Road marking thickness detection method, device, equipment and medium

Abstract

The application discloses a road marking thickness detection method, device, equipment and medium, and relates to the technical field of marking measurement. The method comprises the following steps: acquiring road videos of a plurality of video sources, wherein the road videos comprise a plurality of continuous video frame images; performing two-dimensional image analysis on the road videos to obtain a visual health score of a marking area in a video frame image; determining a target marking area as an abnormal area, wherein the target marking area comprises a marking area with a visual health score lower than a preset threshold; performing reference plane fitting and thickness calculation on the abnormal area to generate a thickness change profile; and performing quality detection on the abnormal area based on the thickness change profile to obtain a detection result. The application can realize rapid detection of road marking thickness and has the technical effect of improving detection efficiency.

Description

Technical Field

[0001] This application belongs to the technical field of road marking measurement, and particularly relates to a method, device, equipment and medium for detecting the thickness of road markings. Background Technology

[0002] Performance evaluation of road traffic safety facilities is crucial for improving road traffic safety. With the rapid development of my country's economy and the rapid increase in the number of motor vehicles, road traffic safety issues have become increasingly prominent. Some traffic safety facilities on road sections with long operating years have experienced performance degradation and are unable to meet usage requirements. Among these issues, the problem of severely worn road markings is particularly prominent. Worn road markings have poor clarity and weaken their guidance function for drivers.

[0003] In existing technologies, the main method for detecting the thickness of road markings is to first conduct a visual inspection of the road by hand, and then use a handheld thickness measuring instrument to measure the thickness. The thickness of the marking is calculated by using the height difference between the selected measurement point and the base. This method not only requires high skills from the inspectors, but also involves manual point-by-point measurement, which is inefficient and labor-intensive. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a method, apparatus, device, and storage medium for detecting the thickness of road markings, thereby rapidly detecting the thickness of road markings and improving detection efficiency.

[0005] In a first aspect, this application provides a method for detecting the thickness of road markings, including:

[0006] Acquire road videos from multiple video sources, wherein the road videos include multiple consecutive video frame images;

[0007] Two-dimensional image analysis was performed on the road video to obtain the visual health score of the marking area in the video frame image;

[0008] The target marking area is identified as an abnormal area, and the target marking area includes the marking area where the visual health score is lower than a preset threshold.

[0009] The abnormal region is fitted to a reference plane and its thickness is calculated to generate a thickness variation profile.

[0010] The abnormal area is subjected to quality inspection based on the thickness variation profile, and the inspection results are obtained.

[0011] Preferably, the step of performing two-dimensional image analysis on the road video to obtain a visual health score for the marking area in the video frame image includes:

[0012] The video frame images are preprocessed to obtain the image to be analyzed;

[0013] Local feature extraction is performed on the image to be analyzed to obtain multidimensional features, which include contrast features, edge sharpness features, surface texture features, and continuity features.

[0014] Based on the preset analysis strategy, scores are calculated for the contrast feature, edge sharpness feature, surface texture feature and continuity feature respectively, to obtain the first score corresponding to the contrast feature, the second score corresponding to the edge sharpness feature, the third score corresponding to the surface texture feature and the fourth score corresponding to the continuity feature.

[0015] The first score, second score, third score, and fourth score are summed according to the preset weighting to obtain the visual health score.

[0016] Preferably, the step of performing two-dimensional image analysis on the road video to obtain a visual health score for the marking area in the video frame image includes:

[0017] The multiple video frame images are input into a preset scoring model;

[0018] In the preset scoring model, for each video frame image, a global feature extraction network extracts global features;

[0019] The spatial attention network analyzes the global features to obtain edge regions, lane marking center regions, and lane marking-road surface intersection regions;

[0020] The regression classification network performs a global score on the edge region, the center region of the marking, and the intersection region of the marking and the road surface to obtain a visual health score.

[0021] Preferably, after determining the target marking area as an abnormal area, the method further includes:

[0022] Obtain the first location information of the abnormal region;

[0023] In the preset cloud map, abnormal locations are marked based on the first location information;

[0024] A detection route is generated based on the abnormal location;

[0025] Based on the detection route and the abnormal area, a detection task list is generated;

[0026] Acquire the second location information and working status of multiple specialized inspection vehicles;

[0027] Based on the second location information and working status, a target professional inspection vehicle is determined, and the inspection task list is sent to the target professional inspection vehicle so that the target professional inspection vehicle can perform line thickness inspection according to the inspection task list.

[0028] Preferably, the step of performing reference plane fitting and thickness calculation on the abnormal region to generate a thickness variation profile includes:

[0029] Obtain the three-dimensional laser point cloud of the abnormal region;

[0030] The three-dimensional laser point cloud is subjected to line extraction to obtain line point cloud clusters;

[0031] Within a local area of ​​each marking cross section in the abnormal region, a road reference surface is dynamically fitted;

[0032] For each cross section of the road marking, calculate the vertical distance from all points in the marking point cloud cluster to the road reference surface corresponding to the cross section;

[0033] Generate a profile of the thickness variation of the markings in the abnormal region based on the vertical distance.

[0034] Preferably, the step of extracting markings from the three-dimensional laser point cloud to obtain a marking point cloud cluster includes:

[0035] The abnormal region is segmented by marking lines using a preset semantic segmentation model to generate a pixel-level mask;

[0036] The three-dimensional laser point cloud is projected onto the video frame image using a preset transformation matrix;

[0037] The datum point cloud cluster is extracted from the three-dimensional laser point cloud using the pixel-level mask.

[0038] Preferably, the step of performing quality inspection on the abnormal area based on the thickness variation profile to obtain the inspection result includes:

[0039] Based on the thickness variation profile, the quality indicators corresponding to the markings are determined, including average thickness, minimum thickness, thickness uniformity index, and cross-sectional uniformity index.

[0040] The quality indicators are compared with preset indicators to obtain the comparison results for each quality indicator;

[0041] The evaluation result of the abnormal area is determined by combining the weights of the comparison results of each of the aforementioned quality indicators.

[0042] Secondly, this application provides a road marking thickness detection device for performing the road marking thickness detection method, the device comprising:

[0043] The acquisition module is used to acquire road videos from multiple video sources, wherein the road videos include multiple consecutive video frame images;

[0044] The analysis module is used to perform two-dimensional image analysis on the road video to obtain a visual health score for the marking area in the video frame image;

[0045] The determination module is used to determine the target marking area as an abnormal area, the target marking area including the marking area where the visual health score is lower than a preset threshold;

[0046] The generation module is used to perform reference plane fitting and thickness calculation on the abnormal region to generate a thickness variation profile.

[0047] The detection module is used to perform quality detection on the abnormal area based on the thickness change profile and obtain the detection result.

[0048] Thirdly, this application provides an electronic device, the device including: a processor and a memory storing computer program instructions;

[0049] The processor implements the road marking thickness detection method when executing the computer program instructions.

[0050] Fourthly, this application provides a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the road marking thickness detection method.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] The road marking thickness detection method, apparatus, equipment, and storage medium described in this application can achieve full-section coverage of road markings through simultaneous acquisition of multiple video streams and extract multi-dimensional visual features of the marking area using image processing algorithms. When the feature value of a certain area deviates from the normal range, a three-dimensional laser scanning device is triggered to perform high-precision point cloud acquisition of that area. In the point cloud data processing stage, a road surface reference plane is dynamically constructed to eliminate road undulation interference, and the vertical height of each point of the marking is accurately calculated to form a thickness distribution map. This map can automatically identify areas of abrupt thickness changes and output detection conclusions in conjunction with preset quality thresholds. That is, by using visual images and thickness change profile maps to detect the thickness of the markings, the thickness data of the markings can be measured quickly, thereby improving the efficiency of marking thickness detection; at the same time, it can also avoid the subjective errors of manual inspection and effectively improve the accuracy of thickness calculation. Attached Figure Description

[0053] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a flowchart illustrating the road marking thickness detection method in an embodiment of this application;

[0055] Figure 2 This is a schematic diagram of the road marking thickness detection device in the embodiments of this application;

[0056] Figure 3 This is a schematic diagram of the structure of the electronic device in the embodiments of this application.

[0057] The attached figures are labeled as follows:

[0058] 201. Acquisition module; 202. Analysis module; 203. Determination module; 204. Generation module; 205. Detection module; 301. Processor; 302. Memory; 303. Communication interface; 304. Bus. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are intended only to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0060] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0061] The road marking thickness detection method provided in the embodiments of this application is described below.

[0062] Reference Figure 1 As shown, the methods for detecting the thickness of road markings include S101-S105:

[0063] S101, acquire road videos from multiple video sources, the road videos include multiple consecutive video frame images.

[0064] In this embodiment of the application, the multiple video sources can be road driving videos uploaded by ordinary vehicles through the cloud, road surface monitoring cameras, or inspection vehicles. There are no limitations here, as long as the video source can provide road video.

[0065] S102, perform two-dimensional image analysis on the road video to obtain the visual health score of the marking area in the video frame image.

[0066] In the embodiments of this application, a neural network model can be used for image recognition to obtain a visual health score of the marking area in the video frame image. It can be understood that the visual health score can be an evaluation index that quantifies the integrity of the marking surface through image features, which can quickly screen potential abnormal areas. The neural network model can be a convolutional neural network that can realize image recognition and output a score, and is not limited here.

[0067] Image recognition using neural network models can quickly monitor marked areas, quantify indicators, and clearly show the overall wear status of the marked area.

[0068] S103, the target marking area is identified as an abnormal area, and the target marking area includes the marking area where the visual health score is lower than a preset threshold.

[0069] In this embodiment of the application, a preset threshold is manually set to determine which areas belong to the target marking area.

[0070] S104: Fit the reference plane and calculate the thickness of the abnormal area to generate a thickness variation profile.

[0071] In this embodiment of the application, since the road may have slopes or potholes, a reference plane can be fitted first, and then the thickness can be calculated based on the reference plane to obtain a thickness change profile of each abnormal area, which facilitates thickness calculation and measurement.

[0072] Specifically, reference plane fitting can refer to establishing a three-dimensional reference surface for road surface smoothness to provide a spatial coordinate system for thickness, and the thickness variation profile can be a continuous distribution map of the vertical dimension of the marking.

[0073] S105, quality inspection of abnormal areas is performed based on thickness variation profile, and the inspection results are obtained.

[0074] In this embodiment of the application, the quality of abnormal areas can be detected based on the thickness change profile of the marking to obtain the detection result, that is, to identify the area of ​​thickness change of the marking, so as to facilitate the detection of marking thickness and marking quality.

[0075] In this embodiment of the application, S102 may specifically include:

[0076] The video frame images are preprocessed to obtain the images to be analyzed;

[0077] Local feature extraction is performed on the image to be analyzed to obtain multidimensional features;

[0078] Based on the preset analysis strategy, scores are calculated for contrast features, edge sharpness features, surface texture features and continuity features respectively, to obtain the first score corresponding to contrast features, the second score corresponding to edge sharpness features, the third score corresponding to surface texture features and the fourth score corresponding to continuity features.

[0079] The visual health score is obtained by summing the first, second, third, and fourth scores according to the preset weight allocation.

[0080] In this embodiment, to obtain an accurate visual health score, the video frame image can first be preprocessed. Preprocessing refers to operations such as noise reduction, brightness adjustment, or geometric correction of the video frame, which can be achieved using Gaussian filtering combined with histogram equalization to eliminate noise and illumination interference generated during image acquisition. Then, feature extraction is performed on the image to be analyzed. A convolutional neural network can be used to extract global and local features to obtain a comprehensive score for the marked area.

[0081] Specifically, the marking region can be segmented and multi-dimensional feature parameters can be extracted using edge detection and texture analysis algorithms. Specifically, the Sobel operator combined with the gray-level co-occurrence matrix can be used to quantify the visual state of the marking in the image.

[0082] The multidimensional features include contrast features, edge sharpness features, surface texture features, and continuity features. Specifically, the contrast feature represents the degree of grayscale difference between the marking and the road surface background, which can be achieved by calculating the grayscale variance between the marking area and the adjacent road surface area, and is used to assess the degree of marking fading. The edge sharpness feature represents the steepness of the marking boundary transition, which can be achieved by calculating the attenuation rate of the edge gradient amplitude, and is used to determine the wear condition of the marking edge. The surface texture feature represents the uniformity of the particle distribution on the marking surface, which can be achieved by analyzing the entropy value and contrast parameter of the gray-level co-occurrence matrix, and is used to identify surface peeling or cracks in the marking. The continuity feature represents the integrity of the marking in the extension direction, which can be achieved by detecting the number of interruptions in the connected domain of the marking area, and is used to assess the marking missingness.

[0083] The calculation methods for the first, second, third, and fourth scores are explained in detail below:

[0084] To address contrast characteristics, morphological dilation of the marking area can be performed to obtain adjacent road surface areas, ensuring comparability of the two areas under lighting conditions. Then, the image is converted to the Lab color space, and the brightness channel is selected for calculation, or a grayscale image is used. The brightness statistics of the marking area and the road surface area are then calculated, and finally normalized to obtain the first score.

[0085] To address edge sharpness features, the inner and outer edges of the marking area can be easily extracted to create an edge neighborhood, typically a strip of 2-5 pixels on each side of the edge. Then, the Sobel operator is applied to the luminance channel to calculate the gradient. Next, the intensity profile is extracted along the edge normal direction, and the rate of change of intensity from the road surface to the center of the marking is calculated. Finally, the gradient features of all edge points are statistically analyzed, and the sum of the mean and variance is calculated to obtain the second score.

[0086] For surface texture features, the edge region can be removed first by morphological erosion to obtain the core region of the marking. Then, multi-scale texture analysis is performed. The method of multi-scale texture analysis is not limited here, as long as it can be performed. After that, defect monitoring is performed. Finally, the sum of multiple defect indicators is combined to obtain the third score.

[0087] To address the continuity characteristics, the marking area can first be skeletonized, then analyzed to detect the endpoints and gaps of the skeleton. Next, the total length and breakage ratio, main continuity indicators and the percentage of the largest continuous segment are calculated. The total length and breakage ratio, main continuity indicators and the percentage of the largest continuous segment are then normalized to obtain the fourth score.

[0088] After obtaining the first, second, third, and fourth scores, the scores can be summed using preset weights to obtain a visual health score. The preset weights are weighting coefficients set according to the degree of influence of different features on visual health status. Specifically, they can be determined using the analytic hierarchy process combined with expert experience to comprehensively reflect the overall quality of the markings.

[0089] In this embodiment of the application, a neural network model can be trained using a large amount of training data to obtain a first score, a second score, a third score, and a fourth score.

[0090] In this embodiment, after Gaussian filtering to remove random noise, the video frame image is enhanced by histogram equalization to form the image to be analyzed. The Sobel operator is used to extract the edge contours of the road markings, and a region growing algorithm is used to segment the marking regions. Within each marking region, the gray-level variance between the marking and adjacent road areas is calculated as a contrast score; the attenuation slope of the edge gradient curve is calculated as an edge sharpness score; the entropy value of the gray-level co-occurrence matrix is ​​extracted as a surface texture score; and the proportion of interrupted points in the connected components of the marking is statistically analyzed as a continuity score. These four scores are weighted and summed according to preset weighting coefficients, and the final output is a health score reflecting the visual state of the road markings.

[0091] In this embodiment of the application, in order to ensure the accuracy of the visual health score, S102 may further include:

[0092] Multiple video frame images are input into a preset scoring model;

[0093] In the preset scoring model, for each video frame image, the global feature extraction network extracts global features;

[0094] Spatial attention network analyzes global features to obtain edge regions, marking center regions, and marking-road surface intersection regions;

[0095] The regression classification network performs global scoring on edge regions, the center region of the marking, and the intersection region of the marking and the road surface to obtain a visual health score.

[0096] In this embodiment, the preset scoring model can be a deep learning-based image analysis model, specifically constructed using a convolutional neural network combined with an attention mechanism, used to automatically extract multi-level features of the marking area and perform comprehensive scoring. The global feature extraction network refers to a feature extraction module used to capture the overall structure of the image, specifically implemented using a residual network or a visual transformer, capable of identifying the overall correlation between the marking area and the surrounding road surface. The spatial attention network is a module used to locate key areas, specifically employing a structure combining channel attention and spatial attention, focusing on feature differences in the marking edges, center, and junction areas by analyzing the spatial distribution of global features. The regression classification network is a module used to output the score, specifically employing a structure combining fully connected layers and regression layers, generating a global score result by fusing local features from different regions.

[0097] Specifically, after video frame images are input into a pre-defined scoring model, a global feature extraction network first extracts global features including color, texture, and spatial relationships. Then, a spatial attention network generates an attention weight map based on the global features, dynamically enhancing the feature responses of edge regions, the center region of the road marking, and the road marking-road interface, while suppressing interference from non-critical regions. A regression classification network receives the enhanced regional features and, through weighted fusion and nonlinear mapping, outputs a score reflecting the visual health status of the road marking. Thus, abnormal conditions such as road marking wear, peeling, or contamination can be quantitatively represented through the score.

[0098] In this embodiment, automated analysis is achieved through a preset scoring model. Key areas are accurately located by global feature extraction and spatial attention mechanism. A comprehensive score is then given by combining a regression classification network, which effectively avoids the instability of manual judgment and improves the detection sensitivity of minor defects in the markings.

[0099] It is worth noting that the first, second, third, and fourth scores of the multidimensional features and the visual health score of the global features can be weighted and summed to obtain a more accurate visual health score.

[0100] In some embodiments, after step 104, the method may further include:

[0101] Obtain the first location information of the abnormal region;

[0102] In the preset cloud map, abnormal locations are marked based on the first location information;

[0103] Generate detection routes based on abnormal locations;

[0104] Based on the detection route and abnormal areas, a list of detection tasks is generated;

[0105] Acquire the second location information and working status of multiple specialized inspection vehicles;

[0106] Based on the second location information and working status, the target professional inspection vehicle is identified, and the inspection task list is sent to the target professional inspection vehicle so that the target professional inspection vehicle can perform the marking thickness inspection according to the inspection task list.

[0107] In this embodiment, the first location information refers to the coordinates or geographic identifier of the abnormal area on the road. Specifically, it can be implemented using GPS positioning or image coordinate system transformation technology to accurately locate the spatial distribution of the abnormal area.

[0108] Preset cloud maps refer to digital map platforms that store road network information. They can be constructed using geographic information systems and are used for real-time updates and visualization of abnormal locations.

[0109] The detection route refers to the path planning result connecting multiple abnormal locations. Specifically, it can be generated using the shortest path algorithm or task priority ranking, and is used to guide detection vehicles to efficiently cover the target area.

[0110] The detection task list is a structured data set that includes the coordinates of abnormal areas, detection priorities, and task parameters. It can be dynamically generated through a database and is used to standardize the detection process.

[0111] A professional inspection vehicle refers to a mobile inspection unit equipped with laser thickness measurement equipment. Specifically, it can report its location and status in real time through an onboard communication module to achieve dynamic task allocation.

[0112] Specifically, once an abnormal area is identified, its location information is extracted and mapped onto a cloud map, forming markers with geographic coordinates. The detection route is automatically generated based on the spatial distribution of these markers, prioritizing connections between adjacent abnormal points or optimizing the path according to road conditions. The detection task list integrates route information with abnormal area parameters, such as lane marking type and detection accuracy requirements. The location and operating status of the specialized detection vehicle are uploaded to the cloud in real time via its onboard terminal. The system selects the most suitable vehicle to perform the task based on proximity or load balancing strategies and pushes the detection task list to the vehicle's control terminal.

[0113] This allows for automatic optimization of detection routes and intelligent vehicle scheduling through cloud maps and dynamic task allocation mechanisms, reducing manual intervention and improving task execution efficiency.

[0114] In this embodiment of the application, S104 may further include:

[0115] Obtain the 3D laser point cloud of the abnormal region;

[0116] Marking lines are extracted from the 3D laser point cloud to obtain a cluster of marking line point clouds;

[0117] Within the local area of ​​each marking cross section in the abnormal region, the road reference surface is dynamically fitted;

[0118] For each cross section of the road marking, calculate the vertical distance from all points in the marking point cloud cluster to the road reference surface corresponding to the cross section;

[0119] Generate a profile of the thickness variation of the marker lines in the abnormal area based on the vertical distance.

[0120] In this embodiment of the application, during the process of generating the thickness change profile, it is first necessary to obtain the three-dimensional laser point cloud of the abnormal area in order to construct the three-dimensional model of the abnormal area. The three-dimensional laser point cloud can be a set of points containing spatial coordinate information obtained by LiDAR scanning. The three-dimensional data of the road surface can be collected by a vehicle-mounted laser scanning system during movement, which can be used to provide spatial geometric data with millimeter-level accuracy for the calculation of the marking thickness.

[0121] Since 3D laser point clouds are discrete points, in order to extract actual road marking data more accurately, road marking extraction can be performed to obtain road marking point cloud clusters. The road marking point cloud clusters are sets of points belonging to road markings separated from the 3D point cloud. Specifically, a deep learning model can be used to perform semantic segmentation on the point cloud, and white or yellow marking areas can be filtered out by color reflection intensity thresholds to achieve accurate differentiation between marking areas and non-marked road surfaces.

[0122] Specifically, the markings are extracted from the 3D laser point cloud to obtain a cluster of marking point clouds, including:

[0123] Anomalies are segmented by marking lines using a pre-defined semantic segmentation model to generate pixel-level masks.

[0124] Using a preset transformation matrix, the three-dimensional laser point cloud is projected onto the video frame image;

[0125] The marker point cloud clusters are extracted from the 3D laser point cloud using a pixel-level mask.

[0126] In this embodiment, a pre-trained small-scale semantic segmentation model can be used to segment the high-definition images acquired synchronously to generate a pixel-level mask. Then, a pre-calibrated transformation matrix is ​​used to project the three-dimensional laser point cloud onto the video frame image. Finally, the pixel-level mask is used to accurately extract the three-dimensional mark point cloud clusters belonging to the mark from the three-dimensional laser point cloud to eliminate interference from the road surface and other objects.

[0127] Specifically, the preset semantic segmentation model can be a deep learning network trained to identify road marking areas. It can be implemented using a convolutional neural network based on the U-Net architecture. This model achieves pixel-level marking area segmentation by analyzing image texture and color features.

[0128] Specifically, the pre-calibrated transformation matrix can be used to jointly calibrate the camera and LiDAR before system initialization to obtain the transformation matrix of rotation and translation between the two. This transformation matrix can project the 3D point cloud of each frame onto the video frame image, so that each point cloud can find the corresponding pixel region on the image.

[0129] The preset transformation matrix is ​​the transformation parameter for spatially aligning point cloud data in the three-dimensional coordinate system with the two-dimensional image coordinate system. Specifically, it can be calculated by calibrating the camera's intrinsic and extrinsic parameters in combination with the LiDAR pose data.

[0130] A pixel-level mask refers to a binary image output by a semantic segmentation model, where white pixels represent marked areas and black pixels represent unmarked areas.

[0131] Specifically, after detecting a marked area with abnormal visual health scores, the video frame image is first processed in real time using a semantic segmentation model deployed on an edge computing device to generate a pixel-level mask covering the marked area outline. Then, the 3D laser point cloud data is mapped to the corresponding video frame coordinate system using a pre-calibrated transformation matrix, forming a pixel-level spatial correspondence between the point cloud and the image. Finally, based on the coordinate range of the marked area in the pixel-level mask, point cloud clusters belonging to the marked area are selected from the projected 3D point cloud data, providing accurate 3D spatial data for subsequent baseline plane fitting.

[0132] Since the entire road section may not be a flat surface and may contain sloping or potholed areas, it is necessary to dynamically fit the road reference surface within the local area of ​​each marking cross section.

[0133] For example, along the direction of the road marking, a cross-section is taken at certain intervals, and the road surface point cloud is extracted in the adjacent areas on both sides of the marking at the cross-section. Then, the RANSAC algorithm is used to independently fit a local road surface reference plane for each cross-section, which can effectively eliminate the influence of road longitudinal slope, cross slope and local unevenness on the calculation of road marking thickness, so as to obtain a true and accurate thickness value.

[0134] In this embodiment, dynamically fitting the road surface reference plane refers to constructing a reference plane in a local area around the cross-section of the road marking. Specifically, the moving window method can be used to select the road point cloud within a 50 cm range on both sides of the cross-section, and the plane equation can be fitted by the least squares method to establish a dynamically changing reference system.

[0135] Vertical distance is the normal distance between a point on the surface of the index line and the reference plane. Specifically, it can be calculated by spatial vector projection to determine the algebraic distance from each point cloud coordinate to the fitting plane. Its function is to transform the three-dimensional spatial thickness into a numerical sequence that can be quantified and analyzed.

[0136] In this embodiment, for each cross section, the vertical distance from all points in the grading point cloud cluster to their corresponding local reference plane is calculated; the median of these distances is taken as the representative thickness of the cross section; by connecting the representative thicknesses of all cross sections, a thickness variation profile of the grading segment can be generated.

[0137] Specifically, a thickness variation profile is a visual chart that continuously displays the thickness values ​​of each cross section along the direction of the scribed line. It can be drawn in the form of a line graph to show the correspondence between the cross section position and the corresponding average thickness. Its purpose is to provide an intuitive expression of the thickness distribution characteristics for quality inspection.

[0138] In this embodiment, upon detecting a marking area with an abnormal visual health score, high-density 3D point cloud data of that area is acquired using an onboard LiDAR. A trained semantic segmentation model is used to identify the marking areas within the point cloud, forming independent point cloud clusters. For each cross-sectional location, a local reference surface is constructed by taking the road surface point cloud within a 50cm range along its lateral extension direction. The vertical distance from each point in the marking point cloud cluster to the corresponding reference surface is calculated, the average thickness of each cross-section is statistically analyzed, and a continuous variation curve along the longitudinal direction of the road is generated. This process replaces manual measurement with automated data processing, achieving continuous detection of marking thickness across the entire cross-section.

[0139] In this embodiment of the application, S105 may include:

[0140] Based on the thickness variation profile, the quality indicators corresponding to the markings are determined. The quality indicators include average thickness, minimum thickness, thickness uniformity index, and cross-sectional uniformity index.

[0141] The quality indicators are compared with the preset indicators to obtain the comparison results for each quality indicator.

[0142] The evaluation results for abnormal areas are determined by combining the weights of the comparison results for each quality indicator.

[0143] In this embodiment, the average thickness is the average thickness of all measurement points within the abnormal area of ​​the marking, which can be calculated using the arithmetic mean or weighted average method; the minimum thickness is the minimum thickness of all measurement points within the abnormal area of ​​the marking, which can be obtained by traversing point cloud data or statistical distribution histograms, and is used to identify the area with the most severe marking wear; the thickness uniformity index is the degree of dispersion of the marking thickness in the longitudinal distribution, which can be achieved by standard deviation calculation or coefficient of variation analysis, and is used to evaluate the stability of marking quality; the cross-sectional uniformity index is the symmetry index of the thickness of each measurement point within a single cross-section, which can be achieved by comparing the thickness difference between the left and right halves or by waveform similarity analysis, and is used to detect the uniformity of the marking coating process.

[0144] Specifically, after generating the thickness variation profile, the data extraction module obtains the set of thickness values ​​for each measurement point. For the average thickness index, the thickness values ​​of all measurement points in the profile are summed and divided by the total number of points, and the result is compared with the preset standard value. For the minimum thickness index, a sorting algorithm is used to select the minimum thickness value in the profile, and it is compared with the minimum thickness threshold required by the specification. The thickness uniformity index is determined by calculating the standard deviation of the thickness data in the profile to determine whether it exceeds the allowable fluctuation range. The cross-sectional uniformity index performs a symmetry analysis of the thickness distribution in the left and right halves of each cross-section, for example, using the Pearson correlation coefficient to assess the consistency of the thickness change trend on both sides. After completing the individual evaluation of the four indicators, a weighted calculation is performed according to preset weight coefficients, for example, the average thickness weight is set to 40%, the minimum thickness to 30%, the thickness uniformity to 20%, and the cross-sectional uniformity to 10%, and finally a comprehensive evaluation score is obtained.

[0145] By establishing a multi-dimensional quality index system, the system can simultaneously cover four dimensions of quality characteristics: mean thickness, limit value, longitudinal stability, and lateral uniformity, thus solving the problem of the one-sidedness of single-index evaluation. Furthermore, by employing a weighted allocation mechanism to comprehensively quantify each index, the evaluation standards can be flexibly adjusted for different road grades or marking types.

[0146] The implementation principle of this application embodiment is as follows: The road marking thickness detection method described in this application achieves full road marking coverage through multi-channel video synchronous acquisition, and extracts multi-dimensional visual features of the marking area using image processing algorithms; when the feature value of a certain area deviates from the normal range, a three-dimensional laser scanning device is triggered to perform high-precision point cloud acquisition of that area; in the point cloud data processing stage, a road surface reference surface is dynamically constructed to eliminate road undulation interference, and the vertical height of each point of the marking is accurately calculated to form a thickness distribution map; this map can automatically identify areas of abrupt thickness change, and output detection conclusions in combination with a preset quality threshold, that is, the thickness of the marking is detected by visual images and thickness change profile maps, which can quickly realize the measurement of the thickness data of the marking, thereby improving the efficiency of marking thickness detection.

[0147] This application also provides a road marking thickness detection device for performing the road marking thickness detection method described above. (Refer to...) Figure 2 As shown, the road marking thickness detection device includes:

[0148] The acquisition module 201 is used to acquire road videos from multiple video sources, the road videos including multiple consecutive video frame images;

[0149] Analysis module 202 is used to perform two-dimensional image analysis on road videos to obtain visual health scores for the marking areas in the video frame images;

[0150] The determination module 203 is used to determine the target marking area as an abnormal area, and the target marking area includes marking areas whose visual health scores are lower than a preset threshold.

[0151] The generation module 204 is used to perform reference plane fitting and thickness calculation on the abnormal area to generate a thickness variation profile.

[0152] The detection module 205 is used to perform quality inspection on abnormal areas based on the thickness change profile and obtain the detection results.

[0153] This application also provides an electronic device. (See reference...) Figure 3 As shown, the electronic device includes a processor 301 and a memory 302 storing computer program instructions. When the processor 301 executes the computer program instructions, it implements the road marking thickness detection method described above.

[0154] Specifically, the processor 301 described above may include a central processing unit (CPU) or an application-specific integrated circuit (ASIC) that can implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application.

[0155] The memory 302 may be a ROM (Read Only Memory) or other type of static storage device that can store static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device that can store information and instructions, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, which is used to store application code that executes the scheme of this application and is controlled to execute by the processor 301.

[0156] In this embodiment, the electronic device may further include a communication interface 303 and a bus 304. The processor 301, memory 302, and communication interface 303 are connected via the bus 304 and communicate with each other.

[0157] Specifically, the communication interface 303 is mainly used to realize communication between various modules, devices, units, and / or equipment in the embodiments of this application. The bus 304 includes hardware, software, or both, coupling the components of the electronic device together. The bus 304 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The symbol is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0158] This application also provides a computer-readable storage medium storing a computer program that, when run on a computer, enables the computer to execute the corresponding content in the aforementioned road marking thickness detection method embodiments.

[0159] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0160] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for detecting the thickness of road markings, characterized in that, include: Acquire road videos from multiple video sources, the road videos including multiple consecutive video frame images, the multiple video sources including ordinary vehicles, road surface monitoring cameras and inspection vehicles; Two-dimensional image analysis was performed on the road video to obtain the visual health score of the marking area in the video frame image; The target marking area is identified as an abnormal area, and the target marking area includes the marking area where the visual health score is lower than a preset threshold. The abnormal region is fitted to a reference plane and its thickness is calculated to generate a thickness variation profile. The abnormal area is subjected to quality inspection based on the thickness variation profile, and the inspection results are obtained. The step of fitting a reference plane and calculating the thickness of the abnormal region to generate a thickness variation profile includes: Obtain the three-dimensional laser point cloud of the abnormal region; The three-dimensional laser point cloud is subjected to line extraction to obtain line point cloud clusters; Within a local area of ​​each marking cross section in the abnormal region, a road reference surface is dynamically fitted; For each cross section of the road marking, calculate the vertical distance from all points in the marking point cloud cluster to the road reference surface corresponding to the cross section; Generate a profile of the thickness variation of the markings in the abnormal region based on the vertical distance.

2. The method according to claim 1, characterized in that, The step of performing two-dimensional image analysis on the road video to obtain a visual health score for the marked area in the video frame image includes: The video frame images are preprocessed to obtain the image to be analyzed; Local feature extraction is performed on the image to be analyzed to obtain multidimensional features, which include contrast features, edge sharpness features, surface texture features, and continuity features. Based on the preset analysis strategy, scores are calculated for the contrast feature, edge sharpness feature, surface texture feature and continuity feature respectively, to obtain the first score corresponding to the contrast feature, the second score corresponding to the edge sharpness feature, the third score corresponding to the surface texture feature and the fourth score corresponding to the continuity feature. The first score, second score, third score, and fourth score are summed according to the preset weighting to obtain the visual health score.

3. The method according to claim 1 or 2, characterized in that, The step of performing two-dimensional image analysis on the road video to obtain a visual health score for the marked area in the video frame image includes: The multiple video frame images are input into a preset scoring model; In the preset scoring model, for each video frame image, a global feature extraction network extracts global features; The spatial attention network analyzes the global features to obtain edge regions, lane marking center regions, and lane marking-road surface intersection regions; The regression classification network performs a global score on the edge region, the center region of the marking, and the intersection region of the marking and the road surface to obtain a visual health score.

4. The method according to claim 1, characterized in that, After identifying the target marking area as an abnormal area, the method further includes: Obtain the first location information of the abnormal region; In the preset cloud map, abnormal locations are marked based on the first location information; A detection route is generated based on the abnormal location; Based on the detection route and the abnormal area, a detection task list is generated; Acquire the second location information and working status of multiple specialized inspection vehicles; Based on the second location information and working status, a target professional inspection vehicle is determined, and the inspection task list is sent to the target professional inspection vehicle so that the target professional inspection vehicle can perform line thickness inspection according to the inspection task list.

5. The method according to claim 1, characterized in that, The step of extracting markings from the three-dimensional laser point cloud to obtain a marking point cloud cluster includes: The abnormal region is segmented by marking lines using a preset semantic segmentation model to generate a pixel-level mask; The three-dimensional laser point cloud is projected onto the video frame image using a preset transformation matrix; The datum point cloud cluster is extracted from the three-dimensional laser point cloud using the pixel-level mask.

6. The method according to claim 1, characterized in that, The quality inspection of the abnormal area based on the thickness variation profile, and the resulting inspection results, include: Based on the thickness variation profile, the quality indicators corresponding to the markings are determined, including average thickness, minimum thickness, thickness uniformity index, and cross-sectional uniformity index. The quality indicators are compared with preset indicators to obtain the comparison results for each quality indicator; The evaluation result of the abnormal area is determined by combining the weights of the comparison results of each of the aforementioned quality indicators.

7. A road marking thickness detection device, used to perform the road marking thickness detection method as described in any one of claims 1-6, characterized in that, The device includes: The acquisition module is used to acquire road videos from multiple video sources, including multiple consecutive video frame images. The multiple video sources include ordinary vehicles, road surface monitoring cameras, and inspection vehicles. The analysis module is used to perform two-dimensional image analysis on the road video to obtain a visual health score for the marking area in the video frame image; The determination module is used to determine the target marking area as an abnormal area, the target marking area including the marking area where the visual health score is lower than a preset threshold; The generation module is used to perform reference plane fitting and thickness calculation on the abnormal region to generate a thickness variation profile. The detection module is used to perform quality detection on the abnormal area based on the thickness change profile and obtain the detection result; The generation module is also specifically used for: Obtain the three-dimensional laser point cloud of the abnormal region; The three-dimensional laser point cloud is subjected to line extraction to obtain line point cloud clusters; Within a local area of ​​each marking cross section in the abnormal region, a road reference surface is dynamically fitted; For each cross section of the road marking, calculate the vertical distance from all points in the marking point cloud cluster to the road reference surface corresponding to the cross section; Generate a profile of the thickness variation of the markings in the abnormal region based on the vertical distance.

8. An electronic device, characterized in that, The device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the road marking thickness detection method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions that, when executed by a processor, implement the road marking thickness detection method as described in any one of claims 1-6.

Images