A machine vision-based road surface construction quality evaluation method and system

Abstract

This invention relates to the field of image quality assessment technology, specifically to a machine vision-based method and system for assessing road construction quality. The method includes the following steps: acquiring image frames and coordinate information to construct a texture feature map; analyzing directional differences to identify indentation changes; outputting abnormal area identifiers and assessing construction quality. In this invention, by synchronously binding image frames with spatial coordinates and temporal information, precise correlation between image data and the construction path is achieved. A unified texture map group is constructed by extracting grayscale features and frequency domain information from image regions. Regional consistency difference indices are obtained by calculating the differences in the main direction angles of the texture. Indentation sequence trends are formed by extracting closed contour shapes and primitive arrangement features, improving the ability to continuously identify the compaction state of the path. Through spatial information mapping, image data is deeply integrated with the work process, accurately outputting abnormal area identifiers, significantly improving the accuracy and timeliness of construction quality assessment.

Description

Technical Field

[0001] This invention relates to the field of image quality assessment technology, and in particular to a method and system for assessing road construction quality based on machine vision. Background Technology

[0002] Image quality assessment technology involves the processing, analysis, and judgment of image data. It is used to identify structural features in images, detect target content, or assess the quality of objects reflected in the image itself. This includes multiple steps such as image acquisition, image preprocessing, feature extraction, target recognition, and quality judgment. Typically, it involves modeling and judging key areas in the image to achieve the assessment and analysis of the physical object or process state. Traditional road construction quality assessment methods rely on manual inspection, measuring tools, or simple photographic equipment to observe the surface of the construction area and compare it with rulers. The assessment content mainly includes road surface smoothness, crack condition, compaction quality, and material laying consistency. Rulers or thickness gauges are typically used to detect surface height differences, and the extent of cracks and damage is judged visually, or the compaction effect is judged based on the vibration frequency and speed of the compaction equipment. However, these methods are limited by human experience and environmental conditions, resulting in high uncertainty and subjectivity.

[0003] Existing road construction quality assessment technologies rely on manual inspections and simple measuring tools to complete road quality assessments. Limited by the operator's experience level and the influence of the site environment, they cannot achieve stable identification of road conditions. Especially in areas with complex textures and small indentations, there are problems of inaccurate judgment and omissions. The image acquisition process lacks a structured binding mechanism, resulting in low matching degree between construction paths and image data, making it difficult to trace and analyze. Image analysis is mainly based on visual inspection, lacking effective texture direction recognition and regional contrast processing methods, making it impossible to accurately assess regional consistency and indentation distribution characteristics. In continuous operation scenarios, it is impossible to track changes in compaction effects, resulting in data fragmentation, assessment distortion, and delayed anomaly identification, affecting the accurate judgment of overall construction quality and risk control. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a machine vision-based method for assessing road construction quality.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for assessing road construction quality based on machine vision, comprising the following steps: S1: Acquire image frame data recorded by the industrial camera on the road roller, collect the construction path coordinates and time identifiers corresponding to each frame in the image sequence, and synchronously bind the image frames with the corresponding coordinate information to generate an image coordinate association matrix. S2: Read the image content in the image coordinate correlation matrix, divide each frame image into equal area structural regions, perform grayscale channel extraction and frequency domain filtering on each structural region image, obtain the region image boundary comparison parameters, perform feature mapping unification, and generate standard region texture feature map group; S3: Extract the image pixels of the structural region in the standard region texture feature map group, calculate the frequency of the main direction angle distribution and determine the main texture direction angle, calculate the difference of the main texture direction angle between adjacent regions, count the proportion of structural regions that exceed the set jump threshold, and generate the main texture direction difference statistical results. S4: Based on the statistical results of the main texture direction difference, locate the end area of ​​the work path and obtain the closed contour curve, calculate the boundary closure, filter and retain the indentation primitives, perform monotonicity judgment on the sequence of indentation primitives arranged along the work path direction, and generate a record of the trend of path indentation quantity change. S5: Read the image region and location information from the path indentation quantity change trend record and the main texture direction difference statistics, output the abnormal area image identifier in the road construction process, and obtain the road construction quality assessment result.

[0006] As a further aspect of the present invention, the main texture direction angle specifically refers to the main direction angle with the highest distribution frequency; The specific steps of filtering and retaining indentation primitives are as follows: filtering contour regions whose boundary closure meets the preset closure judgment conditions; calculating the ratio of the contour area to the minimum circumscribed circle area of ​​the filtered contour regions to obtain the roundness of the target shape; and retaining the contour regions whose roundness of the target shape falls within the indentation morphology range as indentation primitives.

[0007] As a further embodiment of the present invention, the image coordinate association matrix includes an image frame data index, construction path spatial coordinates, and image acquisition time identifier; the standard area texture feature map group includes a grayscale enhanced image, a texture contrast parameter map, and a feature mapping map; the main texture direction difference statistical results include the main texture direction angle distribution, direction angle difference statistics, and structural area difference ratio; the path indentation quantity change trend record includes the indentation primitive quantity sequence, indentation primitive arrangement direction, and indentation quantity change amplitude; and the road construction quality assessment results include abnormal area image identifiers, image spatial location mapping results, and structural feature change information.

[0008] As a further aspect of the present invention, the step of obtaining the image coordinate correlation matrix specifically includes: S111: Acquire image frame data recorded by the on-board industrial camera of the road roller, collect the time stamp and construction path coordinate information corresponding to the image frame data, match the image number in the image frame with the corresponding time stamp, bind each frame of image data with the construction path coordinates pointed to by the corresponding time stamp, and generate an image frame coordinate binding information set. S112: Based on the image frame coordinate binding information set, based on each image frame and its corresponding construction path coordinates and time identifier, the image frames are reordered in ascending order of time identifier to construct a structured image sequence set of continuous image frames and bound coordinates, and the image frame sequence is matched one-to-one with the construction path coordinate sequence to generate a continuous structured image sequence set. S113: Based on the index number of each frame image in the continuous structure image sequence set and the bound construction path coordinates, extract the set of binary pairs consisting of the image frame index and the coordinate position value, aggregate and store all image frame indices and corresponding coordinate positions, and establish an image coordinate association matrix.

[0009] As a further aspect of the present invention, the steps for obtaining the standard region texture feature map group are as follows: S211: Read the image content in the image coordinate association matrix, perform region division processing on each frame of image, divide the image frame into structural region blocks using the equal area division method, construct a block index mapping for the boundary range of each block, record the mapping relationship between the image number and the division region number, and generate an image region division mapping set. S212: Based on each structural region block in the image region division mapping set, extract grayscale channel values ​​from the image content as single-channel image data frames, and use a regional frequency filter to filter the grayscale channel image to obtain the pixel frequency distribution matrix of the image before filtering and the frequency attenuation matrix of the image after filtering. Then, perform grayscale transformation processing on the two types of matrices to generate a regional grayscale enhancement map set. S213: Based on the image content of each structural region in the gray-scale enhancement map set, extract the gray-scale change amplitude at the image edge position as the region boundary parameter set, calculate the amplitude ratio between the boundary parameter of each image region and the boundary parameter of the adjacent region, and normalize the result value to a preset comparison reference interval to complete the texture feature mapping adjustment between regions and establish a standard region texture feature map set.

[0010] As a further aspect of the present invention, the step of obtaining the statistical results of the main texture direction difference is specifically as follows: S311: Extract the image pixels of each structural region in the standard region texture feature map group, calculate the gray-level difference values ​​in the horizontal and vertical directions respectively, construct the arctangent function mapping of the difference values ​​in the two directions, obtain the gradient direction angle distribution map of each pixel, and generate a set of pixel gradient angle matrices. S312: Based on the pixel gradient direction angle of each structural region in the pixel gradient angle matrix set, the angle value is divided into intervals and the pixel frequency in each angle interval is counted. The center angle value corresponding to the angle interval with the highest frequency is selected as the main texture direction angle. The main texture direction angle index corresponding to each structural region is recorded to obtain the main texture direction angle sequence. S313: Based on the main texture direction angle values ​​of adjacent structural regions in the main texture direction angle sequence, calculate the difference sequence of angle values ​​between two adjacent regions, compare the difference of each angle in the difference sequence with the set jump angle threshold, count the number of structural regions with a difference greater than the set jump angle threshold, and calculate the proportion of the corresponding number to the total number of regions to establish the main texture direction difference statistics.

[0011] As a further aspect of the present invention, the step of obtaining the path indentation quantity change trend record specifically includes: S411: Based on the structural region position identified in the main texture direction difference statistics, locate the corresponding end image of the road roller operation path, read the image data frame of the end region of the path, extract the gray-scale difference boundary of the edge pixels and perform a closure detection operation on the edge pixel sequence, obtain the edge set curve with the edge head-tail connection state closed, and generate a closed contour curve set. S412: Based on the boundary of each contour curve in the closed contour curve set, calculate the area of ​​the closed image region enclosed by the curve and its minimum circumscribed circle area, obtain the area ratio and compare it with the lower limit and upper limit of the roundness reference interval, select the contour curve region whose area ratio falls into the interval as the effective morphological primitive region, and generate the indentation primitive position set. S413: Based on the sequence of elements arranged along the working path of the road roller in the set of indentation element positions, the monotonicity of the index number sequence of adjacent elements in the path direction is judged, the length distribution value of the continuous rising or falling subsequence is extracted, the trend of the number of elements is statistically analyzed, the trend value of the number of path indentations is calculated, and the trend of the number of path indentations is established by combining the trend of the element arrangement structure in the path direction.

[0012] As a further aspect of the present invention, the formula for calculating the trend value of the path indentation quantity is as follows: ; in, This value represents the trend of changes in the number of path indentations. Indicates the first direction of the path Number of indentation primitives at each location This represents the average number of primitives. Indicates the first The path direction distance between adjacent indentation elements The average of the spacing between all adjacent primitives. This represents the total number of indentation elements within the path region.

[0013] As a further aspect of the present invention, the steps for obtaining the road construction quality assessment results are specifically as follows: S511: Read the image region number and location information from the path indentation quantity change trend record and the main texture direction difference statistical result, extract the path indentation quantity change trend value of each image region and calculate the main texture direction angle change rate, use the region number as an index to perform joint index mapping, and generate a spatial structure attribute mapping matrix. S512: Based on the spatial structure attribute mapping matrix, compare the trend value of the path indentation quantity change with the rate of change of the main texture direction angle with the preset fluctuation abnormal threshold and angle mutation threshold, determine whether there are regions that simultaneously meet the corresponding thresholds, and record the region image number and path coordinate information to obtain an abnormal region image identifier list. S513: Based on the image number and path coordinate index information in the abnormal area image identifier list, perform quantitative statistics on the trend value and angle change rate of the remaining area in the spatial structure attribute mapping matrix, calculate the proportion of abnormal areas, establish an evaluation sequence with the path coordinate position as the index, and generate the road construction quality evaluation result.

[0014] A machine vision-based road construction quality assessment system includes: The image positioning module is used to perform S1: acquire image frame data recorded by the industrial camera on the road roller, collect the construction path coordinates and time identifiers corresponding to each frame in the image sequence, synchronously bind the image frames with the corresponding coordinate information, and generate an image coordinate association matrix; The texture construction module is used to execute S2: read the image content in the image coordinate correlation matrix, divide each frame image into equal area structural regions, perform grayscale channel extraction and frequency domain filtering on each structural region image, obtain the region image boundary comparison parameters for feature mapping unification, and generate a standard region texture feature map group. The orientation determination module is used to perform S3: extract the structural region image pixels in the standard region texture feature map group, calculate the frequency of the main orientation angle distribution and determine the main texture orientation angle, calculate the main texture orientation angle difference between adjacent regions, count the proportion of structural regions that exceed the set jump threshold, and generate the main texture orientation difference statistical results. The indentation screening module is used to perform S4: based on the statistical results of the main texture direction difference, locate the end area of ​​the work path and obtain the closed contour curve, calculate the boundary closure, screen and retain indentation elements, perform monotonicity judgment on the sequence of indentation element quantity arranged along the work path direction, and generate a record of the path indentation quantity change trend. The quality assessment module is used to execute S5: read the image area and location information from the path indentation quantity change trend record and the main texture direction difference statistics, output the abnormal area image identifier in the road construction operation, and obtain the road construction quality assessment result.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, image frames are synchronously bound with spatial coordinates and temporal information to achieve precise association between image data and construction paths. A unified texture map group is constructed by extracting grayscale features and frequency domain information of image regions to enhance the expressive ability of structural details. Regional consistency difference index is obtained by calculating the difference in the main direction angle of the texture to identify the location of abnormal changes. The indentation sequence trend is formed by extracting closed contour morphology and primitive arrangement features to improve the continuous recognition ability of the compaction state of the path. Image data is deeply integrated with the operation process through spatial information mapping to accurately output abnormal area identifiers. This solves the problems of unstable human experience judgment, lack of image data positioning, and weak continuity analysis, and significantly improves the accuracy and timeliness of construction quality assessment. Attached Figure Description

[0016] Figure 1 This is a flowchart of the main steps of the present invention; Figure 2 This is a flowchart of the image coordinate correlation matrix acquisition process of the present invention; Figure 3 This is a flowchart of the standard region texture feature map acquisition process of the present invention; Figure 4 This is a flowchart of the process for obtaining statistical results of main texture direction differences in this invention; Figure 5 This is a flowchart illustrating the process of recording and obtaining the trend of changes in the number of path indentations in this invention. Figure 6 This is a flowchart illustrating the process of obtaining the road construction quality assessment results according to the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0019] Please see Figure 1 A machine vision-based method for assessing road construction quality includes the following steps: S1: Acquire image frame data recorded by the industrial camera on the road roller, collect the construction path coordinates and time identifiers corresponding to each frame in the image sequence, bind the image frame with the corresponding coordinate information synchronously and store it as a single frame image data packet, construct a continuous structure image set based on the image data packet, and generate an image coordinate association matrix. S2: Read the image content in the image coordinate correlation matrix, divide each frame image into equal area structural regions, extract grayscale channels and perform frequency domain filtering on each structural region image, improve the local detail texture of the image through grayscale transformation, obtain the boundary contrast parameters of the regional image, perform feature mapping unification, and generate standard regional texture feature map group. S3: Extract the image pixels of each structural region in the standard region texture feature map group, extract the gradient angle information in the horizontal and vertical directions, calculate the distribution frequency of the main direction angle and determine the main direction angle with the highest distribution frequency as the main texture direction angle, calculate the difference of the main texture direction angle between adjacent regions and compare it with the set jump threshold, count the proportion of structural regions that exceed the jump threshold, and generate the main texture direction difference statistical results. S4: Based on the structural region location identified in the statistical results of the main texture direction difference, locate the end region of the road roller's working path, perform edge extraction on the end region and obtain a closed contour curve, calculate the boundary closure degree based on the connection between the beginning and end of the contour curve, filter the contour regions whose boundary closure degree meets the preset closure judgment conditions, calculate the ratio of the contour area to the minimum circumscribed circle area of ​​the filtered contour regions to obtain the roundness of the target shape, retain the contour regions whose roundness falls into the indentation pattern range as indentation primitives, perform monotonicity judgment on the sequence of indentation primitives arranged along the working path direction, and generate a record of the trend of changes in the number of path indentations; S5: Read the image region and location information from the path indentation quantity change trend record and the main texture direction difference statistics, construct the spatial region structure information mapping, output the abnormal area image identifier in the road construction operation, and obtain the road construction quality assessment result.

[0020] The image coordinate correlation matrix includes image frame data index, construction path spatial coordinates, and image acquisition time identifier. The standard area texture feature map group includes grayscale enhanced image, texture contrast parameter map, and feature mapping map. The main texture direction difference statistics include the main texture direction angle distribution, direction angle difference statistics, and structural area difference ratio. The path indentation quantity change trend record includes the indentation primitive quantity sequence, indentation primitive arrangement direction, and indentation quantity change amplitude. The road construction quality assessment results include abnormal area image identifiers, image spatial location mapping results, and structural feature change information.

[0021] Please see Figure 2 Step S1 is as follows: S111: Acquire image frame data recorded by the on-board industrial camera of the road roller, collect the time stamp and construction path coordinate information corresponding to the image frame data, match the image number in the image frame with the corresponding time stamp, bind each frame of image data with the construction path coordinates pointed to by the corresponding time stamp, and generate an image frame coordinate binding information set. In the dynamic scenario of a road roller compacting road surface, data acquisition and spatiotemporal synchronization are implemented. First, an industrial-grade CMOS global shutter camera is installed on the top front of the road roller's cab. The specific camera model selected is a shock-resistant gigabit Ethernet industrial camera, and its acquisition resolution is set to [resolution value missing]. Pixels, frame rate set to fps, exposure time set to To prevent motion blur, a hard-wired triggering method is used to synchronously record the current construction path coordinates (including longitude, latitude, and elevation) using a vehicle-mounted RTK-GNSS (Real-time Dynamic Carrier Phase Differential Positioning) receiver. The GNSS data output frequency is set to [frequency value missing]. Hz, and convert latitude and longitude coordinates to UTM plane coordinate system using Gauss-Kruger projection. The unit is meters.

[0022] To synchronize the image stream and the positioning stream, IEEE 1588 PTP (Precise Time Protocol) is used for nanosecond-level timing of camera acquisition and GNSS positioning, unified to the Unix timestamp benchmark. The exposure center time timestamp of each frame is obtained. and timestamp sequences in GNSS data streams When performing a matching operation, iterate through... Searching and Positioning time with the smallest absolute difference That is, satisfying condition point. like Less than the preset time synchronization threshold If the time interval is ms, then the coordinate is directly bound; if it is between two time intervals, then a linear interpolation algorithm is used to calculate the precise coordinates corresponding to the image frame time. Finally, the tuple containing the three data items "unique image frame ID - high-precision timestamp - construction path coordinates" is stored as a single record in the time-series database, completing the batch processing of all acquired frames and generating an image frame coordinate binding information set; among which, , These represent the precise interpolated horizontal and vertical coordinates corresponding to the exposure center time of the image frame, respectively; , These represent the horizontal and vertical coordinates corresponding to the previous GNSS positioning time closest to the image frame timestamp, respectively. , These represent the horizontal and vertical coordinates corresponding to the next GNSS positioning time closest to the image frame timestamp, respectively. This is the center moment of the image frame exposure; The previous GNSS positioning time; This is the time for the next GNSS positioning.

[0023] S112: Based on the image frame coordinate binding information set, based on each image frame and its corresponding construction path coordinates and time identifier, the image frames are reordered in ascending order of time identifier to construct a structured image sequence set of continuous image frames and bound coordinates. The image frame sequence is matched one-to-one with the construction path coordinate sequence to generate a continuous structured image sequence set. Based on the image frame coordinate binding information set, data cleaning is first performed, and the Euclidean distance between adjacent frame coordinates is calculated. ,like If the speed is determined to be stationary or at extremely low speed, redundant image frames are removed. Then, using the "high-precision timestamp" corresponding to each image frame as the key sorting factor, a fast sorting algorithm is employed to reorder the cleaned image frame data in ascending order, ensuring that the image sequence is strictly monotonically increasing in the time dimension. .

[0024] After sorting, a doubly linked list structure is created in memory to store the sorted image frame objects. Each node in the list contains a pointer to the previous frame. A pointer to the next frame Current frame image data pointer and the corresponding construction path coordinates Perform continuity checks on the sequence, examining the time difference between adjacent frames. The known frame rate is... fps, theoretical frame interval is ms, set the tolerance range to (Right now ).like If the value exceeds this range, it is marked as a frame loss breakpoint. Finally, the verified ordered image frames are indexed and aligned with their corresponding construction path coordinate sequences to generate a continuous structure image sequence set.

[0025] S113: Based on the index number of each frame image in the continuous structure image sequence set and the bound construction path coordinates, extract the set of tuples consisting of image frame index and coordinate position value, aggregate and store all image frame indices and corresponding coordinate positions, and establish an image coordinate association matrix. Based on the continuous structure image sequence set, read the image frame index number of each node (e.g., ...). ) and bound construction path coordinates A sparse matrix structure is initialized in memory as a storage container. The row indices of the matrix are defined as the time-series sequential numbers of the image frames, and the columns store the corresponding binary data. During execution, the first... Frame image index Use the primary key to extract the corresponding coordinates. As attribute values, construct tuples All extracted tuples are aggregated in time series order and written into a hash table, where the key sorting key is... The coordinates are This process establishes an efficient retrieval table between the image logical space and the physical construction space, and generates an image coordinate association matrix.

[0026] Please see Figure 3 Step S2 is as follows: S211: Read the image content in the image coordinate correlation matrix, perform region division processing on each frame of the image, divide the image frame into structural region blocks using the equal area division method, construct a block index mapping for the boundary range of each block, record the mapping relationship between the image number and the division region number, and generate an image region division mapping set. Read the image content from the image coordinate correlation matrix, for each frame with a resolution of The original image is used to extract the region of interest (ROI), and the central region is cropped out. Pixels are used to remove lens distortion edges. A grid scanning method is used, and the size of the structural region tiles is set to [value missing]. Pixels. From the top left corner of the image. Start with step size Perform sliding window cropping in both horizontal and vertical directions to physically cut the entire image frame into... Each tile consists of non-overlapping rectangular sub-tiles. A unique local region number is assigned to each tile. ( ), and establish a pixel coordinate range index for the map. Corresponding to the original image number. With the division of the area number The mapping relationship is used to generate an image region partitioning mapping set.

[0027] S212: Based on each structural region patch in the image region division mapping set, extract the grayscale channel values ​​of the image content as a single-channel image data frame, and use a regional frequency filter to filter the grayscale channel image to obtain the pixel frequency distribution matrix of the image before filtering and the frequency attenuation matrix of the image after filtering. Then, perform grayscale transformation processing on the two types of matrices to generate a regional grayscale enhancement map set. Based on the image region partitioning mapping set, for each The RGB color structure region patches are subjected to weighted grayscale processing, and the formula is as follows: Obtain a single-channel grayscale image matrix A regional frequency filter is used to process the grayscale channel image. First, the grayscale channel image is processed... Perform a two-dimensional discrete Fourier transform (DFT) to obtain the frequency domain distribution matrix. This represents the pixel frequency distribution matrix of the image before filtering. A second-order Butterworth high-pass filter is constructed, with the following transfer function: in, The distance from a point in the frequency plane to the origin. Set as cutoff frequency , Set as Perform a dot product operation between the spectrum matrix and the filter: This yields the frequency attenuation matrix of the filtered image. Subsequently, the frequency attenuation matrix of the filtered image is obtained. Perform an inverse Fourier transform and take the real part to obtain the spatial domain image. .right Perform contrast-limited adaptive histogram equalization, with the cropping threshold set to [value missing]. Set the grid size to Normalize pixel values ​​to Generate a grayscale enhancement atlas of the region.

[0028] S213: Based on the image content of each structural region in the regional gray-scale enhancement map set, extract the gray-scale change amplitude at the image edge position as the regional boundary parameter set, calculate the amplitude ratio between each image region boundary parameter and the adjacent region boundary parameter, and normalize the result value to the preset comparison benchmark interval to complete the texture feature mapping adjustment between regions and establish a standard regional texture feature map group. Read the images of each structural region in the grayscale enhancement dataset and use the Canny operator (high threshold) low threshold Extract edge locations. Calculate the gradient magnitude of edge pixels. This is used as the region boundary parameter. For the current tile... Identify its four adjacent regions (top, bottom, left, and right). Calculate the average gradient magnitude of the edge pixels in the current region. Average gradient magnitude of adjacent regions Calculate the amplitude ratio. Set the preset comparison benchmark interval as follows: The Min-Max normalization formula was used to... Adjustments will be made: in (Set based on experience values). If If the value exceeds the range, it will be truncated. Utilizing... As a gain coefficient, the grayscale of pixels at the edge of the region is adjusted to smooth the differences between blocks and establish a standard region texture feature map group.

[0029] Please see Figure 4 Step S3 is as follows: S311: Extract the image pixels of each structural region in the standard region texture feature map group, calculate the gray-level difference values ​​in the horizontal and vertical directions respectively, construct the arctangent function mapping of the difference values ​​in the two directions, obtain the gradient direction angle distribution map of each pixel, and generate a set of pixel gradient angle matrices. For each structural region image pixel in the standard region texture feature map set The Sobel operator is used to calculate the gray-level difference values ​​in the horizontal and vertical directions respectively. Horizontal gradient The convolution kernel is calculated as follows: Vertical gradient The convolution kernel is calculated as follows: Calculate the gradient direction angle of each pixel. : Convert the calculation results from radians to degrees and map the range of values ​​to... The interval is used to generate a set of pixel gradient angle matrices.

[0030] S312: Based on the pixel gradient direction angle of each structural region in the pixel gradient angle matrix set, the angle value is divided into intervals and the pixel frequency in each angle interval is counted. The center angle value corresponding to the angle interval with the highest frequency is selected as the main texture direction angle. The main texture direction angle index corresponding to each structural region is recorded to obtain the main texture direction angle sequence. Based on the pixel gradient angle matrix set, histogram statistics are performed on the pixel gradient direction angles within each structural region. Divided into There are 1 angle intervals, with a step size of 1 for each interval. (like The frequency of pixels falling into each interval is counted, and the angle interval with the highest frequency is selected. The center angle value of this interval is taken as the "primary texture direction angle" of the structural region. For example, if the peak interval is... The principal angle is denoted as Traverse all regions, record angle values ​​by region index, and obtain the main texture direction angle sequence.

[0031] S313: Based on the main texture direction angle values ​​of adjacent structural regions in the main texture direction angle sequence, calculate the difference sequence of angle values ​​between two adjacent regions, compare the difference of each angle in the difference sequence with the set jump angle threshold, count the number of structural regions with a difference greater than the set jump angle threshold, and calculate the proportion of the corresponding number to the total number of regions to establish the main texture direction difference statistics. Read the main texture direction angle sequence Calculate the angular difference sequence between adjacent structural regions. Set the jump angle threshold to... Traverse the difference sequence and count... Number of structural regions Calculate the proportion of the corresponding quantity to the total number of regions: Establish statistical results for the main texture direction differences, which include the number and proportion of abrupt change regions, as well as a list of indexes of regions where abrupt changes occur.

[0032] Please see Figure 5 Step S4 is as follows: S411: Based on the structural region location identified in the statistical results of the main texture direction difference, locate the corresponding end image of the road roller operation path, read the image data frame of the end region of the path, extract the gray-scale difference boundary of the edge pixels and perform a closure detection operation on the edge pixel sequence, obtain the edge set curve with the beginning and end connection state of the edge, and generate a closed contour curve set. Based on the abnormal area locations identified in the main texture direction difference statistics, the image at the end of the roller's working path is located. The image data frame for that area is read, firstly using... Median filtering is used for noise reduction, and an adaptive binarization threshold is calculated using Otsu's algorithm to extract the gray-level difference boundaries of edge pixels. Closure detection is then performed on the edge pixel sequence in the binarized image: the first pixel of the edge sequence is determined. With the last pixel Is the Euclidean distance less than ? (i.e., whether it is an 8-neighbor connection). If it is not closed, perform morphological closing operation (dilation followed by erosion, kernel size). Connect the edges. Obtain the set of closed edge curves and generate a set of closed contour curves.

[0033] S412: Based on the boundary of each contour curve in the closed contour curve set, calculate the area of ​​the closed image region enclosed by the curve and its minimum circumscribed circle area, obtain the area ratio and compare it with the lower limit and upper limit of the roundness reference interval, select the contour curve region whose area ratio falls into the interval as the effective morphological primitive region, and generate the indentation primitive position set. For each contour curve in the closed contour curve set Calculate the area of ​​the closed image region it encloses. (Number of pixels). Calculate the minimum bounding circle of the contour using the minimum bounding circle algorithm, and obtain the area of ​​the bounding circle. Calculate the area ratio (i.e., roundness): Set the lower limit of the roundness reference range to . The upper limit is .Will and interval By comparing and filtering, the contour curve regions that fall within the range are selected and marked as valid morphological primitive regions, and a set of indentation primitive positions is generated.

[0034] S413: Based on the sequence of elements arranged along the roller's operating path in the set of indentation element locations, the monotonicity of the index number sequence of adjacent elements along the path direction is determined, the length distribution value of continuously rising or falling subsequences is extracted, and the trend of element quantity changes is statistically analyzed using the formula: ; The trend value of the path indentation quantity is obtained through calculation, and combined with the trend of the primitive arrangement structure along the path, a record of the path indentation quantity trend is established; among which... This value represents the trend of changes in the number of path indentations. Indicates the first direction of the path Number of indentation primitives at each location This represents the average number of primitives. Indicates the first The path direction distance between adjacent indentation elements The average of the spacing between all adjacent primitives. This represents the total number of indentation elements within the path region; Based on the set of indentation element locations, the elements are arranged along the operating path of the road roller. The index sequence of adjacent elements along the path is extracted, its monotonicity is determined, and the trend of element quantity changes is statistically analyzed. The trend value of the path indentation quantity change is obtained using a formula. To ensure the transparency of the calculation process, Table 1 lists a set of test data actually collected in the embodiment for the calculation.

[0035] Table 1. Test data on the number and distance distribution of path indentations: ; Calculations are based on the data in Table 1: Calculate the number of primitives: mean Numerator (sum of absolute differences): Denominator terms (variance-related): . The first part is valued at .

[0036] Calculate the distance weight term: mean Ratio and The second part is the denominator. The second part is valued at .

[0037] Calculate the final trend value: .

[0038] The calculated Records were created to document the trend of changes in the number of path indentations. The results indicate that the fluctuations in the number of indentations on the current road section are within a stable range.

[0039] Please see Figure 6 The S5 steps are as follows: S511: Read the image region number and location information from the path indentation quantity change trend record and the main texture direction difference statistics result, extract the path indentation quantity change trend value of each image region and calculate the main texture direction angle change rate, use the region number as an index to perform joint index mapping, and generate a spatial structure attribute mapping matrix. Read the path indentation quantity change trend record and the statistical results of the difference between the main texture direction. Using the image region number as an index, extract the path indentation quantity change trend value for each region. And calculate the rate of change of the main texture direction angle. : Region ID, path coordinates , , Perform a joint mapping to generate a spatial structure attribute mapping matrix; where, This represents the main texture direction angle of the i-th structural region; It represents the main texture direction angle of the preceding structural region adjacent to the i-th structural region.

[0040] S512: Based on the spatial structure attribute mapping matrix, compare the trend value of the path indentation quantity change with the rate of change of the main texture direction angle with the preset fluctuation anomaly threshold and angle mutation threshold, determine whether there are regions that simultaneously meet the corresponding thresholds, and record the region image number and path coordinate information to obtain an abnormal region image identifier list. Based on the spatial structure attribute mapping matrix, set the fluctuation anomaly threshold. With angle change threshold (correspond (Changes). Traverse the matrix and determine if there exists a region that satisfies: If all conditions are met simultaneously, it indicates that there is segregation or unevenness in the compaction quality of the area. Record the image number and path coordinate information of the areas that meet the conditions to obtain a list of abnormal area image identifiers.

[0041] S513: Based on the image number and path coordinate index information in the abnormal area image identification list, perform quantitative statistics on the trend value and angle change rate of the remaining area in the spatial structure attribute mapping matrix, calculate the proportion of abnormal areas, establish an evaluation sequence with the path coordinate position as the index, and generate the road construction quality evaluation result. Count the number of abnormal regions based on the list of abnormal region image identifiers. Combined with the total number of regions in the spatial structure attribute mapping matrix. Calculate the percentage of abnormal regions An evaluation sequence is established using path coordinates as an index, and the pavement construction quality score is calculated. For example, if the abnormal percentage is The score is The final result is a pavement construction quality assessment that includes a score and defect coordinates. This result will directly guide construction personnel to perform recompaction or remedial treatment at specific coordinate locations.

[0042] A machine vision-based road construction quality assessment system includes: The image positioning module is used to perform S1: acquire image frame data recorded by the industrial camera on the road roller, collect the construction path coordinates and time identifiers corresponding to each frame in the image sequence, synchronously bind the image frames with the corresponding coordinate information, and generate an image coordinate association matrix; The texture construction module is used to execute S2: read the image content in the image coordinate correlation matrix, divide each frame image into equal area structural regions, perform grayscale channel extraction and frequency domain filtering on each structural region image, obtain the region image boundary comparison parameters for feature mapping unification, and generate standard region texture feature map group; The orientation determination module is used to execute S3: extract the image pixels of the structural region in the standard region texture feature map group, calculate the frequency of the main orientation angle distribution and determine the main texture orientation angle, calculate the difference of the main texture orientation angle of the adjacent region, count the proportion of structural regions that exceed the set jump threshold, and generate the main texture orientation difference statistical results. The indentation screening module is used to perform S4: based on the statistical results of the difference in the main texture direction, locate the end area of ​​the work path and obtain the closed contour curve, calculate the boundary closure, screen and retain indentation elements, perform monotonicity judgment on the sequence of the number of indentation elements arranged along the work path direction, and generate a record of the trend of the number of path indentations. The quality assessment module is used to execute S5: read the image area and location information from the record of the trend of the number of path indentations and the statistical results of the difference in the main texture direction, output the image identifier of the abnormal area in the road construction process, and obtain the road construction quality assessment result.

[0043] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A machine vision-based method for assessing the quality of road construction, characterized in that, Includes the following steps: S1: Acquire image frame data recorded by the industrial camera on the road roller, collect the construction path coordinates and time identifiers corresponding to each frame in the image sequence, and synchronously bind the image frames with the corresponding coordinate information to generate an image coordinate association matrix. S2: Read the image content in the image coordinate correlation matrix, divide each frame image into equal area structural regions, perform grayscale channel extraction and frequency domain filtering on each structural region image, obtain the region image boundary comparison parameters, perform feature mapping unification, and generate standard region texture feature map group; S3: Extract the image pixels of the structural region in the standard region texture feature map group, calculate the frequency of the main direction angle distribution and determine the main texture direction angle, calculate the difference of the main texture direction angle between adjacent regions, count the proportion of structural regions that exceed the set jump threshold, and generate the main texture direction difference statistical results. S4: Based on the structural region location identified in the statistical results of the main texture direction difference, locate the end region of the operation path and obtain the closed contour curve, calculate the boundary closure, filter and retain the indentation primitives, perform monotonicity judgment on the sequence of indentation primitives arranged along the operation path direction, and generate a record of the trend of path indentation quantity change. S5: Read the image region and location information from the path indentation quantity change trend record and the main texture direction difference statistics, output the abnormal area image identifier in the road construction process, and obtain the road construction quality assessment result.

2. The machine vision-based road construction quality assessment method according to claim 1, characterized in that: The main texture direction angle specifically refers to the main direction angle with the highest distribution frequency; The specific steps of filtering and retaining indentation primitives are as follows: filtering contour regions whose boundary closure meets the preset closure judgment conditions; calculating the ratio of the contour area to the minimum circumscribed circle area of ​​the filtered contour regions to obtain the roundness of the target shape; and retaining the contour regions whose roundness of the target shape falls within the indentation morphology range as indentation primitives.

3. The machine vision-based road construction quality assessment method according to claim 1, characterized in that: The image coordinate association matrix includes image frame data index, construction path spatial coordinates, and image acquisition time identifier. The standard area texture feature map group includes grayscale enhanced image, texture contrast parameter map, and feature mapping map. The main texture direction difference statistical results include main texture direction angle distribution, direction angle difference statistics, and structural area difference ratio. The path indentation quantity change trend record includes indentation primitive quantity sequence, indentation primitive arrangement direction, and indentation quantity change amplitude. The road construction quality assessment results include abnormal area image identifier, image spatial location mapping results, and structural feature change information.

4. The machine vision-based road construction quality assessment method according to claim 1, characterized in that, The specific steps for obtaining the image coordinate correlation matrix are as follows: S111: Acquire image frame data recorded by the on-board industrial camera of the road roller, collect the time stamp and construction path coordinate information corresponding to the image frame data, match the image number in the image frame with the corresponding time stamp, bind each frame of image data with the construction path coordinates pointed to by the corresponding time stamp, and generate an image frame coordinate binding information set. S112: Based on the image frame coordinate binding information set, based on each image frame and its corresponding construction path coordinates and time identifier, the image frames are reordered in ascending order of time identifier to construct a structured image sequence set of continuous image frames and bound coordinates, and the image frame sequence is matched one-to-one with the construction path coordinate sequence to generate a continuous structured image sequence set. S113: Based on the index number of each frame image in the continuous structure image sequence set and the bound construction path coordinates, extract the set of binary pairs consisting of the image frame index and the coordinate position value, aggregate and store all image frame indices and corresponding coordinate positions, and establish an image coordinate association matrix.

5. The machine vision-based road construction quality assessment method according to claim 1, characterized in that, The specific steps for obtaining the standard region texture feature map set are as follows: S211: Read the image content in the image coordinate association matrix, perform region division processing on each frame of image, divide the image frame into structural region blocks using the equal area division method, construct a block index mapping for the boundary range of each block, record the mapping relationship between the image number and the division region number, and generate an image region division mapping set. S212: Based on each structural region block in the image region division mapping set, extract grayscale channel values ​​from the image content as single-channel image data frames, and use a regional frequency filter to filter the grayscale channel image and perform grayscale transformation processing to generate a regional grayscale enhancement map set. S213: Based on the image content of each structural region in the gray-scale enhancement map set, extract the gray-scale change amplitude at the image edge position as the region boundary parameter set, calculate the amplitude ratio between the boundary parameter of each image region and the boundary parameter of the adjacent region, and normalize the result value to a preset comparison reference interval to complete the texture feature mapping adjustment between regions and establish a standard region texture feature map set.

6. The machine vision-based road construction quality assessment method according to claim 1, characterized in that, The specific steps for obtaining the statistical results of the main texture direction differences are as follows: S311: Extract the image pixels of each structural region in the standard region texture feature map group, calculate the gray-level difference values ​​in the horizontal and vertical directions respectively, construct the arctangent function mapping of the difference values ​​in the two directions, obtain the gradient direction angle distribution map of each pixel, and generate a set of pixel gradient angle matrices. S312: Based on the pixel gradient direction angle of each structural region in the pixel gradient angle matrix set, the angle value is divided into intervals and the pixel frequency in each angle interval is counted. The center angle value corresponding to the angle interval with the highest frequency is selected as the main texture direction angle. The main texture direction angle index corresponding to each structural region is recorded to obtain the main texture direction angle sequence. S313: Based on the main texture direction angle values ​​of adjacent structural regions in the main texture direction angle sequence, calculate the difference sequence of angle values ​​between two adjacent regions, compare the difference of each angle in the difference sequence with the set jump angle threshold, count the number of structural regions with a difference greater than the set jump angle threshold, and calculate the proportion of the corresponding number to the total number of regions to establish the main texture direction difference statistics.

7. The machine vision-based road construction quality assessment method according to claim 1, characterized in that, The specific steps for obtaining the path indentation quantity change trend record are as follows: S411: Based on the structural region position identified in the main texture direction difference statistics, locate the corresponding end image of the road roller operation path, read the image data frame of the end region of the path, extract the gray-scale difference boundary of the edge pixels and perform a closure detection operation on the edge pixel sequence, obtain the edge set curve with the edge head-tail connection state closed, and generate a closed contour curve set. S412: Based on the boundary of each contour curve in the closed contour curve set, calculate the area of ​​the closed image region enclosed by the curve and its minimum circumscribed circle area, obtain the area ratio and compare it with the lower limit and upper limit of the roundness reference interval, select the contour curve region whose area ratio falls into the interval as the effective morphological primitive region, and generate the indentation primitive position set. S413: Based on the sequence of elements arranged along the working path of the road roller in the set of indentation element positions, the monotonicity of the index number sequence of adjacent elements in the path direction is judged, the length distribution value of the continuous rising or falling subsequence is extracted, the trend of the number of elements is statistically analyzed, the trend value of the number of path indentations is calculated, and the trend of the number of path indentations is established by combining the trend of the element arrangement structure in the path direction.

8. The machine vision-based road construction quality assessment method according to claim 7, characterized in that, The formula for calculating the trend value of the number of path indentations is: ； in, This value represents the trend of changes in the number of path indentations. Indicates the first direction of the path Number of indentation primitives at each location This represents the average number of primitives. Indicates the first The path direction distance between adjacent indentation elements The average of the spacing between all adjacent primitives. This represents the total number of indentation elements within the path region.

9. The machine vision-based road construction quality assessment method according to claim 1, characterized in that, The specific steps for obtaining the road construction quality assessment results are as follows: S511: Read the image region number and location information from the path indentation quantity change trend record and the main texture direction difference statistical result, extract the path indentation quantity change trend value of each image region and calculate the main texture direction angle change rate, use the region number as an index to perform joint index mapping, and generate a spatial structure attribute mapping matrix. S512: Based on the spatial structure attribute mapping matrix, compare the trend value of the path indentation quantity change with the rate of change of the main texture direction angle with the preset fluctuation abnormal threshold and angle mutation threshold, determine whether there are regions that simultaneously meet the corresponding thresholds, and record the region image number and path coordinate information to obtain an abnormal region image identifier list. S513: Based on the image number and path coordinate index information in the abnormal area image identifier list, perform quantitative statistics on the trend value and angle change rate of the remaining area in the spatial structure attribute mapping matrix, calculate the proportion of abnormal areas, establish an evaluation sequence with the path coordinate position as the index, and generate the road construction quality evaluation result.

10. A machine vision-based road construction quality assessment system, characterized in that, The system is used to implement the machine vision-based road construction quality assessment method according to any one of claims 1-9, comprising: The image positioning module is used to perform S1: acquire image frame data recorded by the industrial camera on the road roller, collect the construction path coordinates and time identifiers corresponding to each frame in the image sequence, synchronously bind the image frames with the corresponding coordinate information, and generate an image coordinate association matrix; The texture construction module is used to execute S2: read the image content in the image coordinate correlation matrix, divide each frame image into equal area structural regions, perform grayscale channel extraction and frequency domain filtering on each structural region image, obtain the region image boundary comparison parameters for feature mapping unification, and generate a standard region texture feature map group. The orientation determination module is used to perform S3: extract the structural region image pixels in the standard region texture feature map group, calculate the frequency of the main orientation angle distribution and determine the main texture orientation angle, calculate the main texture orientation angle difference between adjacent regions, count the proportion of structural regions that exceed the set jump threshold, and generate the main texture orientation difference statistical results. The indentation screening module is used to perform S4: based on the statistical results of the main texture direction difference, locate the end area of ​​the work path and obtain the closed contour curve, calculate the boundary closure, screen and retain indentation elements, perform monotonicity judgment on the sequence of indentation element quantity arranged along the work path direction, and generate a record of the path indentation quantity change trend. The quality assessment module is used to execute S5: read the image area and location information from the path indentation quantity change trend record and the main texture direction difference statistics, output the abnormal area image identifier in the road construction operation, and obtain the road construction quality assessment result.

Images