UAV-based municipal facility fault detection system

Abstract

This invention relates to the field of municipal infrastructure fault detection technology, and discloses a municipal infrastructure fault detection system based on unmanned aerial vehicles (UAVs). The key technical points are: using UAVs to collect road surface data; threshold-based road surface crack detection; extracting detected cracks using a cap transform; grouping the extracted cracks using MorphLink-C; measuring the average crack width of each group of fragments; first, data collection is completed; then, the obtained images are clustered using histogram analysis and peak detection, and road surface cracks are detected using a local adaptive threshold; the road surface crack detection process is automated, shortening inspection time and reducing the cost of municipal infrastructure fault detection. Next, grouping the extracted cracks using MorphLink-C improves the accuracy of crack classification results; finally, measuring the average crack width of each group of fragments eliminates the need for complex calculations of crack direction, length, and intensity for arbitrary crack types, further shortening inspection time.

Description

Technical Field

[0001] This invention relates to the field of municipal facility fault detection technology, and more specifically, to a municipal facility fault detection system based on unmanned aerial vehicles (UAVs). Background Technology

[0002] Municipal facilities refer to public facilities constructed with funds from the government, legal entities, or citizens, and generally refer to various urban roads within a planned area. Road surface cracks are often one of the most serious defects in urban roads, and sections with a high density of surface cracks require regular inspection and maintenance to ensure the safe operation of vehicles.

[0003] However, existing crack detection technologies generally rely on visual inspection using multiple ground-based mobile robots. This method is both time-consuming and expensive. Furthermore, the subjective judgment of inspectors limits the objectivity of the detection results and reduces their accuracy.

[0004] Therefore, the present invention provides a municipal facility fault detection system based on unmanned aerial vehicles (UAVs), which improves the above-mentioned technical problems. Summary of the Invention

[0005] The purpose of this invention is to provide a municipal facility fault detection system based on unmanned aerial vehicles (UAVs). This system solves the technical problems of existing crack detection methods, which generally rely on visual inspection using multiple ground mobile robots. This method is both time-consuming and expensive, and the subjective judgment of inspectors limits the objectivity of the detection results, thus reducing their accuracy.

[0006] The above-mentioned technical objective of the present invention is achieved through the following technical solution: a municipal facility fault detection system based on unmanned aerial vehicles (UAVs), comprising the following steps: S1, collecting road surface data using UAVs;

[0007] S2. Threshold-based road surface crack detection;

[0008] S3. Use bottom cap transformation to extract the detected cracks;

[0009] S4. Use MorphLink-C to group the extracted cracks;

[0010] S5. Measure the average crack width of each group of fragments.

[0011] As a preferred embodiment of the present invention, the process of the UAV collecting road surface data is as follows:

[0012] S1.1 The drone takes aerial photos of the road surface from multiple angles;

[0013] S1.2 Point cloud 3D modeling;

[0014] Identify the identical parts and corresponding overlapping points of the road surface in the point cloud, let P. a and P b For the point cloud recorded at locations a and b, the overlap point is determined by the following formula:

[0015]

[0016] in and These are overlapping points used to represent nearest neighbors; τ is a predefined distance threshold; then, the alignment of the point clouds is obtained by applying the iterative nearest point algorithm to find the transformation that minimizes the distance between the point clouds.

[0017] S1.3 Geometric feature extraction;

[0018] Define a plane equation: (ax + by + cz + d = 0); let M represent the parameter vector to be identified; where M = [a, b, c, d]. T ;

[0019] Let ∈ be a positive constant, where ∈ > 0; such that a cluster is defined as a set of points:

[0020] C={p i ∈P|min||p i -p j ||>∈};For any other point that does not belong to this cluster, i.e. It is considered an obstacle;

[0021] S1.4 The cost of moving one pixel to another in the neighborhood of the acquisition path planning is calculated as follows:

[0022] C(k, l, m) = a1k 2 +a2l 2 +a3m 2 ;

[0023] Wherein, coordinates k, l, m∈{-1, 0, 1} represent adjacent positions, and coefficients a1, a2, and a3 are assigned specific weights for each direction; the specific weights assigned to each direction serve as a reference for the motion control of the UAV.

[0024] As a preferred technical solution of the present invention, the threshold-based road surface crack detection process includes: image stitching, stitching pattern removal, and crack detection.

[0025] As a preferred embodiment of the present invention, the stitched image consists of cracked areas, stitched patterns, and blank areas; since the brightness of the cracked areas, stitched patterns, and blank areas varies with illuminance, it is necessary to first detect the main peaks corresponding to these elements; the thresholds t1 and t2 for segmenting the road surface to be detected are calculated as the average intensity values.

[0026]

[0027] Where i b i w and i p These are the peak intensity values ​​corresponding to the blank area, the road surface, and the spliced ​​pattern, respectively.

[0028] The formula for identifying spliced ​​patterns based on histograms is as follows:

[0029]

[0030] Where I gr (x, y) and I r (x, y) are the grayscale intensity and red channel intensity of the stitched image at point P(x, y), respectively, and β is the portion of the image with intensity values ​​selected to distinguish this pattern from other patterns; β is set to 255.

[0031] As a preferred embodiment of the present invention, the crack detection process is as follows:

[0032] First, the image is preprocessed using automatic peak detection, retaining only the background and linear objects. Then, a locally adjusted threshold is applied to identify potential cracks.

[0033] Let the pixels in the x and y directions be (x i y i ); where i = 1, 2, ..., N 2 A window in the neighborhood of pixel P(x,y) is determined by using N×N, and m(x,y) and s(x,y) are used as the mean and standard deviation of this window, respectively; the formulas for calculating m(x,y) and s(x,y) are as follows:

[0034]

[0035]

[0036] As a preferred embodiment of the present invention, the threshold calculation formula for pixel P(x, y) is as follows:

[0037]

[0038] Where R is the dynamic range of the standard deviation, and k is an adjustable parameter used to adjust the effect of the standard deviation;

[0039] Each pixel P(x,y) is evaluated based on its threshold T(x,y); if the gray intensity I of pixel P(x,y) is... gr If pixel P(x, y) is higher than the calculated local threshold, then the pixel belongs to the crack; otherwise, it is considered to be the background.

[0040] As a preferred embodiment of the present invention, the process of extracting the detected cracks by the bottom cap transformation is as follows:

[0041] Let {p} be the set of pixels in a two-dimensional digital image; such that {p} can divide the x, y plane into a grid, where the center coordinates of each grid are a pair of elements (x, y) in Cartesian coordinates; if (x, y) are integers in {p}, and f is a mapping that assigns intensity values ​​to each distinct pair of (x, y), then the function f(x, y) is the shape of the binary image of the digital image applied to the input image, such that the structuring element is a binary image component, where pixels within the boundary of the structuring element are assigned the value 1, and the values ​​of the remaining pixels are assigned zero;

[0042] In image processing, the dilation and erosion of the original image o and the structuring element S are respectively represented as follows: and

[0043] Let the formula for opening operations be: The formula for the closing operation is:

[0044] As a preferred technical solution of the present invention, the detected black and white cracks are extracted by transforming the bottom cap and top cap, where the bottom cap L... b and top hat L t The formulas for calculating the transformation are as follows:

[0045]

[0046]

[0047] Where L represents the grayscale image output as a morphological transformation; O represents the original grayscale image; S {0°，45°，90°，135°} The symbols 0°, 45°, 90°, and 135° represent rotations of linear structural elements; ° represents an open transformation; and · represents a closed transformation.

[0048] As a preferred embodiment of the present invention, the process of grouping the extracted cracks using MorphLink-C is as follows:

[0049] S4.1 Apply the morphological dilation transformation to a binary image containing crack fragments. The calculation formula is as follows:

[0050]

[0051] Where B represents a binary image containing crack fragments; S D Structural elements representing dilation transformations; denoted by morphological dilation operator; D represents the binary image generated after the dilation transformation;

[0052] S4.2. Apply morphological refinement transformation to D to connect the fragments into continuous crack lines within the expansion boundary. The calculation formula is: T = D - hit - and miss(D, S T );

[0053] Wherein, hit-and-miss is the morphological hit-and-miss operator; S T represents the skeletalized structural element; T represents the binary image obtained after the thinning transformation.

[0054] As a preferred embodiment of the present invention, the formula for calculating the average crack width of each group of fragments is as follows:

[0055]

[0056] Where, ∑ i (Area of ​​pixels with border cracks) k The sum of the total areas of the fragment crack pixels within the k-th expansion boundary; i represents the index of the fragment object within the k-th expansion boundary; ∑ i (Length of the connecting crack) k This represents the sum of the total lengths of the crack lines connected within the k-th expansion boundary.

[0057] In summary, this invention offers the following advantages: First, a 3D model of the structure is created using a laser scanner. Then, based on this model, geometric attributes are extracted to generate waypoints needed for navigation drones to capture road surface images, completing data collection. Next, the obtained images are clustered using histogram analysis and peak detection, and road surface cracks are detected using local adaptive thresholding. The road surface crack detection process is automated, shortening inspection time and reducing the cost of municipal facility fault detection. Furthermore, the extracted cracks are grouped using MorphLink-C, improving the accuracy of crack classification results. Finally, the average crack width of each grouped fragment is measured, eliminating the need for complex calculations of crack direction, length, and intensity for any crack type, further shortening inspection time. Attached Figure Description

[0058] Figure 1 A flowchart of a municipal facility fault detection system based on unmanned aerial vehicles (UAVs) provided in an embodiment of the present invention;

[0059] Figure 2An example of a grayscale input image with diamond-shaped structural elements provided in an embodiment of the present invention;

[0060] Figure 3 The original grayscale image provided in the embodiments of the present invention;

[0061] Figure 4 The grayscale image provided in the embodiment of the present invention is an opening operation image;

[0062] Figure 5 The grayscale image provided by the closing operation in the embodiment of the present invention;

[0063] Figure 6 The image showing the result of the top-hat transformation provided in this embodiment of the invention;

[0064] Figure 7 The image showing the result of the cap transformation provided in this embodiment of the invention;

[0065] Figure 8 The original image provided for the embodiments of the present invention;

[0066] Figure 9 This invention provides an embodiment of the method for extracting fragmented objects after cracking using a bottom cap transformation.

[0067] Figure 10 This is the result of segment grouping using dilation transformation provided in an embodiment of the present invention;

[0068] Figure 11 The result of segment concatenation using a thinning transformation is provided in an embodiment of the present invention. Detailed Implementation

[0069] The present application will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application.

[0070] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0071] It should be noted that, unless there is a conflict, the various features in the embodiments of this application can be combined with each other, all of which are within the protection scope of this application. Furthermore, although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than the module division in the device or the order in the flowchart. In addition, the terms "first," "second," and "third" used herein do not limit the IoT data or the execution order, but only distinguish identical or similar items with essentially the same function and effect.

[0072] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0073] Furthermore, the technical features involved in the various embodiments of this application described below can be combined with each other as long as they do not conflict with each other.

[0074] This disclosure aims to address the problems in existing crack detection technologies, which typically rely on visual inspection using multiple ground-based mobile robots. This method is both time-consuming and expensive, and the subjective judgment of inspectors limits the objectivity of the detection results, reducing their accuracy. Therefore, this disclosure proposes a UAV-based municipal facility fault detection system. This system extracts geometric attributes to generate waypoints for road surface images captured by the navigation UAV, completing data collection. Then, the obtained images are clustered using histogram analysis and peak detection, and road surface cracks are detected using a local adaptive threshold. Next, the extracted cracks are grouped using MorphLink-C, improving the accuracy of crack classification results. Finally, the average crack width of each grouped fragment is measured, eliminating the need for complex calculations of crack direction, length, and intensity for any crack type.

[0075] Please refer to Figure 1 , Figure 1 A flowchart of the UAV-based municipal facility fault detection system according to an embodiment of this disclosure is shown. The overall process mainly includes the following five steps:

[0076] Step 1: Use a drone to collect road surface data.

[0077] S1.1 The drone takes aerial photos of the road surface from multiple angles;

[0078] Aerial images are used to obtain sufficient geometric and surface information about the road surface for post-processing to detect potential cracks; they are also used for the navigation of the drone itself.

[0079] S1.2 Point cloud 3D modeling;

[0080] A laser scanner is used to acquire extent information from different locations on the road surface to be inspected and represent it as a point cloud. These point clouds are then merged one by one in a process called registration to create a 3D model. During registration, the first step is to identify identical portions of the road surface and corresponding overlapping points in the point cloud. Let P... a and P b Let be the point cloud recorded at positions a and b. The overlap point is determined by the following formula:

[0081]

[0082] in and These are overlapping points, used to represent nearest neighbors; τ is a predefined distance threshold. Then, the alignment of the point clouds is achieved by applying an iterative nearest-neighbor algorithm to find the transformation that minimizes the distance between the point clouds.

[0083] S1.3 Geometric feature extraction;

[0084] Road surfaces are typically the primary targets for inspection, therefore they need to be extracted from the point cloud. Given a plane equation...

[0085] (ax+by+cz+d=0);

[0086] Let M denote the parameter vector to be identified; where M = [a, b, c, d] T .

[0087] After detecting the road surface, the convex hull algorithm is used to determine its boundary. Then, the remaining point cloud P is clustered into groups to represent obstacles in the UAV's path planning. Here, let a positive constant ∈ , where ∈ > 0, such that a cluster is defined as a set of points:

[0088] C={p i ∈P|min||p i -p j ||>∈};For any other point that does not belong to this cluster, i.e. It is considered an obstacle;

[0089] S1.4, Data Acquisition Path Planning

[0090] Given the road surface to be detected, a list of waypoints needs to be created to navigate the drone. There are two types of waypoints: one corresponds to the shooting points in the captured images, and the other serves as intermediate points for path tracking and obstacle avoidance. Both are generated by first segmenting the operating environment into pixels. Then, a free or occupied state is defined for each pixel to indicate the presence of obstacles within that pixel. Shooting points are then calculated based on intrinsic parameters such as camera focal length, surface area, and minimum resolution. Finally, the shortest path between shooting points is found using the A-star algorithm. The cost of moving from one neighboring pixel to another in each step is calculated as follows:

[0091] C(k, l, m) = a1k 2 +a2l 2 +a3m 2 ;

[0092] Where the coordinates k, l, m ∈ {-1, 0, 1} represent adjacent positions, and the coefficients a1, a2, and a3 are assigned specific weights for each direction. These specific weights serve as a reference for the UAV's motion control.

[0093] Step 2: Threshold-based road surface crack detection.

[0094] The threshold-based crack detection process includes: image stitching, stitching pattern removal, and crack detection.

[0095] S2.1 Image Stitching: Each image captured by the drone only covers a small portion of the road surface being detected. Therefore, each image needs to be stitched together to create a panoramic image. Typically, a certain degree of overlap is required between consecutive images and corresponding features. A feature-invariant approach is employed, including scale-invariant feature transformation feature extraction, homography calculation, and image matching verification. If the road surface to be detected is uniform, a manual pattern can be added to enhance stitching performance and then subsequently removed.

[0096] S2.2. Pattern Removal Before processing the stitched image, removing the pattern used in the stitching can improve detection accuracy. A stitched image typically consists of three elements: crack areas, stitching patterns, and blank areas. The brightness of these three often varies depending on the illuminance; therefore, it is necessary to first detect the dominant peaks corresponding to these elements. The thresholds t1 and t2 for segmenting the road surface to be detected are calculated as the average intensity:

[0097]

[0098] Where i b i w and i p These are the peak intensity values ​​corresponding to the blank area, the road surface, and the spliced ​​pattern, respectively.

[0099] The formula for identifying spliced ​​patterns based on histograms is as follows:

[0100]

[0101] Where I gr (x, y) and I r (x, y) represent the grayscale intensity and red channel intensity of the stitched image at point P(x, y), respectively, and β is the portion of the image with intensity values ​​selected to distinguish this pattern from other patterns. Since the intensity at the crack structure is usually less than the intensity of the non-defective areas of the inspected pavement, β needs to be set to 255 to adequately filter out these features without losing information.

[0102] S2.3 Crack Detection

[0103] First, the image is preprocessed using automatic peak detection, retaining only the background and linear objects. Then, a locally adjusted threshold is applied to identify potential cracks. Unlike global methods, the threshold is calculated here based on the grayscale intensity of each pixel's neighborhood. Specifically:

[0104] Let the pixels in the x and y directions be (x i y i ); where i = 1, 2, ..., N 2 A window in the neighborhood of pixel P(x,y) is determined by using N×N, and m(x,y) and s(x,y) are used as the mean and standard deviation of this window, respectively; the formulas for calculating m(x,y) and s(x,y) are as follows:

[0105]

[0106]

[0107] The threshold calculation formula for pixel P(x, y) is as follows:

[0108]

[0109] Where R is the dynamic range of the standard deviation, and k is an adjustable parameter used to adjust the effect of the standard deviation. Then, each pixel P(x,y) is evaluated based on its threshold T(x,y). If the gray intensity I of pixel P(x,y) is... gr If pixel P(x, y) is higher than the calculated local threshold, then the pixel is considered to be a crack; otherwise, it is considered to be background.

[0110] Step 3: Use the bottom cap transformation to extract the detected cracks.

[0111] Morphological operations include opening and closing operations, and top-hat and bottom-hat (black-hat) transforms. Therefore, mathematical morphology is used as a tool for extracting image components through morphological image processing techniques. The specific steps are as follows:

[0112] Let {p} be the set of pixels in a two-dimensional digital image. Let {p} be such that it divides the x, y plane into a grid, where the center coordinates of each grid are a pair of elements (x, y) in Cartesian coordinates. If (x, y) are integers in {p}, and f is a mapping that assigns intensity values ​​to each distinct pair of (x, y) pairs, then the function f(x, y) is the digital image. Morphological image processing techniques are used to apply the shape (called a structuring element) of the binary image to the input image. This makes the structuring element a binary image component whose shape can be designed for different purposes in image processing. The center of the structuring element is called the pixel of interest (POI). Pixels within the boundaries of the structuring element are assigned the value 1, and the values ​​of the remaining pixels are assigned zero. Figure 2 An example of a grayscale input image with diamond-shaped structural elements is shown.

[0113] In morphological techniques, dilation and erosion are two fundamental operators manipulating f(x, y). Since dilation enlarges or thickens objects in an image, while erosion reduces or thins them, the dilation and erosion of the original image (O) and structuring element (S) in image processing can be represented as follows: and

[0114] Morphological techniques can be used to combine multiple basic operators into another operation for more complex image processing. For example, two commonly used operations are opening and closing operations, which include: and

[0115] Since crack extraction is a pixel-level operation, it is necessary to remove random backgrounds to detect crack-like objects in road surface images. For flexible pavements, random backgrounds typically include asphalt type, pavement coating, pavement texture, and roughness. Therefore, black and white cracks are extracted by using bottom cap and top cap transformations, with the bottom cap L... b and top hat L t The formulas for calculating the transformation are as follows:

[0116]

[0117]

[0118] Where L represents the grayscale image output as a morphological transformation; O represents the original grayscale image; S {0°，45°，90°，135°} The symbols 0°, 45°, 90°, and 135° represent rotations of linear structural elements; ° represents an open transformation; and · represents a closed transformation.

[0119] In this example, a bottom-cap transformation is used to extract the detected cracks. Figure 3The original grayscale image is displayed, containing 10 different horizontal white and black cracks with widths ranging from 10 to 100 pixels, incremented by 10 pixels. This further distinguishes black cracks less than 50 pixels wide from the gray background; thus, the structural elements are designed as vertical lines of size 50×1. Figure 4-6 This illustrates the step-by-step process of the cap transformation. The resulting image of the cap transformation is shown below. Figure 7 As shown.

[0120] Detects black cracks with a width less than 50 pixels. The detected cracks were originally black, but after a background transform, they became gray, and the background changed from gray to black.

[0121] As shown above, the beneficial effects of the cap transform for pavement crack extraction are as follows: by precisely designing the combination of the structural element shape (i.e., the vertical line of a horizontal crack), size (i.e., 50 pixels), and morphological operators (i.e., the cap transform), pavement cracks can be selectively detected based on the following: crack shape (e.g., horizontal cracks), size (e.g., 50-pixel crack width), and intensity (e.g., black cracks). Meanwhile, the cap transform uses a relatively small number of user-defined parameters. In this embodiment, only the structural element length parameter is needed; that is, a vertical line can be used as the structural element to detect black cracks with a crack width less than 50 pixels.

[0122] Step 4: Group the extracted cracks using MorphLink-C.

[0123] The extracted cracks are grouped using MorphLink-C, which consists of the following two steps:

[0124] S4.1 Apply the morphological dilation transformation to a binary image containing crack fragments. The calculation formula is as follows:

[0125]

[0126] Where B represents a binary image containing crack fragments; S D Structural elements representing dilation transformations; denoted as morphological dilation operator; D represents the binary image generated after the dilation transformation.

[0127] Dilation transformation is an image operation that expands or thickens image components; in this invention, the image components are fragmented or cracked pixels. D It was selected as a square structural element.

[0128] S4.2. Apply morphological refinement transformation to D to connect the fragments into continuous crack lines within the expansion boundary. The calculation formula is: T = D - hit - and miss(D, S T );

[0129] Wherein, hit-and-miss is the morphological hit-and-miss operator; S T represents the skeletalized structural element; T represents the binary image obtained after the thinning transformation.

[0130] Step 5: Measure the average crack width of each group of fragments.

[0131] The average crack width of each group of fragments is measured, and the calculation formula is as follows:

[0132]

[0133] Where, ∑ i (Area of ​​pixels with border cracks) k The sum of the total areas of the fragment crack pixels within the k-th expansion boundary; i represents the index of the fragment object within the k-th expansion boundary; ∑ i (Length of the connecting crack line) k This represents the sum of the total lengths of the crack lines connected within the k-th expansion boundary.

[0134] Furthermore, while the cap transform can extract crack components from flexible pavement images, two problems remain. First, the extracted crack objects are fragmented, with multiple non-intersecting sections in their crack paths. Although image B represents a crack shape with abundant local detail, vertical cracks exhibit multiple discontinuities in their crack paths. Therefore, Figure 9 The crack area measured across 8 connected pixels does not represent the area of ​​a true vertical crack. Secondly, even after removing the road surface background, image B still contains non-cracked objects. Figure 9 It can be seen that there is a vertical crack in the middle, and there are false cracks around it. Figure 10 The results of segment grouping using dilation transformation are shown. The boundary range is determined based on the relative proximity between segments, and can be obtained through S... D The size is specified.

[0135] In this invention, a square structural element S is used. D Its size is 10×10 pixels. Figure 10 The numbers shown represent the sum of the number of pixels in each pixel cluster within each boundary. After grouping the extracted cracks using MorphLink-C, it can be observed that... Figure 10 The fragmentation sorting area is more than Figure 9 The fragmented regions in the image more accurately represent real vertical cracks. For example, the largest boundary containing a vertical crack comprises 12 pixel clusters. The largest boundary has 1,842 pixels, which is the sum of the pixels in the 12 clusters. The smallest real crack within the largest boundary has 10 pixels. Without a boundary, this crack could easily be misclassified as a non-crack object because... Figure 9 The largest non-crack segment has 44 pixels. Therefore, MorphLink-C improves the accuracy of crack classification results.

[0136] in, Figure 11 The results of fragment joining using a thinning transform are shown. The numbers shown in the figure represent the lengths of thin lines with a width of one pixel. The joining of crack fragments is adaptive and does not require complex calculations of crack direction, length, and intensity for arbitrary crack types (such as single cracks, branched cracks, and block cracks).

Claims

1. A municipal facility fault detection system based on unmanned aerial vehicles (UAVs), characterized in that, Includes the following steps: S1. Use drones to collect road surface data; S2. Threshold-based road surface crack detection; S3. Use bottom cap transformation to extract the detected cracks; S4. Use MorphLink-C to group the extracted cracks; S5. Measure the average crack width of each group of fragments; The process of the drone collecting road surface data is as follows: S1.1 The drone takes aerial photos of the road surface from multiple angles; S1.2 Point cloud 3D modeling; Identify the identical parts and corresponding overlapping points of the road surface in the point cloud, and set... and For in position and The overlapping points of the point cloud recorded at each location are determined by the following formula: ； in and These are overlapping points, used to represent nearest neighbors; It is a predefined distance threshold; then, by applying the iterative nearest point algorithm, the transformation that minimizes the distance between point clouds is found to obtain the alignment of the point clouds; S1.3 Geometric feature extraction; Define a plane equation: ; Let M denote the parameter vector to be identified; where, ; Let a positive constant be... ,in This allows a cluster to be defined as a set of points: For any other point that does not belong to the cluster, i.e. It is considered an obstacle; S1.4, Data Acquisition Path Planning The cost of moving one pixel in a neighborhood to another pixel is calculated as follows: ； Where, coordinates Indicates adjacent positions, coefficient and Assign weights to each direction; these weights serve as a reference for the drone's motion control. The threshold-based road surface crack detection process includes: image stitching, stitching pattern removal, and crack detection; The stitched image consists of cracked areas, stitched patterns, and blank areas. Since the brightness of these areas varies with illuminance, it's necessary to first detect the main peaks corresponding to these elements. The thresholds t1 and t2 for segmenting the road surface to be detected are calculated as average intensity values. ； in and These are the peak intensity values ​​corresponding to the blank area, the road surface, and the spliced ​​pattern, respectively. The formula for identifying spliced ​​patterns based on histograms is as follows: ； in and They are The grayscale intensity and red channel intensity of the stitched image at the point. It is the portion of the image selected to distinguish this pattern from other patterns based on its intensity values; Set to 255; The crack detection process is as follows: First, the image is preprocessed using automatic peak detection, retaining only the background and linear objects. Then, a locally adjusted threshold is applied to identify potential cracks. set up and The direction of the pixels is ;in, ; Pixels are determined by using N×N Windows in the neighborhood, and and These are respectively used as the mean and standard deviation within this window; and The calculation formula is: ； The pixel The threshold calculation formula is: ； in, It is the dynamic range of the standard deviation. It is an adjustable parameter used to adjust the effect of standard deviation; Based on its threshold Evaluate each pixel If pixels gray intensity If the value is higher than the calculated local threshold, then the pixel... It is considered a crack; otherwise, it is considered part of the background.

2. The municipal facility fault detection system based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The process of extracting the detected cracks by the bottom cap transformation is as follows: set up Let be the set of pixels in a two-dimensional digital image; such that The x, y plane can be divided into a grid, where the center coordinates of each grid are a pair of elements (x, y) in Cartesian coordinates; if (x, y) is Integers in, and If it is a mapping that assigns intensity values ​​to each distinct (x, y) pair, then the function For digital images The shape of the binary image is applied to the input image, such that the structuring element is a binary image component, and the pixels within the boundaries of the structuring element are assigned the value 1, while the values ​​of the remaining pixels are assigned zero. Original image in image processing With structural elements The expansion and corrosion are respectively represented as and ; Let the formula for opening operations be: The formula for calculating the closing operation is: .

3. The municipal facility fault detection system based on unmanned aerial vehicles (UAVs) according to claim 2, characterized in that, Black and white cracks detected by bottom cap and top cap transformation are extracted. and top hat The formulas for calculating the transformation are as follows: ； in, This represents a grayscale image as the output of a morphological transformation. Represents the original grayscale image; The symbols 0°, 45°, 90°, and 135° represent rotations of linear structural elements; ° indicates an open transformation; • indicates a closed transformation.

4. The municipal facility fault detection system based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The process by which MorphLink-C groups the extracted cracks is as follows: S4.1 Apply the morphological dilation transformation to a binary image containing crack fragments. The calculation formula is as follows: ； in, Represents a binary image containing cracked fragments; Structural elements representing dilation transformations; Represents the morphological dilation operator; This represents the binary image generated after the dilation transformation; S4.

2. Apply morphological refinement transformation to D to connect the fragments into continuous crack lines within the expansion boundary. The calculation formula is as follows: ； Among them, hit-and-miss is the morphological hit-and-miss operator; Represents skeletal structural elements; This represents the binary image obtained after the thinning transformation.

5. The municipal facility fault detection system based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The formula for calculating the average crack width of each group of fragments is as follows: ； in, This represents the sum of the total area of ​​the fragment crack pixels within the k-th expansion boundary; i represents the index of the fragment object within the k-th expansion boundary; This represents the sum of the total lengths of the crack lines connected within the k-th expansion boundary.

Images