point The normal vector, Indicates actual point The normal vector of the k-th neighboring point, This represents the dot product operation;

[0028] In formula (2), Indicates actual point The k-th neighboring point, Indicates actual point The direction of the normal vector;

[0029] In formula (3), Indicates actual point A line connecting its k-th neighbor point to the actual point The direction of the cross product of the normal vectors. express , This represents the arctangent function.

[0030] Preferably, the over- or under-excavation area division module includes:

[0031] A point division unit is used to calculate the minimum distance between each standard point in the standard point cloud data and the design point cloud data. If the minimum distance between any standard point and the design point cloud data is greater than a preset division threshold, the standard point is determined to belong to an over-dug area. If the minimum distance between any standard point and the design point cloud data is less than the negative of the preset division threshold, the standard point is determined to belong to an under-dug area. Otherwise, the standard point is determined to belong to a normal area.

[0032] The region division unit is used to combine standard points belonging to over-excavation areas and under-excavation areas to obtain over-excavation areas and under-excavation areas.

[0033] Preferably, the planar screening module includes:

[0034] The plane fitting unit is used to randomly select multiple sets of plane fitting points consisting of three standard points from the standard point cloud data, and to fit multiple random planes through multiple sets of plane fitting points;

[0035] The in-point statistics unit is used to identify multiple standard points in the standard point cloud data whose distance from each random plane is less than the in-point distance threshold as in-points of each random plane, and to count the number of in-points of each random plane.

[0036] The candidate plane selection unit is used to select multiple random planes from all random planes whose ratio of the number of interior points to the number of standard points in the standard point cloud data is greater than a preset ratio as multiple initial candidate planes.

[0037] Preferably, the five actual planes are the front, top, bottom, left, and right sides, and the tunnel dimension calculation module includes:

[0038] The geometric center point acquisition unit is used to acquire the geometric center point of standard point cloud data;

[0039] The classification unit is used to obtain the normal vector of each initial candidate plane, calculate the angle between the normal vector of each initial candidate plane and the direction vector corresponding to the five actual planes in the lidar coordinate system, and select the direction vector with the smallest angle between the normal vector of each initial candidate plane and the five direction vectors as the direction category of each initial candidate plane. The origin of the lidar coordinate system is the geometric center point of the standard point cloud data, the vertical axis is the tunneling direction of the tunnel, the vertical axis is the direction perpendicular to the tunnel to the right, and the horizontal axis is the direction perpendicular to the tunnel upward.

[0040] The noise plane screening unit is used to filter out noise planes of each directional category from all initial candidate planes based on the geometric characteristics of the actual plane, thereby obtaining multiple candidate planes.

[0041] The confidence calculation unit is used to calculate the angular similarity between each candidate plane and its directional category, the plane density of each candidate plane, and the symmetry contribution of each candidate plane. The angular similarity between each candidate plane and its directional category, the plane density of each candidate plane, and the symmetry contribution of each candidate plane are weighted with the corresponding preset confidence weights to obtain the confidence of each candidate plane and its directional category.

[0042] The actual plane screening unit is used to select the candidate plane with the highest confidence in each direction category as the confidence plane for each direction category, and classify each confidence plane into the corresponding actual plane category according to the geometric characteristics of the actual plane, thus obtaining five actual planes of the tunnel.

[0043] The dimension calculation unit is used to take the distance between the front face and the origin of the lidar coordinate system as the length of the tunnel, the plane distance between the left and right faces as the width of the tunnel, and the plane distance between the top and bottom faces as the height of the tunnel. The tunnel dimensions are obtained by combining the length, width and height of the tunnel.

[0044] Preferably, the noise plane filtering unit is used for:

[0045] Based on the geometric characteristic that the direction vectors of the left and right sides are (0,±1,0) and parallel to each other, all initial candidate planes with direction category (0,±1,0) are obtained. Parallel pairs are filtered for all initial candidate planes with direction category (0,±1,0). Initial candidate planes with direction category (0,±1,0) that do not have parallel pairs are regarded as noise planes and are filtered out.

[0046] Based on the geometric property that the direction vectors of the top and bottom surfaces are (±1,0,0) and parallel to each other, all initial candidate planes with direction category (±1,0,0) are obtained. Parallel pairs are filtered for all initial candidate planes with direction category (±1,0,0). Initial candidate planes with direction category (±1,0,0) that do not have parallel pairs are identified as noise planes and are filtered out.

[0047] Based on the geometric characteristics that the direction vector of the front is (0,0,+1) and there are no parallel planes, all initial candidate planes with direction category (0,0,+1) are obtained. Parallel pairs are filtered for all initial candidate planes with direction category (0,0,+1). Initial candidate planes with parallel pairs among all initial candidate planes with direction category (0,0,+1) are regarded as noise planes and filtered out.

[0048] Preferably, the confidence calculation unit, when calculating the angular similarity between each candidate plane and its directional category, the plane density of each candidate plane, and the symmetry contribution of each candidate plane, includes:

[0049] The angle similarity calculation subunit is used to calculate the angle between the normal vector of each candidate plane and the direction vector of its direction category, which is used as the angle similarity between each candidate plane and its direction category.

[0050] The planar density calculation subunit is used to calculate the ratio of the number of interior points of each candidate plane to the total number of standard points in the standard point cloud data, which is used as the planar density of each candidate plane.

[0051] The symmetry contribution calculation subunit is used to obtain the mirror plane of each candidate plane with the geometric center point as the mirror center, take the candidate plane with the most interior points in the direction category of the mirror plane as the comparison plane, obtain all the coincidence points of the mirror plane and the comparison plane, and take the ratio of the number of coincidence points to the number of interior points of the comparison plane as the symmetry contribution of each candidate plane.

[0052] All of the above-mentioned optional technical solutions can be combined arbitrarily, and the present invention will not provide a detailed description of the structure after each combination.

[0053] By means of the above solution, the beneficial effects of the present invention are as follows:

[0054] This invention acquires raw point cloud data of the tunneling roadway using a 3D scanning device. Based on the design point cloud data, it divides the over- and under-excavation areas in the standard point cloud data. Multiple initial random planes are fitted using the standard point cloud data, and five actual planes of the tunneling roadway are identified. The roadway dimensions are then calculated. Combining the over- and under-excavation areas with the roadway dimensions yields the roadway contour information, which is then visualized. This method achieves the construction of a 3D digital contour of the tunneling roadway based on raw point cloud data, solving the problems of large errors and low efficiency caused by manual measurement, while improving the real-time performance of roadway contour construction. By storing the acquisition time and roadway contour information in a 3D scanning database, the roadway contour information at different times can be directly obtained from the database, solving the problem of inconvenient historical data retrieval in traditional technologies. This invention reduces the error in tunneling roadway contour measurement and improves the efficiency and real-time performance of roadway contour construction.

[0055] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0056] Figure 1This is a schematic diagram of the structure of a three-dimensional digital contour detection system for tunneling roadways provided in an embodiment of the present invention.

[0057] Figure 2 This is a schematic diagram of the network architecture of a three-dimensional digital contour detection system for tunneling roadways provided in an embodiment of the present invention.

[0058] Figure 3 This is a schematic diagram of the over-excavation / under-excavation area and the normal area of ​​a certain tunneling roadway in an embodiment of the present invention.

[0059] Figure 4 This is a schematic diagram showing the result of visualizing the tunnel outline information according to an embodiment of the present invention. Detailed Implementation

[0060] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0061] like Figure 1 As shown, this embodiment of the invention provides a three-dimensional digital contour detection system for tunneling roadways, comprising:

[0062] The data acquisition module is used to acquire the original point cloud data of the tunnel through a three-dimensional scanning device pre-deployed on the roof of the tunnel, and to record the acquisition time of the original point cloud data. The three-dimensional scanning device is equipped with a lidar.

[0063] The data preprocessing module is used to preprocess the raw point cloud data to obtain the actual point cloud data. It constructs the design point cloud data according to the design dimensions of the tunnel. The actual point cloud data and the design point cloud data are aligned sequentially through coarse registration and fine registration to obtain the standard point cloud data.

[0064] The over- and under-excavation area division module is used to divide the over- and under-excavation areas in the standard point cloud data according to the design point cloud data;

[0065] The plane filtering module is used to fit multiple random planes using standard point cloud data, count the number of interior points in each random plane, and filter out multiple initial candidate planes from the multiple random planes based on the number of interior points.

[0066] The tunnel size calculation module is used to classify five actual planes from all initial candidate planes to construct the tunnel profile in the lidar coordinate system, and calculate the tunnel size based on the five actual planes.

[0067] The visualization and storage module is used to combine the over-excavation and under-excavation areas and the tunnel dimensions of the tunnel to obtain the tunnel outline information, visualize the tunnel outline information, and store the acquisition time and tunnel outline information in the 3D scanning database.

[0068] Specifically, in the data acquisition module, a 3D scanning device is installed on the roof of the tunnel. This device has a built-in LiDAR scanner for real-time acquisition of raw point cloud data of the tunnel. The raw point cloud data acquired by the 3D scanning device can very accurately reconstruct the shape of the tunnel with minimal error.

[0069] The data preprocessing module employs an edge processing device installed within the tunneling roadway. This device connects to the 3D scanning device via signal lines and is responsible for preprocessing the raw point cloud data acquired by the 3D scanning device. Preprocessing includes noise reduction and weight reduction to ensure the accuracy and transmission efficiency of the actual point cloud data. When constructing the design point cloud data, the design dimensions of the tunneling roadway are obtained, and then the design point cloud data is constructed based on these dimensions. After preprocessing, coarse registration, and fine registration, the standard point cloud data more closely approximates the actual conditions of the tunneling roadway.

[0070] In the over-excavation and under-excavation area division module, the over-excavation and under-excavation areas include over-excavation areas and under-excavation areas. Areas other than over-excavation areas and under-excavation areas are normal areas.

[0071] In the plane selection module, a random plane is obtained by fitting standard points from three standard point cloud datasets. Then, multiple initial candidate planes are selected from these random planes.

[0072] In the tunnel dimension calculation module, the five actual planes of the tunnel outline include the front, top, bottom, left, and right sides. Tunnel dimensions include the length, width, and height of the tunnel. The origin of the lidar coordinate system is the geometric center of the standard point cloud data; the vertical axis represents the tunneling direction; the ordinate axis is perpendicular to the tunnel to the right; and the horizontal axis is perpendicular to the tunnel upwards.

[0073] The network architecture of a three-dimensional digital contour detection system for tunneling roadways provided in this embodiment of the invention is as follows: Figure 2 As shown, the system is divided into an above-ground section and an underground section. The underground section's edge processing device implements the data preprocessing module's functions, and the industrial Ethernet is responsible for transmitting the standard point cloud data generated by the preprocessing module to the above-ground section. In the above-ground section, the server receives and processes the standard point cloud data until the tunnel contour information is generated; a 3D scanning database stores the acquisition time and tunnel contour information. A PC client is used to visualize the tunnel contour information. By setting up an edge processing device in the underground section to process the raw point cloud data, not only can the real-time performance of data processing be improved, but the amount of data processed can also be reduced, thereby reducing the data processing pressure on the above-ground section. The above-ground section processes the lightweight processed data from the underground section, requiring less computation and improving computational efficiency.

[0074] In one specific embodiment, the data preprocessing module includes:

[0075] The outlier threshold calculation unit is used to calculate the average neighborhood distance of each original point in the original point cloud data, and calculate the global mean and global standard deviation of the original point cloud data based on the average neighborhood distance of each original point, and calculate the outlier threshold based on the global mean and global standard deviation.

[0076] The denoising unit removes all original points in the original point cloud data whose average neighborhood distance is greater than the outlier threshold as outliers, thus obtaining the actual point cloud data.

[0077] The feature extraction unit is used to calculate the angle features of each actual point in the actual point cloud data. The angle features of each actual point include the angle between the normal vectors, the angle between the projections, and the azimuth angle of the tangent plane. The unit divides and concatenates the angle features of each actual point to obtain the feature vector of each actual point.

[0078] Design point cloud construction unit, used to construct design point cloud data based on the design dimensions of the tunnel;

[0079] The coarse registration unit is used to obtain the design point corresponding to each actual point in the design point cloud data based on the feature vector of each actual point. Multiple coarse registration point pairs are generated based on the actual points and their corresponding design points. The rigid body transformation matrix between the actual point cloud data and the design point cloud data is calculated based on all coarse registration point pairs. Each actual point is transformed based on the rigid body transformation matrix to obtain each coarse registration point. All coarse registration points constitute the coarse registration point cloud data.

[0080] The fine registration unit is used to generate a multi-resolution pyramid based on the coarse registration point cloud data, and to iteratively update the multi-resolution pyramid to achieve fine registration and obtain standard point cloud data.

[0081] Specifically, the outlier threshold calculation unit calculates any original point average neighborhood distance When, the calculation formula is: ;in, It is the origin point The s-th neighboring point, G represents the original point. The number of neighboring points, Represents the origin point The distance to its s-th neighbor point. Further, when calculating the global mean and global standard deviation of the original point cloud data, the mean and standard deviation of the average neighborhood distances of all original points are calculated to obtain the global mean and global standard deviation. The formula for calculating the outlier threshold is: ;in, This represents the global mean. Indicates the global standard deviation; This indicates the preset outlier coefficient, which is usually 3.

[0082] In the feature extraction unit, it should be noted that since the normal vector of a certain actual point cannot be directly determined, in this embodiment of the invention, PCA surface fitting is performed on the actual point and all its neighboring points to obtain the fitted surface of the actual point. The normal vector of the actual point in the fitted surface is taken as the normal vector of the actual point. The angle between the normal vectors of a certain actual point represents the angle between the normal vector of the actual point and the normal vector of the center point of the fitted surface of the actual point; the projection angle of a certain actual point represents the angle between the normal vector of the actual point and the horizontal plane in which the actual point is located. The azimuth angle of the tangent plane of a certain actual point represents the angle between the tangent plane of the actual point in its fitted surface and the reference direction (in this embodiment of the invention, the abscissa of the lidar coordinate system). For a certain actual point, the value range of all its angular features is [0,π], and its normal vector angle, projection angle, and azimuth angle of the tangent plane are all divided into 11 equal-width intervals. The frequency of each angle falling in each interval is counted to obtain 3 sets of 11-dimensional histograms. The 3 sets of histograms are concatenated into a 33-dimensional vector to obtain the feature vector of the actual point.

[0083] In the design point cloud construction unit, multiple design points are uniformly constructed according to the design dimensions of the tunnel, and all design points are combined to obtain design point cloud data.

[0084] In the coarse registration unit, when obtaining the corresponding design point in the design point cloud data for each actual point in the actual point cloud data based on the feature vector of each actual point, any actual point and any design point are first paired, and then an adaptive threshold T is introduced for judgment. Specifically, after determining the feature vector of the actual point in any pair, the feature vector of the design point in that pair is further calculated, and the distance between them is calculated to obtain the distance between the feature vectors of the pair. The calculation principle of the feature vector of the design point is the same as that of the feature vector of the actual point. Next, the distances of the feature vectors of all pairs of actual points are combined to obtain the distance set. , where o represents the number of actual point pairs. Then, the distance set is calculated. mean of all elements and standard deviation And calculate the adaptive threshold. for: ;in, The adjustment factor is typically set to 1 to 2. Next, the distance set is retained. The elements, and the distance set The actual point pairs corresponding to elements smaller than T are used as multiple coarse registration point pairs. This is used to calculate the rigid body transformation matrix of the actual point cloud data and the design point cloud data. Then, this is achieved through formulas (4) to (7):

[0085] (4);

[0086] (5);

[0087] (6);

[0088] (7);

[0089] In formula (4), and These are the centroids of the actual point cloud data and the design point cloud data, respectively. Represents the covariance matrix. This represents the rotation matrix in the rigid body transformation matrix. This represents the translation vector in the rigid body transformation matrix. This represents the actual point in the i-th coarse registration point pair among all coarse registration point pairs. This represents the design point in the j-th coarse registration point pair among all coarse registration point pairs.

[0090] In formula (5), Let represent the matrix obtained through singular value decomposition (SVD), where and Represents an orthogonal matrix. This represents a singular value diagonal matrix.

[0091] In the fine registration unit, the multi-resolution pyramid processes the coarse registration point cloud data step by step at multiple resolution levels to enable effective analysis at different levels of detail, thereby achieving fine registration.

[0092] In one specific embodiment, the fine registration unit includes:

[0093] The multi-resolution pyramid generation subunit is used to downsample the coarsely registered point cloud data layer by layer to generate a multi-resolution pyramid.

[0094] The error calculation subunit is used to construct a KD tree based on the design point cloud data, query the nearest design point of each coarse registration point in the multi-resolution pyramid in the KD tree, and form a fine registration point pair with each coarse registration point in the multi-resolution pyramid and its nearest design point. The error between the coarse registration point cloud data and the design point cloud data is calculated based on the error function and all fine registration point pairs.

[0095] The fine-fit sub-unit is used to determine whether the error between the coarse-fitted point cloud data and the design point cloud data is greater than the error threshold. If the error is greater than the error threshold, the rigid body transformation matrix and the coarse-fitted point cloud data are updated, and the error between the updated coarse-fitted point cloud data and the design point cloud data is recalculated. The iteration stops when the error is less than the error threshold. The standard point cloud data is determined based on the rigid body transformation matrix and the actual point cloud data after the iteration is completed.

[0096] Specifically, in the multi-resolution pyramid generation sub-unit, layer-by-layer downsampling refers to processing the coarsely registered point cloud data step by step at multiple resolution levels.

[0097] In the error calculation subunit, the KD tree is a spatial data structure used for efficient storage and retrieval of design point cloud data. The formula for calculating the error E between the coarse registration point cloud data and the design point cloud data is as follows: ;in, and This represents the rotation matrix and translation vector in the rigid body transformation matrix of the above embodiments. This represents the coarse registration point in the q-th fine registration point pair. This represents the design point in the q-th fine-fit point pair, and N represents the number of fine-fit point pairs. .

[0098] In the fine-registration sub-unit, when updating the coarse-registration point cloud data and rigid body transformation matrix, multiple actual point pairs are first reconstructed, and then the adaptive threshold T is recalculated based on the actual point pairs. Next, multiple coarse-registration point pairs are redefined, and the rigid body transformation matrix is ​​calculated based on the redefined coarse-registration point pairs. When determining the standard point cloud data, the actual point cloud data is transformed according to the rigid body transformation matrix after iteration to obtain the standard point cloud data.

[0099] In one specific embodiment, the feature extraction unit calculates the actual points in the actual point cloud data. The angle between the normal vectors Projection angle and tangent plane azimuth At that time, this is achieved through formulas (1) to (3):

[0100] (1);

[0101] (2);

[0102] (3);

[0103] In formula (1), Indicates actual point The normal vector, Indicates actual point The normal vector of the k-th neighboring point, This represents the dot product operation;

[0104] In formula (2), Indicates actual point The k-th neighboring point, Indicates actual point The direction of the normal vector;

[0105] In formula (3), Indicates actual point A line connecting its k-th neighbor point to the actual point The direction of the cross product of the normal vectors. express , This represents the arctangent function.

[0106] In one specific embodiment, the over-excavation and under-excavation area division module includes:

[0107] A point division unit is used to calculate the minimum distance between each standard point in the standard point cloud data and the design point cloud data. If the minimum distance between any standard point and the design point cloud data is greater than a preset division threshold, the standard point is determined to belong to an over-dug area. If the minimum distance between any standard point and the design point cloud data is less than the negative of the preset division threshold, the standard point is determined to belong to an under-dug area. Otherwise, the standard point is determined to belong to a normal area.

[0108] The region division unit is used to combine standard points belonging to over-excavation areas and under-excavation areas to obtain over-excavation areas and under-excavation areas.

[0109] Specifically, in the point division unit, calculate any standard point Minimum distance to the design point cloud data The formula is: ;in, Indicates the first point cloud data in the design One design point, Represents standard point cloud data. This represents the function for minimizing the Euclidean distance. The preset partitioning threshold is determined based on historical point cloud data to define over- or under-excavation areas.

[0110] The over- and under-excavation area division module divides standard point cloud data into over- and under-excavation areas, avoiding omissions and errors that may occur when manually dividing over- and under-excavation areas, and ensuring the comprehensiveness and accuracy of over- and under-excavation area division.

[0111] In one specific embodiment, the planar screening module includes:

[0112] The plane fitting unit is used to randomly select multiple sets of plane fitting points consisting of three standard points from the standard point cloud data, and to fit multiple random planes through multiple sets of plane fitting points;

[0113] The in-point statistics unit is used to identify multiple standard points in the standard point cloud data whose distance from each random plane is less than the in-point distance threshold as in-points of each random plane, and to count the number of in-points of each random plane.

[0114] The candidate plane selection unit is used to select multiple random planes from all random planes whose ratio of the number of interior points to the number of standard points in the standard point cloud data is greater than a preset ratio as multiple initial candidate planes.

[0115] Specifically, in a plane fitting unit, a certain random plane can be represented as ;in, It is the normal vector of the random plane. It is the plane offset of the random plane.

[0116] In the in-point statistical unit, the in-point distance threshold is a threshold determined based on historical experience to distinguish in-points. Calculate any standard point... To the above random plane distance When, the calculation formula is: .

[0117] In the candidate plane screening unit, the preset ratio is 5% in this embodiment of the invention.

[0118] In one specific embodiment, the five actual planes are the front, top, bottom, left, and right sides, and the tunnel size calculation module includes:

[0119] The geometric center point acquisition unit is used to acquire the geometric center point of standard point cloud data;

[0120] The classification unit is used to obtain the normal vector of each initial candidate plane, calculate the angle between the normal vector of each initial candidate plane and the direction vector corresponding to the five actual planes in the lidar coordinate system, and select the direction vector with the smallest angle between the normal vector of each initial candidate plane and the five direction vectors as the direction category of each initial candidate plane. The origin of the lidar coordinate system is the geometric center point of the standard point cloud data, the vertical axis is the tunneling direction of the tunnel, the vertical axis is the direction perpendicular to the tunnel to the right, and the horizontal axis is the direction perpendicular to the tunnel upward.

[0121] The noise plane screening unit is used to filter out noise planes of each directional category from all initial candidate planes based on the geometric characteristics of the actual plane, thereby obtaining multiple candidate planes.

[0122] The confidence calculation unit is used to calculate the angular similarity between each candidate plane and its directional category, the plane density of each candidate plane, and the symmetry contribution of each candidate plane. The angular similarity between each candidate plane and its directional category, the plane density of each candidate plane, and the symmetry contribution of each candidate plane are weighted with the corresponding preset confidence weights to obtain the confidence of each candidate plane and its directional category.

[0123] The actual plane screening unit is used to select the candidate plane with the highest confidence in each direction category as the confidence plane for each direction category, and classify each confidence plane into the corresponding actual plane category according to the geometric characteristics of the actual plane, thus obtaining five actual planes of the tunnel.

[0124] The dimension calculation unit is used to take the distance between the front face and the origin of the lidar coordinate system as the length of the tunnel, the plane distance between the left and right faces as the width of the tunnel, and the plane distance between the top and bottom faces as the height of the tunnel. The tunnel dimensions are obtained by combining the length, width and height of the tunnel.

[0125] Specifically, in the geometric center point acquisition unit, the OBB (Directed Bounding Box) algorithm is used to obtain the geometric center points of the standard point cloud data.

[0126] In the classification unit, the five actual planes in the lidar coordinate system correspond to the five direction vectors (0, ±1, 0), (±1, 0, 0), and (0, 0, +1), respectively. The formula for calculating the angle Q between the normal vector (A, B, C) of any initial candidate plane and any direction vector L is as follows: There is a one-to-one correspondence between the direction category and the direction vector. For example, if the angle between the normal vector of an initial candidate plane and (0,0,+1) among the five direction vectors is the smallest, then (0,0,+1) is taken as the direction category of the initial candidate plane.

[0127] In the noise plane screening unit, the geometric characteristics of the five actual planes are as follows: the direction vectors of the left and right sides are (0, ±1, 0) and parallel to each other; the direction vectors of the top and bottom sides are (±1, 0, 0) and parallel to each other; and the direction vector of the front side is (0, 0, +1) and there are no parallel planes. Since there are many initial candidate planes belonging to the same direction category, some of these candidate planes do not meet the geometric characteristics of the actual planes. In this embodiment of the invention, the initial candidate planes belonging to the same direction category but not meeting the geometric characteristics of the actual planes are designated as noise planes.

[0128] In the confidence calculation unit, the pre-set confidence weights for angular similarity, plane density, and symmetry contribution are 0.5, 0.3, and 0.2, respectively. Since the candidate plane with direction vector (0,0,+1) does not have any parallel planes, its symmetry contribution is 0.

[0129] In the actual plane selection unit, since there are multiple candidate planes belonging to the same direction category, this embodiment of the invention selects the candidate plane with the highest confidence level from all candidate planes belonging to the same direction category as the confidence plane for that direction category. The direction vectors of the five actual planes correspond one-to-one with the five direction categories, and each confidence plane is classified into the corresponding category of the actual plane according to the correspondence. For example, if the direction category of a certain confidence plane is (0,0,+1), then that confidence plane is classified as "frontal".

[0130] In the dimension calculation unit, the formula for calculating the planar distance F between the left and right sides is: ; This represents the normal vectors of the left and right sides. and These represent the planar offsets of the left and right sides, respectively. The planar distance between the top and bottom surfaces can be obtained similarly.

[0131] In this embodiment of the invention, the standard point cloud data is divided into five actual planes—front, top, bottom, left, and right—by the roadway size calculation module. This constructs a plane that matches the actual tunnel profile, thereby improving the accuracy of roadway excavation size calculation.

[0132] In one specific embodiment, the noise plane filtering unit is used for:

[0133] Based on the geometric characteristic that the direction vectors of the left and right sides are (0,±1,0) and parallel to each other, all initial candidate planes with direction category (0,±1,0) are obtained. Parallel pairs are filtered for all initial candidate planes with direction category (0,±1,0). Initial candidate planes with direction category (0,±1,0) that do not have parallel pairs are regarded as noise planes and are filtered out.

[0134] Based on the geometric property that the direction vectors of the top and bottom surfaces are (±1,0,0) and parallel to each other, all initial candidate planes with direction category (±1,0,0) are obtained. Parallel pairs are filtered for all initial candidate planes with direction category (±1,0,0). Initial candidate planes with direction category (±1,0,0) that do not have parallel pairs are identified as noise planes and are filtered out.

[0135] Based on the geometric characteristics that the direction vector of the front is (0,0,+1) and there are no parallel planes, all initial candidate planes with direction category (0,0,+1) are obtained. Parallel pairs are filtered for all initial candidate planes with direction category (0,0,+1). Initial candidate planes with parallel pairs among all initial candidate planes with direction category (0,0,+1) are regarded as noise planes and filtered out.

[0136] In a specific embodiment, the confidence calculation unit, when calculating the angular similarity between each candidate plane and its directional category, the plane density of each candidate plane, and the symmetry contribution of each candidate plane, includes:

[0137] The angle similarity calculation subunit is used to calculate the angle between the normal vector of each candidate plane and the direction vector of its direction category, which is used as the angle similarity between each candidate plane and its direction category.

[0138] The planar density calculation subunit is used to calculate the ratio of the number of interior points of each candidate plane to the total number of standard points in the standard point cloud data, which is used as the planar density of each candidate plane.

[0139] The symmetry contribution calculation subunit is used to obtain the mirror plane of each candidate plane with the geometric center point as the mirror center, take the candidate plane with the most interior points in the direction category of the mirror plane as the comparison plane, obtain all the coincidence points of the mirror plane and the comparison plane, and take the ratio of the number of coincidence points to the number of interior points of the comparison plane as the symmetry contribution of each candidate plane.

[0140] Specifically, in the planar density calculation subunit, multiple standard points in the standard point cloud data whose distance from each candidate plane is less than the inlier distance threshold are taken as inliers of each candidate plane.

[0141] If the mirror plane of a candidate plane does not have a direction category, then the symmetry contribution of the candidate plane is 0.

[0142] In one specific embodiment, the visualization and storage module displays the over-excavation / under-excavation areas and normal areas of the tunnel using different colors. For example... Figure 3 The diagram shows the outline of a tunnel. Red areas represent over-excavated areas, green areas represent normal areas, and blue areas represent under-excavated areas. Using different colors to distinguish between over-excavated, under-excavated, and normal areas helps workers differentiate and identify the tunnel's outline, facilitating subsequent adjustments.

[0143] like Figure 4 As shown, it is a schematic diagram of the result of visualizing the tunnel outline information according to an embodiment of the present invention. Figure 4The length, width, and height of the tunnel are 10.85, 2.56, and 2.86 meters, respectively. The tunnel outline information is visualized using a visualization and storage module. Visual color coding and markings provide an intuitive display of the tunnel outline, helping personnel to identify potential problems in real time and take appropriate measures to prevent significant deviations in the project.

[0144] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A three-dimensional digital profile detection system for tunneling a roadway, characterized by, The method comprises the following steps: A data acquisition module is used to collect original point cloud data of the tunneling roadway through a three-dimensional scanning device pre-deployed on the roof of the tunneling roadway, and record the collection time of the original point cloud data, wherein a laser radar is installed in the three-dimensional scanning device; A data preprocessing module is used to preprocess the original point cloud data to obtain actual point cloud data, construct design point cloud data according to the design size of the tunneling roadway, align the actual point cloud data with the design point cloud data through coarse registration and fine registration in sequence, and obtain standard point cloud data; An over-underexcavation area division module is used to divide the over-underexcavation area in the standard point cloud data according to the design point cloud data; A plane screening module is used to fit a plurality of random planes through the standard point cloud data, count the number of inliers of each random plane, and screen a plurality of initial candidate planes from the plurality of random planes according to the number of inliers; A roadway size calculation module is used to classify five actual planes used for constructing the contour of the tunneling roadway in the laser radar coordinate system from all the initial candidate planes, and calculate the roadway size of the tunneling roadway based on the five actual planes; The five actual planes are a front plane, a top plane, a bottom plane, a left side plane and a right side plane, and the roadway size calculation module comprises: A geometric center point acquisition unit is used to acquire a geometric center point of the standard point cloud data; A classification unit is used to acquire a normal vector of each initial candidate plane, calculate the included angle between the normal vector of each initial candidate plane and a direction vector corresponding to the five actual planes in the laser radar coordinate system, and select the direction vector with the smallest included angle between the normal vector of each initial candidate plane and the five direction vectors as the direction category of each initial candidate plane, wherein the origin of the laser radar coordinate system is the geometric center point of the standard point cloud data, the vertical coordinate axis direction is the tunneling direction of the tunneling roadway, the longitudinal coordinate axis direction is the direction perpendicular to the tunneling roadway to the right, and the transverse coordinate axis direction is the direction perpendicular to the tunneling roadway upward; A noise plane screening unit is used to screen out noise planes of each direction category from all the initial candidate planes according to the geometric characteristics of the actual planes, and obtain a plurality of candidate planes; A confidence calculation unit is used to calculate the included angle similarity between each candidate plane and the direction category thereof, the plane density of each candidate plane and the symmetry contribution degree of each candidate plane, and perform weighted calculation on the included angle similarity between each candidate plane and the direction category thereof, the plane density of each candidate plane and the symmetry contribution degree of each candidate plane and the corresponding preset confidence weight, to obtain the confidence of each candidate plane and the direction category thereof; An actual plane screening unit is used to select the candidate plane with the highest confidence in each direction category as the confidence plane of each direction category, classify each confidence plane into the category of the corresponding actual plane according to the geometric characteristics of the actual plane, and obtain the five actual planes of the tunneling roadway. The dimension calculation unit is used to take the distance between the front face and the origin of the lidar coordinate system as the length of the tunnel, the planar distance between the left and right faces as the width of the tunnel, and the planar distance between the top and bottom faces as the height of the tunnel. The tunnel dimensions are obtained by combining the length, width and height of the tunnel. The visualization and storage module is used to combine the over-excavation and under-excavation areas and the tunnel dimensions of the tunnel to obtain the tunnel outline information, visualize the tunnel outline information, and store the acquisition time and tunnel outline information in the 3D scanning database.

2. A three-dimensional digital profile detection system for excavating a roadway according to claim 1, characterized in that, The data preprocessing module includes: The outlier threshold calculation unit is used to calculate the average neighborhood distance of each original point in the original point cloud data, and calculate the global mean and global standard deviation of the original point cloud data based on the average neighborhood distance of each original point, and calculate the outlier threshold based on the global mean and global standard deviation. The denoising unit removes all original points in the original point cloud data whose average neighborhood distance is greater than the outlier threshold as outliers, thus obtaining the actual point cloud data. The feature extraction unit is used to calculate the angle features of each actual point in the actual point cloud data. The angle features of each actual point include the angle between the normal vectors, the angle between the projections, and the azimuth angle of the tangent plane. The unit divides and concatenates the angle features of each actual point to obtain the feature vector of each actual point. Design point cloud construction unit, used to construct design point cloud data based on the design dimensions of the tunnel; The coarse registration unit is used to obtain the design point corresponding to each actual point in the design point cloud data based on the feature vector of each actual point. Multiple coarse registration point pairs are generated based on the actual points and their corresponding design points. The rigid body transformation matrix between the actual point cloud data and the design point cloud data is calculated based on all coarse registration point pairs. Each actual point is transformed based on the rigid body transformation matrix to obtain each coarse registration point. All coarse registration points constitute the coarse registration point cloud data. The fine registration unit is used to generate a multi-resolution pyramid based on the coarse registration point cloud data, and to iteratively update the multi-resolution pyramid to achieve fine registration and obtain standard point cloud data.

3. A three-dimensional digital profile detection system for excavating a roadway according to claim 2, characterized in that, The fine registration unit includes: The multi-resolution pyramid generation subunit is used to downsample the coarsely registered point cloud data layer by layer to generate a multi-resolution pyramid. The error calculation subunit is used to construct a KD tree based on the design point cloud data, query the nearest design point of each coarse registration point in the multi-resolution pyramid in the KD tree, and form a fine registration point pair with each coarse registration point in the multi-resolution pyramid and its nearest design point. The error between the coarse registration point cloud data and the design point cloud data is calculated based on the error function and all fine registration point pairs. The fine-fit sub-unit is used to determine whether the error between the coarse-fitted point cloud data and the design point cloud data is greater than the error threshold. If the error is greater than the error threshold, the rigid body transformation matrix and the coarse-fitted point cloud data are updated, and the error between the updated coarse-fitted point cloud data and the design point cloud data is recalculated. The iteration stops when the error is less than the error threshold. The standard point cloud data is determined based on the rigid body transformation matrix and the actual point cloud data after the iteration is completed.

4. A three-dimensional digital profile detection system for excavating a roadway according to claim 2, characterized in that, The feature extraction unit calculates actual points in the actual point cloud data. The angle between the normal vectors Projection angle and tangent plane azimuth At that time, this is achieved through formulas (1) to (3): （1）； （2）； （3）； In formula (1), Indicates actual point The normal vector, Indicates actual point The normal vector of the k-th neighboring point, This represents the dot product operation; In equation (2), denotes the kth neighbor point of the actual point denotes the normal vector of the actual point in the direction of the normal vector;​ In equation (3), represents the actual point represents the direction of the cross product of the normal vector of the line connecting the kth neighbor point of the actual point , represents , represents the arctangent function.

5. A three-dimensional digital profile detection system for excavating a roadway according to claim 1, characterized in that, The over- and under-excavation area division module includes: A point division unit is used to calculate the minimum distance between each standard point in the standard point cloud data and the design point cloud data. If the minimum distance between any standard point and the design point cloud data is greater than a preset division threshold, the standard point is determined to belong to an over-dug area. If the minimum distance between any standard point and the design point cloud data is less than the negative of the preset division threshold, the standard point is determined to belong to an under-dug area. Otherwise, the standard point is determined to belong to a normal area. The region division unit is used to combine standard points belonging to over-excavation areas and under-excavation areas to obtain over-excavation areas and under-excavation areas.

6. A three-dimensional digital profile detection system for excavating a roadway according to claim 1, characterized in that, The planar filtering module includes: The plane fitting unit is used to randomly select multiple sets of plane fitting points consisting of three standard points from the standard point cloud data, and to fit multiple random planes through multiple sets of plane fitting points; The in-point statistics unit is used to identify multiple standard points in the standard point cloud data whose distance from each random plane is less than the in-point distance threshold as in-points of each random plane, and to count the number of in-points of each random plane. The candidate plane selection unit is used to select multiple random planes from all random planes whose ratio of the number of interior points to the number of standard points in the standard point cloud data is greater than a preset ratio as multiple initial candidate planes.

7. A three-dimensional digital profile detection system for excavating a roadway according to claim 1, wherein, The noise plane filtering unit is used for: Based on the geometric characteristic that the direction vectors of the left and right sides are (0,±1,0) and parallel to each other, all initial candidate planes with direction category (0,±1,0) are obtained. Parallel pairs are filtered for all initial candidate planes with direction category (0,±1,0). Initial candidate planes with direction category (0,±1,0) that do not have parallel pairs are regarded as noise planes and are filtered out. Based on the geometric property that the direction vectors of the top and bottom surfaces are (±1,0,0) and parallel to each other, all initial candidate planes with direction category (±1,0,0) are obtained. Parallel pairs are filtered for all initial candidate planes with direction category (±1,0,0). Initial candidate planes with direction category (±1,0,0) that do not have parallel pairs are identified as noise planes and are filtered out. Based on the geometric characteristics that the direction vector of the front is (0,0,+1) and there are no parallel planes, all initial candidate planes with direction category (0,0,+1) are obtained. Parallel pairs are filtered for all initial candidate planes with direction category (0,0,+1). Initial candidate planes with parallel pairs among all initial candidate planes with direction category (0,0,+1) are regarded as noise planes and filtered out.

8. A three-dimensional digital profile detection system for excavating a roadway according to claim 1 or 7, characterized in that, The confidence calculation unit, when calculating the angular similarity between each candidate plane and its directional category, the planar density of each candidate plane, and the symmetry contribution of each candidate plane, includes: The angle similarity calculation subunit is used to calculate the angle between the normal vector of each candidate plane and the direction vector of its direction category, which is used as the angle similarity between each candidate plane and its direction category. The planar density calculation subunit is used to calculate the ratio of the number of interior points of each candidate plane to the total number of standard points in the standard point cloud data, which is used as the planar density of each candidate plane. The symmetry contribution calculation subunit is used to obtain the mirror plane of each candidate plane with the geometric center point as the mirror center, take the candidate plane with the most interior points in the direction category of the mirror plane as the comparison plane, obtain all the coincidence points of the mirror plane and the comparison plane, and take the ratio of the number of coincidence points to the number of interior points of the comparison plane as the symmetry contribution of each candidate plane.

Images