Under-bridge space unmanned aerial vehicle combined positioning and approaching photography equipment and under-bridge space unmanned aerial vehicle combined positioning and approaching photography method

Abstract

The invention relates to the technical field of engineering detection and surveying and mapping, in particular to an under-bridge space unmanned aerial vehicle combined positioning and approaching photography device and method, and the device comprises a ground reference subsystem, an unmanned aerial vehicle flight platform, a combined positioning subsystem, an approaching photography subsystem and an airborne processing and control unit. The method comprises the steps of arranging a positioning base station and configuring equipment, generating an initial three-dimensional live-action model through coarse scanning, planning an unstructured three-dimensional close route, fusing multi-source sensing data in flight to solve a high-precision pose, adaptively adjusting an optical axis of a camera to collect images, finally generating a component high-precision model, and automatically identifying and measuring diseases. The method solves the problem of under-bridge positioning, improves the image quality and detection precision, improves the operation efficiency, guarantees the safety, and provides reliable support for bridge safety assessment and maintenance decision.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of engineering detection and mapping technology, in particular to a kind of bridge space unmanned aerial vehicle combination positioning and close photography equipment and method. BACKGROUND

[0002] In current bridge detection, due to the complex structure of the space under the bridge and the limited field of view, traditional manual detection requires personnel to enter narrow and dangerous areas for operation, which not only has low safety, but also has low detection efficiency and missed details. Ordinary unmanned aerial vehicle detection relies on single GNSS positioning, and the satellite signal under the bridge is easily blocked, resulting in insufficient positioning accuracy. During flight, it is easy to collide with bridge components. At the same time, ordinary unmanned aerial vehicles mostly use structured flight paths, which cannot adapt to the complex surface morphology of key bridge components, and the image angle is difficult to be perpendicular to the component surface, so the image details are blurred, which cannot meet the needs of detailed detection. In addition, the existing detection technology has low fusion degree of multi-source sensing data, which is difficult to form accurate pose data support, resulting in poor accuracy of subsequent three-dimensional reconstruction, and automatic quantization of disease identification cannot be realized, which is difficult to efficiently support bridge safety evaluation and maintenance decision-making. Therefore, in view of the above problems, a kind of bridge space unmanned aerial vehicle combination positioning and close photography equipment and method is proposed. SUMMARY

[0003] The purpose of the present application is to provide a kind of bridge space unmanned aerial vehicle combination positioning and close photography equipment and method to solve the problems raised in the background art.

[0004] To achieve the above purpose, the present application provides the following technical scheme: In a first aspect, a kind of bridge space unmanned aerial vehicle combination positioning and close photography method, comprising the following steps: S1: data acquisition preparation: positioning reference station is laid out in bridge area, and combination positioning system and shooting system are configured for unmanned aerial vehicle; S2: rough scanning and initial modeling: control unmanned aerial vehicle to execute preliminary scanning flight in the space under the bridge, and synchronously collect wide-angle image and positioning data;Based on wide-angle image and positioning data, generate the initial three-dimensional real scene model of the space under the bridge; S3: close photography path planning: based on the initial three-dimensional real scene model, identify the key components of the bridge, and generate a non-structured three-dimensional close flight path for each key component that is adapted to the surface morphology; S4: tight combination positioning and adaptive photography: control unmanned aerial vehicle to fly along the three-dimensional close flight path;During flight, the following operations are performed: S4-1: synchronously acquire satellite differential signal from positioning reference station, inertial data of airborne inertial measurement unit, image sequence of vision sensor, and distance information from bridge surface of ranging sensor; S4-2: The initial 3D real-world model is used as the spatial constraint prior information and combined with the various sensor data obtained in step S4-1 for tight combination filtering and fusion to calculate the high-precision position and attitude of the UAV in the space under the bridge in real time. S5: Control the drone's shooting system, and adaptively adjust the camera's optical axis direction according to the high-precision attitude calculated in step S4-2, so that it is perpendicular to the local normal of the bridge component surface during shooting, and acquire close-up images.

[0005] As a preferred approach, step S3 generates unstructured 3D proximity flight paths for key components, specifically including: S3-1: Isolate key bridge components to be scanned in detail from the initial 3D reality model; S3-2: Calculate the normal vectors of the surface of the component triangular mesh model; S3-3: Using the normal vector as a guide, extrapolate to generate a parallel three-dimensional virtual photographic surface on the component surface; S3-4: On the 3D virtual photographic surface, generate dense waypoints according to the preset heading and lateral overlap rates, and set the camera orientation of each waypoint to align with the direction of the normal vector of its corresponding component surface.

[0006] As a preferred approach, the compact combination filter fusion in step S4-2 is specifically as follows: Construct a Kalman filter whose state vector includes the UAV's position, velocity, attitude, and inertial sensor errors; The absolute position provided by the satellite differential signal, the relative pose change provided by the visual sensor, and the distance to the bridge surface provided by the ranging sensor are used as the input Kalman filter for multiple observations. The distance information provided by the ranging sensor is converted into a normal distance observation of the UAV relative to the model surface through real-time registration with the initial 3D reality model, which is used to correct the state estimation of the filter.

[0007] As a preferred approach, in the compact combination filtering fusion, the feature points extracted by the visual sensor are matched with the corresponding 3D points in the initial 3D real scene model, and the coordinates of the successfully matched 3D points are used as spatial position constraint observations and input into the Kalman filter together.

[0008] As a preferred option, it also includes: S6: Results Generation and Analysis: Based on the close-up images acquired in step S5 and their corresponding high-precision positions and attitudes, a refined three-dimensional reconstruction is performed to generate a high-precision real-scene model of the bridge components; and based on this model, automatic identification and measurement of defects are carried out.

[0009] Secondly, the equipment for drone-based positioning and close-up photography in the space under the bridge includes: The ground reference subsystem includes positioning reference stations deployed in the bridge area; Unmanned aerial vehicle (UAV) flight platform; The integrated positioning subsystem, integrated into the UAV flight platform, includes: A GNSS receiver is used to receive satellite signals and differential signals from a reference station. Inertial Measurement Unit (IMU); Visual sensors are used to acquire continuous sequences of images; Distance sensors are used to measure distances to the surface of bridges. The close-up photography subsystem is integrated into the drone's flight platform and includes a high-resolution camera and a gimbal for adjusting its attitude. The airborne processing and control unit is configured as follows: Perform compact combination filtering fusion to solve for the high-precision pose of the UAV; Plan unstructured 3D close-up flight routes based on the initial 3D reality model; Based on the calculated pose and planned flight path, the drone flight platform and gimbal are controlled to complete autonomous flight and adaptive shooting.

[0010] As a preferred embodiment, the visual sensor includes a navigation camera for pose calculation and a reconnaissance camera for capturing close-up images, with the navigation camera and the reconnaissance camera being physically isolated and having independent parameters.

[0011] As a preferred option, the ranging sensor is a single-point laser rangefinder or a two-dimensional lidar. Its installation position and beam direction are calibrated so that its measurement value has a known spatial transformation relationship with the optical axis direction of the survey camera.

[0012] As can be seen from the technical solution provided by the present invention above, the beneficial effects of the bridge-under-bridge space UAV combined positioning and close-up photography equipment and method provided by the present invention are: To improve positioning accuracy under the bridge, the problem of satellite signal blockage under the bridge is solved by tightly combining and filtering multi-source sensor data, thereby achieving high-precision pose calculation for UAVs and ensuring flight stability. By optimizing the image quality and adaptively adjusting the camera's optical axis to be perpendicular to the surface of the bridge components, the acquired images are clear in detail and accurate in angle, laying a data foundation for subsequent fine processing. To improve the efficiency and accuracy of bridge inspection, high-quality data is used to complete the fine three-dimensional reconstruction of components, realize automatic identification and quantitative measurement of defects, reduce manual intervention and lower inspection costs; To ensure the safety of inspection operations, there is no need for personnel to enter the narrow and dangerous space under the bridge. Drones can operate autonomously to avoid safety risks such as personnel falling or colliding. A complete closed-loop testing system is formed, with full automation from data collection and processing to output. The output of quantitative testing results can directly support bridge safety assessment and maintenance decisions, thereby improving management efficiency. Attached Figure Description

[0013] Fig. 1 This is a schematic diagram illustrating the steps of a method for combined positioning and close-up photography of unmanned aerial vehicles (UAVs) in space under a bridge according to the present invention. Fig. 2 This is a schematic diagram of the structure of a bridge-under-the-space UAV combined positioning and close-up photography equipment according to the present invention. Detailed Implementation

[0014] 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.

[0015] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific embodiments.

[0016] like Figs. 1-2 As shown, this embodiment of the invention provides a method for combined positioning and close-up photography of unmanned aerial vehicles (UAVs) in the space under a bridge, including the following steps: S1: Data acquisition preparation: Deploy positioning reference stations in the bridge area and equip the UAV with a combined positioning system and a shooting system; S2: Coarse Scan and Initial Modeling: Control the UAV to perform a preliminary scan flight in the space under the bridge, simultaneously acquiring wide-angle images and positioning data; based on the wide-angle images and positioning data, generate an initial 3D reality model of the space under the bridge; S3: Close-up photography path planning: Based on the initial 3D real scene model, identify key bridge components and generate unstructured 3D close-up flight paths for each key component that are adapted to its surface morphology; S4: Compactly Combined Positioning and Adaptive Photography: Control the UAV to fly along a 3D proximity flight path; during flight, perform the following operations: S4-1: Simultaneously acquire satellite differential signals from the positioning reference station, inertial data from the airborne inertial measurement unit, image sequences from the visual sensor, and distance information from the ranging sensor to the bridge surface; S4-2: The initial 3D real-world model is used as the spatial constraint prior information and combined with the various sensor data obtained in step S4-1 for tight combination filtering and fusion to calculate the high-precision position and attitude of the UAV in the space under the bridge in real time. S5: Control the drone's shooting system, and adaptively adjust the camera's optical axis direction according to the high-precision attitude calculated in step S4-2, so that it is perpendicular to the local normal of the bridge component surface during shooting, and acquire close-up images.

[0017] In this embodiment, step S1 aims to establish a stable and reliable data acquisition foundation by scientifically deploying positioning reference stations and precisely configuring the combined positioning and imaging systems of UAVs. This ensures that the sensor data and image data in subsequent stages such as coarse scanning, initial modeling, and close-up photography possess high accuracy and high synchronization, providing solid data support for the entire under-bridge space detection process. The detailed steps are as follows: Step S1-1: Scientific deployment of positioning reference stations: Base station site selection planning: Within the bridge area, select an area with a wide field of view and no obstructions as the base station deployment point. The site should be far away from tall buildings, trees, high-voltage lines and other obstructions to ensure smooth satellite signal reception. At the same time, the site should be in a stable location and not easily affected by factors such as vibration and settlement. The straight-line distance from the detection area under the bridge should be controlled within a reasonable range to ensure the quality of differential signal transmission. Base station equipment installation: Fix the receiving antenna of the positioning base station on a dedicated bracket. The antenna installation height must be higher than surrounding obstructions to ensure sufficient signal reception angle. The bracket installation must be kept vertical and stable. Use a level to calibrate the installation plane and control the error within the specified range. Then connect the receiver and antenna with a dedicated cable. The connection interface must be firmly tightened to avoid signal transmission interruption. Base station parameter configuration: Start the positioning base station receiver host, enter the parameter configuration interface, set the satellite system type to support multi-system joint positioning and improve signal reception stability; configure the sampling frequency and set a reasonable sampling interval according to the detection requirements to ensure the timeliness of positioning data; set the differential signal output format and transmission baud rate to keep them consistent with the GNSS receiver parameters of the UAV to ensure effective resolution of differential signals; Base station data transmission verification: After completing the parameter configuration, start the data transmission function of the base station and send differential signals to the surrounding area through the wireless communication module; operate the drone to fly to the edge of the space detection area under the bridge, turn on the drone's GNSS receiver, receive the differential signal sent by the base station, and check the signal strength, transmission delay and other indicators; the signal strength must reach the specified threshold, and the transmission delay must be controlled within the minimum range to ensure that the differential signal can be stably and reliably transmitted to the drone; Step S1-2: Configuration of the UAV integrated positioning subsystem: Positioning component selection and adaptation: Based on the accuracy requirements of under-bridge space detection, a GNSS receiver with multi-system compatibility is selected to ensure it can simultaneously receive satellite signals and differential signals from the base station; an inertial measurement unit (IMU) is selected, which must have a high sampling rate and high stability to accurately collect inertial data such as the UAV's acceleration and angular velocity; for the visual sensors, a navigation camera and a reconnaissance camera are selected, which are physically isolated and have independent parameters. The navigation camera is used for pose calculation, and the reconnaissance camera is used to capture close-up images; for the ranging sensor, a single-point laser rangefinder or a two-dimensional lidar is selected to ensure that its measurement accuracy meets the distance detection requirements; Component Installation and Calibration: The GNSS receiver antenna is installed on the top of the UAV flight platform, ensuring that the antenna is unobstructed and the received signal is not interfered with by the fuselage structure; the inertial measurement unit (IMU) is fixed at the center of the UAV flight platform to reduce the impact of vibration during flight on the measurement data; the navigation camera and reconnaissance camera are installed below the UAV at preset angles; the installation position and beam direction of the ranging sensor are precisely calibrated so that its measurement value forms a known spatial transformation relationship with the optical axis direction of the reconnaissance camera. A standard calibration board is used during the calibration process, and installation errors are eliminated through multiple measurements and calculations. Component Functional Testing: Start the GNSS receiver, IMU, visual sensor, and ranging sensor respectively, and check whether each component is working properly. The GNSS receiver should be able to stably receive satellite signals and base station differential signals, and output continuous positioning data. The IMU should be able to output accurate inertial data in real time, without drift or abnormal fluctuations. The visual sensor should be able to acquire clear and continuous image sequences. The ranging sensor should be able to accurately measure the distance to the target object, and the data output should be stable and noise-free. Steps S1-3: Configuration of the UAV close-up photography subsystem: Photography equipment selection and installation: A high-resolution camera was selected as the core equipment for close-up photography, ensuring that the camera's pixel count, focal length, and other parameters meet the requirements for shooting details of bridge components; the camera was mounted on an adjustable gimbal, which was installed below the drone flight platform. During installation, it was ensured that the gimbal and the drone body were firmly connected without any looseness to avoid shaking during flight that would affect the shooting effect; Gimbal and camera parameter adjustment: Start the gimbal control system and adjust the gimbal's pitch, roll, and yaw rotation functions to ensure that the gimbal can flexibly adjust the camera's attitude, rotate the angle accurately, and without any stuttering or delay; enter the camera parameter setting interface and configure parameters such as shooting resolution, shutter speed, and ISO, and preset reasonable shooting parameters according to the lighting conditions of the space under the bridge; at the same time, set the camera's shooting mode to support automatic continuous shooting, ensuring that images can be captured at preset intervals during close-up photography; Collaborative testing of the photography system: Establish a connection between the gimbal and the control system of the drone flight platform to test the coordination between the gimbal attitude adjustment and the drone flight status; control the drone to make simple flight attitude adjustments and observe whether the gimbal can respond in time and keep the camera stable; at the same time, start the camera shooting function to collect test images and check the image sharpness, exposure, focus accuracy and other indicators to ensure that the photography system can stably output high-quality images during the drone flight. Step S1-4: Onboard processing and control unit debugging: Hardware connection and system startup: Connect the airborne processing and control unit with the UAV's combined positioning subsystem and close-up photography subsystem through a dedicated interface to ensure smooth data transmission between the components; after connection, start the airborne processing and control unit and check whether the system startup is normal, whether the communication status of each interface is stable, and whether there are any communication failure prompts. Data synchronization and parsing test: Enable the data synchronization function of the airborne processing and control unit, and test the data synchronization transmission effect of equipment such as GNSS receiver, inertial measurement unit (IMU), visual sensor, ranging sensor and camera to ensure that the timestamps of various data are accurately aligned and there is no time deviation; at the same time, test the processing unit's ability to parse various types of data, and be able to accurately extract key contents such as positioning data, inertial data, image information, and distance information. Control command issuance verification: Control commands are issued to the UAV flight platform, gimbal, camera and other devices through the airborne processing and control unit to test the timeliness and accuracy of command transmission; the UAV is controlled to perform basic actions such as takeoff, hovering and turning to observe whether the UAV can respond accurately; the gimbal is controlled to adjust the camera attitude to check whether the attitude adjustment meets the command requirements; the camera is controlled to perform operations such as shooting and stopping shooting to ensure that the camera can respond to commands in a timely manner and complete the corresponding actions.

[0018] In this embodiment, step S2 involves using a drone to perform a preliminary scan of the space under the bridge, simultaneously collecting and fusing multi-source basic data to generate a comprehensive initial 3D reality model with basic accuracy. This provides spatial constraint prior information and basic data support for subsequent key component identification, unstructured 3D close-range flight path planning, and compact combination positioning. The detailed steps are as follows: Step S2-1: Preliminary flight path planning: Flight range definition: Based on the bridge design drawings and on-site survey results, the boundary range of the space under the bridge is defined, including the extension distance at both ends of the bridge, the upper and lower limits of the vertical height under the bridge, and the lateral coverage width, to ensure that the planned path can completely cover the entire area under the bridge without any scanning blind spots; Waypoint distribution design: A uniform grid-like waypoint layout is adopted, with dense and uniform waypoints set within the defined flight range. The waypoint spacing is determined based on the size of the space under the bridge and the field of view of the camera's wide-angle lens, ensuring that the wide-angle images taken by adjacent waypoints can form effective directional and lateral overlap. Flight parameter settings: Set the drone's flight altitude to ensure clear imaging of the overall bridge structure while avoiding collisions with bridge components, and adjust it reasonably based on the clearance height under the bridge; set the flight speed to ensure that the drone has enough time to collect stable images and positioning data at each waypoint, while also taking into account scanning efficiency; enable the drone's flight attitude stabilization mode to ensure the drone remains stable during flight and reduce the impact of attitude fluctuations on data acquisition accuracy. Step S2-2: Synchronous acquisition of multi-source data: Wide-angle image acquisition: Activate the wide-angle camera on the drone and automatically acquire wide-angle images of the space under the bridge at preset shooting intervals. The camera focal length is fixed at the wide-angle end to ensure that each image covers a large area. During the acquisition process, keep the camera optical axis vertically downward to avoid excessive image tilt angle. At the same time, record the shooting time of each image to provide a time reference for subsequent data synchronization. Positioning data acquisition: The GNSS receiver on the UAV receives satellite signals and differential signals sent by the ground reference station in real time, and continuously outputs the absolute position data of the UAV; the inertial measurement unit (IMU) synchronously acquires the UAV's acceleration, angular velocity and other inertial data to supplement and correct the positioning data; the data acquisition frequency of the two devices is kept consistent to ensure that each frame of image corresponds to a set of synchronized positioning and inertial data; Data synchronization and storage: The airborne processing and control unit receives image data from the wide-angle camera, positioning data from the GNSS receiver, and inertial data from the IMU in real time. It achieves accurate synchronization of the three types of data through timestamp alignment technology, and stores the synchronized data in the airborne storage device in a fixed format. A data verification mechanism is used during the storage process to avoid data loss or damage. Step S2-3: Data Preprocessing Wide-angle image preprocessing: Distortion correction is performed on the acquired wide-angle images. Preset camera intrinsic parameters are used to correct geometric distortions in the images and restore the true geometric shape of the images. Image enhancement processing is performed to adjust the brightness, contrast and saturation of the images, improve the clarity of image details, and eliminate the impact of uneven lighting on subsequent processing. Blurry, overexposed or underexposed invalid images are removed, and valid images that meet the quality requirements are retained for modeling. Positioning data preprocessing: Noise filtering is performed on the positioning data output by the GNSS receiver. A moving average filtering algorithm is used to remove high-frequency noise and smooth the positioning data curve. The inertial data collected by the IMU is integrated to obtain the attitude change data of the UAV. The data is then fused with the GNSS positioning data using a loose combination filtering method to improve the continuity and stability of the positioning data. Abnormal data points with positioning accuracy below a preset threshold are removed to ensure the reliability of the positioning data. Data time synchronization calibration: Based on the shooting time of the wide-angle image, a second time synchronization calibration is performed on the preprocessed positioning data and inertial data. The time deviation between each data sequence and the image time sequence is calculated. The timestamps of the positioning data and inertial data are corrected by linear interpolation to ensure that each valid image can accurately correspond to a set of positioning and attitude data. Step S2-4: Initial 3D Reality Model Construction: Image feature extraction and matching: The scale-invariant feature transform algorithm is used to extract feature points from all effective wide-angle images to obtain the key feature points and their descriptors for each image; based on the feature point descriptors, cross-image feature point matching is performed to establish the correspondence between different images; the random sampling consensus algorithm is used to remove mismatched feature point pairs and retain the correct feature matching results. Camera extrinsic parameter calculation: Using feature matching results and preprocessed positioning and attitude data, the structure of motion reconstruction algorithm is used to calculate the camera extrinsic parameters corresponding to each image, including the camera's position and attitude parameters; the calculated camera extrinsic parameters are optimized by the bundle adjustment algorithm to minimize image reprojection error and improve the accuracy of camera extrinsic parameter calculation. Sparse point cloud generation: Based on the optimized camera extrinsic parameters and feature matching results, the three-dimensional spatial coordinates corresponding to the feature points are calculated through the triangulation algorithm to generate sparse point cloud data of the space under the bridge; the sparse point cloud data is denoised to remove isolated points and outliers, and retain the effective point cloud that can reflect the overall structure of the bridge. Dense point cloud and mesh model generation: Based on sparse point clouds, a dense reconstruction algorithm is used to generate dense point clouds in the space under the bridge, improving the density and detail richness of the point cloud data; the dense point cloud data is meshed, and a Poisson surface reconstruction algorithm is used to convert the dense point cloud into a triangular mesh model to construct the three-dimensional geometric structure of the space under the bridge. Texture mapping: The pre-processed wide-angle image is used as the texture data source. The texture mapping algorithm is used to attach the image texture information to the surface of the triangular mesh model, so that the generated initial 3D real scene model has a realistic visual effect, while retaining the basic texture features of the bridge surface, which is convenient for subsequent identification of key components. Step S2-5: Initial model accuracy verification and optimization: Coverage verification: Perform global browsing and partition checks on the generated initial 3D reality model to confirm whether the model completely covers the preset space under the bridge. Check whether key areas of the bridge, such as beams, piers, and supports, are included in the model. If there are gaps in coverage, mark the gap locations and record them. Key component identification and verification: A component identification algorithm based on geometric features is used to initially identify key bridge components in the initial 3D real-world model. The outline of the components is checked to see if it is clear and identifiable and whether it can meet the requirements for component separation in subsequent route planning. If the component identification is blurry, the reasons are analyzed. Model accuracy evaluation: Select feature points with known actual dimensions on the bridge, such as pier spacing and beam thickness, and measure their corresponding dimensions in the initial 3D reality model. Calculate the relative error between the measured values ​​and the actual values. The formula for calculating the relative error is ε = (measured value minus the absolute value of the actual value) divided by the actual value and then multiplied by 100%, where ε is the relative error, the measured value is the measured size of the feature point in the initial model, and the actual value is the true size of the feature point. Simultaneously, the point cloud density and mesh resolution of the model are statistically analyzed to ensure that the point cloud density is not lower than a preset threshold and that the mesh resolution can reflect the overall structural details of the bridge. Model optimization and adjustment: If there are coverage gaps, adjust the initial flight path, supplement the scanning and data acquisition of the missing areas, and rebuild the model; if the component recognition is blurry or the accuracy is not up to standard, optimize the data preprocessing parameters and modeling algorithm parameters, such as adjusting the feature extraction threshold and optimizing the number of bundle adjustment iterations, and regenerate the initial 3D reality model until the model meets the requirements for subsequent steps.

[0019] In this embodiment, step S3 is to accurately identify key bridge components based on the initial three-dimensional real-scene model. Through geometric analysis and spatial planning of the component surface morphology, an unstructured three-dimensional close-up flight path that is highly adapted to the component surface is generated. This provides accurate flight trajectory guidance for subsequent UAV close-up photography, ensuring that the UAV can fly smoothly along a reasonable path on the component surface and collect close-up images with complete coverage and clear details, thus meeting the requirements for refined inspection of key bridge components. Step S3-1: Isolate the key bridge components to be scanned in detail from the initial 3D reality model: Initial model preprocessing: Data purification is performed on the initial 3D real scene model to remove background noise, ground attachments, temporary construction facilities and other non-bridge elements to avoid these elements interfering with the identification of key components; at the same time, the model is simplified and lightweighted to reduce the number of triangular facets while retaining the core geometric shape of the key bridge components, thereby reducing the complexity of subsequent calculations and improving the efficiency of component identification and separation. Key component feature extraction: A multi-dimensional feature fusion and recognition algorithm is used to extract the geometric and semantic features of the structure in the initial 3D real-world model. Geometric features include dimensional parameters such as length, width, and height, morphological features such as straightness, curvature, and cross-sectional shape, and spatial positional relationships such as connection methods and distribution patterns relative to the main bridge body. Semantic features combine knowledge from the field of bridge engineering to associate typical structural attributes of key components such as beams, piers, supports, connection nodes, and crash barriers to establish a key component feature database. Key component classification and identification: The extracted structural features are matched and compared with the features in the key component feature database. Through threshold judgment and logical classification, key components of the bridge are distinguished from non-key components (such as decorative structures attached to the bridge, drainage pipes, etc.). The identified key components are individually marked, assigned a unique identity, and the spatial coordinate range and geometric parameters of each key component are recorded to provide a basis for subsequent separation operations. Key component model separation: Based on the identification and spatial coordinate range of key components, the independent model data of each key component is segmented from the initial 3D reality model to form a 3D model subset of a single component; through a spatial isolation algorithm, residual non-component elements in each component model subset are removed to ensure the purity of the component model; at the same time, a spatial relationship table between components is established to record the connection position and relative attitude of adjacent components, providing a reference for the subsequent connection of different component routes; Separation result verification: The integrity and accuracy of the separated key component models are verified. Visual inspection is used to confirm whether there are problems such as edge breakage or local missing parts in the component models. Dimension comparison is used to confirm whether the deviation of the geometric parameters of the component models from the parameters of the corresponding structures in the initial 3D real scene model is within the allowable range. If there are problems, the previous steps are returned to optimize the feature extraction algorithm or separation parameters again until all key component models meet the basic requirements for fine scanning. Step S3-2: Calculate the normal vectors of the surface of the component triangular mesh model: Component triangulation model acquisition: Extract the triangulation model data of each component from the separated independent model subset of key components. This data consists of a large number of triangular faces. Each triangular face contains the three-dimensional spatial coordinates of three vertices and the topological connection relationship between the faces. This data is the basis for calculating the normal vector. Calculation of normal vector for a single triangular facet: For each triangular facet in the component triangular mesh model, first select its three vertices, and then calculate the two edge vectors of the facet, namely the vector from the first vertex to the second vertex and the vector from the first vertex to the third vertex; obtain the initial normal vector of the triangular facet through the cross product of the two edge vectors, and then normalize the initial normal vector to obtain a unit normal vector with a length of 1, to ensure the consistency of subsequent extrapolation calculations; Unified normal vector direction: Based on the spatial structural characteristics and actual stress logic of key components, determine whether the orientation of the unit normal vector of each triangular facet points to the outside of the component (i.e., away from the center of the component entity). Taking a beam component as an example, the normal vector of its top surface should be vertically upward, the normal vector of its bottom surface should be vertically downward, and the normal vector of its side surface should be vertically pointing to the outside of the beam. If a normal vector is detected to be pointing to the inside of the component, then the unit normal vector is reversed to ensure that the normal vectors of all triangular facets are pointing to the outside of the component, providing a unified directional reference for the subsequent generation of 3D virtual photographic surfaces. Normal vector smoothing optimization: Calculate the angle between the normal vectors of adjacent triangular facets. If the angle is greater than a preset reasonable threshold, it is determined to be a region of abrupt change in normal vectors. For such regions, a weighted average smoothing algorithm is used. Taking the triangular facet of the abrupt change region as the center, the normal vectors of 3 to 5 adjacent facets are selected. Different weights are assigned according to the area ratio of each facet, and the smoothed normal vector is obtained through weighted calculation. Through smoothing, the drastic fluctuations of the normal vectors on the component surface are reduced, making the normal vector distribution more consistent with the actual surface shape of the component, and avoiding unreasonable turns in the subsequently generated flight path. Step S3-3: Using the normal vector as a guide, extrapolate and generate a parallel 3D virtual photographic surface on the component surface: Extrapolation distance determination: The determination of the extrapolation distance needs to comprehensively consider the parameters of the high-resolution camera carried by the UAV, the required image accuracy, and flight safety requirements. Camera parameters include focal length and sensor pixel size. Image accuracy requirements are measured by ground sampling distance (i.e., the actual ground size corresponding to a single pixel in the image). Flight safety requirements require a safety redundancy distance to avoid collisions between the UAV and the surface of the component. Based on these factors, the extrapolation distance is determined through comprehensive evaluation to ensure that the captured images can meet the requirements for detail resolution while ensuring the flight safety of the UAV. Virtual surface mesh generation: Based on the component triangular mesh model, for each triangular facet, the three vertices are moved by a predetermined extrapolation distance along their respective unit normal vector directions to obtain three new vertex coordinates; the new vertices generated after extrapolation of adjacent triangular facets are connected sequentially according to the topological connection relationship of the original triangular mesh model to form a three-dimensional virtual photographic surface triangular mesh similar in structure to the component surface triangular mesh model; this generation method ensures that the virtual photographic surface is completely parallel to the component surface in terms of geometry, and is only translated by a set extrapolation distance along the normal vector direction; Virtual surface integrity check: The generated 3D virtual photographic surface is traversed globally to check for issues such as mesh breaks, missing vertices, or surface holes. If missing areas are found, the coordinates of the missing vertices and triangles are supplemented by interpolation calculation based on the extrapolation rules of the surrounding triangles, ensuring that the virtual photographic surface can completely cover the entire surface of the key components without any blind spots. At the same time, it is checked whether the edges of the surface form a reasonable connection with the virtual surfaces of adjacent components. If gaps exist, the extrapolation distance of the edge area is adjusted to keep the virtual surfaces of different components continuous, facilitating the connection of subsequent flight paths. Virtual surface smoothing: The Laplacian smoothing algorithm is used to optimize the triangular mesh of the virtual photographic surface. By calculating the average coordinates of each vertex and its neighboring vertices, the coordinates of that vertex are updated to make the surface smoother. During the smoothing process, the parallel relationship between the virtual surface and the component surface must be strictly maintained. Only small bumps or depressions on the mesh surface are eliminated to make the virtual surface more in line with the trajectory requirements of stable UAV flight and reduce the drastic adjustment of UAV attitude after subsequent waypoint generation. Step S3-4: On the 3D virtual photographic surface, generate dense waypoints according to the preset heading and lateral overlap rates, and set the camera orientation of each waypoint to align with the direction of the normal vector of its corresponding component surface: Overlap rate parameter presets: Based on the complexity of the surface details of key bridge components, preset the forward overlap rate and the lateral overlap rate; for component areas with rich surface textures and prone to defects such as cracks and peeling (such as the bottom surface of the beam and the side of the pier), preset a higher overlap rate to ensure that the overlapping area of ​​adjacent images is large enough to meet the matching requirements of subsequent 3D reconstruction; for component areas with simple surface structures and low risk of defects (such as the top surface of the support), appropriately reduce the overlap rate to improve the efficiency of flight path planning and shooting while ensuring detection accuracy; Waypoint spacing determination: The waypoint spacing is determined based on the field of view of the camera at the height of the virtual photographic surface, combined with the preset overlap rate. First, the field of view of the camera in the heading and lateral directions is calculated based on the camera's focal length and extrapolation distance. Then, the heading field of view width and lateral field of view height of the camera at the virtual photographic surface are calculated based on the field of view angle and extrapolation distance. Combined with the preset heading overlap rate and lateral overlap rate, the heading waypoint spacing and lateral waypoint spacing are calculated respectively to ensure that the images captured by adjacent waypoints can meet the preset overlap rate requirements. Dense waypoint generation: On the 3D virtual photogrammetric surface, dense waypoints are generated according to the principle of "heading priority, lateral connection". First, the first row of waypoints is laid out along the length direction of the component (i.e., the heading), and the distance between adjacent waypoints is the predetermined heading waypoint spacing to ensure that the waypoints uniformly cover the heading range of the component. Then, the next row of waypoints is laid out along the width direction of the component (i.e., the lateral direction), and the distance between the next row of waypoints and the previous row of waypoints is the predetermined lateral waypoint spacing, while ensuring that the overlap rate of the images captured by the adjacent rows of waypoints meets the preset requirements. For components with complex curved surfaces (such as arc supports), curve sampling is used to generate waypoints so that the waypoint trajectory fits the curvature change of the surface and avoids the waypoints deviating from the surface, which would cause abnormal shooting distances. After generation, the waypoints are numbered, a waypoint sequence is established, and the 3D coordinates of each waypoint are recorded. Waypoint camera orientation settings: Obtain the unit normal vector of the component surface corresponding to each waypoint on the 3D virtual photographic surface, and determine the direction of the normal vector as the pointing direction of the camera optical axis; calculate the pitch angle, roll angle and yaw angle of the camera at the waypoint based on the spatial attitude of the normal vector: the pitch angle is determined by the angle between the normal vector and the horizontal plane, the roll angle is adjusted according to the projection direction of the normal vector on the horizontal plane, and the yaw angle ensures that the camera optical axis is always accurately aligned with the component surface; by adjusting the gimbal control parameters, the calculated attitude parameters are assigned to the corresponding waypoint, so that when the camera reaches the waypoint, the optical axis can be perpendicular to the local normal of the component surface, and a close-up image at the orthogonal angle can be acquired, avoiding image tilting and distortion; Waypoint sequence verification and optimization: Simulate the UAV flying along the generated waypoint sequence and check whether the distance between waypoints meets the minimum turning radius requirement of the UAV. If the waypoint spacing is too small and the UAV cannot turn smoothly, the waypoint position needs to be adjusted. Check whether the camera attitude adjustment angle is within the maximum rotation range of the gimbal. If it exceeds the range, the normal vector direction needs to be recalculated or the extrapolation distance needs to be adjusted. Correct the waypoints with problems until the waypoint sequence fully meets the actual flight and shooting requirements of the UAV, forming the final unstructured 3D close-up flight path.

[0020] In this embodiment, step S4 functions by controlling the UAV to fly along a preset three-dimensional proximity flight path, simultaneously acquiring multi-source sensor data and performing tight-combination filtering and fusion, to calculate the UAV's high-precision position and attitude in the complex space under the bridge in real time. This provides a precise pose reference for subsequent adaptive adjustment of the camera optical axis and acquisition of high-quality proximity images, solving the problem of insufficient positioning accuracy caused by the susceptibility of satellite signals under the bridge, and ensuring the stability and accuracy of flight and photography; including: Step S4-1: The UAV initiates flight along the three-dimensional proximity flight path and simultaneously acquires multi-source data: Flight path loading and flight initialization: The airborne processing and control unit retrieves the unstructured 3D proximity flight path data generated in step S3 from the storage module, including the 3D coordinates of all waypoints, the preset flight speed, and the transition method between adjacent waypoints; it parses the flight path data into UAV flight control commands, and initializes the UAV flight platform, including calibrating the inertial measurement unit zero bias, adjusting the gimbal initial attitude, and starting the preheating of each sensor device to ensure that the equipment is in a stable working state; after receiving the start command, the UAV takes off from the take-off and landing point, flies smoothly at the preset initial speed according to the coordinates of the first waypoint of the flight path, and enters the flight area under the bridge. Satellite differential signal reception: The GNSS receiver onboard the UAV receives satellite differential signals in real time from the positioning reference station deployed in the bridge area, and simultaneously receives raw satellite signals from the Global Navigation Satellite System. The receiver's built-in signal processing module demodulates and decodes the two signals synchronously, removing systematic errors such as satellite orbit errors and atmospheric delay errors, and generating preliminary absolute position data for the UAV. In response to potential obstruction or multipath effects of satellite signals in the space under the bridge, the receiver activates an anti-interference algorithm to filter valid satellite signals, ensuring the continuity of absolute position data and maintaining data output even in areas with weak signals to avoid interruptions. Real-time acquisition of inertial data: The inertial measurement unit (IMU) acquires the three-dimensional acceleration and three-dimensional angular velocity data of the drone at a high frequency (usually not less than 200Hz). These data directly reflect the real-time motion state of the drone, including acceleration, deceleration, turning, and attitude tilt. The IMU converts the acquired inertial data into digital signals and transmits them to the onboard processing and control unit in real time through a high-speed data interface. To reduce the interference of vibration on the inertial data, the vibration damping module on the IMU filters out high-frequency vibrations generated by the operation of the drone motors and airflow disturbances. At the same time, the onboard processing unit performs preliminary noise reduction processing on the inertial data to remove instantaneous pulse noise and ensure data stability. Image sequence acquisition by the visual sensor: The navigation camera and the reconnaissance camera in the visual sensor simultaneously start continuous shooting mode. The navigation camera acquires a continuous image sequence at a high frame rate (usually 15-30 frames / second) for subsequent calculation of the relative pose change of the UAV. Its shooting range covers the UAV's flight direction and surrounding area to ensure that enough environmental feature points can be captured. The reconnaissance camera adjusts the shooting interval according to the UAV's flight speed and the preset image overlap rate, usually taking one image every 0.5-1 meter of flight. The image resolution is no less than 20 megapixels to ensure that the surface details of the bridge components are clearly presented. Both cameras use a timestamp module to mark the precise shooting time of each image for subsequent data synchronization. Distance information acquisition by the ranging sensor: The ranging sensor (single-point laser rangefinder or two-dimensional lidar) continuously measures the straight-line distance from the UAV to the surface of the bridge component at a preset sampling frequency (usually 10-50Hz); the beam direction of the sensor is pre-calibrated and always aligned with the surface of the component in front of the UAV's flight path to ensure that the measured distance reflects the actual distance between the UAV and the component surface; the measured distance data is transmitted to the onboard processing and control unit in real time via the data bus, and the calibration module built into the sensor will periodically correct the drift of the measured values ​​to avoid the accumulation of measurement errors caused by long-term operation; for two-dimensional lidar, it will also output multiple distance point data within the scanning range simultaneously to form a local point cloud, providing more spatial information for subsequent registration with the initial three-dimensional real scene model; Multi-source data synchronization and integration: The airborne processing and control unit activates the time synchronization module, using the time signal from the GNSS receiver as a reference (its time accuracy is calibrated by the satellite signal, ensuring high stability). It uniformly calibrates the timestamps of satellite differential signals, inertial data, visual image sequences, and ranging information, ensuring that all data are aligned in the time dimension with errors controlled within milliseconds. The synchronized multi-source data is packaged and stored according to a preset format. Each data packet contains various sensor data at the corresponding time point. Simultaneously, a cache module temporarily stores the latest 10-20 seconds of data to prevent data loss due to instantaneous data transmission delays, providing a complete and synchronized data source for subsequent compact combination filtering and fusion. Step S4-2: Compact combination filtering fusion and high-precision UAV pose calculation: Kalman Filter Construction and Initialization: The airborne processing and control unit constructs a Kalman filter based on a preset algorithm. The state vector of this filter includes the UAV's three-dimensional position (coordinates along the east, north, and sky directions), three-dimensional velocity (velocities in the three directions), three-dimensional attitude (roll, pitch, and yaw angles), and error parameters of the inertial measurement unit (accelerometer bias, gyroscope bias, and calibration coefficient error). During the filter initialization phase, the initial values ​​of the state vector are set using the GNSS absolute position data and IMU initial attitude data (horizontal attitude calibrated by gravity vector, and yaw angle calibrated by GNSS heading). At the same time, the initial covariance matrix of the filter is set according to the technical parameters of each sensor device (such as GNSS positioning accuracy and IMU error characteristics) to reflect the degree of uncertainty of the initial state vector. Multiple observations preprocessing and input: Satellite differential signal observation processing: The preliminary absolute position data output by the GNSS receiver is subjected to outlier detection. By comparing the position changes at multiple consecutive time points, jump outliers caused by signal blockage are eliminated. The processed absolute position data is used as absolute position observation and input into a Kalman filter to constrain the global position of the UAV and avoid long-term drift. Visual relative pose observation processing: Feature extraction is performed on the continuous image sequence acquired by the navigation camera. Stable feature points such as corners and edges in each image are extracted using feature detection algorithms (such as scale-invariant feature transform algorithm). Feature points in adjacent images are matched using feature matching algorithm, and the displacement of feature points in the image coordinate system is calculated. Combining the intrinsic parameters (focal length, pixel size) and extrinsic parameters (camera mounting position and attitude relative to the UAV body) of the navigation camera, the image displacement is converted into the relative pose change of the UAV in space (including relative position and relative attitude), which is used as the input filter for the relative pose observation. Ranging and Normal Distance Observation Processing: The distance data collected by the ranging sensor is registered in real time with the initial 3D reality model. The onboard processing unit calls the surface data of the components near the current position of the UAV in the initial model (triangular mesh model) to calculate the angle between the measurement direction of the ranging sensor and the normal vector of the component surface. Based on the measured distance and this angle, the straight-line distance is converted into the normal distance of the UAV relative to the component surface (i.e., the distance along the normal vector direction of the component surface), eliminating the distance deviation caused by the inconsistency between the measurement direction and the normal vector. The converted normal distance is used as the normal distance observation and input into the Kalman filter to constrain the relative position of the UAV and the component surface, avoiding the distance being too close or too far. Visual feature point spatial constraint observation processing: For feature points extracted from navigation camera images, search for corresponding 3D feature points in the initial 3D reality model (by matching feature descriptors and associating them with spatial positions); for successfully matched feature point pairs, obtain the absolute coordinates of the 3D feature points in the initial model, use them as spatial position constraint observations, and input them into the Kalman filter; this observation can provide additional spatial position reference for the UAV when GNSS signals are weak (such as in the shadow area under a bridge), thus improving positioning stability; Filtering Prediction and Iterative Update: The Kalman filter operates according to an iterative "prediction-update" process. In the prediction phase, the acceleration and angular velocity data from the inertial measurement unit (IMU) are used in conjunction with preset motion equations to predict the UAV's state vector (position, velocity, attitude, and IMU error) for the next moment, and the prediction covariance matrix is ​​updated to reflect the uncertainty of the predicted state. In the update phase, the preprocessed multiple observations are compared with the predicted state vector, and the observation residual (the difference between the observed and predicted values) is calculated. The Kalman gain is calculated based on the observation residual and the observation covariance matrix (reflecting the accuracy of the observations). The predicted state vector is adjusted by the Kalman gain to reduce uncertainty and obtain the optimal state estimate (i.e., the current high-precision position and attitude of the UAV). The iteration cycle is consistent with the sampling frequency of the multi-source data to ensure that a filter update is completed for each set of synchronous data received, realizing real-time attitude calculation. Pose calculation result verification and error correction: The airborne processing and control unit verifies the rationality of the position and attitude results obtained from each filtering update. By comparing the position change from two consecutive calculations with the theoretical position change calculated from the velocity, it determines whether there is any abnormal deviation. If the deviation exceeds a preset threshold (e.g., position deviation greater than 0.1 meters, attitude deviation greater than 1 degree), an anomaly handling mechanism is activated, increasing the weight of visual feature point matching and ranging data registration to reduce the impact of abnormal GNSS data. At the same time, based on the calculated inertial measurement unit error parameters, the inertial data output by the IMU is corrected in real time to compensate for accelerometer zero bias and gyroscope zero bias, improving the accuracy of subsequent prediction stages. The verified high-precision pose results are transmitted in real time to the UAV flight control system and gimbal control system. The flight control system adjusts the flight attitude and velocity according to the pose results to ensure that the UAV flies along the preset three-dimensional close-fitting flight path. The gimbal control system prepares to adjust the camera optical axis direction based on the pose results to prepare for adaptive photography in step S5.

[0021] In this embodiment, step S5 is based on the high-precision attitude data of the UAV calculated in step S4. By adaptively adjusting the direction of the camera optical axis, it ensures that the optical axis of the camera is perpendicular to the local normal of the bridge component surface when the camera is captured, thereby acquiring close-up images with clear details and accurate angles. This provides high-quality image data support for the subsequent refined three-dimensional reconstruction and defect identification of the bridge component, while ensuring a high degree of adaptation between the image acquisition process and the UAV flight trajectory and the surface morphology of the component. Step S5-1: Real-time reception and analysis of high-precision UAV pose data: Attitude data reception: The airborne processing and control unit receives the high-precision UAV attitude data output from step S4-2 in real time. This data includes the UAV's three-dimensional position coordinates, namely the coordinates along the east, north, and sky directions, three-dimensional attitude parameters, namely roll angle, pitch angle, yaw angle, and the accuracy evaluation value of the attitude calculation. This evaluation value can reflect the reliability of this set of attitude data. The data is transmitted through the airborne internal high-speed data bus, and the transmission rate matches the attitude calculation frequency, usually not less than 50 Hz, to ensure that the data reception is without delay or loss. Pose data parsing: The airborne processing and control unit parses the received pose data, converting the raw data into standardized parameters that can be directly used for camera optical axis calculation. Among them, the three-dimensional position coordinates are used to determine the surface area of ​​the component where the UAV is currently located, the three-dimensional attitude parameters are used to establish the relationship between the UAV's body coordinate system and the local coordinate system of the bridge component, and the accuracy evaluation value is used as a reference for judging the accuracy of subsequent optical axis adjustment. If the accuracy evaluation value is lower than the preset threshold, it indicates that the reliability of the current pose data is insufficient. At this time, image acquisition should be paused and the operation should be restarted after the pose data accuracy recovers. Data timeliness verification: Compare the timestamp of the pose data with the current system time to determine if there is any data lag; if the time difference exceeds the preset threshold, usually within ten milliseconds, the data is deemed insufficient in timeliness, the data set is discarded and the system waits for the next set of latest pose data; at the same time, continuously monitor data lag, if lag occurs multiple times, trigger the cache optimization mechanism of the airborne processing and control unit to reduce intermediate links in data transmission and ensure that the pose data and camera control commands are synchronized in time; Step S5-2: Real-time acquisition of local normal directions on the surface of bridge components: Component surface area matching: Based on the UAV's three-dimensional position coordinates analyzed in step S5-1, locate the local area of ​​the component surface directly below the UAV or the corresponding shooting area in the bridge key component triangular network model constructed in step S3; quickly match the triangular patch set corresponding to the area through the spatial coordinate retrieval algorithm to ensure that the matching result accurately corresponds to the component range currently covered by the UAV's shooting, and avoid deviation of normal direction caused by cross-region matching; Local normal direction extraction: For the set of triangular facets in the local area of ​​the matched component surface, extract the unit normal vector of each triangular facet. The direction of this normal vector has been unified in step S3-2, and they all point to the outside of the component. The weighted average algorithm is used to calculate the comprehensive normal direction of the local area. The unit normal vectors of all triangular facets are weighted and summed, and the summation result is normalized to obtain the average unit normal direction of the local area. This direction is the reference direction that the camera optical axis needs to be perpendicularly aligned with. Dynamic update of normal direction: As the UAV flies along the 3D close-up flight path, the surface area matching and local normal direction extraction of the component are repeated in real time to ensure that the normal direction is dynamically updated as the UAV's shooting position changes; the update frequency is adapted to the UAV's flight speed. For example, when the UAV's flight speed is one meter per second, the normal direction update frequency is no less than ten hertz, ensuring that the normal direction of the current area is reacquired every 0.1 meters of flight, providing a real-time reference for optical axis adjustment; Step S5-3: Calculation of the target direction of the camera optical axis: Coordinate system transformation: Establish a multi-coordinate system relationship to realize the transformation from the component's local coordinate system to the UAV's fuselage coordinate system; First, transform the local average unit normal direction of the component surface obtained in step S5-2 from the component's local coordinate system to the global coordinate system. The component's local coordinate system is established with a fixed point of the component as the origin and along the length, width, and height directions of the component, while the global coordinate system is consistent with the positioning reference station's coordinate system; Then, combined with the UAV's three-dimensional attitude parameters analyzed in step S5-1, transform the normal direction in the global coordinate system to the UAV's fuselage coordinate system to obtain the component's local normal direction vector in the fuselage coordinate system; Optical axis target direction determination: Based on the local normal direction vector of the component in the camera coordinate system, determine the target direction vector of the camera optical axis; since the camera needs to photograph the surface of the component, the target direction of the optical axis must be opposite to the local normal direction of the component, that is, pointing towards the surface of the component; if the local normal direction vector of the component in the camera coordinate system is upward and away from the surface of the component, then the target direction vector of the camera optical axis is downward and pointing towards the surface of the component, and the two are completely collinear in space and opposite in direction, ensuring that the optical axis is perpendicular to the local normal of the component surface; Target direction accuracy verification: Calculate the angle between the optical axis target direction vector and the component local normal direction vector. If the angle is not equal to 180 degrees, the allowable error range is ±0.5 degrees. Then, re-examine the coordinate system transformation process and vector calculation steps to eliminate coordinate transformation errors or vector calculation deviations. If the angle is within the allowable error range, the optical axis target direction is confirmed to be valid. The camera attitude parameters corresponding to this direction, namely roll angle, pitch angle, and yaw angle, are used as the target command parameters for gimbal adjustment. Step S5-4: Adaptive attitude adjustment and control of the gimbal: Target attitude command issuance: The airborne processing and control unit converts the camera target attitude parameters confirmed in step S5-3 into control commands that the gimbal can recognize, such as motor rotation angle and speed, and issues them in real time to the gimbal close to the photography subsystem via the gimbal control bus; a check code mechanism is used during command transmission to ensure that the command data transmission is accurate and free from errors or loss. Real-time gimbal attitude adjustment: After receiving the target attitude command, the gimbal activates its built-in motor drive system to adjust the gimbal's attitude along the roll, pitch, and yaw axes. During the adjustment process, the gimbal's built-in miniature inertial measurement unit collects the current gimbal attitude data in real time at a sampling frequency of no less than 100 Hz, and feeds the data back to the airborne processing and control unit to form a closed-loop control. The airborne processing and control unit compares the current attitude data with the target attitude parameters, calculates the attitude deviation, and dynamically adjusts the rotation speed and force of the gimbal motors according to the magnitude of the deviation. When the deviation is large, such as greater than two degrees, the adjustment speed is increased; when the deviation is small, such as less than 0.5 degrees, the speed is reduced to avoid overshoot. Accuracy Adjustment Confirmation: When the deviation between the current gimbal attitude data and the target attitude parameters is less than the preset accuracy threshold, typically within a ±0.3 degree range, the gimbal attitude is considered to be adjusted correctly. The onboard processing and control unit sends a stop adjustment command to the gimbal, and the gimbal maintains its current stable attitude. If the attitude deviation continues to exceed the threshold during the adjustment process, such as for more than five seconds, the gimbal fault diagnosis mechanism is activated to check whether the motor is stuck or whether the micro inertial measurement unit is abnormal. If it is a slight sticking, the motor drive torque is increased to attempt adjustment. If it is a hardware malfunction, the current shooting point is marked as an abnormal area, and reshooting is performed later. Step S5-5: Precise acquisition control of close-up images: Data Acquisition Trigger Condition Judgment: The airborne processing and control unit monitors in real time whether two key trigger conditions are met: First, whether the UAV has reached the waypoint position preset in step S3. By comparing the current three-dimensional position coordinates of the UAV with the waypoint coordinates, a deviation of less than 0.2 meters indicates arrival. Second, whether the gimbal attitude has been adjusted to the correct position, i.e., the attitude deviation confirmed in step S5-4 is less than the threshold. Only when both conditions are met simultaneously will the close-up image acquisition command be triggered to avoid image angle shift and detail blurring due to position deviation or attitude inaccuracy. Adaptive adjustment of shooting parameters: Before the camera receives the acquisition command, the onboard processing and control unit obtains the light intensity data of the space under the bridge in real time through the camera's built-in metering sensor, and adaptively adjusts the camera shooting parameters according to the light intensity; for example, when the light is sufficient and the light intensity is greater than 5,000 lux, the sensitivity is reduced, such as by setting it to ISO 100, and the shutter speed is shortened, such as by setting it to 1 / 1000th of a second, to avoid overexposure of the image; when the light is dim and the light intensity is less than 1,000 lux, the sensitivity is appropriately increased, not exceeding ISO 800 to reduce noise, and the shutter speed is extended, not exceeding 1 / 100th of a second to avoid blur, while the camera's image stabilization function is turned on to ensure uniform image brightness and clear details; Acquisition frequency coordinated control: Based on the UAV's flight speed and the forward and lateral overlap rates preset in step S3, the camera acquisition frequency is determined; for example, if the UAV's flight speed is one meter per second, the preset forward overlap rate is 80%, and the camera's field of view at the shooting height is five meters, then the shooting interval between adjacent images is one meter, that is, the acquisition frequency is one frame per second, ensuring that the forward overlap area of ​​adjacent images reaches four meters, and five meters multiplied by 80% meets the overlap rate requirements; the airborne processing and control unit dynamically adjusts the shooting interval by associating the UAV's flight speed and acquisition frequency, avoiding insufficient or excessive overlap rates due to flight speed fluctuations, as excessive overlap rates will increase data redundancy; Step S5-6: Real-time association and storage of image data and pose data: Data association and tagging: After each close-up image is acquired, the airborne processing and control unit immediately associates and tags the image data with the corresponding high-precision pose data of the UAV and the camera shooting parameters at that time. Through a unified timestamp, accurate to the millisecond level, the three types of data are bound to ensure that each image can accurately correspond to its spatial position, attitude state and shooting settings at the time of shooting, providing a data association basis for the subsequent refined 3D reconstruction in step S6. Among them, the high-precision pose data of the UAV includes 3D position coordinates and 3D attitude parameters, and the camera shooting parameters include shutter speed, ISO, and aperture size. Data format standardization processing: The associated data is converted into a standardized storage format; among them, image data adopts a lossless compression format, such as TIFF format, to retain all the detailed information of the image; pose data and shooting parameters adopt an extensible markup language format, such as XML format, to clearly record the names and values ​​of each parameter; the three types of data are named with the same file name prefix, such as using a timestamp as a prefix, to facilitate subsequent retrieval and retrieval; Secure Data Storage: Correlated and standardized data is stored through a dual-path storage mechanism. On one hand, it is stored on the UAV's onboard storage device, such as a high-speed solid-state drive, ensuring local data accessibility. On the other hand, it is uploaded in real time to the storage server at the ground control center via a wireless communication module, such as a 4G or 5G module, achieving off-site data backup. Data encryption algorithms are used during storage to prevent unauthorized access or tampering. A storage caching mechanism is also enabled. If wireless transmission is interrupted, the data is temporarily stored in the onboard buffer and uploaded again after transmission is restored to avoid data loss. Step S5-7: Real-time verification of data acquisition quality and anomaly handling: Real-time image quality verification: After each image is stored, the onboard processing and control unit starts the image quality verification algorithm to automatically detect the image's sharpness, exposure, and color reproduction. Sharpness detection calculates the gradient value of the edges of components in the image; the higher the gradient value, the sharper the edges. If the gradient value is lower than a preset threshold, the image is considered blurry. Exposure detection analyzes the image's grayscale histogram; if the histogram peak is biased towards the brightest or darkest area, the image is considered overexposed or underexposed. Color reproduction detection compares the colors of known components in the image, such as the natural color of bridge concrete, with standard color values; if the deviation is too large, the color is considered abnormal. Pose and optical axis matching verification: Verify the matching between pose data and optical axis direction by reverse verification of image content; for example, observe whether the edge of the bridge component in the image is orthogonal. If the optical axis is perpendicular to the surface of the component, the edge of the component should not be tilted or deformed. If there is obvious tilt, it is determined that the optical axis direction is deviated; compare the correspondence between the image shooting position and waypoint coordinates. If the component area in the image does not match the preset coverage area of ​​the waypoint, it is determined that the pose data deviates from the actual position. Anomaly Handling Mechanism: If the verification detects abnormal image quality or pose / optical axis mismatch, the airborne processing and control unit immediately triggers anomaly handling. For images with slight blurring or slight exposure deviation, they are marked for manual review, and the data is retained for subsequent manual judgment on usability. For images with severe blurring, overexposure, underexposure, or severe optical axis deviation, they are marked as invalid images, and a re-acquisition command is sent to the gimbal and flight control system. The flight control system controls the UAV to pause its forward movement and fine-tune its position back to the original waypoint. The gimbal re-executes the attitude adjustment process, and after adjustment, it re-triggers image acquisition until a qualified image is acquired. If the same waypoint is acquired three times consecutively and all images are invalid, the coordinates of the waypoint and the cause of the anomaly are recorded and fed back to the ground control center for subsequent manual planning of a reshoot path.

[0022] In this embodiment, step S6 is based on the close-up image and its corresponding high-precision position and attitude data collected in step S5. Through data processing, three-dimensional reconstruction, defect identification and measurement, it outputs quantitative results of bridge inspection, provides a basis for bridge safety assessment and maintenance decisions, and forms a complete closed loop from data collection to analysis and application. Step S6-1: Preprocessing optimization of close-up image and pose data: Close-up image quality optimization: Denoising is performed on the close-up images acquired in step S5 to eliminate noise while preserving details such as component edges and textures; to address the issue of uneven lighting under the bridge, lighting equalization is performed to make the overall brightness and darkness distribution of the image more uniform; overexposed, underexposed, or blurry abnormal images are removed to ensure that the image quality for subsequent processing meets the standards. Post-processing optimization of pose data: Secondary correction is performed on the associated high-precision pose data of UAVs, and global smoothing is performed in combination with multi-source sensor data to eliminate pose fluctuations caused by instantaneous disturbances; the pose data is aligned with the coordinate system of the ground base station to ensure that the coordinates of the subsequent 3D reconstruction results are consistent with the actual position of the bridge and to avoid coordinate system offset. Data association consistency verification: Re-associate the optimized images with pose data using timestamps to confirm that each image corresponds to unique and accurate pose data; check the matching between the image shooting angle and the component surface. If an association anomaly is found, re-match the correct data to ensure the accurate correspondence between images and poses. Step S6-2: Refined 3D Reconstruction of Bridge Components Multi-view image feature extraction and matching: Extract stable key feature points from the preprocessed images to generate feature description information; match feature points between different images, eliminate mismatched points by combining geometric constraints, and retain the correct matching results that can reflect the spatial relationship of the images; Sparse point cloud optimization and dense point cloud generation: Based on feature matching results and optimized pose data, sparse point clouds of bridge components are generated; the sparse point cloud is optimized by an algorithm to reduce reprojection error; on this basis, a denser point cloud with higher density is generated to clearly present the surface details of the components. 3D mesh model construction: Denoising and smoothing of dense point clouds, and using surface reconstruction algorithms to construct a continuous 3D mesh model; Optimizing the mesh model, filling in small holes and eliminating irregular protrusions to ensure that the model can accurately reflect the actual geometric shape of the components; High-precision texture mapping: The pre-processed image is used as the texture data source. Based on the spatial relationship between the camera pose and the mesh model, the texture is accurately mapped onto the mesh surface. The textures in overlapping areas are fused to avoid stitching marks, so that the 3D model presents a visual effect consistent with the real components. Step S6-3: Automatic identification of bridge component defects: 3D model component segmentation: The refined 3D mesh model is decomposed into independent key bridge component models. Segmentation rules are established based on the size, shape, spatial location, and other characteristics of the components. Mesh patches belonging to the same component are classified to achieve precise separation of components. Disease feature extraction: Analyze the texture and point cloud data of individual component models to extract features that may indicate diseases, such as gray-scale abrupt changes and surface unevenness; fuse texture features and point cloud features to form a feature set of candidate disease regions to avoid missing small or hidden diseases. Disease classification and identification: A machine learning-based classification model is used to determine the disease type of the candidate region by taking the features of the candidate region as input; the identification results are post-processed to eliminate misidentified regions, merge adjacent diseases of the same type, and generate a disease distribution map to intuitively show the location and type of disease; Step S6-4: Precise Measurement of Bridge Defects Geometric parameter calculation of diseases: Based on the three-dimensional model and dense point cloud data, the identified diseases are measured, and the length and width of cracks, the area of ​​the spalling area, and the depth of diseases such as exposed reinforcement and depressions are calculated according to the three-dimensional coordinates to obtain the quantitative indicators of diseases. Measurement accuracy optimization: Improve measurement accuracy through cross-validation of multi-source data, compare measurement results from different perspectives, and correct measurement values ​​by combining previously collected distance data; perform statistical analysis on measurement data, remove outliers, and ensure that measurement errors are controlled within a reasonable range; Step S6-5: Results Verification and Output: 3D model accuracy verification: Select control points on the bridge site, measure their actual 3D coordinates, and compare them with the coordinates of the corresponding points in the model to verify the geometric accuracy of the model; compare the actual dimensions of the components with the model dimensions to ensure that the deviations meet the inspection accuracy requirements; Disease identification accuracy verification: Professional inspectors manually mark diseases in some areas, compare the manual marking results with the automatic identification results, and calculate the identification accuracy, precision, and recall rate to ensure the reliability of the identification results; analyze and optimize the identification error cases to improve the subsequent identification effect; Inspection results output: The 3D model, disease distribution and measurement data are organized into standardized results, and output 3D model files, disease statistics tables and distribution maps in a general format are output; an inspection report containing inspection methods, accuracy descriptions, disease analysis conclusions and maintenance suggestions is generated to provide a comprehensive reference for bridge management.

[0023] The equipment for drone positioning and close-up photography in the space under the bridge includes: The ground reference subsystem includes positioning reference stations deployed in the bridge area; Unmanned aerial vehicle (UAV) flight platform; The integrated positioning subsystem, integrated into the UAV flight platform, includes: A GNSS receiver is used to receive satellite signals and differential signals from a reference station. Inertial Measurement Unit (IMU); Visual sensors are used to acquire continuous sequences of images; Distance sensors are used to measure distances to the surface of bridges. The close-up photography subsystem is integrated into the drone's flight platform and includes a high-resolution camera and a gimbal for adjusting its attitude. The airborne processing and control unit is configured as follows: Perform compact combination filtering fusion to solve for the high-precision pose of the UAV; Plan unstructured 3D close-up flight routes based on the initial 3D reality model; Based on the calculated pose and planned flight path, the drone flight platform and gimbal are controlled to complete autonomous flight and adaptive shooting.

[0024] Furthermore, the visual sensors include a navigation camera for pose calculation and a reconnaissance camera for capturing close-up images. The navigation camera and the reconnaissance camera are physically isolated and have independent parameters.

[0025] Furthermore, the ranging sensor is a single-point laser rangefinder or a two-dimensional lidar, and its installation position and beam direction are calibrated so that its measurement value has a known spatial transformation relationship with the optical axis direction of the survey camera.

[0026] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for positioning and close-up photography of an unmanned aerial vehicle in a bridge space, characterized in that: The method comprises the following steps: S1: data collection preparation: positioning reference stations are arranged in the bridge area, and a combined positioning system and a shooting system are configured for the unmanned aerial vehicle; S2: coarse scanning and initial modeling: the unmanned aerial vehicle is controlled to perform preliminary scanning flight in the space under the bridge, and wide-angle images and positioning data are synchronously collected; based on the wide-angle images and the positioning data, an initial three-dimensional real scene model of the space under the bridge is generated; S3: close-up photography path planning: based on the initial three-dimensional real scene model, key components of the bridge are identified, and an unstructured three-dimensional close-up flight path that is adapted to the surface morphology of each key component is generated; S4: tight combined positioning and adaptive photography: the unmanned aerial vehicle is controlled to fly along the three-dimensional close-up flight path; in the process of flight, the following operations are performed: S4-1: satellite differential signals from the positioning reference stations, inertial data of an onboard inertial measurement unit, image sequences of a visual sensor, and distance information of a ranging sensor to the surface of the bridge are synchronously acquired; S4-2: the initial three-dimensional real scene model is taken as spatial constraint prior information, and is combined with the various sensing data acquired in step S4-1 to perform tight combined filtering fusion, so that the high-precision position and attitude of the unmanned aerial vehicle in the space under the bridge are solved in real time; S5: the shooting system of the unmanned aerial vehicle is controlled to adaptively adjust the direction of the camera optical axis according to the high-precision attitude solved in step S4-2, so that the camera optical axis is perpendicular to the local normal of the surface of the bridge component when shooting, and close-up images are collected. 2.The method of claim 1, wherein the method further comprises: determining a distance between the UAV and the target object; and determining a distance between the UAV and the target object based on the distance between the UAV and the target object and the height of the target object. In step S3, the unstructured three-dimensional close-up flight path is generated for the key component, and specifically comprises: S3-1: the key component of the bridge to be finely scanned is separated from the initial three-dimensional real scene model; S3-2: the normal vector of the surface of the component triangular mesh model is calculated; S3-3: a three-dimensional virtual photography curved surface parallel to the component surface is generated by extrapolation guided by the normal vector; S3-4: dense flight points are generated on the three-dimensional virtual photography curved surface according to the preset forward and lateral overlap ratios, and the camera orientation of each flight point is set to be aligned with the normal vector direction of the corresponding component surface. 3.The method of claim 1, wherein the method further comprises: determining a distance between the UAV and the target object; and determining a distance between the UAV and the target object based on the distance between the UAV and the target object and the height of the target object. In step S4-2, the tight combined filtering fusion specifically comprises: a Kalman filter is constructed to include the position, velocity, attitude of the unmanned aerial vehicle and the inertial sensor error; the absolute position provided by the satellite differential signals, the relative attitude change provided by the visual sensor, and the distance from the ranging sensor to the surface of the bridge are taken as multiple observation inputs of the Kalman filter; wherein the distance information provided by the ranging sensor is converted into the normal distance observation of the unmanned aerial vehicle relative to the model surface by real-time registration with the initial three-dimensional real scene model, and is used to correct the state estimation of the filter.

4. The bridge space unmanned aerial combination positioning and close photography method according to claim 3, characterized in that: In the tight combined filtering fusion, the feature points extracted by the visual sensor are matched with the corresponding three-dimensional points in the initial three-dimensional real scene model, and the coordinates of the matched three-dimensional points are taken as spatial position constraint observations and input into the Kalman filter.

5. The method of claim 1, wherein the method further comprises: determining a distance between the UAV and the target object; and determining a distance between the UAV and the target object based on the distance between the UAV and the target object and the height of the target object. Further comprising: S6: result generation and analysis: based on the close-up images collected in step S5 and the corresponding high-precision position and attitude, fine three-dimensional reconstruction is performed to generate a high-precision real scene model of the bridge component; and automatic identification and measurement of diseases are performed based on the model.

6. A bridge space unmanned aerial vehicle combined positioning and close-up photography equipment for implementing the method of any one of claims 1-5, characterized in that: Comprise: The ground reference subsystem comprises a positioning reference station arranged in the bridge area; The UAV flight platform; The integrated positioning subsystem comprises: A GNSS receiver for receiving satellite signals and differential signals of the reference station; An inertial measurement unit (IMU); A visual sensor for collecting a continuous image sequence; A ranging sensor for measuring the distance to the bridge surface; The close-up photography subsystem comprises a high-resolution camera and a gimbal for adjusting the attitude of the camera; The onboard processing and control unit is configured to: Perform tight combined filtering fusion to calculate the high-precision position and pose of the UAV; Plan an unstructured three-dimensional close-up flight path based on an initial three-dimensional real scene model; Control the UAV flight platform and the gimbal according to the calculated position and pose and the planned flight path to complete autonomous flight and adaptive photography.

7. The bridge space unmanned aerial combination positioning and close-up photography equipment according to claim 6, characterized in that: The visual sensor comprises a navigation camera for position and pose calculation and an exploration camera for close-up image shooting, and the navigation camera is physically separated from the exploration camera and has independent parameters.

8. The bridge space unmanned aerial combination positioning and close-up photography equipment according to claim 6, characterized in that: The ranging sensor is a single-point laser range finder or a two-dimensional laser radar, and the installation position and beam direction thereof are calibrated so that the measurement value thereof and the optical axis direction of the exploration camera have a known spatial transformation relationship.

Images