Bridge bearing cushion stone construction precision intelligent detection method based on unmanned aerial vehicle oblique photography

Abstract

The application discloses a bridge support cushion stone construction precision intelligent detection method based on unmanned aerial vehicle oblique photography, which comprises the following steps: S100, a space coordinate system is established, and a feature point on a bridge pier is selected; S200, a three-dimensional model of the bridge pier is established in the space coordinate system, and a reference model of the bridge support cushion stone is added, and reference point cloud data of the bridge support cushion stone are acquired; S300, the position of the unmanned aerial vehicle in the space coordinate system is acquired, and the unmanned aerial vehicle is used to synchronously collect image information of the bridge support cushion stone; S400, an actual model of the bridge support cushion stone is established in the space coordinate system, and actual point cloud data of the bridge support cushion stone are acquired; and S500, the reference point cloud data and the actual point cloud data of the bridge support cushion stone are compared, and whether the bridge support cushion stone is qualified or not is judged. Through the introduction of the unmanned aerial vehicle non-contact image collection and three-dimensional space digital comparison, the automation and the intelligentization of the detection process are realized.

Description

Technical Field

[0001] This invention belongs to the field of visual inspection technology, and more specifically, relates to an intelligent inspection method for the construction accuracy of bridge bearing pad stones based on UAV oblique photography. Background Technology

[0002] Bridge bearing pads are critical load-bearing components connecting the superstructure and substructure of a bridge. Their primary function is to safely and evenly transfer the weight and live load of the bridge superstructure to the piers. The quality of their construction directly affects the overall stability, stress state, and service life of the bridge structure. Therefore, during the bridge construction acceptance and handover process, as well as the subsequent safe operation and maintenance phases, it is essential to conduct rigorous and high-precision testing of the geometric dimensions and spatial position of each bearing pad to ensure that it meets the original engineering drawing specifications.

[0003] Currently, the industry's standard testing methods for bridge bearing pad stones mainly rely on traditional manual, on-site contact measurements. In practice, inspectors typically need to use auxiliary equipment such as aerial work platforms, scaffolding, or bridge inspection vehicles to climb to the top of bridge piers tens or even hundreds of meters high. They then manually use measuring instruments such as steel tape measures, levels, total stations, or RTK rover stations to directly contact the pad stone surface for manual point marking and sampling measurements. Finally, engineering personnel bring the scattered dimensional data recorded on-site back indoors and manually convert and compare it with two-dimensional construction drawings to determine whether the pad stone construction is up to standard.

[0004] However, the aforementioned existing technologies have many insurmountable drawbacks in practical engineering applications: First, the high-altitude working environment is complex, and the risk of manually climbing to the top of the bridge piers is extremely high, posing serious safety hazards; second, for large-scale bridge projects, manually climbing and surveying each pier and each bearing stone is time-consuming and laborious, with extremely low detection efficiency, seriously hindering the overall project progress; most importantly, the instruments cannot measure the bearing stones from below, resulting in blind spots, so it is necessary to use drones to collect images for measuring the bearing stones, which places high demands on the coordinate accuracy and modeling accuracy of the drones. Summary of the Invention

[0005] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides an intelligent detection method and system for bridge bearing pad construction accuracy based on UAV oblique photography. By introducing UAV non-contact image acquisition and three-dimensional spatial digital comparison, the detection process is automated and intelligent. This changes the dangerous operation mode of traditional manual high-altitude climbing and on-site contact measurement, solves the problem of blind spots in the field of vision when the instrument is measured from below, eliminates the safety hazards of high-altitude operations, and upgrades time-consuming and laborious local spot checks into efficient and high-precision comprehensive inspections, significantly improving the overall efficiency of bridge engineering quality inspection.

[0006] To achieve the above objectives, according to a first aspect of the present invention, a method for intelligent detection of bridge bearing pad stone construction accuracy based on UAV oblique photography is provided, comprising: S100. Set up one or more control points, establish a spatial coordinate system, select three or more feature points on the bridge pier and mark their positions in the spatial coordinate system. S200. Based on the construction drawings, establish a reference model of the pier cap beam and bearing pad in the spatial coordinate system, and obtain the reference point cloud data of the bridge bearing pad from the reference model. S300: Establish a connection between the UAV and the spatial coordinate system, obtain the position of the UAV in the spatial coordinate system in real time, and use the UAV to synchronously collect image information of the bridge bearing pad stone in five directions: top, front, back, left, and right. S400: Using the collected image information, the actual model of the pad stone and the cap beam is reconstructed. Then, by using the absolute position coordinates of the control points and the relative position relationship between the actual model and the control points, the absolute position coordinates and absolute elevation of the final pad stone are calculated in reverse, thereby generating actual point cloud data with high-precision absolute elevation information in the spatial coordinate system. S500: Compare the reference point cloud data and the actual point cloud data of the bridge bearing pads to obtain the deviation value of each bridge bearing pad and determine whether the bridge bearing pads are qualified.

[0007] Furthermore, in step S300, the airborne positioning module needs to be associated with the spatial coordinate system established in step S100, and the corresponding coordinate transformation parameters need to be input so that the UAV's control and data storage unit can receive differential correction signals in real time, thereby achieving high-precision three-dimensional spatial positioning; at the same time, the UAV's built-in inertial measurement unit is used to acquire the UAV's attitude angle data during flight in real time at high frequency.

[0008] Furthermore, the drone performs automated autonomous flight according to the preset route. When it reaches each preset exposure point on the route, the control and data storage unit sends a synchronous trigger signal to control the five-lens camera to expose simultaneously, thereby synchronously completing the image acquisition of five different perspectives from the top, front, back, left and right of the bridge bearing pad stone. At the moment each exposure is triggered by the five-lens camera, the drone's time synchronization unit binds the current exposure time, spatial coordinates, and attitude angle data to generate POS data corresponding to each image. This POS data is then used to generate an independent data log file, thereby assigning each acquired image a high-precision absolute spatial position and initial attitude value. This provides essential exterior orientation elements for the accurate spatial establishment of the actual point cloud model.

[0009] Furthermore, in step S400, when establishing the actual model of the bridge bearing pad stone in the spatial coordinate system, the exposure timestamp is used as the association medium to match and align the collected images in the five dimensions of top, front, back, left, and right with the corresponding three-dimensional coordinates and IMU attitude angle data to determine the absolute position and rotation angle of the image in space. Then, computer vision algorithms were used to perform feature point detection and cross-view matching on the full set of images, identify the same feature points corresponding to the same geometric part of the pad stone in different images, and construct a complex network of image connection points.

[0010] Furthermore, using the initial pose data of the image as a constraint, aerial triangulation calculations are performed through bundle adjustment. The actual spatial position and lens distortion parameters of each photo are accurately corrected through iterative calculation, thereby generating a sparse point cloud that reflects the basic outline of the pad stone. Based on this, by using dense matching of multi-view images, and based on geometric consistency and pixel grayscale information, the depth of each pixel of the pad stone is estimated, and the two-dimensional image information is projected back into the three-dimensional space. Finally, by combining the coordinates of the feature points marked in step S100 for absolute orientation and scale scaling, background noise points are filtered out, thereby extracting high-density, high-precision actual point cloud data of bridge bearing pad stones, and completing the actual model construction in the spatial coordinate system.

[0011] Furthermore, in step S500, when comparing the reference point cloud data and the actual point cloud data of the bridge bearing pad stone, it is necessary to preprocess the actual point cloud data obtained in step S400, and use statistical filtering or radius filtering algorithms to remove floating noise points and redundant point clouds in non-pad stone areas generated during UAV acquisition and 3D reconstruction, so as to obtain clean actual point cloud data of pad stone.

[0012] Furthermore, in order to eliminate systematic errors, point cloud registration is required to achieve fine fitting. At this time, the closest point pair is found in the reference point cloud and the actual point cloud, the rotation and translation matrix between the two sets of point clouds is calculated, and the process is iterated until the mean square error between the two sets of point clouds is minimized or the convergence condition is met.

[0013] Furthermore, deviation calculation is required. A distance calculation algorithm is used to calculate the shortest normal distance from each 3D point in the registered actual point cloud to the reference point cloud. When calculating deviation values, for the engineering inspection requirements of bridge bearing pad stones, the following core dimensions of deviation values ​​are extracted and calculated: Planar position deviation: Extract the coordinates of the center point of the top surface of the pad stone, and calculate the planar offset of the X and Y axes with the coordinates of the center point of the reference model; Elevation deviation: Extract the average value of the Z-axis coordinate set of the point cloud on the top surface of the pad stone, and calculate the difference with the design elevation of the reference model; Flatness deviation: Calculate the variance of the vertical distance from each point on the top surface of the actual point cloud to the fitted plane.

[0014] Furthermore, a chromatographic mapping relationship is established to generate a visualized three-dimensional chromatographic deviation map of the pad stone; At the same time, the calculated deviation values ​​are automatically compared with the preset allowable error thresholds of the bridge construction specifications. If all deviation values ​​are within the allowable threshold range, the system determines that the bridge bearing pad is qualified. If any value exceeds the threshold, it is determined to be unqualified, and a report of the specific location and deviation value of the deviation is output.

[0015] According to a second aspect of the present invention, an intelligent detection system for the construction accuracy of bridge bearing pad stones based on UAV oblique photography is provided, comprising: Reference establishment module: used to set up one or more control points, establish a spatial coordinate system, select three or more feature points on the bridge pier and mark their positions in the spatial coordinate system; Data acquisition module: used to establish reference models of bridge pier cap beams and bearing pads in a spatial coordinate system in conjunction with construction drawings, and to acquire reference point cloud data of bridge bearing pads from the reference model; Image acquisition module: used to establish a connection between the UAV and the spatial coordinate system, obtain the position of the UAV in the spatial coordinate system in real time, and use the UAV to synchronously acquire image information of the bridge bearing pad stone in five directions: top, front, back, left, and right. Point cloud acquisition module: It is used to reconstruct the actual model of the pad stone and cap beam using the acquired image information. Then, by using the absolute position coordinates of the control points and the relative position relationship between the actual model and the control points, it reversely calculates the final absolute position coordinates and absolute elevation of the pad stone, thereby generating actual point cloud data with high-precision absolute elevation information in the spatial coordinate system. Data Judgment Module: Used to compare the reference point cloud data and the actual point cloud data of bridge bearing pad stones, obtain the deviation value of each bridge bearing pad stone, and determine whether the bridge bearing pad stone is qualified.

[0016] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: 1. The intelligent inspection method for bridge bearing pad stones of the present invention, by introducing non-contact image acquisition and three-dimensional spatial digital comparison by UAV, realizes the automation and intelligence of the inspection process, changes the dangerous operation mode of traditional manual high-altitude climbing and on-site contact measurement, solves the problem of blind spots in the field of vision of the instrument when measuring from below, eliminates the safety hazards of high-altitude operation, and upgrades the time-consuming and laborious local sampling inspection to a high-efficiency and high-precision comprehensive survey, which greatly improves the overall efficiency of bridge engineering quality inspection.

[0017] 2. The intelligent detection method for bridge bearing pad stones of the present invention uses known and accurate control points to calculate the elevation of the pad stones. By using a drone to take close-up photos, the actual model of the pad stones can be measured. Only the actual relative coordinates of the model measured by the drone are needed. The absolute position coordinates of the final pad stones can be calculated by using the absolute position coordinates and relative position relationships of the known control points, thereby avoiding the drawbacks caused by the insufficient accuracy of the absolute coordinates of the drone itself.

[0018] 3. The intelligent detection method for bridge bearing pad stones of the present invention achieves synchronous acquisition of images of the top surface and surrounding sides of the pad stone without blind spots through the linkage of a multi-view array camera and a high-precision spatial positioning module. This effectively overcomes the problems of visual blind spots and omission of details that are easily generated by traditional single-view shooting, and ensures the integrity of the underlying image features.

[0019] 4. The intelligent detection method for bridge bearing pad stones of the present invention adopts high-precision time alignment technology to bind multi-view two-dimensional images with three-dimensional spatial absolute coordinates and attitude data in milliseconds. Through the three-dimensional feature space solution of multi-view images, the two-dimensional images are reverse-converted into high-density three-dimensional actual point clouds with high fidelity, restoring the micro-undulation morphology of the pad stone surface and effectively eliminating the subjective error caused by sparse manual measurement points. Attached Figure Description

[0020] Figure 1 This is a schematic diagram illustrating the steps of an intelligent detection method for bridge bearing pad stones using oblique photography by an unmanned aerial vehicle (UAV) according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating an intelligent detection method for bridge bearing pad stones using oblique photography by an unmanned aerial vehicle (UAV) according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the distribution of bridge bearing pad stones in an intelligent detection method for bridge bearing pad stones using oblique photography by unmanned aerial vehicles, according to an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0022] Example 1 like Figure 1-3 As shown, this embodiment of the invention provides an intelligent detection method for the construction accuracy of bridge bearing pad stones based on UAV oblique photography, including the following steps: S100. Set up one or more control points, establish a spatial coordinate system, select three or more feature points on the bridge pier and mark their positions in the spatial coordinate system. S200. Based on the construction drawings, establish a reference model of the pier cap beam and bearing pad in the spatial coordinate system, and obtain the reference point cloud data of the bridge bearing pad from the reference model. S300: Establish a connection between the UAV and the spatial coordinate system, obtain the position of the UAV in the spatial coordinate system in real time, and use the UAV to synchronously collect image information of the bridge bearing pad stone in five directions: top, front, back, left, and right. S400: Using the collected image information, the actual model of the pad stone and the cap beam is reconstructed. Then, by using the absolute position coordinates of the control points and the relative position relationship between the actual model and the control points, the absolute position coordinates and absolute elevation of the final pad stone are calculated in reverse, thereby generating actual point cloud data with high-precision absolute elevation information in the spatial coordinate system. S500: Compare the reference point cloud data and the actual point cloud data of the bridge bearing pads to obtain the deviation value of each bridge bearing pad and determine whether the bridge bearing pads are qualified.

[0023] In step S100, control points are established on the bottom surface of the cap beam or on the pier. The coordinates of each control point can be directly measured below the cap beam, thereby acquiring the three-dimensional coordinates of each control point in the engineering independent coordinate system or the national geodetic coordinate system, and establishing a spatial coordinate system for global data registration. Then, at least three non-collinear points are selected as feature points at structurally stable locations such as the top of the pier, the edge of the cap beam, or the foundation stone base. These feature points can be natural features such as obvious geometric corners on the pier surface, or special targets with high-contrast colors that are pre-pasted manually. Finally, using non-contact surveying equipment such as a total station, with the established ground control points as the station reference, the distance and angle of each selected feature point are accurately measured, the absolute three-dimensional coordinate values ​​of these three feature points in the spatial coordinate system are calculated and recorded, thereby completing the position marking of the feature points in the spatial coordinate system.

[0024] In step S200, since the purpose of this invention is to inspect bridge bearing pads, the prerequisite is that the error of the pier structure is within the design range. Therefore, the design dimensions of the pier are directly used for modeling. Then, a reference model of the bridge bearing pads is built on the 3D model of the pier according to the design dimensions. When modeling the pier, the feature points obtained in step S100 are used for position calibration. With the shape and size of the pier determined, determining the coordinates of its three feature points completes the modeling of the entire pier. In addition, using the design dimensions of the pier to directly inspect the bridge bearing pads can eliminate the errors inherent in the pier itself and avoid the overall error exceeding the design value after the superposition of multiple structural errors.

[0025] In step S200, key design parameters are first extracted from the bridge construction drawings or BIM design files, including the design length, width, and height dimensions, design elevation, and relative geometric positional relationship between the piers and bridge bearing pads. Then, using 3D modeling software, a standard 3D solid model of the piers and bearing pads is constructed based on these design parameters. Combined with the feature points already marked in step S100, a 3D coordinate system transformation algorithm is used to accurately map and anchor this standard 3D solid model to the established spatial coordinate system, forming a reference model with absolute spatial coordinate positions. Next, according to a preset sampling resolution or point spacing, the outer surface of the bearing pad reference model in the spatial coordinate system is discretized or virtually scanned to extract the theoretical 3D coordinate values ​​of a large number of discrete points on the pad surface, thereby generating and exporting reference point cloud data for the bridge bearing pads.

[0026] In step S300, the UAV is equipped with a multi-lens oblique photography camera matrix, an airborne positioning unit, an inertial measurement unit, a time synchronization unit, and a control and data storage unit.

[0027] The multi-lens tilting camera matrix comprises five independent high-resolution camera modules, which are mounted together on a lightweight gimbal beneath the drone. One camera lens captures orthophotos vertically downwards, while the other four camera lenses capture tilted images at preset fixed tilt angles in the forward, backward, left, and right directions of the flight path.

[0028] The airborne positioning unit includes a GNSS antenna and an RTK data processing motherboard. By receiving satellite navigation signals and real-time differential correction signals from ground reference stations, it performs carrier phase differential calculation to provide the UAV with real-time absolute three-dimensional coordinates, and then transforms the coordinates into a spatial coordinate system.

[0029] The inertial measurement unit is equipped with a high-precision three-axis gyroscope and a three-axis accelerometer, which measure the three-dimensional attitude angles of the UAV in flight and at the moment of camera exposure in real time, including heading angle, pitch angle and roll angle.

[0030] The time synchronization unit adds a high-precision timestamp at the instant the five-lens camera shutter triggers exposure, and extracts the center coordinate data output by the onboard positioning unit and the attitude data output by the IMU at the same moment. These three data are forcibly bound together to generate a positioning and orientation data log.

[0031] The control and data storage unit is used to execute pre-planned wrap-around 3D flight paths, send trigger commands to the camera according to the exposure interval set by the flight path, and perform high-speed writing and secure storage of the massive amounts of high-resolution image data and corresponding POS attitude log files.

[0032] In step S300, the airborne positioning module needs to be associated with the spatial coordinate system established in step S100, and the corresponding coordinate transformation parameters need to be input so that the UAV's control and data storage unit can receive differential correction signals in real time, thereby achieving high-precision three-dimensional spatial positioning; at the same time, the UAV's built-in inertial measurement unit is used to acquire the UAV's attitude angle data during flight in real time at high frequency.

[0033] During image acquisition by the UAV, based on the 3D model of the bridge pier and the location of the bearing pads in step S200, a 3D orbital flight path encircling the bridge bearing pads is pre-planned in the UAV ground station software. During path planning, a safe flight distance and speed between the UAV and the bearing pads are set, along with high overlap parameters that meet photogrammetry requirements, such as a forward overlap of at least 80% and a lateral overlap of at least 70%, to ensure no blind spots during subsequent model reconstruction. The UAV performs automated autonomous flight according to the preset flight path. Upon reaching each preset exposure point on the path, the control and data storage unit sends a synchronization trigger signal, controlling the five-lens camera to simultaneously expose, thereby synchronously acquiring images of the bridge bearing pads from five different perspectives: top, front, back, left, and right. At the moment each exposure is triggered by the five-lens camera, the drone's time synchronization unit binds the current exposure time, spatial coordinates, and attitude angle data to generate POS data corresponding to each image. This POS data is then used to generate an independent data log file, thereby assigning each acquired image a high-precision absolute spatial position and initial attitude value. This provides essential exterior orientation elements for the accurate spatial establishment of the actual point cloud model.

[0034] In step S300, the set of images acquired by the UAV's five-lens camera during one orbital flight is as follows: Each image The exposure time is given a unique absolute timestamp by the hardware time synchronization unit: in, This is a collection of all bridge bearing pad stone images acquired by five cameras from a drone. For the first Zhang Image For the first The exposure timestamp of the image. This represents the total number of images. For the time domain; The position time series output by the airborne GNSS / RTK is as follows: in, Output timestamp sequences for GNSS positioning. For the first GNSS positioning time , Number of GNSS positioning sampling points; The attitude time series output by the IMU is as follows: in, For the first IMU attitude moment , This represents the number of IMU sampling points.

[0035] In step S400, the spatial coordinates of the UAV and the acquired image information need to be time-aligned. Then, an actual model of the bridge bearing pad stone is established in the spatial coordinate system. When establishing the actual model of the bridge bearing pad stone in the spatial coordinate system, the exposure timestamp is used as the association medium to match and align the acquired images in five dimensions (top, front, back, left, and right) with the corresponding three-dimensional coordinates and IMU attitude angle data at the corresponding time, determining the absolute position and rotation angle of the image in space. Then, computer vision algorithms are used to perform feature point detection and cross-view matching on the full set of images, identifying the same feature points corresponding to the same geometric part of the pad stone in different images, and constructing a complex network of image connection points.

[0036] Using the initial pose data of the images as constraints, aerial triangulation calculations are performed through bundle adjustment. Iterative calculations precisely correct the actual spatial position and lens distortion parameters of each image, generating a sparse point cloud reflecting the basic contour of the bearing pad. Based on this, multi-view image dense matching is used, and depth estimation is performed on each pixel of the bearing pad based on geometric consistency and pixel grayscale information, inversely projecting the 2D image information back into 3D space. Finally, absolute orientation and scale are performed using the coordinates of the feature points marked in step S100, filtering out background noise points, thereby extracting high-density, high-precision actual point cloud data of the bridge bearing pad, completing the construction of the actual model in the spatial coordinate system.

[0037] In step S400, when matching and aligning the acquired images in five dimensions (top, front, back, left, and right) with the corresponding three-dimensional coordinates and IMU attitude angle data, the images are... ,exist Find the nearest GNSS time in the time frame, specifically: in, In order to be with the first The index of the time nearest GNSS sampling point for each image. For a two-dimensional Euclidean norm, Similarly, in Find the nearest IMU time in the time frame, specifically: in, This is the index of the IMU sampling point that is temporally nearest to the k-th image. Obtain the image-position-pose triplet binding: in, Let k be the spatiotemporal pose triplet for the k-th image. :image The corresponding three-dimensional spatial coordinates of the photography center :image The corresponding IMU attitude angle.

[0038] The camera attitude rotation follows the aerial measurement standard of rolling-pitch-heading rotation in sequence. The overall external orientation rotation matrix of the camera is obtained by multiplying three single-axis rotation matrices: in, For the first The camera exterior orientation rotation matrix of the image. For roll rotation matrix, For pitch and rotation matrices, For the heading rotation matrix, Let X be the roll angle of the camera's rotation about the X-axis. Let Y be the pitch angle of the camera's rotation around the Y-axis. Let be the yaw angle of the camera's rotation about the Z-axis. The rotation matrix must satisfy the orthogonality and identity determinant constraints: in, It is a 3x3 identity matrix, representing that the spatial scale remains unchanged after rotation. This is a matrix determinant operation; a value of 1 indicates no mirror flip.

[0039] The camera itself has fixed optical parameters. It is necessary to establish a relationship between the two-dimensional image plane coordinates and the three-dimensional spatial coordinates. Homogeneous coordinate transformation is used to eliminate the translation limitations of Cartesian coordinates, achieving accurate mapping from the two-dimensional image to three-dimensional space. These fixed optical parameters are obtained from factory calibration and remain unchanged. The focal length and principal point parameters are integrated into an intrinsic parameter matrix, characterizing the camera's inherent optical properties. Specifically: in, This is the matrix of in-camera orientation elements. , The equivalent focal length in the u / v direction of the camera's image plane is calculated by converting the camera sensor pixel size to the physical focal length. These are the principal point coordinates, i.e., the projection position of the camera's optical center onto the image plane.

[0040] Converting the two-dimensional image point coordinates to three-dimensional homogeneous coordinates and adapting matrix transformation operations yields: in, Let be the homogeneous coordinates of any point in the image plane. Using Cartesian coordinates on the image plane, In the spatial coordinate system, the four-dimensional homogeneous coordinates of any point on the surface of the pad stone are: in, Let these be the homogeneous coordinates of the object point in space. This represents the three-dimensional rectangular coordinates of the object points on the surface of the bearing stone in the global spatial coordinate system of the bridge.

[0041] Based on the principle of collinearity of three points, the scalar relationship between image plane coordinates and spatial object coordinates is derived, directly reflecting the spatial position of a single pixel, specifically: For the first Image rotation matrix No. Line number The elements of the column.

[0042] The collinearity condition is the core principle of photogrammetry, which means that the image point, the center of the photograph, and the object point in space are collinear. This equation can accurately determine the absolute position and rotation angle of each image in the global spatial coordinate system of the bridge, thus completing the spatial anchoring of the image.

[0043] To adapt to batch computing on computers and more efficiently complete image spatial pose calculation, it is also necessary to integrate the scalar equations into matrix-vector form, specifically: in, A non-zero scale factor, representing the projected depth from the center of the photograph to the object point in space. The camera's exterior orientation matrix integrates rotation and translation parameters to determine the image space pose.

[0044] When constructing a complex image connectivity network, geometric features, including the edges and corners of the pad stones, are stably extracted from each image. The Harris corner response function is constructed, and local extrema are detected as feature points. The structure tensor for: The response value is the corner point value; if it is greater than the threshold, it is determined to be a feature point. For local structure tensors, describe the gradient distribution in the neighborhood. The gradient of the image in the x and y directions. It is a sliding window. Gaussian weighted kernel, This is an empirical coefficient; For images The feature point set obtained by detection is: in, For the first The set of feature points of an image For the first Image number 1 feature point.

[0045] When identifying identifiable feature points corresponding to the same geometric location of the foundation stone in different images, the feature points... Generate a 128-dimensional SIFT descriptor: Feature point descriptor vector, , The L2 norm is used to calculate the descriptor distance, and the nearest neighbor is selected. and the next nearest neighbor Make a ratio decision: If the conditions are met, it is considered a pair of points with the same name: in, For matching threshold, This refers to the same spatial point on the surface of the pad stone.

[0046] To ensure the correctness of corresponding points, epipolar geometry is used to eliminate incorrect matches that do not satisfy the projection constraints. For image pairs Fundamental matrix Satisfying polar constraints: Iterative estimation using RANSAC Remove outliers and retain the set of interior points: in, Based on the matrix, For image A reliable set of pairs of points with the same name between them.

[0047] Finally, all images and all corresponding points are constructed into a global connectivity network:

[0048] In bundle adjustment, a strict projection relationship between the image point observations and the object point is established using the initial pose as a constraint. Lens distortion correction is also introduced to obtain the adjusted projected image points. : in, It is the radial + tangential distortion vector. This is the distortion parameter vector; The objective function for bundle adjustment is: The condition for stopping the iteration is: , For all parameters to be estimated, After adjustment convergence, the three-dimensional coordinates of all connection points are obtained, forming a sparse point cloud of the pad stones:

[0049] When considering geometric consistency and pixel grayscale information, for images medium pixel ,depth satisfy: For pixels The corresponding spatial point, Matching quality is measured by neighborhood gray-level similarity, and the optimal depth is obtained by minimizing the cost. in, For matching cost, For pixels The optimal depth By substituting the depth of each pixel into the projection model, dense 3D points are generated:

[0050] When filtering out noise points, it is necessary to calculate the neighborhood statistical characteristics, remove outlier noise points, and retain the effective point cloud of the pad stones. This requires calculating the mean neighborhood distance. with standard deviation Specifically: When the condition is met At that time, a high-precision actual point cloud model of the bridge bearing pad stone in the global spatial coordinate system was obtained. .

[0051] In step S500, when comparing the reference point cloud data and the actual point cloud data of the bridge bearing pad stone, it is necessary to preprocess the actual point cloud data obtained in step S400. By using statistical filtering or radius filtering algorithms, floating noise points and redundant point clouds in non-pad stone areas generated during UAV acquisition and 3D reconstruction are removed to obtain clean actual point cloud data of the pad stone.

[0052] To eliminate systematic errors, point cloud registration is required to achieve fine fitting. This involves finding the closest point pair between the reference point cloud and the actual point cloud, calculating the rotation and translation matrix between these two sets of point clouds, and iterating this process until the mean square error between the two sets of point clouds is minimized or the convergence condition is met, so that the actual point cloud and the reference model achieve optimal fitting in spatial pose.

[0053] Deviation calculation is also required. A distance calculation algorithm is used to calculate the shortest normal distance from each 3D point in the registered actual point cloud to the reference point cloud.

[0054] When calculating deviation values, for the engineering inspection requirements of bridge bearing pad stones, the following core dimensions of deviation values ​​are extracted and calculated: Planar position deviation: Extract the coordinates of the center point of the top surface of the pad stone, and calculate the planar offset of the X and Y axes with the coordinates of the center point of the reference model.

[0055] Elevation deviation: The average value of the Z-axis coordinate set of the point cloud on the top surface of the pad stone is extracted and the difference is calculated with the design elevation of the reference model.

[0056] Flatness deviation: Calculate the variance of the vertical distance from each point on the top surface of the actual point cloud to the fitted plane.

[0057] Finally, a chromatographic mapping relationship is established to generate a visualized three-dimensional chromatographic deviation map of the bearing pad. Simultaneously, the calculated deviation values ​​are automatically compared with preset allowable error thresholds in bridge construction specifications. If all deviation values ​​are within the allowable threshold range, the system determines that the bridge bearing pad is qualified; if any value exceeds the threshold, it is determined to be unqualified, and a report detailing the specific location and deviation value of the deviation is output.

[0058] To denoise the reconstructed point cloud and remove floating points and non-bed redundant points, each point... Select its surroundings Find the nearest neighbor and calculate the distance from that point to this nearest neighbor. The average distance between the neighboring points is as follows: Its global mean with standard deviation for: Confidence coefficients also need to be set. Preserve the average neighborhood distance plus the global mean. For points within a range of 1 / 2 standard deviation, remove noisy points that exceed this range, and retain them based on the following criteria:

[0059] Radius filtering is used to remove isolated floating noise points, specifically: For the center of the ball, Let the radius be , count the number of neighboring points contained within the sphere. If the number of neighboring points satisfies: If the conditions are met, the point is considered a valid point for the foundation stone and should be retained; otherwise, it is considered noise and should be removed. After dual processing using statistical filtering and radius filtering, all noise and redundant points are removed, resulting in a pure, actual point cloud containing only the pad stone region. , denoted as: This is a filtering operation.

[0060] The actual point cloud and the reference point cloud may have rotation and translation deviations in spatial pose. Direct comparison will produce large systematic errors. Therefore, fine registration is required by ICP (Iterative Closest Point) algorithm to adjust the spatial pose of the actual point cloud so that the two can achieve optimal fit in space and minimize systematic errors.

[0061] The core of point cloud registration is to find the correspondence between actual points and reference points, that is, for each point in a clean actual point cloud. In the reference point cloud Find the nearest reference point to this point. These form the closest point pair, which is used for subsequent calculations of the rotation and translation matrices. The formula for finding the closest point is: ,in This indicates taking the reference point that minimizes the Euclidean distance.

[0062] The goal of the ICP algorithm is to find the optimal rotation matrix. Translation vector The objective function is to minimize the mean square error between all nearest point pairs, that is, to minimize the deviation between the actual point and its corresponding reference point after rotation and translation. in, To be practical, This is the corresponding nearest reference point; It optimizes the rotation matrix step by step through an iterative approach. Translation vector After each iteration, the deviation between the actual point's position after rotation and translation and the position of the previous iteration is calculated. When this deviation is less than a preset convergence threshold, the convergence is considered complete. When the iteration reaches its optimal point, the iteration stops. The convergence condition formula is: in, , For the first The rotation matrix and translation vector of the next iteration , Let be the rotation matrix and translation vector for the (k+1)th iteration.

[0063] Once the ICP iteration reaches the convergence condition, the optimal rotation matrix is ​​obtained. Translation vector Each point in the clean actual point cloud is subjected to this rotation and translation transformation to obtain a registered actual point cloud that best fits the spatial pose of the reference point cloud. : .

[0064] The planar position deviation is used to detect the horizontal offset of the pad stone. By comparing the coordinates of the center point of the top surface of the reference pad stone and the actual pad stone, the difference between the two in the X and Y axis directions is calculated. First, the reference center point of the top surface of the pad stone is extracted. and actual center point Then calculate Axis offset and Axis offset, The total deviation in the plane is , The composite value of axis offset reflects the degree of overall planar offset, specifically: This value is directly compared with the plane allowable error threshold.

[0065] The elevation deviation is used to detect the difference between the actual elevation and the design elevation of the top surface of the pad stone. First, the average Z-axis coordinate of all points on the top surface of the actual point cloud after registration is calculated, which is the average elevation of the actual top surface: ,in This represents the actual number of points in the top surface point cloud. For each top surface point Axis coordinates; The design elevation of the foundation stone in the reference model; elevation deviation. for: A positive difference indicates that the actual elevation is too high, while a negative difference indicates that the actual elevation is too low. The absolute value is compared with the allowable error threshold for elevation.

[0066] The flatness deviation is used to detect the flatness of the top surface of the pad stone. First, a plane fitting is performed on the top surface points of the registered actual point cloud to obtain the fitted plane equation of the top surface: ,in The parameters of the fitted plane are then calculated; subsequently, the perpendicular distance from each point on the top surface to the fitted plane is calculated. : The flatness deviation is represented by the variance of these vertical distances. The larger the variance, the worse the flatness of the top surface. The formula is: ,in The variance is the average of all vertical distances, and this variance is compared to the flatness tolerance threshold.

[0067] Example 2 This invention provides an intelligent detection system for the construction accuracy of bridge bearing pad stones based on UAV oblique photography, comprising: Reference establishment module: used to set up one or more control points, establish a spatial coordinate system, select three or more feature points on the bridge pier and mark their positions in the spatial coordinate system; Data acquisition module: used to establish reference models of bridge pier cap beams and bearing pads in a spatial coordinate system in conjunction with construction drawings, and to acquire reference point cloud data of bridge bearing pads from the reference model; Image acquisition module: used to establish a connection between the UAV and the spatial coordinate system, obtain the position of the UAV in the spatial coordinate system in real time, and use the UAV to synchronously acquire image information of the bridge bearing pad stone in five directions: top, front, back, left, and right. Point cloud acquisition module: It is used to reconstruct the actual model of the pad stone and cap beam using the acquired image information. Then, by using the absolute position coordinates of the control points and the relative position relationship between the actual model and the control points, it reversely calculates the final absolute position coordinates and absolute elevation of the pad stone, thereby generating actual point cloud data with high-precision absolute elevation information in the spatial coordinate system. Data Judgment Module: Used to compare the reference point cloud data and the actual point cloud data of bridge bearing pad stones, obtain the deviation value of each bridge bearing pad stone, and determine whether the bridge bearing pad stone is qualified.

[0068] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent detection of bridge bearing pad stone construction accuracy based on UAV oblique photography, characterized in that, Includes the following steps: S100. Set up one or more control points, establish a spatial coordinate system, select three or more feature points on the bridge pier and mark their positions in the spatial coordinate system. S200. Based on the construction drawings, establish a reference model of the pier cap beam and bearing pad in the spatial coordinate system, and obtain the reference point cloud data of the bridge bearing pad from the reference model. S300: Establish a connection between the UAV and the spatial coordinate system, obtain the position of the UAV in the spatial coordinate system in real time, and use the UAV to synchronously collect image information of the bridge bearing pad stone in five directions: top, front, back, left, and right. S400: Using the collected image information, the actual model of the pad stone and the cap beam is reconstructed. Then, by using the absolute position coordinates of the control points and the relative position relationship between the actual model and the control points, the absolute position coordinates and absolute elevation of the final pad stone are calculated in reverse, thereby generating actual point cloud data with high-precision absolute elevation information in the spatial coordinate system. S500: Compare the reference point cloud data and the actual point cloud data of the bridge bearing pads to obtain the deviation value of each bridge bearing pad and determine whether the bridge bearing pads are qualified.

2. The intelligent detection method for bridge bearing pad stone construction accuracy based on UAV oblique photography according to claim 1, characterized in that, In step S300, the airborne positioning module needs to be associated with the spatial coordinate system established in step S100, and the corresponding coordinate transformation parameters need to be input so that the UAV's control and data storage unit can receive differential correction signals in real time, thereby achieving high-precision three-dimensional spatial positioning; at the same time, the UAV's built-in inertial measurement unit is used to acquire the UAV's attitude angle data during flight in real time at high frequency.

3. The intelligent detection method for bridge bearing pad stone construction accuracy based on UAV oblique photography according to claim 2, characterized in that, The drone flies automatically and autonomously along a preset route. When it reaches each preset exposure point on the route, the control and data storage unit sends a synchronous trigger signal to control the five-lens camera to expose simultaneously, thereby synchronously completing the image acquisition of five different perspectives from the top, front, back, left and right of the bridge bearing pad stone. At the moment each exposure is triggered by the five-lens camera, the drone's time synchronization unit binds the current exposure time, spatial coordinates, and attitude angle data to generate POS data corresponding to each image. This POS data is then used to generate an independent data log file, thereby assigning each acquired image a high-precision absolute spatial position and initial attitude value. This provides essential exterior orientation elements for the accurate spatial establishment of the actual point cloud model.

4. The intelligent detection method for bridge bearing pad stone construction accuracy based on UAV oblique photography according to claim 3, characterized in that, In step S400, when establishing the actual model of the bridge bearing pad stone in the spatial coordinate system, the exposure timestamp is used as the association medium to match and align the collected images in the five dimensions of top, front, back, left, and right with the corresponding three-dimensional coordinates and IMU attitude angle data to determine the absolute position and rotation angle of the image in space. Then, computer vision algorithms were used to perform feature point detection and cross-view matching on the full set of images, identify the same feature points corresponding to the same geometric part of the pad stone in different images, and construct a complex network of image connection points.

5. The intelligent detection method for bridge bearing pad stone construction accuracy based on UAV oblique photography according to claim 4, characterized in that, Then, using the initial pose data of the image as a constraint, aerial triangulation calculation is performed through bundle adjustment. The actual spatial position and lens distortion parameters of each photo are accurately corrected through iterative calculation, thereby generating a sparse point cloud that reflects the basic outline of the pad stone. Based on this, by using dense matching of multi-view images, and based on geometric consistency and pixel grayscale information, the depth of each pixel of the pad stone is estimated, and the two-dimensional image information is projected back into the three-dimensional space. Finally, by combining the coordinates of the feature points marked in step S100 for absolute orientation and scale scaling, background noise points are filtered out, thereby extracting high-density, high-precision actual point cloud data of bridge bearing pad stones, and completing the actual model construction in the spatial coordinate system.

6. The intelligent detection method for bridge bearing pad stone construction accuracy based on UAV oblique photography according to claim 5, characterized in that, In step S500, when comparing the reference point cloud data and the actual point cloud data of the bridge bearing pad stone, it is necessary to preprocess the actual point cloud data obtained in step S400. By using statistical filtering or radius filtering algorithms, floating noise points and redundant point clouds in non-pad stone areas generated during UAV acquisition and 3D reconstruction are removed to obtain clean actual point cloud data of the pad stone.

7. The intelligent detection method for bridge bearing pad stone construction accuracy based on UAV oblique photography according to claim 6, characterized in that, To eliminate systematic errors, point cloud registration is required to achieve fine fitting. This involves finding the closest point pair between the reference point cloud and the actual point cloud, calculating the rotation and translation matrix between these two sets of point clouds, and iterating this process until the mean square error between the two sets of point clouds is minimized or the convergence condition is met.

8. The intelligent detection method for bridge bearing pad stone construction accuracy based on UAV oblique photography according to claim 7, characterized in that, It is also necessary to calculate the deviation value. Using a distance calculation algorithm, the shortest normal distance from each 3D point in the registered actual point cloud to the reference point cloud is calculated one by one. When calculating deviation values, for the engineering inspection requirements of bridge bearing pad stones, the following core dimensions of deviation values ​​are extracted and calculated: Planar position deviation: Extract the coordinates of the center point of the top surface of the pad stone, and calculate the planar offset of the X and Y axes with the coordinates of the center point of the reference model; Elevation deviation: Extract the average value of the Z-axis coordinate set of the point cloud on the top surface of the pad stone, and calculate the difference with the design elevation of the reference model; Flatness deviation: Calculate the variance of the vertical distance from each point on the top surface of the actual point cloud to the fitted plane.

9. The intelligent detection method for bridge bearing pad stone construction accuracy based on UAV oblique photography according to claim 8, characterized in that, Finally, a chromatographic mapping relationship is established to generate a visualized three-dimensional chromatographic deviation map of the foundation stone; At the same time, the calculated deviation values ​​are automatically compared with the preset allowable error thresholds of the bridge construction specifications. If all deviation values ​​are within the allowable threshold range, the system determines that the bridge bearing pad is qualified. If any value exceeds the threshold, it is determined to be unqualified, and a report of the specific location and deviation value of the deviation is output.

10. An intelligent detection system for bridge bearing pad stone construction accuracy based on UAV oblique photography, used to implement the intelligent detection method for bridge bearing pad stone construction accuracy based on UAV oblique photography as described in any one of claims 1-9, characterized in that, include: Reference establishment module: used to set up one or more control points, establish a spatial coordinate system, select three or more feature points on the bridge pier and mark their positions in the spatial coordinate system; Data acquisition module: used to establish reference models of bridge pier cap beams and bearing pads in a spatial coordinate system in conjunction with construction drawings, and to acquire reference point cloud data of bridge bearing pads from the reference model; Image acquisition module: used to establish a connection between the UAV and the spatial coordinate system, obtain the position of the UAV in the spatial coordinate system in real time, and use the UAV to synchronously acquire image information of the bridge bearing pad stone in five directions: top, front, back, left, and right. Point cloud acquisition module: It is used to reconstruct the actual model of the pad stone and cap beam using the acquired image information. Then, by using the absolute position coordinates of the control points and the relative position relationship between the actual model and the control points, it reversely calculates the final absolute position coordinates and absolute elevation of the pad stone, thereby generating actual point cloud data with high-precision absolute elevation information in the spatial coordinate system. Data Judgment Module: Used to compare the reference point cloud data and the actual point cloud data of bridge bearing pad stones, obtain the deviation value of each bridge bearing pad stone, and determine whether the bridge bearing pad stone is qualified.

Images