Two-dimensional and three-dimensional pavement crack recognition method based on enhanced depth edge features

Abstract

The present invention provides a two-dimensional and three-dimensional road crack recognition method based on enhanced deep edge features, including: reading three-dimensional road data / two-dimensional road data acquired by a road acquisition system; data preprocessing; removing the information of three-dimensional road data posture fluctuation / two-dimensional road data uneven illumination; edge enhancement data acquisition; overall nested edge detection deep network training, and obtaining the edge probability map of the test data according to the training model; S6, converting and processing the edge probability map to obtain strong edge information of the road data; S7, selecting and connecting strong edges; S8, extracting crack attribute information of the two-dimensional / three-dimensional road surface. The present invention integrates the frequency component characteristics of the road cross section and the multi-scale edge characteristics acquired by the deep learning network, enhances the edge of the road crack, and is helpful for the accurate extraction of cracks in two-dimensional and three-dimensional road data. It is suitable for the automatic recognition of cracks in a variety of typical roads acquired by two-dimensional and three-dimensional data acquisition systems.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic detection and identification of road damage, and in particular to a two-dimensional and three-dimensional road crack identification method based on enhanced depth edge features. Background Art

[0002] With the development of road pavement damage measurement equipment and automatic detection technology, vehicle-mounted road pavement measurement systems can obtain more and more two-dimensional and three-dimensional road surface data, including data of different road surface backgrounds and different crack types. Two-dimensional optical road surface data acquisition equipment is economical, mature in technology, and suitable for the introduction of typical machine learning methods, but it is easily affected by lighting, shadows, tire wear marks and oil stains on the road surface in actual applications. Line scanning three-dimensional road surface detection method can overcome the interference of uneven lighting and shadows in actual road surface detection projects, and can obtain three-dimensional information of road surface damage, which is the current development trend. However, high-precision three-dimensional road surface data contains serious vehicle posture fluctuation information in the actual dynamic acquisition environment, and the three-dimensional data contains elevation information of road surface damage, road surface texture, marking lines and other road surface targets. Compared with the rapid development of two-dimensional optical road surface damage detection algorithms, the current three-dimensional road surface damage detection technology is still relatively lagging behind. Cracks, as the most common type of damage in road surface damage, are the most concerned detection indicators in actual road maintenance and detection engineering applications. Since two-dimensional road surface data and three-dimensional road surface data have certain advantages in actual highway detection engineering applications, they are currently playing a role in specific detection projects. However, the current methods for crack detection in two-dimensional and three-dimensional pavement are independent of each other, and there is a lack of methods that can simultaneously process crack detection in two-dimensional and three-dimensional pavement data.

[0003] Due to the massive big data characteristics of two-dimensional and three-dimensional pavement data in actual pavement crack detection engineering applications, the use of machine learning methods to quickly and automatically detect crack locations and attribute information from massive pavement data has important theoretical and application value. Since two-dimensional optical pavement data is more in line with the data set requirements of computer vision machine learning models, the current two-dimensional optical pavement machine learning deep learning methods have been widely studied and applied in actual production. However, it is difficult for machine learning to directly obtain good results in the application of crack detection in online scanning three-dimensional pavement data. On the one hand, the line scanning three-dimensional data contains interference from factors such as driving posture and deformation diseases. The actual crack detection task requires a high degree of robustness of the method, and requires data suitable for different types of cracks and different pavement backgrounds; on the other hand, due to the large difference in crack characteristics between three-dimensional data and two-dimensional data, three-dimensional pavement annotation is generally more difficult to obtain, and traditional supervised machine learning methods and even deep learning methods have limited applicability to cracks in different data and different pavements.

[0004] It should be noted that the above introduction to the technical background is only for the convenience of providing a clear and complete description of the technical solutions of the present application and for the convenience of understanding by those skilled in the art. It cannot be considered that the above technical solutions are well known to those skilled in the art simply because they are described in the background technology section of the present application. Summary of the invention

[0005] The purpose of the present invention is to provide a two-dimensional and three-dimensional pavement crack automatic detection method based on enhanced depth edge features to address the deficiencies in the above-mentioned background technology, so as to fully enhance and utilize the continuous edge characteristics of cracks in pavement data, and can be suitable for crack detection applications of various typical pavement cracks extracted from two-dimensional and three-dimensional pavement data.

[0006] In order to achieve the above object, the present invention provides a two-dimensional and three-dimensional pavement crack recognition method based on enhanced depth edge features, comprising the following steps:

[0007] S1, reading in the 3D road surface data / 2D road surface data acquired by the road surface acquisition system;

[0008] S2, data preprocessing;

[0009] S3, removing the undulation of 3D road surface data and the uneven illumination of 2D road surface data;

[0010] S4, edge-enhanced data acquisition;

[0011] S5, overall nested edge detection deep network training, obtaining the edge probability map of the test data according to the training model;

[0012] S6, transforming the edge probability map to obtain strong edge information of the road surface data;

[0013] S7, strong edge selection and connection;

[0014] S8, extracting crack attribute information of the two-dimensional / three-dimensional road surface.

[0015] Furthermore, the two-dimensional road surface data is road surface optical data acquired using a vehicle-mounted two-dimensional camera, and the three-dimensional road surface data is acquired by splicing a series of road surface cross-sectional profiles together using a line scanning three-dimensional measurement sensor to collect a series of road surface cross-sectional profiles along the measurement direction.

[0016] Furthermore, it is characterized in that in S2, abnormal values ​​caused by system abnormalities and environmental abnormalities are corrected, the cross-sectional contour is converted from image space to object space through a calibration file, the cross-sectional data of the object space is obtained, and the system errors in the measurement system are corrected.

[0017] Furthermore, in S3, the components of each cross section of the acquired three-dimensional road surface data / two-dimensional road surface data are decomposed to remove the influence of low-frequency components and enhance the high-frequency edge features of cracks.

[0018] Furthermore, in S4, the high-frequency components of each cross section are spliced ​​to form high-frequency component data with enhanced edge characteristics, and the edge-enhanced data are uniformly converted into a grayscale image with a value distribution in the range of 0-255.

[0019] Furthermore, in S5, the data obtained in S4 is input into the overall nested edge detection deep network to obtain the edge probability map. The overall nested edge detection deep network is trained by an open source two-dimensional pavement dataset and corresponding crack annotations. The dataset input in the training process is also processed by steps S3 and S4.

[0020] Furthermore, in S6, the acquired edge probability map is converted into a grayscale map with values ​​distributed in the range of 0-255, and a threshold value Th1=3*Ave is obtained in combination with the edge map mean value Ave.

[0021] Furthermore, in S7, based on the acquired and transformed edge probability map and threshold Th1, weak edges below the threshold are removed, strong edges not less than the threshold are retained, the connected domain of the selected strong edges is used as the image edge object, the image edge object position and its direction are used as the input of the tensor voting algorithm, and the tensor encoding of the edge object area and object direction is combined to obtain the tensor significance field of the input edge object data. Combined with the more significant line segments in the acquired tensor significance field, possible connection results are constructed, the connection results are combined with the strong edge map, and the connected domain area parameter is used to delete the edge map with a smaller connected domain.

[0022] Furthermore, in S8, based on the crack edge connection result obtained in S7 and the high-frequency component data obtained in S4, the crack area and severity attribute information are obtained by using the crack object region attribute characteristics.

[0023] Furthermore, based on the final acquired position of the inferred crack, the edge sub-object position and its connected domain information are correspondingly obtained; for two-dimensional data, the length and area attributes of the crack are determined according to the connected domain characteristics of the crack edge object, and the severity of the crack is determined in combination with the grayscale of the high-frequency component data; for three-dimensional data, the length and area attributes of the crack are determined according to the connected domain characteristics of the crack edge object, and the depth information of the crack is determined in combination with the elevation map of the high-frequency component data, and then the severity of the crack and its distribution attributes are determined.

[0024] The above scheme of the present invention has the following beneficial effects:

[0025] The two-dimensional and three-dimensional pavement crack recognition method based on enhanced deep edge features provided by the present invention combines the two-dimensional and three-dimensional pavement data obtained in the actual application environment, and can be applied to the automatic recognition of cracks in various typical pavements such as asphalt pavement, cement pavement, coarse texture, fine texture and ordinary texture pavement obtained by two-dimensional and three-dimensional data acquisition systems; the pavement cross-section frequency component characteristics and the multi-scale edge characteristics obtained by the deep learning network are integrated to enhance the pavement crack edges, which can remove the submerged influence of the pavement data acquisition system on the crack edge characteristics due to uneven illumination (two-dimensional pavement data situation) and vehicle posture fluctuations (three-dimensional pavement data situation); on the other hand, it can take into account the edge characteristics of cracks of different thicknesses and depths, which is helpful for the accurate extraction of cracks in two-dimensional and three-dimensional pavement data;

[0026] In the present invention, the pavement crack edge characteristics are enhanced after cross-section component decomposition and enhancement. Based on the enhanced component data and a small amount of crack annotation data, a deep learning edge detection network is trained. Based on this deep learning edge detection network, the crack edge probability results of the two-dimensional pavement enhancement data and the three-dimensional pavement enhancement data can be processed simultaneously, so that the existing two-dimensional pavement annotation data set is used to train the deep learning model, overcome the defect that it is difficult to train the network due to insufficient pavement annotation samples, and obtain accurate and comprehensive two-dimensional and three-dimensional pavement crack edge probability information under limited annotation conditions, which can provide a robust crack detection method for actual highway pavement detection applications;

[0027] Other beneficial effects of the present invention will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a flow chart of the steps of the present invention;

[0029] Figure 2 A schematic diagram of the overall nested edge detection deep network (HED) structure of the present invention and the process of HED processing road edge enhancement data to obtain edge feature maps;

[0030] Figure 3 The tensor diagram and edge object level tensor direction encoding results of the present invention, (a) voting generation; (b) two-dimensional stick tensor significance field; (c) HED strong edge object diagram; (d) data tensor significance field obtained by tensor encoding combined with edge object characteristics; (e) HED strong edge connection result based on tensor voting results;

[0031] Figure 4This is an example of high-precision line scanning 3D pavement crack recognition obtained by the present invention, (a) pre-processed 3D pavement data; (b) edge enhancement result after removing low-frequency components after frequency component decomposition; (c) edge probability result obtained by HED deep network; (d) crack recognition result after tensor voting and morphological post-processing; (e) crack attribute result obtained based on edge-enhanced pavement base map and crack recognition;

[0032] Figure 5 This is an example of two-dimensional pavement crack recognition obtained by the present invention, (a) two-dimensional pavement data after preprocessing; (b) edge enhancement result after removing uneven illumination after frequency component decomposition; (c) edge probability result obtained by HED deep network; (d) crack recognition result after tensor voting and morphological post-processing; (e) crack attribute result obtained based on edge enhanced pavement base map and crack recognition result. DETAILED DESCRIPTION

[0033] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0034] In the description of the present invention, for the sake of simplicity, the method or rule is depicted or described as a series of operations, and its purpose is neither to exhaustively list the experimental operations nor to limit the order of the experimental operations. For example, the experimental operations can be performed in various orders and / or simultaneously, and include other experimental operations that are not described again. In addition, the steps described are not all necessary for the methods and algorithms described herein. Those skilled in the art will recognize and understand that these methods and algorithms can be represented as a series of unrelated states through state diagrams or projects.

[0035] like Figure 1 As shown, embodiment 1 of the present invention provides a two-dimensional and three-dimensional pavement crack recognition method based on enhanced depth edge features, which specifically includes the following steps:

[0036] S1, read in the 3D road data / 2D road data acquired by the road acquisition system. The 2D road data refers to the road optical data acquired by the vehicle-mounted 2D camera. The 3D road data refers to the 3D data acquired by splicing a series of road section profiles along the measurement direction using a line scanning 3D measurement sensor to obtain the 3D data.

[0037] In this embodiment, the two-dimensional road surface data is obtained by an optical camera to reflect the brightness information of the road surface; the three-dimensional road surface data is obtained by a line scanning three-dimensional measurement sensor, and the relative elevation of the surface of the measured object is measured based on the triangulation principle. The obtained three-dimensional data can reflect the elevation information of the measured road surface. The cross-sectional accuracy of the two-dimensional road surface data involved in this embodiment is not less than 2mm, the cross-sectional accuracy of the three-dimensional data involved is not less than 1mm, and the elevation resolution is higher than 0.5mm.

[0038] S2, data preprocessing. Correct the abnormal values ​​(zero value and convex burrs) caused by system abnormalities and environmental abnormalities, convert the cross-sectional profile from image space to object space through the calibration file, and obtain the cross-sectional data of the object space to correct the systematic errors in the measurement system.

[0039] Due to the interference of the measurement environment (water stains, oil stains on the road surface or foreign objects in the measured area), the collected data may contain a small amount of abnormal noise (zero value points). In order to reduce the false detection of the crack edge caused by these zero value points, this embodiment replaces the abnormal noise points with non-abnormal sampling points in the cross section after mean filtering to obtain the pre-processed road section data. The actual vehicle-mounted road data acquisition system has system errors such as sensor installation angle and uneven light intensity distribution. These system errors will weaken the characteristics of the road crack target, so the data collected by the measurement sensor needs to be corrected through the calibration file.

[0040] S3: Removal of attitude fluctuations in 3D road data / uneven illumination information in 2D road data. When the vehicle-mounted system collects road data, due to the impact of the bumps and undulations of driving and deformation diseases on the road surface, the collected original 3D data contains obvious low-frequency amplitude attitude fluctuation information, and cracks are hidden in these macroscopic change information. Due to the uneven illumination and shadow of the road surface, as well as the impact of potholes, deformation diseases, etc., the 2D road surface also has obvious uneven brightness, and the crack edge information is easily disturbed by this low-frequency uneven grayscale distribution.

[0041] Therefore, in order to reduce the impact of such low-frequency fluctuations in road surface data on crack location and attribute information extraction, it is necessary to use relevant algorithms to remove the low-frequency attitude fluctuation components contained in the data, such as the patent CN108765376A, which only retains the sum of sparse components and vibration components as high-frequency components. The above processing can reduce the impact of driving fluctuations, uneven lighting and road surface deformation on crack detection without losing crack information, so as to facilitate crack edge feature enhancement and attribute information extraction.

[0042] Specifically, for the three-dimensional road surface data obtained by S2, each cross-section is decomposed into components to remove the influence of low-frequency components caused by the posture fluctuations of the vehicle during dynamic collection and the smooth deformation of the road surface, thereby enhancing the high-frequency edge characteristics of the cracks; for the two-dimensional road surface data obtained by S2, each cross-section is decomposed according to the frequency components to remove the influence of low-frequency components caused by uneven lighting and shadows on the road surface, thereby enhancing the high-frequency edge characteristics of the cracks.

[0043] S4, edge enhancement data acquisition. The road surface data component analysis is performed on the basis of the road surface cross-section data, and the high-frequency components of each cross-section are spliced ​​to form high-frequency component data with enhanced edge characteristics as input for subsequent processing.

[0044] In order to make the acquired edge enhanced data better connected with the deep learning network, the edge enhanced data is uniformly converted into a grayscale image with a value distribution in the range of 0-255.

[0045] S5, overall nested edge detection deep network training, obtain the test data edge probability map according to the training model. The data obtained in S4 is input into the overall nested edge detection deep network to obtain the edge probability map. The deep network is trained by the open source two-dimensional pavement dataset and the corresponding crack annotations. The dataset input in the training process is also processed by steps S3 and S4 to enhance the crack edge characteristics of the input data.

[0046] Among them, the open source road data sets participating in pre-training contain corresponding annotations, including: two-dimensional asphalt road data set FISSURESDataset, two-dimensional asphalt road data set DeepCrack Dataset, ESAR two-dimensional road data set, and LCMS three-dimensional road data set.

[0047] The overall nested edge detection deep network (HED) structure is as follows Figure 2 As shown in the figure, it is improved based on the 16-layer visual group network (VGG). The output layer is connected to the convolution layer, and the last pooling layer and all the subsequent connection layers of VGG are removed. The training and prediction of the overall image can increase the perception of the feature map and alleviate the problem of low resolution of the feature map. At the same time, HED belongs to a multi-level and multi-scale network and is robust to noise. HED connects a side output layer to the last convolution layer of each submodule (stage). The edge features contained in different output layers are different. The side output image is deconvolved to make the size of each side output layer consistent with the input image. Finally, the feature fusion layer is used to fuse the multi-scale edge features to form a fused edge output layer. For actual road data, edge pixels and non-edge pixels are seriously unbalanced. HED uses a class balance weight strategy to automatically balance the loss between edge / non-edge classes.

[0048] HED can obtain the edge feature map of the road surface data output by S4. Due to the differences in the thickness and depth of the cracks themselves, as well as the interference of the road surface texture background, the edge feature map not only contains edges of different degrees, but also contains edge information of other targets, which needs to be further processed.

[0049] S6, further process the edge probability map to obtain strong edge information of the road surface data. The fusion layer result obtained in S5 is used as the edge probability map of the input road surface data. Due to the uneven degree of strong and weak edges caused by weak road background texture, edge influence and different depths of cracks themselves, it is necessary to further judge and process the edge probability map output by the deep network in combination with the continuity of the cracks. According to the edge probability map output by the deep network, it is converted into a grayscale map with a value distribution in the range of 0-255, and combined with the edge map mean Ave, the threshold Th1=3*Ave is obtained.

[0050] S7: Strong edge selection and connection. Based on the edge probability map obtained in S6 and the threshold Th1, the weak edges below the threshold are removed, and the strong edges not less than the threshold are retained. The connected domain of the selected strong edge is used as the image edge object, and the position and direction of the image edge object are used as the input of the tensor voting algorithm. Combined with the edge object area and object direction tensor encoding, the tensor significance field of the input edge object data can be obtained. Combined with the more significant line segments in the obtained tensor significance field, possible connection results are constructed. The connection results are combined with the strong edge map, and the edge map with a smaller connected domain is deleted using the connected domain area parameter.

[0051] Figure 3 (a) and (b) show the generation of the tensor voting (TV) algorithm and the 2D stick tensor saliency field. The TV algorithm infers saliency structure from data through tensor representation of features and nonlinear voting. The TV algorithm generally uses two-dimensional points as input. This embodiment improves the TV algorithm and uses image edge object sub-blocks with direction and position features as input. Specifically, the position and direction features of the image sub-object are used to form a second-order symmetric non-negative tensor field T, T = (λ1-λ2)e1e1T+λ2(e1e1T+e2e2T). By decomposing the tensor field, meaningful curve information can be obtained. λ1 and λ2 are eigenvalues ​​arranged in descending order, and e1 and e2 are the corresponding eigenvectors. (λ1-λ2) represents the stick tensor saliency, which shows the basic curve unit with the eigenvector λ1 as the curve normal. According to the curve structure saliency value of each image object element, a probability map representing the curve structure appearing at a certain position can be obtained, which is called the tensor saliency field of the sub-object, such as Figure 3 (d) as shown.

[0052] The direction of the pixels in each sub-object is obtained from the overall direction of the sub-object. The improved object-level direction tensor voting can combine the proximity and direction characteristics of the crack to appropriately extend the edge features of the discontinuity. Figure 3 (c) shows the selected sub-object, and the sub-object position and its orientation are used as the input of TV. Therefore, the sub-object region and the object orientation tensor encoding are combined to obtain the result after strong edge connection, as shown in Figure 3 (e) As shown in the figure, the connected domain area parameter is used to delete the edge map with a smaller connected domain. The connected domain area threshold Th2 set here can be set according to the needs. For high-precision two-dimensional and three-dimensional data with a lateral sampling resolution accuracy of 1-2mm, the Th2 threshold is 100 by default.

[0053] S8: Extract crack attribute information of 2D / 3D pavement. Based on the crack edge connection results obtained in S7 and the high-frequency component data obtained in S4, the crack object area attribute features are used to obtain attribute information such as crack area and severity. Specifically, based on the position of the inferred crack finally obtained, the edge sub-object position and its connected domain information are correspondingly obtained. For 2D data, the length, area attributes, etc. of the crack can be determined based on the connected domain features of the crack edge object, and the severity of the crack can be determined in combination with the grayscale of the high-frequency component data. For 3D data, the length, area attributes, etc. of the crack can be determined based on the connected domain features of the crack edge object, and the depth information of the crack can be determined in combination with the high-frequency component data elevation map, and then the severity of the crack and its distribution attributes can be determined. 2D / 3D pavement crack attribute information can provide more comprehensive results for the severity and distribution of pavement cracks.

[0054] The effectiveness and accuracy of this embodiment are further illustrated below in combination with line scan three-dimensional road surface data obtained from actual road surfaces and open source two-dimensional road surface data sets.

[0055] 1. Test data. In order to verify the universality of the invented method, based on the HED trained with the above-mentioned FISSURES Dataset, DeepCrackDataset, ESAR 2D dataset and LCMS 3D road dataset, the test data is the Aigle-RN 2D road dataset and the 3D road dataset obtained by the equipment involved in CN112270677A (lateral resolution 1mm, elevation resolution ≤0.5mm).

[0056] 2. Three-dimensional pavement data crack detection test. Figure 4 (a) shows an example of a high-precision line scan of a 3D road surface affected by carrier posture fluctuations and road surface deformation. The edge enhancement result after frequency component decomposition with low-frequency components removed is shown in the figure. Figure 4 As shown in (b), the edge probability results obtained by HED are as follows Figure 4(c) shows the crack identification results after tensor voting and morphological post-processing. Figure 4 As shown in (d), the crack attribute information obtained based on the edge-enhanced pavement base map and crack identification results is as follows: Figure 4 (e) as shown.

[0057] 3. Two-dimensional pavement book crack identification experiment. Figure 5 (a) shows an example of a 2D road data set in Aigle-RN that is affected by uneven illumination. The edge enhancement result obtained after frequency component decomposition is shown in Figure 1. Figure 5 As shown in (b), the edge probability results obtained by the HED deep network are as follows Figure 5 (c) shows the crack identification results after tensor voting and morphological post-processing. Figure 5 As shown in (d), the crack attribute information obtained based on the edge-enhanced pavement base map and crack identification results is as follows: Figure 5 (e) as shown.

[0058] From the above practical operation results, it can be seen that making full use of edge enhancement processing and deep network edge features can effectively remove the influence of environment and vehicle in actual road surface data, and rely on dynamically acquired two-dimensional and three-dimensional road surface data to carry out accurate, effective and universal crack identification method.

[0059] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A two-dimensional and three-dimensional pavement crack recognition method based on enhanced depth edge features, characterized in that: The steps include: S1, reading in the 3D road surface data / 2D road surface data acquired by the road surface acquisition system; S2, data preprocessing; S3, removing the undulation of the 3D road surface data / the uneven illumination of the 2D road surface data; for the acquired 3D road surface data / 2D road surface data, decomposing each cross section into components, removing the influence of low-frequency components, and enhancing the high-frequency edge features of cracks; S4, edge enhancement data acquisition: splicing the high-frequency components of each cross section to form high-frequency component data with enhanced edge characteristics, and uniformly converting the edge-enhanced data into a grayscale image with a value distribution in the range of 0-255; S5, overall nested edge detection deep network training, obtain the test data edge probability map according to the training model; input the data obtained in S4 into the overall nested edge detection deep network to obtain the edge probability map, the overall nested edge detection deep network is trained by the open source two-dimensional pavement data set and the corresponding crack annotations, and the data set input in the training process is also processed by steps S3 and S4; S6, converting the edge probability map to obtain strong edge information of the road surface data; converting the obtained edge probability map into a grayscale map with a value distribution in the range of 0-255, and combining the edge map mean Ave to obtain a threshold Th1; S7, strong edge selection and connection; based on the acquired and converted edge probability map and the threshold Th1, the weak edges below the threshold are removed, and the strong edges not less than the threshold are retained, and the connected domain of the selected strong edge is used as the image edge object, and the image edge object position and direction are used as the input of the tensor voting algorithm, and the tensor significance field of the input edge object data is obtained by combining the edge object area and object direction tensor encoding, and the possible connection results are constructed by combining the more significant line segments in the acquired tensor significance field, and the connection results are combined with the strong edge map, and the edge map with a smaller connected domain is deleted by using the connected domain area parameter; S8, extracting crack attribute information of the two-dimensional / three-dimensional road surface.

2. The two-dimensional and three-dimensional pavement crack identification method based on enhanced depth edge features according to claim 1 is characterized in that: The two-dimensional road surface data is the optical road surface data obtained by using a vehicle-mounted two-dimensional camera, and the three-dimensional road surface data is obtained by using a line scanning three-dimensional measurement sensor to collect a series of road surface cross-sectional profiles along the measurement direction and splicing a series of road surface cross-sectional profiles.

3. The two-dimensional and three-dimensional pavement crack identification method based on enhanced depth edge features according to claim 2 is characterized in that: In S2, the abnormal values ​​caused by system anomalies and environmental anomalies are corrected, the cross-sectional contour is converted from image space to object space through the calibration file, the cross-sectional data of the object space is obtained, and the systematic error in the measurement system is corrected.

4. The two-dimensional and three-dimensional pavement crack identification method based on enhanced depth edge features according to claim 3 is characterized in that: In S8, based on the crack edge connection result obtained in S7 and the high-frequency component data obtained in S4, the crack area and severity attribute information are obtained by using the crack object area attribute characteristics.

5. The two-dimensional and three-dimensional pavement crack identification method based on enhanced depth edge features according to claim 4 is characterized in that: Based on the final acquired inference crack position, the edge sub-object position and its connected domain information are obtained accordingly; for two-dimensional data, the length and area attributes of the crack are determined according to the connected domain characteristics of the crack edge object, and the severity of the crack is determined in combination with the grayscale of the high-frequency component data; for three-dimensional data, the length and area attributes of the crack are determined according to the connected domain characteristics of the crack edge object, and the depth information of the crack is determined in combination with the elevation map of the high-frequency component data, and then the severity of the crack and its distribution attributes are determined.

Images