Engineering structure crack detection method and system based on image processing technology

Abstract

The invention discloses an engineering structure crack detection method and system based on an image processing technology, and relates to the technical field of image processing, and the method comprises the following steps: S100, collecting image data of a plurality of time points in an engineering structure monitoring area, obtaining a corresponding sky brightness change curve and a reflector reflection angle change sequence, and obtaining a corresponding sky brightness change curve; generating a time-varying illumination profile covering a monitoring period and a reflector trajectory map, and constructing an illumination baseline for subsequent time sequence reference; and S200, deriving a brightness fluctuation sensitivity field according to the illumination baseline, screening an area with stable brightness response in the sensitivity field as an edge anchor point set, and carrying out initial position calibration on the crack candidate boundary by using the edge anchor point set. The crack identification stability is improved through the steps of dynamic illumination modeling, brightness sensitivity analysis, radiation calibration, boundary reestimation and the like, and high-precision, self-adaptive and sustainable optimization of crack detection in a complex environment is realized by introducing a closed-loop control and self-learning mechanism.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to an engineering structure crack detection method and system based on image processing technology. BACKGROUND

[0002] Engineering structure crack detection based on image processing technology refers to using digital imaging devices to obtain high-resolution images of the surface of engineering structures (such as bridges, tunnels, buildings, etc.), and through a series of image processing and analysis methods, automatically identifying, accurately positioning and quantitatively measuring the cracks on the structure surface. The process usually includes image preprocessing to remove uneven illumination and noise interference, crack feature enhancement to highlight texture details, edge and contour extraction to separate crack regions, geometric and texture feature analysis to judge crack morphology and direction, and combining pixel calibration or three-dimensional reconstruction technology to calculate crack width, length, expansion trend and other indicators, so as to realize non-destructive, rapid, objective and repeatable detection of structure surface damage. This method can replace or assist manual inspection, significantly improve the efficiency and accuracy of crack detection, and provide reliable data support for structure health assessment and maintenance decision-making.

[0003] The prior art has the following disadvantages: In the long-term crack monitoring process of outdoor engineering structures in the prior art, the monitoring environment cannot be kept stable, and is often significantly affected by weather conditions and changes in surrounding light sources. When the cloud layer moves rapidly above the monitoring area, the sky background brightness will fluctuate significantly in a short time; at the same time, the position and angle of the reflected light source from the glass curtain wall, vehicle metal surface, construction machinery, etc. will also change constantly, causing the light intensity of the crack edge area to present high-frequency flicker in the image sequence. This dynamic unevenness of the light condition will directly interfere with the stable identification of the crack boundary by the image processing algorithm, resulting in continuous pixel-level shifts in the crack edge detection results. In the dynamic monitoring sequence, such small shifts will accumulate frame by frame and form systematic drift, which will affect the accurate positioning of the crack contour and cause persistent deviations in the measurement of crack width and length. More seriously, this boundary drift caused by dynamic changes in light is often not effectively suppressed in the prior art, which can easily lead to directional misjudgment in the trend analysis of the structure health state assessment, such as misjudging a stable crack as rapidly expanding, or misjudging a crack that is expanding as tending to be stable, thereby affecting the accuracy of safety warning and the rationality of maintenance decision-making.

[0004] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] The purpose of the present application is to provide an engineering structure crack detection method and system based on image processing technology to solve the problems in the background art.

[0006] In order to achieve the above-mentioned purpose, the present application provides the following technical scheme: an engineering structure crack detection method based on image processing technology, comprising the following steps: S100, collecting image data at multiple time points in the engineering structure monitoring area, obtaining corresponding sky brightness variation curves and reflector reflection angle variation sequences, generating time-varying illumination profiles and reflector trajectory maps covering the monitoring period, and constructing an illumination baseline for subsequent time sequence reference; S200, deriving a brightness fluctuation sensitivity field according to the illumination baseline, selecting a region with stable brightness response in the sensitivity field as an edge anchor point set, and using the edge anchor point set to perform initial position calibration on the crack candidate boundary; S300, based on the edge anchor point set, inverting the exposure drift and reflected light intensity flicker characteristics in the image sequence, constructing a time sequence radiation calibrator, uniformly processing the brightness of the multiple time images, and generating a continuous image sequence with consistent brightness characteristics; S400, in the continuous image sequence with uniform brightness, extracting crack tracing information that is invariant to illumination, re-estimating the position of the crack boundary based on the edge anchor point set, and correcting the boundary position offset caused by illumination changes; S500, taking the re-estimated results of the crack boundary as a control signal to form a closed-loop acquisition control strategy, and adjusting the exposure time point, polarization angle parameter and observation viewing angle to actively suppress the subsequent illumination flicker interference in the image acquisition process; S600, after executing the closed-loop acquisition control strategy, feeding the current round of radiation calibration residual and boundary drift rate back to the illumination baseline, and dynamically updating the brightness fluctuation sensitivity field and the edge anchor point set based on the updated illumination baseline, so that the crack detection system has the ability of self-learning and continuous optimization of precision.

[0007] Preferably, step S100 comprises: A fixed observation position is set above or in front of the engineering structure monitoring area, and an image acquisition device with high dynamic range imaging capability is deployed, the sensor of each image acquisition device has a brightness sampling accuracy of not less than 12 bits, supports automatic exposure adjustment function and continuous shooting frequency of not less than 5 frames per second; The image acquisition device is controlled to continuously acquire image data at multiple time points during 30 minutes before sunrise to 30 minutes after sunset, and the acquisition frequency is 1 frame per minute when the illumination is stable and 1 frame per 10 seconds when the illumination changes drastically, so as to cover multiple day-night cycles and capture key illumination disturbance nodes; The image region division operation is performed on each frame of image, the sky region is extracted based on the brightness gradient distribution, the boundary contour shape and the color channel proportion, the average brightness, the brightness variance and the maximum edge brightness gradient of the region are calculated, the sky brightness change curve is formed, the reflection region in the image which meets the brightness threshold, the area and the shape continuity condition is identified, the boundary position, the center point, the brightness peak and the change parameter with time of the reflection region are extracted, and the reflection angle change sequence of the reflector is constructed; The image at each time point is indexed with the corresponding sky brightness statistical data and the reflector feature data to generate the time-varying illumination profile and the reflector trajectory atlas covering the monitoring period, and a three-dimensional data matrix with time as the main axis and image space position as the auxiliary axis is constructed to form the illumination baseline for subsequent crack boundary identification.

[0008] Preferably, the step S200 comprises: Based on the constructed illumination baseline, all image sequences in the monitoring period are extracted, and all images are uniformly sized and coordinate aligned to ensure that the pixel position is consistent with the physical structure surface area; The brightness values of each pixel in all image frames are counted, the brightness fluctuation amplitude and the change frequency are calculated, and the brightness fluctuation sensitivity field is constructed by combining the gray scale contrast and the spatial position attribute to form a two-dimensional distribution map representing the degree of illumination response; In the brightness fluctuation sensitivity field, the region with low brightness fluctuation amplitude, small change frequency, stable gray scale contrast and continuous spatial structure is screened, and the region with a main direction texture change amplitude less than 10 degrees is further screened to determine the edge anchor point set; The boundary line segment is extracted in the crack candidate region, and the direction matching and spatial pairing with the anchor points in the edge anchor point set are performed, when the difference between the boundary direction and the main direction of the anchor point is less than 15 degrees and the distance is less than 20 pixels, the initial position of the crack boundary is calibrated according to the anchor point center coordinates.

[0009] Preferably, the step S300 comprises: A number of anchor point regions with high position stability and strong gray scale texture continuity in the edge anchor point set are selected, the gray scale mean value, the gray scale extreme value and the gray scale gradient change rate of each region in the corresponding frame image of the image sequence are extracted, and the anchor point gray scale feature time sequence is generated; The offset of each frame of image and the reference image in the anchor point gray scale feature is calculated, and the brightness linear translation processing of the whole image is performed according to the offset, and the highlight region with a gray scale value higher than three times the average gray scale value of the whole image is identified, and the boundary sharpness, area and appearance time sequence are extracted; The identified highlight reflection region is locally corrected by replacing the edge gray scale mean value and reducing the brightness threshold to avoid brightness mutation or structure fault in the image, and the image sequence with uniform brightness is generated; The quality of the image sequence after brightness correction is inspected to ensure that the gray value of the anchor point area is within the set range, the boundary gray contrast of the crack area remains consistent, and the reflection interference in the highlight area is effectively reduced, thereby generating a continuous image sequence with consistent brightness.

[0010] Preferably, step S400 comprises: In the generated brightness-unified image sequence, the region with an elongated shape, a significant edge gray mutation, and a stable dark line feature before and after the brightness enhancement processing is extracted, the spatial position of the crack edge point in the region is identified, and the start point, end point, central trend line, and edge contour information of the crack are recorded; Based on the established edge anchor point set, the spatial position, boundary direction, gray range, and distance between the crack region of each anchor point in the image coordinate system are extracted, the spatial angle and distance between the crack line segment and one or more anchor points are calculated, and the reference relationship between the crack and the anchor point is established; According to the reference relationship, the spatial position of the crack boundary is fine-tuned, the geometric relationship between the crack line segment and the corresponding anchor point meets the preset direction consistency and distance constraint standard, and the time continuity and spatial position consistency of the crack boundary are corrected in combination with the crack position data in the previous and subsequent image frames; The consistency of all crack boundary re-estimation results is verified to check whether there are abnormal boundary extension, position offset exceeding the limit, width mutation, or incorrect connection with the structure main body, and for the crack boundary that does not meet the verification standard, a rollback and re-adjustment operation is performed, and finally the corrected crack boundary identification result is output.

[0011] Preferably, step S500 comprises: Based on the crack boundary re-estimation result, the crack boundary spatial position correction value in the image sequence is counted, the image frame in which the boundary position changes discontinuously and the spatial offset is large is identified, and the shooting time point thereof is extracted; The shooting time point is compared with the constructed illumination baseline, the reflection body trajectory and the sky brightness fluctuation are analyzed, and a period with gentle brightness change is selected as the optimal shooting time; The angle between the current imaging direction and the reflection path is judged, the device observation direction and the polarization plate rotation angle are adjusted to avoid the reflection high-occurrence area, and the crack boundary contrast is improved; According to the difference between the crack boundary gray and the background gray, the exposure time and the exposure compensation value are dynamically adjusted, and after the exposure time point, the polarization angle parameter, and the observation viewing angle are set, a new round of image acquisition is performed.

[0012] Preferably, step S600 comprises: Based on the brightness correction residual error and the boundary position drift rate of the crack region in the collected image, a feedback data set for the current image processing round is constructed; The feedback data set is used to update the constructed illumination baseline, correct the original brightness curve and adjust the stability identification of the image acquisition period; On the updated illumination baseline, the brightness fluctuation sensitivity field is reconstructed, and the illumination response fluctuation score of each pixel is calculated combined with the radiation residual error; According to the updated sensitivity field, the point with serious drift is removed, and the area with stable gray scale change and obvious boundary structure is selected as a new edge anchor point set for the next round of crack detection.

[0013] The engineering structure crack detection system based on image processing technology comprises an illumination modeling module, an anchor point extraction and boundary calibration module, a time sequence radiation calibration module, a boundary re-estimation module, a closed-loop control module and a model updating module: The illumination modeling module collects image data at multiple time points in the engineering structure monitoring area, obtains corresponding sky brightness change curves and reflector reflection angle change sequences, generates time-varying illumination profiles and reflector trajectory maps covering the monitoring period, and constructs an illumination baseline for subsequent time sequence reference; The anchor point extraction and boundary calibration module derives a brightness fluctuation sensitivity field according to the illumination baseline, selects an area with stable brightness response in the sensitivity field as an edge anchor point set, and uses the edge anchor point set to calibrate the initial position of the crack candidate boundary; The time sequence radiation calibration module inverses the exposure drift and reflected light intensity flicker characteristics in the image sequence based on the edge anchor point set, constructs a time sequence radiation calibrator, uniformly processes the brightness of the images at multiple time points, and generates a continuous image sequence with consistent brightness characteristics; The boundary re-estimation module extracts crack tracing information that is invariant to illumination in the continuous image sequence with uniform brightness, re-estimates the position of the crack boundary based on the edge anchor point set, and corrects the boundary position offset caused by illumination changes; The closed-loop control module uses the re-estimated results of the crack boundary as a control signal to form a closed-loop acquisition control strategy, adjusts the exposure time point, polarization angle parameter and observation viewing angle, and realizes active suppression of subsequent illumination flicker interference in the image acquisition process; The model updating module feeds back the radiation calibration residual error and boundary drift rate of the current round to the illumination baseline after executing the closed-loop acquisition control strategy, and dynamically updates the brightness fluctuation sensitivity field and the edge anchor point set based on the updated illumination baseline, so that the crack detection system has the ability of self-learning and continuous optimization of precision.

[0014] In the above technical solution, the technical effects and advantages provided by the present application are: The application establishes a dynamic updating illumination baseline by constructing a time-varying illumination profile and a reflector trajectory atlas, comprehensively captures the environmental illumination disturbance law; then, in combination with a brightness fluctuation sensitivity field and a spatial anchor point structure, a robust boundary positioning reference is provided to realize spatial precision constraint of crack profile identification; further, through radiation calibration and boundary position re-estimation steps, the brightness inconsistency between images and crack identification drift risk are significantly reduced; finally, a closed-loop acquisition control driven by crack identification result feedback and a dynamic model self-learning mechanism are introduced, so that the detection process has adaptive adjustment ability and precision evolution ability for different environments. In the long-term, multi-scene crack image monitoring task, the whole method not only effectively improves the repeatability and reliability of crack geometric feature measurement, but also provides a higher quality data basis for subsequent crack expansion trend analysis, structure health assessment and maintenance decision based on image sequence. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0016] Figure 1 The method flow chart of the crack detection method of the engineering structure based on image processing technology of the present application.

[0017] Figure 2 The module schematic diagram of the crack detection system of the engineering structure based on image processing technology of the present application. DETAILED DESCRIPTION

[0018] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations may, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these example implementations are provided so that this disclosure will be thorough and complete, and will fully convey the inventive aspects to those skilled in the art.

[0019] The present application provides a crack detection method of an engineering structure based on image processing technology as shown in Figure 1 The method comprises the following steps: S100, acquiring image data at multiple time points in the engineering structure monitoring area, obtaining corresponding sky brightness change curves and reflector reflection angle change sequences, generating time-varying illumination profiles and reflector trajectory atlases covering the monitoring period, constructing an illumination baseline for subsequent time sequence reference; In order to solve the technical problems of crack boundary identification drift and measurement error accumulation caused by unstable light conditions in outdoor environment, an illumination baseline construction method for long-term crack monitoring of engineering structures is proposed. This method can accurately record and restore the time sequence change process of environmental illumination during monitoring in natural environment, with significant time continuity, spatial resolution and response ability to reflection interference behavior. The specific steps are as follows: A fixed observation position is set above or in front of the surface of the engineering structure that needs to be monitored, and a set of image acquisition devices with high dynamic range imaging capability are deployed. The sensor of each device should have a brightness sampling accuracy of not less than 12 bits, and be able to support automatic exposure adjustment function and continuous shooting frequency of at least 5 frames per second. The viewing angle of the image acquisition device should cover the target structure surface, and include part of the open background around the structure to ensure that the sky area and the external environment that may cause reflection interference can be acquired. The image acquisition work should cover continuous multiple day-night cycles, and the daily acquisition time starts 30 minutes before sunrise and ends 30 minutes after sunset. The acquisition frequency is dynamically adjusted according to the intensity of environmental light change. In sunny and stable light period, the acquisition frequency is 1 frame per minute, and in cloudy, windy or frequent sunny and cloudy period, the acquisition frequency is 1 frame per 10 seconds, so as to ensure that the key time nodes of light change can be captured.

[0020] In each frame of image collected, first, according to the installation parameters of the imaging device and the geometric features in the image, the sky area above the image is identified. The identification of this area can be based on the brightness gradient distribution, the boundary contour shape and the proportion of blue channel after color space conversion. The brightness of all pixels in the sky area is counted, and the average brightness value, brightness variance and maximum brightness gradient of the edge area in each frame are extracted, and their change process in continuous time sequence is recorded, so as to form a brightness change curve representing the change of sky brightness with time. At the same time, the high-brightness reflection area in the non-sky area of the image is identified. The identification method is based on pixel brightness threshold, shape continuity and area judgment standard to judge whether it belongs to typical reflection source. For the area confirmed to belong to the reflector, further extract its boundary position, center point position, brightness peak value, edge diffusion radius, brightness gradient direction and their change with time in the image coordinate system, record the moving track of the reflection area in the continuous image frame and its brightness change trend.

[0021] Based on the sky brightness variation curve extracted in the above steps and the movement trajectory and light intensity variation data of the plurality of reflectors in the time sequence, a time sequence index is established, and each frame of image is synchronously paired with its corresponding brightness statistical data. The sky brightness data in each frame and the reflector reflected light intensity data are jointly encoded to construct a three-dimensional data matrix with time as the main axis and image space position as the auxiliary axis, which is used to represent the actual light influence degree of each region of the structure surface at each time point. By recording the change trend of the data matrix at a plurality of continuous time points, a set of time-varying light profile data is formed, which includes: the average brightness evolution curve of the sky region, the spatial position transformation trajectory of the reflector, the brightness variation amplitude trend, and the boundary displacement of the projection position of the reflection region in the image coordinate system. All the above data are mapped to a unified reference plane according to the structure coordinate system, so as to obtain the time-varying light profile covering the entire monitoring period.

[0022] After the construction of the light profile is completed, the behavior of the reflector in the image of the monitoring area is further modeled. The high-light region repeatedly appearing at a plurality of times is selected, the spatial appearance frequency, duration, brightness peak value distribution and directionality change of the high-light region are analyzed, the moving mode of the reflection source in the actual environment is determined, and the trajectory curve of the reflection source in the image coordinate system is integrated with the light profile data. Finally, a complete data set covering the sky brightness curve, the trajectory of the reflector, and the light interference features of the local region is established as a light baseline. The light baseline takes a unified time index as the core, takes the brightness variation trend of each key region in the image space as the reference, records the light state at each time through a high-resolution data structure, and can support the accurate restoration of the light interference on the crack region in the subsequent image sequence processing process. Compared with the existing technology which only relies on the fixed brightness mean value or the static image template, the light baseline has the response ability to dynamic light disturbance, and can provide reliable time reference basis and spatial correction conditions for the subsequent stable identification and measurement of the crack boundary.

[0023] The role of this step is to provide a light state basis with time reference and spatial distribution characteristics for the stable identification, accurate positioning and consistency of the measurement results of the crack boundary of the subsequent engineering structure. During the long-term monitoring process of the outdoor engineering structure, dynamic light fluctuation problems caused by cloud movement, sunlight angle change, reflector interference and the like are often encountered, which directly affect the gray distribution and edge features of the crack boundary in the image, causing the crack contour to shift, the boundary to be discontinuous, and even misidentification in the image processing algorithm in multiple frames of images, thereby causing systematic errors in the crack width and length measurement results. Most of the traditional image processing methods rely on single-frame static images for processing, lack the modeling ability of dynamic light changes, and cannot effectively eliminate the visual drift caused by external environmental fluctuations.

[0024] The illumination baseline constructed by this step can systematically record the trend of sky brightness change at each moment in the monitoring period, the distribution position and reflection intensity of the reflection source in the image, thereby providing a reference framework with time continuity and spatial positioning capability. In the subsequent image processing process, it can be judged whether there is abnormal illumination interference in the current image based on the illumination baseline, and radiation calibration, image uniform processing or boundary position correction can be carried out accordingly. Especially when analyzing the crack expansion trend and comparing the image sequence changes, the illumination baseline can be used as a high-precision time background information source, which can significantly improve the stability of crack identification and the comparability of measurement data, and reduce the risk of misjudgment. Therefore, this step is not only the basis for realizing the full-process automatic crack detection of the present application, but also an important technical link for realizing the robustness of dynamic monitoring, which is different from the existing static processing method.

[0025] S200, deriving a brightness fluctuation sensitivity field according to the illumination baseline, screening a region with stable brightness response in the sensitivity field as an edge anchor point set, and using the edge anchor point set to calibrate the initial position of the crack candidate boundary; After the construction of the illumination baseline, in order to extract the region with stable brightness response from the time sequence image and calibrate the initial position of the crack candidate boundary of the engineering structure surface, a method for constructing the brightness fluctuation sensitivity field and screening the edge anchor point is proposed. This method analyzes the brightness fluctuation of each pixel in the image in the monitoring period, identifies the image region with high stability, and uses it as the structure reference region, thereby realizing the spatial positioning consistency of the crack profile in the image sequence, and providing a high-robustness reference for subsequent boundary tracking and measurement calibration. The specific steps are as follows: From the illumination baseline built in the previous stage, all image sequences covering the monitoring period are extracted, and all images are arranged in chronological order under the premise of maintaining consistent spatial resolution, forming an image set indexed by frame. All images are aligned in a unified image size and coordinate system to ensure that the same pixel position can correspond to the same physical structure surface area. The brightness of each pixel in each frame of the image set is extracted, and the brightness time sequence corresponding to each pixel point in the monitoring period is recorded. Then, based on the brightness change value of each pixel point in all image frames, the brightness fluctuation amplitude of the point is calculated, and the specific calculation method is the difference between the maximum brightness and the minimum brightness; further, the brightness change frequency of the point is counted, and the number of times the brightness rises or falls beyond the set change threshold is recorded. In addition, the local gray scale contrast of the region to which the pixel point belongs in each frame of image is extracted, and whether its position in the image space is close to the high reflection region or the sky region is recorded. These parameters comprehensively reflect the brightness response stability of the pixel in the entire monitoring period. The above brightness response indicators are summarized by pixel position to generate a two-dimensional sensitivity distribution map with the same size as the image. The value corresponding to each pixel in the distribution map represents its sensitivity to changes in illumination in the monitoring period, i.e., the brightness fluctuation sensitivity field.

[0026] According to the constructed brightness fluctuation sensitivity field, the brightness response stable region in the image is screened. The specific screening process is as follows: in the sensitivity distribution map, all pixel points with a brightness fluctuation amplitude below a preset stability threshold, a brightness change frequency below a set upper limit, and a local gray scale contrast always maintained within a reasonable range are extracted as candidate points. These candidate pixel points are aggregated according to spatial adjacency, and isolated regions with an area less than 5x5 pixels and irregularly shaped scatter sets are removed, and regions with clear boundaries, continuous gray scale, and regular geometric shape are retained as brightness response stable regions. In each region block, the consistency of the main direction texture distribution is further detected, and regions with a texture direction change amplitude less than 10 degrees are retained as edge anchor point candidate areas. The above screening criteria ensure that the selected regions not only maintain temporal stability in brightness, but also have good morphological continuity in spatial texture structure.

[0027] On the basis of the candidate edge anchor region, spatial position extraction and stability verification are performed. For each anchor candidate region, the geometric center position in the image coordinate system is extracted, and whether there is obvious position drift in all image frames, i.e., whether the region center point has a translation of more than 3 pixels in the sequence, is counted. If there is obvious movement, the region is excluded. The regions with constant spatial position, stable boundary shape and controlled brightness change in the entire image sequence are retained, and the spatial position and main direction angle are marked. These regions are used as the final determined edge anchor set. The center point coordinates, main boundary direction, average gray value and gray standard deviation of each anchor point are recorded for subsequent crack boundary alignment and position calibration.

[0028] After extracting the crack candidate region in the image sequence, the above edge anchor set is combined as a spatial positioning reference to perform initial position calibration of the crack boundary. The specific implementation process is as follows: the boundary line segment extracted from the edge of the candidate crack region is analyzed, the boundary direction is calculated, and the angle is compared with the main direction of the edge anchor point. When the difference between the crack boundary direction and the main direction of the anchor point is less than 15 degrees and the distance is less than 20 pixels, the anchor point is used as the reference point of the crack boundary. According to the center coordinates of the anchor point, the crack boundary starting point is spatially paired with the anchor point to construct the mapping relationship between the crack contour and the anchor point. For the crack boundary segment that is successfully matched, according to the time stability characteristics of the anchor point, the initial position of the crack boundary is aligned in the entire image sequence to make the crack boundary have a unified reference. This processing method effectively avoids the problem of crack contour drift, breakage or offset in the image caused by light changes. Compared with the traditional image processing method based on single-frame edge extraction, it has significant time consistency and recognition accuracy advantages.

[0029] The role of this step is to form an edge anchor set with time stability and spatial continuity by constructing a brightness fluctuation sensitivity field and selecting image regions with stable brightness response, thereby providing reliable and accurate positioning reference for the initial position of the crack candidate boundary in the image sequence. In the long-term crack monitoring of engineering structures, the light conditions in the outdoor natural environment are uncontrollable and are often affected by factors such as cloud cover, changes in solar angle, and reflection interference, which causes the gray distribution and boundary shape of the crack edge in the image to fluctuate. Especially in the image sequence, the edge position of the crack contour may have a slight but continuous shift. This small boundary drift is easily accumulated frame by frame in the continuous image, which can cause false expansion or compression of the boundary, thereby affecting the measurement accuracy of the crack length, width and trend, and even misleading the trend judgment of the structure safety evaluation.

[0030] The present step is designed to solve the above problems. By constructing a luminance fluctuation sensitivity field, the response degree of each pixel in the image to the light change in the monitoring period can be systematically evaluated, and the regions with the smallest amplitude of luminance change, the lowest frequency of change, and the most stable gray scale distribution can be identified. These regions are not sensitive to light disturbance and usually maintain stable morphology and consistent position in the image sequence, so they can be used as "image reference points". Further filtering the spatial structure characteristics and texture directionality of these regions can obtain a set of edge anchor points with clear spatial structure and stable texture direction. Using these edge anchor points as the calibration reference of the crack candidate boundary can not only improve the accuracy of the initial boundary extraction of the crack, but also provide a stable alignment reference in the subsequent processing process to prevent boundary drift and recognition errors caused by light disturbance during crack recognition. In summary, the present step plays a core role as a "stabilizer" and "positioning reference" in the entire crack detection method, and is a basic supporting link for accurate crack recognition in multi-time sequence images.

[0031] S300, based on the edge anchor point set, the exposure drift and reflected light intensity flicker characteristics in the image sequence are inverted to construct a time sequence radiation calibrator for luminance uniform processing of multi-time images to generate a continuous image sequence with consistent luminance characteristics; After completing the construction of the edge anchor point set based on the luminance fluctuation sensitivity field and the initial position calibration of the crack candidate boundary, in order to eliminate the overall exposure inconsistency and local highlight reflection interference problems in the image sequence caused by light fluctuation, further luminance characteristic calibration processing is performed on the multi-time image sequence to ensure that each image has consistency in the luminance space, thereby forming a continuous image sequence that can be directly compared and analyzed in space. This processing is implemented by the following steps: Using the edge anchor point set obtained in the previous step, select several anchor point regions with the highest position stability, the strongest boundary clarity, and the best gray scale texture continuity in the image sequence. For each frame of image in these anchor point regions, extract the average gray scale value, the gray scale extreme value (minimum and maximum gray scale), and the gray scale gradient change rate of the region edge, and record the sequence values of these gray scale characteristics changing with time as the index. By comparing the gray scale characteristics of these anchor point regions at different time points with the corresponding characteristic values in the first frame (or a selected frame as the reference frame), it can be judged whether the current frame image has overexposure or underexposure in the overall brightness, and the luminance offset between the current frame and the reference frame can be calculated accordingly. The gray scale offset data extracted in this step reflects the exposure drift trend of the entire image, providing a quantitative adjustment basis for subsequent correction operations.

[0032] Identify the local highlight interference area in each frame of image that may be caused by the surface of high-reflective material (such as metal equipment shell, glass curtain wall or water surface). The specific method is: taking each image as the processing object, traversing all pixel regions, marking the continuous region whose brightness value exceeds three times the average gray value of the whole image; then judging the shape, area, boundary sharpness of these regions and their change trend in the previous and next frames, eliminating the bright spot area with fuzzy boundary, small area or continuous position in multiple frames of image, retaining those areas that appear in a single frame, with sharp boundary, abnormal high gray and suddenly appear or disappear in the previous and next frames, and determining as the reflection flicker area. For these areas, record their position in the image coordinate system, the maximum area gray value, the edge gray change rate and the occurrence time, and mark as high-risk light interference area.

[0033] After identifying the exposure drift and highlight flicker features existing in the image, take a frame of image with uniform brightness, stable edge anchor gray and unaffected by local strong reflection interference in the image sequence as the brightness calibration reference image. In the reference image, the gray value of each anchor point is taken as the standard reference value, and the gray of the corresponding anchor point in other frames of image is compared to calculate the brightness increment or decrement required by each frame. On this basis, perform brightness calibration operation on each pixel in the image, specifically: according to the average gray difference of anchor points, linearly shift the brightness of the whole image to make all anchor points gray close to the reference image level after correction; for the local strong light area, i.e. the high-risk area marked in the previous step, use edge mean replacement or brightness threshold reduction method to pull its gray value back to the anchor point gray interval range, and maintain the natural transition of its surrounding pixel gray gradient, avoiding the appearance of corrected image fault or unnatural bright spot. The above processing ensures that the overall brightness and local structure brightness of the image remain consistent with the reference image after correction, without over-correction or color scale distortion.

[0034] A unified quality inspection is performed on all the images after brightness calibration, including the following three aspects: (1) whether the gray value of the edge anchor point region is within the upper and lower 10 gray units of the recorded range of the reference image, if it exceeds this range, repeat the brightness translation correction; (2) whether the boundary gray contrast of the original crack candidate region in each frame image remains consistent before and after correction, that is, the gray difference between the background and the crack should not be reversed or significantly reduced; (3) whether the high-light reflection region still has a sudden boundary or a local extremely bright region after correction, if so, the region is reprocessed by neighborhood replacement until the correction meets the requirements of overall gray distribution uniformity and boundary clarity. Through the above strict threefold verification, the image sequence has horizontal consistency and vertical comparability in brightness distribution, forming a standard image sequence that can be used for structural crack time series evolution analysis, strike extraction and expansion rate calculation. Compared with the existing technology which only performs average correction on the overall brightness of the image or only relies on image histogram equalization, this method combines anchor point gray reference, local area reflection interference suppression and image overall reconstruction, which can more comprehensively and stably restore the structural intrinsic features of the image at each time, effectively eliminating the interference of light condition changes on crack identification.

[0035] The role of this step is to solve the problem of inconsistent image brightness caused by changes in environmental lighting conditions in long-term monitoring of engineering structures, thereby ensuring that the images collected at multiple times have uniform brightness characteristics, making the identification and measurement of crack boundaries in time series images comparable and stable. In actual engineering environments, image acquisition is often in outdoor scenes with severe changes in natural light, and the strength and direction of light are affected by many factors such as sunlight angle, cloud cover, and reflection interference. Since the imaging device is sensitive to light, the images collected often have significant differences in brightness, especially in some image frames with overexposure, underexposure, or local high-light flicker. These phenomena will directly affect the gray distribution, texture features, and edge clarity of the crack boundaries in the image, thereby reducing the accuracy of crack identification by image processing algorithms and possibly causing systematic bias in the measurement of crack geometric parameters such as width and length.

[0036] To overcome this problem, this step builds a radiation characteristic inversion method based on time series gray scale feature analysis by referencing the edge anchor point set extracted in the previous stage, which can identify the exposure drift trend and local highlight reflection interference behavior in the image, and accordingly calibrate the brightness of the image. In the process of image brightness adjustment, not only the overall brightness level of the whole image is unified, but also the identified high reflection area is suppressed to avoid the masking effect of local strong light on the structure texture. Finally, all images are processed into standard images that are consistent in overall brightness distribution, have stable edge anchor point gray scale values, and have effective control of local reflection interference. In this way, images taken at different times can be compared in the same brightness reference for structural damage identification, crack profile extraction, and expansion trend analysis. In summary, this step plays a key role in the "light standardization" of the whole crack detection process, and is the key link to ensure the quality of image sequence processing and achieve time-dimension comparative analysis. It is significantly different from the traditional histogram equalization or simple brightness normalization based on static images, and has higher dynamic environmental adaptability and practical engineering application value.

[0037] S400, in the continuous image sequence with uniform brightness, extract the crack recording information that is invariant to light, re-estimate the position of the crack boundary based on the edge anchor point set, and correct the boundary position offset caused by light changes; After completing the brightness uniformity processing of the image sequence, to ensure the consistency of the crack boundaries extracted under different shooting conditions in space position, and effectively offset the slight profile offset caused by changes in lighting conditions, the position of the crack boundary is re-estimated to obtain a crack identification result with time continuity and spatial accuracy. This process is based on the crack morphology information extracted from the uniformly bright image sequence, combined with the edge anchor point set constructed earlier, through spatial alignment and offset correction, to achieve stable positioning of the crack boundary. Specifically, the following steps are included: Select the image sequence processed by brightness uniformity as the processing object, and extract the significant gray difference area of the crack area from each frame of image in turn. This gray difference area usually appears as an elongated structure in the image background with clear edges and obvious gray scale changes. To ensure that the extracted crack boundary is not affected by local reflection or exposure changes, only those areas that exhibit stable dark line features before and after image brightness enhancement processing are extracted in the image. Perform pixel-level analysis on these areas to identify the position of the crack edge point, record the crack start point, end point, center line, edge contour line, and local width. To further improve the recognition accuracy, exclude the gray scale gradual change fuzzy area, shadow overlapping area, and component joint area during crack boundary extraction. Through multi-frame comparison, the crack line segments that do not change significantly over time, have stable morphology, and have clear boundary characteristics are retained as the crack recording information that is invariant to light.

[0038] The edge anchor set constructed based on the brightness fluctuation sensitivity field in the previous stage is called, and the spatial coordinates, boundary direction, gray value range, and relative spatial distance between each anchor point and the crack region in the image are extracted. For each crack trace, the Euclidean distance and the direction angle between it and all anchor points are calculated, and the two or more anchor points with the highest correlation are marked. If a crack line segment is in the spatial region formed by two anchor points and the consistency with the main direction of the anchor points is greater than a certain threshold (for example, within 10 degrees), the line segment is determined as a re-estimation target with reference value. In this process, the gray gradient change on the anchor point boundary is extracted and matched with the gray change characteristics of the crack edge region to enhance the accuracy of the re-estimation reference and prevent position correction deviation caused by false pairing.

[0039] According to the reference relationship between the crack line segment and the edge anchor point established above, the crack boundary position is re-estimated and corrected. The spatial offset between the crack trace and the anchor point is taken as the reference to calculate the position of the crack line segment in the image coordinate system. The specific operation includes: for each crack boundary to be re-estimated, the direction of the line connecting the endpoints and the associated anchor points is calculated, and then the crack boundary is adjusted in space as a whole along the direction until the theoretical geometric relationship between it and the anchor point is consistent in space. If the crack boundary has continuous offset phenomenon in the previous and subsequent image frames, further reference is introduced in the time dimension, and the crack position in the current frame is compared with the crack position in the previous and subsequent frames by difference value comparison to ensure that the re-estimated crack boundary is not only accurate in spatial position, but also maintains the trend continuity and smooth width change in time sequence. For the gaps between multiple crack boundaries, the gaps are connected by completing the line segments and smoothing the boundary curvature, so that the final re-estimated crack boundary appears as a clear and complete contour line in the entire image.

[0040] The logical consistency of all crack re-estimation results is verified within the entire image range to ensure that the crack position adjustment operation does not cause boundary jumping, excessive translation, or distortion. The specific verification contents include: whether the crack line appears abnormal extension out of the original texture region, whether the re-estimated boundary crosses the anchor point region to cause gray inconsistency, whether the crack width appears mutation in the time sequence, and whether the crack start and end points maintain the correct connection relationship with the surrounding structure. If a segment of the re-estimated crack boundary does not pass the above check items, it is rolled back to the initial position and the translation vector is recalculated combined with the local gray information. After all the crack boundaries are re-estimated and corrected and pass the consistency check, the corrected crack boundary result is stored as the formal recognition result for subsequent crack trend judgment, structure evaluation, and change rate calculation.

[0041] The role of this step is to solve the problem of unstable identification of crack boundary, spatial position drift and measurement error accumulation in the image caused by changes in environmental lighting, so as to realize consistent expression and stable positioning of the crack boundary in multiple time images. In long-term crack monitoring of engineering structures, image acquisition is often limited by natural light conditions, such as changes in sunlight angle, cloud cover, and the influence of surrounding reflectors. Although these changes can be partially suppressed after uniform brightness processing, they may still cause slight deformation and edge blurring of the crack texture in the image, especially in the details of the boundary, which can easily cause pixel-level shifts. If this small shift is continuously accumulated in multiple images, it will cause spatial drift of the crack profile, resulting in systematic deviations in crack length, width, and orientation, ultimately interfering with the accuracy of the judgment of the crack development trend in structural health assessment.

[0042] This step extracts illumination invariant crack tracing information from the image sequence after uniform brightness, and combines the previously extracted edge anchor point set with spatial stability and illumination robustness to establish a spatial mapping relationship between the crack boundary and the fixed reference point in the image space, thereby realizing the re-estimation and correction of the crack boundary position. By positioning and adjusting the crack boundary based on the anchor point, it can ensure that the same crack in the image has consistent identification results in the spatial coordinate system under different shooting times and different lighting conditions, eliminating the boundary drift phenomenon caused by lighting disturbance. Further, the re-estimated crack profile can maintain morphological continuity, texture clarity and width stability in the image sequence, greatly improving the temporal consistency and structural comparability of crack identification, providing a high-precision data basis for subsequent crack expansion speed judgment, dynamic change analysis, and structural damage trend assessment based on image sequences.

[0043] S500, with the re-estimated results of the crack boundary as the control signal, a closed-loop acquisition control strategy is formed to actively suppress the interference of subsequent lighting flicker in the image acquisition process by adjusting the exposure time point, polarization angle parameter and observation angle; After completing the crack boundary position re-estimation and obtaining reliable spatial correction results, in order to improve the adaptability to lighting changes in the subsequent image acquisition process and reduce image distortion caused by environmental factors, an acquisition parameter adjustment method based on the feedback of the crack boundary re-estimation results is used. This method takes identification quality as the target, actively controls the imaging time point, observation angle and optical parameters of the acquisition device to realize prediction and avoidance of possible future lighting interference, which includes the following steps: According to the boundary position correction value recorded in the previous stage crack boundary re-estimation result, the crack space drift trend in each frame image in the image sequence is analyzed. The specific operation is: traversing each frame image in the image sequence, reading the coordinate position difference value before and after the crack boundary re-estimation, and calculating the angle change amount, width change ratio and edge definition index of the crack center line. The above data reflects the difference in crack recognition stability in images collected at different time periods, which can be used to infer the degree of light disturbance in image collection at a certain time point. Arrange all image frames in the order of shooting time, extract the image frames with large crack boundary position offset, poor recognition continuity and unclear texture, mark them as poor recognition period, and correspond to the actual collection time to form the mapping relationship between collection time and crack recognition stability.

[0044] Compare the above-mentioned poor recognition time period with the previously constructed light baseline to determine the reason for the crack recognition deviation caused by the light change. By matching the corresponding sky brightness fluctuation curve and reflector trajectory in the light baseline at these time points, it can be found whether there are strong reflection incidence, rapid cloud movement, backlight direct reflection and other interference situations. Based on the analysis result, select the time interval with small light fluctuation amplitude and gentle brightness change slope in the adjacent time period as the preferred window for the next round of image collection. Take the specific shooting time of each image as the unit, sample several frames of images forward and backward in time, analyze the stability of the crack boundary, comprehensively evaluate the optimal shooting time, and update the collection time schedule based on this.

[0045] After determining the optimal collection time, according to the reflection interference distribution characteristics recorded in the poor image frames, combined with the spatial displacement information of the reflector in the light baseline, the angle relationship between the current collection direction and the main reflection path is inferred. If it is found that the angle between the current device imaging direction and the reflection path is less than 30 degrees, that is, there is a high probability of reflection overlap risk, the imaging direction is adjusted to an alternative observation direction with an angle greater than 60 degrees with the reflection direction. Combined with the actual spatial position of the crack area, the camera equipment is physically adjusted to ensure that the shooting direction avoids the reflection hotspot area. At the same time, analyze the reflection intensity difference of the crack area under different polarization angle conditions, calculate the optimal polarization plate rotation angle according to the light source incidence direction and the reflector material characteristics, adjust the polarization plate to the angle that can filter out the reflected light component to the maximum (for example, rotate to 45 degrees or 90 degrees around the optical axis), in order to improve the contrast of the crack texture boundary.

[0046] According to the light intensity and image gray distribution characteristics of the selected acquisition time point, the exposure parameters of the image acquisition device are adjusted. The specific method is: in low light conditions, in order to ensure the complete imaging of crack details, the exposure time is appropriately extended, for example, from the original setting of 1 / 250 seconds to 1 / 125 seconds; in high reflection conditions, in order to prevent local overexposure of the image, the exposure time is shortened to within 1 / 500 seconds. At the same time, according to the contrast between the crack gray range and the background gray range identified in the previous round of sampling, the exposure compensation value is dynamically adjusted, so that the crack profile maintains high gray separation in the final image, and the quality of subsequent image processing is improved. After completing the joint optimization of the above-mentioned time point, imaging direction, polarization angle and exposure parameter, the next round of image acquisition task is started, and the collected image sequence is sent to the crack identification process. The real-time evaluation of the crack boundary re-estimation in the new image is carried out, and if the identification quality is obviously improved, it means that the current parameter combination is effective; if the identification quality is still not ideal, the new round of identification result is fed back to the acquisition control process, and the shooting parameters are adjusted again to realize closed-loop control.

[0047] The role of this step is to dynamically adjust the key parameters in the image acquisition process through the introduction of real-time feedback mechanism of crack boundary identification results, to build a closed-loop control mechanism with adaptive ability, so as to realize the active suppression of subsequent light interference and guarantee the continuous stability of image quality in the process of crack detection. In the actual long-term monitoring scene of engineering structure, the imaging device is usually in a complex and uncontrollable outdoor environment, and is easily disturbed by natural light changes, reflector interference, weather influence and other factors. These disturbances will cause problems such as inconsistent exposure, local highlight and abnormal gray distribution in the image, thereby affecting the boundary clarity and identification accuracy of the cracks in the image. Especially in continuous acquisition at multiple times, if the acquisition conditions remain unchanged, the identification quality will decrease frequently when the light changes dramatically, resulting in insufficient data reliability and measurement result deviation, and finally affecting the accuracy of the judgment of the structure crack propagation trend.

[0048] In order to solve this problem, this step uses the boundary offset, width fluctuation and trend change in the crack boundary re-estimation result to identify the image frame with unstable crack identification, and analyzes the correlation between the image frame and the acquisition time, imaging angle and reflection interference. Based on this analysis result, the key parameters of image acquisition are finely adjusted, including reselecting the most suitable image acquisition time point to avoid periods of strong sky brightness fluctuation or high reflection intensity; adjusting the imaging direction to avoid the observation angle coinciding with the strong reflection path; controlling the rotation angle of the polarizer to effectively weaken the pollution of the mirror reflection to the image; and adjusting the exposure time according to the current light intensity to balance the acquisition of brightness and detail information. Through the comprehensive adjustment of the above parameters, the transformation from "passive response to light conditions" to "active intervention in image quality" is realized.

[0049] Finally, the acquisition control mechanism established in this step not only improves the stability, clarity and continuity of the crack boundary in the image, but also enables the entire monitoring process to dynamically adapt to environmental changes. This strategy can continuously optimize image input quality, providing a high-consistency and high-reliability data foundation for subsequent crack identification, measurement and evaluation, and is particularly suitable for long-term continuous monitoring, day-night changes, or complex high-reflectivity scene applications in bridge, tunnel, large concrete components and other engineering environments. It is the key link in the entire image crack detection process to improve the anti-interference and robustness.

[0050] S600, after executing the closed-loop acquisition control strategy, the current round of radiation calibration residual and boundary drift rate are fed back to the illumination baseline, and the brightness fluctuation sensitivity field and edge anchor point set are dynamically updated based on the updated illumination baseline, so that the crack detection system has the ability of self-learning and precision continuous optimization; After completing the image acquisition and processing process guided by the closed-loop acquisition control strategy, in order to gradually improve the accuracy of subsequent crack detection and identification, data feedback processing is required on the image processing results of this round, and the illumination response reference model and the reference structure of crack identification in the monitoring system are adjusted accordingly. The specific process includes the following steps: After completing the processing of the continuous image sequence after brightness unification and obtaining the crack boundary re-estimation results, the brightness correction residual and spatial offset in the crack area of each image frame are extracted. The specific operation is: compare the original image brightness and the brightness value after radiation calibration pixel by pixel, calculate the brightness error value, and form the radiation calibration residual map of the image frame. At the same time, compare the two-dimensional coordinate positions of the crack boundary before and after re-estimation, quantify the drift distance in the horizontal and vertical directions, and further calculate the overall drift rate of the crack profile, expressed as the average position offset of the crack boundary per unit time. The above brightness residual and boundary drift rate data are classified and sorted according to the image acquisition time, image sequence number, imaging angle and other attributes to build a complete feedback data set.

[0051] The feedback dataset is fused with the previously constructed illumination baseline to correct the original recording errors and prediction biases in the illumination baseline. The specific process is as follows: for each image frame, the sky brightness curve and reflector trajectory data in the illumination baseline are retrieved for the time period corresponding to the image frame, and the radiation residual data for the time period is used to correct the original brightness curve, so that the recorded illumination intensity in the baseline is more consistent with the actual image response. At the same time, the boundary drift rate is used to judge the imaging stability in the time period, and the reliability label of the time period in the illumination baseline is adjusted. If a time period repeatedly appears with a high drift rate, it indicates that the ambient light in the time period has a dramatic fluctuation, and is marked as a "high fluctuation section". The subsequent collection will reduce the priority of this section. In this way, the illumination baseline is transformed from a static illumination record to a dynamic feedback data model, which continuously reflects the changing characteristics of the actual monitoring environment.

[0052] On the basis of the updated illumination baseline, the brightness fluctuation sensitivity field is reconstructed to evaluate the illumination response fluctuation of each region in the image during the monitoring period. The specific method is as follows: for each image frame, the brightness response sequence of the corresponding region at different time points is extracted based on its spatial coordinates, which is compared with the reference brightness curve of the region in the illumination baseline, and the amplitude, frequency and direction trend of the brightness fluctuation are calculated. In this process, the radiation residual is used as a correction factor to introduce the actual error of each pixel point, improving the accuracy of the sensitivity evaluation. Finally, the brightness stability score of each pixel point is mapped to the image space to form a complete two-dimensional sensitivity field distribution map. This map can be used in the subsequent identification stage to screen stable regions in the image, improving the immunity of the identification model to adverse light effects.

[0053] Based on the newly constructed brightness fluctuation sensitivity field, the edge anchor point set with high illumination stability is reselected. The selection process includes the following operations: first, identify in the image space those regions with low sensitivity values, smooth gray change curves, and fluctuation ranges not exceeding the set threshold; second, analyze the texture features and spatial structure of these regions to confirm whether they have obvious edge structures in the crack contour neighborhood, and preferentially select those edge points that have been stable for multiple monitoring periods and have high recognition accuracy as new anchor points; third, remove the points in the last round of anchor point set that have significant drift or have lost recognition value due to strong light reflection in this round of monitoring, and maintain the spatial distribution balance of the points in the set; finally, the updated anchor point set is used as a spatial reference for position reestimation and radiation calibration in the next round of crack identification, achieving robustness improvement in the crack boundary identification process.

[0054] The role of this step is to effectively feed back the error information generated in each round of image acquisition and crack identification process to the basic level of light modeling and identification reference structure, so as to realize the continuous optimization and self-learning upgrade of the crack detection process, and ensure that the identification accuracy and light adaptation ability are always maintained at a high level in the long-term operation. Specifically, in the long-term monitoring of actual engineering structures, the light environment presents obvious nonlinear and non-periodic change characteristics, such as seasonal change, weather fluctuation, light source reflection path change, etc., which will cause deviation between the original light baseline and the current environment. If the crack identification still relies on the old static model and fixed anchor point area, it will inevitably lead to accumulation of identification error, drift of crack profile, inaccurate width measurement, and even misjudgment. Therefore, effectively capturing the residual brightness error and identification error in each round of identification task and using it for model updating in a timely manner is the core of ensuring detection accuracy and long-term adaptability.

[0055] By feeding back the radiation calibration residual and boundary drift rate to the light baseline, the errors caused by initial modeling deviation or environmental changes in the baseline can be effectively corrected to make it more consistent with the current actual lighting conditions, improving the accuracy of subsequent acquisition time point selection and light prediction. At the same time, based on the updated light baseline, the brightness fluctuation sensitivity field is reconstructed, which can evaluate the response degree of each region in the image to the light change in real time, identify more stable and robust identification reference regions, and filter out more reliable edge anchor point set, providing stable support for subsequent crack boundary extraction and position re-estimation.

[0056] More importantly, this process is not one-time modeling or static updating, but continuously feeds back the error information generated in each identification task to promote the dynamic evolution of the light model and anchor point structure, so that the entire detection method has self-adaptive learning ability. This ability not only effectively suppresses the error accumulation problem caused by long-term monitoring environmental changes, but also ensures that the identification mechanism can maintain high identification accuracy and image processing consistency under different seasons, different climates, and different day-night lighting conditions. Therefore, this step is the key to transforming passive identification mechanism into active evolution identification mechanism, and is the precision closed loop and learning core throughout the whole process of the crack detection method, which has obvious innovation and practical value different from the prior art.

[0057] The engineering structure crack detection method based on the image processing technology realizes high stability, high consistency and high precision of crack boundary recognition results in a complex outdoor lighting environment, and overcomes key problems such as crack recognition drift, measurement error accumulation and trend judgment error caused by factors such as sky brightness fluctuation and reflected light interference in the prior art. The method constructs a time-varying light profile and a reflector trajectory atlas, establishes a dynamic updated light baseline, and comprehensively captures the environmental light disturbance law; combined with the brightness fluctuation sensitivity field and the spatial anchor point structure, a robust boundary positioning reference is provided to realize the spatial precision constraint of crack contour recognition; further through the radiation calibration and boundary position re-estimation steps, the brightness inconsistency between images and the crack recognition drift risk are significantly reduced; finally, the crack recognition result feedback driven closed loop acquisition control and dynamic model self-learning mechanism are introduced, so that the detection process has the self-adaptive adjustment ability and precision evolution ability for different environments. In the long-term, multi-scene crack image monitoring task, the method not only effectively improves the repeatability and reliability of crack geometric feature measurement, but also provides a higher quality data basis for subsequent crack expansion trend analysis, structure health assessment and maintenance decision based on image sequences.

[0058] The application provides an engineering structure crack detection system based on an image processing technology as shown in the formula (I): Figure 2 The engineering structure crack detection system based on the image processing technology comprises a light modeling module, an anchor point extraction and boundary calibration module, a time sequence radiation calibration module, a boundary re-estimation module, a closed loop control module and a model updating module. The light modeling module collects image data at multiple time points in the engineering structure monitoring area, obtains corresponding sky brightness change curves and reflector reflection angle change sequences, generates a time-varying light profile and a reflector trajectory atlas covering a monitoring period, and constructs a light baseline for subsequent time sequence reference. The anchor point extraction and boundary calibration module derives a brightness fluctuation sensitivity field according to the light baseline, selects a region with stable brightness response in the sensitivity field as an edge anchor point set, and uses the edge anchor point set to perform initial position calibration on a crack candidate boundary. The time sequence radiation calibration module inverses exposure drift and reflected light intensity flicker characteristics in the image sequence based on the edge anchor point set, constructs a time sequence radiation calibrator, uniformly processes brightness of the images at multiple time points, and generates a continuous image sequence with consistent brightness characteristics. The boundary re-estimation module extracts crack tracing information that is invariant to light in the continuous image sequence with uniform brightness, re-estimates the position of the crack boundary based on the edge anchor point set, and corrects boundary position offset caused by light changes. A closed-loop control module takes the re-estimation result of the crack boundary as a control signal, forms a closed-loop acquisition control strategy, and realizes active suppression of subsequent light flicker interference in the image acquisition process by adjusting the exposure time point, the polarization angle parameter and the observation angle. A model updating module feeds the current round of radiation calibration residual and the boundary drift rate to the light baseline after executing the closed-loop acquisition control strategy, and dynamically updates the brightness fluctuation sensitivity field and the edge anchor point set based on the updated light baseline, so that the crack detection system has the ability of self-learning and precision continuous optimization.

[0059] The engineering structure crack detection method based on image processing technology provided by the embodiment of the present application is realized through the engineering structure crack detection system based on image processing technology, and the specific method and process of the engineering structure crack detection system based on image processing technology are described in the embodiment of the engineering structure crack detection method based on image processing technology, which will not be described here.

[0060] The above only describes some exemplary embodiments of the present application by way of illustration, and it is needless to say that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present application. Therefore, the above drawings and descriptions are illustrative in nature and should not be understood as limiting the scope of protection of the claims of the present application.

Claims

1. A method for detecting cracks in engineering structures based on image processing technology, characterized in that, Includes the following steps: S100 collects image data at multiple time points in the engineering structure monitoring area, obtains the corresponding sky brightness change curve and reflector reflection angle change sequence, generates a time-varying illumination profile and reflector trajectory map covering the monitoring period, and constructs an illumination baseline for subsequent time series reference; S200: Derive the brightness fluctuation sensitivity field based on the illumination baseline, select regions with stable brightness response in the sensitivity field as the edge anchor point set, and use the edge anchor point set to initially calibrate the candidate crack boundary. S300, based on the set of edge anchor points, inverts the exposure drift and reflected light intensity flicker features in the image sequence, constructs a temporal radiometric calibrator, performs brightness unification processing on images at multiple time points, and generates a continuous image sequence with consistent brightness characteristics. S400 extracts crack delineation information with constant illumination from a continuous image sequence with uniform brightness, re-estimates the position of crack boundaries based on the set of edge anchor points, and corrects the boundary position offset caused by illumination changes. The S500 uses the re-estimation result of the crack boundary as the control signal to form a closed-loop acquisition control strategy. By adjusting the exposure time point, polarization angle parameter and observation angle, it can actively suppress the interference of subsequent illumination flicker during the image acquisition process. After executing the closed-loop acquisition and control strategy, the S600 feeds back the current round of radiation calibration residuals and boundary drift rates to the illumination baseline, and dynamically updates the brightness fluctuation sensitivity field and edge anchor point set based on the updated illumination baseline, enabling the crack detection system to have the ability to learn and continuously optimize accuracy.

2. The method for detecting cracks in engineering structures based on image processing technology according to claim 1, characterized in that, Step S100 includes: A fixed observation position is set up above or in front of the engineering structure monitoring area, and image acquisition equipment with high dynamic range imaging capability is deployed. The sensor of each image acquisition device has a brightness sampling accuracy of no less than 12 bits, supports automatic exposure adjustment function and a continuous shooting frequency of no less than 5 frames per second. The image acquisition device is controlled to continuously acquire image data at multiple time points from 30 minutes before sunrise to 30 minutes after sunset. When the light is stable, it acquires data at a frequency of 1 frame per minute, and when the light changes drastically, it acquires data at a frequency of 1 frame per 10 seconds, so as to cover multiple day and night cycles and capture key light disturbance nodes. For each frame of the image, an image region segmentation operation is performed. The sky region is extracted based on the brightness gradient distribution, boundary contour shape and color channel ratio. The average brightness, brightness variance and edge brightness gradient maximum of the region are calculated to form the sky brightness change curve. The reflective regions in the image that meet the brightness threshold, area and shape continuity conditions are identified. Their boundary position, center point, brightness peak and time change parameters are extracted to construct the reflective body reflection angle change sequence. A time-series index is established for the images at each time point, along with their corresponding sky brightness statistics and reflector feature data. This generates a time-varying illumination profile and reflector trajectory map covering the monitoring period. Based on this, a three-dimensional data matrix is ​​constructed with time as the main axis and image spatial location as the auxiliary axis, forming an illumination baseline for subsequent crack boundary identification.

3. The method for detecting cracks in engineering structures based on image processing technology according to claim 1, characterized in that, Step S200 includes: Based on the constructed illumination baseline, all image sequences within the monitoring period are extracted, and all images are processed to uniformly size and coordinate alignment to ensure that the pixel position is consistent with the surface area of ​​the physical structure. The brightness values ​​of each pixel in all image frames are statistically analyzed to calculate the brightness fluctuation amplitude and frequency of change. Combined with grayscale contrast and spatial location attributes, a brightness fluctuation sensitivity field is constructed to form a two-dimensional distribution map representing the degree of illumination response. In the brightness fluctuation sensitivity field, regions with low brightness fluctuation amplitude, low change frequency, stable grayscale contrast and continuous spatial structure are selected. Regions with a main direction texture change amplitude of less than 10 degrees are further selected and identified as the edge anchor point set. Boundary line segments are extracted from the crack candidate region and matched with anchor points in the edge anchor point set in terms of direction and space. When the difference between the boundary direction and the main direction of the anchor point is less than 15 degrees and the distance is less than 20 pixels, the crack boundary is initially calibrated based on the center coordinates of the anchor point.

4. The method for detecting cracks in engineering structures based on image processing technology according to claim 1, characterized in that, Step S300 includes: Select several anchor point regions with high positional stability and strong grayscale texture continuity from the set of edge anchor points, and extract the grayscale mean, grayscale extreme value and grayscale gradient change rate of each region in the corresponding frame image in the image sequence to generate the anchor point grayscale feature time series. Calculate the offset of each frame image from the reference image in terms of anchor point grayscale features, and perform linear brightness translation processing on the entire image based on the offset. At the same time, identify the bright areas with grayscale values ​​higher than three times the average grayscale value of the entire image, and extract their boundary sharpness, area and occurrence sequence. For the identified bright reflection areas, local correction is performed by replacing the average gray level at the edge and reducing the brightness threshold to avoid abrupt changes in brightness or structural breaks in the image, and to generate an image sequence with uniform brightness. The image sequence after brightness correction is quality checked to ensure that the gray value of the anchor point area is within the set range, the gray value contrast of the crack area boundary is consistent, the reflection interference of the bright area is effectively reduced, and finally a continuous image sequence with consistent brightness is generated.

5. The method for detecting cracks in engineering structures based on image processing technology according to claim 1, characterized in that, Step S400 includes: In the generated brightness-uniform image sequence, regions with elongated shapes, obvious edge gray-scale abrupt changes, and stable dark line features that remain stable before and after brightness enhancement are extracted from the images. The spatial location of crack edge points in these regions is identified, and the crack's start point, end point, center direction line, and edge contour information are recorded. Based on the established set of edge anchor points, the spatial position, boundary direction, gray range and distance between each anchor point and the crack area in the image coordinate system are extracted. The spatial angle and distance between the crack line segment and one or more anchor points are calculated, and the reference relationship between the crack and the anchor points is established. Based on this reference relationship, the position of the crack boundary is spatially fine-tuned so that the geometric relationship between the crack line segment and the corresponding anchor point meets the preset direction consistency and distance constraint standards. Combined with the crack position data in the previous and next image frames, the temporal continuity and spatial position consistency of the crack boundary are corrected. All crack boundary re-estimation results are subjected to consistency verification to check for abnormal boundary extension, excessive positional offset, abrupt width changes, or incorrect connection with the main structure. For crack boundaries that do not meet the verification criteria, rollback and readjustment operations are performed, and finally, the corrected crack boundary identification results are output.

6. The method for detecting cracks in engineering structures based on image processing technology according to claim 1, characterized in that, Step S500 includes: Based on the crack boundary re-estimation results, the spatial position correction values ​​of crack boundaries in the image sequence are statistically analyzed. Image frames with inconsistent boundary position changes and large spatial offsets between consecutive image frames are identified, and their shooting time points are extracted. By comparing the shooting time with the constructed illumination baseline, analyzing the trajectory of the reflector and the fluctuation of sky brightness, the period of time with gentle brightness change is selected as the preferred shooting time. Determine the angle between the current imaging direction and the reflection path, adjust the observation direction of the equipment and the rotation angle of the polarizer to avoid areas with high reflection rates, and improve the contrast of the crack boundary; Based on the difference in grayscale between the crack boundary and the background, the exposure time and exposure compensation value are dynamically adjusted. After the exposure time point, polarization angle parameters and observation angle are set, a new round of image acquisition is performed.

7. The method for detecting cracks in engineering structures based on image processing technology according to claim 1, characterized in that, Step S600 includes: Based on the brightness correction residual and boundary position drift rate of the crack region in the acquired image, a feedback dataset for the current image processing round is constructed. The constructed illumination baseline was updated using this feedback dataset to correct the original brightness curve and adjust the stability indicator for the image acquisition period. On the updated illumination baseline, the brightness fluctuation sensitivity field is reconstructed, and the illumination response fluctuation score of each pixel is calculated by combining the radiation residual. Based on the updated sensitivity field, points with severe drift are removed, and areas with stable grayscale changes and clear boundary structures are selected as a new set of edge anchor points for use in the next round of crack detection.

8. An engineering structure crack detection system based on image processing technology, used to implement the engineering structure crack detection method based on image processing technology according to any one of claims 1-7, characterized in that, It includes a lighting modeling module, an anchor point extraction and boundary calibration module, a temporal radiometric calibration module, a boundary reestimation module, a closed-loop control module, and a model update module. The illumination modeling module collects image data at multiple time points in the engineering structure monitoring area, obtains the corresponding sky brightness change curve and reflector reflection angle change sequence, generates a time-varying illumination profile and reflector trajectory map covering the monitoring period, and constructs an illumination baseline for subsequent time series reference. The anchor point extraction and boundary calibration module derives the brightness fluctuation sensitivity field based on the illumination baseline, selects regions with stable brightness response in the sensitivity field as the edge anchor point set, and uses this edge anchor point set to perform initial position calibration of the crack candidate boundary. The temporal radiometric calibration module, based on the set of edge anchor points, inverts the exposure drift and reflected light intensity flicker characteristics in the image sequence, constructs a temporal radiometric calibrator, performs brightness unification processing on images at multiple time points, and generates a continuous image sequence with consistent brightness characteristics. The boundary re-estimation module extracts crack delineation information with constant illumination from a continuous image sequence with uniform brightness, and re-estimates the position of the crack boundary based on the set of edge anchor points to correct the boundary position offset caused by illumination changes. The closed-loop control module uses the re-estimation result of the crack boundary as the control signal to form a closed-loop acquisition control strategy. By adjusting the exposure time point, polarization angle parameter and observation angle, it can actively suppress the interference of subsequent illumination flicker during the image acquisition process. The model update module, after executing the closed-loop acquisition control strategy, feeds back the current round of radiation calibration residuals and boundary drift rates to the illumination baseline, and dynamically updates the brightness fluctuation sensitivity field and edge anchor point set based on the updated illumination baseline, enabling the crack detection system to have the ability to learn and continuously optimize accuracy.

Images