Tunnel scanning type detection method based on three-dimensional point cloud

Abstract

The invention relates to the field of intelligent detection, and provides a tunnel scanning type detection method based on three-dimensional point cloud. The method aims to solve the problems of high data processing complexity, high working intensity, low accuracy, low detection efficiency and the like in the existing tunnel detection method. According to the main scheme, a conventional four-wheel motor vehicle is used as a carrier, an identification reading device and three scanners distributed in a circumferential array mode are installed on the rear portion of the carrier, a laser scanning face is formed, and three-dimensional coordinates of dense points on a laser contour line are synchronously obtained. And the detection step comprises the steps of calculating binomial residual errors of all points to identify abnormal points, realizing sectional detection through a reading device, carrying out clustering and quantitative analysis on the abnormal points, and adopting a threshold method and a reference template comparison method to detect the abnormality and variation of tunnel morphology and diseases. The method is used for efficiently, accurately and comprehensively detecting the inner wall of the tunnel, and important support is provided for tunnel engineering acceptance, inner wall state evaluation and daily maintenance.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent detection, and in particular relates to a tunnel scanning detection method based on three-dimensional point cloud. Background Art

[0002] Detecting the inner wall morphology and disease conditions is conducive to maintaining tunnel traffic safety. The inner wall morphology of the tunnel often undergoes abnormal changes due to factors such as creep, erosion, and aging, while diseases refer to local defects such as cracks, delamination, shedding, holes, depressions, bulges, and bulges distributed on the inner wall of the tunnel. Timely detection of morphological and disease abnormalities and effective assessment of their deterioration degree will effectively avoid the occurrence of severe accidents such as collapse. At the same time, targeted maintenance measures can be taken before the morphological deterioration and disease further expand to increase the service life of the tunnel and reduce subsequent maintenance costs. However, manual regular inspections or some existing tunnel inspection methods and equipment often have shortcomings such as high work intensity, low accuracy, and low inspection efficiency, and it is difficult to compare the changes before and after the same location. Therefore, the development of efficient and reliable tunnel inspection methods is of great significance for tunnel project acceptance, inner wall status and aging degree assessment, and daily repair and maintenance. Summary of the invention

[0003] The purpose of the present invention is to realize the monitoring of the inner wall of the tunnel with economy, accuracy, comprehensiveness and real-time, so as to effectively overcome the shortcomings of the traditional tunnel monitoring method. The present invention designs a simple and efficient data acquisition and processing mode to solve the problem of high data processing complexity of the existing three-dimensional point cloud detection method; designs a tunnel inner wall morphology quantification method that fits the linear point cloud data attributes, and combines the fitting binomial mean residual and normal vector distribution characteristics to characterize the scanning contour morphology of the tunnel inner wall; in addition to the inner wall morphology evaluation, by clustering and quantitatively analyzing the abnormal points on the inner wall of the tunnel, the diseased area in each detection unit is effectively quantified to achieve a more comprehensive tunnel detection function; a unique identification reading method is proposed to realize segmented detection, so as to facilitate and accurately locate the detection position; combining the two monitoring modes can simultaneously accurately identify and locate the abnormalities and changes of the inner wall of the tunnel, which greatly facilitates the engineering acceptance, maintenance and repair of the tunnel and ensures traffic safety.

[0004] In order to achieve the above purpose, the present invention adopts the following technical means:

[0005] The present invention provides a tunnel scanning detection method based on three-dimensional point cloud, comprising:

[0006] Vehicle: A conventional four-wheeled motor vehicle with a logo reading device fixed on the rear to read the logos arranged equidistantly on the inner wall of the tunnel;

[0007] Scanning device: installed at the rear of the vehicle 1, it consists of three scanners distributed in a circular array, namely the first scanner, the second scanner and the third scanner. The laser source on each scanner projects a fan-shaped laser vertically to form a laser scanning surface. The scanning device synchronously obtains the y and z coordinates of the dense points distributed on the laser contour line, and their x coordinates are obtained by combining the real-time speed of the vehicle and the timestamp of each frame of linear point cloud data. Thus, the linear three-dimensional point cloud data representing the tunnel inner wall information is obtained by line scanning;

[0008] The detection steps include:

[0009] Step (1) calculating the binomial residual of each point of each frame of linear point cloud data, and identifying abnormal points based on this; calculating the average residual of the fitted binomial of each frame of data to quantify the smoothness of the inner wall of the tunnel; calculating the normal vector distribution eleven-tuple of each frame of data to characterize the morphology of the inner wall of the tunnel;

[0010] Step (2) reading the identification by the reading device to realize segmented detection, and clustering and quantitative analysis of the abnormal points in each detection unit obtained by the division;

[0011] Step (3) detects abnormalities and changes in tunnel morphology and defects through two modes: threshold method and reference template comparison method.

[0012] In the above scheme, in step (1), the abnormal points in the linear array point cloud are identified based on the residual of the fitted binomial: the zy fitted binomial is calculated based on a point on the linear array and several neighboring points on its left and right. The difference between the z coordinate of the point and the corresponding value z(y) of the fitted binomial is the binomial residual. When the value exceeds the deviation threshold, the coordinate point is considered to be an abnormal point on the inner wall of the tunnel.

[0013] In the above scheme, for any point p in the linear point cloud data i，j (x i ,y i，j , z i，j ) Calculate the binomial fit residual e i，j , specifically, with p i，j (x i ,y i，j , z i，j ) and its left and right neighboring points p i，h-k ~p i，j+k , k is an integer, calculate zy fitting binomial

[0014] z(y)=ay 2 +by+c (1)

[0015] Then p i,j The corresponding binomial residual is

[0016] e i，j =zi，j -z(y i，j ) (2)

[0017] z(y i，j )=ay i，j 2 +by i，j +c

[0018] where i represents the frame number and j represents the index of the point in the specific frame i; if |e i，j |≥T 3σ , then p i，j is an abnormal point, according to the deviation threshold T 3σ The current frame line point cloud P can be identified i All abnormal points.

[0019] In the above scheme, the deviation threshold is determined by the Raida criterion: sampling is performed on the disease-free inner wall of the tunnel to be inspected, the binomial residual of each coordinate point is calculated and their standard deviation is calculated, and the deviation threshold is 3 times the standard deviation.

[0020] In the above scheme, in step (1), the process of solving the normal vector distribution eleven-tuple is:

[0021] Take any point p in the current linear point cloud i，j Calculate the fitting straight line with its left and right adjacent points. The vertical direction of the straight line is the normal vector n corresponding to the point. i，j ;

[0022] Then calculate the normal vector n i，j The vector n perpendicular to the ground v The angle θ i，j , angle θ i，j The value range of is [-110°, 110°];

[0023] Divide [-110°, 110°] into 11 intervals with a step size of 20°, namely, [-110°, -90°), [-90°, -70°), ..., [90°, 110°];

[0024] Calculate all coordinate points p of the current linear point cloud i,l ~p i，n The corresponding normal vector n i，1 ~n i，n With n v The angle formed is θ i，1 ~θ i，n , statistics θ i，1 ~θ i，n The number of points that fall into the above 11 intervals respectively, and the ratio of the number of points in each interval to the total number of points n of the current linear array is calculated Thus, we get the normal vector distribution eleven-tuple

[0025] In the above scheme, in step (2), the position of the abnormal point is determined by a segmented detection mechanism: by reading the mark through the mark reading device fixed on the carrier, it can be determined between which two marks the current detection position is located, and then combined with the frame where the abnormal point is located and the corresponding real-time speed information, the distance from the abnormal position to the mark is determined. The marks are arranged on the inner wall of one side of the tunnel at equal intervals of Δx, and a detection unit is between two marks.

[0026] In the above scheme, in step (2), the abnormal points in each detection unit divided by the mark are clustered, the abnormal points belonging to different disease areas are classified into different sets, and the sets are quantitatively analyzed to determine the length, width and maximum fluctuation of each disease area in the detection unit.

[0027] In the above scheme, the clustering process is specifically as follows:

[0028] Step a1. Randomly select an outlier point p from the current detection unit and use this point as the seed point to construct the set C k , all the unvisited outliers in the region with a three-dimensional radius of r and centered on p are classified into the set C k , mark point p as visited;

[0029] Step a2. Traverse all the outliers in the seed point area described in step a1, and repeat step a1 until no new outliers are included in C. k , so the set C k Build completed;

[0030] Step a3. Randomly select one abnormal point from the remaining unvisited points as a new seed point, construct a new set, and repeat steps a1 and a2 until all points in the detection unit are visited, that is, all abnormal points are classified into corresponding sets according to the proximity relationship, and all abnormal points in the current detection unit are respectively classified into sets {C1, C2, C3...}, where each abnormal point set represents a tunnel inner wall disease area.

[0031] In the above scheme, quantitative analysis is performed on each set obtained by clustering to obtain the length, width and maximum fluctuation of the diseased area on the inner wall of each tunnel. Specifically: principal component analysis is performed on the aforementioned sets {C1, C2, C3...} to obtain the first principal component direction e1 and the second principal component direction e2. Each point in the set is projected onto the two principal component directions. The distance between the first and last projection points of e1 is the length L of the diseased area, and the distance between the first and last projection points of e2 is the width W of the diseased area. The maximum absolute value of the binomial residual corresponding to all abnormal points in the set is the maximum fluctuation of the diseased area at that location.

[0032] In the above scheme, in step (3), the two modes A and B are combined to detect abnormalities and changes in tunnel morphology and diseases. Mode A: Threshold method detects abnormalities in tunnel inner wall morphology and diseases.

[0033] Specifically, in terms of morphological anomaly detection, the average residual of the fitted binomial and the normal vector distribution eleven-tuple corresponding to each frame of the linear point cloud data are calculated. If the average residual or any element of the eleven-tuple exceeds the allowable range, it is considered that there is a morphological anomaly at the corresponding frame;

[0034] In terms of abnormal disease detection, when the number of diseases or the size of the diseased area in the detection unit exceeds the normal allowable value, it is determined that the current detection unit has abnormal tunnel wall diseases, and the detection unit needs to be inspected, and the location of the disease is located in combination with the distance from the abnormal point to the mark for targeted repair;

[0035] Mode B: Benchmark template comparison method to detect changes in tunnel inner wall morphology before and after and changes in disease before and after

[0036] Specifically, firstly, a preliminary scan of the inner walls of all the detection units in the entire detection tunnel is completed;

[0037] In terms of morphological variation detection, the morphological vector corresponding to each frame in each section is obtained based on the fitted binomial average residual and normal vector distribution eleven-tuple corresponding to the linear point cloud data of each frame, so as to construct a reference template;

[0038] Then, in the subsequent actual detection scan, the interpolation method is used to calculate the difference vector between the current frame shape vector and the reference template;

[0039] According to the difference vector, the magnitude of the change of each vector element compared with the initial reference is determined, thereby judging the change of the tunnel inner wall morphology before and after on the time scale;

[0040] In terms of defect mutation detection, during the pre-scan, according to the number of defects in the detection unit and the size of each defect area, the same processing is performed on each detection unit again during the subsequent actual detection scan. The same defect areas in the previous and subsequent scans are paired through the centroid position relationship, and the corresponding changes in length, width, maximum fluctuation and number of defects of the same defect area before and after are compared. In this way, the changes of the tunnel inner wall defects before and after on the time scale are judged, so as to make corresponding maintenance decisions.

[0041] Beneficial effects:

[0042] The present invention is aimed at tunnel inner wall detection and designs a unique line scanning detection method based on three-dimensional point cloud processing technology, including:

[0043] The abnormal points on the inner wall of the tunnel are intelligently identified based on the fitted binomial residual. A method is designed to quantitatively characterize the scanning contour morphology of the inner wall of the tunnel based on the linear point cloud data by calculating the average residual of the fitted binomial and the eleven-tuple of the normal vector distribution. A marked segmented detection mechanism is designed to divide the tunnel into several detection units for easy positioning of the detection position. The abnormal points on the inner wall of the tunnel are clustered and quantitatively analyzed to effectively evaluate the defective area in each detection unit. Two detection modes, the threshold judgment method and the benchmark comparison method, are designed to respectively identify the abnormalities and variations in the morphology and diseases of the inner wall of the tunnel.

[0044] Compared with the existing intelligent tunnel detection methods, the present invention has a simple layout, low computational complexity, good real-time performance, and high detection efficiency. It can accurately and simultaneously detect both the inner wall morphology and the disease of the tunnel. In addition, it can not only use the threshold method to determine the abnormal inner wall morphology of the tunnel finish and normal vector distribution, as well as the abnormal length, width, and undulation of the diseased area, but also accurately obtain the morphology and changes before and after the disease within a period of time through the reference template comparison method. It can also accurately locate the location of abnormalities and changes, which is convenient for repair and maintenance. In addition, compared with the conventional image processing detection method based on the RGB color value of pixels, the present invention can avoid the influence of lighting conditions and color differences, and is expected to achieve more reliable detection in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic diagram of the method of the present invention;

[0046] Figure 2 is a diagram illustrating the abnormal point identification mechanism of the present invention;

[0047] Figure 3 It is a schematic diagram of solving the normal vector eleven-tuple of the present invention;

[0048] Figure 4 The marked segmented detection mechanism of the present invention;

[0049] Figure 5 This is an illustration of the clustering of abnormal points of the present invention;

[0050] Figure 6 This is a diagram illustrating the quantitative analysis of the diseased area of ​​the present invention;

[0051] Figure 7 This is an illustration of an example of disease mutation detection in mode B of the present invention.

[0052] Description of Figure Numbers:

[0053] 1-carrier, 101-label reading device, 2-scanning device, 2A-laser source, 201-first scanner, 202-second scanner, 203-third scanner, 3-label, 301-label I, 302-label II, 303-label III. DETAILED DESCRIPTION

[0054] like Figure 1 As shown, the present invention proposes a tunnel scanning detection method based on three-dimensional point cloud, which mainly includes a carrier 1 and a scanning device 2.

[0055] The vehicle 1 is a conventional four-wheeled motor vehicle on which a reading device 101 is fixed. When the vehicle 1 passes by, the reading device 101 can read the information of the identification array (identification I301, identification II302, identification III303...) arranged equidistantly on the inner wall of the tunnel in sequence, thereby realizing the segmentation of the detected tunnel and facilitating the detection and positioning of abnormalities on the inner wall of the tunnel.

[0056] The scanning device 2 is also installed on the vehicle, and is composed of three scanners distributed in a circular array, namely the first scanner 201, the second scanner 202, and the third scanner 203. The laser source installed on the scanner projects a fan-shaped laser vertically, which together form a laser scanning surface. The scanning surface covers the entire inner wall of the tunnel and forms a laser contour line that can characterize the contour of the inner wall of the tunnel. The three scanners synchronously acquire the y and z coordinates of the dense points distributed on the laser contour line in real time, and their x coordinates are obtained by combining the real-time speed of the vehicle and the timestamp of each frame of linear point cloud data (the coordinate system of the scanning device is defined as: the direction of vehicle movement is x, the direction of the tunnel cross section is v, and the direction perpendicular to the ground is z). In this way, the linear array three-dimensional point cloud data {P1{(x1, y 1，1 , z 1，1 ), (x1, y 1，2 , z 1，2 )...,(x1,y 1，n , z 1，n )}, P2{(x2, y 2，1 , z 2，1 ), (x2, y 2，2 , z 2，2 )...,(x2,y 2，n , z 2，n )}, ...P i {(x i ,y i，1 , z i，1 ), (x i ,y i，2 , z i，2 )...,(x i ,y i，n , z i，n The vehicle speed can be obtained by detecting the wheel linear speed or rotation speed through a dedicated speed sensor, or can be directly obtained in real time by communicating with the instrument provided by the vehicle 1.

[0057] like Figure 2As shown, in the line scan data acquisition mode, an abnormal point recognition mechanism based on binomial fitting residual is designed to identify tunnel inner wall defects. i {p i，1 (x i ,y i,l , z i，1 ), p i，2 (x i ,y i，2 , z i,2 )..., p i，n (x i ,y i，n , z i，n )} as an example (Note: (x i ,y i，1 , z i，l ) is the left starting point of the arc section of the tunnel inner wall cross section, and (x i ,y i，n , z i，n ) is the right end point of the arc segment, and the point cloud data of the road surface is shielded). i，j (x i ,y i，j , z i，j ) Calculate the binomial fit residual e i,j Specifically, P i，j (x i ,y i，j , z i，j ) and its left and right neighboring points P i，j-k ~P i，j+k Calculate zy to fit the binomial

[0058] f i,j :z(y)=ay 2 +by+c (1)

[0059] Then p i，j The corresponding binomial residual is

[0060] e i，j =z i，j -z(y i，j ) (2)

[0061] If |e i，j |≥T 3σ , then p i，j is an outlier point. Threshold T 3σ According to the Laida criterion, the inner wall of the test tunnel without disease is sampled, and the standard deviation σ of the sample point residual is calculated according to formula (1) and formula (2). According to the statistical principle, in the case of no disease, |e i，j <T3σ The probability is 99.7%, that is, when |e i，j |≥T 3σ When p i，j The probability of being misidentified as an abnormal point is only 0.3%. Based on this threshold method, the linear point cloud P of the current frame can be identified. i All abnormal points.

[0062] Furthermore, in order to effectively detect abnormalities in the inner wall morphology of the tunnel, a method for evaluating the inner wall morphology of the tunnel through linear point cloud data is designed, which is achieved by calculating the fitting binomial mean residual and the normal vector distribution nine-tuple:

[0063] Fitting binomial mean residual e i :For the current line array P i All coordinate points p i，1 ~p i，n The absolute value of the residual (determined by the above formula (1) and (2)) is averaged to obtain:

[0064]

[0065] e i It reflects the smoothness of the inner wall of the tunnel. The smaller the value, the smoother the inner wall of the tunnel.

[0066] Normal vector distribution eleven-tuple like Figure 3 As shown in (a), all the coordinate points p of the current linear point cloud i，1 ~p i，n Any point p in i，j Its left and right adjacent points p i，j-1 and p i，j+1 Calculate the fitting line, the normal vector n corresponding to the point i,l The direction is perpendicular to the fitting line; then calculate n i，j The vector n perpendicular to the ground v (direction is opposite to the z-axis, i.e. vertically downward from the road surface) i，j , and stipulate that n i，j Located in v Clockwise, θ i，j When it is negative and located counterclockwise, θ i，j is positive. Usually, the angle θ i，j The value range of is [-110°, 110°], and [-110°, 110°] is divided into 11 intervals with a step size of 20°, namely [-110°, -90°), [-90°, -70°), ..., [90°, 110°]; calculate all the coordinate points p of the current linear point cloud i，1 ~p i，n The corresponding normal vector n i，1 ~ni，n With n v The angle formed is θ i，1 ~θ i，n ; Statistics θ i，1 ~θ i，n The number of points that fall into the above 11 intervals respectively, and the ratio of the number of points in each interval to the total number of points n of the current linear array is calculated Thus, we get the normal vector distribution eleven-tuple

[0067] like Figure 3 As shown in (b), when the inner wall of the tunnel is uneven or creeps, the normal vector distribution is bound to change. The symmetry of the eleven-tuple of normal vector distribution can be used to intuitively evaluate the abnormal morphology of the inner wall of the tunnel.

[0068] Furthermore, the present invention proposes a segmented detection mechanism that divides the tunnel into several detection units along the length direction (x-axis direction) and realizes accurate positioning of the scanning position. Figure 4 As shown, soft markers I301, II302, III303, etc. are arranged at equal intervals of Δx on the inner wall of one side of the tunnel. The markers can be RFID electronic tags, bar codes, QR codes, color codes, etc. A marker reading device 101 is fixed on the carrier 1, and the signal receiving direction of the device is coplanar with the laser scanning surface. When 101 is facing the marker, the serial number corresponding to the marker can be read. The current marker reading time corresponds to the starting point of the current detection unit, and the next marker reading time is the end point of the detection unit. Assume that the reading time of a frame of linear point cloud data is t m At the time of reading the mark I and reading the mark II (t I and t II ), and the frame data is the mth frame after the identification I is read. Then the current scanning surface position can be accurately located according to the real-time speed information:

[0069]

[0070] Where v m represents the real-time speed of vehicle (1) when the mth frame of data from tag I is read, t1 and v1 represent the time of the first frame from tag I to tag I and the real-time speed of vehicle 1, respectively. m_I represents the x-distance from the outlier point to the marker I, v i Indicates the real-time speed of vehicle l when acquiring the i-th frame (calculated from reading to identification I) of the linear point cloud data.

[0071] Furthermore, based on the above segmentation mechanism combined with the above abnormal point identification mechanism (see Figure 2 ), all abnormal points on the inner wall of each detection unit tunnel can be identified. Figure 5(a) shows all abnormal points on the inner wall of the tunnel in the detection unit divided by markers I and II. Next, all abnormal points in each detection unit are clustered, that is, abnormal points belonging to different disease areas are classified into different sets based on the three-dimensional distance.

[0072] The clustering process is as follows:

[0073] 1. Randomly select an outlier point p from the current detection unit and use this point as the seed point to construct the set C k , all the unvisited outliers in the region with a three-dimensional radius of r and centered on p are classified into the set C k , mark point p as visited; (such as Figure 5 As shown in (b), all outliers in the neighborhood centered on point p are classified into set C1).

[0074] 2. Traverse all the outliers in the seed point area described in step 1 and repeat step 1 until no new outliers are included in C. k , so the set C k Construction completed; (such as Figure 5 As shown in (c), the set C1 is constructed).

[0075] 3. Randomly select another outlier point from the remaining unvisited points as a new seed point, construct a new set, and repeat steps 1 and 2 until all points in the detection unit are visited, that is, all outliers are classified into corresponding sets according to the neighbor relationship. Figure 5 As shown in (d), all abnormal points in the current detection unit are classified into sets (C1, C2, C3, ...), where each abnormal point set represents a tunnel inner wall defect area.

[0076] Furthermore, quantitative analysis is performed on each clustering set to obtain the length, width and maximum fluctuation of each tunnel inner wall defect area. Figure 6 As shown in Figure 1, principal component analysis is performed on the aforementioned sets (C1, C2, C3, ...), thereby obtaining the first principal component direction e1 and the second principal component direction e2. Each point in the set is projected onto the two principal component directions. The distance between the first and last projection points of e1 is the length of the diseased area L, and the distance between the first and last projection points of e2 is the width of the diseased area W. The maximum absolute value of the binomial residual (determined by formula (2)) corresponding to all abnormal points in the set is the maximum fluctuation of the diseased area at that location.

[0077] Furthermore, according to the aforementioned tunnel inner wall morphology assessment and disease quantitative analysis method, the morphology and disease abnormalities and variations of the tunnel inner wall are detected through mode A and mode B:

[0078] Mode A: Threshold method to detect abnormalities in tunnel inner wall morphology and abnormalities in disease

[0079] In terms of morphological anomaly detection, the above-mentioned tunnel inner wall morphology evaluation is performed on the detection unit frame by frame, that is, the average residual of the fitted binomial corresponding to the linear point cloud data of each frame (Formula (3)) and the normal vector distribution eleven-tuple ( Figure 3 As shown), if the average residual e i If the maximum allowable value is exceeded, it means that the smoothness of the current scan is significantly deteriorated; if the normal vector distribution is 11-tuple If an element in exceeds the allowable range, it can be considered that the morphology of the inner wall of the tunnel at the scanning location is abnormal. Combined with the distance from the scanning surface to the mark (Formula (4)), the abnormal frame can be accurately located, so that inspection and maintenance can be carried out separately.

[0080] In terms of disease anomaly detection, the detection units are clustered as described above ( Figure 5 shown) and quantitative analysis ( Figure 6 (as shown in the figure), determine the number of defects (the number of sets obtained by the aforementioned clustering process) and the size of the defect area (length, width, and maximum fluctuation), and use the threshold to determine whether the current detection unit is abnormal. If the number of defects in a certain detection unit exceeds the maximum allowable value, or the length, width, or maximum fluctuation of a certain defect area exceeds the allowable value, it means that the current detection unit has abnormal road surface defects, and the detection unit needs to be checked, and the defect position is located in combination with formula (4) and targeted repairs are carried out.

[0081] Mode B: Benchmark template comparison method to detect changes in tunnel inner wall morphology before and after and changes in disease before and after

[0082] In terms of morphological variation detection, when the tunnel is completed or the inner wall morphology is in compliance with the requirements, the vehicle 1 is operated slowly and evenly to complete the pre-scan of the entire tunnel to be inspected, and the fitting binomial average residual (Formula (3)) and the normal vector distribution eleven-tuple ( Figure 3 ). Figure 4 Taking the detection unit divided by the identifiers I and II as an example, the 1st to qth frames of point cloud data are collected in the detection unit, and the shape vectors S1, S2, ... S corresponding to each of the q frames of data are obtained. i , ... S q , forming a reference template, where any vector

[0083]

[0084] Among them, x i Indicates S i The x-direction distance from the corresponding linear point cloud data to the marker I (can be obtained from the x in formula (4) m_IThe solution method is determined), and the second element e i is the average residual of the fitted binomial which can be determined by equation (3). The remaining elements These are the 11 elements of the normal vector distribution eleven-tuple.

[0085] In actual detection, when the vehicle passes through the detection unit divided by identifiers I and II, the parameter vector corresponding to all frame linear point cloud data in the detection unit will also be obtained. m For example, it is expressed as

[0086]

[0087] If x i <x m <x i+1 , that is, S n The corresponding scanning position is between the reference template vector S i and S i+1 The difference vector ΔS can be calculated by interpolation. m :

[0088]

[0089] It is expressed as

[0090]

[0091] ΔS m It represents the difference between the actual detection and the initial benchmark of the pre-scan. The morphological changes of the tunnel inner wall during the pre-scan and re-scan can be judged by its second to last element. m is larger, it means that the scanning position (x after the mark I) m The smoothness of the inner wall of the tunnel changes greatly. If the absolute value of any element is large, it means that the morphology of the inner wall of the tunnel at the scanning position has changed significantly compared with the pre-scan.

[0092] In terms of disease mutation detection, first, when the tunnel is completed or the inner wall morphology of the tunnel is in compliance, let the vehicle l run slowly and evenly to complete the preliminary scan of the entire tunnel to be inspected, and use the above-mentioned disease area clustering ( Figure 5 ) and quantitative analysis ( Figure 6 ) The processing method determines the size of each diseased area in the detection unit. In the subsequent actual detection, each detection unit is scanned again and processed in the same way. Figure 7 As shown, the detection unit divided by identification I and identification II is also used as an example to calculate the corresponding set C of each disease during actual detection and pre-scanning. k The centroid position (i.e. C kThe average value of the three-dimensional coordinates of all coordinate points in the image). The same diseased area can be found by the proximity of the centroid position. The length, width, and maximum fluctuation of the same diseased area before and after are compared. If a certain value changes greatly before and after, it means that the corresponding disease has expanded greatly. For example Figure 7 As shown in the figure, the length L of the defective area C2 is significantly increased in the re-scan compared to the pre-scan, and timely inspection and repair are required to prevent the further deterioration of the defect and potential hazards to driving safety. In addition, the number of defects in the previous and subsequent scans can be compared through the aforementioned clustering process. If the number changes, it means that a new defective area has been generated. Timely repair will also effectively control the defect at a smaller level and reduce the cost of tunnel repair and maintenance.

[0093] The above line scanning detection method for tunnel morphology and defects can selectively or simultaneously work in mode A and mode B: mode A is used to identify abnormalities in tunnel morphology and defects, while mode B is used to detect abnormal changes in tunnel morphology and defects.

[0094] In summary, by integrating three-dimensional scanning technology, binomial residual analysis, quantitative characterization method, segmented detection mechanism and threshold and reference template comparison method, the present invention proposes a tunnel scanning detection method based on three-dimensional point cloud, which aims to effectively overcome the technical difficulties such as real-time data acquisition, anomaly recognition, quantitative characterization, segmented detection and system integration in tunnel inner wall morphology and disease detection, and realizes efficient and accurate tunnel detection.

[0095] Tunnel scanning detection method based on 3D point cloud: By installing a scanner array on the rear of the vehicle, the linear point cloud data of the tunnel inner wall is obtained to achieve real-time detection of the tunnel inner wall morphology and defects. The advantage of this method is that it can accurately obtain 3D information of the tunnel inner wall, and compared with the traditional 2D image detection method, it provides richer and more accurate detection data.

[0096] Binomial residual analysis: Use binomial fitting residuals to identify abnormal points. By calculating the residual of each point and comparing it with the threshold, the defective area of ​​the tunnel wall can be effectively identified. The advantage of this method is that it can accurately and low-complexity identify small deformations and defects, improving the accuracy and reliability of detection.

[0097] Quantitative characterization method of tunnel inner wall morphology: The morphology of the tunnel inner wall is quantitatively characterized by calculating the fitted binomial mean residual and the normal vector distribution eleven-tuple. The advantage of this method is that it can quantify the morphology of the tunnel inner wall from multiple dimensions.

[0098] Quantitative characterization method for tunnel inner wall defects: Clustering and quantitative analysis of abnormal points on the tunnel inner wall can effectively quantify the defect area in each detection unit, thereby gaining a more comprehensive understanding of the status of the tunnel inner wall.

[0099] Segmented detection mechanism: segmented detection is achieved through the identification reading device, dividing the tunnel into several detection units, which is convenient for locating and evaluating the diseased area. The advantage of this method is that it improves the efficiency and pertinence of the detection, and helps to manage and maintain the diseases on the inner wall of the tunnel more finely.

[0100] Threshold method and reference template comparison method: Combining the two modes of threshold method and reference template comparison method, it is possible to simultaneously detect anomalies and changes in tunnel morphology and defects. The advantage of this method is that it can simultaneously achieve a comprehensive assessment of the tunnel inner wall morphology and defects, improving the comprehensiveness and practicality of the detection.

[0101] In summary, the present invention provides an efficient, accurate and comprehensive method for detecting tunnel inner wall morphology and defects. Compared with the existing technology, it has the advantages of simple layout, low calculation complexity, good real-time performance and high detection efficiency. It is of great significance for tunnel project acceptance, inner wall status assessment and daily maintenance.

Claims

1. A tunnel scanning detection method based on three-dimensional point cloud, characterized in that: include: Vehicle (1): A motor vehicle is used, and a marking reading device (101) is fixedly mounted on the rear of the vehicle to read markings arranged at equal intervals on the inner wall of the tunnel; Scanning device (2): installed at the rear of the vehicle 1, and composed of three scanners distributed in a circular array, namely, a first scanner (201), a second scanner (202), and a third scanner (203). The laser sources on each scanner vertically project fan-shaped lasers to form a laser scanning surface. The scanning device synchronously obtains the y and z coordinates of the dense points distributed on the laser contour line, and their x coordinates are obtained by combining the real-time speed of the vehicle (1) and the timestamp of each frame of linear array point cloud data, thereby obtaining linear array three-dimensional point cloud data representing the tunnel inner wall information in a line scanning manner; The detection steps include: Step (1) calculating the binomial residual of each point of each frame of linear point cloud data, and identifying abnormal points based on this; calculating the average residual of the fitted binomial of each frame of data to quantify the smoothness of the inner wall of the tunnel; calculating the normal vector distribution eleven-tuple of each frame of data to characterize the morphology of the inner wall of the tunnel; Step (2) reading the identification by the reading device (101) to realize segmented detection, and clustering and quantitative analysis of the abnormal points in each detection unit obtained by the division; Step (3) detects abnormalities and changes in tunnel morphology and defects through two modes: threshold method and reference template comparison method.

2. The tunnel scanning detection method based on three-dimensional point cloud according to claim 1, characterized in that: In step (1), the abnormal points in the linear array point cloud are identified based on the residual of the fitted binomial: the zy fitted binomial is calculated based on a point on the linear array and several neighboring points on its left and right. The difference between the z coordinate of the point and the corresponding value z(y) of the fitted binomial is the binomial residual. When the value exceeds the deviation threshold, the coordinate point is considered to be an abnormal point on the inner wall of the tunnel.

3. The tunnel scanning detection method based on three-dimensional point cloud according to claim 2, characterized in that: For any point p in the linear point cloud data i，j (x i ,y i，j , z i，j ) Calculate the binomial fit residual e i，j , specifically, with p i，j (x i ,y i，j , z i，j ) and its left and right neighboring points p i，，-k ~p i，j+k , k is an integer, calculate zy fitting binomial z(y)=is 2 +by+c (1) Then p i，j The corresponding binomial residual is e i，j =z i，j -z(y i，j ) (2) z(y i，j )=is i，j 2 +by i，j +c Where i represents the frame number and j represents the index of the point in the specific frame i; if |e ij |≥T 3σ , then p i，j is an abnormal point, according to the deviation threshold T 3σ The current frame line point cloud P can be identified i All abnormal points.

4. The tunnel scanning detection method based on three-dimensional point cloud according to claim 2, characterized in that: The deviation threshold T 3σ Determined by the Raida criterion: sample the disease-free inner wall of the tunnel to be inspected, calculate the binomial residual of each coordinate point and their standard deviation, and the deviation threshold is 3 times the standard deviation.

5. The tunnel scanning detection method based on three-dimensional point cloud according to claim 1, characterized in that: In step (1), the process of solving the normal vector distribution eleven-tuple is: Take any point p in the current linear point cloud i，j Calculate the fitting straight line with its left and right adjacent points. The vertical direction of the straight line is the normal vector n corresponding to the point. i，j ; Then calculate the normal vector n i，j The vector n perpendicular to the ground v The angle θ i，j , angle θ i，j The value range of is [-110°, 110°]; Divide [-110°, 110°] into 11 intervals with a step size of 20°, namely, [-110°, -90°), [-90°, -70°), ..., [90°, 110°]; Calculate all coordinate points p of the current linear point cloud i，1 ~p i，n The corresponding normal vector n i，l ~n i，n , and n v The angle formed is θ i，1 ~θ i，n , statistics θ i，1 ~θ i，m The number of points that fall into the above 11 intervals respectively, and the ratio of the number of points in each interval to the total number of points n of the current linear array is calculated Thus, we get the normal vector distribution eleven-tuple 6. The tunnel scanning detection method based on three-dimensional point cloud according to claim 1, characterized in that: In step (2), the position of the abnormal point is determined by a segmented marking detection mechanism: by reading the mark through the mark reading device (101) fixed on the vehicle (1), it can be determined between which two marks the current detection position is located, and then combined with the frame where the abnormal point is located and the corresponding real-time speed information, the distance from the abnormal position to the mark is determined. The marks are arranged on the inner wall of one side of the tunnel at equal intervals of Δx, and the space between the two marks is a detection unit.

7. A tunnel scanning detection method based on three-dimensional point cloud according to any one of claims 1 to 6, characterized in that: In step (2), the abnormal points in each detection unit divided by the mark are clustered, and the abnormal points belonging to different disease areas are classified into different sets, and the sets are quantitatively analyzed to determine the length, width and maximum fluctuation of each disease area in the detection unit.

8. The tunnel scanning detection method based on three-dimensional point cloud according to claim 7, characterized in that: The specific process of clustering is as follows: Step a1. Randomly select an outlier point p from the current detection unit and use this point as the seed point to construct the set C k , all the unvisited outliers in the region with a three-dimensional radius of r and centered on p are classified into the set C k , mark point p as visited; Step a2. Traverse all the outliers in the seed point area described in step a1, and repeat step a1 until no new outliers are included in C. k , so the set C k The build is complete; Step a3. Randomly select one abnormal point from the remaining unvisited points as a new seed point, construct a new set, and repeat steps a1 and a2 until all points in the detection unit are visited, that is, all abnormal points are classified into corresponding sets according to the proximity relationship, and all abnormal points in the current detection unit are respectively classified into sets {C1, C2, C3...}, where each abnormal point set represents a tunnel inner wall disease area.

9. The tunnel scanning detection method based on three-dimensional point cloud according to claim 8, characterized in that: A quantitative analysis is performed on each set obtained by clustering to obtain the length, width and maximum fluctuation of the diseased area on the inner wall of each tunnel. Specifically: a principal component analysis is performed on the aforementioned sets {C1, C2, C3...} to obtain the first principal component direction e1 and the second principal component direction e2. Each point in the set is projected onto the two principal component directions. The distance between the first and last projection points of e1 is the length L of the diseased area, and the distance between the first and last projection points of e2 is the width W of the diseased area. The maximum absolute value of the binomial residual corresponding to all abnormal points in the set is the maximum fluctuation of the diseased area at that location.

10. A tunnel scanning detection method based on three-dimensional point cloud according to any one of claims 1 or 6, characterized in that: In step (3), the two modes A and B are combined to detect abnormalities and changes in tunnel morphology and diseases. Mode A: Threshold method detects abnormalities in tunnel inner wall morphology and diseases. Specifically, in terms of morphological anomaly detection, the average residual of the fitted binomial and the normal vector distribution eleven-tuple corresponding to each frame of the linear point cloud data are calculated. If the average residual or any element of the eleven-tuple exceeds the allowable range, it is considered that there is a morphological anomaly at the corresponding frame; In terms of abnormal disease detection, when the number of diseases or the size of the diseased area in the detection unit exceeds the normal allowable value, it is determined that the current detection unit has abnormal tunnel wall diseases, and the detection unit needs to be inspected, and the location of the disease is located in combination with the distance from the abnormal point to the mark for targeted repair; Mode B: Benchmark template comparison method to detect changes in tunnel inner wall morphology before and after and changes in disease before and after Specifically, firstly, a preliminary scan of the inner walls of all the detection units in the entire detection tunnel is completed; In terms of morphological variation detection, the morphological vector corresponding to each frame in each section is obtained based on the fitted binomial average residual and normal vector distribution eleven-tuple corresponding to the linear point cloud data of each frame, so as to construct a reference template; Then, in the subsequent actual detection scan, the interpolation method is used to calculate the difference vector between the current frame shape vector and the reference template; According to the difference vector, the magnitude of the change of each vector element compared with the initial reference is determined, thereby judging the change of the tunnel inner wall morphology before and after on the time scale; In terms of defect mutation detection, during the pre-scan, according to the number of defects in the detection unit and the size of each defect area, the same processing is performed on each detection unit again during the subsequent actual detection scan. The same defect areas in the previous and subsequent scans are paired through the centroid position relationship, and the corresponding changes in length, width, maximum fluctuation and number of defects of the same defect area before and after are compared. In this way, the changes of the tunnel inner wall defects before and after on the time scale are judged, so as to make corresponding maintenance decisions.

Images