CN110084844B - Airport pavement crack detection method based on depth camera - Google Patents

Abstract

A Depth Camera Based Crack Detection Method for Airport Pavement. It includes using the depth camera to collect the depth image of the airport road surface, and then dividing it into multiple grids; expanding the grid; constructing the road surface model of the expanded grid; connecting the road surface model of each grid The overall surface model of the pavement in the depth image is obtained; the difference between the depth image and the overall surface model is calculated to obtain candidate crack pixels; the candidate crack pixel points are screened to obtain the real crack area and other steps. The method of the invention has the advantages that it is applied to the detection of cracks in airport pavement, and the manual operation is replaced by the high-performance automatic detection method of airport pavement, which can improve the detection accuracy and work efficiency, and further improve the safety performance of the airport pavement. It is not affected by illumination changes and is more robust to environmental noise. The pavement structure is more accurately modeled, thus making crack detection based on pavement model reconstruction more accurate.

Description

A method for detecting cracks in airport pavement based on depth camera

Technical Field

The present invention belongs to the technical field of nondestructive testing, and in particular relates to a method for detecting cracks on an airport pavement based on a depth camera.

Background Art

Airport runways are the most important infrastructure for aircraft flight. Due to the increase in flight frequency and the destruction of the natural environment, many airport runway pavements have suffered varying degrees of damage, causing major safety hazards. Cracks are the main defect of airport runway pavements. Therefore, airport runway crack detection technology has received increasing attention.

At present, the detection of cracks on airport runways mainly relies on manual inspection with manual observation and manual recording. However, manual inspection has many problems such as poor accuracy, easy to miss detection, strong subjectivity, and low efficiency. It is urgent to develop a high-performance automated airport runway defect detection method to replace manual operation and improve work efficiency and safety performance. With the rapid development of visual sensor technology and related technologies such as pattern recognition, some scholars have begun to pay attention to crack detection technology based on computer vision. For the detection of cracks on highway and bridge pavements, which are similar to them, most of the current related research is based on visible light sensors, and some results have been achieved. However, the robustness problem in complex environments is the biggest difficulty faced by existing research. Pavement crack detection methods based on visible light cameras can be divided into four categories: grayscale threshold-based methods, edge detection-based methods, machine learning methods, and morphological methods. Grayscale threshold-based methods are sensitive to noise, especially when the lighting conditions are poor, the crack detection effect is very unreliable. The main disadvantage of edge detection-based methods is that since the connectivity of cracks is not considered, edge detection can only extract discontinuous crack features, and this method often fails in scenes with low contrast and strong noise interference. Machine learning methods cannot guarantee the extraction of global crack information of the entire image. In addition, the learning process requires a large number of accurately labeled samples, which is difficult to achieve in applications with significant changes in lighting and scenes. The effect of crack detection based on morphology is greatly affected by parameter selection and is difficult to apply in practice. Most of the existing studies, such as the work mentioned above, are conducted on roads or bridges. Although there are also literatures claiming that the road pavement crack detection method can also be used for airport runways, there is no actual experimental verification. Compared with roads and bridge pavements, airport runways have obvious traces of oil stains and rubber residues due to frequent aircraft takeoffs and landings, and the contrast between cracks and background is very low due to the runway material. In addition, pavement detection can only be performed at night under artificial light conditions, and the crack features are usually relatively small. These factors make visual detection of cracks on airport runway pavements very difficult. In recent years, some people have begun to use depth sensors to detect pavement defects, but they are mainly used for pothole detection on roads, and the pavement model is often approximated as a plane, resulting in unsatisfactory detection results.

Summary of the invention

In order to solve the above problems, an object of the present invention is to provide an airport pavement crack detection method based on a depth camera.

In order to achieve the above object, the airport pavement crack detection method based on a depth camera provided by the present invention comprises the following steps performed in sequence:

Step 1) using a depth camera to collect a depth image of the airport pavement, and then dividing the collected depth image into multiple grids;

Step 2) expanding each of the above grids;

Step 3) constructing a road surface model for the expanded grid;

Step 4) connecting the road surface models of the above grids to obtain the overall surface model of the road surface in the depth image;

Step 5) calculating the difference between the depth image obtained in step 1) and the overall surface model obtained in step 4) to obtain candidate crack pixel points;

Step 6) Screen the candidate crack pixel points to obtain the real crack area.

In step 1), the method of using a depth camera to collect a depth image of the airport pavement and then dividing the collected depth image into a plurality of grids is:

First, a depth camera is used to collect a depth image of the airport pavement, and then the collected depth image is divided into n grids _Yi with a size of k*k pixels each, and the grid area is defined as _Yi , i = 1, 2, ..., n.

In step 2), the method for expanding each of the above grids is:

的大小为2k*2k个像素点，将扩充后的网格区域定义为

i＝1,2,…,n。The size of each grid is increased by k/2 pixels around it, so that each expanded grid

The size of is 2k*2k pixels, and the expanded grid area is defined as

i＝1,2,…,n.

In step 3), the method of constructing a road surface model for the expanded grid is:

i＝1,2,…,n，定义P(x,y,z)为位于扩充后的网格

的道面曲面上的点坐标，其中x是深度图像中像素点的横坐标，y是深度图像中像素点的纵坐标，z是道面曲面模型在坐标(x,y)处的深度值；For the expanded grid area

i＝1,2,…,n，define P(x,y,z) as the grid after expansion

The coordinates of the point on the road surface, where x is the horizontal coordinate of the pixel in the depth image, y is the vertical coordinate of the pixel in the depth image, and z is the depth value of the road surface model at the coordinate (x, y);

The pavement surface model of the airport pavement is established using the cubic surface equation.

Let A＝[a ₀ ,a ₁ ,a ₂ ,a ₃ ,a ₄ ,a ₅ ,a ₆ ,a ₇ ,a ₈ ,a ₉ ] ^T ,H＝[1,x,y,x ² ,xy, y ² ,x ³ ,x ² y,xy ² ,y ³ ], then:

z＝H·A

表示由c个像素点的坐标组成的自变量矩阵；则关于未知道面曲面模型参数向量的线性方程组为：Where A is the parameter vector of the road surface model to be solved, and H is the independent variable vector composed of the horizontal and vertical coordinates of the pixel points in the expanded grid _Yi ^* . The present invention adopts the RANSAC algorithm framework and the least squares method to estimate the road surface model. The specific method is as follows: Under the RANSAC algorithm framework, 9 pixels are randomly selected each time to estimate the road surface model, and _Zc = [ _z1 , _z2 , _z3 , ..., _zc ] ^T represents the depth value vector of c pixels, and

represents the independent variable matrix composed of the coordinates of c pixel points; then the linear equations for the unknown surface model parameter vector are:

Z _c =X _c ·A

Then, the least squares method is used to solve the above linear equations to obtain the pavement surface model parameter vector:

求得扩充后的网格

内所有坐标的深度值：Then, the road surface model parameter vector

Get the expanded grid

Depth values of all coordinates within:

表示由扩充后的网格

内所有像素点的坐标组成的自变量矩阵；令Z₀表示原始深度值，则可计算出道面曲面模型与原始深度值之间的最短距离：in

Represents the expanded grid

The independent variable matrix composed of the coordinates of all pixel points in ; let Z ₀ represent the original depth value, then the shortest distance between the road surface model and the original depth value can be calculated:

的道面曲面模型参数向量

If the distance from a certain pixel to the road surface is less than the distance threshold _Ti , the pixel is judged as an interior point, otherwise it is an exterior point; finally, the road surface model with the largest number of interior points is selected as the true model, and the interior point set composed of these interior points is used to re-optimize the estimated expanded grid using the least squares method

The road surface model parameter vector

In step 4), the method of connecting the road surface models of the above-mentioned grids to obtain the overall surface model of the road surface in the depth image is:

i＝1,2,…,n全部计算完毕，得到每个网格的道面曲面模型

i＝1,2,...,n，根据每个网格的道面曲面模型

和深度图像像素点坐标组成自变量向量:Take out the original grid in the road surface fitting model of the expanded grid obtained in step 3) as the road surface model of the original grid; according to the position of the original grid, splice the respective road surface models together to form the overall surface model of the road surface in the depth image. The specific method is: repeat step 3) until the expanded grid area

i＝1,2,…,n All calculations are completed, and the road surface model of each grid is obtained

i＝1,2,...,n, according to the road surface model of each grid

And the depth image pixel coordinates form an independent variable vector:

的道面曲面的深度值：Where i represents the i-th grid, j represents the j-th pixel in the i-th grid, and each coordinate in the depth image is calculated

Depth value of the road surface:

In step 5), the method for calculating the difference between the depth image obtained in step 1) and the overall surface model obtained in step 4) to obtain candidate crack pixel points is:

The depth image obtained in the above step 1) and the overall surface model obtained in step 4) are subtracted at the corresponding positions of the pixels and the absolute value is taken. Let d(x, y) be the absolute value of the above difference at the pixel point (x, y); if the absolute value d(x, y) is greater than a certain set threshold T _d , the pixel point (x, y) is considered to be a candidate crack pixel point, otherwise, the pixel point is a non-crack pixel.

In step 6), the method of screening the candidate crack pixel points to obtain the real crack area is:

则标记该连通区域为真实裂缝区域，否则标记该连通区域为非裂缝区域并删除，其中T_m、T_l和T_r为设定的阈值。The candidate crack pixels are divided into several connected regions according to connectivity. The real crack regions are screened out by calculating the area, length and aspect ratio of each connected region. The specific method is as follows: first, the skeleton of each connected region is extracted, and the number of skeleton pixels is defined as the length of the connected region, denoted as l. Then, the distance from each pixel on the skeleton to the edge of the connected region along its normal direction is calculated, and the average of all distances is taken as the width of the connected region, denoted as w. The total number of pixels in the connected region is denoted as m. If the following conditions are met at the same time: m>T _m and l>T _l and

Then the connected area is marked as a real crack area, otherwise the connected area is marked as a non-crack area and deleted, where T _m , T _l and _Tr are the set thresholds.

In step 5), the threshold value _Td represents the minimum depth of detectable cracks, and according to the inspection requirements of airport pavement, the value is set to 5 mm.

In step 7), the threshold value _Tm is an empirical value, ranging from 30 to 60; the threshold value _Tl is an empirical value, ranging from 20 to 30; the threshold value _Tr is an empirical value, ranging from no less than 7.

Compared with the prior art, the airport pavement crack detection method based on a depth camera provided by the present invention has the following advantages: ① The method of the present invention is applied to airport pavement crack detection, and a high-performance automated airport pavement detection method is used to replace manual operation, which can improve detection accuracy and work efficiency, thereby improving the safety performance of the airport pavement. ② The present invention is not affected by changes in illumination and is more robust to environmental noise. ③ The method of the present invention models the pavement structure more accurately, thereby making crack detection based on pavement model reconstruction more accurate.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of a method for detecting cracks on an airport pavement based on a depth camera provided by the present invention;

Fig. 2 is a schematic diagram of the main steps of the present invention;

Figure 3 shows the crack detection results of the airport pavement.

DETAILED DESCRIPTION

The airport pavement crack detection method based on a depth camera provided by the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

The depth camera emits continuous near-infrared pulses to the target scene, and then uses a sensor to receive the light pulses reflected by the object. By comparing the phase difference between the emitted light pulse and the light pulse reflected by the object, the transmission delay between the light pulses can be calculated and then the distance of the object relative to the transmitter can be obtained, and finally a depth image is obtained. The grayscale value of each pixel in the depth image can be used to characterize the distance of a certain point in the target scene from the depth camera. The crack area on the airport pavement appears in the depth image as a grayscale value greater than the grayscale value of the airport pavement. However, in order to increase friction, the roughness of the runway surface is increased in the design of the airport runway, resulting in the grayscale value of some airport pavement areas in the collected depth image being similar to or even lower than the deep grayscale value of the crack area, making it difficult to directly extract cracks from the depth image. The problem solved by the method of the present invention is the problem of airport pavement crack detection using a depth camera.

As shown in FIG1 , the airport pavement crack detection method based on a depth camera provided by the present invention comprises the following steps performed in sequence:

Step 1) using a depth camera to collect a depth image of the airport pavement, and then dividing the collected depth image into multiple grids;

First, a depth camera is used to collect a depth image of the airport pavement, and then the collected depth image is divided into n grids _Yi, each with a size of k*k pixels, as shown in Figure 2(a). If the value of k is too large, the consistency of the depth image grayscale value model inside the grid will deteriorate, which is not conducive to subsequent surface fitting and crack detection; if the value of k is too small, it may cause the crack pixels to occupy most of the grid area, so subsequent pavement surface fitting cannot be performed. According to multiple experimental tests, k=100 is taken in the experiment of the present invention. The grid area is defined as _Yi , i=1,2,…,n.

Step 2) expanding each of the above grids;

的大小为2k*2k个像素点，如图2(b)所示。扩充后的网格区域定义为

i＝1,2,…,n。Since the grids are independent of each other, the road surface models established by each grid are not connected, resulting in discontinuity of the road surface models of adjacent grids at the edge and reduced accuracy. Therefore, each grid needs to be expanded to strengthen the connection between adjacent grids and improve the accuracy of the road surface model at the grid edge area. The specific method for expanding each of the above grids is to increase the size of each grid by k/2 pixels on all sides, so that each expanded grid

The size of is 2k*2k pixels, as shown in Figure 2(b). The expanded grid area is defined as

i＝1,2,…,n.

Step 3) constructing a road surface model for the expanded grid;

i＝1,2,…,n，定义P(x,y,z)为位于扩充后的网格

的道面曲面上的点坐标，其中x是深度图像中像素点的横坐标，y是深度图像中像素点的纵坐标，z是道面曲面模型在坐标(x,y)处的深度值。For the expanded grid area

i＝1,2,…,n，define P(x,y,z) as the grid after expansion

The coordinates of a point on the road surface, where x is the horizontal coordinate of the pixel in the depth image, y is the vertical coordinate of the pixel in the depth image, and z is the depth value of the road surface model at the coordinate (x, y).

As shown in Figure 2(c), the pavement surface model of the airport pavement is established using the cubic surface equation.

Let A＝[a ₀ ,a ₁ ,a ₂ ,a ₃ ,a ₄ ,a ₅ ,a ₆ ,a ₇ ,a ₈ ,a ₉ ] ^T ,H＝[1,x,y,x ² ,xy, y ² ,x ³ ,x ² y,xy ² ,y ³ ], then:

z＝H·A

中像素点的横纵坐标组成的自变量向量。由于扩充后的网格

中不仅仅包含道面坐标，还可能含有裂缝区域的坐标，因此，本发明采用RANSAC算法框架和最小二乘法来估计道面曲面模型，具体方法为：在RANSAC算法框架下，每次随机选择9个像素点来估计道面曲面模型，令Z_c＝[z₁,z₂,z₃,...,z_c]^T表示c个像素点的深度值向量，令

表示由c个像素点的坐标组成的自变量矩阵。则关于未知道面曲面模型参数向量的线性方程组为：Where A is the parameter vector of the road surface model to be solved, and H is the expanded grid

The independent variable vector is composed of the horizontal and vertical coordinates of the pixel points in the grid.

contains not only the road surface coordinates, but also the coordinates of the crack area. Therefore, the present invention adopts the RANSAC algorithm framework and the least squares method to estimate the road surface model. The specific method is as follows: Under the RANSAC algorithm framework, 9 pixels are randomly selected each time to estimate the road surface model. Let Z _c = [z ₁ , z ₂ , z ₃ , ..., z _c ] ^T represent the depth value vector of c pixels, let

represents the independent variable matrix composed of the coordinates of c pixel points. Then the linear equations for the unknown surface model parameter vector are:

Z _c =X _c ·A

Then, the least squares method is used to solve the above linear equations to obtain the pavement surface model parameter vector:

求得扩充后的网格

内所有坐标的深度值：Then, the road surface model parameter vector

Get the expanded grid

Depth values of all coordinates within:

表示由扩充后的网格

内所有像素点的坐标组成的自变量矩阵。令Z₀表示原始深度值，则可计算出道面曲面模型与原始深度值之间的最短距离：in

Represents the expanded grid

The independent variable matrix composed of the coordinates of all pixel points in . Let Z ₀ represent the original depth value, then the shortest distance between the road surface model and the original depth value can be calculated:

的道面曲面模型参数向量

距离阈值T_i的选取需综合考虑裂缝实际深度以及道面设计的粗糙纹理的深度差异，此处取值为5毫米。If the distance from a certain pixel to the road surface is less than the distance threshold _Ti , the pixel is judged as an interior point, otherwise it is an exterior point; finally, the road surface model with the largest number of interior points is selected as the true model, and the interior point set composed of these interior points is used to re-optimize the estimated expanded grid using the least squares method

The road surface model parameter vector

The selection of the distance threshold _Ti needs to comprehensively consider the actual depth of the crack and the depth difference of the rough texture of the pavement design. The value here is 5 mm.

Step 4) connecting the road surface models of the above grids to obtain the overall surface model of the road surface in the depth image;

i＝1,2,…,n全部计算完毕，得到每个网格的道面曲面模型

i＝1,2,...,n，根据每个网格的道面曲面模型

和深度图像像素点坐标组成自变量向量:Take out the original grid in the road surface fitting model of the expanded grid obtained in step 3) as the road surface model of the original grid; according to the position of the original grid, splice the respective road surface models together to form the overall surface model of the road surface in the depth image, as shown in Figure 2(d). The specific method is: repeat step 3) until the expanded grid area

i＝1,2,…,n All calculations are completed, and the road surface model of each grid is obtained

i＝1,2,...,n, according to the road surface model of each grid

And the depth image pixel coordinates form an independent variable vector:

的道面曲面的深度值：Where i represents the i-th grid, j represents the j-th pixel in the i-th grid, and each coordinate in the depth image is calculated

Depth value of the road surface:

Step 5) calculating the difference between the depth image obtained in step 1) and the overall surface model obtained in step 4) to obtain candidate crack pixel points;

Since there is a significant difference between the distance from the pavement area point to the pavement surface model and the distance from the crack area point to the pavement surface model in the actual data, the depth image obtained in the above step 1) and the overall surface model obtained in step 4) are subtracted according to the corresponding positions of the pixels and the absolute value is taken, and d(x,y) is the absolute value of the above difference at the pixel point (x,y). If the absolute value d(x,y) is greater than a certain set threshold _Td , the pixel point (x,y) is considered to be a candidate crack pixel point, otherwise, the pixel point is a non-crack pixel. The threshold _Td represents the minimum depth of the crack that can be detected. According to the detection requirements of the airport pavement, this value is set to 5 mm. As shown in Figure 2(e).

Step 6) Screening the candidate crack pixel points to obtain the real crack area;

则标记该连通区域为真实裂缝区域，否则标记该连通区域为非裂缝区域并删除，其中T_m、T_l和T_r为设定的阈值。阈值T_m为经验值，取值范围为30～60；阈值T_l为经验值，取值范围为20～30；阈值T_r为经验值，取值范围为不小于7。The obtained candidate crack pixel points are divided into several connected regions according to connectivity, and the real crack regions are screened out by calculating the area, length and aspect ratio of each connected region, as shown in Figure 2(f). The specific method is as follows: first, the skeleton of each connected region is extracted, and the number of skeleton pixels is defined as the length of the connected region, denoted as l. Then, the distance from each pixel point on the skeleton to the edge of the connected region along its normal direction is calculated, and the average value of all distances is taken as the width of the connected region, denoted as w; the total number of pixels in the connected region is denoted as m. If the following conditions are met at the same time: m>T _m and l>T _l and

If the connected area is a real crack area, the connected area is marked as a non-crack area and deleted. _Tm , _Tl and _Tr are the set thresholds. The threshold _Tm is an empirical value with a range of 30 to 60; the threshold _Tl is an empirical value with a range of 20 to 30; the threshold _Tr is an empirical value with a range of not less than 7.

The effect of the airport pavement crack detection method based on a depth camera provided by the present invention can be further illustrated by the following experimental results. The inventor collected a total of 42 depth images of the airport pavement to verify the method of the present invention. The experimental results are shown in Figure 3, where the left image in the same row is the depth image of the airport pavement, and the right image is the crack area extraction result.

Claims (8)

1. A method for detecting cracks on an airport pavement based on a depth camera, characterized in that: the method for detecting cracks on an airport pavement based on a depth camera comprises the following steps performed in sequence: Step 1) using a depth camera to collect a depth image of the airport pavement, and then dividing the collected depth image into multiple grids; Step 2) expanding each of the plurality of grids; Step 3) constructing a road surface model for the expanded grid; Step 4) connecting the road surface models of the expanded grids to obtain an overall surface model of the road surface in the depth image; Step 5) calculating the difference between the depth image obtained in step 1) and the overall surface model obtained in step 4) to obtain candidate crack pixel points; Step 6) Screening the candidate crack pixel points to obtain the real crack area; In step 3), the method of constructing a road surface model for the expanded grid is: For the expanded grid area _Yi ^* , = 1, 2, ..., n, define P(x, y, z) as the coordinates of a point on the road surface of the expanded grid _Yi ^* , where x is the horizontal coordinate of the pixel in the depth image, y is the vertical coordinate of the pixel in the depth image, and z is the depth value of the road surface model at the coordinate (x, y); The pavement surface model of the airport pavement is established using the cubic surface equation. Let A＝[a ₀ ,a ₁ ,a ₂ ,a ₃ ,a ₄ ,a ₅ ,a ₆ ,a ₇ ,a ₈ ,a ₉ ] ^T ,H＝[1,x,y,x ² ,xy, y ² ,x ³ ,x ² y,xy ² ,y ³ ], then: z＝H·A Where A is the parameter vector of the road surface model to be solved, and H is the independent variable vector composed of the horizontal and vertical coordinates of the pixel points in the expanded grid _Yi ^* . The present invention adopts the RANSAC algorithm framework and the least squares method to estimate the road surface model. The specific method is as follows: Under the RANSAC algorithm framework, 9 pixels are randomly selected each time to estimate the road surface model, and _Zc = [ _z1 , _z2 , _z3 , ..., _zc ] ^T represents the depth value vector of c pixels, and

represents the independent variable matrix composed of the coordinates of c pixel points; then the linear equations for the unknown surface model parameter vector are: Z _c =X _c ·A Then the least squares method is used to solve the above linear equations to obtain the parameter vector of the road surface model:

Then, the road surface model parameter vector

Obtain the depth values of all coordinates in the expanded grid _Yi ^* :

in

represents the independent variable matrix composed of the coordinates of all pixels in the expanded grid _Yi ^* ; let _Z0 represent the original depth value, then the shortest distance between the road surface model and the original depth value can be calculated:

If the distance from a certain pixel to the road surface is less than the distance threshold _Ti , the pixel is judged as an interior point, otherwise it is an exterior point; finally, the road surface model with the largest number of interior points is selected as the true model, and the interior point set consisting of all the interior points on the road surface model with the largest number of interior points is used to re-optimize and estimate the road surface model parameter vector of the expanded grid _Yi ^* using the least squares method

2. The airport pavement crack detection method based on a depth camera according to claim 1 is characterized in that: in step 1), the method of using the depth camera to collect a depth image of the airport pavement and then dividing the collected depth image into a plurality of grids is: First, a depth camera is used to collect a depth image of the airport pavement, and then the collected depth image is divided into n grids _Yi with a size of k*k pixels each, and the grid area is defined as _Yi , i = 1, 2, ..., n. 3. The airport pavement crack detection method based on a depth camera according to claim 1, characterized in that: in step 2), the method of expanding each of the plurality of grids is: The size of each grid is increased by k/2 pixels in all directions, so that the size of each expanded grid _Yi ^* is 2k*2k pixels. The expanded grid area is defined as Yi _* ^, i=1, 2, ..., n. 4. The airport pavement crack detection method based on a depth camera according to claim 1, characterized in that: in step 4), the method of connecting the pavement surface model of the expanded grid to obtain the overall surface model of the pavement in the depth image is: Take out the original grid in the road surface fitting model of the expanded grid obtained in step 3) as the road surface model of the original grid; according to the position of the original grid, splice the respective road surface models together to form the overall surface model of the road surface in the depth image. The specific method is: repeat step 3) until all the expanded grid areas _Yi ^* , i = 1, 2, ..., n are calculated, and the road surface model of each grid is obtained.

According to the road surface model of each grid

And the depth image pixel coordinates form an independent variable vector:

Where i represents the i-th grid, j represents the j-th pixel in the i-th grid, and each coordinate in the depth image is calculated

Depth value of the road surface:

5. The airport pavement crack detection method based on a depth camera according to claim 1, characterized in that: in step 5), the method of calculating the difference between the depth image obtained in step 1) and the overall surface model obtained in step 4) to obtain candidate crack pixel points is: The depth image obtained in the above step 1) and the overall surface model obtained in step 4) are subtracted at the corresponding positions of the pixels and the absolute value is taken. Let d(x, y) be the absolute value of the difference between the depth image obtained in step 1) and the overall surface model obtained in step 4) at the pixel point (x, y); if the absolute value d(x, y) is greater than a certain set threshold T _d , the pixel point (x, y) is considered to be a candidate crack pixel point, otherwise, the pixel point is a non-crack pixel. 6. The airport pavement crack detection method based on a depth camera according to claim 1 is characterized in that: in step 6), the method of screening the candidate crack pixel points to obtain the real crack area is: The obtained candidate crack pixel points are divided into several connected regions according to connectivity, and the real crack regions are screened out by calculating the area, length and aspect ratio of each connected region. The specific method is as follows: first, the skeleton of each connected region is extracted, and the number of skeleton pixels is defined as the length of the connected region, recorded as l. Then, the distance from each pixel point on the skeleton to the edge of the connected region along its normal direction is calculated, and the average value of all distances is taken as the width of the connected region, recorded as w; the total number of pixels in the connected region is recorded as m; if the following conditions are met at the same time: m＞T _m and l＞T _l and

Then the connected area is marked as a real crack area, otherwise the connected area is marked as a non-crack area and deleted, where T _m , T _l and _Tr are the set thresholds. 7. The airport pavement crack detection method based on a depth camera according to claim 5 is characterized in that: the threshold value _Td represents the minimum detectable crack depth, and according to the airport pavement detection requirements, the threshold value _Td is set to 5 mm. 8. The airport pavement crack detection method based on a depth camera according to claim 6 is characterized in that: the threshold _Tm is an empirical value, and the value range is 30 to 60; the threshold _Tl is an empirical value, and the value range is 20 to 30; the threshold _Tr is an empirical value, and the value range is not less than 7.

Images