Road disease identification method

Abstract

The invention relates to a road disease identification method, and relates to the technical field of road maintenance, and the method comprises the following steps: an image collection step: carrying out the image collection of a road surface; a brightness conversion step: converting the RGB pixel value of the acquired image into brightness information L, and improving the contrast between the disease in the image and the normal road surface; a processing step: dividing the image subjected to the brightness conversion step into N square areas with the side length of a, performing edge extension on non-integral multiple areas, ensuring complete area division, judging the boundary of a disease, and extracting a disease area; and a classification step: classifying the extracted disease areas, and outputting the classified disease areas. The method has the advantages of low maintenance cost and high response speed.

Description

Technical Field

[0001] This application relates to the technical field of road maintenance, and in particular to a method for identifying highway defects. Background Technology

[0002] The timely and accurate identification of highway defects (such as cracks, potholes, transverse / longitudinal cracks, ruts, undulations, broken slabs, and craters) is a core element in ensuring road safety and improving maintenance efficiency. However, mainstream inspection methods still heavily rely on manual inspections or vehicle-mounted semi-automatic inspection equipment, resulting in insufficient overall automation and intelligence. This inspection mode, primarily manual with semi-automatic assistance, generally suffers from significant drawbacks such as low inspection efficiency, slow response times, and high labor and time costs, making it difficult to meet the urgent needs of efficient and precise maintenance of modern road networks. This constitutes the core background and key driving force for developing an AI-based intelligent identification algorithm for highway defects. Summary of the Invention

[0003] In view of the shortcomings of the existing technology, one of the purposes of this application is to provide a method for identifying highway defects, which has the advantages of low maintenance cost and fast response speed.

[0004] The above-mentioned objective of this application is achieved through the following technical solution:

[0005] A method for identifying road defects includes the following steps: an image acquisition step, which acquires an image of the road surface; a brightness conversion step, which converts the RGB pixel values ​​of the acquired image into brightness information L to improve the contrast between defects and normal road surfaces in the image; a processing step, which divides the image after the brightness conversion step into N square regions with side length a, extends the edges of non-integer multiple regions to ensure complete region division, judges the boundaries of defects, and extracts the defect regions; and a classification step, which classifies the extracted defect regions and outputs the classification results.

[0006] In a preferred embodiment, this application can be further configured such that, in the brightness conversion step, the following formula is used:

[0007] In a preferred embodiment, this application can be further configured such that, during the processing step,

[0008]

[0009] In a preferred embodiment, this application can be further configured such that, during the processing step, when determining the boundary of the disease, three-dimensional coordinates (px, py, pz) are constructed for the pixels of each square region, where pz...

[0010] For the brightness value, fit the plane equation to the three vertices p1, p2, p3 of the region:

[0011]

[0012] In a preferred embodiment, this application can be further configured as follows: in the processing step, the diseased area is extracted using the following function, and a 3×3 convolution kernel T with variable center weights is designed. k Used to extract the differences between the characteristics of road defects and those of normal road surfaces: Where g is an adaptive weighting parameter, dynamically solved using the brightness matrix characteristics of the current image block, for any 3×3 image block reg i Construct the brightness matrix T l,i , Calculate the rank of the matrix product σ = r(T) k ×T l,i Construct the normal distribution probability density function: Combining matrix determinant d i =|T k ×T l,i | Solve for the adaptive parameter g using an integral equation: For the current road surface image B0 to be detected and the standard normal road surface image B_std, respectively, the adaptive kernel function T is applied. k Perform convolution to obtain the feature matrix X B0 and X B_std ; Calculate the similarity before and after convolution:

[0013] Where X' is the original image matrix; if thr0 <thr ， A value of 0 indicates a significant difference between the current image and the standard image, suggesting the presence of defects; conversely, a value of 0 indicates a normal road surface.

[0014] This application has the following advantages:

[0015] High-precision recognition: Based on brightness field modeling and adaptive kernel function, it overcomes interference from changes in lighting and road surface stains, achieving an accuracy of over 97%; Strong real-time performance: Region division and parallel computing design support real-time processing of image streams by on-board equipment, with a processing speed of ≥25 frames / second; Wide adaptability: No need to manually adjust thresholds for different road conditions, automatically adapting to various scenarios such as highways, national roads, and urban roads; Low cost: It can replace manual inspection, significantly reducing maintenance costs and improving the level of intelligent highway maintenance. Detailed Implementation

[0016] This application discloses a method for identifying road defects, comprising the following steps: an image acquisition step, which acquires an image of the road surface; a brightness conversion step, which converts the RGB pixel values ​​of the acquired image into brightness information L to improve the contrast between defects and normal road surfaces in the image; a processing step, which divides the image after the brightness conversion step into N square regions with side length a, extends the edges of non-integer multiple regions (filling with black pixels of R=0, G=0, B=0) to ensure complete region division, judges the boundaries of defects, performs connected component analysis on all non-continuous planar regions in the image, extracts regions with an area greater than a threshold, and, combined with the morphological characteristics of road defects (such as linear cracks and closed potholes), filters out potential defect regions and extracts the defect regions; and a classification step, which classifies the extracted defect regions and outputs the classification.

[0017] The following formula is used in the brightness conversion step:

[0018] In the processing steps,

[0019]

[0020] , where x, y are the image pixel coordinates.

[0021] In the processing steps, when determining the boundary of the lesion, a three-dimensional coordinate system (px, py, pz) is constructed for the pixels of each square region, where pz is the brightness value. The three vertices p1, p2, and p3 of the region are then fitted to the horizontal plane.

[0022] Surface equation:

[0023] Calculate the plane normal vector (A, B, C) and randomly select 4 points within the region to verify the plane fit. If more than 3 points do not satisfy the plane equation, the region is determined to contain disease boundaries (such as crack edges or pothole outlines). Perform connected component analysis on all discontinuous planar regions in the image, extract regions with areas greater than a threshold, and combine this with the morphological characteristics of highway diseases (such as the linearity of cracks and the closed regions of potholes) to screen out potential disease regions.

[0024] In the processing steps, the diseased areas are extracted using the following function, and a 3×3 convolution kernel T with variable center weights is designed. k Used to extract the differences between the characteristics of road defects and those of normal road surfaces: Where g is an adaptive weighting parameter, dynamically solved using the brightness matrix characteristics of the current image block, for any 3×3 image block reg i Construct the brightness matrix T l,i , Calculate the rank of the matrix product σ = r(T)k ×T l,i Construct the normal distribution probability density function: Combining matrix determinant d i =|T k ×T l,i | Solve for the adaptive parameter g using an integral equation: For the current road surface image B0 to be detected and the standard normal road surface image B_std, respectively, the adaptive kernel function T is applied. k Perform convolution to obtain the feature matrix X B0 and X B_std ; Calculate the similarity before and after convolution: Where X' is the original image matrix; if thr0 <thr ， A value of 0 indicates a significant difference between the current image and the standard image, suggesting the presence of defects; conversely, a value of 0 indicates a normal road surface.

[0025] Taking a highway inspection system in a certain province as an example, the specific implementation process is as follows:

[0026] Step 1: Image Acquisition

[0027] The vehicle-mounted high-definition camera was used to capture images of the road surface, with a resolution of 1920×1080.

[0028] Step 2: Brightness Conversion

[0029] Convert the RGB image into a luminance matrix L;

[0030] Step 3: Processing Steps

[0031] Set the region side length a = 16 pixels; calculate the number of regions N, and extend the image edges with black fill; initialize the region index i = 1; loop (i ≤ N):

[0032] 1. Take the i-th square region, extract the coordinates of the three vertices, and calculate the normal vector based on the fitted plane equation;

[0033] 2. Randomly select 4 points to verify the continuity of the plane. If ≥3 points do not satisfy the equation, mark the region as a "potential disease boundary".

[0034] 3. i = i + 1, until all regions have been traversed;

[0035] Connectivity analysis was performed on the marked areas, and the top 10% of the areas were extracted as candidate disease areas.

[0036] Divide the candidate region into M 3×3 convolutional blocks and initialize the block index i = 1;

[0037] Loop (i≤M):

[0038] 1. Solve for the adaptive kernel parameter g of the current block;

[0039] 2. Using the kernel function T k Perform convolution on the current block to extract features;

[0040] 3. i = i + 1, until all blocks have been traversed;

[0041] Calculate the similarity thr0 between the convolutional candidate region and the standard image, and the similarity thr between the original image and the standard image. ， 0.

[0042] Step 4: Sorting Steps

[0043] If thr0 <thr ， 0 indicates the presence of disease, and classifies it as cracks or pits based on the area's morphology (linear or blocky); outputs the assessment results of disease location, size, type, and severity to complete the detection.

[0044] Actual tests were conducted on three highways with different road conditions (expressway, national highway, and provincial highway) to compare the recognition accuracy of the invention with that of manual inspection (by three senior maintenance engineers):

[0045] Detection scenario Manual inspection (A) Manual inspection (B) Manual inspection (C) This application high speed 96％ 94％ 95％ 97％ national highway 98％ 97％ 96％ 98％ Provincial highway (complex environment) 95％ 93％ 94％ 96％

[0046] The implementation principle of this embodiment is as follows: by collecting road surface images and using intelligent recognition algorithms for accurate identification, maintenance departments can accurately locate the location of defects and achieve targeted maintenance. This technical approach effectively overcomes the drawbacks of traditional extensive maintenance, significantly reduces maintenance costs (such as reducing ineffective inspections and optimizing material usage), and greatly improves emergency response and daily maintenance efficiency.

[0047] The embodiments described in this specific implementation are preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for identifying highway defects, characterized in that: The steps include the following: image acquisition step, which involves acquiring images of the road surface; The brightness conversion step converts the RGB pixel values ​​of the acquired image into brightness information L, improving the contrast between the damaged and normal road surfaces in the image. The processing step divides the image after the brightness conversion step into N square regions with a side length of a, extends the edges of non-integer multiple regions to ensure complete region division, judges the boundaries of the damaged areas, and extracts the damaged areas. The classification step classifies the extracted damaged areas and outputs the classification results.

2. The method for identifying highway defects according to claim 1, characterized in that: The following formula is used in the brightness conversion step:

3. The method for identifying highway defects according to claim 1, characterized in that: In the processing steps, 4. The method for identifying highway defects according to claim 1, characterized in that: In the processing steps, when determining the boundary of the lesion, three-dimensional coordinates (px, py, pz) are constructed for the pixels of each square region, where pz is the brightness value. The three vertices p1, p2, and p3 of the region are used to fit the plane equation:

5. The method for identifying highway defects according to claim 1, characterized in that: In the processing steps, the diseased areas are extracted using the following function, and a 3×3 convolution kernel T with variable center weights is designed. k Used to extract the differences between the characteristics of road defects and those of normal road surfaces: Where g is an adaptive weighting parameter, dynamically solved using the brightness matrix characteristics of the current image block, for any 3×3 image block reg i Construct the brightness matrix T l,i , Calculate the rank of the matrix product σ = r(T) k ×T l,i Construct the normal distribution probability density function: Combining matrix determinant d i =|T k ×T l,i | Solve for the adaptive parameter g using an integral equation: For the current road surface image B0 to be detected and the standard normal road surface image B_std, respectively, the adaptive kernel function T is applied. k Perform convolution to obtain the feature matrix X B0 and X B_std ; Calculate the similarity before and after convolution: Where X' is the original image matrix; if thr0 <thr ， A value of 0 indicates a significant difference between the current image and the standard image, suggesting the presence of defects; conversely, a value of 0 indicates a normal road surface.

Images