CN-121982596-A - Pavement crack detection method and system based on unmanned aerial vehicle image

CN121982596ACN 121982596 ACN121982596 ACN 121982596ACN-121982596-A

Abstract

The invention relates to the technical field of image processing, in particular to a pavement crack detection method and system based on unmanned aerial vehicle images, wherein the method comprises the steps of acquiring visible light images and thermal infrared images synchronously acquired by unmanned aerial vehicles, positioning original pavement thermal anomaly areas and seed points in the thermal infrared images, mapping the original pavement thermal anomaly areas and seed points to the visible light images to obtain mapped pavement thermal anomaly areas, and executing area growth to acquire first crack areas; and extracting a pavement crack skeleton based on the second crack region and mapping the pavement crack skeleton back to the original pavement thermal anomaly region, and estimating the damage state of the target pavement according to the temperature distribution of the pavement crack skeleton to finish crack detection. The method improves the accuracy of the pavement crack detection result.

Inventors

LI XIAOCHEN
ZHANG SHUNWEN
YANG BO
YAO XIONG
ZHENG CHENGGONG
ZHOU RAN
ZENG XIANLING

Assignees

贵州交建信息科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260409

Claims (10)

1. The pavement crack detection method based on the unmanned aerial vehicle image is characterized by comprising the following steps of: Acquiring a visible light image and a thermal infrared image of a target pavement synchronously acquired by an unmanned aerial vehicle; Positioning an original pavement thermal anomaly region in the thermal infrared image through histogram analysis, extracting seed points of the original pavement thermal anomaly region, and mapping the original pavement thermal anomaly region and the seed points to the visible light image to obtain a mapped pavement thermal anomaly region in the visible light image; performing heat source superposition analysis on pixel points in the first crack region based on a Gaussian heat diffusion rule, and generating a crack heat influence diagram corresponding to the first crack region in the map pavement thermal anomaly region; obtaining an image structure similarity index of the crack heat influence graph and the original pavement heat abnormal region through image structure similarity analysis, and correcting the first crack region according to the image structure similarity index to obtain a second crack region; And performing morphological operation on the second crack region to extract a pavement crack skeleton, mapping the pavement crack skeleton back to the original pavement thermal anomaly region, and evaluating the damage state of the target pavement according to the temperature distribution of the pavement crack skeleton in the original pavement thermal anomaly region so as to finish pavement crack detection.
2. The pavement crack detection method as set forth in claim 1, wherein the method of locating an original pavement thermal anomaly region in the thermal infrared image by histogram analysis and extracting seed points of the original pavement thermal anomaly region is as follows: Constructing a temperature distribution histogram of the thermal infrared image by taking the temperature value as an abscissa and the number of pixel points corresponding to the temperature value as an ordinate, and extracting a temperature value corresponding to a maximum peak value as a background temperature after carrying out smoothing treatment on the temperature distribution histogram; The method comprises the steps of taking the background temperature as a boundary, forming a first temperature sequence by taking all temperature values with temperature values lower than the background temperature in the temperature distribution histogram, and forming a second temperature sequence by taking all temperature values with temperature values higher than the background temperature in the temperature distribution histogram, wherein the temperature values are independent variables, the number of pixels corresponding to the temperature values is taken as dependent variables, respectively calculating discrete second-order differences for the first temperature sequence and the second temperature sequence, setting a temperature value corresponding to an inflection point with zero first second-order difference in the first temperature sequence as a first temperature threshold, setting a temperature value corresponding to an inflection point with zero first second-order difference in the second temperature sequence as a second temperature threshold, and setting the first temperature threshold to be smaller than the second temperature threshold; The method comprises the steps of taking a section of a first temperature threshold value and a section of a second temperature threshold value as a normal temperature section, judging temperature values outside the normal temperature section as abnormal temperature values, obtaining all connected areas formed by pixel points corresponding to all abnormal temperature values in the thermal infrared image, respectively carrying out edge smoothing treatment on all the connected areas to obtain a plurality of original road surface thermal abnormal areas, and selecting the pixel points with the largest forward difference value between the temperature values and the background temperature and meeting the first quantity ratio as seed points in each original road surface thermal abnormal area according to a preset first quantity ratio.
3. The pavement crack detection method as set forth in claim 1, wherein the method of performing region growing based on the seed points in the visible light image to obtain the first crack region in the mapped pavement thermal anomaly region comprises: Setting a local neighborhood by taking each seed point in the visible light image as a center according to a preset size, extracting gradient characteristics of each pixel point in the local neighborhood, carrying out rationality verification on the seed points based on the gradient characteristics of each pixel point, carrying out region growing on the basis of the rational seed points if the rationality verification is passed, generating a first crack region in the map pavement thermal anomaly region if the rationality verification is not passed, and not carrying out region growing if the rationality verification is not passed.
4. The pavement crack detection method as set forth in claim 3, wherein extracting gradient features of each pixel in the local neighborhood, verifying rationality of the seed points based on the gradient features of each pixel, comprises: The gradient characteristics of each pixel point in the local neighborhood comprise gradient values and gradient directions, pixel points with the maximum gradient value and the second number of the pixel points meeting the second number of the pixel points are selected in the local neighborhood according to a preset second number of the pixel points to serve as gradient significant pixel points, the direction of each gradient significant pixel point pointing to the seed point is obtained and is recorded as a pointing direction, the included angle between the gradient direction of each gradient significant pixel point and the pointing direction is calculated, if all the included angles are larger than a preset angle threshold value, the rationality verification of the seed point is judged to be passed, and if at least one included angle is not larger than the preset angle threshold value, the rationality verification of the seed point is judged to be failed.
5. The pavement crack detection method as set forth in claim 3, wherein performing region growing based on the rational seed points, generating the first crack region within the mapped pavement thermal anomaly region includes: Performing region growing by taking the reasonable seed points as starting points, and in each round of region growing, acquiring a current grown pixel point and all pixel points to be judged, and normalizing gradient values of the pixel points to be judged for each pixel point to be judged to obtain dynamic decision weights of the pixel points to be judged; Taking the direction of the current grown pixel points to the pixel point to be judged as a space growth direction, and determining the growth direction consistency of the pixel point to be judged based on the space growth direction and the gradient direction of the pixel point to be judged; the method comprises the steps of obtaining the nearest gray value of a preset number of grown pixel points for extracting local gray statistical characteristics from the pixel points to be judged, and extracting the local gray statistical characteristics; Weighting the consistency of the growing direction by using the dynamic decision weight, weighting the local gray scale similarity by using a difference value between the dynamic decision weight and the dynamic decision weight, and then performing summation operation to obtain the growing decision probability of the pixel point to be judged; Judging whether the pixel points to be judged exist, which have the maximum growth decision probability and are larger than all growth decision probabilities in the continuously preset historical rounds, updating the pixel points to the grown pixel points to continue to execute the next round if the pixel points to be judged exist, ending the region growth if the pixel points to be judged do not exist, and executing morphological processing on the connected regions formed by all the grown pixel points to obtain the first crack region.
6. The pavement crack detection method as set forth in claim 1, wherein performing heat source superposition analysis on the pixel points in the first crack region based on a gaussian heat diffusion rule, generating a crack heat influence map corresponding to the first crack region in the mapped pavement heat anomaly region, comprises: In the mapping pavement thermal anomaly area, taking any one pixel point as a target pixel point, calculating Euclidean distances between the target pixel point and all crack thermal source points, and constructing a Gaussian kernel function decaying along with the Euclidean distances so as to represent a diffusion superposition process of temperature through the Euclidean distances; And calculating independent crack heat influence values generated by each crack heat source point on the target pixel point based on the Gaussian kernel function, accumulating all the independent crack heat influence values to generate a comprehensive crack heat influence value of the target pixel point, and generating the crack heat influence graph based on the comprehensive crack heat influence values of all the target pixel points.
7. The pavement crack detection method as set forth in claim 5, wherein correcting the first crack region based on the image structural similarity index to obtain a second crack region includes: The method comprises the steps of counting image structure similarity indexes based on a plurality of groups of visible light images and thermal infrared images in advance, and determining an image structure similarity evaluation threshold based on a three-time standard deviation principle; If the image structure similarity index of the crack thermal influence map and the original pavement thermal anomaly area is larger than the image structure similarity evaluation threshold, judging that the first crack area is not required to be corrected, and directly taking the first crack area as the second crack area; If the image structure similarity index of the crack thermal influence map and the original pavement thermal anomaly region is not greater than the image structure similarity evaluation threshold, dividing the crack thermal influence map into a plurality of subareas, and taking the corresponding part of each subarea in the original pavement thermal anomaly region as an original subarea of the subarea; and screening out the subareas needing to be re-grown by comparing the heat distribution intensity of each subarea with the heat distribution intensity of the original subarea, and re-executing the area growth on the corresponding parts of the subareas needing to be re-grown in the first crack area until the image structure similarity index is larger than the image structure similarity evaluation threshold value, so as to obtain a second crack area.
8. The pavement crack detection method as set forth in claim 7, wherein the steps of screening out the subregion requiring the re-regional growth, and re-executing the regional growth on the corresponding portion of the subregion requiring the re-regional growth in the first crack region, include: Calculating image structure similarity indexes of each sub-region and each original sub-region, selecting the sub-region with the lowest image structure similarity index as a sub-region to be corrected, determining respective heat distribution intensity based on respective temperature distribution of the sub-region to be corrected and the corresponding original sub-region, judging whether the heat distribution intensity of the sub-region to be corrected is larger than that of the corresponding original sub-region, if so, judging that the sub-region to be corrected excessively grows, reducing the preset number for extracting local gray statistical features according to the preset numerical value, and if not, judging that the sub-region to be corrected is insufficient in growth, and increasing the preset number for extracting the local gray statistical features according to the preset numerical value; And in the first crack region, starting from the correction starting pixel point, re-executing region growth on the corresponding part of the sub-region to be corrected in the first crack region by adopting the increased/decreased preset quantity.
9. The pavement crack detection method as set forth in claim 2, wherein evaluating the damaged condition of the target pavement in the original pavement thermal anomaly region based on the temperature distribution of the pavement crack skeleton to complete pavement crack detection includes: Traversing each pixel point of the pavement crack skeleton, setting a maximum temperature value in a local neighborhood of each pixel point as a local thermal peak value of the pixel point, and taking a difference value between the local thermal peak value and the background temperature as a background temperature difference of the pixel point; Identifying and breaking cross branch nodes on the pavement crack skeleton, decomposing the pavement crack skeleton into a plurality of independent skeleton segments, restricting the minimum pixel point length of the skeleton segments, taking the minimum sum of variances of background temperature differences of pixel points of each analysis segment as a cost function, and dividing all the skeleton segments into a plurality of analysis segments with consistent semantics; Calculating an average value and a range of the background temperature difference of each analysis segment, dividing the range by the average value to obtain a relative fluctuation ratio, taking the sum of one and the relative fluctuation ratio as a product coefficient, and multiplying the average value to obtain a local structural abnormality index of the analysis segment; And carrying out weighted average calculation on the local structural abnormality indexes based on the lengths of all the analysis sections to obtain weighted average indexes, obtaining the ratio of the total number and the total length of all the analysis sections as the segmentation frequency, multiplying the sum of one and the segmentation frequency as the expansion coefficient by the weighted average indexes to obtain the global structural abnormality indexes of the pavement crack skeleton, and evaluating the damage state of the target pavement based on the global structural abnormality indexes.
10. A pavement crack detection system based on unmanned aerial vehicle images, characterized in that the system comprises a memory and a processor, the memory having stored thereon a computer program, the processor executing the computer program to implement the steps of the pavement crack detection method according to any of claims 1-9.

Description

Pavement crack detection method and system based on unmanned aerial vehicle image Technical Field The invention relates to the technical field of image processing. In particular to a pavement crack detection method and system based on unmanned aerial vehicle images. Background The road is used as an important traffic infrastructure, and the structural health condition of the road is related to the safety and stability of public transportation. In the long-term service process of the road, the road surface is easily subjected to stress concentration and uneven stress due to the continuous influence of internal infrastructure defect evolution or external complex environmental loads (such as heavy load, alternating temperature and rain erosion) so as to initiate cracks. If the road cracks are used as early surface morphology manifestations of deep road disasters, the early road disasters such as pits and subsidence are finally caused because moisture permeates into a base layer or even a roadbed along the cracks and accelerates the integral degradation of a road surface structure if the road cracks are not accurately detected at the early stage and corresponding repair operations are timely implemented. In recent years, the use of unmanned aerial vehicle platforms to carry visual sensors for road surface inspection on a large scale has become an important development trend in the field of traffic infrastructure maintenance. The existing pavement crack detection method based on unmanned aerial vehicle images generally mainly depends on a single visible light image (RGB image), and most of detection logics directly extract textures, gray scales and edge geometric features of cracks on a two-dimensional visible light image plane through a traditional image edge detection algorithm or a deep learning model, so that the positioning and segmentation of the cracks are completed. However, the existing detection method has the defects that the real road surface is often covered with pavement marking lines, water stains, greasy dirt, tire friction marks and historical repairing marks due to the fact that the visible light image is extremely sensitive to road background noise, and the complex non-crack textures are similar to real cracks in gray distribution and geometric edges, so that the conventional image edge detection algorithm is extremely easy to generate misjudgment and over-segmentation phenomena. If a deep learning model is used, the deep learning model has high computational power requirements, and a high-quality labeling sample is needed, and if the recognition result deviates from the actual situation, the model is difficult to adjust, and the iteration cost is high. Therefore, a pavement crack detection method capable of breaking through the limitation of a single visible light mode, effectively inhibiting the background noise interference of a complex pavement, identifying the geometric form of a pavement crack with high precision and not needing to rely on huge calculation force and massive marking data is needed. Disclosure of Invention The invention provides a pavement crack detection method and a pavement crack detection system based on unmanned aerial vehicle images, which are used for solving the technical problems that the existing pavement crack detection method is easy to be interfered by complex background noise to cause segmentation misjudgment and is limited by the fact that two-dimensional surface visual characteristics cannot quantitatively evaluate the damage state of the internal structure of a crack. In a first aspect, the present invention provides a pavement crack detection method based on unmanned aerial vehicle images, including: Acquiring a visible light image and a thermal infrared image of a target pavement synchronously acquired by an unmanned aerial vehicle; Positioning an original pavement thermal anomaly region in the thermal infrared image through histogram analysis, extracting seed points of the original pavement thermal anomaly region, and mapping the original pavement thermal anomaly region and the seed points to the visible light image to obtain a mapped pavement thermal anomaly region in the visible light image; performing heat source superposition analysis on pixel points in the first crack region based on a Gaussian heat diffusion rule, and generating a crack heat influence diagram corresponding to the first crack region in the map pavement thermal anomaly region; obtaining an image structure similarity index of the crack heat influence graph and the original pavement heat abnormal region through image structure similarity analysis, and correcting the first crack region according to the image structure similarity index to obtain a second crack region; And performing morphological operation on the second crack region to extract a pavement crack skeleton, mapping the pavement crack skeleton back to the original pavement thermal anomaly region, and evaluating the damage st