Dike inspection robot environment detection method and system

Abstract

The invention relates to the technical field of embankment inspection, and provides an embankment inspection robot environment detection method and system, which can determine a motion blur direction, extract linear features of an image and determine a linear feature direction by acquiring an embankment image and performing motion blur judgment when the image has motion blur. And on the basis, according to the geometrical relationship between the motion blur direction and the linear feature direction, whether the linear feature is the embankment crack or not is judged. By using the motion blur direction as a judgment basis, a pseudo crack and a real crack caused by motion blur can be effectively distinguished, and early-stage potential safety hazard detection omission caused by image quality reduction is avoided, so that the accuracy and the reliability of embankment crack detection are remarkably improved, and the continuity and the effectiveness of an embankment inspection task are guaranteed.

Description

Technical Field

[0001] This application relates to the field of dike inspection technology, and more specifically, to an environmental detection method and system for a dike inspection robot. Background Technology

[0002] In water conservancy project management, daily inspections of dikes are crucial for ensuring flood control safety and structural stability. To improve inspection efficiency and coverage, dike inspection robots incorporating image recognition technology have become an important means of automation. When the dike inspection robot travels from a smooth road surface to a new type of ecological slope protection brick road surface, the grooves of varying depths between its wheels or tracks and the bricks generate periodic, high-frequency impacts and bumps, resulting in slight motion blur in the images captured by the robot's vision system.

[0003] The robot's built-in image quality assessment program typically evaluates each frame of image in real time based on a preset blur level standard. Once the blur level exceeds this standard, the system deems the image "unacceptable" and immediately triggers a "stop-retake" command. However, because the root cause of motion blur—the high-frequency bumps from the new road surface—is persistent, even if the robot stops and attempts to retake the image, the newly acquired image is highly likely to still be unacceptable. This repeated "stop-retake" cycle makes it almost impossible for the robot to move effectively on that section of road, resulting in prolonged periods of stagnation and severely impacting the continuity and efficiency of the inspection task.

[0004] After the embankment renovation was completed and put into use for a period of time, the new ecological slope protection bricks, under the combined effects of long-term water erosion and repeated thermal expansion and contraction stress caused by temperature differences, are prone to developing extremely narrow, hairline-like micro-cracks at the edges of the bricks. Due to their extremely small width and concealed location (deeply hidden in the shadows of the grooves), these cracks are very difficult to detect during routine visual inspections. When the image is blurred, the background image analysis program struggles to accurately identify these extremely narrow and uniquely shaped new hairline cracks located within the shadow areas of the slope protection brick grooves. Although the system log records poor image quality for this section, this record is merely a general quality description and cannot reveal the missed detection of specific early safety hazards caused by the decline in image quality.

[0005] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0006] This application discloses an environmental detection method for a dike inspection robot, which aims to solve the problem that the image quality of the dike inspection robot is reduced due to motion blur in complex road environments, which in turn affects the accuracy of dike crack identification and the inability to effectively assess the risk of road surface anomalies and cracks.

[0007] The technical solution of this application is as follows: In a first aspect, this application discloses an environmental detection method for a dike inspection robot, the method comprising: Acquire an image of the dike and determine motion blur; the dike image is an image of the dike surface. In response to the presence of motion blur in the embankment image, the direction of motion blur is determined, and the linear features of the embankment image are extracted and their directions are determined. Based on the geometric relationship between the motion blur direction and the linear feature direction, determine whether the position corresponding to the linear feature is detected as a crack in the embankment.

[0008] Furthermore, based on the above method, in response to the existence of motion blur in the levee image, the direction of motion blur is determined, including: A two-dimensional Fourier transform is performed on the image of the dike to obtain the frequency domain Fourier spectrum; Detect the dark lines formed by periodic zeros caused by motion blur in the Fourier spectrum of this frequency domain; The direction of motion blur in the embankment image is deduced from the direction of the dark line; wherein the direction of motion blur is perpendicular to the direction of the dark line.

[0009] In some preferred embodiments, the levee image has multiple linear features; based on the geometric relationship between the motion blur direction and the linear feature direction, determining whether the location corresponding to the linear feature is detected as a levee crack includes: Determine the angle between the direction of each linear feature and the direction of motion blur. If the angle is within a preset range, determine whether the position corresponding to the linear feature is detected as a crack in the embankment.

[0010] Based on the above, this application further proposes that the dike inspection robot includes an inertial measurement unit (IMU); Acquire images of the dike and perform motion blur detection, including: The robot continuously acquires images of the levee at a first frequency, and simultaneously records the inertial measurement information of the levee inspection robot at a second frequency while acquiring the levee images, based on the inertial measurement unit (IMU). The inertial measurement information includes triaxial linear acceleration and triaxial angular velocity. For each frame of the embankment image, the variance of the Laplacian operator is calculated. If it is lower than a preset threshold, it is determined to be a blurred image. The blur type is determined based on the inertial measurement information synchronized with the time of the embankment image. The blur type includes single linear blur and composite motion blur.

[0011] More specifically, in some implementations, the method further includes: if the fuzziness type is a composite motion fuzziness, calculating its composite motion index based on the inertial measurement information; and storing the composite motion index in association with the geographical location corresponding to the current levee image. The inspection path of the dike inspection robot is divided into multiple segments, and the composite motion index is accumulated for each segment. When the composite motion index of a road section is higher than the abnormal threshold in multiple inspections, or when the average composite motion index of the road section shows a continuous upward trend, it is determined that there is a potential road surface anomaly in the road section.

[0012] In one implementation, determining whether the location corresponding to the linear feature is detected as a levee crack based on the geometric relationship between the motion blur direction and the linear feature direction includes: Acquire a continuous sequence of dike images, determine whether the sequence contains motion blur, and if motion blur is present, align two frames in the sequence. Calculate the pixel difference between the two aligned images, and determine whether the linear feature is a crack based on the pixel difference.

[0013] To enhance functionality, determining whether the location corresponding to the linear feature is detected as a levee crack includes: Based on the size, location, and historical development trend of the first linear feature identified as a crack, a risk assessment is conducted on the first linear feature. Based on the risk assessment results, the first linear feature is prioritized. High-risk crack information after priority sorting is highlighted and an immediate alarm is triggered.

[0014] Based on the above, this application further proposes to conduct a risk assessment of the first linear feature identified as a crack, based on its size, location, and historical development trend. This includes: retrieving historical inspection data from a historical database that is adjacent to the location of the first linear feature identified as a crack, and obtaining historical crack features; comparing the similarity between the first linear feature and the historical crack features; performing crack matching based on the similarity and the size and direction of the current crack; identifying the current crack as the crack corresponding to the first linear feature; updating the historical development trend of the crack based on the crack matching result; and conducting a risk assessment based on the updated historical development trend, the size of the crack, and the location of the crack.

[0015] In one implementation, a risk assessment is performed based on the updated historical trend, the size of the crack, and the location of the crack, including: determining the initial risk level of the current crack based on the current crack size, location, and the updated historical trend data; acquiring the current environmental parameters of the dike inspection robot when the initial risk level is at a critical state; correcting the initial risk level based on the environmental parameters to obtain a corrected risk level; and performing a risk assessment on the current crack based on the corrected risk level.

[0016] Secondly, this application also discloses an environmental detection system for a dike inspection robot. The system includes: a fuzzy judgment module for acquiring a dike image and performing motion fuzzy judgment on the image; the dike image is an image of the dike surface; a direction determination module for determining the motion fuzzy direction in response to the existence of motion fuzz in the dike image, extracting linear features of the dike image, and determining the direction of the linear features; and a detection module for determining whether the position corresponding to the linear feature is detected as a dike crack based on the geometric relationship between the motion fuzzy direction and the direction of the linear features. Beneficial effects

[0017] The environmental detection method for dike inspection robots disclosed in this application acquires dike images and performs motion blur assessment. When motion blur is present in the image, the method determines the direction of the motion blur and extracts linear features from the image, identifying the direction of these linear features. Based on this, the method determines whether the linear features represent dike cracks according to the geometric relationship between the motion blur direction and the linear feature direction. This method effectively solves the problem in existing technologies where motion blur caused by road surface bumps interferes with the identification of novel hairline cracks in dike inspection robots operating in "forced passage" mode. By using the motion blur direction as the criterion, it effectively distinguishes between false cracks and real cracks caused by motion blur, avoiding the omission of early safety hazards due to image quality degradation. This significantly improves the accuracy and reliability of dike crack detection, ensuring the continuity and effectiveness of dike inspection tasks. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the steps of the environmental detection method for a dike inspection robot disclosed in an embodiment of the present invention; Figure 2This is a schematic diagram of the environmental monitoring system of the dike inspection robot disclosed in an embodiment of the present invention. Detailed Implementation

[0020] The implementation details of the technical solution in this embodiment are described in detail below: This application proposes an environmental detection method for a dike inspection robot, such as... Figure 1 As shown, the method includes: S101, acquire an image of the dike and perform motion blur determination; the image of the dike is an image of the surface of the dike; S102, in response to the existence of motion blur in the embankment image, the direction of motion blur is determined, and the linear features of the embankment image are extracted and the direction of the linear features is determined; S103, based on the geometric relationship between the motion blur direction and the linear feature direction, determine whether the position corresponding to the linear feature is detected as a levee crack.

[0021] This application determines dike cracks by performing motion blur detection on dike images and, when motion blur is present, by combining the geometric relationship between the motion blur direction and the linear feature direction. This effectively solves the problem that traditional methods are difficult to accurately identify dike cracks in motion-blurred images, thus improving the accuracy and efficiency of dike inspection.

[0022] The environmental detection method for dike inspection robots proposed in this application aims to address the impact of image motion blur caused by robot movement or environmental factors on the accuracy of crack detection during dike inspection. Here, "dike image" refers to the surface image data of the dike acquired by the vision sensor mounted on the dike inspection robot; these images form the basis for subsequent analysis and crack detection. "Motion blur judgment" refers to analyzing the acquired dike images to determine whether blurring occurs due to the relative motion between the camera and the target object. When motion blur is present, it is necessary to further "determine the direction of motion blur," i.e., identify the direction of the motion trajectory causing the image blur. Simultaneously, "extracting linear features from the dike image" refers to identifying information with linear structures from the image; these linear features may be potential cracks or other non-crack linear textures. Finally, "determining the direction of linear features" refers to identifying the extension direction of these linear features. By comprehensively analyzing the "geometric relationship" between the direction of motion blur and the direction of linear features, it is possible to more accurately "determine whether the corresponding position of the linear feature is detected as a dike crack," thereby distinguishing between genuine cracks and artifacts caused by motion blur.

[0023] The core of the environmental detection method for the dike inspection robot in this application lies in the accurate identification and judgment of linear features in motion-blurred images.

[0024] Several methods can be employed to acquire images of dikes and determine motion blur. For example, a dike inspection robot can periodically capture images of the dike surface using its onboard camera. After image acquisition, each frame can be analyzed in real time, for instance, by calculating image sharpness, contrast, or specific frequency components to assess its clarity. When the image's clarity index falls below a preset threshold, it can be preliminarily determined that the image is blurry. Another approach is to equip the robot with an image quality assessment module. This module, while acquiring images, uses image processing algorithms (such as the Laplacian operator and gradient magnitude) to quantify the blur. When the quantification value exceeds a specific threshold, the system determines that the image exhibits motion blur.

[0025] In response to motion blur in the levee image, the direction of motion blur is determined, and linear features of the levee image are extracted to determine the direction of these linear features. When the system determines that the levee image has motion blur, it needs to further determine the direction of the blur. One approach is to infer the direction of motion blur by analyzing the frequency domain information of the blurred image. For example, a Fourier transform can be performed on the blurred image to observe the energy distribution in its spectrum. Motion blur usually produces specific dark lines or energy attenuation patterns in the spectrum, and the directions of these patterns have a specific geometric relationship with the direction of motion blur (e.g., perpendicular). By identifying the directions of these patterns, the direction of motion blur can be deduced. Simultaneously, to identify potential levee cracks, linear features need to be extracted from the image. This can be achieved using edge detection algorithms (such as Canny, Sobel, etc.), which can identify regions with drastic changes in pixel intensity in the image, thereby outlining linear structures in the image. For each extracted linear feature, its direction can be further calculated, for example, by fitting a straight line or analyzing its local gradient direction.

[0026] Based on the geometric relationship between the motion blur direction and the linear feature direction, it is determined whether the corresponding position of the linear feature is detected as a crack in the embankment. After obtaining the motion blur direction and the directions of each linear feature, it is necessary to use the geometric relationship between them to determine which linear features are real cracks. One implementation is to calculate the angle between each linear feature direction and the motion blur direction. Since motion blur usually stretches objects in an image along a specific direction, the direction of artifacts caused by motion blur (e.g., stretched textures or noise) is often parallel or nearly parallel to the motion blur direction. However, the direction of a real embankment crack is usually random and does not necessarily have a strong correlation with the motion blur direction. Therefore, a preset angle range can be set. If the angle between the direction of a linear feature and the motion blur direction is outside this range (e.g., nearly perpendicular), then the linear feature is more likely to be a real crack. Conversely, if the angle is within the preset range (e.g., nearly parallel), then the linear feature is more likely to be an artifact caused by motion blur, and thus it is excluded.

[0027] This application's environmental detection method for a dike inspection robot effectively solves the problem of accurately identifying dike cracks in motion-blurred images by introducing the judgment of the image motion blur direction and combining the geometric relationship between the linear feature direction and the motion blur direction. Specifically, when the dike inspection robot travels from a smooth road surface to a new type of ecological slope protection brick road surface, the image becomes motion-blurred due to the bumpy road surface. Traditional methods often simply judge blurred images as unacceptable and trigger retake or forced passage modes, causing potential crack information to be ignored. The method of this application first acquires the dike image and performs motion blur judgment. When motion blur is detected in the image, it no longer simply discards the image, but further analyzes its motion blur direction. At the same time, all possible linear features and their directions are extracted from the blurred image. The key is that this application uses the geometric relationship between the motion blur direction and the linear feature direction for judgment. Since motion blur usually stretches the image content in a specific direction, artifacts or noise caused by blur often extend along the motion blur direction. Genuine levee cracks are formed unrelated to robot motion, and their directions are usually random, not necessarily parallel to the motion blur direction. Therefore, by calculating the angle between the linear feature direction and the motion blur direction, artifacts caused by motion blur and genuine levee cracks can be effectively distinguished. For example, if the direction of a linear feature is almost parallel to the motion blur direction, it is likely an artifact; while if the direction of a linear feature has a large angle with the motion blur direction, it is more likely a genuine levee crack. This method enables the effective identification of novel, extremely narrow, and uniquely shaped hairline cracks even in motion-blurred images, avoiding detection omissions due to poor image quality.

[0028] This application further proposes that, in response to the existence of motion blur in the embankment image, the step of determining the direction of motion blur includes: Perform a two-dimensional Fourier transform on the embankment image to obtain the frequency domain Fourier spectrum; Detect the dark lines formed by periodic zeros caused by motion blur in the frequency domain Fourier spectrum; The motion blur direction of the embankment image is deduced from the direction of the dark line; wherein the motion blur direction is perpendicular to the direction of the dark line.

[0029] Specifically, the two-dimensional Fourier transform is a mathematical tool that converts an image from the spatial domain to the frequency domain, aiming to reveal the frequency components of the image. By performing a two-dimensional Fourier transform on the embankment image, its frequency domain Fourier spectrum can be obtained, which can intuitively show the distribution of different frequency components in the image. When the image has linear motion blur, its frequency domain Fourier spectrum will produce a series of periodic zeros due to the motion blur. These zeros typically appear as a series of parallel and equidistant dark lines in the frequency domain. These dark lines are the unique "fingerprint" of motion blur in the frequency domain, and their direction has a specific geometric relationship with the direction of the motion blur. Detecting these dark lines can be achieved through various image processing techniques, such as thresholding the Fourier spectrum, edge detection, or Hough transform to identify these linear structures.

[0030] In practical applications, once these dark lines are detected, the motion blur direction of the embankment image can be deduced from their direction. For example, the motion blur direction is perpendicular to the direction of the dark line. This means that if the dark line is horizontal, the motion blur direction is vertical; if the dark line is at a 45-degree angle, the motion blur direction is at a 135-degree angle. This inverse deduction mechanism utilizes the properties of the Fourier transform to accurately recover the spatial domain motion blur features from the frequency domain information.

[0031] This application's solution transforms the embankment image from the spatial domain to the frequency domain. Utilizing the dark lines formed by the unique periodic zeros generated by motion blur in the frequency domain, it can identify the direction of motion blur in an objective and quantitative manner. Traditional methods for determining the direction of motion blur may rely on local image features or statistical information, making them susceptible to noise or the complexity of image content. However, through Fourier transform, the global characteristics of motion blur are concentrated in the dark lines of the frequency domain, making the detection of motion blur direction more robust and accurate. It is precisely because of this characteristic of frequency domain analysis that the determination of motion blur direction is no longer limited by local interference in the spatial domain, thus providing a more reliable foundation for subsequent linear feature analysis.

[0032] In some preferred embodiments, suppose a dike inspection robot captures an image of the dike surface, which exhibits a degree of linear motion blur due to slight shaking during the robot's movement. To accurately determine the direction of this motion blur, a two-dimensional Fourier transform is first performed on the blurred dike image. In the resulting frequency domain Fourier spectrum, a series of clear, parallel dark lines are observed. These dark lines are caused by linear motion blur in a specific direction. For example, if these dark lines appear approximately horizontal in the frequency domain, then according to the principles of this application, the actual motion blur direction can be deduced to be vertical. In this way, even when the blur direction is difficult to determine precisely by the naked eye, the accurate motion blur direction can be obtained through frequency domain analysis, providing an accurate correction basis for subsequent crack detection.

[0033] Traditional environmental detection methods for levee inspection robots, after acquiring levee images and performing motion blur assessment, determine whether the corresponding locations of linear features are levee cracks based on the geometric relationship between the motion blur direction and the linear feature direction. However, in practical applications, levee images may contain multiple linear features, some of which may be real cracks, while others may be artifacts caused by the motion blur itself. Relying solely on generalized geometric relationships may make it difficult to accurately distinguish these linear features, leading to false positives or false negatives and affecting the accuracy of crack detection.

[0034] In response, this application further proposes that the linear features of the aforementioned levee image are multiple; the determination of whether the position corresponding to the linear feature is detected as a levee crack based on the geometric relationship between the motion blur direction and the linear feature direction includes: determining the angle between each linear feature direction and the motion blur direction; if the angle is within a preset range, then determining whether the position corresponding to the linear feature is detected as a levee crack.

[0035] Specifically, the linear features of the levee image can be understood as straight or curved segments of a certain length and direction extracted from the levee image using image processing algorithms (such as edge detection, Hough transform, etc.). These linear features may represent the texture and structural edges of the levee surface, or they may represent potential cracks or stripes caused by motion blur. Determining the angle between the direction of each linear feature and the direction of motion blur refers to calculating the angle difference between the principal direction of each detected linear feature and the overall motion blur direction of the image determined by methods such as Fourier transform. This angle reflects the relative orientation of the linear feature and the direction of motion blur. In practical applications, the preset range is an angle interval set based on experience or experimental data. The purpose of setting this range is to distinguish linear artifacts caused by motion blur from real levee cracks. For example, if motion blur typically produces linear stripes parallel or perpendicular to the blur direction in the image, then the preset range can be set to exclude these angles, thereby identifying linear features at specific non-parallel or non-perpendicular angles to the blur direction, which are more likely to be real cracks. The aim is to provide a clear discrimination criterion to improve the accuracy of crack detection.

[0036] This application's solution addresses the challenge of effectively distinguishing between real cracks and motion blur artifacts when multiple linear features are present by introducing precise judgment of the angle between the direction of each linear feature and the direction of motion blur. Motion blur typically introduces linear stripes or artifacts in an image that align with the blur direction. Real levee cracks, as defects in the levee structure itself, do not necessarily have a strong correlation between their direction and the movement direction of the inspection robot (and consequently, the direction of motion blur). Therefore, by calculating the angle between each linear feature and the direction of motion blur and comparing it to a preset range, linear artifacts highly correlated with the direction of motion blur (e.g., parallel or perpendicular) can be effectively filtered out, highlighting linear features with weaker correlation to the direction of motion blur—features more likely to be real levee cracks. Thus, this solution improves the accuracy and reliability of crack detection based on more refined geometric relationships.

[0037] Through the above technical solution, this application can effectively distinguish between linear artifacts caused by motion blur and real levee cracks in levee images, especially when multiple linear features exist in the image. This method, by introducing precise angle judgment and preset range filtering, significantly reduces the interference of motion blur on crack detection, reduces the false alarm rate, and thus improves the accuracy and reliability of levee crack detection. This enables levee inspection robots to more accurately identify and locate potential levee safety hazards, providing more reliable data support for subsequent maintenance and management.

[0038] In some preferred embodiments, a specific example is given below. Suppose that during a dike inspection, the captured dike image exhibits horizontal motion blur due to camera shake or robot movement. Using methods such as two-dimensional Fourier transform, the direction of the motion blur can be determined to be horizontal (e.g., 0 degrees). At this point, multiple linear features may be detected in the image, including a real vertical crack (direction approximately 90 degrees) and some horizontal stripes caused by the horizontal motion blur (direction approximately 0 or 180 degrees). According to the scheme of this application, the direction of each linear feature is first determined. For example, the linear feature direction of the vertical crack is 90 degrees, and the linear feature direction of the horizontal stripes is 0 degrees. Next, the angle between each linear feature direction and the motion blur direction (0 degrees) is calculated. For the vertical crack, the angle is 90 degrees; for the horizontal stripes, the angle is 0 degrees. A preset range can be set, such as [60 degrees, 120 degrees], which aims to identify linear features significantly different from the motion blur direction. The calculated included angle is compared with the preset range: the included angle of the vertical crack is 90 degrees, which falls within the preset range [60 degrees, 120 degrees]. Therefore, this linear feature is identified as a levee crack. The included angle of the horizontal stripe is 0 degrees, which does not fall within the preset range [60 degrees, 120 degrees]. Therefore, this linear feature is excluded and not detected as a levee crack. In this way, the solution of this application can effectively distinguish real levee cracks from artifacts caused by motion blur, ensuring the accuracy of the detection results.

[0039] In some embodiments described above, this application proposes acquiring images of dikes and determining motion blur. However, in practical applications, relying solely on the visual features of the image itself to determine motion blur may not accurately distinguish between different types of blur, such as linear blur in a single direction and composite motion blur caused by the superposition of multiple motions. Misjudgment or insufficient information regarding this type of blur may lead to decreased accuracy in subsequent image processing and crack detection, or even false alarms or missed detections.

[0040] In response, this application further proposes an environmental detection method for a dike inspection robot, wherein the dike inspection robot includes an inertial measurement unit (IMU). Acquire images of the dike and perform motion blur detection, including: The system continuously acquires images of the levee at a first frequency, and simultaneously records the inertial measurement information of the levee inspection robot at a second frequency based on the inertial measurement unit (IMU) while acquiring the levee images; the inertial measurement information includes triaxial linear acceleration and triaxial angular velocity. For each frame of the embankment image, the variance of the Laplacian operator is calculated. If it is lower than a preset threshold, it is determined to be a blurred image. The blur type is determined based on the inertial measurement information synchronized with the time of the embankment image. The blur type includes single linear blur and composite motion blur.

[0041] Specifically, the aforementioned dike inspection robot is equipped with an inertial measurement unit (IMU) for real-time sensing and recording of the robot's motion. During the inspection, the robot continuously acquires images of the dike surface at a first frequency, for example, 30 frames per second. Simultaneously, the IMU synchronously records the robot's inertial measurement information during image acquisition at a second frequency. This second frequency can be the same as or a multiple of the first frequency to ensure time synchronization between the inertial data and the image data. The inertial measurement information typically includes the robot's linear acceleration along the X, Y, and Z axes and its angular velocity around these axes, comprehensively reflecting the robot's translational and rotational motion.

[0042] During the inspection process, the levee inspection robot's vision system (e.g., a Baslerac A2040-90gc camera equipped with a CMOS sensor) continuously acquires images of the levee surface. Simultaneously, the robot's internal inertial measurement unit (IMU, e.g., the ADIS16460 series, providing three-axis linear acceleration and three-axis angular velocity data) synchronously records the vehicle's motion data at a high frequency (e.g., 200Hz) at the moment of image acquisition. These images and synchronized motion data are transmitted to the robot's internal image processing unit (e.g., an embedded computer equipped with an NVIDIA Jetson Xavier NX module). The image processing unit precisely timestamps each frame and matches it with the timestamps in the IMU data stream, ensuring a high degree of temporal consistency between the images and the corresponding vehicle motion data. This can be achieved through hardware synchronization triggering mechanisms, such as ensuring camera exposure and IMU data sampling occur within the same clock cycle, or through software interpolation to align the timestamps, ensuring a timestamp difference of less than 1 millisecond.

[0043] The image processing unit first performs a preliminary blur assessment on each received image frame. If an image is determined to be blurry (for example, by calculating that the variance of the Laplacian operator of the image is below a preset threshold), the system further combines synchronized IMU data to identify and quantify the type of image blur.

[0044] In this embodiment, the blur type determination includes: the system analyzes the IMU data synchronized with the blurred image. If only the instantaneous peak value of the Z-axis linear acceleration (vertical swaying) exceeds a preset threshold (e.g., |Az| > 0.5g), while the peak values ​​of the Y-axis linear acceleration (lateral sliding) and the angular velocity around the Z-axis (vehicle body torsion) are both below their respective thresholds (e.g., |Ay| < 0.1g, |Gz| < 5 deg / s), then it is determined to be a single linear blur.

[0045] Composite fuzzy quantization: If the Z-axis acceleration exceeds the limit while the Y-axis acceleration or angular velocity around the Z-axis also significantly exceeds the limit (e.g., |Ay| > 0.1g or |Gz| > 5 deg / s), it is determined to be composite motion fuzziness. In this case, the system will further quantify the "complexity" of the composite fuzziness. For example, a "composite motion index" (CMI) can be calculated as follows: CMI = sqrt( (Ay_peak / Ay_threshold) 2 + (Gz_peak / Gz_threshold) 2 ); Here, Ay_peak and Gz_peak are the peak values ​​of the Y-axis acceleration and Z-axis angular velocity at the instant of image acquisition, respectively, while Ay_threshold and Gz_threshold are preset judgment thresholds. The larger the CMI value, the more intense the composite motion and the higher the degree of composite image blur.

[0046] This application's solution, by introducing an inertial measurement unit (IMU) and combining it with the image's own blur assessment, can more comprehensively and accurately evaluate the motion blur of an image. Traditional image blur assessment methods, such as Laplacian variance, can only determine whether an image is blurry, but cannot provide information on the cause and type of blur. However, by synchronously recording inertial measurement information, such as three-axis linear acceleration and three-axis angular velocity, the actual motion state of the dike inspection robot at the moment of image acquisition can be objectively reflected. When an image is determined to be blurry, the system can combine IMU data to analyze the motion pattern causing the blur. For example, if IMU data shows that the robot mainly exhibits linear motion along a certain direction, the image may show single linear blur; if IMU data shows that the robot also exhibits significant translational and rotational motion, the image is more likely to show composite motion blur. This judgment method, which combines image features and motion data, makes the identification of blur types more accurate, providing crucial input information for subsequent image deblurring processing or adjustments to crack detection algorithms.

[0047] In some preferred embodiments, it is assumed that during the inspection process, the levee inspection robot's onboard camera continuously acquires levee images at a frequency of 30 frames per second. Simultaneously, the robot's internal inertial measurement unit (IMU) synchronously records the robot's three-axis linear acceleration and three-axis angular velocity data at a frequency of 100 times per second. When a frame of the levee image is acquired, the system first calculates the Laplacian variance of that image. If this variance is lower than a preset threshold (e.g., set to 100), the image is preliminarily determined to be blurry. At this point, the system retrieves the IMU data closest to the timestamp of the blurry image.

[0048] For example, if IMU data shows that at the moment of image acquisition, the robot exhibits a large linear acceleration primarily along its forward direction, while its angular velocity is close to zero, then the system can determine that the blur is a single linear blur. This type of blur can usually be processed using a simple linear deblurring algorithm.

[0049] On the other hand, if IMU data shows that at the moment of image acquisition, the robot not only exhibits linear acceleration but also a significant angular velocity around its vertical axis (e.g., the robot is turning), then the system will classify the blur as a complex motion blur. For complex motion blur, more complex deblurring algorithms may be required, or the algorithm parameters may need to be adjusted more significantly during crack detection to accommodate more complex blur patterns in the image. In this way, the system can accurately classify image blur based on the actual motion state, thus providing more targeted strategies for subsequent image processing and crack detection.

[0050] In some embodiments described above, the levee inspection robot can determine the blur type of the acquired levee image based on inertial measurement information recorded by the inertial measurement unit (IMU), including single linear blur and composite motion blur. However, merely identifying the type of composite motion blur cannot directly provide in-depth information about the levee road surface condition, nor can it proactively warn of potential road surface anomalies, potentially leading to reduced inspection efficiency or failure to detect safety hazards in a timely manner. Therefore, this application further proposes a scheme to evaluate road surface anomalies using composite motion blur information, aiming to achieve early warning of potential road surface anomalies along the levee inspection path through the accumulation and analysis of composite motion indices.

[0051] If the blur type is a composite motion blur, calculate its composite motion index based on the inertial measurement information; store the composite motion index and the geographical location corresponding to the current embankment image. The inspection path of the dike inspection robot is divided into multiple segments, and the composite motion index is accumulated for each segment. When the composite motion index appears to be higher than the abnormal threshold in multiple inspections of a road segment, or when the average composite motion index of the road segment shows a continuous upward trend, it is determined that there is a potential road surface anomaly in the road segment.

[0052] In this embodiment, when a dike image is determined to have complex motion blur, the dike inspection robot uses inertial measurement unit (IMU) data, including triaxial linear acceleration and triaxial angular velocity, to calculate a complex motion index characterizing the complexity of the complex motion. This complex motion index can be understood as a quantification of the robot's motion state instability or complexity. For example, it can be obtained by comprehensively calculating the variance, standard deviation, and spectral energy distribution of inertial measurement information within a specific time window. The calculated complex motion index is then associated with the geographical location information corresponding to the current dike image and stored in a database for subsequent querying and analysis.

[0053] The inspection path of the levee inspection robot is pre-divided or dynamically into multiple discrete segments. Each segment can have a preset length, such as 50 meters, 100 meters, or be divided according to geographical features. For each segment, the system continuously accumulates the composite motion index corresponding to all levee images collected within that segment. The accumulation method can include calculating the average, maximum, median, or performing a weighted sum to reflect the overall motion blur of the segment.

[0054] For example, for an image identified as having complex motion blur, the system correlates its quantified Complex Motion Index (CMI) with the precise geographic location information of the current image (obtained through the robot's GPS module and odometry data) and records this data in local storage (e.g., a solid-state drive or flash memory). The system continuously monitors the accumulation of CMI within the same geographic area.

[0055] Data storage: The data structure of the records can be a key-value database, where the key is the geographic coordinates (e.g., latitude and longitude accurate to 6 decimal places), and the values ​​are a list containing the CMI value and timestamp recorded for each inspection at that location. For example, it can be stored as { "Longitude, Latitude": [ { "CMI": 2.1, "Timestamp": "2023-10-26T10:00:00"},... ]}.

[0056] Regional Division and Cumulative Judgment: The levee inspection path can be logically divided into a series of small geographical regions (e.g., regions every 5 meters, or grid division based on GPS coordinates). The system maintains a CMI history record for each region. When the robot inspects a region multiple times consecutively (e.g., more than 3 times), and the CMI value recorded for each inspection is higher than the preset "anomaly threshold" (e.g., CMI > 1.5), or the average CMI of the region shows a continuous upward trend over a period of time (e.g., one month) (e.g., a slope greater than 0.1 through linear regression analysis), the system determines that there is a potential road surface anomaly pattern in that region.

[0057] When a geographic area is identified as having a potential road surface anomaly pattern, the system will generate a diagnostic report, highlighting the possibility of localized deformation or settlement of the road surface in that area, and send it to the management personnel.

[0058] The report will include the precise geographic coordinates of the abnormal area, the average or trend of the detected CMI (Complex Motion Index), the time when the anomaly was first detected, and a suggestion for managers to conduct on-site verification. The report may also include a recent, blurry image of the area (even if blurry, it serves as evidence) and a historical CMI trend chart. The report can be uploaded in real-time to a remote management platform via a wireless communication module (e.g., a 5G cellular module) or sent to designated personnel via email / SMS. Furthermore, different alert levels can be set based on the cumulative value or upward trend of the CMI (e.g., a CMI average of 1.5-2.0 is "mild concern," 2.0-3.0 is "moderate alert," and greater than 3.0 is "high alert"), and these levels can be distinguished by different colors or icons in the report to assist managers in prioritizing tasks.

[0059] Furthermore, in practical applications, after multiple inspections of a specific road segment, if the cumulative composite motion index of each inspection within that segment consistently exceeds a preset anomaly threshold, or if the average composite motion index of that segment shows a continuous upward trend, then that segment will be identified as having potential road surface anomalies. The anomaly threshold can be determined based on historical data, empirical values, or through experimental calibration. For example, when the composite motion index exceeds a certain value, it indicates that the robot's movement in that segment is abnormally drastic. A continuous upward trend can be determined by comparing the average composite motion index of several consecutive inspections; for example, if the average of three consecutive inspections is higher than the previous one, and the increase exceeds a certain percentage. Potential road surface anomalies can refer to areas such as uneven surfaces, potholes, areas with dense cracks, and areas of settlement that may affect the robot's stable movement or indicate structural problems with the embankment.

[0060] The solution proposed in this application is effective in detecting potential road surface anomalies because the motion state of the embankment inspection robot becomes more complex and unstable when driving on uneven or damaged roads, making the acquired images more prone to complex motion blur. Since complex motion blur is caused by the superposition of multiple motion components (such as translation, rotation, and vibration), the corresponding inertial measurement information exhibits higher complexity and volatility. This complex motion state can be quantified by calculating the complex motion index. When the complex motion index of a specific road segment is consistently high or shows an upward trend, it directly reflects that the road surface condition of that segment is deteriorating, leading to a decrease in the robot's driving stability in that area, and consequently causing more frequent or severe complex motion blur. Therefore, through long-term monitoring and trend analysis of the complex motion index, changes in road surface conditions can be indirectly and effectively reflected, thereby achieving early identification of potential road surface anomalies.

[0061] Through the above technical solution, this application overcomes the limitation of merely identifying fuzzy types without providing in-depth information on road surface conditions. By introducing the calculation, associated storage, and cumulative analysis based on road segments, intelligent early warning of potential road surface anomalies along the embankment inspection route is achieved. This enables embankment management personnel to obtain timely quantitative assessments of road surface conditions, thereby proactively planning maintenance work, avoiding equipment damage or decreased inspection data quality due to road surface problems, significantly improving the intelligence level and efficiency of embankment inspection, and ensuring the safety and stability of the embankment structure.

[0062] In some preferred embodiments, a specific example is given below. Assume a levee inspection robot is conducting regular inspections along a 10-kilometer-long levee. The inspection path is automatically divided into 100 segments, each 100 meters long. During one inspection, when the robot reaches segment 25, its inertial measurement unit (IMU) records severe, non-linear fluctuations in triaxial linear acceleration and triaxial angular velocity data, causing the levee image acquired for that segment to be classified as having complex motion blur. At this point, the system calculates the complex motion index of that segment as 0.85 based on this inertial measurement information and stores it along with the geographical location information of segment 25 (e.g., latitude and longitude range).

[0063] In subsequent inspections, the system continuously accumulated and analyzed the composite motion index of road segment 25. For example, in the first inspection, the average composite motion index of this segment was 0.85; in the second inspection, the average rose to 0.92; and in the third inspection, the average further increased to 1.05. Assuming a preset anomaly threshold of 0.90, and the system is set to trigger an alert when the average of two consecutive inspections is higher than the anomaly threshold, or when the average shows a continuous upward trend, in this example, the average composite motion index of road segment 25 not only continuously exceeded the anomaly threshold but also showed a continuous upward trend. Based on this, the system immediately determined that there was a potential road surface anomaly in road segment 25 and triggered an immediate alarm, while simultaneously highlighting the geographical location information of the road segment, prompting maintenance personnel to conduct on-site inspection and repair. In this way, early warning and intervention can be achieved before road surface problems worsen, effectively reducing maintenance costs and risks.

[0064] This application further proposes a more robust method for detecting cracks in dikes, which improves the accuracy of detection by analyzing a series of consecutive dike images.

[0065] According to the above-mentioned environmental detection method for levee inspection robots, based on the geometric relationship between the motion fuzzy direction and the linear feature direction, it is determined whether the location corresponding to the linear feature is detected as a levee crack, including: Acquire a continuous sequence of dike images, determine whether the dike image sequence has motion blur, and when the dike image sequence has motion blur, align two frames in the dike image sequence. Calculate the pixel difference between the two aligned images, and determine whether the linear feature is a crack based on the pixel difference.

[0066] Specifically, acquiring a continuous sequence of levee images refers to the process by which a levee inspection robot continuously acquires images of the levee surface at a certain frame rate during inspection, forming a time-series image data stream. This image sequence can consist of multiple consecutive image frames, for example, acquiring 30 frames per second. Determining whether motion blur exists in the levee image sequence can employ various image quality assessment algorithms. For example, calculating the variance of the Laplacian operator for each frame in the sequence; if it is lower than a preset threshold, the frame is determined to have motion blur. When motion-blurred image frames exist in the levee image sequence, it is necessary to align the two frames. The purpose of image alignment is to eliminate or reduce the relative displacement between images caused by robot movement, camera shake, etc., so that the pixel positions of the same scene point in different image frames are as consistent as possible. Alignment methods can include feature-point-based matching (e.g., SIFT, SURF, ORB feature-point matching algorithms) or pixel-based registration (e.g., optical flow, phase correlation methods). The pixel difference between the two aligned frames refers to the difference in grayscale or color values ​​at the same pixel position in the two aligned frames. For example, the absolute or squared difference of corresponding pixels in two frames can be calculated. Based on this pixel difference, it can be determined whether the linear feature is a crack. The principle is that a real crack in a levee is an inherent physical structure of the levee surface. In a continuous image sequence, even with motion blur, the shape and position of the crack should remain relatively stable after image alignment. However, image artifacts or noise caused by motion blur may exhibit inconsistent pixel difference patterns between different frames. Therefore, by analyzing the pixel differences in the aligned images, it is possible to effectively distinguish between real cracks and blur artifacts. For example, if a linear feature exhibits a continuous and stable pixel difference pattern in two aligned frames, it is more likely to be identified as a crack.

[0067] The proposed solution acquires a continuous sequence of dike images and aligns images with motion blur, effectively compensating for image displacement caused by robot motion. This, to some extent, eliminates the interference of motion blur on the determination of the position and shape of linear features in the image. Furthermore, by calculating the pixel differences between the aligned two frames, the stability of cracks as fixed physical features of the dike surface can be utilized to distinguish them from instantaneous pixel changes caused by motion blur. Real cracks exhibit a relatively stable pixel difference pattern in the aligned image, while blur artifacts may show random or unstable differences. This method, based on sequence image analysis and pixel difference determination, significantly improves the accuracy and robustness of dike crack detection in motion-blurred environments.

[0068] In some preferred embodiments, the dike inspection robot continuously acquires a sequence of dike images at a rate of 25 frames per second during the inspection process. When the system detects motion blur in the image sequence, for example, by calculating the variance of two consecutive frames using the Laplacian operator and finding that both are below a preset threshold, image alignment and pixel difference analysis are triggered. Specifically, the system selects two adjacent frames in the blurred image sequence, such as frame N and frame N+1. First, the two frames are precisely aligned using a SIFT-based feature point matching algorithm to eliminate minor displacements and jitters of the robot during acquisition. After alignment, the pixel grayscale value difference between the two frames is calculated. For example, for each pixel (x, y) in the image, the absolute value of the grayscale value difference between frame N and frame N+1 is calculated. Then, the linear feature regions extracted from the image are analyzed. If the pixel difference values ​​within a linear feature region exhibit a consistent and significant pattern across two aligned frames (e.g., the difference values ​​exceed a certain threshold, and the pattern remains consistent across multiple consecutive frame pairs), then the linear feature is identified as a levee crack. Conversely, if the pixel difference pattern is unstable or random, it may be considered an artifact or noise caused by motion blur.

[0069] This application further proposes a method for determining whether the location corresponding to the linear feature is detected as a levee crack, including: A risk assessment is conducted on the first linear feature identified as a crack, based on its size, location, and historical development trend. Based on the risk assessment results, the first linear features are prioritized. The high-risk crack information after priority sorting is highlighted and an immediate alarm is triggered.

[0070] Specifically, the first linear feature refers to the linear structure identified as a crack in the dike using the aforementioned method. Risk assessment involves comprehensively considering the crack's physical properties (such as size, length, width, and depth), its specific location on the dike (such as whether it is located in a critical structural area or near the waterline), and its trend over time (such as whether it is widening, deepening, or extending), to quantify the potential hazard the crack may pose to dike safety. For example, cracks that are larger, located in critical areas, and exhibit a rapid development trend will have a higher risk assessment result.

[0071] Prioritizing the first linear feature based on the risk assessment results can be understood as arranging all detected cracks from high to low risk level according to the quantitative results obtained from the risk assessment. For example, cracks can be divided into different levels such as "high risk," "medium risk," and "low risk," and a corresponding processing priority can be set for each level.

[0072] In practical applications, the high-risk crack information, after being prioritized, is highlighted and triggers an immediate alarm. Specifically, when a crack is assessed as high-risk and assigned a high priority, the system will display its location and detailed information in a prominent manner (such as a red marker or flashing icon) in the inspection report, monitoring interface, or Geographic Information System (GIS). Simultaneously, it will automatically trigger a preset alarm mechanism, such as sending SMS or email notifications to relevant management personnel, or activating an audible and visual alarm, to ensure that high-risk situations are detected promptly and handled urgently. The aim is to ensure that cracks posing the most direct threat to dike safety receive the highest priority and the most rapid response.

[0073] This application's solution effectively addresses the problem of failing to differentiate the severity and urgency of cracks simply by conducting crack detection. Specifically, by comprehensively considering the crack's size, location, and historical development trend, a scientific and quantitative risk assessment can be performed on each crack, avoiding the pitfalls of treating all cracks the same. This refined risk assessment enables subsequent prioritization, ensuring that limited maintenance resources are allocated to the areas most in need. Furthermore, the real-time highlighting and alarm function for high-risk cracks further shortens the time from problem detection to action, significantly improving the response speed and efficiency of dike safety management, thereby effectively reducing potential dike safety risks.

[0074] In some preferred embodiments, a specific example is given below. Suppose that the levee inspection robot detects three linear features using the method described above and determines that they are all levee cracks, which are labeled as crack A, crack B, and crack C, respectively.

[0075] First, the system conducts a risk assessment based on the size, location, and historical development trend of each crack. For example, crack A is located in a critical load-bearing area of ​​the levee's upstream slope, is relatively large, and historical data analysis shows its width has increased by 20% in the past month; crack B is located in a non-critical area of ​​the levee's downstream slope, is relatively small, and its historical development trend is stable; crack C is located on the levee crest, is of medium size, but is close to an important sluice gate structure and shows a slight tendency to extend. Based on this information, the system calculates a risk assessment score for each crack. For example, crack A has the highest risk assessment score (e.g., 90 points, classified as high risk), crack C is next (e.g., 70 points, classified as medium risk), and crack B is the lowest (e.g., 30 points, classified as low risk). Next, the system prioritizes the cracks based on these risk assessment results. Crack A is given the highest priority, followed by crack C, and crack B the lowest. Finally, for crack A, which is identified as high-risk, the system will immediately highlight its location on the electronic map of the dike management platform with a flashing red icon and automatically send an SMS alert to the on-duty engineer's mobile phone, including the detailed location, size, risk level, and suggested emergency handling measures for crack A. For crack C, which is of medium risk, the system may mark it as "requiring close attention" in the daily inspection report and recommend a detailed inspection during the next routine maintenance. For crack B, which is of low risk, it may simply be recorded as a long-term monitoring target. In this way, dike management personnel can quickly identify and prioritize the most pressing threats, thereby effectively ensuring dike safety.

[0076] This application further proposes a method for risk assessment of a first linear feature identified as a crack, based on its size, location, and historical development trend. The method includes: retrieving historical inspection data from a historical database that is adjacent to the location of the first linear feature identified as a crack, and obtaining historical crack features; comparing the similarity between the first linear feature and the historical crack features; performing crack matching based on the similarity and the size and direction of the current crack; wherein the current crack is the crack corresponding to the first linear feature; updating the historical development trend of the crack based on the crack matching result; and performing a risk assessment based on the updated historical development trend, the size of the crack, and the location of the crack.

[0077] Specifically, the system retrieves historical inspection data from a historical database that is adjacent to the location of the first linear feature identified as a crack, and obtains historical crack features. The historical database can be configured to store information on all cracks detected during previous dike inspections, including their geographic coordinates, geometric parameters (e.g., length, width, depth), crack direction, first discovery time, and previous observation records. When the dike inspection robot detects a new first linear feature and initially identifies it as a crack, the system uses the center location of this first linear feature as a reference and retrieves all crack features recorded in historical inspection data within a preset geographical area (e.g., an area with a radius of several meters centered on the first linear feature). This step aims to collect historical records that may correspond to the currently detected crack, laying the data foundation for subsequent accurate comparison and matching.

[0078] Furthermore, the similarity between the first linear feature and the historical crack features is compared. This similarity comparison can be based on multiple dimensions, such as comparing the crack's geometry, size, orientation, texture features, and local contextual information within the image. For example, the similarity can be quantified by calculating metrics such as the distance between the center points of two cracks, the percentage difference in length, the percentage difference in width, and the directional angle. By setting an appropriate similarity threshold, historical crack features with high similarity to the first linear feature can be preliminarily selected, thereby effectively reducing the computational complexity of subsequent matching processes and improving matching accuracy.

[0079] Based on this, crack matching is performed according to the similarity and the current crack's size and orientation; the current crack is the crack corresponding to the first linear feature. In practical applications, crack matching is a crucial step, aiming to determine whether the currently detected first linear feature is the same crack as a crack in the historical database at different points in time. The matching algorithm can comprehensively consider the above similarity indicators and combine them with the latest size and orientation information of the current crack. For example, if the current crack is geographically very close to a historical crack, and its size and orientation are also highly similar and conform to a certain evolutionary pattern, then they can be determined to be the same crack. The matching result will directly provide a reliable basis for subsequent updates to the historical development trend of cracks.

[0080] Therefore, based on the crack matching results, the historical development trend of the crack is updated. Once the current crack successfully matches a crack in the historical database, the latest observation data of the current crack (including its size, location, orientation, etc.) can be used to update the development trend record of that crack in the historical database. For example, the rate of change of crack length and width over time, or the direction and speed of crack expansion can be recorded. This continuous update mechanism ensures that the historical development trend data of the crack is always up-to-date and can accurately reflect the dynamic changes and evolution patterns of the crack.

[0081] Finally, a risk assessment is conducted based on the updated historical development trends, the size of the cracks, and their location. Furthermore, the final stage of the risk assessment comprehensively considers the updated historical development trend data, the current size of the cracks, and their specific location on the dike. For example, a crack that may currently be small but shows a rapid growth trend might be assessed as having a higher risk level than a larger crack that has remained stable over a long period. Simultaneously, the location of the cracks (e.g., whether they are located in critical structural areas of the dike, areas prone to water erosion, or areas susceptible to external influences) is also taken into consideration to affect the risk assessment results. Through this multi-dimensional, dynamic, and comprehensive assessment, a more accurate and reliable crack risk level can be obtained.

[0082] This application's solution effectively addresses the problem of inaccurate acquisition and updating of historical development trend data in traditional risk assessment by introducing historical database retrieval, similarity comparison, and crack matching mechanisms. Specifically, when a new first linear feature is detected, historical inspection data from the historical database is retrieved to obtain historical crack features, providing a comprehensive historical context for subsequent comparison and matching. Subsequently, by comparing the similarity between the first linear feature and historical crack features, the performance of the same potential crack at different time points can be identified. Furthermore, crack matching is performed based on similarity and the current crack size and direction, ensuring accurate identification of the crack and thus enabling precise updating of the crack's historical development trend. It is precisely because of this systematic historical data integration and updating mechanism that risk assessment can be based on a more comprehensive and accurate crack development trajectory, thereby avoiding the bias that may result from judging solely based on the current state or vague historical information.

[0083] As a specific implementation method, a concrete example is given below. Suppose that during a dike inspection, the dike inspection robot detects a new linear feature, which is initially identified as a crack, referred to as the first linear feature. This first linear feature is located at a specific coordinate point on the dike and has a certain length and width. To assess its risk, the system first retrieves all historical inspection data near this coordinate point (e.g., within a 5-meter radius) from the historical database and obtains previously detected crack features. For example, the historical database may record a short and narrow crack A detected in this area three years ago, and a slightly longer and wider crack B detected a year ago. Next, the system compares the similarity of the currently detected first linear feature with cracks A and B. If the first linear feature is highly similar to crack B in location, size, and orientation, but less similar to crack A, the system determines that the first linear feature and crack B are different stages of the same crack. Based on this matching result, the system updates the historical development trend of the crack, for example, recording that its length has increased by X millimeters and its width by Y millimeters in the past year, while maintaining approximately the same orientation. Finally, the system will conduct a comprehensive risk assessment based on the updated historical development trend (e.g., the crack continues to expand), the current size of the crack (e.g., it has reached a certain length), and its location (e.g., it is located on the water-facing side of the dike). If the assessment results show that the crack is expanding rapidly and has reached a certain size, its risk level may be determined as medium to high risk, triggering a corresponding warning.

[0084] In some embodiments described above, crack risk assessments are conducted based on the crack's historical development trend, size, and location. However, in practical applications, relying solely on these static or historical data for risk assessment may have limitations. In particular, in certain situations, when the initial risk level of the crack is critical, failure to fully consider the influence of current environmental parameters may result in inaccurate risk assessment results or an inability to reflect the true degree of danger of the crack in a timely manner.

[0085] In this regard, this application further proposes that the steps for risk assessment based on the updated historical development trend, crack size, and crack location include: Based on the current crack size, location, and the updated historical development trend, determine the initial risk level of the current crack; When the initial risk level is at a critical state, obtain the current environmental parameters of the dike inspection robot; The initial risk level is corrected based on the environmental parameters to obtain a corrected risk level; and a risk assessment is performed on the current crack based on the corrected risk level.

[0086] Specifically, determining the initial risk level of a crack involves preliminary risk quantification based on its physical properties (such as size, depth, and width), its specific location on the dike (e.g., whether it is located in a critical structural area), and its trend over time (e.g., growth rate, stability). This initial risk level can be divided into several preset levels, such as low risk, medium risk, and high risk. The critical state can be understood as the initial risk level falling within a fuzzy region between two risk levels, or approaching a certain high-risk threshold, requiring further judgment to avoid misjudgment. For example, if the initial risk level is determined to be medium risk, but its value is very close to the high-risk threshold, it is considered to be in a critical state. In practical applications, dike inspection robots can be equipped with various environmental sensors to acquire current environmental parameters. These parameters can specifically include temperature, humidity, rainfall, wind speed, soil moisture content, and groundwater level. These parameters can directly or indirectly affect the stability and development of cracks. For example, continuous rainfall may lead to soil saturation, increasing water pressure inside the dike and thus accelerating crack expansion; low temperatures may lead to freeze-thaw cycles, exacerbating crack damage. Furthermore, revising the initial risk level based on environmental parameters refers to adjusting the initial risk level using the acquired environmental parameters as input through a pre-set correction model or rule. For example, if the initial risk level is medium risk and in a critical state, but the current environmental parameters show heavy rain or high humidity, the revised risk level may be upgraded to high risk. The revised risk level can more accurately reflect the actual degree of danger of the crack under the current environmental conditions.

[0087] This application's solution overcomes the limitations of relying solely on historical and static data for risk assessment by incorporating consideration of current environmental parameters. Specifically, once the initial risk level of a crack is determined, the system assesses whether that level is in a critical state. This assessment is necessary because cracks in a critical state are most susceptible to external environmental factors and require the most refined evaluation. By acquiring the current environmental parameters of the dike inspection robot at this critical moment, the system can capture immediate external conditions that may affect crack stability. Subsequently, by revising the initial risk level based on these environmental parameters, dynamic and real-time environmental impacts can be incorporated into the risk assessment system. This avoids underestimating or overestimating risks due to environmental changes, resulting in a final risk assessment that more closely reflects the true severity of the crack.

[0088] In some preferred embodiments, a specific example is illustrated below. Suppose a levee inspection robot detects a 5 mm crack located in the middle of the levee's water-facing slope, and historical data shows that the crack has grown at a rate of 0.5 mm per month over the past three months. First, based on the crack's size, location, and updated historical trend data, the system determines its initial risk level to be "medium risk." This "medium risk" value is very close to the "high risk" threshold, thus it is judged to be in a critical state. Given that this crack is in a critical state, the levee inspection robot then acquires the current environmental parameters. For example, the sensor detects that the current temperature is -5°C, and there has been moderate to heavy snowfall in the past 24 hours, resulting in high soil moisture content. Based on these environmental parameters, the system invokes a preset correction model. This model may contain rules, such as "if a medium-risk crack is in a critical state and the ambient temperature is below 0°C with snowfall, the risk level is increased by one level." According to this rule, the initial "medium risk" level is corrected to "high risk." Ultimately, based on the revised "high-risk" level, a risk assessment will be conducted on the current crack, which may trigger an immediate alarm, prompting management personnel to take immediate preventative measures, such as antifreeze treatment or emergency reinforcement, to prevent the crack from further deteriorating under freeze-thaw cycles and high humidity conditions, thereby avoiding more serious damage to the levee structure.

[0089] This application proposes an environmental detection system for a dike inspection robot, such as... Figure 2 As shown, the system includes: The fuzzy judgment module 201 is used to acquire the embankment image and perform motion blur judgment on the image; the embankment image is an image of the embankment surface; The direction determination module 202 is used to determine the direction of motion blur in response to the existence of motion blur in the embankment image, and to extract the linear features of the embankment image and determine the direction of the linear features. The detection module 203 is used to determine whether the position corresponding to the linear feature is detected as a levee crack based on the geometric relationship between the motion blur direction and the linear feature direction.

[0090] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. An environmental detection method for a dike inspection robot, characterized in that, The method includes: Acquire an image of the dike and perform motion blur detection; the dike image is an image of the dike surface; In response to the presence of motion blur in the embankment image, the direction of motion blur is determined, and the linear features of the embankment image are extracted to determine the direction of the linear features. Based on the geometric relationship between the motion blur direction and the linear feature direction, determine whether the position corresponding to the linear feature is detected as a levee crack.

2. The environmental detection method for a dike inspection robot according to claim 1, characterized in that, In response to motion blur in the levee image, determining the direction of motion blur includes: Perform a two-dimensional Fourier transform on the embankment image to obtain the frequency domain Fourier spectrum; Detect the dark lines formed by periodic zeros caused by motion blur in the frequency domain Fourier spectrum; The motion blur direction of the embankment image is deduced from the direction of the dark line; wherein the motion blur direction is perpendicular to the direction of the dark line.

3. The environmental detection method for a dike inspection robot according to claim 1 or 2, characterized in that, The levee image has multiple linear features; determining whether the location corresponding to the linear feature is detected as a levee crack based on the geometric relationship between the motion blur direction and the linear feature direction includes: Determine the angle between the direction of each linear feature and the direction of motion blur. If the angle is within a preset range, determine whether the position corresponding to the linear feature is detected as a crack in the embankment.

4. The environmental detection method for a dike inspection robot according to claim 1 or 2, characterized in that, The dike inspection robot includes an inertial measurement unit (IMU). Acquire images of the dike and perform motion blur detection, including: The system continuously acquires images of the levee at a first frequency, and simultaneously records the inertial measurement information of the levee inspection robot at a second frequency based on the inertial measurement unit (IMU) while acquiring the levee images; the inertial measurement information includes triaxial linear acceleration and triaxial angular velocity. For each frame of the embankment image, the variance of the Laplacian operator is calculated. If it is lower than a preset threshold, it is determined to be a blurred image. The blur type is determined based on the inertial measurement information synchronized with the time of the embankment image. The blur type includes single linear blur and composite motion blur.

5. The environmental detection method for a dike inspection robot according to claim 4, characterized in that, The method further includes: If the blur type is a composite motion blur, calculate its composite motion index based on the inertial measurement information; and associate and store the composite motion index with the geographical location corresponding to the current embankment image. The inspection path of the dike inspection robot is divided into multiple segments, and the composite motion index is accumulated for each segment. When the composite motion index appears to be higher than the abnormal threshold in multiple inspections of a road segment, or when the average composite motion index of the road segment shows a continuous upward trend, it is determined that there is a potential road surface anomaly in the road segment.

6. The environmental detection method for a dike inspection robot according to claim 1, characterized in that, Based on the geometric relationship between the motion blur direction and the linear feature direction, determining whether the position corresponding to the linear feature is detected as a levee crack includes: Acquire a continuous sequence of dike images, determine whether the dike image sequence has motion blur, and when the dike image sequence has motion blur, align two frames in the dike image sequence. Calculate the pixel difference between the two aligned images, and determine whether the linear feature is a crack based on the pixel difference.

7. The environmental detection method for a dike inspection robot according to claim 1, characterized in that, Determining whether the location corresponding to the linear feature is detected as a levee crack includes: A risk assessment is conducted on the first linear feature identified as a crack, based on its size, location, and historical development trend. Based on the risk assessment results, the first linear features are prioritized. The high-risk crack information after priority sorting is highlighted and an immediate alarm is triggered.

8. The environmental detection method for a dike inspection robot according to claim 7, characterized in that, The risk assessment of the first linear feature, determined based on its size, location, and historical development trend, includes: Retrieve historical inspection data from the historical database that are adjacent to the location of the first linear feature identified as a crack, and obtain historical crack features; Compare the similarity between the first linear feature and the historical crack feature; Crack matching is performed based on the similarity and the current crack size and orientation; the current crack is the crack corresponding to the first linear feature. Update the historical development trend of the cracks based on the crack matching results; A risk assessment is conducted based on the updated historical trends, the size of the crack, and the location of the crack.

9. The environmental detection method for a dike inspection robot according to claim 8, characterized in that, A risk assessment is conducted based on the updated historical trends, the size of the crack, and the location of the crack, including: Based on the current crack size, location, and the updated historical development trend, determine the initial risk level of the current crack; When the initial risk level is at a critical state, obtain the current environmental parameters of the dike inspection robot; The initial risk level is corrected based on the environmental parameters to obtain a corrected risk level; and a risk assessment is performed on the current crack based on the corrected risk level.

10. An environmental monitoring system for a dike inspection robot, characterized in that, The system includes: The fuzzy detection module is used to acquire the embankment image and perform motion blur detection on the image; the embankment image is an image of the embankment surface; The direction determination module is used to determine the direction of motion blur in response to the existence of motion blur in the embankment image, and to extract the linear features of the embankment image and determine the direction of the linear features. The detection module is used to determine whether the position corresponding to the linear feature is detected as a crack in the dike based on the geometric relationship between the motion blur direction and the linear feature direction.

Images