Unmanned aerial vehicle auxiliary image control measurement method

Abstract

The invention discloses an unmanned aerial vehicle auxiliary image control measurement method, and relates to the technical field of unmanned aerial vehicle aerial survey. An unmanned aerial vehicle auxiliary image control measurement method comprises an image control point arrangement method and an image control point arrangement method. According to the method, historical aerial photography data and a YOLO model are introduced for automatic identification, multi-scale texture features and multi-view matching are combined, high-credibility image control point initial particles are automatically generated, and the problems that traditional image control point distribution depends on artificial experience, efficiency is low and quality is uneven are solved; by constructing a GNSS shielding probability graph, a reflecting surface density graph and an ionization disturbance graph and based on a group optimization algorithm, an optimal point distribution position with the minimum error influence is intelligently screened out, self-adaptive analysis and layout optimization of a signal environment are achieved, and error source aggregation is effectively avoided.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) aerial surveying technology, and in particular to a UAV-assisted image control measurement method. Background Technology

[0002] Currently, UAV aerial surveying is widely used in large-scale digital mapping in areas with complex terrain. However, in such areas, traditional image control point (ADC) deployment and measurement methods face the following technical challenges due to the frequent influence of GNSS signals from factors such as mountain obstruction, tree canopy cover, building reflection, and ionospheric disturbance: ADC deployment relies on manual experience, is inefficient, and it is difficult to deploy reasonable control points in areas with degraded signals; GNSS obstruction and reflection effects cause fluctuations in the accuracy of control point coordinates, resulting in unstable aerial triangulation and geometric distortion or error accumulation in the mapping results; existing methods are unable to automatically adapt deployment and measurement strategies to different GNSS environments and lack intelligent error identification and compensation mechanisms.

[0003] Therefore, a method for selecting control points that takes into account different environments is needed. This method enables adaptive optimization of control point deployment, intelligent selection of regional classification measurement strategies, and separation and correction of error sources in complex environments, fundamentally improving the control accuracy and measurement reliability of UAV mapping in areas with unstable GNSS signals. Summary of the Invention

[0004] Technical problems to be solved This invention aims to provide a UAV-assisted image control measurement method, which realizes adaptive optimization of image control point deployment and intelligent selection of regional classification measurement strategies.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a UAV-assisted image control measurement method, comprising: Step S1: Image control point deployment method; Acquire historical aerial photography data of the area to be measured; generate initial control point particles using the historical aerial photography data; acquire the occlusion probability map of GNSS signals in the area to be measured; analyze the occlusion probability map and the initial control point particles using a population optimization algorithm to obtain the preferred control point particles; use the preferred control point particles as pre-selected points for control point measurement. Obtain the real-time coordinates of the pre-selected points for image control measurement; classify all pre-selected points into regions with stable reception and unstable reception. Step S2: UAV image control measurement method; For the pre-selected points of the image control measurement in the stable receiving area, the standard UAV image control measurement method is used to measure and obtain the measurement data of the stable area. For pre-selected points for image control measurement in unstable areas, auxiliary measurement rules are established; measurements are then performed using the auxiliary measurement rules and standard UAV image control measurement methods to obtain measurement data for unstable areas.

[0006] As a preferred embodiment of the present invention, the specific steps for generating initial particles for image control points using historical aerial photography data include: Historical aerial photography data contains image data of the area to be measured and geographic reference information of the measurement area; The neighborhood image of the area to be measured is obtained based on the geographic reference information of the measurement area; a multi-scale texture histogram is generated based on the neighborhood image of the area to be measured; and a pre-trained YOLO model is used to identify the image data of the area to be measured and the multi-scale texture histogram to obtain the location of candidate control points. Each candidate control point position is marked as a candidate control point particle; coordinate matching is performed in multiple image data of the area to be measured containing candidate control point particles; if the candidate control point particle satisfies at least three viewpoints, the candidate control point particle is retained as the initial control point particle; otherwise, the candidate control point particle is discarded. By iterating through all candidate image control point particles, the final image control point initial particles are obtained.

[0007] As a preferred embodiment of the present invention, the specific steps for analyzing the occlusion probability map and the initial particles of the control points using a population optimization algorithm include: Within the area to be measured, acquire GNSS signals, historical DSM data, and historical meteorological data; use a three-dimensional geographic model to simulate the GNSS signals, output the occlusion probability value of all unit pixels within the area to be measured in unit pixels, and generate an occlusion probability map; Historical DSM data were processed using a 3D geographic model to obtain a reflectance density map; historical meteorological data were processed using physical calculation methods to obtain an ionization wave map. By fusing image features from the occlusion probability map, the reflectance density map, and the ionization wave map, a simulated image control deployment map is obtained; The initial particles of the control points are screened in the simulated control point layout map using a swarm optimization algorithm to obtain the preferred control point particles.

[0008] As a preferred embodiment of the present invention, the specific steps for measurement using a standard UAV image-controlled measurement method include: The three-dimensional geographic coordinates of the pre-selected points for image control measurement within the stable receiving area are obtained, resulting in multiple three-dimensional measurement coordinates; Set standard flight parameters for the UAV; obtain multiple control point images from the UAV for selective barrage; Multiple control point puncture images and multiple measurement 3D coordinates are paired to define the image coordinates of the control point puncture images; N pre-selected control measurement points are randomly selected as coordinate verification points; the reprojection error of the control point puncture images containing coordinate verification points is calculated to obtain the control measurement error factor; Aerial triangulation is performed on the selected images of the control points with image coordinates to obtain the aerial triangulation measurement results; the aerial triangulation measurement results are calibrated using the control point measurement error factor to obtain the adjusted aerial triangulation measurement results; Based on the adjusted aerial triangulation results, corresponding DOM, DSM, and DEM are generated and combined to obtain stable region measurement data.

[0009] As a preferred embodiment of the present invention, the specific steps for establishing auxiliary measurement rules include: Step A1: For the pre-selected image control measurement points in the unstable receiving area, M auxiliary control points are automatically generated within a radius of r around the pre-selected image control measurement points; the auxiliary control points and the corresponding pre-selected image control measurement points are combined to obtain the pre-selected image control measurement point cluster. Step A2: Based on the original standard flight parameters, add flight path guidance rules, specifically by randomly using different headings or shooting angles to obtain no less than J images containing pre-selected points for image control measurement; Determine the fluctuation complexity of the receiving unstable region; based on the fluctuation complexity, establish the mapping rules with steps A1 and A2 to obtain the auxiliary measurement rules.

[0010] As a preferred embodiment of the present invention, the specific steps for measurement using auxiliary measurement rules and standard UAV image control measurement methods include: For receiving unstable areas, auxiliary measurement rules are used to conduct UAV measurements to obtain the image set of pre-selected control points corresponding to the control point selection image set; Acquire real-time GNSS signals; construct a pre-selected 3D plane for image control measurement based on the pre-selected point cluster; perform calibration in the pre-selected 3D plane for image control measurement using the anti-occlusion interpolation method to obtain the occlusion error calibration factor; A direction-weighted projection model is constructed based on real-time GNSS signals and image sets of selected control points, and a reflection error calibration factor is output based on the direction-weighted projection model. Ionization characteristics of real-time GNSS signals are extracted to obtain real-time ionization interference characteristics; short-time compensation is performed based on real-time ionization interference characteristics to obtain ionization error calibration factors; The occlusion error calibration factor, reflection error calibration factor, ionization error calibration factor, and image control measurement error factor are fused to obtain the fused error calibration factor. Based on the fused error calibration factor, subsequent operations are performed on the aerial triangulation measurement results to obtain measurement data for unstable regions.

[0011] As a preferred embodiment of the present invention, the swarm optimization algorithm is the particle swarm optimization algorithm.

[0012] Compared with the prior art, the present invention provides a UAV-assisted image control measurement method, which has the following beneficial effects: 1. This invention introduces historical aerial photography data and the YOLO model for automatic identification, and combines multi-scale texture features and multi-view matching to automatically generate high-reliability initial particles for ground control points, avoiding the problems of traditional ground control point placement relying on manual experience, low efficiency, and uneven quality. By constructing GNSS occlusion probability maps, reflector density maps, and ionization disturbance maps, and based on a swarm optimization algorithm, it intelligently selects the optimal placement locations with the least error impact, realizing adaptive analysis and placement optimization of the signal environment, and effectively avoiding the accumulation of error sources.

[0013] 2. This invention enhances measurement redundancy and geometric constraints by introducing auxiliary measurement rules in unstable regions, thereby improving the stability of aerial triangulation and reducing the risks of registration failure and model breakage caused by GNSS anomalies. By integrating a composite calibration mechanism that combines occlusion error, reflection error, ionization error and image control error, calibration factors are generated based on anti-occlusion interpolation, direction weight projection and ionization disturbance prediction methods, respectively, to achieve the separation modeling and real-time compensation of error sources, significantly improving the accuracy in unstable regions. Attached Figure Description

[0014] Figure 1 This is a diagram illustrating the steps of the method of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] A method for UAV-assisted image control measurement, see [link to relevant documentation] Figure 1 As shown, it includes: Step S1: Image control point deployment method; Acquire historical aerial photography data of the area to be measured; generate initial control point particles using the historical aerial photography data; acquire the occlusion probability map of GNSS signals in the area to be measured; analyze the occlusion probability map and the initial control point particles using a population optimization algorithm to obtain the preferred control point particles; use the preferred control point particles as pre-selected points for control point measurement. Historical aerial photography data refers to aerial photographic images and related auxiliary data of the area to be measured acquired by aircraft, drones or satellite platforms in the past period. Historical aerial photography data mainly includes the following: (1) Image data of the area to be measured, that is, aerial images covering the area to be measured, with a resolution usually between 0.1 meters and 1 meter, depending on the flight altitude and camera parameters. These images can be used to identify stable feature points; (2) Geographic reference information of the measurement area: including the image exterior orientation elements, including the position and attitude of the camera center when shooting, used to project the image onto the real ground coordinate system; camera interior orientation parameters, including focal length, principal point position, distortion parameters, etc., used to perform pixel coordinate and image space geometry conversion; geographic reference information can be used to assist image positioning and ground feature identification, and improve the initial particle recognition accuracy.

[0017] The methods for obtaining historical aerial photography data typically include the following: obtaining publicly available geographic aerial photography data; or archiving data from missions previously performed by existing drones or manned flight platforms.

[0018] The specific steps for generating initial particles for image control points using historical aerial photography data include: Historical aerial photography data contains image data of the area to be measured and geographic reference information of the measurement area; The neighborhood image of the area to be measured is obtained based on the geographic reference information of the measurement area; a multi-scale texture histogram is generated based on the neighborhood image of the area to be measured; and a pre-trained YOLO model is used to identify the image data of the area to be measured and the multi-scale texture histogram to obtain the location of candidate control points. Based on georeferenced information, a neighborhood range with a fixed radius (e.g., 50 to 200 meters) is set at the boundary of the existing area to be measured. Taking the area to be measured as the center, image data blocks covering its surroundings are extracted to form a neighborhood image. This neighborhood image is used to supplement the problems that the area to be measured may have, such as weak texture and poor image quality, and to improve the stability and robustness of subsequent candidate point matching.

[0019] On the extracted neighborhood images, a multi-scale sliding window strategy is used to perform texture feature analysis on each image patch. Local texture operators (such as LBP, Gabor filtering, and Haar features) can be applied to extract the local texture response of the image at multiple window sizes (such as 16×16, 32×32, and 64×64 pixels). Gray-level co-occurrence matrix or edge intensity statistics are calculated for image patches at each scale to form corresponding texture histograms. Finally, the histogram features at different scales are concatenated into a set of multi-scale texture histograms with contextual feature expression capabilities. This multi-scale texture histogram not only reflects the changes in local pattern texture but also includes the edge sharpness and structural complexity of candidate regions at different resolutions, providing a basis for the YOLO model to make judgments.

[0020] After extracting the multi-scale texture histogram, it is combined with the image data of the region to be tested and input into a specially trained YOLO object detection model. The model input includes: the image data of the region to be tested (main image channel) and the corresponding multi-scale texture histogram (as an additional channel or conditional label input). The model extracts spatial and semantic features through a convolutional neural network and outputs a bounding box containing the location of candidate control points and its classification confidence. The training set for training the YOLO object detection model consists of the following parts: positive samples are manually annotated control point image regions, and negative samples are background regions (such as trees, water surfaces, and bare ground) and target-like patterns that may cause misclassification.

[0021] Each candidate control point position is marked as a candidate control point particle; coordinate matching is performed in multiple image data of the area to be measured containing candidate control point particles; if the candidate control point particle satisfies at least three viewpoints, the candidate control point particle is retained as the initial control point particle; otherwise, the candidate control point particle is discarded. Each candidate control point (CNP) identified by the object detection model is labeled as a candidate CNP particle. This particle contains its image coordinates in the main view (i.e., the image being identified) and includes attributes such as recognition confidence and texture feature encoding. All aerial photographs containing the region of this particle are automatically retrieved from historical aerial photography data, constructing a corresponding multi-view image set. Each image must contain the ground location of the candidate particle, and the shooting angles differ; these angle differences can be calculated using attitude parameters in the exterior orientation elements or flight path records. In the constructed multi-view set, a matching method based on local image features (such as SIFT, SuperPoint, ORB, etc.) is used to identify the image location corresponding to the candidate particle in each image. To improve accuracy and stability, the matching process can employ bidirectional nearest neighbor matching combined with the RANSAC algorithm to eliminate mismatches. For each candidate CNP particle, the number of locations where it successfully completes a high-confidence match in different images is counted. If a candidate particle is successfully identified in at least three images from different viewpoints (i.e., the difference between the three shooting positions exceeds a set angle threshold, such as 10°~15°), then the particle is considered to meet the three-view reconstruction conditions, has a good stereo geometric foundation, and is retained as the initial particle of the control point; otherwise, if the number of matches is insufficient for three views or the difference in the viewpoints is too small (i.e., slight offset in the same direction), the particle is discarded to avoid model instability due to weakened spatial geometry in subsequent aerial triangulation calculations.

[0022] By iterating through all candidate image control point particles, the final image control point initial particles are obtained.

[0023] The specific steps for analyzing the occlusion probability map and initial particles of control points using a population optimization algorithm include: Within the area to be measured, GNSS signals, historical DSM data, and historical meteorological data are acquired. A 3D geographic model is used to simulate the GNSS signals, outputting the occlusion probability value of all unit pixels within the area to be measured, generating an occlusion probability map. Environmental input data related to GNSS signals within the area to be measured needs to be acquired. GNSS signal data includes ephemeris information, satellite orbit distribution, and transmission characteristics of commonly used frequency bands (such as L1 / L2), which can be downloaded from relevant publicly available databases. Historical DSM (Digital Surface Model) data provides elevation information of terrain, buildings, and vegetation within the area; the higher the resolution, the more sensitive it is to occlusion accuracy. This data usually originates from lidar mapping or high-resolution aerial survey results. Historical meteorological data covers local temperature, humidity, air pressure, and ionospheric activity index within a specific time period; this data can be obtained from relevant publicly available databases.

[0024] Based on the aforementioned DSM data, a 3D geographic model of the area to be measured is constructed, and a GNSS visible satellite simulation is overlaid on the model. For each ground pixel, the field of view of a simulated GNSS receiver is determined, and the average number of visible GNSS satellites within a 24-hour period is calculated. For each pixel, the proportion of its area obscured (i.e., obscured by ground features at low elevation angles) per unit time is calculated, and the GNSS obscuration probability value for that pixel is output, ranging from 0 to 1. An obscuration probability map is generated from the obscuration probability values ​​corresponding to all pixels in the entire image. The formula for calculating the obscuration probability value is: ;in, Let i represent a unit pixel, where i = 1, 2, ..., I, and I represents the total number of unit pixels. This represents the occlusion probability value. This represents the number of GNSS satellites occluded by the 3D model at time t at that pixel location; This represents the total number of GNSS satellites that are theoretically visible at time t, where t = 1, 2, ..., T. t is a simulation time slice within a 24-hour period and can be set according to actual needs. For example, if t is set to once every 10 minutes, then T = 144.

[0025] Historical DSM data is processed using a 3D geographic model to obtain a reflectance surface density map; historical meteorological data is processed using physical calculation methods to obtain an ionization fluctuation map; the surface slope and normal variation within the radius of each pixel (e.g., 10-20 meters) are analyzed using a 3D geographic model; if there are many steep structures (e.g., walls, cliffs, metal roofs, etc.) in the area, their surfaces can be considered as potential GNSS reflection sources; a reflectance surface density map is generated by statistically analyzing the number or area density of patches with tilt angles greater than a certain threshold (e.g., 30°) within a unit area, with each pixel value representing the corresponding multipath interference risk index.

[0026] Ionization disturbances cause GNSS signal propagation delays and path refraction, especially in high-altitude, tropical, and electromagnetically frequent areas. By analyzing historical meteorological data (such as ionospheric TEC, solar activity Kp index, regional temperature and humidity distribution, etc.), and combining the Klobuchar model or NeQuick model to simulate ionospheric delay, time-division statistics are performed and normalized for each pixel region to output a spatial map of the degree of ionospheric disturbance, i.e., an ionization fluctuation map. This map can be understood as a probability assessment of GNSS delay error in a certain region over a specific time period. The physical calculation method refers to establishing a mathematical expression model of the GNSS signal propagation process in the ionosphere based on known physical laws and parameter models, such as choosing the Klobuchar model for calculation; the formula for calculating the degree of ionospheric disturbance per unit pixel is: ,in, The TEC increment at time t compared to the previous time indicates a disturbance. This indicates the normalized degree of ionospheric perturbation per unit pixel. This represents the average TEC value of pixel p within period T.

[0027] By fusing image features from the occlusion probability map, the reflectance density map, and the ionization wave map, a simulated image control deployment map is obtained; The occlusion probability map, reflectance density map, and ionization fluctuation map are fused at the pixel level. The fusion method can be weighted linear combination, principal component analysis (PCA), or deep neural network feature fusion method. Each pixel has a comprehensive score in the fused map, which is used to represent the signal stability and measurement feasibility of the area as a location for GNSS point deployment. The fusion result is called a simulated GNSS deployment map, which visually shows the suitability of GNSS point deployment in different areas.

[0028] The initial particles of the control points are screened in the simulated control point layout map using a swarm optimization algorithm to obtain the preferred control point particles; the swarm optimization algorithm is the particle swarm optimization algorithm. On the simulated image control deployment map, a particle swarm optimization algorithm is executed to globally screen all generated initial particles of the image control points. The optimization objective function considers factors such as the comprehensive GNSS suitability score of each particle, the uniformity of spatial distribution among particles (such as maximum and minimum distances), and the representativeness of elevation changes. In multiple iterations, the particle swarm continuously adjusts the search path through an elite screening mechanism to find the optimal deployment subset in the solution space. Finally, the group of particles with the highest score, reasonable distribution, and minimal GNSS signal interference is retained as the preferred image control point particles for subsequent GNSS coordinate measurement and image selection.

[0029] Obtain the real-time coordinates of the pre-selected points for image control measurement; classify all pre-selected points into regions with stable reception and unstable reception. Once the preferred image control point particles are determined, these points are pushed to the field execution phase as pre-selected image control measurement points. Professional technicians conduct static or dynamic observations at the pre-selected point locations to obtain their three-dimensional coordinates. For each measurement point, in addition to acquiring its coordinates, GNSS reception quality information is also recorded, such as the number of satellites, signal-to-noise ratio (SNR), and position accuracy factor (DOP). This additional information provides a basis for subsequent area classification.

[0030] Once the on-site coordinates of all pre-selected points are obtained, the point set is classified according to environmental suitability. The occlusion probability value of the corresponding pixel for each pre-selected point is extracted from the actual GNSS occlusion probability map. This value reflects the degree to which the GNSS line of sight is obstructed at that point in space. According to the preset occlusion threshold (such as 0.4~0.6, which is manually set by professional technicians based on experience or mission accuracy requirements), the points are classified and judged: if the occlusion probability value is less than the occlusion threshold, the GNSS signal reception at that point is considered stable, and it is marked as a stable reception area point; otherwise, the point is considered to have obvious occlusion or signal unavailability risk, and it is marked as a non-stable reception area point.

[0031] Step S2: UAV image control measurement method; For the pre-selected points of the image control measurement in the stable receiving area, the standard UAV image control measurement method is used to measure and obtain the measurement data of the stable area. For pre-selected points for image control measurement in unstable areas, auxiliary measurement rules are established; measurements are then performed using the auxiliary measurement rules and standard UAV image control measurement methods to obtain measurement data for unstable areas.

[0032] The specific steps for conducting measurements using standard UAV image-controlled measurement methods include: The three-dimensional geographic coordinates of the pre-selected points for image control measurement within the stable receiving area are obtained, resulting in multiple three-dimensional measurement coordinates; These points are measured using RTK or PPK equipment to obtain high-precision three-dimensional geographic coordinates for each point. This coordinate information will be used for subsequent steps such as image matching, aerial triangulation, and model calibration, forming the geometric control basis of the survey area. Set standard flight parameters for the UAV; obtain multiple control point images from the UAV for selective barrage; After control point preparation is complete, standard aerial survey parameters for the UAV are established. Flight parameter settings include: forward overlap (e.g., ≥75%) to ensure sufficient overlap between adjacent images in the forward and backward directions; lateral overlap (e.g., ≥65%) to ensure lateral coverage between flight strips; flight altitude determined based on the desired ground resolution (GSD) and camera parameters, typically 80-150 meters; exposure interval matched to shutter speed to ensure sufficient sampling frequency without creating holes; and camera parameter configuration, such as setting a fixed focus mode, shutter speed, aperture value, etc., to ensure stable and consistent settings.

[0033] Multiple control point stab images and multiple measurement 3D coordinates are paired to define image coordinates for the control point stab images. N pre-selected control measurement points are randomly selected as coordinate verification points. The reprojection error of the control point stab images containing coordinate verification points is calculated to obtain the control measurement error factor. After image acquisition, all image regions containing pre-selected control points are identified in the acquired images, and a stab image set for each point is constructed. Using the 3D geographic coordinates of the point and the exterior orientation elements of the image, the geographic coordinates are mapped to image coordinates through projection transformation to mark the precise position of each control point on the image (i.e., "stab point"). At the same time, information such as the image number, shooting angle, and control point image coordinates of each image is recorded for subsequent aerial triangulation modeling.

[0034] To verify the spatial geometric accuracy of the entire measurement system, N points (e.g., 5-10%) are randomly selected from the pre-selected points of the image control measurement as coordinate verification points. These points participate in imaging but not in aerial triangulation calculations, and are only used for post-hoc accuracy evaluation. In the selected images, these verification points are back-projected from image coordinates to ground coordinates. The reprojection error in the model is calculated using the known coordinates. The mean and variance of the errors of all verification points constitute the error factor of this batch of image control measurements, which is used to determine whether model correction or data resampling is required. Specifically: the coordinates of the coordinate verification point are set as follows: , The coordinates are: For each coordinate verification point, its known 3D ground coordinates are back-projected into the image space to obtain the corresponding predicted image coordinates. ;use and Calculate reprojection error in the plane The formula is: ; Calculate all reprojection errors The mean error and variance can be used to further derive the root mean square error, which constitutes the error factor for this batch of image control measurements.

[0035] Aerial triangulation is performed on the selected images of ground control points with image coordinates to obtain aerial triangulation measurement results. The aerial triangulation results are then calibrated using ground control point measurement error factors to obtain adjusted aerial triangulation results. Aerial triangulation is performed on all selected ground control point images with image coordinates and their corresponding ground coordinates as input. This process includes feature point extraction, matching optimization, camera pose estimation, and spatial point inversion. The final output is the exterior orientation elements, sparse 3D point cloud, and solution residuals for each image. This step establishes a high-precision geometric relationship between the image and geospatial data, a crucial prerequisite for the generation of DSM, DOM, and DEM. The aerial triangulation results are then corrected using the reprojection error factors of the verification points. Correction strategies include, but are not limited to: global weighted adjustment based on the error field; elevation surface fitting correction for Z-direction deviation; and posterior least squares optimization using verification points as soft constraints to improve model consistency. This process outputs calibrated aerial triangulation measurement results, with both exterior and interior orientation elements and point cloud geometric accuracy reaching a high level of reliability.

[0036] Analyze the distribution characteristics of the image control measurement error factors in the plane and elevation directions to determine whether there is a systematic deviation. Then, construct an error calibration model, such as a parameter correction model based on least squares adjustment, or use a locally fitted elevation offset surface to adjust the point cloud coordinates and camera exterior orientation elements in the aerial triangulation results. Finally, recalculate the reprojection error of the calibrated model. If the error is lower than the set threshold, the current calibration result is accepted.

[0037] Based on the adjusted aerial triangulation measurement results, corresponding DOM, DSM and DEM are generated, and combined to obtain the measurement data of the stable region; The specific steps for establishing auxiliary measurement rules include: Step A1: For the pre-selected image control measurement points in the unstable receiving area, M auxiliary control points are automatically generated within a radius of r around the pre-selected image control measurement points; the auxiliary control points and the corresponding pre-selected image control measurement points are combined to obtain the pre-selected image control measurement point cluster. When a pre-selected image control point is identified as being located in an unstable GNSS reception area, a spatial radius r (e.g., 5-10 meters) is set around that point, and M auxiliary control points are automatically generated within this radius. The spatial positions of the generated points must meet certain geometric constraints, such as a maximum interior angle of no more than 135° and a minimum boundary distance of no less than 2 meters, to prevent the control point set from being too concentrated or collinear. The position of each auxiliary control point can be randomly generated through geometric configuration rules or determined based on three-dimensional terrain analysis to identify representative points. These auxiliary control points, together with the central pre-selected point, form a pre-selected image control point cluster.

[0038] Step A2: Based on the existing standard flight parameters, add flight path guidance rules. Specifically, randomly use different headings or shooting angles to acquire at least J images containing the pre-selected control point. Considering that there may be local obstruction, reflection, or delay in unstable GNSS signal areas, images under the standard flight path may not fully cover the geometric information around the control point. Therefore, flight path guidance rules are introduced based on the existing standard flight parameters. Specifically, during the automatic flight path generation process, an angle perturbation mechanism is introduced to randomly select additional flight paths with different headings (e.g., ±20° deviation from the main heading) and camera attitudes (e.g., ±15° pitch angle) to ensure that images containing the pre-selected point are acquired at different angles. Finally, each pre-selected point in an unstable area must be covered by at least J images from different perspectives (J is generally 3~5) to construct the control point selection image set.

[0039] Determine the fluctuation complexity of the receiving unstable region; establish mapping rules based on the fluctuation complexity and steps A1 and A2 to obtain auxiliary measurement rules; the specific content of the auxiliary measurement rules is as follows: The fluctuation complexity of each pre-selected point in an unstable region is quantitatively analyzed. The normalized occlusion probability value, multipath reflection density, and ionospheric disturbance index are used as the regional fluctuation score. When the regional fluctuation score is in the first scoring interval, it indicates that the measurement environment is relatively stable, and the standard measurement method can be used. When the regional fluctuation score is in the second scoring interval, only step A2 is triggered, that is, image redundancy is enhanced through multi-view flight. When the regional fluctuation score is in the third scoring interval, step A1 is triggered, and auxiliary control points are automatically deployed to enhance geometric stability. When the regional fluctuation score is in the fourth scoring interval, it indicates that the measurement environment is extremely complex, and steps A1 and A2 need to be triggered simultaneously to achieve robust control point deployment and multi-angle image coverage guarantee for highly interfered areas. The first, second, third, and fourth scoring intervals are set manually.

[0040] The specific steps for conducting measurements using auxiliary measurement rules and standard UAV image control measurement methods include: For receiving unstable regions, UAV measurements are performed using auxiliary measurement rules. In this embodiment, the cluster of pre-selected control points corresponding to the control point selection image set is obtained, taking the most unstable fluctuation (i.e., satisfying the fourth scoring interval) as an example. After the auxiliary measurement rules are executed, the pre-selected control points in the unstable region and multiple auxiliary control points automatically deployed around them have been obtained. For each pre-selected point, it is associated with the set of images that have been successfully identified and selected in the image to form a complete control point selection image set. Acquire real-time GNSS signals; construct a pre-selected 3D plane for image control measurement based on the pre-selected point cluster; perform calibration in the pre-selected 3D plane for image control measurement using the anti-occlusion interpolation method to obtain the occlusion error calibration factor; The receiving equipment collects the real-time coordinate information of the control points in each pre-selected point cluster, and combines it with observation data such as the number of surrounding satellites, DOP value, and signal-to-noise ratio (SNR) to identify changes in signal stability. Taking the main control point in the point cluster as the center, it combines the coordinates of its auxiliary points to form a local three-dimensional measurement reference plane, which is the image control measurement pre-selected three-dimensional plane. A direction-weighted projection model is constructed based on real-time GNSS signals and image sets of selected control points, and a reflection error calibration factor is output based on the direction-weighted projection model. To address the issue of missing or offset GNSS signals caused by obstruction from buildings, vegetation, etc., an anti-occlusion interpolation method is performed within a constructed 3D plane. Specifically, the method involves using GNSS measurement results from points in the cluster that are not obstructed (occlusion probability value < occlusion threshold) as the observation benchmark to estimate the spatial location of obstructed points through interpolation. Interpolation methods can include IDW (inverse distance weighted), bidirectional weighted plane fitting, or triangulation interpolation. The deviation between the theoretical coordinates obtained through interpolation and the actual GNSS coordinates of the obstructed points is the occlusion error calibration factor, which is used to compensate for measurement errors caused by obstruction.

[0041] Ionization characteristics of real-time GNSS signals are extracted to obtain real-time ionization interference characteristics; short-time compensation is performed based on real-time ionization interference characteristics to obtain ionization error calibration factors; To address GNSS signal errors caused by multipath reflections, a directional weighted projection model is constructed based on real-time GNSS signals (real-time GNSS ephemeris), reflector density maps, and image sets of selected ground control points. This model analyzes the angle between the GNSS signal arrival path and the normals of surrounding reflectors, estimating the signal reflection probability and its impact on the Z-direction coordinates. The model projects this error back to the three-dimensional plane according to its direction and performs a weighted average based on the reflector distribution, outputting a reflection error calibration factor for each point. This factor is used to adjust for systematic offsets caused by multipath effects, particularly vertical deviations.

[0042] By utilizing the ionospheric delay frequency correlation of GNSS dual-frequency carriers in real-time GNSS signals, the ionospheric delay on each satellite observation path can be directly calculated. This delay is then transformed into the total electron content in the vertical direction using a mapping function, yielding real-time ionization interference characteristics. Based on these characteristics, the ionization delay error of the signal under a specific path within that time period is predicted. A fitting analysis is then performed using the GNSS observation residuals to obtain a short-time error curve within a time window. Finally, the ionization error calibration factor for each measurement point is calculated to compensate for measurement errors caused by ionospheric refraction or delay in the signal propagation path.

[0043] The occlusion error calibration factor, reflection error calibration factor, ionization error calibration factor, and image control measurement error factor are fused to obtain the fused error calibration factor. Based on the fused error calibration factor, subsequent operations are performed on the aerial triangulation measurement results to obtain measurement data for unstable regions.

[0044] After obtaining the occlusion error calibration factor, reflection error calibration factor, and ionization error calibration factor, the three types of errors are uniformly modeled through a weighted fusion method. The fusion method can be based on the confidence of the error source (such as GNSS signal-to-noise ratio and occlusion probability), error directionality (whether it is dominated by the Z direction), and scene weight (such as the weight corresponding to mountainous scenes) to perform multi-factor weighting, and finally output the fusion error calibration factor for each control point. The fusion error calibration factor is applied to the aerial triangulation (ALT) results to adjust the geographic coordinates of the image control points. This adjustment can be achieved by directly offsetting the control point coordinates, updating the control network constraints, or introducing an error weight model during the adjustment process. The corrected ALT results will have higher geometric consistency and consistency with actual observations, providing a reliable control basis for the subsequent generation of high-precision DOM, DSM, and DEM.

[0045] It should be noted that for the area in the second scoring interval, the GNSS signal is identified as having some instability but still possessing a measurement basis. In this case, only step A2 in the auxiliary measurement rules is triggered, which guides the UAV to perform a multi-view image acquisition task. Specifically, this includes setting different headings and shooting angles (such as ±20° tilt, lateral yaw, etc.) to ensure that each pre-selected ground control measurement point is successfully identified and selected in no less than J images, forming a redundant image set. Subsequently, aerial triangulation calculations are performed, and ground control measurement error factors are introduced for preliminary calibration. Since no auxiliary control points are deployed, the GNSS signal in this area is not further corrected for occlusion, reflection, and ionization errors. However, due to sufficient image viewing angles and enhanced aerial triangulation beam intersection conditions, the calculation accuracy is improved. Finally, DOM, DSM, and DEM are generated from the enhanced aerial triangulation results, forming measurement data for the transition area with moderate accuracy.

[0046] For the third scoring interval, the GNSS signal is identified as relatively unstable, with severe occlusion or reflection. At this point, step A1 in the auxiliary measurement rules is triggered, which automatically deploys M auxiliary control points within a radius r around the pre-selected image control measurement points, forming a local control point cluster to enhance spatial geometric constraints. Images are acquired along the standard flight path, and image control point selection and coordinate pairing are performed. Subsequently, real-time GNSS data is acquired, and a local 3D measurement plane is constructed using the clustered control points. Occlusion error correction (such as IDW interpolation) and reflection error correction under the direction-weighted projection model are performed, but ionization compensation is not performed. By fusing the occlusion and reflection error calibration factors with the image control measurement error factors, multi-source error compensation and correction of the aerial triangulation results are completed, generating higher-precision control-enhanced measurement data and ensuring the stability and reliability of terrain modeling in complex areas.

[0047] In this embodiment, for example, in a high-speed railway site selection project in a mountainous area, the engineering corridor is approximately 30 kilometers long. The survey area has complex terrain, significant elevation variations, and covers various GNSS signal environments, including open valleys, forests, canyons, and tunnels, requiring high-precision 1:500 digital mapping. A UAV-assisted image control (ADC) survey method was employed, and ADC placement was optimized based on the GNSS signal environment. The survey area was divided into two categories: stable areas, such as open valleys, exposed rock surfaces, and fields, where GNSS signals are stable; and unstable areas, such as canyons, steep slopes, forests, or tunnel exits, where GNSS obstruction / reflection is strong. A total of 162 ADCs were deployed, of which 102 (63%) were in stable areas and 60 (37%) were in unstable areas. Before the deployment optimization, the average spacing between control points was about 420 meters. After the optimized deployment of control points, the average spacing between control points was reduced from the original 420 meters to 250 meters, and the control point density was increased by 68% (specifically: (420 / 250-1)*100%=60%), which significantly improved the computational redundancy and spatial balance.

[0048] High-precision DOM (Digital Orthophoto Map), DSM (Digital Surface Model), and DEM (Digital Elevation Model) are generated in the stable GNSS receiving area to form the stable area measurement data. With a flight altitude of 120 meters, a camera focal length of 35mm, and a pixel size of 2.4μm, the calculated ground resolution (GSD) is approximately 1.5cm / pixel. Combining image measurements of actual selected sniping points with the coordinates of 3D control points, the image coordinates are inversely calculated using the collinearity equation, and the average reprojection error of all participating aerial triangulation verification points is calculated. There are 12 verification points in this area. The average reprojection error of the verification points, e, is calculated using the reprojection error formula. 平均=1.08 pixels, the worst is 1.21 pixels, and the best is 0.89 pixels, so the reprojection error is controlled within 1.1 pixels; the elevation accuracy is obtained by comparing the three-dimensional coordinates of the verification point with the aerial triangulation inversion value. The standard deviation of the elevation residual of all points is less than ±8cm, which meets the accuracy requirements of the 1:500 digital mapping specification.

[0049] In unstable regions, auxiliary measurement rules are triggered: by deploying redundant auxiliary control points, guiding UAVs to perform multi-angle flight missions, and acquiring rich image data, the robustness of the solution is enhanced. In unstable regions, through the auxiliary measurement mechanism, 4 to 6 redundant auxiliary control points are automatically deployed around each main control point. The UAVs use multi-angle (vertical, oblique, backlight) flight missions to acquire images. In traditional methods, the number of images covered per point is about 4, while the average number of images covered per control point cluster is increased to 9, which is 2.25 times that of traditional methods.

[0050] Subsequently, ionization characteristics, occlusion index, and reflector density were extracted from real-time GNSS observation data. Multi-source error compensation was then applied to the aerial triangulation results using a fusion error calibration factor, generating calibrated DOM, DSM, and DEM datasets to form a measurement dataset for the unstable region. Modeling and analysis were performed on the occlusion index (average 0.58), reflector density map (0.32), and ionization interference characteristics. After applying the fusion error calibration model, statistical analysis of the reprojection errors at 32 verification points in the unstable region revealed an average error of 2.71 pixels in the original aerial triangulation results. After applying the fusion error calibration factor (including weighted models for occlusion, ionization, and reflection), the average error decreased to 1.31 pixels. The error was reduced by more than 51% ((2.71-1.31) / 2.71*100%=51.66%). Independent accuracy verification was performed on the planar error of the DOM and the elevation error of the DEM in the unstable area (a total of 21 measured points). The RMSE was calculated as follows: the original mean of the DOM planar error was ±23cm, and the optimized mean was ±9cm; the original mean of the DEM elevation error was ±28cm, and the optimized mean was ±11cm. The accuracy improvements were 60.87% ((23-9) / 23*100%=60.87%) and 60.71% ((28-11) / 28*100%=60.71%), respectively, which greatly improved the mapping quality of the unstable area.

[0051] The practical significance of this zonal measurement mechanism is that: data from stable areas serves as a global geometric benchmark, ensuring the overall control accuracy of the survey area; data from unstable areas achieves local quality improvement through error compensation, solving geometric distortion problems caused by occlusion, multipath, and ionospheric disturbances, thereby ensuring that the entire survey area has a unified, high-precision, and interference-resistant geographic information foundation, providing reliable support for subsequent applications such as orthophoto mapping, 3D reconstruction, and terrain analysis.

[0052] 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 unmanned aerial vehicle (UAV)-assisted image control measurement, characterized in that, include: Step S1: Image control point deployment method; Acquire historical aerial photography data of the area to be measured; use the historical aerial photography data to generate initial particles for image control points; Obtain the occlusion probability map of GNSS signals in the area to be measured; based on the occlusion probability map and the initial particles of the control points, use a swarm optimization algorithm to analyze and obtain the preferred control point particles; use the preferred control point particles as the pre-selected points for control point measurement; Obtain the real-time coordinates of the pre-selected points for image control measurement; classify all pre-selected points into regions with stable reception and unstable reception. Step S2: UAV image control measurement method; For the pre-selected points of the image control measurement in the stable receiving area, the standard UAV image control measurement method is used to measure and obtain the measurement data of the stable area. For pre-selected points for image control measurement in unstable areas, auxiliary measurement rules are established; measurements are then performed using the auxiliary measurement rules and standard UAV image control measurement methods to obtain measurement data for unstable areas.

2. The UAV-assisted image control measurement method according to claim 1, characterized in that, The specific steps for generating initial particles for image control points using historical aerial photography data include: the historical aerial photography data contains image data of the area to be measured and geographic reference information of the measurement area; The neighborhood image of the area to be measured is obtained based on the geographic reference information of the measurement area; a multi-scale texture histogram is generated based on the neighborhood image of the area to be measured; and a pre-trained YOLO model is used to identify the image data of the area to be measured and the multi-scale texture histogram to obtain the location of candidate control points. Each candidate control point position is marked as a candidate control point particle; coordinate matching is performed in multiple image data of the area to be measured containing candidate control point particles; if the candidate control point particle satisfies at least three viewpoints, the candidate control point particle is retained as the initial control point particle; otherwise, the candidate control point particle is discarded. By iterating through all candidate image control point particles, the final image control point initial particles are obtained.

3. The UAV-assisted image control measurement method according to claim 2, characterized in that, The specific steps for analyzing the occlusion probability map and initial particles of ground control points using a population optimization algorithm include: acquiring GNSS signals, historical DSM data, and historical meteorological data within the area to be measured; simulating the GNSS signals using a three-dimensional geographic model, outputting the occlusion probability value of all unit pixels within the area to be measured in unit pixels, and generating an occlusion probability map. Historical DSM data were processed using a 3D geographic model to obtain a reflectance density map; historical meteorological data were processed using physical calculation methods to obtain an ionization wave map. By fusing image features from the occlusion probability map, the reflectance density map, and the ionization wave map, a simulated image control deployment map is obtained; The initial particles of the control points are screened in the simulated control point layout map using a swarm optimization algorithm to obtain the preferred control point particles.

4. The UAV-assisted image control measurement method according to claim 3, characterized in that, The specific steps for measuring using standard UAV image control measurement methods include: acquiring the three-dimensional geographic coordinates of pre-selected points within the stable receiving area to obtain multiple measurement three-dimensional coordinates; Set standard flight parameters for the UAV; obtain multiple control point images from the UAV for selective barrage; Multiple control point puncture images and multiple measurement 3D coordinates are paired to define the image coordinates of the control point puncture images; N pre-selected control measurement points are randomly selected as coordinate verification points; the reprojection error of the control point puncture images containing coordinate verification points is calculated to obtain the control measurement error factor; Aerial triangulation is performed on the selected images of the control points with image coordinates to obtain the aerial triangulation measurement results; the aerial triangulation measurement results are calibrated using the control point measurement error factor to obtain the adjusted aerial triangulation measurement results; Based on the adjusted aerial triangulation results, corresponding DOM, DSM, and DEM are generated and combined to obtain stable region measurement data.

5. The UAV-assisted image control measurement method according to claim 4, characterized in that, The specific steps for establishing auxiliary measurement rules include: Step A1: For the pre-selected image control measurement points in the unstable receiving area, M auxiliary control points are automatically generated within a radius of r around the pre-selected image control measurement points; the auxiliary control points and the corresponding pre-selected image control measurement points are combined to obtain a cluster of pre-selected image control measurement points; Step A2: Based on the original standard flight parameters, add flight path guidance rules, specifically by randomly using different headings or shooting angles to obtain no less than J images containing pre-selected points for image control measurement; Determine the fluctuation complexity of the receiving unstable region; based on the fluctuation complexity, establish the mapping rules with steps A1 and A2 to obtain the auxiliary measurement rules.

6. The UAV-assisted image control measurement method according to claim 5, characterized in that, The specific steps for measurement using auxiliary measurement rules and standard UAV image control measurement methods include: using auxiliary measurement rules to perform UAV measurement in areas with unstable reception, and obtaining the image control point selection image set corresponding to the image control measurement pre-selected point cluster; Acquire real-time GNSS signals; construct a pre-selected 3D plane for image control measurement based on the pre-selected point cluster; perform calibration in the pre-selected 3D plane for image control measurement using the anti-occlusion interpolation method to obtain the occlusion error calibration factor; A direction-weighted projection model is constructed based on real-time GNSS signals and image sets of selected control points, and a reflection error calibration factor is output based on the direction-weighted projection model. Ionization characteristics of real-time GNSS signals are extracted to obtain real-time ionization interference characteristics; short-time compensation is performed based on real-time ionization interference characteristics to obtain ionization error calibration factors; The occlusion error calibration factor, reflection error calibration factor, ionization error calibration factor, and image control measurement error factor are fused to obtain the fused error calibration factor. Based on the fused error calibration factor, subsequent operations are performed on the aerial triangulation measurement results to obtain measurement data for unstable regions.

7. The UAV-assisted image control measurement method according to claim 6, characterized in that, The swarm optimization algorithm is the particle swarm optimization algorithm.

Images