Rut depth estimation method and system based on road surface crack curvature feature inversion

Abstract

The application discloses a rut depth estimation method and system based on road surface crack curvature characteristic inversion, the method extracts the crack contour through collecting high-resolution road surface images, pretreatment and crack detection, and uses differential geometry to calculate the minimum curvature radius and curvature change rate of the crack and other parameters, and constructs a rut crack geometric response index R-CGRI to represent the response strength of the crack to the subsidence deformation. The image is divided into M*N sub-regions, and the R-CGRI values of each region are calculated to form a two-dimensional response matrix. By regression fitting of each row of R-CGRI values, the minimum point of the fitting curve is extracted as the rut depth estimation value, and finally the longitudinal profile morphology of the rut is reconstructed. The method does not need to rely on high-cost three-dimensional measuring equipment, and can realize accurate estimation and morphology modeling of the rut depth through two-dimensional image analysis, and is suitable for road intelligent inspection and deformation risk assessment.

Description

Technical Field

[0001] The present invention relates to the field of road rutting detection, and in particular to a rutting depth estimation method and system based on inversion of pavement crack curvature characteristics. Background Art

[0002] Rutting is a permanent deformation of roads caused by long-term traffic loads. It seriously affects vehicle comfort and safety and is a key indicator for monitoring during pavement maintenance. Rutting can damage the integrity and stability of the pavement structure, affecting not only the surface but also the underlying layers and even the base layer. Severe rutting can even cause structural damage, significantly reducing the service life of highways and significantly increasing maintenance costs, resulting in significant economic losses.

[0003] Rut depth detection technology has evolved from traditional manual inspection to modern automated detection. Early detection methods relied primarily on manual contact measurement using rulers and feeler gauges. While simple to use, these methods suffered from significant drawbacks such as low accuracy, poor efficiency, and inadequate safety. With technological advancements, non-contact measurement methods have gradually become mainstream. Point laser detection technology, based on the principle of laser triangulation, uses a deployment of several laser sensors to collect road elevation data and calculate rut depth using a fitted cross-section. While offering advantages such as high accuracy and stability, it also carries significant data processing costs and high costs. Line laser detection technology, similarly employing laser triangulation, uses a CCD camera to capture the deformation of a laser line to measure ruts. While maintaining a simple structure and low cost, its accuracy is highly susceptible to ambient light. Ultrasonic detection technology, utilizing the principle of sound wave reflection, allows for rapid multi-parameter detection, but suffers from issues such as weak interference immunity and complex equipment maintenance. Currently, automated detection technology, with its advantages of efficiency and accuracy, has gradually replaced traditional manual methods. However, further improvements are needed in terms of detection costs, resistance to environmental interference, and optimized data processing to better meet the needs of modern road inspection.

[0004] In summary, traditional rutting depth detection methods primarily include contact measurement (such as rulers, levels / total stations), and non-contact measurement (such as laser scanners, ultrasonic detection technology, and linear structured light sensors). While non-contact methods offer certain advantages in accuracy and efficiency, they rely heavily on high-precision equipment, resulting in high costs and limited widespread adoption. Furthermore, obtaining high-precision 3D point cloud data of the road surface requires fusion and pose correction with a high-precision inertial navigation system (INS). This computationally complex process can compromise measurement accuracy.

[0005] With the recent development of image recognition technology, pavement defect detection technology based on image processing has become relatively low-cost and offers advantages in both speed and accuracy. Rutting areas, due to concentrated forces, are often accompanied by significant crack deformation characteristics. In particular, cracks that run along the rutting direction or form a semi-arc are prone to appear within the rutting groove. Due to the subsidence effect caused by rutting, the geometry of these cracks (particularly their curvature) is inherently correlated with rutting depth. Therefore, based on this physical phenomenon, the present invention efficiently extracts and analyzes crack curvature to estimate rutting depth based solely on pavement crack images, independent of high-precision 3D measurement equipment. This provides a new solution for large-scale, low-cost rutting detection. Summary of the Invention

[0006] In order to overcome the shortcomings of the existing technology, the present invention provides a rutting depth estimation method and system based on the inversion of pavement crack curvature characteristics, which utilizes the geometric curvature characteristics of cracks in images to achieve inversion estimation of rutting depth.

[0007] The rutting depth estimation method designed by the present invention based on the inversion of pavement crack curvature characteristics includes the following steps:

[0008] Collect high-resolution road surface images including rutting and crack features;

[0009] Preprocessing the high-resolution road surface image;

[0010] Extracting crack contour segments from the pavement image and obtaining a centerline coordinate sequence of the cracks;

[0011] The curvature characteristic parameters of the crack are calculated based on the differential geometry principle, including the minimum curvature radius, curvature change rate and average curvature absolute value;

[0012] The rutting crack geometric response index R-CGRI is constructed to characterize the response intensity of the crack to the subsidence deformation:

[0013]

[0014] Among them, w1, w2, w3 are weight parameters, and the minimum curvature radius R min , curvature change rate , the absolute value of the mean curvature ; A regression model based on multi-layer perceptron MLP is used for training;

[0015] Dividing the pavement image into multiple sub-regions, calculating the R-CGRI value of each sub-region to form a two-dimensional crack response matrix;

[0016] Fit the R-CGRI value of each row of sub-areas and extract the minimum point of the fitting curve as the estimated rutting depth;

[0017] The longitudinal profile of the rut is reconstructed by combining the rut depth estimation of each row to achieve structured modeling of the rut geometry.

[0018] Preferably, the resolution of the high-resolution road surface image is not less than 1920×1080, and is capable of distinguishing crack information with a minimum width of 1 mm.

[0019] Preferably, the pre-processing first performs grayscale processing, then adopts median filtering to perform denoising, and applies contrast-limited adaptive histogram equalization to enhance the contrast of local details of the image.

[0020] Preferably, the extraction of the crack contour segments adopts a crack semantic segmentation model based on deep learning combined with Canny edge detection and morphological processing, and the extracted crack contours are skeletonized.

[0021] Preferably, the curvature characteristic parameters are calculated by estimating the first-order and second-order derivatives using the finite difference method, and the curvature value of each pixel is calculated using a two-dimensional plane curvature formula.

[0022] Preferably, fitting the R-CGRI value of each row of sub-regions and extracting the minimum point of the fitting curve as the estimated rutting depth specifically includes:

[0023] The longitudinal profile of the rut is reconstructed by combining the rut depth estimation of each row to achieve structured modeling of the rut geometry. Specifically, the following steps are involved:

[0024] For each row, the vertical index j (1~N) is used as the horizontal axis coordinate, the R-CGRI value is the target fitting value, and a univariate polynomial curve is fitted:

[0025]

[0026] Extract the minimum point position of each line of the fitting curve , reversely infer the rutting depth value at this location through the mapping relationship , and finally obtain a one-dimensional depth distribution sequence .

[0027] Preferably, the structured modeling of the rut geometry includes interpolating and smoothing the estimated depths at multiple lateral positions to generate a complete rut profile curve.

[0028] Based on the same inventive concept, the present invention also designs a rutting depth estimation system based on the inversion of pavement crack curvature characteristics, comprising:

[0029] Image acquisition module, which collects high-resolution road surface images including rutting areas;

[0030] A preprocessing module, for preprocessing the road surface image;

[0031] A crack extraction module extracts the crack contour segments from the pavement image and obtains a centerline coordinate sequence of the cracks;

[0032] The curvature calculation module calculates the curvature characteristic parameters of the crack based on the differential geometry principle, including the minimum curvature radius, curvature change rate and average curvature absolute value;

[0033] The response index building module constructs the rutting crack geometric response index R-CGRI, which is used to characterize the response intensity of the crack to the subsidence deformation:

[0034]

[0035] Among them, w1, w2, w3 are weight parameters, and the minimum curvature radius R min , curvature change rate , the absolute value of the mean curvature ; A regression model based on multi-layer perceptron MLP is used for training;

[0036] a matrix generation module, which divides the pavement image into a plurality of sub-regions, calculates the R-CGRI value of each sub-region, and forms a two-dimensional crack response matrix;

[0037] The fitting module fits the R-CGRI value of each row of sub-areas and extracts the minimum point of the fitting curve as the estimated rutting depth;

[0038] The reconstruction module combines the rutting depth estimation of each row to reconstruct the longitudinal profile of the rutting, thus realizing the structured modeling of the rutting geometry.

[0039] Based on the same inventive concept, the present invention also designs an electronic device, including: a memory and a processor, the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to execute the rutting depth estimation method based on the inversion of the curvature characteristics of pavement cracks.

[0040] The present invention has the following advantages:

[0041] 1) A method for automatically inverting rutting depth based on the geometric characteristics of pavement crack curvature is proposed. The correlation between crack morphology and pavement subsidence is modeled by constructing the crack response index (R-CGRI). This method extracts continuous crack curves from images and calculates curvature characteristics, thus avoiding the traditional rutting analysis that relies on high-cost 3D measurement or laser profiling instruments, and offers greater adaptability and scalability.

[0042] 2) The Rutting Crack Geometric Response Index (R-CGRI) was introduced as an evaluation metric to quantify the nonlinear morphological characteristics of cracks and the response strength of rutting depth in each region of the image. Compared to traditional methods that rely on image grayscale, texture, or morphological operations, this metric more stably reflects the actual deformation state of the road surface, improving the spatial consistency and engineering interpretation capabilities of the estimated results.

[0043] 3) Based on an image space partitioning strategy, the entire image is divided into multiple subregions. The crack response curve for each column is extracted and regressed. Ultimately, a spatial distribution matrix and longitudinal profile of the rutting depth are constructed, achieving a structured reconstruction of the rutting surface morphology. This method not only determines the local maximum rutting depth but also assesses the geometric trends of the entire rutting section and identifies risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is the original road surface image preprocessing process of the present invention.

[0045] Figure 2 This is the calculation of the two-dimensional crack-rutting correlation response R-CGRI matrix of the present invention.

[0046] Figure 3 This is the lateral fitting and longitudinal morphology reconstruction of the rut based on the R-CGRI matrix of the present invention.

[0047] Figure 4 It is an overall flow chart of the technical implementation scheme of the present invention. DETAILED DESCRIPTION

[0048] The technical solution of the present invention will be further specifically described below through embodiments and in conjunction with the accompanying drawings.

[0049] Example 1:

[0050] This embodiment proposes a rutting depth estimation method based on the inversion of pavement crack curvature characteristics. The overall flow chart is shown in the attached figure. Figure 4As shown, a high-resolution road surface image containing rutting areas is first acquired. After preprocessing and crack detection, crack contour segments are extracted. Subsequently, based on differential geometry principles, crack parameters such as the minimum curvature radius and curvature change rate are calculated to construct a rut crack geometric response index (R-CGRI), which reflects the local curvature morphological changes. This index characterizes the intensity of the crack response to subsidence deformation within each subregion. The image is divided into M×N subregions in both the horizontal and vertical directions, and the R-CGRI value is calculated for each subregion to construct a two-dimensional rutting response matrix. Within each row, a regression fit is performed on the longitudinal R-CGRI distribution, and the minimum point of the fitted curve is extracted as the estimated rutting depth at that location. Rutting depth values ​​are then obtained at multiple longitudinal locations. This method not only estimates the maximum rutting depth in a local area but also reconstructs the longitudinal profile trend of the entire rutting line, achieving structured modeling and quantitative expression of the rutting surface geometry. This method is suitable for intelligent road inspection and deformation risk assessment scenarios. Specifically, it includes the following steps:

[0051] Step 1: Image acquisition and preprocessing.

[0052] High-resolution imaging equipment is deployed on a vehicle or ground detection platform to collect road surface images containing obvious rutting and crack features. The image resolution should be no less than 1920×1080 to ensure that crack information with a minimum width of less than 1 mm can be distinguished. Figure 1 The preprocessing process shown in the figure preprocesses the collected raw pavement images. First, grayscale processing is performed to reduce channel redundant information. Median filtering is then used to denoise the image, suppressing high-frequency interference and enhancing edge stability. Next, contrast-limited adaptive histogram equalization (CLAHE) is applied to enhance the contrast of local details in the image, making small cracks easier to identify. This improves crack texture clarity and suppresses background interference, providing reliable input for subsequent crack identification.

[0053] Step 2: Crack identification and geometric feature extraction.

[0054] After preprocessing the pavement image, a deep learning-based crack semantic segmentation model (such as U-Net or DeepLabv3+) is used in conjunction with traditional Canny operators and morphological processing to extract the crack areas in the image. The crack edges are then skeletonized, and a pixel chain tracking algorithm is used to extract the crack centerline. For each crack curve, the curvature value of each pixel is calculated based on differential geometry principles. , extract typical geometric feature indicators, including the minimum curvature radius R min , curvature change rate and the absolute value of the mean curvature The details are as follows:

[0055] After the crack skeleton line is extracted, the pixel point sequence of the crack center line is obtained ( ), to facilitate subsequent geometric analysis, the curvature characteristics of this discrete curve need to be quantified. Considering that cracks often exhibit smooth but locally curved features, the present invention uses differential geometry to calculate the curvature of the pixel chain curve. The specific process is as follows:

[0056] First, for each point =( ), and use finite difference methods to estimate the first and second derivatives:

[0057]

[0058]

[0059] Use the two-dimensional plane curvature formula to calculate the curvature value of each pixel , the sign of the curvature reflects the bending direction, and the absolute value can be used to obtain the bending strength.

[0060]

[0061] To facilitate the subsequent construction of regional indicators, the following typical curvature features are extracted from each fracture segment:

[0062] Minimum curvature radius:

[0063]

[0064] Curvature change rate (fluctuation degree):

[0065]

[0066] Absolute value of mean curvature:

[0067]

[0068] The above characteristics can reflect the local variation and nonlinearity of cracks and are used to construct the rutting crack response index R-CGRI, which serves as a quantitative indicator input for the sensitivity of cracks to subsidence deformation.

[0069] Step 3: Construct the Rut Crack Geometric Response Index (R-CGRI).

[0070] After obtaining the curvature characteristic parameters of each crack (minimum curvature radius, curvature change rate, average curvature), in order to quantify the crack's response intensity to rutting deformation, the sensitivity of the crack geometry to subsidence deformation is comprehensively considered. The rutting crack geometric response index (R-CGRI) is defined to measure the crack's response intensity to rutting depth. The calculation formula is as follows:

[0071]

[0072] Where w1, w2, and w3 are weight parameters, trained using a regression model based on a multi-layer perceptron (MLP). Crack curvature features extracted from measured images and manually annotated true rutting depths are used as training samples. By minimizing the mean squared error (MSE) loss function between the predicted and measured depths, backpropagation optimization is used to determine the optimal weighting coefficients. After training, the network model can automatically generate the optimal R-CGRI index for any combination of crack geometric features.

[0073] Step 4: Regional division and index matrix construction.

[0074] The entire road surface image is divided into M×N equally spaced sub-regions, and the R-CGRI value in each region block is calculated to form a two-dimensional crack response matrix R(i, j), which is used to describe the relationship between the spatial variation of crack morphology and rutting response in each region of the image, as shown in the attached figure. Figure 2 shown.

[0075] Step 5: Response fitting and depth extraction.

[0076] For each row R(i, j) of the segmented road image (fixed lateral position i), a univariate polynomial is fitted using the least squares method to construct a lateral curve profile. Specifically, for each row, with the longitudinal index j (1-N) as the horizontal axis coordinate and the R-CGRI value as the target fitting value, a univariate polynomial curve (e.g., quadratic / cubic) is fitted:

[0077]

[0078] in, a j ，b j 、c j are polynomial coefficients. Extract the minimum point position of each line of fitting curve , reversely infer the rutting depth value at this location through the mapping relationship , and finally obtain a one-dimensional depth distribution sequence .

[0079] Step 6: Rutting morphology reconstruction and structural evaluation.

[0080] By analyzing the whole map in multiple longitudinal (column) directions, a series of estimated rutting profiles can be obtained, which can reflect the maximum rutting depth at each position and reconstruct the morphological curve of the entire rutting. By performing interpolation and smooth reconstruction, a complete rutting profile curve can be obtained, and the structured modeling of the rutting morphology can be realized, as shown in the attached figure. Figure 3 As shown in the figure, the analysis of indicators such as maximum depth, curvature change, and symmetry can be further used for pavement settlement trend assessment and risk identification.

[0081] Example 2

[0082] Based on the same inventive concept, the present invention also designs a rutting depth estimation system based on the inversion of pavement crack curvature characteristics, comprising:

[0083] Image acquisition module, which collects high-resolution road surface images including rutting areas;

[0084] A preprocessing module, for preprocessing the road surface image;

[0085] A crack extraction module extracts the crack contour segments from the pavement image and obtains a centerline coordinate sequence of the cracks;

[0086] The curvature calculation module calculates the curvature characteristic parameters of the crack based on the differential geometry principle, including the minimum curvature radius, curvature change rate and average curvature absolute value;

[0087] The response index construction module constructs the rutting crack geometric response index R-CGRI, which is used to characterize the response intensity of the crack to the subsidence deformation;

[0088] a matrix generation module, which divides the pavement image into a plurality of sub-regions, calculates the R-CGRI value of each sub-region, and forms a two-dimensional crack response matrix;

[0089] The fitting module fits the R-CGRI value of each row of sub-areas and extracts the minimum point of the fitting curve as the estimated rutting depth;

[0090] The reconstruction module combines the rutting depth estimation of each row to reconstruct the longitudinal profile of the rutting, thus realizing the structured modeling of the rutting geometry.

[0091] The system described in this embodiment is a system for implementing the rutting depth estimation method based on the inversion of the curvature characteristics of the pavement cracks in embodiment 1. The specific process content can be found in the corresponding part of the above method embodiment and will not be repeated here.

[0092] Example 3

[0093] Based on the same inventive concept, the present invention also provides an electronic device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the aforementioned rutting depth estimation method based on inversion of pavement crack curvature characteristics. The system described in this embodiment is an electronic device that implements the image data compression and decompression method based on the drone-mounted generative model described in Example 1. The specific process details are described in the corresponding sections of the aforementioned method embodiments and will not be repeated here.

[0094] The above embodiments are intended only to illustrate the design concepts and features of the present invention. Their purpose is to enable those skilled in the art to understand the contents of the present invention and implement them accordingly. The scope of protection of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design concepts disclosed in the present invention are within the scope of protection of the present invention.

Claims

1. A rutting depth estimation method based on inversion of pavement crack curvature characteristics, characterized by: Collect high-resolution road surface images including rutting and crack features; Preprocessing the high-resolution road surface image; Extract crack contour lines from preprocessed pavement images and obtain centerline coordinate sequences of the contour lines; In combination with the centerline coordinate sequence, the curvature characteristic parameters of the crack are calculated based on the differential geometry principle, including the minimum curvature radius, the curvature change rate and the average curvature absolute value; The rutting crack geometric response index (R-CGRI) is constructed by combining the minimum curvature radius, curvature change rate, and average curvature absolute value to characterize the crack response intensity to subsidence / rutting. A regression model based on a multi-layer perceptron (MLP) is used to train the R-CGRI, and an effective mapping relationship between the R-CGRI and pavement subsidence / rutting is established. The rutting crack geometric response index R-CGRI is: Among them, w1, w2, w3 are weight parameters, and the minimum curvature radius R min , curvature change rate , the absolute value of the mean curvature ; Dividing the high-resolution pavement image into multiple sub-regions, calculating the R-CGRI value of each sub-region to form a two-dimensional crack response matrix; Fit the R-CGRI value of each row sub-area and extract the minimum point of the fitting curve as the estimated rutting depth, including: For each row, the vertical index j (1~N) is used as the horizontal axis coordinate, the R-CGRI value is the target fitting value, and a univariate polynomial curve is fitted: in, a j ，b j 、c j is the polynomial coefficient; extract the minimum point position of each line of fitting curve , reversely infer the rutting depth value at this location through the mapping relationship , and finally obtain a one-dimensional depth distribution sequence ; The longitudinal profile of the rut is reconstructed by combining the estimated rut depth of each row to achieve structured modeling of the rut geometry.

2. The rutting depth estimation method based on inversion of pavement crack curvature characteristics according to claim 1 is characterized in that: The preprocessing includes first performing grayscale processing, then performing denoising using median filtering, and applying contrast-limited adaptive histogram equalization to enhance the contrast of local details of the image.

3. The rutting depth estimation method based on inversion of pavement crack curvature characteristics according to claim 1 is characterized in that: The crack contour line segments of the pavement image are extracted by using a crack semantic segmentation model based on deep learning combined with Canny edge detection and morphological processing, and the extracted crack contours are skeletonized.

4. The rutting depth estimation method based on inversion of pavement crack curvature characteristics according to claim 1 is characterized in that: The curvature characteristic parameters are calculated by estimating the first-order and second-order derivatives using the finite difference method, and the curvature value of each pixel is calculated using a two-dimensional plane curvature formula.

5. The rutting depth estimation method based on inversion of pavement crack curvature characteristics according to claim 1 is characterized in that: In the fitting process, the least square method is used to fit a univariate polynomial curve to each row of R-CGRI values.

6. The rutting depth estimation method based on inversion of pavement crack curvature characteristics according to claim 1 is characterized in that: The structured modeling of the rutting geometry includes interpolating and smoothing the estimated depths at multiple lateral positions to generate a complete rutting profile curve.

7. A rutting depth estimation system based on inversion of pavement crack curvature characteristics, characterized in that: include: Image acquisition module, which collects high-resolution road surface images including rutting areas; A preprocessing module, for preprocessing the high-resolution road surface image; A crack extraction module extracts the crack contour segments from the pavement image and obtains a centerline coordinate sequence of the cracks; The curvature calculation module calculates the curvature characteristic parameters of the crack based on the differential geometry principle, including the minimum curvature radius, curvature change rate and average curvature absolute value; The response index construction module combines the minimum curvature radius, curvature change rate and average curvature absolute value to construct the rutting crack geometric response index R-CGRI, which is used to characterize the response intensity of the crack to the subsidence deformation. The R-CGRI is trained using a regression model based on a multi-layer perceptron (MLP). The rutting crack geometric response index R-CGRI is: Among them, w1, w2, w3 are weight parameters, and the minimum curvature radius R min , curvature change rate , the absolute value of the mean curvature ; a matrix generation module, which divides the high-resolution pavement image into a plurality of sub-regions, calculates the R-CGRI value of each sub-region, and forms a two-dimensional crack response matrix; The fitting module fits the R-CGRI value of each row sub-area and extracts the minimum point of the fitting curve as the estimated rutting depth. Specifically, it includes: For each row, the vertical index j (1~N) is used as the horizontal axis coordinate, the R-CGRI value is the target fitting value, and a univariate polynomial curve is fitted: in, a j ，b j 、c j is the polynomial coefficient; extract the minimum point position of each line of fitting curve , reversely infer the rutting depth value at this location through the mapping relationship , and finally obtain a one-dimensional depth distribution sequence ; The reconstruction module combines the rutting depth estimation of each row to reconstruct the longitudinal profile of the rutting, thus realizing the structured modeling of the rutting geometry.

8. An electronic device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the rutting depth estimation method based on inversion of pavement crack curvature characteristics as described in any one of claims 1 to 6 by executing the computer instructions.

Images