Asphalt pavement crack detection method and system

Abstract

The invention relates to the technical field of image recognition, in particular to an asphalt pavement crack detection method and system. The method comprises the following steps: collecting a road surface image of a target road section, executing illumination compensation, and generating a to-be-analyzed road surface image; determining the road indication texture of the target road section according to the gray level change condition of the pixel points in the to-be-analyzed road surface image; removing non-pavement texture features in the to-be-analyzed pavement image by using the road indication texture; identifying a linear crack region according to pavement texture features in the to-be-analyzed pavement image, associating lane information, and generating a crack feature set of each lane of the target road section; and identifying crack connectivity between adjacent lanes according to the crack feature set of each lane, and dividing maintenance priorities of cracks of the target road section based on the crack connectivity between the adjacent lanes to obtain a pavement maintenance report of the target road section. According to the invention, the precision and reliability of crack identification are obviously improved, and accurate identification of cracks of a single lane and adjacent lanes can be realized.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method and system for detecting cracks in asphalt pavement. Background Technology

[0002] Asphalt pavement crack detection technology has evolved from manual visual inspection and semi-automatic imaging recognition to a system based on intelligent recognition and multi-source information fusion. Early crack detection relied primarily on highway maintenance personnel visually recording and manually marking pavement surface defects through on-site inspections or slow-moving vehicle checks. While intuitive, this method was limited by the experience level of maintenance personnel, resulting in high labor intensity, low efficiency, strong subjectivity, and difficulty in generating structured historical data. Consequently, it could not meet the demands of the expanding highway network and refined maintenance management.

[0003] Subsequently, with the application of line scan cameras, area scan cameras, and vehicle-mounted acquisition equipment, semi-automatic crack detection schemes based on road surface images emerged. These schemes utilize industrial cameras, light sources, and acquisition control units deployed on inspection vehicles to process road surface images acquired during vehicle operation offline, achieving preliminary crack identification and labeling. However, these methods often employ image processing techniques based on grayscale thresholds, edge operators, or simple morphological operations. They are poorly adaptable to complex lighting conditions, road surface texture interference, oil stains, water stains, and rut shadows, leading to frequent false positives and false negatives in crack identification. Furthermore, their generalization ability to different grades of highways and different types of asphalt pavements is limited. Summary of the Invention

[0004] Therefore, the present invention needs to provide a method and system for detecting cracks in asphalt pavement to solve at least one of the above-mentioned technical problems.

[0005] To achieve the above objectives, a method for detecting cracks in asphalt pavement includes the following steps: Step S1: Acquire road surface images of the target road section, and perform illumination compensation based on the grayscale changes of adjacent pixels in the road surface images to generate road surface images to be analyzed; Step S2: Determine the road indicator texture of the target road segment based on the grayscale changes of pixels in the road surface image to be analyzed; use the road indicator texture to remove non-road surface texture features from the road surface image to be analyzed; Step S3: Based on the road surface texture features in the road surface image to be analyzed, identify linear crack areas and associate the linear crack areas with the corresponding lane information of the target road segment to generate a crack feature set for each lane of the target road segment; Step S4: Identify the crack connectivity between adjacent lanes based on the crack feature set of each lane, and classify the maintenance priority of the cracks in the target road section based on the crack connectivity between adjacent lanes to obtain the pavement maintenance report of the target road section.

[0006] This application effectively eliminates the impact of uneven lighting and shadow interference on crack identification by dividing the road surface image into sub-block regions along the driving direction and the lateral lane direction, and combining brightness anomaly detection and adaptive illumination compensation. Furthermore, it utilizes road marking texture extraction technology to eliminate non-road surface texture features, such as shadow textures, short-term textures, and road marking areas, ensuring that crack detection is only performed on the actual road surface area. Based on this, it identifies dark texture pixels by statistically analyzing local neighborhood grayscale features and aggregates them into linear texture segments. Then, it filters linear crack regions based on length, width, and aspect ratio, achieving accurate identification of cracks in single lanes and adjacent lanes. Simultaneously, it determines crack connectivity based on the lateral spacing between cracks in adjacent lanes and the overlap length in the driving direction, combining connected cracks into continuous crack paths. Finally, it calculates a comprehensive risk score based on the total length in the driving direction, lateral span, and aspect ratio, ranking and prioritizing maintenance to generate a quantifiable and traceable road maintenance report. This application not only significantly improves the accuracy and reliability of crack identification, but also enables its generalization under complex lighting conditions, different asphalt pavement types, and diverse lane conditions, providing reliable data support for refined maintenance and scientific decision-making of highway networks.

[0007] Optionally, this application also provides an asphalt pavement crack detection system for performing the asphalt pavement crack detection method described above, the asphalt pavement crack detection system comprising: The data acquisition module is used to acquire road surface images of the target road section and perform illumination compensation based on the grayscale changes of adjacent pixels in the road surface image to generate a road surface image to be analyzed. The texture analysis module is used to determine the road indicator texture of the target road segment based on the grayscale changes of pixels in the road surface image to be analyzed; and to remove non-road surface texture features from the road surface image to be analyzed using the road indicator texture. The crack recognition module is used to identify linear crack areas based on the road texture features in the road surface image to be analyzed, and associate the linear crack areas with the corresponding lane information of the target road segment to generate a crack feature set for each lane of the target road segment. The maintenance priority classification module is used to identify the crack connectivity between adjacent lanes based on the crack feature set of each lane, and classify the maintenance priority of cracks in the target road section based on the crack connectivity between adjacent lanes, so as to obtain the pavement maintenance report of the target road section.

[0008] The asphalt pavement crack detection system of the present invention can implement any of the asphalt pavement crack detection methods of the present invention. It serves as a medium for the operation and signal transmission between various modules to complete the asphalt pavement crack detection method. The internal modules of the system cooperate with each other, thereby improving the accuracy and reliability of crack identification. Attached Figure Description

[0009] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the steps in the asphalt pavement crack detection method of the present invention; Figure 2 This is a non-road texture feature marker map of the road surface image to be analyzed in this embodiment of the invention; Figure 3 This is a block diagram of the asphalt pavement crack detection system in an embodiment of the present invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0010] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0011] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0012] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0013] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a method for detecting cracks in asphalt pavement, the method comprising the following steps: Step S1: Acquire road surface images of the target road section, and perform illumination compensation based on the grayscale changes of adjacent pixels in the road surface images to generate road surface images to be analyzed; In this embodiment, the vehicle is controlled to travel along the target road segment at a speed of approximately 60 km / h. A linear array industrial camera mounted on the vehicle's acquisition beam acquires continuous road image frames at a spatial resolution of 0.5 m. The camera exposure time is controlled at 1 / 500 s, and the ISO sensitivity is set to 800 to reduce motion blur and noise interference. Each image frame is first divided into sub-blocks according to vertical and horizontal pixel height in the image coordinate system, and the average grayscale value of each sub-block is calculated. And calculate its grayscale mean with the entire image. deviation When the deviation is greater than the preset brightness offset threshold When there are 20 gray levels (e.g.,), the sub-block is marked as a brightness abnormality region. Further, the gray-level change gradient in the driving direction and the lateral lane direction is calculated between the brightness abnormality region and adjacent sub-blocks. If the gradient magnitude is less than the gradient threshold... (For example, 5) then a low-frequency brightness distribution is constructed based on the average grayscale value of adjacent sub-blocks and fitted as a bilinear brightness reference surface. Grayscale correction is then performed pixel-by-pixel on the entire image. If the gradient magnitude is greater than Then, only the sub-blocks with abnormal brightness are compensated for overall grayscale shift, thus obtaining the road surface image to be analyzed.

[0014] In another embodiment, continuous road surface image frames can be captured using a fixed line array industrial camera pre-deployed on the target road section.

[0015] Step S2: Determine the road indicator texture of the target road segment based on the grayscale changes of pixels in the road surface image to be analyzed; use the road indicator texture to remove non-road surface texture features from the road surface image to be analyzed; In a further embodiment, the gradient intensity under the Sobel operator is calculated for each pixel in the road image to be analyzed, and the gray level and gradient are accumulated pixel by pixel in the transverse lane direction to obtain a gray level accumulation sequence. With gradient accumulation sequence The brightest columns with H(x) in the top 10 percentiles are selected as candidate brightness bands, and then further selections are made from these candidate brightness bands. The largest columns form candidate road marking strips with a width of approximately 20–40 pixels. Then, Hough line detection or least-squares line fitting is performed within these candidate strips along the driving direction to filter out approximate straight line textures with a continuous length greater than 3m, a width between 10–20cm (converted pixel range), and an angle less than 5° with the road's extension direction; these are marked as road indicator textures. A binary mask is constructed by expanding 0.5 times the road marking width to both sides of the road indicator texture. Textures within this mask are marked as road indicator texture masks, while the area outside the mask is used as candidate road background regions. Connected component analysis can be used to identify connected components with a lateral span greater than 1.2 times the width of a single lane and whose outlines are parallel to the road indicator texture as shadow textures. Simultaneously, short-term textures appearing only in a few frames are detected using adjacent time frame difference detection. Road markings, shadow textures, and short-term textures are all removed from the image.

[0016] Step S3: Based on the road surface texture features in the road surface image to be analyzed, identify linear crack areas and associate the linear crack areas with the corresponding lane information of the target road segment to generate a crack feature set for each lane of the target road segment; In a further embodiment, the average grayscale value of each pixel is calculated using a 9×9 pixel local neighborhood window. The gray value is lower than the mean of the corresponding neighborhood. Subtract threshold Pixels with 8 gray levels (e.g., 8 gray levels) are labeled as dark texture candidate pixels. A connectivity search is performed on these dark texture candidate pixels based on 8-adjacency relationships, constraining the center-to-center distance between adjacent candidate pixels to no more than 2 pixels. The set of pixels meeting this condition is aggregated into linear texture segments, and the continuous length along the driving direction is calculated for each segment. Horizontal width and aspect ratio The preset crack length threshold is... =0.5m equivalent pixel count, crack width range of 3-15mm corresponding pixel count, linear feature threshold If a certain linear texture segment simultaneously satisfies 10, > , Falling within the width range and > If the cracks are identified, they are marked as linear crack areas. Subsequently, based on the camera calibration parameters and lane geometry data, the image coordinates of the crack areas are converted into world coordinates and projected onto the lane centerline coordinate system. The corresponding lane number and mileage markers along the route are matched, and attributes such as crack length, width, and direction are recorded to form a crack feature set for each lane of the target road segment.

[0017] Step S4: Identify the crack connectivity between adjacent lanes based on the crack feature set of each lane, and classify the maintenance priority of the cracks in the target road section based on the crack connectivity between adjacent lanes to obtain the pavement maintenance report of the target road section.

[0018] In a further embodiment, the start and end mileage of each crack in the direction of travel is recorded. , and horizontal position For any pair of cracks A and B on adjacent lanes, first calculate their overlap length in the driving direction. And calculate the horizontal spacing based on the horizontal coordinate. ,when Less than the lateral connectivity threshold (e.g., 0.5 times the lane width), and Greater than the vertical connectivity threshold (For example, at a depth of 1m), crack A and crack B are determined to be spatially connected. The letter A represents the starting mileage of crack A, and the other letters follow the same logic. Using a graph traversal approach, all cracks satisfying the connectivity condition are clustered according to lane number and mileage direction to form continuous crack pathways. The total length of each pathway in the driving direction is then calculated. Lateral span corresponding to the number of lanes crossed and overall length-to-width ratio The comprehensive risk score is calculated according to the preset weights α, β, and γ. For example, α=0.5, β=0.3, and γ=0.2 can be used. Based on the score from high to low, the continuous crack path is divided into first-level, second-level, and third-level maintenance priorities, and a pavement maintenance report for the target road section is generated, which includes the location and size of the cracks and the recommended maintenance priorities.

[0019] It is worth noting that the illumination compensation step in this application mainly performs "smoothing and equalization" on the entire image or large area based on the average grayscale value of sub-blocks and low-frequency brightness distribution. The purpose is to eliminate large-scale brightness deviations caused by camera exposure, street light distribution, and changes in the sun's altitude angle. Shadow textures (vehicle shadows, guardrail shadows, overpass shadows, etc.) usually exhibit drastic spatial variations, clear boundaries, and significant grayscale abrupt changes, belonging to local high-gradient structures. During illumination compensation, in order not to destroy the real texture, strong smoothing is deliberately avoided to "erase" this structure, so it is preserved. Therefore, in this method, the retention of shadow textures after illumination compensation is intentional: first, illumination compensation is used to flatten large-scale unevenness in brightness and darkness; then, geometric constraints such as "connected domain span + parallel to road indicator texture" are used to specifically identify shadow textures with large lateral spans and extending along the road direction; finally, they are removed separately in step S2 to avoid accidentally deleting real cracks.

[0020] Optionally, performing illumination compensation in step S1 includes: The road surface image is divided into sub-block regions along the driving direction and the lateral lane direction; In this embodiment, a single-frame road surface image is captured when the vehicle is traveling at approximately 60 km / h. After calibrating the camera's intrinsic and extrinsic parameters, the vertical direction of the image coordinate system is aligned with the road's driving direction, and the horizontal direction is aligned with the transverse lane direction. Based on the camera's imaging resolution and the desired spatial resolution, the image is divided into a regular grid of 64 pixels vertically and 64 pixels horizontally. Each grid corresponds to a road surface unit of approximately 0.5m × 0.5m on the ground, thus forming several rectangular sub-block regions distributed along the driving direction and transverse lane direction. During the division process, the sub-block boundaries are determined by integer division using pixel indices to ensure that adjacent sub-blocks do not overlap and cover the entire road surface image.

[0021] The average pixel grayscale value of each sub-block region is calculated, and sub-block regions whose average pixel grayscale value deviates from the average grayscale value of the overall road surface image by more than a preset brightness offset threshold are marked as brightness abnormal regions. In a further embodiment, for each sub-block region that has been divided, the grayscale values ​​of all pixels within it are summed and divided by the number of pixels to obtain the average grayscale value of that sub-block. Simultaneously, the global grayscale mean is calculated for the entire road surface image. The brightness offset is obtained by subtracting the average grayscale value of each sub-block from the global average grayscale value. ,when Greater than the preset brightness offset threshold When there are 20 gray levels, the sub-block area is determined to be significantly brighter or lower than the overall road surface brightness, and it is marked as a brightness abnormal area.

[0022] Based on the grayscale change gradient between the brightness anomaly region and the adjacent sub-block region, adaptive pixel brightness compensation is performed to obtain the road surface image to be analyzed.

[0023] In a further embodiment, for a marked brightness anomalous region, its adjacent normal sub-blocks in the driving direction and the transverse lane direction are used together as a local illumination estimation window. First, the gray-level change gradients of the brightness anomalous sub-block and its adjacent sub-blocks in both directions are calculated. , ,like and All are less than the gradient magnitude threshold (For example, 5) If the brightness change in this area is considered to be gradual, the average gray value of each sub-block within the window can be used as a sampling point. A brightness reference surface that changes slowly with the direction of travel and the lateral lane direction can be obtained using bilinear interpolation. Then, the gray value of the corresponding pixels in the entire image is subtracted from or added to the reference surface offset value point by point to achieve fine-grained brightness correction. If the gradient magnitude at the abnormal brightness sub-block is greater than... If a local area contains road surface texture or structural edges, then only all pixels within that sub-block are uniformly added or subtracted. The reverse compensation value is used to perform overall translational brightness adjustment, and finally output a road surface image with a more uniform brightness distribution.

[0024] Optionally, dividing the road surface image into sub-block regions along the driving direction and the lateral lane direction includes: Based on the camera field of view of the target road segment and the geometric distribution characteristics of lane lines in the road surface image, the road extension direction of the target road segment is determined, and the image coordinate system of the road extension direction of the target road segment and the road surface image is calibrated. In this embodiment, the camera is mounted in a fixed posture at the front of the detection vehicle, at a height of approximately 1.8m, with a pitch angle of approximately -5°. Its horizontal field of view is approximately 60°, and its vertical field of view is approximately 40°. Intrinsic and extrinsic parameters are calibrated using a calibration board. First, based on the horizontal and vertical field of view, as well as the camera's mounting height and pitch angle, the coverable road surface area in the current frame image is geometrically estimated. For example, within a longitudinal distance of approximately 3m to 25m in front of the vehicle, one or more lane areas are defined on either side of the vehicle's centerline. The corresponding pixel area for this road surface area is then deduced in the image coordinate system. This area is used as a candidate lane line search zone, located in the lower half of the image and with its horizontal width limited to the middle 60% of the image width. Subsequently, a Hough line transform is performed within the candidate lane line search zone to obtain several candidate lines, each corresponding to a direction vector in the image coordinate system. Subsequently, each candidate straight line is matched with typical lane line features based on its position and brightness characteristics in the image. Straight lines that clearly belong to structures such as guardrails or curbs are eliminated, retaining only straight lines located within the lanes and with higher brightness as the lane line candidate set. The direction vectors in the lane line candidate set are then clustered and weighted (weights can be determined by the straight line length and gradient response intensity) to obtain a normalized principal direction vector, which serves as the road's extension direction in the image plane. Then, combining the camera's mounting height, pitch angle, and the projection positions of nearby road reference points (such as the known ground position near the vehicle's front wheel contact point) in the image coordinate system, a geometric relationship is used to determine an axis parallel to the road's extension direction as the longitudinal direction and an axis perpendicular to this axis as the lateral direction, thus aligning the road's extension direction with the image coordinate system.

[0025] It is worth noting that the typical lane line features in this embodiment are derived from the standardized lane marking designs of highways and urban arterial roads. They are typically white or yellow stripes with a brightness significantly higher than the surrounding road surface texture, distributed in an approximately straight line along the road's extension direction, with a relatively stable width and a length much greater than the width, and their position changes smoothly in adjacent frames. In one-way multi-vehicle road sections, lane lines also exhibit a grouped distribution characteristic parallel to the road centerline and with spacing close to the designed lane width. The image coordinate system originates from the camera's imaging plane. After camera calibration, the top-left pixel of the image is used as the origin, with the horizontal direction as the x-axis and the vertical direction as the y-axis. The x-axis increases to the right, and the y-axis increases downwards. Each pixel coordinate is composed of an integer row and column index, and a one-to-one correspondence is established between the camera's intrinsic and extrinsic parameters and its installation posture and the road's three-dimensional coordinates.

[0026] Based on the calibration results, the vertical direction of the image coordinate system is set as the driving direction, and the horizontal direction is set as the lateral lane direction; In a further embodiment, based on the calibration results, the axis approximately parallel to the road extension direction in the image coordinate system is defined as the longitudinal direction (driving direction), and the axis orthogonal to it is defined as the transverse direction (lateral lane direction). Then, based on the camera resolution and the target road segment's ground width, the longitudinal and transverse pixel intervals are set (example values: longitudinal pixel interval = 64 pixels, transverse pixel interval = 64 pixels, corresponding to a ground length of approximately 0.4–0.6m after calibration conversion). When determining the pixel interval, the design lane width of the target road segment (e.g., 3.5m) is referenced, and the single-lane pixel width is calculated as lane width / single-pixel ground size. The single-pixel ground size is calculated using camera calibration parameters and known ground geometry. For example, selecting a lane width of 3.5m or a 5m longitudinal road marking in the image, counting the corresponding pixel span in the image, and dividing the actual length by the number of pixels yields the ground length corresponding to each pixel in the longitudinal and transverse directions within that area.

[0027] The road surface image is divided into strip-shaped areas along the direction of travel, and then further divided into rectangular sub-blocks along the transverse lane direction. These rectangular sub-blocks are called sub-block regions.

[0028] In a further embodiment, the image is divided longitudinally from the starting boundary according to the longitudinal pixel interval to generate several strip-shaped regions; each strip is then divided laterally into rectangular sub-blocks according to the horizontal pixel interval to form several rectangular sub-blocks, which are the sub-block regions.

[0029] Optionally, performing adaptive pixel brightness compensation includes: Calculate the grayscale gradient between the brightness anomaly region and the adjacent sub-block region in the driving direction and the lateral lane direction; If the magnitude of the grayscale change gradient in both the driving direction and the transverse lane direction at the brightness abnormal area is less than the preset gradient magnitude threshold, then a low-frequency brightness distribution estimation map is constructed based on the grayscale mean distribution of the brightness abnormal area and the adjacent sub-block areas. In this embodiment, several normal sub-blocks adjacent to the abnormal brightness area are selected in both the driving direction and the transverse lane direction, taking sub-blocks as units. The difference in the average gray value of adjacent sub-blocks is calculated according to the sub-block index order, and then divided by the sub-block spacing in the corresponding direction to obtain the gray value change gradient in the driving direction and the transverse lane direction. If the gradient magnitude in both directions at a certain abnormal brightness area is less than the gradient magnitude threshold 5, it is considered that the brightness change in the area is gradual. The average gray value of the abnormal brightness area and its adjacent sub-blocks can be used as sampling points. A low-frequency brightness distribution estimation map is constructed on the sub-block grid of the entire image by interpolation to guide the subsequent brightness reference surface fitting.

[0030] The difference between the mean gray value of each sub-block region and the brightness estimate at the corresponding position in the low-frequency brightness distribution estimation map is fitted as a brightness reference surface. The pixel gray values ​​of the road surface image are then corrected point by point using the brightness reference surface to obtain the road surface image to be analyzed. In a further embodiment, based on the aforementioned low-frequency brightness distribution estimation map, the brightness estimate of each sub-block at its corresponding position in the estimation map is read, and the difference between this estimate and the actual grayscale mean of the sub-block is calculated to obtain the brightness offset of the sub-block relative to the low-frequency brightness distribution. The brightness offsets of all sub-blocks are fitted according to their spatial positions to obtain a brightness reference surface that changes slowly along the driving direction and the lateral lane direction. For example, piecewise linear or quadratic surface fitting is used to make the brightness reference surface continuous and smooth across the entire image. Subsequently, each pixel in the road image is corrected point-by-point according to the offset of its sub-block on the brightness reference surface, reducing excessively high grayscale values ​​and increasing excessively low grayscale values, thereby outputting the road image to be analyzed.

[0031] If the magnitude of the grayscale change gradient in both the driving direction and the transverse lane direction at the area of ​​abnormal brightness is greater than the gradient magnitude threshold, then translation compensation is performed on the area of ​​abnormal brightness to obtain the road surface image to be analyzed.

[0032] In another embodiment, if the grayscale gradient magnitude in both the driving direction and the transverse lane direction within a certain brightness abnormal area is detected to be greater than the gradient magnitude threshold 5, it is determined that there is obvious road surface texture or structural edge within the area. If point-by-point correction is used, it may weaken the contrast of cracks or aggregate texture. Therefore, a translational compensation method is adopted for this type of brightness abnormal area: First, the difference between the grayscale mean of the area and the brightness estimate at the same position in the low-frequency brightness distribution estimation map is calculated. This difference is used as the compensation amount. The compensation amount is uniformly added or subtracted for all pixels in the brightness abnormal area to control the grayscale after compensation to still be within the effective range of 0 to 255, thereby outputting the road surface image to be analyzed.

[0033] Optionally, determining the road sign texture for the target road segment in step S2 includes: Calculate the gray value and gradient intensity of each pixel in the road image to be analyzed, and accumulate the gray value and gradient intensity of each pixel column along the driving direction, taking each pixel column in the transverse lane direction as a unit, to obtain the accumulated gray value and the accumulated gradient value respectively. In this embodiment, the grayscale value of each pixel in the road image to be analyzed is extracted, and the gradient intensity of the pixel is calculated using the Sobel operator. Then, the image is processed in the direction of travel, with each pixel column in the transverse lane direction as the unit, and the grayscale value and gradient intensity are accumulated in the direction of travel to obtain the accumulated grayscale value and gradient value of each column.

[0034] The column of pixels with the highest brightness is selected based on the grayscale accumulation value, and the column of pixels with the largest corresponding gradient accumulation value in the column of pixels with the highest brightness is selected as the candidate strip for road markings. In a further embodiment, all grayscale accumulated values ​​are sorted, and the pixel column with the top 5% of accumulated values ​​is selected as the pixel column with the highest brightness. Then, the pixel column with the largest gradient accumulated value is selected from these columns as the candidate strip for road markings. This candidate strip represents the most likely lane line position in the image.

[0035] The approximate straight line texture of the candidate road marking strip along the road extension direction is determined. Approximate straight line textures with a continuous length exceeding a preset marking length threshold in the driving direction and a lateral width within a preset marking width range are determined as road indication textures.

[0036] In a further embodiment, a Hough line transform is performed on the candidate road marking strips to extract approximate straight line textures along the road extension direction, and straight line segments with a continuous length greater than 50 pixels and a lateral width in the driving direction are selected and identified as road indicator textures.

[0037] Optionally, removing non-road texture features from the road surface image to be analyzed in step S2 includes: A road sign texture mask is constructed based on the geometric position of the road sign texture in the image coordinate system. Pixels inside the road sign texture mask are marked as road marking areas, and pixels outside the road sign texture mask are marked as road surface background candidate areas. In this embodiment, a binary mask is generated in the image coordinate system based on the pixel coordinates of the extracted road indication texture. The pixel value inside the mask is set to 1 to represent the road marking area, and the pixel value outside the mask is set to 0 to represent the road background candidate area. This mask is used to distinguish lane lines from background texture. The structure of the mask is consistent with the aforementioned image coordinate system, and each pixel is mapped to the image space through its row and column index.

[0038] Perform connected component analysis on the texture features of the candidate area of ​​the road background, and take the texture features whose span in the transverse lane direction exceeds the upper limit of the lane width of the target road segment and whose edge contour in the driving direction is parallel to the road indicator texture as the shadow texture features. In a further embodiment, connected component analysis is performed on the candidate regions of the road surface background. The pixel span of each connected component in the lateral lane direction and the edge direction in the driving direction are counted. If the lateral span of the connected component exceeds the number of pixels corresponding to the upper limit of lane width of 3.5 meters, and the edge contour direction is parallel to the road indicator texture, then the connected component is marked as a shadow texture feature. This method can effectively eliminate interference such as vehicle shadows and roadside bright reflections.

[0039] Detect short-term texture features of the road indicator texture mask and adjacent road surface background candidate regions on the time axis; In a further embodiment, by analyzing the grayscale and gradient changes inside the road indication texture mask and adjacent background candidate regions in consecutive frame images, short-term texture features with a duration of less than 3 frames are identified and marked as short-term texture features to eliminate temporary texture interference caused by driving bumps or instantaneous lighting changes.

[0040] Remove shadow texture features, short-term texture features, and road indicator texture masks from the road surface image to be analyzed.

[0041] In a further embodiment, pixels marked as shadow texture features, short-term texture features, and road indicator texture mask areas in the road surface image to be analyzed are removed or set to zero. At the same time, the spatial positions corresponding to the shadow texture feature and short-term texture feature areas are shifted by the corresponding pixel values ​​on the time axis to maintain image continuity, thereby obtaining a road surface image after removing non-road surface textures.

[0042] Of particular importance, the non-road surface texture features include shadow texture features, short-term texture features, and road indication textures.

[0043] Figure 2This is a non-road texture feature marker map of the road surface image to be analyzed in this embodiment of the invention; such as... Figure 2 As shown in the figure, the texture contained in the area circled in yellow is the road indicator texture, the texture features contained in the areas circled in red and yellow are the shadow texture features, and the texture features contained in the area circled in blue are the short-term texture features. In practical applications, the number of road indicator textures, shadow texture features, and short-term texture features is not unique. The non-road surface texture feature marking map in this application is merely an example, and this application does not limit the color marking and number of each texture.

[0044] Of particular importance, after removing shadow texture features and short-term texture features, the following are also included: On the time axis, the pixels corresponding to the shadow texture features and short-term texture features in the road surface image to be analyzed that do not contain shadow texture features and short-term texture features are shifted to the corresponding pixels in the road surface image to be analyzed that have had the shadow texture features and short-term texture features removed.

[0045] In this embodiment, a time series of continuous frame road surface images is established, assuming a frame rate of 25 frames / second. The pixel positions in each frame, after removing shadow texture features and short-term texture features, are marked to form a binary mask. For each frame of the road surface image to be analyzed, it is detected whether the corresponding pixel position has a valid grayscale value in the preceding and following frames. If the position has pixels that do not contain shadow textures and short-term textures in at least two preceding and following frames, the pixel is translated along the time axis to the corresponding position in the current frame to achieve pixel value compensation, thereby reducing texture loss caused by instantaneous illumination or occlusion. In this embodiment, the translation operation uses linear interpolation to perform a weighted average of the pixel values ​​of continuous frames, for example, using a weight of 0.4 for the previous frame, 0.4 for the following frame, and 0.2 for the current frame, ensuring smooth pixel transition while preserving the detailed structure of the road surface texture and avoiding excessive smoothing that could lead to the loss of crack edge information.

[0046] Optionally, identifying the linear crack region in step S3 includes: The gray values ​​of each pixel in the road surface texture feature are statistically analyzed using a preset local neighborhood window. Pixels with gray values ​​lower than the gray average of the corresponding local neighborhood are marked as dark texture candidate pixels. In this embodiment, a local neighborhood window of 9×9 pixels (or 7×7 pixels) is used to scan the road surface image to be analyzed after removing non-road surface textures. The gray value of each pixel in its neighborhood is counted, and pixels with gray values ​​more than 20 gray units lower than the neighborhood average are marked as dark texture candidate pixels.

[0047] Determine the pixel spacing between adjacent dark texture candidate pixels, aggregate adjacent dark texture candidate pixels with a pixel spacing less than a preset maximum spacing threshold into a linear texture fragment, and count the continuous length, horizontal width and aspect ratio of the linear texture fragment. In a further embodiment, the candidate pixels of the dark texture are aggregated according to their spatial positions in the driving direction and the lateral lane direction. The pixel spacing between adjacent candidate pixels is calculated. If the spacing is less than 5 pixels, these candidate pixels are merged to form a linear texture fragment. The continuous length, lateral width and aspect ratio of each fragment are then calculated.

[0048] Linear texture fragments with a continuous length greater than a preset crack length threshold, a lateral width falling within a preset crack width range, and an aspect ratio greater than a preset linear feature threshold are marked as linear crack regions.

[0049] In a further embodiment, linear texture fragments with a continuous length greater than 50 pixels, a horizontal width in the range of 2 to 10 pixels, and an aspect ratio greater than 4 are marked as linear crack regions.

[0050] Optionally, identifying the crack connectivity between adjacent lanes in step S4 includes: The overlap length of the linear crack region in the driving direction of each lane is determined based on the crack feature set of each lane; the spacing of the linear crack regions of adjacent lanes in the transverse lane direction is calculated. In this embodiment, the linear crack region in each lane is pixel-projected in the driving direction, the length of the crack region along the driving direction is mapped into a one-dimensional vector, and the number of overlapping pixels of each pair of crack region vectors in adjacent lanes is calculated. If the number of overlapping pixels exceeds 30 pixels, it is determined that there is sufficient continuity in the longitudinal direction.

[0051] In a further embodiment, the minimum edge spacing between adjacent lane crack regions in the lateral lane direction is measured. Specifically, the pixel boundary coordinates of each crack region in the lateral direction are calculated, and the minimum difference between the boundary coordinates of the two regions is found. If the spacing is less than 15 pixels, the two regions are considered to be close in the lateral direction and can form a potential connected path.

[0052] If the distance between the linear crack regions of adjacent lanes in the lateral lane direction is less than a preset lateral connectivity threshold, and the overlap length in the driving direction is greater than a preset longitudinal connectivity threshold, then the linear crack regions of the adjacent lanes are determined to be connected.

[0053] In a further embodiment, the longitudinal continuity and lateral proximity are combined for determination. When the longitudinal overlap length is greater than 30 pixels and the lateral spacing is less than 15 pixels, the crack areas of adjacent lanes are marked as connected crack areas, and a cross-lane crack path structure is generated, recording the start and end pixel positions and span information of each path.

[0054] Optionally, the maintenance priority for classifying cracks in the target road section in step S4 includes: Connecting linear crack regions in adjacent lanes are combined into a continuous crack path, and the total length, lateral span and aspect ratio of the continuous crack path in the driving direction are calculated. In this embodiment, adjacent lane crack regions that are determined to be connected are sequentially combined into continuous crack paths according to their start and end pixel positions in the driving direction to form a two-dimensional crack topology. The total pixel length of each path in the driving direction, the maximum span in the transverse lane direction, and the aspect ratio are recorded to characterize the spatial distribution characteristics of the cracks.

[0055] According to the preset calculation weight ratio, the comprehensive risk score of the continuous crack passage in the driving direction is calculated as the total length, lateral span and aspect ratio, and the maintenance priority of each continuous crack passage is determined according to the ranking of the comprehensive risk scores.

[0056] In a further embodiment, based on the driving direction length, lateral span, and aspect ratio of the continuous crack pathways, a comprehensive risk score is calculated according to a preset weighting ratio, for example, driving direction length accounts for 50%, lateral span accounts for 30%, and aspect ratio accounts for 20%. The spatial characteristics of each pathway are normalized and then weighted and summed to obtain the quantitative risk value of each pathway. All continuous crack pathways are sorted from high to low according to their comprehensive risk scores, and maintenance priorities are assigned based on the sorting results. For example, scores greater than 0.7 are classified as Level 1 priority, 0.5-0.7 as Level 2 priority, and scores less than 0.5 as Level 3 priority. The priority, location, and spatial characteristics of each pathway are recorded in the maintenance report.

[0057] Optionally, this application also provides an asphalt pavement crack detection system 100 for performing the asphalt pavement crack detection method described above, the asphalt pavement crack detection system comprising: The data acquisition module 101 is used to acquire road surface images of the target road section and perform illumination compensation based on the grayscale changes of adjacent pixels in the road surface image to generate a road surface image to be analyzed. The texture analysis module 102 is used to determine the road indicator texture of the target road segment based on the grayscale changes of pixels in the road surface image to be analyzed; and to remove non-road surface texture features from the road surface image to be analyzed using the road indicator texture. The crack recognition module 103 is used to identify linear crack areas based on the road texture features in the road surface image to be analyzed, and associate the linear crack areas with the corresponding lane information of the target road segment to generate a crack feature set for each lane of the target road segment. The maintenance priority division module 104 is used to identify the crack connectivity between adjacent lanes based on the crack feature set of each lane, and to divide the maintenance priority of the cracks in the target road section based on the crack connectivity between adjacent lanes, so as to obtain the pavement maintenance report of the target road section.

[0058] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0059] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for detecting cracks in an asphalt pavement, characterized by, The method comprises the following steps: Step S1: Collecting a road surface image of a target road section, and performing illumination compensation according to the gray level change of adjacent pixels in the road surface image to generate a to-be-analyzed road surface image; Step S2: According to the gray level change of the pixel points in the to-be-analyzed road surface image, the road indicating texture of the target road section is determined; the non-road surface texture features in the to-be-analyzed road surface image are removed by using the road indicating texture; Step S3: According to the road surface texture features in the to-be-analyzed road surface image, a linear crack region is recognized, and the linear crack region is associated with the corresponding lane information of the target road section to generate a crack feature set of each lane of the target road section; Step S4: The crack connectivity between adjacent lanes is recognized according to the crack feature set of each lane, and the maintenance priority of the cracks of the target road section is divided based on the crack connectivity between adjacent lanes to obtain a road surface maintenance report of the target road section.

2. The asphalt pavement crack detection method according to claim 1, characterized by, The illumination compensation in step S1 comprises: Dividing the road surface image into sub-block regions along the driving direction and the transverse lane direction; Statistically calculating the mean value of the pixel gray values in each sub-block region, and marking the sub-block region whose mean value of the pixel gray values deviates from the gray mean value of the entire road surface image by more than a preset luminance deviation threshold as a luminance abnormal region; According to the gray level change gradient between the luminance abnormal region and the adjacent sub-block region, adaptive pixel luminance compensation is performed to obtain the to-be-analyzed road surface image.

3. The asphalt pavement crack detection method according to claim 2, characterized by, The division of the road surface image into sub-block regions along the driving direction and the transverse lane direction comprises: According to the camera field of view angle of the target road section and the geometric distribution characteristics of the lane lines in the road surface image, the road extension direction of the target road section is determined, and the road extension direction of the target road section and the image coordinate system of the road surface image are calibrated; According to the calibration result, the longitudinal direction of the image coordinate system is set as the driving direction, and the transverse direction is set as the transverse lane direction; The road surface image is divided into strip-shaped regions along the driving direction, and the strip-shaped regions are divided into rectangular sub-blocks along the transverse lane direction, and the rectangular sub-blocks are the sub-block regions.

4. The asphalt pavement crack detection method according to claim 2, characterized by, The adaptive pixel luminance compensation comprises: Calculating the gray level change gradient between the luminance abnormal region and the adjacent sub-block region in the driving direction and the transverse lane direction; If the amplitude of the gray level change gradient of the luminance abnormal region in the driving direction and the transverse lane direction is less than a preset gradient amplitude threshold, a low-frequency luminance distribution estimation map is constructed according to the gray mean value distribution of the luminance abnormal region and the adjacent sub-block region; The difference between the gray mean value of each sub-block region and the luminance estimation value at the corresponding position in the low-frequency luminance distribution estimation map is fitted as a luminance reference surface, and the pixel gray of the road surface image is point-by-point corrected by using the luminance reference surface to obtain the to-be-analyzed road surface image; If the amplitude of the gray level change gradient of the luminance abnormal region in the driving direction and the transverse lane direction is greater than the gradient amplitude threshold, a translational compensation is performed on the luminance abnormal region to obtain the to-be-analyzed road surface image.

5. The asphalt pavement crack detection method according to claim 1, characterized by, The determination of the road indicating texture of the target road section in step S2 comprises: Calculating the gray value and gradient intensity of each pixel point in the to-be-analyzed road surface image, and accumulating the gray value and gradient intensity in each pixel column in the driving direction as a unit in the transverse lane direction to obtain the gray accumulation value and the gradient accumulation value, respectively; Screening the pixel column with the highest brightness according to the gray value accumulation, and taking the pixel column with the largest gradient accumulation value in the pixel column with the highest brightness as a road marking candidate strip; Determining the approximate straight line texture of the road marking candidate strip along the extension direction of the road, and taking the approximate straight line texture with a continuous length exceeding a preset marking length threshold in the driving direction and a lateral width within a preset marking width range as a road indication texture.

6. The asphalt pavement crack detection method according to claim 1, characterized by, The step S2 includes: According to the geometric position of the road indication texture in the image coordinate system, a road indication texture mask is constructed, and pixels within the road indication texture mask are marked as road marking regions, and pixels outside the road indication texture mask are marked as road surface background candidate regions; Performing connected component analysis on the texture features of the road surface background candidate region, and taking the texture features with a span in the lateral lane direction exceeding the upper limit of the lane width of the target road section and an edge profile in the driving direction parallel to the road indication texture as shadow texture features; Detecting short-time texture features in the time axis inside the road indication texture mask and adjacent road surface background candidate regions; Removing the shadow texture features, short-time texture features, and road indication texture mask from the road surface image to be analyzed.

7. The asphalt pavement crack detection method according to claim 1, characterized by, The step S3 includes: Statistically determining the gray value of each pixel in the road surface texture feature with a preset local neighborhood window, and marking the pixels with a gray value lower than the average gray value of the corresponding local neighborhood as dark texture candidate pixels; Determining the pixel spacing of adjacent dark texture candidate pixels, aggregating adjacent dark texture candidate pixels with a pixel spacing less than a preset maximum spacing threshold as linear texture segments, and statistically determining the continuous length, lateral width, and aspect ratio of the linear texture segments; Marking the linear texture segments with a continuous length greater than a preset crack length threshold, a lateral width within a preset crack width range, and an aspect ratio greater than a preset linear feature threshold as linear crack regions.

8. The asphalt pavement crack detection method of claim 1, wherein, The step S4 includes: Determining the overlapping length of the linear crack regions of each lane in the driving direction according to the crack feature set of each lane; and calculating the spacing of the linear crack regions of adjacent lanes in the lateral lane direction; If the spacing of the linear crack regions of adjacent lanes in the lateral lane direction is less than a preset lateral connectivity threshold, and the overlapping length in the driving direction is greater than a preset longitudinal connectivity threshold, it is determined that the linear crack regions of the adjacent lanes are connected.

9. The asphalt pavement crack detection method of claim 1, wherein, The step S4 includes: Combining the connected linear crack regions in adjacent lanes into continuous crack paths, and calculating the total length, lateral span, and aspect ratio of the continuous crack paths in the driving direction; According to a preset calculation weight proportion, the comprehensive risk score of the total length, lateral span, and aspect ratio of the continuous crack paths in the driving direction is calculated, and the maintenance priority of each continuous crack path is divided according to the ordering of the comprehensive risk score.

10. An asphalt pavement crack detection system characterized by, A system for performing the asphalt pavement crack detection method of claim 1, the asphalt pavement crack detection system comprising: The data acquisition module is configured to acquire a road surface image of the target road section, perform light compensation according to a gray level change of adjacent pixels in the road surface image, and generate a to-be-analyzed road surface image. The texture analysis module is configured to determine road indication texture of the target road section according to a gray level change of the pixels in the to-be-analyzed road surface image, and remove non-road surface texture features in the to-be-analyzed road surface image by using the road indication texture. The crack identification module is configured to identify a linear crack region according to road surface texture features in the to-be-analyzed road surface image, associate the linear crack region with corresponding lane information of the target road section, and generate a crack feature set of each lane of the target road section. The maintenance priority division module is configured to identify crack connectivity between adjacent lanes according to the crack feature set of each lane, divide a maintenance priority of cracks of the target road section based on the crack connectivity between the adjacent lanes, and obtain a road surface maintenance report of the target road section.

Images