A method for verifying the quality of highway slope section parameters based on aerial image and photogrammetry point cloud

Abstract

The application discloses a kind of highway slope section parameter quality verification method based on aerial image and photogrammetry point cloud, belong to highway engineering digitization and construction period quality management field.This method includes obtaining design parameter, aerial image, dense point cloud and camera parameter;Identify slope surface area mask and platform area mask on image;Based on the semantic assignment of three-dimensional points based on re-projection relationship, obtain slope surface point cloud and platform area point cloud;Using platform area point cloud to classify and segment slope surface point cloud, form each level slope surface object point cloud;Fitting is carried out to each level slope surface, and slope top line and slope foot line are extracted;Calculate the intersection of law plane and each level slope surface, slope top line and slope foot line, obtain slope top slope foot key point;Calculate measured slope height and measured slope gradient, determine whether cross section is out of limit.The application improves the automation level and result stability of highway slope section parameter verification.

Description

Technical Field

[0001] This invention relates to the field of digital highway engineering and construction period quality management, specifically to a method for verifying the quality of highway slope cross-sectional parameters based on UAV aerial images and photogrammetric point clouds. Background Technology

[0002] In highway reconstruction and expansion projects, after slope construction is completed, it is usually necessary to compare and verify the actual formed state of the slope with the slope design parameters to determine whether the cross-sectional parameters such as slope height and slope ratio meet the design requirements. Current slope quality verification work largely relies on manual cross-sectional sampling, typically involving on-site personnel measuring, recording, and comparing data at representative cross-sections. This method suffers from problems such as untimely inspections, missed inspections, incorrect recording, and limited sampling coverage, making it difficult to generate continuous, comprehensive, and stable quality verification results for long-distance, multi-level slopes.

[0003] On the other hand, quality management during the construction period not only requires determining whether individual sections meet design requirements, but also requires clarifying the specific mileage location, continuous distribution range, and statistical results of sections exceeding limits, in order to support problem identification, rectification review, and process management. Existing manual verification methods still have shortcomings in terms of continuous section location, batch result statistics, and full-process traceability, making it difficult to meet the requirements of automation, quantification, and traceability for slope quality verification during the construction period of highway reconstruction and expansion projects. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a method for verifying the quality of highway slope cross-sectional parameters based on aerial imagery and photogrammetric point clouds, which solves the problems of unstable positioning of key points of slopes during construction, difficulty in automatically extracting cross-sectional parameters, and difficulty in quantifying and outputting excess sections.

[0005] To solve the above problems, the present invention adopts the following technical solution: A method for verifying the quality of highway slope cross-sectional parameters based on aerial imagery and photogrammetric point clouds is applied to the quality verification of slopes during highway reconstruction and expansion construction. The technical solution is as follows: S1. Obtain the route centerline data and slope design parameters of the road section to be verified; S2. Acquire drone aerial images of the road section and dense point clouds reconstructed from the aerial images by photogrammetry, and acquire camera parameters obtained by photogrammetry calculation; S3. Perform two-dimensional element recognition on the aerial image and output the slope surface area mask and the platform area mask. S4. Based on the camera parameters, the three-dimensional points in the dense point cloud are reprojected onto the aerial image plane covering the area where they are located. According to the correspondence between the reprojection landing point and the slope surface area mask and the platform area mask, the three-dimensional points are semantically assigned to obtain the slope surface point cloud and the platform area point cloud respectively. S5. In a multi-level slope scenario, the slope surface point cloud is segmented using the platform area point cloud described in step S4 to form slope surface object point clouds at each level; in a single-level slope scenario, the slope surface point cloud as a whole is treated as a single-level slope surface object point cloud. S6. Perform slope surface fitting on the point cloud of slope surface objects at all levels described in step S5. S7. Extract the slope crest line and slope toe line corresponding to each level of the fitted slope surface; S8. Extract multiple normal planes perpendicular to the tangent of the route along the centerline of the route at fixed station intervals; S9. Calculate the intersection points of each plane with the slope surface, slope top line and slope toe line of each level of slope, and use the intersection points as the key points of the slope top and slope toe of each level of slope, respectively. S10. Based on the key points at the top and bottom of the slope described in step S9, calculate the measured slope height and measured slope ratio, and compare the measured slope height and measured slope ratio with the designed slope height and designed slope ratio respectively to obtain the slope height deviation and slope ratio deviation. Determine whether the cross section exceeds the limit according to the preset threshold rule, and output the mileage location, the range of the exceeding section, and the statistical results of the exceeding section.

[0006] Optionally, in step S1, the route centerline data includes at least route station information and the spatial coordinates corresponding to each station; the slope design parameters include at least the design slope height and the design slope ratio.

[0007] Optionally, in step S2, the aerial image is subjected to photogrammetric processing, which includes image feature extraction, corresponding point matching, aerial triangulation, sparse point cloud reconstruction, and multi-view image dense reconstruction.

[0008] Optionally, in step S3, the two-dimensional element recognition is achieved using a semantic segmentation method based on deep learning.

[0009] Optionally, in step S3, the platform area includes the sluice gate, the slope protection road, and the slope platform areas of each level between two adjacent slope surfaces in a multi-level slope.

[0010] Optionally, in step S4, if the reprojection point of the three-dimensional point is located within the slope surface area mask, then the three-dimensional point is assigned a slope surface semantic label; if the reprojection point is located within the slope platform area mask at each level, then the three-dimensional point is assigned a corresponding slope platform area semantic label.

[0011] Optionally, in step S5, the point cloud of each level of slope platform area is used as the hierarchical boundary constraint to divide the slope surface point cloud into multiple independent slope object point clouds corresponding to the adjacent slope platform areas.

[0012] Optionally, in step S6, the slope fitting is a least squares plane fitting.

[0013] Optionally, in step S7, the top line and toe line of the slope are extracted based on the boundary position between the fitted slope surface of each level and its adjacent platform area or outer boundary. For general slopes, the upper boundary line is used as the top line and the lower boundary line is used as the toe line. For the last level of cut slopes and the first level of fill slopes, the toe line is extracted based on the boundary position between the slope surface of the slope level and its adjacent platform area, and the outer boundary on the other side of the slope level is used as the top line.

[0014] Optionally, in step S8, the normal plane passes through the spatial coordinate point of the corresponding station number, and its normal vector is consistent with the tangential direction of the route centerline at that station number.

[0015] Optionally, in step S9, the intersection of each normal plane with the slope surface of each level is used to obtain the cross-sectional curve of the corresponding level slope, and the intersection of the cross-sectional curve with the top line and toe line of the corresponding level slope is further calculated as the key points of the top and toe of the slope of that level slope.

[0016] Optionally, in step S10, the measured slope height is calculated based on the elevation difference between the key point at the top of the slope and the key point at the bottom of the slope, and the measured slope ratio is calculated based on the vertical elevation difference and lateral distance of the corresponding slope in the normal plane; when multiple adjacent cross sections continuously exceed the limit, they are merged and output as a continuous exceeding section according to the mileage continuity.

[0017] Compared with the prior art, the present invention has at least the following beneficial effects: (1) Addressing the problem of "difficulty in accurately classifying multi-level slopes": In existing technologies, multi-level slopes usually rely on manual interpretation or coarse classification based on elevation thresholds, making it difficult to accurately identify the spatial boundary relationships between different levels of slopes. This is especially true when there is shading, vegetation cover, or irregular morphology, resulting in poor stability of the classification results. This invention, through the processing mechanisms of steps S3 and S4, first performs semantic segmentation of the slope surface and platform area based on aerial imagery. Then, it uses camera parameters to reproject the semantic information onto a 3D point cloud, achieving semantic assignment of the point cloud and thus clearly distinguishing the slope surface and platform area in 3D space. Based on this, combined with step S5 using the platform area point cloud as the classification boundary constraint, the slope surface point cloud is structurally segmented to form point clouds of slope surface objects at each level. Through the above technical measures, the problems of reliance on experience and ambiguous classification boundaries in traditional methods for slope classification are solved, realizing automated and refined classification expression of multi-level slopes in 3D space, significantly improving the accuracy and stability of classification.

[0018] (2) Regarding the problem of "high noise in point clouds and geometric instability leading to cross-section calculation errors": Existing point cloud-based cross-section analysis methods usually directly use the original point cloud for truncation and calculation, which is easily affected by the discreteness of point clouds, local missing points, and noise points, resulting in unstable extraction of slope top and slope toe positions, thus affecting the accuracy of cross-section parameter calculation. This invention constructs a stable geometric expression through steps S6 to S7: First, least squares plane fitting is performed on the point cloud of slope objects at all levels to obtain a continuous and smooth slope model; then, the slope top line and slope toe line are extracted based on the spatial relationship between the fitted slope and the platform area. Furthermore, a normal plane constraint mechanism is introduced in steps S8 and S9, and a cross-section analysis framework is uniformly constructed through the normal plane perpendicular to the route tangent, and a unique key point is determined by the method of "slope intersection → cross-section line → intersection with slope top / slope toe line → elevation screening". Through the above technical measures, the original discrete point cloud is transformed into a stable geometric system under the constraints of "fitting surface + structure line + normal plane", which effectively reduces the impact of noise and local anomalies on the results and improves the robustness and geometric consistency of key point extraction at the top and bottom of the slope.

[0019] (3) Addressing the problem of "low automation level and difficulty in engineering application of cross-sectional parameter verification": In existing technologies, slope height and slope ratio verification mostly rely on manual cross-sectional measurement or semi-automatic tools, which suffers from low efficiency, strong subjectivity, and difficulty in batch processing, making it difficult to meet the needs of large-scale and rapid quality verification during the construction period. This invention forms a complete automatic cross-sectional analysis chain through steps S8 to S10: using the route centerline as a reference, a normal plane is automatically constructed at fixed station intervals; based on the normal plane, key points of the slope top and slope toe of each level of slope are automatically obtained; further, the measured height and slope ratio of each level of slope are calculated and compared with the design parameters to complete the over-limit judgment. At the same time, by sorting the over-limit cross-sections by mileage and merging continuous sections, the expansion from single-point detection to section-level quality analysis is realized. Through the above technical measures, the full-process automated calculation and quality judgment of slope cross-sectional parameters are realized, significantly improving detection efficiency, reducing the degree of manual intervention, and possessing good engineering application value. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 The present invention provides a method flowchart.

[0022] Figure 2 This is a schematic diagram showing the two-dimensional identification results of the slope surface area and the slope platform area. Figure 3 A schematic diagram of point cloud reprojection and semantic assignment; Figure 4 A schematic diagram for finding the intersection of the normal plane and determining the key points. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are all within the protection scope of the present invention.

[0024] Taking the multi-level slope on the right side of the K3499+000~K3499+100 section as an example, this paper illustrates the application of the method of the present invention in the quality verification of slope cross-sectional parameters during the highway reconstruction and expansion construction period. This section already has route centerline data, slope design parameters, aerial imagery, dense point clouds, and camera parameters. Based on the above data, the following steps are performed sequentially: two-dimensional element identification, point cloud semantic assignment, slope grading and segmentation, slope surface fitting, extraction of slope crest and toe lines, intersection of normal planes, and verification of cross-sectional parameters.

[0025] In this invention, the sequence of multi-level slopes is determined according to the slope type. For multi-level cut slopes, each level of slope is numbered sequentially from the toe to the crest, i.e., the lowest level slope is the first level slope, the adjacent level above it is the second level slope, and so on. For multi-level fill slopes, each level of slope is numbered sequentially from the crest to the toe, i.e., the highest level slope is the first level slope, the adjacent level below it is the second level slope, and so on.

[0026] In this invention, the slope type can be determined based on slope design parameters, design cross-section tables, or design model data. First, the platform area mask is identified, and combined with the slope type determination results, the platform areas at each level are sorted according to the corresponding direction. For excavation, the lowest level platform is the breakwater, and from the breakwater upwards are the first-level platform, the second-level platform, and so on. For embankment, the highest level platform is the first-level platform, and the lowest level platform is the slope protection channel. Then, the slope surface object point cloud of the corresponding level is determined by the slope surface range between adjacent platform areas. Specifically, for multi-level excavation slopes, the slope surface range between the lowest platform (breakwater) area and its adjacent slope platform area above it corresponds to the first-level slope object point cloud, and so on, corresponding to the second-level slope object point cloud above it. For multi-level embankment slopes, the slope surface range between the highest slope and the first-level slope platform area corresponds to the first-level slope object point cloud, and so on, corresponding to the second-level slope object point cloud below it.

[0027] This invention provides a method for verifying the quality of highway slope cross-sectional parameters based on aerial imagery and photogrammetric point clouds, such as... Figure 1 As shown, the steps include the following.

[0028] S1. Obtain the route centerline data and slope design parameters of the road section to be verified.

[0029] Route centerline data can be obtained from at least one of the following: design drawings, route design deliverables, or delivered BIM models; route design deliverables, for example, are route deliverables that can be directly read and accessed by the JSL-Route Expert System. Route centerline data includes at least the route station information and the spatial coordinates corresponding to each station, used to determine the mileage position and cross-sectional direction of each section, forming the basis for establishing the geometric positioning along the route. Based on the station numbers and their corresponding spatial coordinates, the mileage position of subsequent normal planes can be determined; based on the route tangential direction calculated from the spatial coordinates of adjacent stations, the direction of each normal plane can be further determined.

[0030] Slope design parameters include at least the design slope height and design slope ratio. In multi-level slope scenarios, they may also include information such as the design height, design slope ratio, platform width, and number of levels for each slope. Slope design parameters can be extracted from design drawings, route design deliverables, or delivered BIM models. Route design deliverables, for example, are route deliverables that can be directly read and accessed by the JSL-Route Expert System.

[0031] In one specific embodiment, the section from K3499+000 to K3499+100 is selected as the section to be verified. The route centerline data and slope design parameters are directly read and output by the JSL-Route Expert System, including design parameters for typical station numbers such as K3499+000, K3499+020, K3499+040, K3499+060, K3499+080 and K3499+100.

[0032] S2. Acquire drone aerial images of the road segment and dense point clouds reconstructed from the aerial images via photogrammetry, and obtain camera parameters obtained from photogrammetric calculations.

[0033] Specifically, the first step is to plan the drone flight path based on the spatial extent, length, height, and occlusion of the slope to be verified, ensuring that the aerial imagery covers the slope area. During aerial photography, one or more flight paths can be set along the slope direction, ensuring sufficient forward overlap (no less than 70%) and lateral overlap (no less than 60%) between adjacent images to guarantee that the slope surface area can be completely captured. After acquisition, the aerial imagery undergoes photogrammetric processing, which includes steps such as image feature extraction, corresponding point matching, aerial triangulation, sparse point cloud reconstruction, and multi-view image dense reconstruction, as detailed below: (1) Image feature extraction and corresponding point matching: Local feature extraction is performed on each aerial image. The feature detection and description algorithm (SIFT) with scale and rotation invariance is used to obtain key points and their feature descriptors in the image.

[0034] Similarity matching is performed between different images based on feature descriptors. The nearest neighbor distance ratio criterion is used to filter matching point pairs, eliminate false matches, and obtain a reliable set of corresponding points.

[0035] (2) Aerial Triangulation (SfM): Based on the matching results of corresponding points, the collinearity equations between multi-view images are constructed, and the interior and exterior orientation parameters of the camera are jointly solved, while simultaneously restoring the initial 3D structure of the scene. The optimization process is achieved by minimizing the reprojection error of image points: in, For image observation points, For camera projection matrix, These are the coordinates of a point in space.

[0036] Through the above calculations, the camera parameters for each aerial image are obtained, including interior orientation parameters (focal length, principal point position, and distortion parameters) and exterior orientation parameters (camera spatial position and attitude angle).

[0037] (3) Sparse point cloud reconstruction: Based on the camera parameters and corresponding point relationships obtained from aerial triangulation, the spatial three-dimensional coordinates of the matching points are recovered by multi-view geometric triangulation method to generate a sparse point cloud describing the overall structure of the slope.

[0038] (4) Dense Reconstruction of Multi-View Imagery (MVS): Based on the obtained camera parameters and sparse point cloud, a dense three-dimensional reconstruction of the slope area is performed using a multi-view stereo reconstruction method. Specifically, this includes: View selection: Select a neighboring image with sufficient parallax for each reference image; Depth estimation: Based on photometric consistency constraints, the depth value of each pixel is estimated at the pixel level; Deep fusion: Fusion of depth information from multiple views to restore high-density 3D points; Its optimization objective function can be expressed as: in, The first term represents the matching cost of pixels under different viewpoints, and the second term represents the depth smoothing constraint.

[0039] The final result is a dense point cloud covering the slope area.

[0040] In one specific embodiment, drone aerial photography was conducted on the right-side slope of the road section from K3499+000 to K3499+100, acquiring a total of 128 aerial images. Through the aforementioned photogrammetric processing flow, dense point cloud data covering the slope of this section was obtained, and the focal length, principal point coordinates, distortion parameters, camera spatial position, and attitude angle corresponding to each image were acquired.

[0041] S3. Perform two-dimensional feature recognition on aerial images and output slope surface area masks and platform area masks.

[0042] Specifically, a deep learning-based semantic segmentation method is used to perform pixel-level classification of aerial images. First, a training dataset is constructed, and the aerial images are manually labeled, with at least two categories: slope surface and platform. During model training, a semantic segmentation network (DeepLabv3+) is selected and trained under supervision using the labeled data as input, resulting in a semantic segmentation model capable of pixel-level classification of aerial images. During inference, the aerial images to be processed are input into the trained model, and the model outputs the labeled category for each pixel, thus obtaining multi-class segmentation results. Based on the segmentation results, the pixel set belonging to the slope surface is extracted to generate a slope surface region mask; simultaneously, the platform region between adjacent slope surfaces is extracted to generate a platform region mask.

[0043] When focusing on a single target (such as a slope surface), the corresponding pixel can be marked as 1, and the remaining pixels as 0, forming a binary mask. When it is necessary to distinguish multiple land cover categories (such as slope surfaces and platforms), different label values ​​can be assigned to different categories, forming a multi-category mask. In multi-level slope scenarios, the platform area mask is used to represent the spatial range and boundary location of different platforms, providing constraint information for subsequent point cloud-based hierarchical segmentation. For single-level slopes, only the slope surface area mask is generated, and the platform area mask does not participate in subsequent hierarchical processing.

[0044] In one specific embodiment, semantic segmentation is performed on aerial images with a resolution of 2 cm / pixel in the K3499+000~K3499+100 section to obtain slope surface area masks, and the first-level slope platform areas are identified, forming corresponding platform area masks. Through the above processing, two-dimensional slope recognition results can be obtained, such as... Figure 2 As shown.

[0045] S4. Point cloud reprojection and semantic assignment based on camera parameters: Based on the camera intrinsic and extrinsic parameters obtained by photogrammetry, the 3D points in the dense point cloud are reprojected onto the aerial image plane covering the area where they are located. According to the correspondence between the reprojected pixels and the slope surface mask and the platform area mask, the 3D points are semantically assigned, thereby obtaining the slope surface point cloud and the platform area point cloud respectively.

[0046] Specifically, for any three-dimensional point P(X, Y, Z) in the dense point cloud, based on its three-dimensional coordinates and the internal and external parameters of the camera corresponding to the aerial image, it is projected onto the image plane to obtain pixel coordinates (u, v). Among them, only when the three-dimensional point meets the projection validity condition does it participate in the subsequent semantic determination; the projection validity conditions include: the three-dimensional point is in front of the camera (i.e., the depth value in the camera coordinate system is greater than 0), and its projected pixel coordinates meet the image range constraint conditions 0 ≤ u < W, 0 ≤ v < H; where W and H are the width and height of the image respectively.

[0047] On this basis, based on the semantic mask with the same resolution and coordinate system as the aerial image, the category determination of the projected pixels is performed. The semantic mask is a pixel-level label map, where different pixel values correspond to the slope surface category and different levels of platform categories respectively. When the mask label corresponding to the projected pixel (u, v) is the preset slope surface category, then this three-dimensional point is given the slope surface semantic label; when the mask label corresponding to the projected pixel is the k-th level slope platform category, then this three-dimensional point is given the platform area semantic label corresponding to the level.

[0048] When the same three-dimensional point can be projected onto multiple overlapping images, the aerial image that meets the following conditions is preferentially selected as the basis for semantic assignment: the projected pixel is within the valid range of the image, the image spatial resolution (ground sampling distance) is higher, and the imaging angle has a smaller included angle with the slope surface normal.

[0049] When the three-dimensional point meets any of the following conditions, no semantic assignment is performed on it: the projected pixel coordinates exceed the image range; the three-dimensional point is behind the camera; the mask label corresponding to the projected pixel is an invalid category (such as an occlusion area, an unrecognized area, or a background area); the multi-image projection results are inconsistent and do not meet the consistency determination conditions.

[0050] Through the above method, the semantic recognition result in the two-dimensional aerial image is mapped to the three-dimensional point cloud, so as to obtain the point cloud data with slope surface and semantic labels of each level of platform, providing a basis for subsequent slope area segmentation, structural analysis, and objectified expression. The process of point cloud reprojection and semantic assignment is shown as Figure 3 shown.

[0051] S5. In the multi-level slope scenario, use the platform area point cloud described in step S4 to segment the slope surface point cloud of the slope, forming the slope surface object point cloud of each level; in the single-level slope scenario, the slope surface point cloud of the slope is taken as the single-level slope surface object point cloud as a whole.

[0052] Specifically, in multi-level slope scenarios, the platform region point cloud is used as the hierarchical boundary constraint to divide the slope surface point cloud into multiple independent slope surface object point clouds corresponding to adjacent platform regions. In other words, the point clouds of different platform regions are used to represent the separation positions between different levels of slopes. With the help of the separation positions, the overall slope surface point cloud can be hierarchically segmented to obtain the first-level slope surface object point cloud, the second-level slope surface object point cloud, and so on. In single-level slope scenarios, hierarchical segmentation based on platform regions is no longer performed. The entire slope surface point cloud is treated as a single-level slope surface object point cloud and enters the subsequent slope fitting and cross-sectional parameter calculation steps.

[0053] In one specific embodiment, based on the identified first-level platform area point cloud, the slope surface point cloud is divided into two levels of independent slope object point clouds, corresponding to the first-level slope surface and the second-level slope surface, respectively.

[0054] S6. Slope Fitting: Perform slope fitting on the point cloud of slope objects at each level in step S5.

[0055] Specifically, planar models are fitted to the point clouds of slope objects at each level to obtain fitted slope surfaces representing the geometric morphology of each level of slope. Slope surface fitting is based on the least squares method, and the specific process is as follows: First, for a point cloud of a slope surface object of a certain level, let it contain 3D points. Construct the planar model expression: Secondly, substitute all points into the above planar model to construct the error function: By minimizing the error function E, the plane parameters a, b, c are solved to obtain the optimal fitting plane.

[0056] Furthermore, to improve the robustness of the fitting, a weighted least squares method is introduced during the fitting process to reduce the impact of noise points on the fitting results.

[0057] Finally, the above fitting process is performed on the point clouds of slope objects at each level to obtain the corresponding fitted slope surfaces, which are then used for the stable extraction of the slope top line and slope toe line.

[0058] In one specific embodiment, least squares plane fitting is performed on the point cloud of the first-level slope surface object and the point cloud of the second-level slope surface object to obtain the corresponding fitted slope surface.

[0059] S7. Extraction of slope top and slope toe lines: Extract the slope top and slope toe lines corresponding to each level of slope from the fitted slope surface.

[0060] Specifically, based on the boundary positions between the fitted slope surfaces of each level and their adjacent platform areas or outer boundaries, the corresponding slope top lines and slope toe lines for each level of slope are extracted. For a general-level slope, its slope toe line corresponds to the boundary line between the lower side of the slope and the adjacent platform, and its slope top line corresponds to the boundary line between the upper side of the slope and the adjacent platform.

[0061] Specifically, for the last level of the excavation slope, the toe line of the slope is extracted based on the boundary between the slope surface and the platform of the last level of the slope, and the outer boundary of the other side of the slope is used as the top line of the slope; for the first level of the fill slope, the toe line of the slope is extracted based on the boundary between the slope surface and the platform of the first level of the slope, and the outer boundary of the other side of the slope is used as the top line of the slope.

[0062] In a multi-level slope scenario, different levels of slopes have their own corresponding top and bottom lines. Through the above extraction process, the top and bottom lines of the first-level slope, the top and bottom lines of the second-level slope, etc., can be obtained respectively, providing a line element basis for subsequent intersection of the normal plane.

[0063] In one specific embodiment, taking the right-side secondary excavation slope as an example, the toe line of the secondary slope is extracted based on the boundary position between the fitted slope surface of the secondary slope and the platform of the primary slope, and the outer boundary of the toe line of the secondary slope is extracted as the top line of the secondary slope; the top line and toe line of the primary slope are extracted based on the boundary position between the fitted slope surface of the primary slope and the platform of the primary slope and the toe platform (falling platform).

[0064] S8. Normal plane extraction: Extract multiple normal planes perpendicular to the tangent of the route along the centerline of the route at fixed station intervals.

[0065] Specifically, based on the route centerline data obtained in step S1, multiple normal planes are laid out along the route direction at fixed station intervals. Each normal plane passes through the spatial coordinate point of the corresponding station, and its normal vector is consistent with the tangential direction of the route centerline at that station. The fixed station interval can be set according to the verification accuracy requirements, such as 5m, 10m, 20m, or other suitable spacing.

[0066] In engineering, the normal plane is used to construct the cross-sectional analysis surface at the corresponding station number so that the intersection of the normal plane with the slope surface, slope top line and slope toe line of each level can be calculated later.

[0067] In one specific embodiment, the plane is extracted along the K3499+000~K3499+100 section at 20m station intervals.

[0068] S9. Intersection Point Determination and Key Point Selection: Calculate the intersection points of each plane with the slope surface, slope top line, and slope toe line of each slope level. Use these intersection points as the key points at the slope top and slope toe of each slope level, respectively.

[0069] Specifically, for any normal plane, its plane equation can be expressed as: The slope surfaces at each level are planes obtained through fitting in step S6, and their equations are as follows: First, the normal plane π is compared with the slope surface of the i-th level slope. Solving the system of equations simultaneously yields the intersection line of the two planes, which serves as the cross-sectional line Li of the slope at this level within the normal plane. The cross-sectional line can be expressed in parametric form: in, The direction vector of the intersection line. For any point on the intersection line.

[0070] Furthermore, the cross-sectional line Li is intersected with the corresponding slope crest line and slope toe line for that slope level. The slope crest line and slope toe line are three-dimensional polylines composed of multiple line segments. Parametric equations are established for each line segment and solved simultaneously with the cross-sectional line to obtain a set of candidate intersection points.

[0071] When multiple intersections exist between the cross-sectional line and the slope crest or toe line, key points are selected based on elevation: candidate intersections are sorted by elevation value, the intersection with the highest elevation is selected as the slope crest key point, and the intersection with the lowest elevation is selected as the slope toe key point. After the above selection, the slope crest and slope toe key points of that level of slope are determined in the same method plane. For multi-level slopes, the above steps are performed on each level of slope in the same method plane to obtain multiple sets of slope crest and slope toe key points, corresponding to different levels of slope structure. These key points can be further used to calculate the slope height and slope ratio of each level of slope.

[0072] In one specific embodiment, for a given normal plane, it is first intersected with the slope surface of the first-level slope to obtain the cross-sectional line of the slope within the normal plane. Then, the cross-sectional line is intersected with the top and bottom lines of the first-level slope to obtain a set of candidate intersection points. Based on the elevation criterion, a unique top and bottom key point are selected from the candidate intersection points. Similarly, the above process can be repeated for the second and subsequent levels of slopes, calculating their corresponding cross-sectional lines and intersecting them with the top and bottom lines of each level of slope. The same selection rules are used to determine the top and bottom key points of each level of slope. The process of normal plane intersection and key point selection is illustrated below. Figure 4 As shown.

[0073] S10. Calculation of cross-sectional parameters and determination of exceeding limits: Based on the key points at the top and bottom of the slope in step S9, calculate the measured slope height and measured slope ratio, and compare the measured slope height and measured slope ratio with the designed slope height and designed slope ratio respectively to obtain the slope height deviation and slope ratio deviation. Determine whether the cross-section exceeds the limit according to the preset threshold rules, and output the mileage location, the range of the exceeding section, and the statistical results of the exceeding section.

[0074] Specifically, the measured slope height is calculated based on the elevation difference between the key points at the top and bottom of the slope; the measured slope ratio is calculated based on the vertical elevation difference and lateral distance of the corresponding slope levels within the normal plane. For multi-level slopes, the measured height and slope ratio of each level can be calculated separately and compared with the corresponding design parameters to obtain the height and slope deviations of each level of slope.

[0075] Then, the system determines whether the corresponding cross-section exceeds the limits based on preset threshold rules. The threshold rules are as follows: if the absolute value of the slope height deviation exceeds the height threshold, the height is considered excessive; if the absolute value of the slope rate deviation exceeds the slope rate threshold, the slope rate is considered excessive; if either the height or slope rate exceeds the limit, the corresponding cross-section is considered excessive. After determining all cross-sections, the excessive cross-sections are organized in mileage order. If several adjacent cross-sections continuously exceed the limits, they are merged and output as continuously excessive sections based on mileage continuity, and the number of excessive cross-sections, the number of continuously excessive sections, and the corresponding section range are counted. The slope height deviation threshold is generally 0.15m, and the slope rate deviation threshold is generally 0.05m, which can be manually set.

[0076] In one specific embodiment, taking the secondary slope of the excavation to the right of K3499+000, K3499+020, K3499+040, K3499+060, K3499+080 and K3499+100 as an example, the slope height and slope ratio obtained by plane calculation of each station number are shown in Table 1.

[0077] Table 1 Slope Calculation Results The slope height deviation threshold is set at ±0.15m, and the slope ratio deviation threshold is set at ±0.05m. The output conclusion is as follows: In the secondary slope of the right-side excavation from K3499+000 to K3499+100, the station numbers and segments where the height of the primary slope exceeds the limit are K3499+060 to K3499+100; there are no station numbers or continuous exceeding sections where the slope ratio of the primary slope exceeds the limit. The station numbers and continuous exceeding sections where the height of the secondary slope exceeds the limit are K3499+000 to K3499+020 and K3499+080; the station numbers and continuous exceeding sections where the slope ratio of the secondary slope exceeds the limit are K3499+000 to K3499+080.

[0078] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for verifying the quality of highway slope cross-sectional parameters based on aerial imagery and photogrammetric point clouds, characterized in that, Includes the following steps: S1. Obtain the route centerline data and slope design parameters of the road section to be verified; S2. Acquire drone aerial images of the road segment and dense point clouds reconstructed from the aerial images via photogrammetry; and acquire camera parameters obtained from photogrammetry calculations. S3. Perform two-dimensional element recognition on the aerial image and output the slope surface area mask and the platform area mask. S4. Based on the camera parameters, the three-dimensional points in the dense point cloud are reprojected onto the aerial image plane covering the area where they are located. According to the correspondence between the reprojection landing point and the slope surface area mask and the platform area mask, the three-dimensional points are semantically assigned to obtain the slope surface point cloud and the platform area point cloud respectively. S5. In a multi-level slope scenario, the slope surface point cloud is segmented using the platform area point cloud described in step S4 to form slope surface object point clouds at each level; in a single-level slope scenario, the slope surface point cloud as a whole is treated as a single-level slope surface object point cloud. S6. Perform slope surface fitting on the point cloud of slope surface objects at all levels described in step S5. S7. Extract the slope crest line and slope toe line corresponding to each level of the fitted slope surface; S8. Extract multiple normal planes perpendicular to the tangent of the route along the centerline of the route at fixed station intervals; S9. Calculate the intersection points of each plane with the slope surface, slope top line and slope toe line of each level of slope, and use the intersection points as the key points of the slope top and slope toe of each level of slope, respectively. S10. Based on the key points at the top and bottom of the slope described in step S9, calculate the measured slope height and measured slope ratio, and compare the measured slope height and measured slope ratio with the designed slope height and designed slope ratio respectively to obtain the slope height deviation and slope ratio deviation. Determine whether the cross section exceeds the limit according to the preset threshold rule, and output the mileage location, the range of the exceeding section, and the statistical results of the exceeding section.

2. The method according to claim 1, characterized in that, In step S1, the route centerline data includes at least the route station information and the spatial coordinates corresponding to each station; the slope design parameters include at least the design slope height and the design slope ratio.

3. The method according to claim 1, characterized in that, In step S2, the aerial image is subjected to photogrammetric processing, which includes image feature extraction, corresponding point matching, aerial triangulation, sparse point cloud reconstruction, and multi-view image dense reconstruction.

4. The method according to claim 1, characterized in that, In step S3, the two-dimensional element recognition is achieved using a semantic segmentation method based on deep learning. The platform area includes the debris platform, the slope protection road, and the slope platform areas of each level between two adjacent slopes in a multi-level slope.

5. The method according to claim 1, characterized in that, In step S4, if the reprojection point of the three-dimensional point is located within the slope surface area mask, then the three-dimensional point is assigned a slope surface semantic label; if the reprojection point is located within the slope platform area mask at each level, then the three-dimensional point is assigned a corresponding slope platform area semantic label.

6. The method according to claim 1, characterized in that, In step S5, the point cloud of each level of slope platform area is used as the hierarchical boundary constraint to divide the slope surface point cloud into multiple independent slope object point clouds corresponding to the adjacent slope platform areas.

7. The method according to claim 1, characterized in that, In step S6, the slope fitting is a least squares plane fitting; in step S7, the top line and toe line of the slope are extracted based on the boundary position between the fitted slope surface of each level and its adjacent platform area or outer boundary; for general level slopes, the upper boundary line is used as the top line and the lower boundary line is used as the toe line; for the last level of cut slopes and the first level of fill slopes, the toe line is extracted based on the boundary position between the slope surface of the slope level and its adjacent platform area, and the outer boundary on the other side of the slope level is used as the top line.

8. The method according to claim 1, characterized in that, In step S8, the normal plane passes through the spatial coordinate point of the corresponding station number, and its normal vector is consistent with the tangential direction of the route centerline at that station number.

9. The method according to claim 1, characterized in that, In step S9, the intersection of each normal plane with the slope surface of each level is used to obtain the cross-sectional curve of the corresponding level slope. The intersection of the cross-sectional curve with the top line and toe line of the corresponding level slope is further calculated as the key points of the top and toe of the slope for that level slope.

10. The method according to claim 1, characterized in that, In step S10, the measured slope height is calculated based on the elevation difference between the key point at the top of the slope and the key point at the bottom of the slope, and the measured slope ratio is calculated based on the vertical elevation difference and lateral distance of the corresponding slope level in the normal plane. When multiple adjacent cross sections exceed the limit consecutively, they are merged and output as a continuous over-limit section according to the mileage continuity.

Images