A method for determining the maximum crack width of a road surface crack

Abstract

A kind of maximum crack width determination method suitable for road surface crack, road surface image is collected, crack area is obtained by neural network detection segmentation, the coordinates of the center point of the line segment of the maximum width of crack are obtained after binarization and corrosion treatment, the edge point is obtained by taking point around the starting point with the coordinates, the slope of the edge of the maximum crack width is obtained by the least square method with the obtained edge point, the slope of crack is obtained after averaging, the normal slope is calculated, the coordinates of the two edge points corresponding to the maximum crack width are determined according to the normal slope and the coordinates of the center point of the line segment, the distance between two points is the maximum crack width of crack.The present application automatically selects crack and calculates the maximum crack width, is suitable for various cracks, and can also achieve relatively accurate detection effect under complex road surface conditions, solves the problems of high cost and long time of traditional manual investigation and determination of road surface crack, and overcomes the defects of existing detection based on image processing analysis.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of video image processing, and relates to the application of a neural network, and is a maximum crack width determination method suitable for road surface cracks. BACKGROUND

[0002] In recent years, with the construction and laying of highway networks, the completion of large-scale highways brings heavy highway maintenance work. When cracks appear on the road surface, road runoff caused by rain and the like will seep into the interior of the road along the cracks, erode the roadbed, and make the roadbed unstable, and even cause the roadbed to collapse, which endangers personal and property safety. Crack characteristics are an important indicator for evaluating the condition of the road surface, and the maximum crack width is one of the most representative characteristics of cracks. The size of the crack width directly reflects the current engineering quality of the road, and when the maximum crack width exceeds a safety threshold, the staff responsible for road surface maintenance need to be reminded to repair the road surface as soon as possible, and especially for high-risk road sections such as highways, the maximum crack width of the road surface needs to be known in a timely manner and disposed of in a timely manner to prevent traffic accidents.

[0003] At present, crack determination methods include manual timing survey and image processing analysis. The manual timing survey method is to measure the width of the crack at regular intervals by using a steel ruler, a vernier caliper or a crack width measuring instrument, and to record the direction, position and width of the crack and the like. However, the manual determination method is time-consuming and labor-intensive, has low determination efficiency, and is affected by subjective factors, such as the experience of the operator, which may lead to different measurement results. The image processing analysis method is to process the crack image by means of a computer, to analyze and process the crack image by using an image processing algorithm, and to automatically calculate the maximum width of the crack. The main methods include a segmented width evaluation method and a maximum inscribed circle calculation method. The segmented width evaluation method is to evaluate the width of each segment of the crack after the image is segmented vertically, but cannot solve the problem of different direction cracks being spliced together. The maximum inscribed circle calculation method is to calculate the maximum inscribed circle of each crack, and to represent the maximum width of the crack by the diameter of the maximum inscribed circle. The maximum inscribed circle is more suitable for the case where the upper and lower boundaries of the crack are nearly parallel, but is not suitable for the case where the slopes of the upper and lower boundaries of the crack are quite different. The above image processing analysis determination methods mostly use the steps of traditional greying, edge extraction, median filtering and morphological opening operation to pre-process the image and obtain the crack edge information. However, these steps need to pre-set a threshold and are easily affected by the surrounding background, and the error of edge extraction is large, which affects the accuracy of the subsequent maximum crack width calculation and has poor universality. SUMMARY

[0004] The present application aims to provide a method for automatically screening cracks according to pictures taken by a road inspection vehicle and obtaining more accurate maximum crack width by detecting crack direction.

[0005] The technical solution of the present application is a method for measuring the maximum crack width of a road surface, which collects image information of the road surface, detects and segments the crack shape and area through a neural network, performs binary processing, image erosion processing and connected domain processing, obtains the coordinates of the center point of the line segment at the maximum crack width, takes the surrounding pixel points around the center point as the center for traversal, obtains the edge points of the two edges of the crack, obtains the slope of the two edge curves at the maximum crack width by combining the obtained edge points with the least square method, averages the slope as the slope of the crack, calculates the normal slope perpendicular to the crack, determines the coordinates of the two edge points corresponding to the maximum crack width according to the normal slope, the crack slope and the coordinates of the center point of the line segment at the maximum crack width, and the distance between the two points is the maximum crack pixel width.

[0006] Further, the present application comprises the following steps:

[0007] Step 1: Obtain image information collected by a road inspection vehicle;

[0008] Step 2: Construct a target detection neural network model, input all collected image information into the model, and screen out pictures containing road cracks;

[0009] Step 3: Construct a deep enhanced neural network for road crack segmentation, realize pixel-level segmentation of the crack, perform binary processing on the segmentation result to make the crack edge clearer, obtain a binary image of the crack area, record the minimum value of the binary image as m1 and the maximum value as m2;

[0010] Step 4: Perform erosion on the binary image of the crack area, set the number of erosion operations on the crack as N, record the coordinates of the pixel points equal to m2 after N-1 times of edge erosion, wherein the region where the pixel points are gathered and connected is a connected domain, filter out isolated points deviating from the connected domain by coordinate comparison, and take the average value of the pixel point coordinates in the connected domain as the coordinates of the center point of the maximum pixel width, denoted as (x0, y0);

[0011] Step 5: Take points every other pixel point to the left and right of (x0, y0), and take two pixel points in each direction. Starting from these five pixel points, traverse vertically upwards and downwards, respectively. Meanwhile, take points every other pixel point upwards and downwards of (x0, y0), and take two pixel points in each direction. Starting from these five pixel points, traverse horizontally to the left and right, respectively. The traversal determines whether the current point is a point on the edge of the crack. If the upper and lower edges of the crack are traversed first, record the coordinates of the five pixel points on the upper edge of the crack and the five pixel points on the lower edge of the crack. If the left and right edges of the crack are traversed first, record the coordinates of the five pixel points on the left edge of the crack and the five pixel points on the right edge of the crack. According to the recorded ten pixel point coordinates, the slopes k1, k2 of the two edge curves of the crack are calculated by the least squares method. The average of the two slopes is taken as the slope of the crack, and the normal slope k of the crack is calculated.

[0012] Step 6: Extract all edge points of the crack. Divide the four quadrants with (x0, y0) as the center. According to the normal slope k, determine the quadrants in which the two edge points corresponding to the maximum crack width are located. Then determine the coordinates of the edge points corresponding to the maximum crack width in the two quadrants. The slopes of the straight lines formed by these two coordinates and (x0, y0) in their respective quadrants are both closest to k. Determine the edge points in this way. The distance between the two edge points is obtained by the distance formula, which is the maximum crack width.

[0013] Further, in Step 2, collect road crack pictures, label the cracks with a rectangular frame and indicate their categories, and put them into the data set as labels to build a target detection convolutional neural network. Set the learning rate and iteration number parameters, train the constructed target detection convolutional neural network with the data set, and obtain a neural network model for automatically detecting road cracks. This model can automatically detect pictures containing road cracks and the areas where the cracks are located from newly collected road pictures.

[0014] Further, in Step 3, a deep enhanced neural network for road crack segmentation is constructed. The network layers are optimized using an attention mechanism. The road cracks and corresponding label pictures are input into the deep enhanced neural network for training to obtain a road crack segmentation model for automatically segmenting road cracks and extracting the areas of road cracks, and obtaining crack shape information. In this process, only grayscale images containing road cracks are used as label pictures. The deep enhanced neural network selects DAUNet. The DAUNet is trained twice with road crack pictures. The loss functions used in the two training processes are different. The first training result is saved in a weight file. The second training optimizes the network based on the loss function of the predicted value and the true value based on the weight file.

[0015] Further, in Step 6, based on the point (x0, y0), four quadrants are determined, and according to the normal slope k, two quadrants in which two edge points corresponding to the maximum crack width are located are obtained, when the normal slope k is 0, i.e., the edge point is located on the x-axis, the edge point traversed to the left and the edge point traversed to the right of the center (x0, y0) are obtained according to Step 5, and the distance between the two points is the maximum crack width value; when the normal slope k is infinite, i.e., the edge point is located on the y-axis, the edge point traversed upward and the edge point traversed downward of the center (x0, y0) are obtained according to Step 5, and the distance between the two points is the maximum crack width value; when the normal slope k is neither 0 nor infinite, two edge points corresponding to the maximum crack width are determined according to the normal slope k, and the distance formula is used to obtain the value of the maximum crack width.

[0016] The application also provides a maximum crack width measuring device suitable for road cracks, wherein a computer program is arranged in the measuring device, the measuring device receives a road image shot by a road inspection vehicle, and the computer program is executed to implement the above measuring method on the received road image.

[0017] The maximum crack width calculated by the application can effectively reflect the current road damage condition and provide a basis for road maintenance work. The application first realizes extraction of a road crack region, automatically segments a road crack by using a segmentation neural network with an attention mechanism, can effectively exclude interference of a lane line and the like background, and can improve the accuracy of the segmentation network by using the attention mechanism, and then performs a binarization operation on the segmentation result to accurately extract the road crack region. The application corrodes the crack region by using an image corrosion method, removes isolated points by using a connected domain discrimination, and processes a point set in the connected domain, so that the position information of a line segment corresponding to the maximum crack width can be accurately positioned. When the center pixel point and the pixel points segmented around are traversed to obtain two edge points of the crack, the application does not need to traverse globally, but only needs to traverse in a specific range, so that the traversal efficiency is improved. When the slope of the two edge curves is solved, four pixel points around are selected, and the least square method is used for fitting, so that the slope of the two edge curves of the crack is combined, which greatly improves the calculation accuracy.

[0018] The present application only needs the road surface picture shot by the inspection vehicle, can automatically screen the road surface crack picture, and automatically calculates the maximum crack width of the crack, so as to objectively reflect the crack condition of the current road, and the maintenance personnel can determine the road condition according to the calculated width value. The present application is suitable for various cracks, and can achieve good detection effect under complex road surface conditions, and overcomes the defects of the existing detection based on image processing analysis. If the safety threshold of the maximum crack width of the road surface crack is exceeded, the maintenance personnel can repair in time, eliminates the safety hidden danger of road driving, reduces the possibility of traffic accidents caused by cracks, solves the problems of time-consuming and labor-consuming, subjective influence and low measurement efficiency of traditional personnel in time checking and measuring the maximum width of the crack, and realizes the intelligentization of road crack detection and maximum crack width calculation. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 The flow chart for measuring the maximum crack width of the road surface crack of the present application.

[0020] Figure 2 The schematic diagram for measuring the crack of the present application, (a) is the original image of the collected road surface crack, and (b) is the schematic diagram for detecting the road surface crack based on the deep learning YOLO algorithm.

[0021] Figure 3 The schematic diagram for segmenting the road surface crack based on the DAUNet neural network of the present application.

[0022] Figure 4 The schematic diagram for fitting the slope of the two edges of the crack by using the least square method of the present application.

[0023] Figure 5 The schematic diagram for obtaining the edge coordinates by dividing the quadrant of the present application.

[0024] Figure 6 The schematic diagram for labeling the maximum crack pixel width of the present application, (a) is the maximum crack width labeling diagram of the multi-branch crack, and (b) is the maximum crack width labeling diagram of the oblique crack. DETAILED DESCRIPTION

[0025] The application firstly installs a camera perpendicular to the ground at the tail of a road inspection vehicle, collects image information of the road through inspection, selects a part of road crack pictures for labeling, trains a neural network, the division ratio of the training set and the test set is 8 to 2, saves the trained weight file, then detects the road crack by using the trained model, and evaluates the accuracy. A part of the screened crack pictures is labeled and converted into a gray picture as a label, input into the neural network for training twice, different optimizers and error functions are used in the two times, and a deep neural network capable of automatically segmenting the road crack area is obtained. After segmentation, binary processing is adopted, and then the maximum crack width corresponding position information is obtained by edge erosion, five upper and lower edge points are obtained through image processing, the slope is calculated by using the least square method, the slope and the inclination of the normal line are calculated from the slope, and the final maximum crack width pixel distance is obtained by using the relationship between the distance and the inclination.

[0026] The implementation of the application will be described below in combination with examples, and the specific implementation steps of the application are as shown in Figure 1 , and the description is as follows.

[0027] Step 1: Obtain image information, the inspection vehicle drives on the road, the camera at the tail of the vehicle vertically shoots the road surface, generates video images, and obtains road area pictures by extracting frames from the video through a program.

[0028] Step 2: label the pictures containing road cracks in the road pictures by using a labeling tool Labelimg, use a one-stage detection model YOLO algorithm as a detection network model, train the labeled crack pictures, the division ratio of the training set and the test set is 8 to 2, save the trained weight file, and use the network model and the corresponding network weight to automatically detect which road pictures contain road cracks, so as to achieve the purpose of screening. As shown in Figure 2 , Figure 2 (a) is the original image of the collected road crack, Figure 2 (b) is a schematic diagram of detecting the road crack based on the deep learning YOLO algorithm.

[0029] Step3: The region of the pavement crack is labeled by the labeling tool Labelimg, and the json file after labeling needs to be converted into a gray picture containing the region of the pavement crack. The gray picture is input into the DAUNet (Deep Augmented Neural Network for Pavement Crack Segmentation) deep network together with the original crack picture for training. The attention mechanism is added to the network to optimize the accuracy. The training is divided into two steps. In the first stage, the Adam optimizer is used to train with an adaptive learning rate, and the training result is saved in the form of a weight file. In the second stage, based on the weight file, the network is optimized again according to the loss function of the predicted value and the true value. In this embodiment, the Dice Loss and the SGD optimizer are used for training, and the pavement crack segmentation network model is obtained, which segments the region of the crack, as shown in FIG. 8. The first training is to make the network model better balance the background and the crack, because the crack pixel ratio in the picture is very small. The second training is because the predicted probability of the crack in the first training result is low, and it is difficult to select a suitable threshold to filter out the crack. Therefore, the model is trained with a new loss function to increase the predicted probability of the crack. Figure 3

[0030] After the crack region is segmented, a binarization operation is performed to make the crack edge clearer. The calculation formula of the binarization is as follows:

[0031]

[0032] In the formula, θ is the threshold value of binarization, which is set to 240 by default, f m (x a ,y a ) is the pixel value corresponding to the position of the pixel point (x a ,y a ) in the mth frame, and f m (x a ,y a ) is set to 255 when it is greater than the threshold value, and is changed to 0 when it is less than the threshold value.

[0033] Step4: The binarized image is subjected to edge erosion operation. Through iterative erosion, the coordinates of the center point of the line segment with the maximum width of the crack in the image are determined. Each iteration will delete the pixel points on the edge of the crack until the total number of white pixel points in the eroded crack is 0. The total number of operations is recorded as N, and the result of the edge erosion in the N-1th time is recorded. Isolated points are filtered out through connected domain discrimination. For the pixel points of the connected domain, the average value is taken as the center point coordinate (x0, y0) of the line segment with the maximum pixel width, and the coordinates are recorded for the next calculation operation.

[0034] ​Step5: taking the (x0, y0) obtained in Step4 as the center, taking points every other pixel point to the left and right, two pixel points are taken in each direction, starting from the five pixel points, respectively vertically traversing upwards and downwards, while taking the (x0, y0) as the center, taking points every other pixel point upwards and downwards, two pixel points are taken in each direction, starting from the five pixel points, respectively horizontally traversing left and right, the traversal determines whether the current point is a point on the edge of the crack, if the upper and lower edges of the crack are traversed first, record the five pixel point coordinates of the upper edge of the crack and the five pixel point coordinates of the lower edge of the crack, if the left and right edges of the crack are traversed first, record the five pixel point coordinates of the left edge of the crack and the five pixel point coordinates of the right edge of the crack, according to the recorded ten pixel point coordinates, the slopes k1, k2 of the two edge curves of the crack are calculated by the least square method:

[0035]

[0036] where k 1,2 represents the curve slope k1 or k2, both of which are the same calculation formula, x i , y i are the horizontal and vertical coordinates of the five edge points on an edge curve, is the average of the corresponding five horizontal and vertical coordinates, as shown in Figure 4 .

[0037] The normal slope k is calculated from the values of k1 and k2:

[0038] k = -1 / ((k1+k2) / 2)

[0039] Step6: Since the normal slope k may not be an integer, and the pixel coordinates must be integers, the normal line cannot be accurately marked in the pixel image, and the crack edge point coordinates are extracted. As shown in Figure 5 , four quadrants are defined with (x0, y0) as the center, because the maximum crack width is calculated as a line segment, so the corresponding two edge points must be in the first, third, or second, fourth quadrant or on the coordinate axis. When the normal slope k = 0, i.e. the two edge points fall on the x-axis, the edge points (x x1 , y x1 ) traversed to the left and (x x2 , y x1 ) traversed to the right are obtained from Step5, the distance between the two points is the maximum crack width w:

[0040] w = |x x2 -x x1 |

[0041] When the normal slope k is infinite, i.e. both edge points are on the y-axis, the edge points (x y1 ,y y1 ) and (x y1 ,y y2 ) traversed upward and downward from the center (x0, y0) obtained in Step 5 are obtained, and the distance between the two points is the maximum crack width value w:

[0042] w = |y y2 -y y1 |

[0043] When the normal slope k is not 0 and not infinite, i.e. both edge points are not on the coordinate axis, the two quadrants in which the two edge points corresponding to the maximum crack width are located are determined according to the normal slope k of the crack, such as the second and fourth quadrants in Figure 5 , and the coordinates of the edge points corresponding to the maximum crack width in the two quadrants are determined, the slopes of the straight lines formed by the coordinates and (x0, y0) in the respective quadrants are both closest to k, so as to determine the edge points (x1, y1) and (x2, y2), and finally the distance between the two edge points, i.e. the maximum crack width value w, is obtained by the distance formula:

[0044]

[0045] The calculation result of the maximum crack pixel width is shown as the width at the top left corner of (a) and (b) of Figure 6 . (a) and (b) of Figure 6 respectively show the maximum crack pixel width maps of two cracks, which reflect the calculation results of the maximum crack width under different crack shapes.

[0046] Based on the above determination method, the application further provides a maximum crack width determination device suitable for road surface cracks, wherein a computer program is arranged in the determination device and communicates with a camera on a road inspection vehicle, and the computer program is executed to realize the above method. The determination device first realizes the screening of the road surface pictures collected by the inspection vehicle, and automatically detects the pictures containing road surface cracks, and can achieve good detection effect in complex road surface conditions such as lane line interference and shadow interference. And the subsequent can be accurately positioned to the position of the crack by combining GPS, so as to enable maintenance personnel to find the crack position and repair in time.

[0047] The current maximum crack width measurement method is mostly artificial, time-consuming, low efficiency and high maintenance cost, the maximum crack width measurement method suitable for the pavement crack is provided, the disease condition of the current road can be effectively detected, the maximum crack width is measured by the found crack, and reference basis is provided for pavement maintenance work. The road image is obtained by the camera of the pavement inspection vehicle, the trained deep convolution network is used for automatic segmentation of the crack to obtain regional information when the crack is detected, image preprocessing is carried out, part of the interference is removed to make the edge clearer, the region where the maximum crack width is located is obtained by using the image erosion of the cycle mode, after the coordinate points in the region are obtained, the isolated points are filtered out by connected domain discrimination, the curve slope of the edges of the two cracks is calculated, the quadrant is demarcated, and the maximum crack pixel width is calculated by using the related formula. Compared with artificial measurement, the method can effectively save personnel cost and time cost. In addition, the curvature information of the upper edge and the lower edge of the crack is considered in the method, so that the error caused by the large difference between the upper edge and the lower edge is reduced. The pavement disease information is obtained by the image detection method, the personnel screening of the pavement crack area can be reduced, the edge information of the pavement crack is obtained by the image segmentation method, the interference of light and other factors can be effectively excluded, the method can be suitable for different pavement areas, the image extraction crack edge does not need to select different threshold values manually, and the robustness is high. Meanwhile, the segmentation network is improved in the application, the attention mechanism is added in the network downsampling process, and the network is trained by using the crack picture after the attention mechanism is added, so that the accuracy of the pavement crack edge segmentation is improved.

Claims

1. A method for measuring the maximum crack width of a crack in a pavement, characterized by Collecting road surface image information, detecting and segmenting the crack shape and area through a neural network, binarizing, then performing image erosion processing and connected domain processing to obtain the coordinates of the center point of the line segment where the maximum crack width is located, taking the center point as the center to traverse the surrounding pixel points to obtain the edge points of the two edges of the crack, obtaining the slope of the two edge curves at the maximum crack width through the least square method, averaging the slope as the crack slope, calculating the normal slope perpendicular to the crack, determining the coordinates of the two edge points corresponding to the maximum crack width according to the normal slope, the crack slope and the coordinates of the center point of the line segment where the maximum crack width is located, and the distance between the two points is the maximum crack pixel width, including the following steps: Step 1: obtaining image information collected by a road inspection vehicle; Step 2: constructing a target detection neural network model, inputting all collected image information into the model, and screening out pictures containing road cracks; Step3: Construct a deep enhanced neural network for pavement crack segmentation, realize pixel-level segmentation of the crack, and then binarize the segmentation result to make the crack edge clearer, obtain the binarized image of the crack area, and record the minimum value of the binarized image as , and the maximum value as ; Step4: erode the binary image of the crack region, and set the number of times of crack operation after eroding as , record ; the pixel point coordinates in the image after eroding the edge for the first time, wherein the region of the pixel points gathered and connected is a connected domain, and the isolated points deviating from the connected domain are filtered out by the coordinate comparison method, and the average value of the pixel point coordinates in the connected domain is taken as the coordinates of the center point of the line segment of the maximum width of the crack, denoted as ;​ Step5: taking points every other pixel point to the left and right of , two pixel points in each direction, starting from the five pixel points, vertically traversing up and down respectively, at the same time taking points every other pixel point up and down from , two pixel points in each direction, starting from the five pixel points, horizontally traversing left and right respectively, the traversal determines whether the current point is a point on the edge of the crack, if the upper and lower edges of the crack are traversed first, record the five pixel point coordinates of the upper edge of the crack and the five pixel point coordinates of the lower edge of the crack, if the left and right edges of the crack are traversed first, record the five pixel point coordinates of the left edge of the crack and the five pixel point coordinates of the right edge of the crack, according to the recorded ten pixel point coordinates, the slope of the two edge curves of the crack is calculated by the least square method , taking the average of the two slope as the crack slope, and then calculating the normal slope of the crack ; Step 6: Extract all edge points of the crack, to Divide the center into four quadrants, based on the slope of the normal. Obtain the quadrants where the two edge points corresponding to the maximum seam width are located, then determine the coordinates of the edge points corresponding to the maximum seam width in each of the two quadrants. These two coordinates lie in their respective quadrants. The slopes of the lines that make up the line are all closest to The edge points are determined in this way, and the distance between the two edge points is obtained by the distance formula, which is the maximum crack pixel width.

2. The method for measuring the maximum crack width of a road surface crack according to claim 1, wherein In Step 2, collect road crack pictures, label the cracks with a rectangular frame and mark their categories, put them into the data set as labels, construct a target detection convolutional neural network, set the learning rate and iteration number parameters, train the constructed target detection convolutional neural network with the data set, and obtain a neural network model for automatically detecting road cracks, which is used to automatically detect pictures containing road cracks and crack areas from newly collected road pictures.

3. The maximum crack width determination method for road cracks according to claim 1, characterized in that in Step 3, a deep enhanced neural network for road crack segmentation is constructed, the network layers are optimized using an attention mechanism, road cracks and corresponding label pictures are input into the deep enhanced neural network for training, a road crack segmentation model is obtained, which is used to automatically segment road cracks and extract road crack areas, and crack shape information is obtained at the same time, wherein only grayscale images containing road cracks are used as label pictures, the deep enhanced neural network selects DAUNet, and the road crack pictures are input into the DAUNet for two-time training, the loss functions used in the two-time training are different, the first training result is saved in a weight file, and the second training is based on the weight file to optimize the network according to the loss function of the predicted value and the true value.

4. The method for measuring the maximum crack width of a road surface crack according to claim 1, wherein In Step 6, based on the point four quadrants are drawn, and according to the normal slope get the two quadrants where the two edge points corresponding to the maximum seam width are located, when the normal slope that is, the edge point is located on the x-axis, and from Step 5, the edge point traversed to the left and the edge point traversed to the right with as the center, the distance between the two points is the maximum seam width value; when the normal slope is infinite, that is, the edge point is located on the y-axis, and from Step 5, the edge point traversed upward and the edge point traversed downward with as the center, the distance between the two points is the maximum seam width value; when the normal slope is not 0 nor infinite, according to the normal slope determine the two edge points corresponding to the maximum seam width, and use the distance formula to get the maximum seam width value.

5. A method for determining the maximum crack width of a crack in a pavement according to any one of claims 1 to 4, characterized in that The road surface information is collected by a road inspection vehicle, and a measuring device is arranged in the road inspection vehicle, wherein a computer program is configured in the measuring device, the measuring device receives road surface images taken by the road inspection vehicle, and the computer program is executed to implement the measuring method of any one of claims 1-4 on the received road surface images.

Images