Asphalt pavement base layer crack identification method and system

Abstract

The present application relates to the technical field of road maintenance, and particularly relates to an asphalt pavement base crack identification method and system, which comprises: acquiring an asphalt pavement base image and preprocessing the image to obtain a gray image, then performing edge detection on the gray image by using a Canny operator to obtain all edge connected domains, calculating the corresponding crack similarity based on the contour shape features of each edge connected domain, screening out suspected crack edges and non-crack edges according to the crack similarity, then extracting the HOG feature values of each edge point in the two types of edges and calculating the crack confidence value of the suspected crack edges, determining the suspected crack edges with the crack confidence value greater than a preset threshold as target crack edges, and finally completing crack positioning in the asphalt pavement base image according to the target crack edges to realize accurate identification of the asphalt pavement base cracks. The present application effectively eliminates the noise interference of material particle edges in the asphalt pavement image through the double technical mechanisms of coarse screening of crack similarity and fine screening of crack cluster degree.

Description

A method and system for identifying cracks in the base course of asphalt pavement Technical Field

[0001] This invention relates to the field of road maintenance technology. In particular, it relates to a method and system for identifying cracks in the base course of asphalt pavement. Background Technology

[0002] With the continuous advancement of my country's highway network construction, the high-grade highways built in the early stages have gradually entered a critical stage of maintenance and repair. Asphalt pavement base layer cracks, as a typical structural defect affecting the service quality of highways, have made their detection and identification a core aspect of road maintenance. These cracks are characterized by their high degree of concealment, rapid development, and high repair costs. If they are not identified and addressed promptly and accurately, they can easily lead to structural damage to the pavement, significantly shortening the service life of the highway. Therefore, achieving efficient and accurate identification of asphalt pavement base layer cracks is of great significance for ensuring road traffic safety, reducing maintenance costs, and promoting the precision and intelligent development of preventative road maintenance.

[0003] However, current traditional detection methods are difficult to meet actual maintenance needs: manual inspection is not only inefficient, but also greatly affected by subjective factors such as the experience and sense of responsibility of the inspectors, resulting in a high rate of missed and false detections; detection technology based on two-dimensional images has the problem of insufficient penetration, and cannot effectively penetrate the asphalt surface layer to capture the true state of the base layer cracks, thus failing to achieve the ideal detection effect.

[0004] While existing technologies include schemes for crack region identification using the Canny operator, conventional Canny operators only perform edge detection based on the gradient magnitude of pixels. However, asphalt pavement base images contain numerous material particle edges, which create strong noise fields that severely interfere with the effective identification of crack edges. This significantly reduces the accuracy of crack identification and fails to meet the refined and intelligent requirements of modern road maintenance for detection technology. Summary of the Invention

[0005] To address the problem that existing asphalt pavement base layer crack identification schemes based on the Canny operator suffer from insufficient accuracy due to interference from noise at the edges of material particles in the image, making it difficult to meet the refined and intelligent requirements of modern road maintenance, this invention provides solutions in the following aspects.

[0006] In a first aspect, a method for identifying cracks in the base course of asphalt pavement includes: acquiring an image of the base course of asphalt pavement; preprocessing the image to obtain a grayscale image; performing edge detection on the grayscale image using the Canny operator to obtain the connected components of all edges; calculating the corresponding crack similarity based on the contour shape features of the connected components of each edge; determining whether an edge is a suspected crack edge based on the crack similarity; identifying the connected components of the remaining edges as non-crack edges; extracting the HOG feature values ​​of each edge point in the suspected crack edges and non-crack edges; calculating the crack confidence value of the suspected crack edges based on the HOG feature values; identifying suspected crack edges with crack confidence values ​​greater than a preset confidence threshold as target crack edges; and locating the cracks in the base course image based on the target crack edges to complete the identification of cracks in the base course of asphalt pavement.

[0007] Preferably, the step of obtaining the contour shape features includes:

[0008] For each edge-connected region, traverse all the edge points it contains, calculate the Euclidean distance between any two edge points, and filter out the region with the largest Euclidean distance. The largest Euclidean distance is the longest axis of the edge-connected region.

[0009] Determine the direction perpendicular to the longest axis, obtain the maximum extension distance of the edge points on both sides of the edge connected region relative to the longest axis in the vertical direction, and add the maximum extension distances on both sides to obtain the vertical Euclidean distance of the edge connected region.

[0010] Perform convex hull operation on all edge points contained in the edge connected region to obtain the smallest convex polygon that completely encloses all edge points of the edge connected region. Calculate the area of ​​the smallest convex polygon to obtain the convex hull area of ​​the edge connected region.

[0011] Preferably, the calculation of the crack similarity includes the calculation of a contour deviation from circularity factor, and the calculation method of the contour deviation from circularity factor includes:

[0012] The area of ​​the corresponding maximum circle is derived from the number of edge points of the edge connected region. The ratio of the convex hull area of ​​the edge connected region to the area of ​​the maximum circle is used as the degree of closeness to the circle. The difference between 1 and the degree of closeness to the circle is used as the factor of the degree of deviation of the contour from the circle.

[0013] Preferably, the calculation of the crack similarity includes the calculation of a narrowing factor, and the calculation method of the narrowing factor includes:

[0014] The ratio between the maximum Euclidean distance and the vertical Euclidean distance of the edge connected domains is used as the elongation of the contour. The elongation is exponentially decayed by a negative exponential function. Subtracting the exponential decay result from 1 yields the normalized elongation factor.

[0015] Preferably, the specific method for determining whether a crack edge is a suspected crack edge based on crack similarity is as follows:

[0016] A preset crack threshold is set. If the crack similarity is greater than the preset crack threshold, it is marked as a suspected crack edge; otherwise, if it is less than or equal to the preset crack threshold, it is marked as a non-crack edge.

[0017] Preferably, the specific steps for calculating the crack confidence value include:

[0018] Take any suspected crack edge as the edge to be analyzed, and any edge point of the edge to be analyzed as the target point. Iterate through the absolute value of the difference between the target point and the mean of the HOG feature values ​​of any other suspected crack edge to obtain the first absolute deviation of the target point from any other suspected crack edge. Then, perform standardization processing and take the reciprocal of the standardization as the feature fit of the target point from any other suspected crack edge.

[0019] The absolute value of the difference between the mean HOG feature values ​​of the target point and any other non-crack edge is traversed to obtain the second absolute deviation of the target point from any other non-crack edge. This deviation is then standardized, and the reciprocal of the standardized value is taken as the feature fit of the target point to any other non-crack edge.

[0020] The ratio of the maximum feature fit of the target point with all other suspected crack edges to the maximum feature fit of the target point with all other non-crack edges is used as the clustering degree of the target point with either the suspected crack edge or the non-crack edge.

[0021] Traverse all edge points of the edge to be analyzed, calculate the clustering degree of each edge point one by one, and perform root mean square operation on the clustering degree of all edge points to obtain the crack confidence value of the edge to be analyzed.

[0022] Preferably, the preprocessing includes:

[0023] The asphalt pavement base image is segmented to remove the non-asphalt pavement background area; the segmented image is then subjected to Gaussian filtering for noise reduction; and the denoised image is converted to a grayscale image.

[0024] Secondly, an asphalt pavement base course crack identification system includes: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned asphalt pavement base course crack identification method is implemented.

[0025] The present invention has the following effects:

[0026] 1. This invention extracts three core contour shape features of the edge connected region: convex hull area, maximum Euclidean distance, and vertical Euclidean distance. The convex hull area directly reflects the compactness of the edge region. The convex hull of particle edges is close to a circle, while the convex hull of crack edges is elongated. The combination of maximum Euclidean distance and vertical Euclidean distance quantifies the difference in the length-to-width ratio of the edge. By calculating crack similarity, it can accurately distinguish between thin and elongated crack edges and nearly circular particle edges, quickly filtering particle noise. On this basis, it further extracts the HOG texture features of all edge points and compares the texture distribution patterns of suspected crack edges and non-crack edges by calculating crack clustering, thus verifying the essential attributes of the edge a second time. This effectively eliminates the interference of remaining particle edges that are similar in shape but do not match the texture, completely solving the problem of false detection and false negative detection caused by the traditional Canny operator relying solely on pixel gradient detection, and significantly improving the accuracy of crack recognition.

[0027] 2. This invention calculates crack similarity by accurately quantifying the convex hull shape and elongation of the edge connected regions. Convex hull shape quantification focuses on the geometric differences between cracks that deviate from a circle and particle edges that are close to a circle, while elongation quantification highlights the morphological features of cracks that are thin and long and particle edges that are compact. This dual quantification dimension enables a preliminary and accurate distinction between the two types of edges. Furthermore, by extracting HOG texture features and comparing the feature distribution patterns of suspected crack edges and non-crack edges, crack clustering is calculated, further verifying the essential attributes of the edges from a texture perspective. Through a multi-dimensional feature judgment system that combines geometric shape quantification and texture feature comparison, the differences in noise interference caused by different particle distribution densities and the impact of changes in lighting conditions on edge detection can be effectively offset. This completely overcomes the limitation of single feature judgment being only applicable to specific scenarios. Even in complex road conditions such as dense particles and uneven lighting, it can still stably output accurate recognition results, significantly improving the robustness of the method in practical engineering applications. Attached Figure Description

[0028] Figure 1 is a flowchart of steps S1-S3 in an asphalt pavement base course crack identification method according to an embodiment of the present invention.

[0029] Figure 2 is a structural block diagram of an asphalt pavement base course crack identification system according to an embodiment of the present invention. Detailed Implementation

[0030] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0031] Referring to Figure 1, a method for identifying cracks in the base course of asphalt pavement includes steps S1-S3, as detailed below:

[0032] S1: Obtain the image of the asphalt pavement base course, preprocess the image of the asphalt pavement base course, and obtain a grayscale image.

[0033] A drone equipped with image acquisition equipment is used to capture original images of the asphalt pavement at a preset flight altitude and shooting angle, ensuring that the original images completely cover the asphalt pavement area to be detected and are clear. For the acquired original images, an image segmentation algorithm is used to segment the asphalt pavement area, removing background areas other than the asphalt pavement (such as road shoulders, green belts, surrounding buildings, etc.) to extract the target image containing only the asphalt pavement, ensuring the targeted nature of subsequent detection. Considering that the target image may be affected by factors such as ambient light and equipment vibration during drone acquisition, and may contain noise interference points, Gaussian filtering is performed on the target image. Gaussian functions are used to perform weighted smoothing calculations on image pixels, effectively filtering out the aforementioned noise interference points and improving image quality. To adapt to the image format requirements of the subsequent edge detection algorithm based on the Canny operator, the target image after Gaussian filtering and denoising is converted to grayscale, transforming the color target image into a grayscale image containing only grayscale information. The grayscale image can retain the edge feature information of the asphalt pavement base layer.

[0034] S2: Perform edge detection on the grayscale image using the Canny operator to obtain the connected components of all edges. Calculate the corresponding crack similarity based on the contour shape features of the connected components of each edge, and determine whether it is a suspected crack edge based on the crack similarity. Identify the connected components of the remaining edges as non-crack edges. Extract the HOG feature values ​​of each edge point in the suspected crack edges and non-crack edges. Calculate the crack confidence value of the suspected crack edges based on the HOG feature values. Identify the suspected crack edges with crack confidence values ​​greater than a preset confidence threshold as target crack edges.

[0035] When identifying asphalt pavement cracks using the Canny operator, directly performing edge detection can easily lead to the detection of asphalt particle edges along with crack edges, creating an interfering noise field and causing false crack detections. Therefore, this step uses edge detection, similarity filtering, and clustering verification to achieve accurate extraction of the true crack edges. The specific process is as follows:

[0036] The Canny operator is characterized by its accurate edge detection capability, making it suitable for preliminary edge detection of asphalt pavement base layers. Specifically, the Canny operator is used to perform edge detection on grayscale images, extracting edge points, grouping interconnected edge points into a single connected component, and counting the total number of edge connected components in the image. At the same time, the first The number of edge points contained in each edge connected region is denoted as . .

[0037] The edge connectivity of cracks typically exhibits a thin and elongated shape, while the edge connectivity of asphalt particles tends to be more circular. Based on this morphological difference, initial screening can be performed using crack similarity. Simultaneously, to avoid the limitations of a single threshold judgment, HOG feature values ​​are further combined to calculate crack clustering, achieving more rigorous feature verification. The specific process is as follows:

[0038] First, perform convex hull operation on the edge connected region containing all edge points to obtain the smallest convex polygon (i.e., the convex hull) that completely encloses all edge points of the edge connected region. Calculate the area of ​​the smallest convex polygon to obtain the convex hull area of ​​the edge connected region. The convex hull area reflects the compactness and spatial occupancy characteristics of the edge connected region's outline. The edge connected region of cracks, due to its thin and elongated shape, has a smaller convex hull area corresponding to the number of edge points in terms of space ratio. On the other hand, the edge connected region of asphalt material particles, due to its more circular outline, has a larger convex hull area corresponding to the number of edge points in terms of space ratio.

[0039] Traverse all edge points within the edge connected region and calculate the Euclidean distance (i.e., the straight-line distance between any two edge points). Select the maximum value among all Euclidean distances and define it as the maximum Euclidean distance of the edge connected region. The line connecting the two edge points corresponding to the maximum Euclidean distance forms the longest axis of the edge connected region. The maximum Euclidean distance characterizes the longest extension dimension of the edge connected region, directly reflecting the length characteristics of the connected region in a two-dimensional image. Cracks in the base layer of asphalt pavement, being linear defects, have a significantly longer longest axis in their edge connected regions.

[0040] Using the longest axis corresponding to the maximum Euclidean distance as a reference, a direction perpendicular to the longest axis is determined. Along this vertical direction, the maximum extension distances of the edge points on both sides of the edge connected region relative to the longest axis are measured. These two maximum extension distances are then added together to obtain the vertical Euclidean distance of the edge connected region. The vertical Euclidean distance characterizes the total width of the edge connected region in the direction perpendicular to the longest axis, directly reflecting the thickness of the connected region. Crack edge connected regions, due to their narrow shape, have limited extension distances in the direction perpendicular to the longest axis, resulting in smaller vertical Euclidean distances. In contrast, asphalt particle edge connected regions, with their nearly circular outlines, have vertical extension distances similar to the longest axis dimension, resulting in larger vertical Euclidean distances.

[0041] Using the number of edge points in the edge-connected domain, derive the area of ​​the corresponding maximum circle. If the edge points are uniformly distributed in a circle (a typical outline of asphalt particles), the circumference of the circle is... ,but ; Circular area The maximum area when the outline is most compact (circular) with the same number of edge points.

[0042] The ratio of the convex hull area of ​​the edge connected region to the area of ​​the largest circle is used as the degree of closeness to the circle. Subtracting the degree of closeness to the circle from 1 yields the factor of the degree of deviation of the contour from the circle.

[0043] The elongation of the contour is obtained by calculating the ratio between the maximum Euclidean distance and the vertical Euclidean distance of the edge connected domains. The elongation is then exponentially decayed by a negative exponential function and normalized to obtain the normalized elongation factor.

[0044] The crack similarity is calculated by multiplying the factor of the degree of deviation of the outline from the circle by the factor of the degree of elongation.

[0045] The crack similarity of each edge connected region is calculated based on the maximum Euclidean distance and the perpendicular Euclidean distance. Specifically, the crack similarity satisfies the following relationship:

[0046] ;

[0047] In the formula, Indicates the first Crack similarity of edge-connected domains Indicates the first The area of ​​the convex hull of each edge-connected region. Indicates the first The number of edge points contained in each edge connected component Indicates the first The maximum Euclidean distance of each edge-connected region Indicates the first The vertical Euclidean distance of each edge-connected domain.

[0048] Indicates the first The largest circle area that can be enclosed by the number of edge points of a connected edge region. The aspect ratio of the edge connected domain is quantified to reflect the degree of narrowness. Since cracks are linear structural diseases, the longest axis of their edge connected domain is much larger than the vertical axis, resulting in a significantly larger aspect ratio.

[0049] If the crack similarity is greater than a preset crack threshold, it is marked as a suspected crack edge; otherwise, if it is less than or equal to the preset crack threshold, it is marked as a non-crack edge.

[0050] Using a fixed threshold for crack similarity alone to screen suspected crack edges is insufficient to avoid misjudgment risks in specific scenarios. For example, the edges of irregularly shaped material particles that resemble cracks may be mistakenly included as suspected crack edges because they meet the similarity threshold; simultaneously, some weak crack edges may be missed because their similarity is slightly below the threshold. To improve recognition accuracy, a secondary verification is needed, combining morphological screening with the clustering of texture features of edge points. By judging the consistency of features between suspected crack edges and those of most suspected crack edges, and their differences from non-crack edges, precise crack edge identification can be achieved. The specific steps are as follows:

[0051] For each edge point in all connected edge regions, including both suspected and non-crack edges, HOG (Histogram of Oriented Gradients) feature values ​​are extracted to quantify the local texture features of the edge points. Among these, the ... In the nth edge connected component The HOG feature values ​​of the edge points are denoted as follows: HOG feature values ​​can accurately depict the gradient direction distribution of pixels around edge points, and are the core texture basis for distinguishing different types of edge crack edges from grain edges.

[0052] Regarding the first For each suspected crack edge, the HOG eigenvalues ​​of all edge points are statistically analyzed, and the mean and standard deviation of the HOG eigenvalues ​​for the suspected crack edge are calculated. The mean reflects the central tendency of the texture features of all edge points within the suspected crack edge, while the standard deviation reflects the dispersion of the texture features of the edge points within the edge.

[0053] Regarding the first For each non-crack edge, the HOG eigenvalues ​​of all edge points are statistically analyzed, and the mean and standard deviation of the HOG eigenvalues ​​for the non-crack edges are calculated. The mean reflects the central tendency of the texture features within the non-crack edges, while the standard deviation reflects the dispersion of the texture features within the non-crack edges.

[0054] Using any suspected crack edge as the edge to be analyzed, and any edge point on that edge as the target point, the crack confidence value of the edge to be analyzed is calculated through the following steps. This confidence value is used to quantify the reliability that the edge to be analyzed is a real crack. The specific steps are as follows:

[0055] The process iterates through the target point and every other suspected crack edge. It's important to note that other suspected crack edges do not include the edge being analyzed itself, to avoid interference from their own features in the comparison results. The absolute value of the difference between the HOG feature value of the target point and the average HOG feature value of the currently visited suspected crack edges is calculated. This absolute value is defined as the first absolute deviation. The physical meaning of this first absolute deviation is the degree of absolute difference between the local gradient features (HOG features) of the target point and the average overall gradient features of the currently visited suspected crack edges. The larger the value, the more significant the feature difference between the two.

[0056] The first absolute deviation is standardized using the standard deviation of the HOG eigenvalues ​​of the currently traversed suspected crack edges. Due to differences in the physical morphology of different suspected crack edges (e.g., some cracks have large width variations and small branches, resulting in high feature dispersion; some cracks have regular shapes and more concentrated features), directly using the first absolute deviation cannot objectively compare the feature correlation between the target point and different suspected crack edges. The standard deviation, as a core statistic characterizing feature dispersion, reflects the natural fluctuation range of the HOG features of the current suspected crack edges. Through standardization, absolute differences can be transformed into relative differences, eliminating the interference of heterogeneous feature distributions among different suspected crack edges and making the standardization results horizontally comparable.

[0057] The reciprocal of the standardized first absolute deviation is used as the feature fit between the target point and the currently traversed suspected crack edge. The physical meaning of feature fit is: the degree of consistency between the features of the target point and the current suspected crack edge. Feature fit is negatively correlated with the standardized first absolute deviation. The smaller the standardized deviation, the larger the feature fit, indicating that the target point fits the feature distribution pattern of the current suspected crack edge more closely, and the higher the probability that it belongs to the same type of crack group.

[0058] Traverse the target point and every other non-crack edge, calculate the absolute value of the difference between the HOG feature value of the target point and the mean HOG feature value of the currently traversed non-crack edge, and define this absolute value as the second absolute deviation. Its physical meaning is consistent with the first absolute deviation, that is, the degree of absolute difference between the overall characteristics of the target point and the current non-crack edge (such as the edge of asphalt particles).

[0059] The second absolute deviation is standardized using the standard deviation of the HOG eigenvalues ​​of the currently traversed non-crack edges. Similarly, the physical morphology of non-crack edges (such as differences in particle size and shape) will also cause discrete differences in their HOG characteristics. Standardization using the standard deviation can eliminate the interference of these differences, transforming the second absolute deviation into a comparable relative difference index.

[0060] The reciprocal of the standardized second absolute deviation is taken as the feature fit between the target point and the currently traversed non-crack edge. The feature fit is used to characterize the degree of feature consistency between the target point and the non-crack edge; the larger the value, the more likely the target point belongs to the non-crack noise group.

[0061] From the feature fit between the target point and all other suspected crack edges, the maximum value is selected and recorded as the maximum feature fit of the suspected crack. This value reflects the highest feature consistency level between the target point and all suspected crack edges. From the feature fit between the target point and all other non-crack edges, the maximum value is selected and recorded as the maximum feature fit of the non-crack edge. This value reflects the highest feature consistency level between the target point and all non-crack edges.

[0062] The ratio of the maximum feature fit of the suspected crack to the maximum feature fit of the non-crack is calculated, and this ratio is defined as the clustering degree of the target point. The physical meaning of clustering degree is the feature attribution tendency of the target point: when the clustering degree is greater than 1, it indicates that the target point's fit with crack features is higher than its fit with non-crack features, and the larger the ratio, the more significant the attribution tendency; when the clustering degree is less than 1, it indicates that the target point is more biased towards non-crack noise features. Selecting the maximum feature fit for ratio calculation can avoid the random errors caused by distortion of single edge features (such as feature anomalies caused by local noise), ensuring the reliability of the comparison results.

[0063] Since the core characteristic of the target crack is that the features of all edge points are consistent across the group, it is necessary to comprehensively analyze the judgment results of all edge points of the edge to be analyzed. The specific operation is as follows:

[0064] The algorithm iterates through all edge points on the edge to be analyzed, calculating the clustering degree of each point. A root mean square (RMS) operation is then performed on the clustering degrees of all edge points; the result is the crack confidence value of the edge to be analyzed. The advantage of the RMS operation is that it integrates the clustering degree information of all edge points and assigns higher weights to edge points with high clustering degrees (edge ​​points with clearly defined characteristics), effectively highlighting the overall consistency of the target crack's features while suppressing interference from local anomalous edge points. A higher crack confidence value indicates that all edge points of the edge to be analyzed generally match the crack characteristics more closely, and the higher the reliability of it being a real crack.

[0065] In this embodiment, the preset confidence threshold is 1, and the suspected crack edge corresponding to the crack confidence value being greater than the preset confidence threshold is marked as the target crack edge.

[0066] S3: Locate the asphalt pavement base layer image based on the edge of the target crack to complete the identification of asphalt pavement base layer cracks.

[0067] Traverse all connected components that are identified as the edge of the target crack, extract the pixel coordinates of all edge points in each connected component, and form the coordinate set corresponding to each target crack edge; at the same time, record the key coordinate parameters of each connected component.

[0068] For the coordinate set of a single target crack edge, adjacent edge points are sequentially connected according to pixel connectivity rules to reconstruct a complete and continuous crack outline. If the coordinate ranges of multiple adjacent target crack edge connected regions overlap, or the spacing is less than a preset pixel threshold (in this embodiment, it is set to 3 pixels), the corresponding coordinate sets are merged and integrated to form a single complete crack outline, avoiding segmentation recognition problems caused by local crack breaks and ensuring the integrity of the crack outline.

[0069] In the original asphalt pavement base layer image, a preset visual marking method is used to intuitively present the location and attributes of cracks. Specifically, this includes drawing the crack outline with a red solid line, marking the minimum bounding area of ​​the crack with a green rectangle, and labeling key parameters such as crack length, width, and center coordinates. At the same time, a structured recognition result is output, which includes the number of cracks, the core parameters corresponding to each crack (pixel coordinate range, actual physical size, geographic coordinates), and recognition confidence (based on the calculation of crack similarity and clustering). Finally, a standardized dataset that can be directly used for road maintenance analysis is formed, thus fully realizing the identification process of cracks in the asphalt pavement base layer.

[0070] This invention also provides an asphalt pavement base course crack identification system. As shown in Figure 2, the system includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement an asphalt pavement base course crack identification method according to the first aspect of this invention. The system also includes other components well-known to those skilled in the art, such as a communication bus and a communication interface. Their configuration and functions are known in the art and will not be described further here.

[0071] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for identifying cracks in the base course of asphalt pavement, characterized in that, include: Acquire images of the asphalt pavement base course, and preprocess the asphalt pavement base course images to obtain grayscale images; Edge detection is performed on the grayscale image using the Canny operator to obtain the connected components of all edges. Based on the contour shape features of the connected components of each edge, the corresponding crack similarity is calculated, and whether it is a suspected crack edge is determined based on the crack similarity. The connected components of the remaining edges are determined as non-crack edges. The HOG feature values ​​of each edge point in the suspected crack edge and non-crack edge are extracted. The crack confidence value of the suspected crack edge is calculated based on the HOG feature value. The suspected crack edge with a crack confidence value greater than the preset confidence threshold is determined as the target crack edge. The identification of cracks in the asphalt pavement base course is achieved by locating the crack edges in the base course image. The specific steps for calculating the crack confidence value include: taking any suspected crack edge as the edge to be analyzed, and any edge point of the edge to be analyzed as the target point; iterating through the absolute values ​​of the differences between the target point and the mean of the HOG feature values ​​of any other suspected crack edge to obtain the first absolute deviation of the target point from any other suspected crack edge; standardizing this deviation; and using the reciprocal of the standardized value as the feature fit of the target point to any other suspected crack edge; iterating through the mean of the HOG feature values ​​of the target point and any other non-crack edge... The absolute value of the difference is used to obtain the second absolute deviation of the target point from any other non-crack edge. This deviation is then standardized, and the reciprocal of the standardized value is used as the feature fit of the target point to any other non-crack edge. The ratio between the maximum feature fit of the target point to all other suspected crack edges and the maximum feature fit of the target point to all other non-crack edges is used as the clustering degree of the target point with the suspected crack edge or belonging to the non-crack edge. All edge points of the edge to be analyzed are traversed, and the clustering degree of each edge point is calculated one by one. The root mean square operation is performed on the clustering degree of all edge points to obtain the crack confidence value of the edge to be analyzed.

2. The method for identifying cracks in the base course of asphalt pavement according to claim 1, characterized in that, The steps for obtaining the contour shape features include: for each edge connected region, traversing all the edge points it contains, calculating the Euclidean distance between any two edge points, filtering out the maximum Euclidean distance, which is the longest axis of the edge connected region; determining the direction perpendicular to the longest axis, obtaining the maximum extension distance of the edge points on both sides of the edge connected region relative to the longest axis in the vertical direction, adding the maximum extension distances on both sides to obtain the vertical Euclidean distance of the edge connected region; performing convex hull operation on all the edge points contained in the edge connected region to obtain the minimum convex polygon that completely encloses all the edge points of the edge connected region, calculating the area of ​​the minimum convex polygon to obtain the convex hull area of ​​the edge connected region.

3. The method for identifying cracks in the base course of asphalt pavement according to claim 2, characterized in that, The calculation of the crack similarity includes the calculation of the contour deviation from the circle factor. The calculation method of the contour deviation from the circle factor includes: deriving the corresponding maximum circle area based on the number of edge points of the edge connected domain, taking the ratio of the convex hull area of ​​the edge connected domain to the maximum circle area as the degree of closeness to the circle, and taking the difference between 1 and the degree of closeness to the circle as the contour deviation from the circle factor.

4. The method for identifying cracks in the base course of asphalt pavement according to claim 2, characterized in that, The calculation of the crack similarity includes the calculation of the elongation factor. The calculation method of the elongation factor includes: taking the ratio between the maximum Euclidean distance and the vertical Euclidean distance of the edge connected domain as the elongation of the contour, exponentially decaying the elongation through a negative exponential function, and subtracting the exponential decay result from 1 to obtain the normalized elongation factor.

5. The method for identifying cracks in the base course of asphalt pavement according to claim 1, characterized in that, The specific method for determining whether a crack edge is a suspected crack edge based on crack similarity is as follows: a crack threshold is preset. If the crack similarity is greater than the preset crack threshold, it is marked as a suspected crack edge; otherwise, if it is less than or equal to the preset crack threshold, it is marked as a non-crack edge.

6. The method for identifying cracks in the base course of asphalt pavement according to claim 1, characterized in that, The preprocessing includes: segmenting the asphalt pavement base image into pavement regions and removing non-asphalt pavement background regions; performing Gaussian filtering on the segmented image for noise reduction; and converting the denoised image into a grayscale image.

7. A system for identifying cracks in the base course of asphalt pavement, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the asphalt pavement base course crack identification method according to any one of claims 1-6.

Images