Tunnel disease detection system and method based on three-dimensional laser scanning

Abstract

The invention provides a tunnel disease detection system and method based on three-dimensional laser scanning. The method comprises the following steps: inputting tunnel lining point cloud data obtained by laser three-dimensional scanning; performing noise reduction processing on the tunnel lining point cloud data; according to the tunnel lining point cloud data after noise reduction processing, lining peeling, lining cracks and lining water leakage diseases are identified through image feature processing; misjudgment processing and removal are carried out on lining stripping, lining cracks and lining water leakage diseases; and distinguishing and marking disease positions and / or areas in the three-dimensional point cloud model by using different modes. According to the invention, the position, size, shape and other information of the tunnel lining disease can be extracted and analyzed, so that the tunnel lining disease can be quickly and accurately identified and positioned, targeted prevention and repair can be carried out, potential safety hazards can be eliminated, and the driving safety in the tunnel can be ensured.

Description

[0001] Technology Neighborhood

[0002] The present invention relates to the field of tunnel monitoring technology, and in particular to a tunnel disease detection system and method based on three-dimensional laser scanning. Background Art

[0003] Tunnel lining is a permanent support structure built with reinforced concrete and other materials along the perimeter of the tunnel body to prevent deformation or collapse of the tunnel surrounding rock. It bears the pressure and stress distribution of the surrounding rock, prevents the rock from loosening and falling, and maintains the structural stability of the tunnel and driving safety.

[0004] After the tunnel is put into use, it is easily affected by changes in surrounding rock pressure, erosion and corrosion by rainwater, voids / cracks behind the lining, aging of the lining structure over time, concrete shrinkage, uneven thickness / misalignment of the concrete layer, etc., and is prone to falling blocks, cracking and falling off. At the same time, during the tunnel maintenance and construction process, the poor adhesion of concrete repairs and repair seams on the lining surface can easily lead to falling blocks, cracking and falling off of the lining, posing a great threat to driving safety.

[0005] The existing non-destructive testing technologies for tunnel linings mainly include video rapid testing technology (with area array cameras / line array cameras as the main acquisition and detection methods), laser rapid testing technology (with laser video scanning as the main acquisition and detection methods) and radar rapid testing technology (with ground penetrating radar as the main acquisition and detection methods), which can realize non-destructive rapid testing of tunnel lining surface defects and internal contours. Radar detection can also be used to detect internal defects at the same time. Summary of the invention

[0006] The purpose of the present invention is to provide a tunnel disease detection system and method based on three-dimensional laser scanning. Based on the tunnel laser three-dimensional scanning point cloud data, cracks, peeling and water leakage of the tunnel lining can be detected, and the location, size, shape and other information of the disease can be extracted and analyzed to quickly and accurately identify and locate tunnel lining diseases, carry out targeted prevention and repair, eliminate safety hazards, and ensure driving safety in the tunnel.

[0007] According to a first aspect of the present invention, a tunnel defect detection method based on three-dimensional laser scanning is proposed, comprising the following steps:

[0008] Input the tunnel lining point cloud data obtained by laser 3D scanning;

[0009] De-noising the tunnel lining point cloud data;

[0010] Based on the tunnel lining point cloud data after noise reduction, the lining peeling, lining cracks and lining water leakage defects are identified through image feature processing;

[0011] Misjudgment and treatment of lining peeling, lining cracks and lining water leakage; and

[0012] Different methods are used to distinguish and mark the diseased locations and / or areas in the 3D point cloud model.

[0013] In a further embodiment, according to the tunnel lining point cloud data after noise reduction, the lining spalling disease is identified by image feature processing, including:

[0014] Based on the least squares method, the local quadratic surface is fitted, and the concave-convex deformation is identified for each point, and the candidate set of concave-convex change regions is constructed through region growing expansion;

[0015] Based on the distance judgment of adjacent points and region growing combined with geometric constraints, a candidate set of lining missing areas is constructed.

[0016] In a further embodiment, the least square method is used to fit the local quadratic surface, and concave-convex deformation is identified for each point, and a candidate set of concave-convex change regions is constructed by region growing expansion, including:

[0017] For point p i , after determining the neighborhood N(i), perform local quadratic surface fitting based on least squares;

[0018] For the fitted quadratic surface, calculate its Gaussian curvature K and mean curvature H;

[0019] The normal vector is calculated based on the coefficients of the fitted quadratic surface and normalized to obtain the unit normal vector;

[0020] Based on point p i If the curvature difference or normal vector difference with its neighboring points exceeds the preset standard range, the point p i Mark possible concave-convex change points;

[0021] Traverse the point cloud data P in the above manner, combine with the neighborhood set N, identify possible concave and convex change points in the point cloud data P and mark them;

[0022] Create a labeling matrix M with the same scale as the point cloud data, with an initial value of 0; when the i-th point is judged as a possible concave-convex change point, set M(i) = 1; then, through the region growing algorithm, the adjacent points marked as 1 are grouped into a region to form a candidate set of concave-convex change regions.

[0023] In a further embodiment, the distance judgment based on adjacent points and region growing combined with geometric constraints to construct a candidate set of lining missing regions includes:

[0024] Calculate the two adjacent points p in the point cloud data i With pj The distance between ij ;

[0025] If the distance d ij If the distance exceeds the set threshold, it is determined that there may be a missing point between the two points;

[0026] When possible missing points are found, a region growing algorithm combined with geometric constraints is used to extract the missing points from p i First, check its neighboring points. If the angle between the normal vector of a neighboring point and the average normal vector of the current missing area is within the preset range and the distance meets the distance threshold condition, then add it to the missing area set;

[0027] During the growing process, if the neighborhood points newly added to the missing region set cause the curvature estimate of the missing region to exceed the curvature threshold curvature of the lining surface, the growth in that direction is stopped.

[0028] In a further embodiment, the identifying of lining cracks by image feature processing based on the tunnel lining point cloud data after noise reduction includes:

[0029] Generate point cloud slices based on the tunnel lining point cloud data after noise reduction, determine the slice plane to which each point in the point cloud data belongs, form a two-dimensional point set with points belonging to the same slice plane, convert it into a two-dimensional image through vertical projection, and use the reflection intensity of each point as the pixel value of the point in the two-dimensional image;

[0030] Perform image processing on the two-dimensional image to obtain the edge image of the lining crack and identify the foreground and background of the crack area to obtain the crack image;

[0031] By extracting the features of the crack area, the crack geometric features and depth estimation features are obtained.

[0032] In a further embodiment, the extraction of crack region features to obtain crack geometry features and depth estimation features includes:

[0033] For the obtained crack image, the pixel point set of the crack is determined by pixel connectivity analysis; for each crack pixel point, the crack edge is tracked using the chain code method, and the crack length is obtained by calculating the number of crack edge pixel points or accumulating the length of the tracking path;

[0034] The obtained crack image is subjected to a morphological dilation operation using a structural element, and the dilated image is subtracted from the original crack image to obtain the boundary dilation area of ​​the crack; the maximum distance perpendicular to the crack direction in the boundary dilation area is calculated as an estimated value of the crack width;

[0035] The crack positions at different perspectives are identified by ICP registration of point cloud data from different perspectives. The crack depths are estimated using the triangulation principle based on the coordinate differences of the crack points at different perspectives and the known geometric relationship of the perspectives.

[0036] In a further embodiment, the identifying of lining water leakage damage by image feature processing based on the tunnel lining point cloud data after noise reduction processing includes:

[0037] Based on the point cloud slicing and converting the points belonging to the same slicing plane into a two-dimensional image by vertical projection, an adaptive threshold detection algorithm based on the local mean and standard deviation of the image is used to identify potential water leakage pixels;

[0038] For the marked potential water leakage pixels, a seed point is selected and its neighboring pixels are checked starting from the seed point. If the difference between the gray value of the neighboring pixel and the gray value of the seed point is less than the leakage threshold T diff , then add the neighboring pixels to the current water leakage area, and continue to check the neighborhood of the newly added pixels until no new pixels meet the conditions; repeat the above process until all potential water leakage pixels are marked in the corresponding water leakage area;

[0039] For each water leakage area, the pixel area of ​​water leakage is obtained by counting the number of pixels in the area;

[0040] For each water leakage area, the boundary pixels of the water leakage area are tracked and its pixel perimeter is calculated.

[0041] In a further embodiment, for the identified water leakage area, its texture features are calculated for verification. If the contrast of a water leakage area is greater than the set contrast threshold and the energy is less than the energy threshold, then the area is confirmed as a water leakage area, otherwise it is discarded; and if the pixel area of ​​the water leakage is less than the minimum water leakage area threshold, it is considered to be a misjudged area and is removed.

[0042] According to a second aspect of the present invention, a computer system is provided, comprising:

[0043] one or more processors;

[0044] A memory storing operable instructions, wherein when the instructions are executed by the one or more processors, the one or more processors are caused to perform operations, wherein the operations include the process of executing the aforementioned tunnel defect detection method based on three-dimensional laser scanning.

[0045] It should be understood that all combinations of the aforementioned concepts and the additional concepts described in more detail below can be considered as part of the inventive subject matter of the present disclosure as long as such concepts are not mutually inconsistent. In addition, all combinations of the claimed subject matter are considered as part of the inventive subject matter of the present disclosure.

[0046] The foregoing and other aspects, embodiments and features of the present invention can be more fully understood from the following description in conjunction with the accompanying drawings. Other additional aspects of the present invention, such as the features and / or beneficial effects of the exemplary embodiments, will be apparent from the following description or learned from the practice of the specific embodiments according to the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings are not intended to be drawn to scale. In the accompanying drawings, each identical or nearly identical component shown in each figure may be represented by the same reference numeral. For clarity, not every component is labeled in each figure. Embodiments of various aspects of the present invention will now be described by way of example and with reference to the accompanying drawings.

[0048] Figure 1 It is a flow chart of a tunnel defect detection method based on three-dimensional laser scanning according to an embodiment of the present invention.

[0049] Figure 2 It is a schematic diagram of a process for identifying lining spalling damage according to an embodiment of the present invention.

[0050] Figure 3 It is a schematic diagram of a process for identifying lining crack damage according to an embodiment of the present invention.

[0051] Figure 4 It is a schematic diagram of a process of obtaining crack geometric features and depth estimation features by extracting features of a crack region according to an embodiment of the present invention.

[0052] Figure 5 It is a schematic diagram of a process for identifying lining water leakage hazards according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to better understand the technical content of the present invention, specific embodiments are given and described as follows in conjunction with the accompanying drawings.

[0054] Various aspects of the present invention are described in this disclosure with reference to the accompanying drawings, in which many illustrative embodiments are shown. The embodiments of the present disclosure are not necessarily intended to include all aspects of the present invention. It should be understood that the various concepts and embodiments introduced above, as well as those described in more detail below, can be implemented in any of many ways, because the concepts and embodiments disclosed by the present invention are not limited to any implementation. In addition, some aspects disclosed by the present invention can be used alone or in any appropriate combination with other aspects disclosed by the present invention.

[0055] {Example 1}

[0056] Combination Figure 1 As shown, the tunnel defect detection method based on three-dimensional laser scanning according to an embodiment of the present invention includes the following steps:

[0057] Input the tunnel lining point cloud data obtained by laser 3D scanning;

[0058] De-noising the tunnel lining point cloud data;

[0059] Based on the tunnel lining point cloud data after noise reduction, the lining peeling, lining cracks and lining water leakage defects are identified through image feature processing;

[0060] Misjudgment and treatment of lining peeling, lining cracks and lining water leakage; and

[0061] Different methods are used to distinguish and mark the diseased locations and / or areas in the 3D point cloud model.

[0062] In a further embodiment, the tunnel lining point cloud data is subjected to noise reduction processing, including:

[0063] For any point p in the point cloud data P i , use the k-nearest neighbor search algorithm to determine p i The initial neighborhood point set N k (i); P = {p i =(x i ,y i ,z i ,r i )}, i = 1, 2, 3, ..., n, n represents the total number of points in the point cloud data, (x i ,y i ,z i ,) represents point p i The three-dimensional coordinates, r i Represents point p i The reflection intensity;

[0064] Traverse and check the initial neighborhood point set N kFor each point in (i), the distance from point p i Points larger than the set radius R are removed from the initial neighborhood point set N k (i) is removed to obtain the final point p i Neighborhood of;

[0065] The point cloud data P is traversed according to the above denoising process to obtain a neighborhood set N.

[0066] like Figure 2 The example shown is based on the tunnel lining point cloud data after noise reduction, and the lining spalling disease is identified through image feature processing, including:

[0067] Based on the least squares method, the local quadratic surface is fitted, and the concave-convex deformation is identified for each point, and the candidate set of concave-convex change regions is constructed through region growing expansion;

[0068] Based on the distance judgment of adjacent points and region growing combined with geometric constraints, a candidate set of lining missing areas is constructed.

[0069] In a further embodiment, the least square method is used to fit the local quadratic surface, and concave-convex deformation is identified for each point, and a candidate set of concave-convex change regions is constructed by region growing expansion, including:

[0070] For point p i , after determining the neighborhood N(i), perform local quadratic surface fitting based on least squares;

[0071] For the fitted quadratic surface, calculate its Gaussian curvature K and mean curvature H;

[0072] The normal vector is calculated based on the coefficients of the fitted quadratic surface and normalized to obtain the unit normal vector;

[0073] Based on point p i If the curvature difference or normal vector difference with its neighboring points exceeds the preset standard range, the point p i Mark possible concave-convex change points;

[0074] Traverse the point cloud data P in the above manner, combine with the neighborhood set N, identify possible concave and convex change points in the point cloud data P and mark them;

[0075] Create a labeling matrix M with the same scale as the point cloud data, with an initial value of 0; when the i-th point is judged as a possible concave-convex change point, set M(i) = 1; then, through the region growing algorithm, the adjacent points marked as 1 are grouped into a region to form a candidate set of concave-convex change regions.

[0076] In a further embodiment, the point p iIf the curvature difference or normal vector difference with its neighboring points exceeds the preset standard range, the point p i Possible bump change points are marked, including:

[0077] For any point p in the point cloud data i , calculate the curvature difference ΔK between it and the neighboring points i And the normal vector angle difference:

[0078]

[0079] in, represents the average normal vector of the neighborhood points, Indicates normalization to obtain the unit normal vector;

[0080] if and Then point p i Mark possible concave-convex change points;

[0081] if and Then point p i Mark possible concave-convex change points;

[0082] in, and They represent the curvature standard deviation threshold and the normal vector angle cosine value standard deviation threshold respectively; T K and They are the curvature threshold and the normal vector change threshold respectively.

[0083] As an optional example, based on the distance judgment of adjacent points and the region growing combined with geometric constraints, a candidate set of lining missing regions is constructed, including:

[0084] Calculate the two adjacent points p in the point cloud data i With p j The distance between ij ;

[0085] If the distance d ij If the distance exceeds the set threshold, it is determined that there may be a missing point between the two points;

[0086] When possible missing points are found, a region growing algorithm combined with geometric constraints is used to extract the missing points from p i First, check its neighboring points. If the angle between the normal vector of a neighboring point and the average normal vector of the current missing area is within the preset range and the distance meets the distance threshold condition, then add it to the missing area set;

[0087] During the growing process, if the neighborhood points newly added to the missing region set cause the curvature estimate of the missing region to exceed the curvature threshold curvature of the lining surface, the growth in that direction is stopped.

[0088] like Figure 3 The example shown is based on the tunnel lining point cloud data after noise reduction, and the lining crack disease is identified through image feature processing, including:

[0089] Generate point cloud slices based on the tunnel lining point cloud data after noise reduction, determine the slice plane to which each point in the point cloud data belongs, form a two-dimensional point set with points belonging to the same slice plane, convert it into a two-dimensional image through vertical projection, and use the reflection intensity of each point as the pixel value of the point in the two-dimensional image;

[0090] Perform image processing on the two-dimensional image to obtain the edge image of the lining crack and identify the foreground and background of the crack area to obtain the crack image;

[0091] By extracting the features of the crack area, the crack geometric features and depth estimation features are obtained.

[0092] like Figure 4 In the example shown, the crack geometric features and depth estimation features are obtained by extracting the features of the crack area, including:

[0093] For the obtained crack image, the pixel point set of the crack is determined by pixel connectivity analysis; for each crack pixel point, the crack edge is tracked using the chain code method, and the crack length is obtained by calculating the number of crack edge pixel points or accumulating the length of the tracking path;

[0094] The obtained crack image is subjected to a morphological dilation operation using a structural element, and the dilated image is subtracted from the original crack image to obtain the boundary dilation area of ​​the crack; the maximum distance perpendicular to the crack direction in the boundary dilation area is calculated as an estimated value of the crack width;

[0095] The crack positions at different perspectives are identified by ICP registration of point cloud data from different perspectives. The crack depths are estimated using the triangulation principle based on the coordinate differences of the crack points at different perspectives and the known geometric relationship of the perspectives.

[0096] like Figure 5 The example shown is based on the tunnel lining point cloud data after noise reduction, and through image feature processing, the lining leakage disease is identified, including:

[0097] Based on the point cloud slicing and converting the points belonging to the same slicing plane into a two-dimensional image by vertical projection, an adaptive threshold detection algorithm based on the local mean and standard deviation of the image is used to identify potential water leakage pixels;

[0098] For the marked potential water leakage pixels, a seed point is selected and its neighboring pixels are checked starting from the seed point. If the difference between the gray value of the neighboring pixel and the gray value of the seed point is less than the leakage threshold T diff , then add the neighboring pixels to the current water leakage area, and continue to check the neighborhood of the newly added pixels until no new pixels meet the conditions; repeat the above process until all potential water leakage pixels are marked in the corresponding water leakage area;

[0099] For each water leakage area, the pixel area of ​​water leakage is obtained by counting the number of pixels in the area;

[0100] For each water leakage area, the boundary pixels of the water leakage area are tracked and its pixel perimeter is calculated.

[0101] In a further embodiment, for the identified water leakage area, its texture features are calculated for verification. If the contrast of a water leakage area is greater than the set contrast threshold and the energy is less than the energy threshold, then the area is confirmed as a water leakage area, otherwise it is discarded; and if the pixel area of ​​the water leakage is less than the minimum water leakage area threshold, it is considered to be a misjudged area and is removed.

[0102] In an embodiment of the present invention, based on the identification of lining spalling, lining cracks and lining water leakage defects, misjudgment processing and removal of lining spalling, lining cracks and lining water leakage defects are performed, and the false alarm removal processing can be performed by setting a minimum area threshold.

[0103] For example, for spalling damage, if the area of ​​the marked area (estimated by calculating the number of points in the area and the distance between the points) is less than the area threshold, it is considered a false alarm and removed. At the same time, check whether the shape of the spalling area meets the spalling characteristics, such as whether it is a narrow crack or a block.

[0104] Similarly, for water leakage detection, a minimum water leakage area threshold can be set for judgment. If the marked water leakage area (the actual area can be obtained by slowly dividing the pixel area) is smaller than the minimum water leakage area threshold, it is considered to be a misjudged area and removed.

[0105] For crack detection, the area of ​​each detected crack region is calculated (in pixels), and the minimum area threshold can be set for judgment (such as a pixel value of 10). If the crack region area is smaller than the minimum area threshold, it is considered to be noise or a false crack that is misjudged and removed from the result.

[0106] {Example 2}

[0107] In this embodiment, in combination with the accompanying drawings, the implementation process of the tunnel disease detection method of the present invention is further described in detail.

[0108] The tunnel defect detection method based on 3D laser scanning of the embodiment of the present invention uses a 3D laser scanner as the main acquisition device to quickly obtain the 3D spatial information inside the tunnel, and carries a vehicle-mounted mobile platform to perform multi-point scanning in the tunnel according to the set scanning station positions to obtain the surface 3D point cloud data in the tunnel. After pre-processing, denoising and splicing, the point cloud data collected at different positions are spliced ​​into a complete tunnel model, and the lining part of the tunnel is segmented for subsequent defect detection.

[0109] (I) Tunnel lining point cloud data input and noise reduction processing

[0110] Step 1.1: Import the tunnel lining point cloud data obtained by 3D laser scanning in a standard point cloud format (such as PLY, LAS, etc.), and record the point cloud data as P, where P = {p i =(x i ,y i ,z i ,r i )}, i = 1, 2, 3, ..., n, n represents the total number of points in the point cloud data, (x i ,y i ,z i ,) represents the i-th point p in the point cloud data i The three-dimensional coordinates, r i Represents point p i The reflection intensity.

[0111] Step 1.2: Combine radius search to further optimize the neighborhood range, including the following process:

[0112] For any point p in the point cloud data P i , use the k-nearest neighbor search algorithm to determine p i The initial neighborhood point set N k (i); P = {p i =(x i ,y i ,z i ,r i )}, i = 1, 2, 3, ..., n, n represents the total number of points in the point cloud data, (x i ,y i ,z i ,) represents point p i The three-dimensional coordinates, r i Represents point p i The reflection intensity; the neighbor range coefficient k value can be selected according to the point cloud density, and in this example, k=20;

[0113] Traverse and check the initial neighborhood point set N kFor each point in (i), the distance from point p i Points larger than the set radius R are removed from the initial neighborhood point set N k (i) is removed to obtain the final point p i Neighborhood of;

[0114] The point cloud data P is traversed according to the above denoising process to obtain a neighborhood set N.

[0115] Through the above processing, the k-nearest neighbor search algorithm is first used to preliminarily determine the approximate neighborhood of each point in P. On this basis, the neighborhood range is further optimized in combination with the radius search, thereby ensuring that the neighborhood contains sufficient information while reducing the errors caused by noise or outliers.

[0116] (II) Identify lining spalling defects through image feature processing

[0117] In an embodiment of the present invention, tunnel lining spalling disease is manifested as unevenness and missing parts on the lining surface in three-dimensional point cloud data. The unevenness changes and missing areas of the lining surface are detected by means of a geometric feature analysis method of the point cloud data, and the curvature and normal vector of each point in the point cloud data are calculated. According to the changes in the curvature and the normal vector, the unevenness changes of the lining surface are judged. At the same time, the missing areas of the lining surface are detected by comparing the positional relationship of adjacent point cloud data.

[0118] Step 2.1: For any point p in the point cloud data i , local quadratic surface fitting is performed based on least squares, which specifically includes the following processes:

[0119] For any point p in the point cloud data i , considering the complex geometric shape and changes of the lining surface, the improved least squares fitting quadratic surface is adopted, and the quadratic surface equation is set as:

[0120] z=a0+a1x+a2y+a3x 2 +a4xy+a5y 2 ;

[0121] Then construct a weighted least squares problem, with weights w j According to point p j To point p i The distance d ij Determine, and follow the principle that points closer to each other contribute more to the fit:

[0122] The error function E is:

[0123]

[0124] In order to find a set of coefficients of the quadratic surface equation that minimizes the error function E, the error function E is calculated with respect to each coefficient a. k (k=1,2,3,4,5) find the partial derivative and set it equal to 0 to get a set of linear equations. This set of linear equations together constitutes the normal equation (A T WA) a =A T Wz.

[0125] Finally, by solving the normal equation (A T WA)a=A T Wz determines the coefficients of the quadratic surface equation a=[a0,a1,…,a5] T , so that the fitting surface is as close to point p as possible i A is the design matrix, W is the diagonal weight matrix, and z is the Z-axis coordinate vector of the neighborhood point:

[0126]

[0127]

[0128] z=[z1,z2,…,z m ] T ;

[0129] Where m is the point p i The number of neighboring points in the neighborhood point set N(i).

[0130] Therefore, the robustness of the local quadratic surface fitting is further improved by weighting, especially when the neighborhood points are unevenly distributed or there is noise.

[0131] Step 2.2, calculating the Gaussian curvature K and the mean curvature H of the surface, and calculating the curvature according to the first basic form and the second basic form of the surface, specifically comprising the following steps:

[0132] First, according to the first basic form of the surface I = Edu 2 +2Fdudυ+Gdυ 2 , compute the coefficients of the first fundamental form:

[0133]

[0134] Under parameterization (u=x,υ=y), we further have:

[0135]

[0136] Then: E=1+(a1+2a3x+a4y) 2 ;

[0137]

[0138] Then, according to the second basic form of the surface II = Ldu 2 +2Mdudυ+Ndυ 2 , compute the coefficients of the second fundamental form:

[0139]

[0140] in, is the surface normal vector;

[0141] Then: L = 2a3;

[0142]

[0143] Finally, calculate the Gaussian curvature K and the mean curvature H:

[0144]

[0145] Therefore, the curvature of the surface at the point is characterized by the Gaussian curvature K and the mean curvature H, and the normal vector is further estimated on this basis.

[0146] Step 2.3: Use the coefficients of the fitted surface to further calculate the normal vector and normalize it to obtain the unit normal vector, which specifically includes the following process:

[0147] First, the normal vector is calculated using the coefficients of the fitted surface The normal vector is the gradient direction of the surface. For the fitted local quadratic surface z=a0+a1x+a2y+a3x 2 +a4xy+a5y 2 , its gradient is expressed as:

[0148]

[0149] From this we can get:

[0150] n x = -(a1+2a3x+a4y);

[0151] n y = -(a2+a4x+2a5y);

[0152] n z =1.

[0153] Then, in order to ensure the consistency of the normal vector and the accuracy of subsequent calculations, the normal vector is normalized:

[0154]

[0155] in,

[0156] Thus, the obtained unit normal vector can be used to compare the normal vector changes between the point and the neighboring points as one of the bases for judging the concavity and convexity changes. At the same time, the direction information of the normal vector is used as the basis for judging the concavity and convexity direction of the surface in the subsequent region growing algorithm.

[0157] Through the above steps 2.1 to 2.3, accurate curvature and normal vector are obtained, which provide key feature information for the subsequent judgment of the concave-convex changes and missing areas of the lining surface, so as to accurately detect the spalling disease of the tunnel lining.

[0158] Step 2.4, based on point p i If the curvature difference or normal vector difference with its neighboring points exceeds the preset standard range, the point p i Mark as possible concave-convex change points, traverse the point cloud data P, combine with the neighborhood set N, identify and mark possible concave-convex change points in the point cloud data P.

[0159] In the embodiment of the present invention, for any point p in the point cloud data i When calculating the difference between the curvature and normal vector with the neighboring points, not only the simple difference is considered, but also the statistical characteristics of the neighborhood. The standard deviation of the curvature and the standard deviation of the cosine value of the normal angle in the neighborhood are calculated. By comparing the statistical characteristics, the misjudgment caused by local noise or measurement error can be effectively eliminated.

[0160] For any point p in the point cloud data i , calculate the curvature difference ΔK between it and the neighboring points i And the normal vector angle difference:

[0161]

[0162] in, represents the average normal vector of the neighborhood points, Indicates normalization to obtain the unit normal vector;

[0163] if and Then point p i Mark possible concave-convex change points;

[0164] if and Then point p i Mark possible concave-convex change points;

[0165] in, and They represent the curvature standard deviation threshold and the normal vector angle cosine value standard deviation threshold respectively; TK With T ii They are the curvature threshold and the normal vector change threshold respectively.

[0166] In this example, in the analysis of multiple sample tunnels, it was found that the Gaussian curvature of the normal lining surface has a small range of variation, while the Gaussian curvature near the spalling area will change significantly. For the normal vector change threshold, the normal vector angle change is converted into a form that is easier to calculate, such as calculating the cosine value of the normal vector angle through the dot product. Under normal circumstances, the cosine value of the normal vector angle in the neighborhood is close to 1.

[0167] Step 2.5, create a labeling matrix M with the same scale as the point cloud data, with an initial value of 0; when the i-th point is judged as a possible concave-convex change point, set M(i) = 1; then use the region growing algorithm to group the adjacent points marked as 1 into a region to form a candidate set of concave-convex change regions.

[0168] For example, using the region growing algorithm of eight neighborhoods or twenty-six neighborhoods (in three-dimensional space), starting from a marked point, check the mark value of its neighboring points, if it is 1, add it to the current region, and continue to check the neighborhood of the newly added point until there are no new marked points to add; finally, the adjacent points marked as 1 are grouped into a region to form a candidate set of concave-convex change regions. This effectively combines the scattered possible concave-convex change points into continuous regions, which is convenient for subsequent analysis and processing.

[0169] Step 2.6: Based on the distance judgment of adjacent points and the region growing combined with geometric constraints, a candidate set of lining missing regions is constructed, which specifically includes the following steps:

[0170] Calculate the two adjacent points p in the point cloud data i With p j The distance between ij ;

[0171] If the distance d ij Exceeding the set distance threshold T d , then it is judged that there may be a missing between the two points;

[0172] When possible missing points are found, a region growing algorithm combined with geometric constraints is used to extract the missing points from p i First, check its neighboring points. If the angle between the normal vector of a neighboring point and the average normal vector of the current missing area is within the preset range and the distance meets the distance threshold condition, then add it to the missing area set;

[0173] During the growing process, if the neighborhood points newly added to the missing region set cause the curvature estimate of the missing region to exceed the curvature threshold curvature of the lining surface, the growth in that direction is stopped.

[0174] Among them, two adjacent points p in the point cloud data i With p j The distance between ij ,as follows:

[0175]

[0176] In a further optional embodiment, the aforementioned distance threshold Td can be adaptively adjusted according to these geometric characteristics. For example, in the axial direction, since the point spacing may be relatively small under normal circumstances, a smaller axial distance threshold, such as 0.01 meters, can be set; in the radial direction, according to the thickness and design curvature of the lining, a larger radial distance threshold, such as 0.03 meters, can be set. By comparing the distances of adjacent points in different directions with the corresponding thresholds, possible missing areas can be detected more accurately.

[0177] When possible missing points are found, the region is expanded based on the traditional region growing algorithm combined with the geometric constraints of the tunnel lining.

[0178] During the growth process, the distance between points and the direction of the normal vector of the points are considered. If the angle between the normal vector of a neighboring point and the average normal vector of the current missing area is within a certain range (such as less than 45°), and the distance meets the threshold condition, it will be added to the missing area. In this way, the region is prevented from growing to a place that does not meet the geometric characteristics of the missing area, and the accuracy of detection is improved.

[0179] At the same time, the range of regional growth is limited according to the curvature estimate of the lining surface. If it is found during the growth process that the curvature estimate of the region caused by the newly added point exceeds the normal range of curvature change of the lining surface, the growth in this direction is stopped.

[0180] Finally, the determined missing area is represented in the point cloud data or related data structure in a specific marking method, which can be unified with the marking method of the concave-convex change area, for example, in the marking matrix M, the points in the missing area are marked as 2.

[0181] In the subsequent processing, the points marked as 1 (concave-convex change) and 2 (missing area) are uniformly regarded as potential spalling areas for visualization.

[0182] (III) Identify lining spalling defects through image feature processing

[0183] In the embodiment of the present invention, a method based on image processing is used for crack detection of tunnel lining. First, the three-dimensional point cloud data is sliced ​​to determine that the point cloud data of each tunnel slice constitutes a two-dimensional point set and is converted into a two-dimensional image; then, the two-dimensional image is processed using an image processing algorithm to extract the characteristic information of the crack, and based on the extracted characteristic information, the existence of the crack is judged, and the parameters such as the length, width, and depth of the crack are estimated to obtain the crack disease detection result.

[0184] Step 3.1, generating point cloud slices according to the tunnel lining point cloud data after noise reduction, and determining the slice plane to which each point in the point cloud data belongs, specifically comprising the following steps:

[0185] First, the i-th point p in the point cloud data i The reflection intensity r i Mapped to gray value g i , using linear mapping:

[0186]

[0187] where r max and r min are the minimum and maximum values ​​of the point cloud reflection intensity, respectively, and the reflection intensity is normalized to a grayscale value range of 0 to 255;

[0188] Then, according to the shape of the tunnel lining and the possible flow direction of the leakage water, the slicing direction is determined. Usually, the slicing can be performed along the tunnel axial direction, radial direction, or at a certain angle (such as 30°) to the axial direction. In this embodiment, axial slicing and vertical projection are taken as an example, and the slicing interval Δs (such as a value of 0.02 meters) and the slicing thickness h (such as a value of 0.02 meters) are set according to the detection accuracy and point cloud density;

[0189] For each point, determine the slice plane it belongs to based on its coordinates. The points belonging to the same plane are grouped into a two-dimensional point set and converted into a two-dimensional image by projection or other methods. For example, when slicing along the axial direction and projecting vertically, point p i =(x i ,y i , z i ) in the two-dimensional image. i , v i ) can be (x i , z i ), and the reflection intensity of the point is used as the pixel value of the point in the two-dimensional image, ignoring the y coordinate.

[0190] Step 3.2: Perform image processing on the two-dimensional image to obtain the lining crack edge image and identify the foreground (possible crack area) and background of the crack area to obtain the crack image.

[0191] As an optional example, the process of obtaining a lining crack edge image includes the following steps:

[0192] First, the two-dimensional image is Gaussian filtered to smooth the image and reduce the impact of noise on edge detection; for each pixel (u, υ) in the image, its pixel value I after filtering is smooth (u, υ) is:

[0193]

[0194] Where k is the radius of the Gaussian kernel, and I(u,υ) is the pixel value of the original image at (u,υ);

[0195] Then, the Sobel operator is used to calculate the gradient G of the image in the x and y directions. x With G y :

[0196]

[0197] Among them, * represents the convolution operation, and the calculation formulas of the gradient amplitude G and the θ direction are:

[0198]

[0199] Then, by setting the high threshold T high (such as 0.5, which can be determined according to the image pixel value range) and the low threshold T low (such as 0.2). Set the gradient amplitude to be greater than T high The pixel point is determined as a strong edge point, which is less than T low The points between the two are determined as non-edge points, and the points between them are considered edge points if they are connected to strong edge points. By connecting the edges in this way, a preliminary crack edge image is obtained.

[0200] Step 3.3: Based on the crack edge image histogram, extract the foreground and background of the crack area to obtain a crack image, which specifically includes the following steps:

[0201] Calculate the grayscale histogram H(i) of the image, i=0,1,2,…,L-1, where L is the number of gray levels.

[0202] Calculate the between-class variance

[0203]

[0204] in,

[0205]

[0206] In the formula, t is the set threshold;

[0207] Traverse all possible thresholds t and find the variance between classes The maximum threshold T otsu . Set the pixel values ​​in the image greater than T otsu The pixels with a value smaller than T are set as foreground (possible crack area). otsu Set the pixels as background.

[0208] Step 3.4: Extract the features of the crack area to obtain the crack geometry features and depth estimation features.

[0209] As an example, the process includes the following:

[0210] For the obtained crack image, the pixel point set of the crack is determined by pixel connectivity analysis (such as eight-connected domain analysis); for each crack pixel point, the crack edge is tracked using the chain code method, and the crack length is obtained by calculating the number of crack edge pixel points or accumulating the length of the tracking path;

[0211] The obtained crack image is subjected to a morphological dilation operation using a structural element, and the dilated image is subtracted from the original crack image to obtain the boundary dilation area of ​​the crack; the maximum distance perpendicular to the crack direction in the boundary dilation area is calculated as an estimated value of the crack width;

[0212] The crack positions at different perspectives are identified by ICP registration of point cloud data from different perspectives. The crack depths are estimated using the triangulation principle based on the coordinate differences of the crack points at different perspectives and the known geometric relationship of the perspectives.

[0213] (IV) Identification of lining leakage

[0214] In this example, based on the slicing of the aforementioned point cloud data, taking axial slicing and vertical projection as an example, on the basis of converting the points belonging to the same slicing plane into a two-dimensional image through vertical projection, the grayscale value of the point is used as the pixel value of the point in the two-dimensional image.

[0215] Step 4.1, based on the adaptive threshold detection algorithm of the local mean and standard deviation of the image, identify potential water leakage pixels. As an example, the specific process includes the following:

[0216] For each pixel (u,υ) in the image, calculate the grayscale mean μ in its neighborhood (such as a 5*5 pixel neighborhood) uυ and standard deviation σ uυ . Set the threshold T adaptive =μ uυ +qσ uυ , q is a constant, and the setting range is between 2-5. If the pixel value g(u,υ) of pixel (u,υ)>T adaptiυe, then the pixel (u, υ) is marked as a potential water leakage pixel.

[0217] Step 4.2: For the marked potential water leakage pixels, select a seed point and start from the seed point to check its neighboring pixels. If the difference between the gray value of the neighboring pixels and the gray value of the seed point is less than the leakage threshold T diff , then add the neighboring pixels to the current water leakage area, and continue to check the neighborhood of the newly added pixels until no new pixels meet the conditions;

[0218] The above process is repeated until all potential water leakage pixels are marked to the corresponding water leakage areas.

[0219] Step 4.3: Extract features for each water leakage area:

[0220] For each water leakage area, the pixel area of ​​water leakage is obtained by counting the number of pixels in the area;

[0221] For each water leakage area, the boundary pixels of the water leakage area are tracked and its pixel perimeter is calculated.

[0222] In a further embodiment, for the identified water leakage area, its texture features are calculated for verification. If the contrast of a water leakage area is greater than the set contrast threshold and the energy is less than the energy threshold, then the area is confirmed as a water leakage area, otherwise it is discarded; and if the pixel area of ​​the water leakage is less than the minimum water leakage area threshold, it is considered to be a misjudged area and is removed.

[0223] As an optional example, the texture features are calculated by the gray level co-occurrence matrix (GLCM). On the two-dimensional image of the slice, the gray level co-occurrence matrix P(d, θ) is calculated, where d is the distance between pixel pairs (such as d = 1 pixel), and θ = 0° (horizontal direction). For pixel pairs with gray values ​​of g1 and g2, distances d1 and d2, and direction θ, the number of their occurrences is counted as an element in P(d, θ). The gray level co-occurrence matrix is ​​used to further extract energy and contrast features:

[0224] The energy characteristics are expressed as:

[0225] The contrast feature is expressed as:

[0226] Wherein, G is the number of gray levels, for example, G=256.

[0227] At this point, the detection and identification of lining spalling, lining cracks and lining water leakage diseases have been achieved.

[0228] {Example 3}

[0229] In combination with the detection methods of lining spalling, lining cracks and lining water leakage in the above embodiments, the present invention further provides a computer system, including: one or more processors and a memory.

[0230] The aforementioned memory is used to store operable instructions, which, when executed by one or more processors, enable the one or more processors to perform operations, including the process of executing the tunnel defect detection method based on three-dimensional laser scanning of the aforementioned embodiment.

[0231] Although the present invention has been disclosed as above with preferred embodiments, it is not intended to limit the present invention. A person with ordinary knowledge in the technical field to which the present invention belongs may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be determined by the definition of the claims.

Claims

1. A tunnel defect detection method based on three-dimensional laser scanning, characterized in that: include: Input the tunnel lining point cloud data obtained by laser 3D scanning; De-noising the tunnel lining point cloud data; Based on the tunnel lining point cloud data after noise reduction, the lining peeling, lining cracks and lining water leakage defects are identified through image feature processing; Misjudgment and treatment of lining peeling, lining cracks and lining water leakage; as well as Different methods are used to distinguish and mark the diseased locations and / or areas in the 3D point cloud model.

2. The tunnel defect detection method based on three-dimensional laser scanning according to claim 1 is characterized in that: The tunnel lining point cloud data is subjected to noise reduction processing, including: For any point p in the point cloud data P i , use the k-nearest neighbor search algorithm to determine p i The initial neighborhood point set N k (i); P = {p i =(x i ,y i ,z i ,r i )}, i = 1, 2, 3, ..., n, n represents the total number of points in the point cloud data, (x i ,y i ,z i ,) represents point p i The three-dimensional coordinates, r i Represents point p i The reflection intensity; Traverse and check the initial neighborhood point set N k For each point in (i), the distance from point p i Points larger than the set radius R are removed from the initial neighborhood point set N k (i) is removed to obtain the final point p i Neighborhood of; The point cloud data P is traversed according to the above denoising process to obtain a neighborhood set N.

3. The tunnel defect detection method based on three-dimensional laser scanning according to claim 1 is characterized in that: Based on the tunnel lining point cloud data after noise reduction, the lining spalling disease is identified through image feature processing, including: Based on the least squares method, the local quadratic surface is fitted, and the concave-convex deformation is identified for each point, and the candidate set of concave-convex change regions is constructed through region growing expansion; Based on the distance judgment of adjacent points and region growing combined with geometric constraints, a candidate set of lining missing areas is constructed.

4. The tunnel defect detection method based on three-dimensional laser scanning according to claim 1 is characterized in that: The method of fitting a local quadratic surface based on the least squares method, identifying concave-convex deformation at each point, and constructing a candidate set of concave-convex change regions through regional growth expansion includes: For point p i , after determining the neighborhood N(i), perform local quadratic surface fitting based on least squares; For the fitted quadratic surface, calculate its Gaussian curvature K and mean curvature H; The normal vector is calculated based on the coefficients of the fitted quadratic surface and normalized to obtain the unit normal vector; Based on point p i If the curvature difference or normal vector difference with its neighboring points exceeds the preset standard range, the point p i Mark possible concave-convex change points; Traverse the point cloud data P in the above manner, combine with the neighborhood set N, identify possible concave and convex change points in the point cloud data P and mark them; Create a labeling matrix M with the same scale as the point cloud data, with an initial value of 0; when the i-th point is judged as a possible concave-convex change point, set M(i) = 1; then, through the region growing algorithm, the adjacent points marked as 1 are grouped into a region to form a candidate set of concave-convex change regions.

5. The tunnel defect detection method based on three-dimensional laser scanning according to claim 4 is characterized in that: Based on point p i If the curvature difference or normal vector difference with its neighboring points exceeds the preset standard range, the point p i Possible bump change points are marked, including: For any point p in the point cloud data i , calculate the curvature difference ΔK between it and the neighboring points i And the normal vector angle difference: in, represents the average normal vector of the neighborhood points, Indicates normalization to obtain the unit normal vector; if And |σ K |>σ Kth , then point p i Mark possible concave-convex change points; if And |σ cos θ|>σ cos θ th , then point p i Mark possible concave-convex change points; Among them, σK th With σ cos θ th They represent the curvature standard deviation threshold and the normal vector angle cosine value standard deviation threshold respectively; T K and They are the curvature threshold and the normal vector change threshold respectively.

6. The tunnel defect detection method based on three-dimensional laser scanning according to claim 4 is characterized in that: The method of constructing a candidate set of lining missing regions based on the distance judgment of adjacent points and the region growth combined with geometric constraints includes: Calculate the two adjacent points p in the point cloud data i With p j The distance between ij ; If the distance d ij If the distance exceeds the set threshold, it is determined that there may be a missing point between the two points; When possible missing points are found, a region growing algorithm combined with geometric constraints is used to extract the missing points from p i First, check its neighboring points. If the angle between the normal vector of a neighboring point and the average normal vector of the current missing area is within the preset range and the distance meets the distance threshold condition, then add it to the missing area set; During the growing process, if the neighborhood points newly added to the missing region set cause the curvature estimate of the missing region to exceed the curvature threshold curvature of the lining surface, the growth in that direction is stopped.

7. The tunnel defect detection method based on three-dimensional laser scanning according to claim 1 is characterized in that: The method of identifying lining crack defects by image feature processing based on the tunnel lining point cloud data after noise reduction processing includes: Generate point cloud slices based on the tunnel lining point cloud data after noise reduction, determine the slice plane to which each point in the point cloud data belongs, form a two-dimensional point set with points belonging to the same slice plane, convert it into a two-dimensional image through vertical projection, and use the reflection intensity of each point as the pixel value of the point in the two-dimensional image; Perform image processing on the two-dimensional image to obtain the edge image of the lining crack and identify the foreground and background of the crack area to obtain the crack image; By extracting the features of the crack area, the crack geometric features and depth estimation features are obtained.

8. The tunnel defect detection method based on three-dimensional laser scanning according to claim 7 is characterized in that: The method of extracting the features of the crack region to obtain the crack geometric features and depth estimation features includes: For the obtained crack image, the pixel point set of the crack is determined by pixel connectivity analysis; for each crack pixel point, the crack edge is tracked using the chain code method, and the crack length is obtained by calculating the number of crack edge pixel points or accumulating the length of the tracking path; The obtained crack image is subjected to a morphological dilation operation using a structural element, and the dilated image is subtracted from the original crack image to obtain the boundary dilation area of ​​the crack; the maximum distance perpendicular to the crack direction in the boundary dilation area is calculated as an estimated value of the crack width; The crack positions at different perspectives are identified by ICP registration of point cloud data from different perspectives. The crack depths are estimated using the triangulation principle based on the coordinate differences of the crack points at different perspectives and the known geometric relationship of the perspectives.

9. The tunnel defect detection method based on three-dimensional laser scanning according to claim 7 is characterized in that: The method of identifying lining water leakage damage by image feature processing based on the tunnel lining point cloud data after noise reduction processing includes: Based on the point cloud slicing and converting the points belonging to the same slicing plane into a two-dimensional image by vertical projection, an adaptive threshold detection algorithm based on the local mean and standard deviation of the image is used to identify potential water leakage pixels; For the marked potential water leakage pixels, a seed point is selected and its neighboring pixels are checked starting from the seed point. If the difference between the gray value of the neighboring pixel and the gray value of the seed point is less than the leakage threshold T diff , then add the neighboring pixels to the current water leakage area, and continue to check the neighborhood of the newly added pixels until no new pixels meet the conditions; repeat the above process until all potential water leakage pixels are marked in the corresponding water leakage area; For each water leakage area, the pixel area of ​​water leakage is obtained by counting the number of pixels in the area; For each water leakage area, the boundary pixels of the water leakage area are tracked and its pixel perimeter is calculated.

10. The tunnel defect detection method based on three-dimensional laser scanning according to claim 9, characterized in that: For the identified water leakage area, its texture features are calculated for verification. If the contrast of a water leakage area is greater than the set contrast threshold and the energy is less than the energy threshold, the area is confirmed to be a water leakage area, otherwise it is discarded; and if the pixel area of ​​the water leakage is less than the minimum water leakage area threshold, it is considered to be a misjudged area and is removed.

11. A computer system, characterized in that: include: one or more processors; A memory storing operable instructions, wherein when the instructions are executed by the one or more processors, the one or more processors are caused to perform operations, wherein the operations include the process of executing the tunnel defect detection method based on three-dimensional laser scanning as described in any one of claims 1 to 10.

Images