Unmanned aerial vehicle positioning and labeling method based on total station and RTK GNSS

Abstract

The application provides a kind of unmanned plane positioning and marking method based on total station and RTK GNSS, it is related to unmanned plane positioning technical field, to solve the defect of low accuracy of positioning unmanned plane. Including: time alignment processing of time log based on RTK GNSS and TS;Based on the position information and the line of sight direction of TS, satellite ephemeris and line of sight blocking parameter of RTK GNSS, remove NLOS data in the first observation data of RTK GNSS to the unmanned plane, obtain the first observation data after optimization;Convert the first observation data after optimization to target coordinate system, and convert the second observation data of TS to the unmanned plane to target coordinate system;In target coordinate system, based on dynamic weight, the first observation data and the second observation data are fused to obtain fusion data;Based on the out-of-order or delay data of unmanned plane, the fusion data is smoothed to determine the real-time positioning data of unmanned plane.

Description

Technical Field

[0001] This invention relates to the field of UAV positioning technology, and in particular to a UAV positioning and labeling method based on total station and RTK GNSS. Background Technology

[0002] When conducting automated inspections, logistics delivery, and 3D mapping in urban low-altitude environments, drones must first possess high positioning accuracy to achieve autonomous flight. However, the densely built environment, multi-source electromagnetic interference, and complex weather conditions together constitute a typical urban scenario characterized by "strong obstruction, strong reflection, and strong dynamics," posing systemic challenges to the positioning system in terms of accuracy, continuity, and robustness.

[0003] Therefore, in complex urban environments with obstructions, abnormal observation data collected due to obstacle blockage can cause filter bias, slow convergence, or even divergence. This results in low accuracy for drone positioning in urban environments. Summary of the Invention

[0004] This invention provides a method for UAV positioning and labeling based on total station and RTK GNSS, which solves the problem of low accuracy in UAV positioning in urban environments in the prior art, and improves the accuracy of UAV positioning in urban environments.

[0005] This invention provides a method for UAV positioning and labeling based on total station and RTK GNSS, comprising the following steps.

[0006] Based on the time synchronization logs of the real-time dynamic global navigation satellite system RTK GNSS and the total station TS, time alignment processing is performed on RTK GNSS and TS; Based on the position information and line-of-sight direction of TS in time-aligned state, and the satellite ephemeris and line-of-sight occlusion parameters of RTK GNSS in time-aligned state, non-line-of-sight propagation (NLOS) data in the first observation data of UAV by RTK GNSS are removed to obtain optimized first observation data; The optimized first observation data is transformed from the first coordinate system to the target coordinate system, and the second observation data of the UAV by the TS is transformed from the second coordinate system to the target coordinate system. The first coordinate system is the coordinate system corresponding to RTK GNSS, and the second coordinate system is the coordinate system corresponding to TS. In the target coordinate system, the optimized first and second observation data are fused based on dynamic weights to obtain fused data. The dynamic weights are determined based on the data quality of the optimized first and second observation data. The continuity parameter of the covariance of the fused data satisfies the preset continuity parameter. The fused data is smoothed based on the disordered or delayed data of the UAV to determine the real-time positioning data of the UAV. Based on the real-time positioning data of the drone, the drone's location is marked in the images captured by the drone's onboard camera.

[0007] According to the present invention, a UAV positioning and labeling method based on total station and RTK GNSS is provided. Based on the time synchronization logs of the real-time dynamic global navigation satellite system RTK GNSS and the total station time synchronization logs, time alignment processing of RTK GNSS and TS is performed, including: Based on the time logs of RTK GNSS, the sampling time drift and first time uncertainty of RTK GNSS are determined; Based on the time log of TS, the equivalent center time correction and second time uncertainty of the measurement window of TS are determined; Time alignment is performed on RTK GNSS and TS based on sampling time drift, first time uncertainty, equivalent center time correction, and second time uncertainty.

[0008] According to the present invention, a UAV positioning and annotation method based on total station and RTK GNSS is provided. Based on the position information and line-of-sight direction of the time-aligned TS (Tracking Station) and the satellite ephemeris and line-of-sight occlusion parameters of the RTK GNSS in the time-aligned state, non-line-of-sight (NLOS) propagation data from the first observation data of the UAV by the RTK GNSS is removed to obtain optimized first observation data, including: Based on location information and line-of-sight direction, as well as satellite ephemeris and line-of-sight occlusion parameters, sub-observation data for each satellite in the first observation data are determined. The sub-observation data includes at least one of the following: residual, residual change rate, satellite carrier-to-noise ratio, and elevation angle threshold. The optimized first observation data is obtained by removing NLOS data from the first observation data based on the target parameter set. The target parameter set includes at least one of the following: location information and line-of-sight occlusion parameters, residuals and residual change rate, satellite carrier-to-noise ratio and elevation angle threshold, line-of-sight direction and UAV velocity direction.

[0009] According to the UAV positioning and labeling method based on total station and RTK GNSS provided by the present invention, the dynamic weights are determined in the following way: The optimized data quality of the first observation data and the data quality of the second observation data are input into a preset function to obtain the initial weights. The initial weights are low-pass filtered to obtain dynamic weights, and the rate of change of the dynamic weights is less than the preset rate of change.

[0010] According to the present invention, a method for UAV positioning and labeling based on total station and RTK GNSS is provided. This method smooths the fused data based on out-of-order or delayed data from the UAV to determine the UAV's real-time positioning data, including: Based on a preset sliding window depth and a preset insertion frequency, out-of-order or delayed data is inserted into the fused data to obtain the fused data after data insertion; The fused data after data insertion is smoothed to determine the real-time positioning data of the UAV.

[0011] According to the present invention, a method for UAV positioning and labeling based on total station and RTK GNSS is provided, the method further includes: During the fusion of the optimized first and second observation data, the fusion parameters are monitored. The fusion parameters include at least one of the following: the condition number of the information matrix, the normalized innovation square (NIS), and the normalized estimation error square (NEES). If the condition number of the information matrix is ​​greater than the preset condition number, and / or the consistency parameter between NIS and NEES does not match the preset consistency parameter, the real-time positioning data of the UAV is determined based on the first observation data or the second observation data.

[0012] According to the present invention, a method for UAV positioning and labeling based on total station and RTK GNSS is provided, which labels the UAV's position in images captured by the UAV's onboard camera based on the UAV's real-time positioning data, including: Determine the first positioning coordinates of the real-time positioning data in the world coordinate system; Based on the camera extrinsic parameters, the first positioning coordinates are transformed into the camera coordinate system to obtain the second positioning coordinates of the real-time positioning data in the camera coordinate system; Based on camera intrinsic parameters and distortion parameters, pinhole projection and distortion processing are performed on the second positioning coordinates to obtain a set of pixels, which is used to construct the UAV. Based on the set of pixels, a bounding box for the drone is generated. The bounding box is used to mark the location of the drone in the image.

[0013] The present invention also provides a UAV positioning and labeling device based on a total station and RTK GNSS, comprising the following modules: a processing module, an optimization module, a conversion module, a fusion module, and a determination module; The processing module is used to perform time alignment processing on RTK GNSS and TS based on the time logs of the real-time dynamic global navigation satellite system RTK GNSS and the total station TS. The optimization module is used to remove non-line-of-sight (NLOS) data from the first observation data of the UAV by RTK GNSS based on the position information and line-of-sight direction of TS in time-aligned state, as well as the satellite ephemeris and line-of-sight occlusion parameters of RTK GNSS in time-aligned state, so as to obtain optimized first observation data. The conversion module is used to convert the optimized first observation data from the first coordinate system to the target coordinate system, and to convert the second observation data of the UAV by the TS from the second coordinate system to the target coordinate system. The first coordinate system is the coordinate system corresponding to RTK GNSS, and the second coordinate system is the coordinate system corresponding to TS. The fusion module is used to fuse the optimized first observation data and the second observation data in the target coordinate system based on dynamic weights to obtain fused data. The dynamic weights are determined based on the data quality of the optimized first observation data and the data quality of the second observation data. The continuity parameter of the covariance of the fused data satisfies the preset continuity parameter. The determination module is used to smooth the fused data based on the out-of-order or delayed data of the UAV and determine the real-time positioning data of the UAV. The processing module is also used to mark the drone's location in images captured by the drone's onboard camera based on the drone's real-time positioning data.

[0014] According to the present invention, a UAV positioning and labeling device based on a total station and RTK GNSS, the processing module is specifically used for: Based on the time logs of RTK GNSS, the sampling time drift and first time uncertainty of RTK GNSS are determined; Based on the time log of TS, the equivalent center time correction and second time uncertainty of the measurement window of TS are determined; Time alignment is performed on RTK GNSS and TS based on sampling time drift, first time uncertainty, equivalent center time correction, and second time uncertainty.

[0015] According to the present invention, an UAV positioning and labeling device based on a total station and RTK GNSS is provided, wherein the optimization module is specifically used for: Based on location information and line-of-sight direction, as well as satellite ephemeris and line-of-sight occlusion parameters, sub-observation data for each satellite in the first observation data are determined. The sub-observation data includes at least one of the following: residual, residual change rate, satellite carrier-to-noise ratio, and elevation angle threshold. The optimized first observation data is obtained by removing NLOS data from the first observation data based on the target parameter set. The target parameter set includes at least one of the following: location information and line-of-sight occlusion parameters, residuals and residual change rate, satellite carrier-to-noise ratio and elevation angle threshold, line-of-sight direction and UAV velocity direction.

[0016] According to the present invention, a UAV positioning and labeling device based on a total station and RTK GNSS, the processing module is further used for: The optimized data quality of the first observation data and the data quality of the second observation data are input into a preset function to obtain the initial weights. The initial weights are low-pass filtered to obtain dynamic weights, and the rate of change of the dynamic weights is less than the preset rate of change.

[0017] According to the present invention, a UAV positioning and marking device based on a total station and RTK GNSS is provided, wherein the determining module is specifically used for: Based on a preset sliding window depth and a preset insertion frequency, out-of-order or delayed data is inserted into the fused data to obtain the fused data after data insertion; The fused data after data insertion is smoothed to determine the real-time positioning data of the UAV.

[0018] According to the present invention, a UAV positioning and labeling device based on a total station and RTK GNSS, the processing module is further used for: During the fusion of the optimized first and second observation data, the fusion parameters are monitored. The fusion parameters include at least one of the following: the condition number of the information matrix, the normalized innovation square (NIS), and the normalized estimation error square (NEES). If the condition number of the information matrix is ​​greater than the preset condition number, and / or the consistency parameter between NIS and NEES does not match the preset consistency parameter, the real-time positioning data of the UAV is determined based on the first observation data or the second observation data.

[0019] According to the present invention, a UAV positioning and labeling device based on a total station and RTK GNSS, the processing module is further used for: Determine the first positioning coordinates of the real-time positioning data in the world coordinate system; Based on the camera extrinsic parameters, the first positioning coordinates are transformed into the camera coordinate system to obtain the second positioning coordinates of the real-time positioning data in the camera coordinate system; Based on camera intrinsic parameters and distortion parameters, pinhole projection and distortion processing are performed on the second positioning coordinates to obtain a set of pixels, which is used to construct the UAV. Based on the set of pixels, a bounding box for the drone is generated. The bounding box is used to mark the location of the drone in the image.

[0020] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-described methods for UAV positioning and labeling based on a total station and RTKGNSS.

[0021] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described UAV positioning and labeling methods based on total station and RTK GNSS.

[0022] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements any of the above-described methods for UAV positioning and labeling based on total station and RTK GNSS.

[0023] The UAV positioning and annotation method based on total station and RTK GNSS provided by this invention first requires time alignment processing of RTK GNSS and TS based on the time logs of the real-time dynamic global navigation satellite system RTK GNSS and the total station TS. Then, in the time-aligned state, based on the position information and line-of-sight direction of TS, as well as the satellite ephemeris and line-of-sight occlusion parameters of RTK GNSS, non-line-of-sight (NLOS) propagation data in the first observation data of the UAV from RTK GNSS is removed, resulting in optimized first observation data. Further, the optimized first observation data is transformed from the first coordinate system corresponding to RTK GNSS to the target coordinate system, and the second observation data of TS from the UAV is transformed from the second coordinate system corresponding to TS to the target coordinate system. In the target coordinate system, the optimized first and second observation data are fused based on dynamic weights to obtain fused data. Finally, the fused data is smoothed based on the out-of-order or delayed data of the UAV to determine the real-time positioning data of the UAV. Thus, this application achieves time consistency between RTK GNSS and TS by performing time alignment processing. Then, the RTK GNSS and TS observation data are converted to the same target coordinate system and fused to accurately determine the UAV's real-time positioning data. This improves the accuracy of UAV positioning in urban environments. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0025] Figure 1 This is one of the flowcharts illustrating the UAV positioning and annotation method based on total station and RTK GNSS provided by this invention.

[0026] Figure 2 This is the second flowchart of the UAV positioning and annotation method based on total station and RTK GNSS provided by the present invention.

[0027] Figure 3 This is a schematic diagram of the mapping relationship of the dual time scales on the time axis provided by the present invention.

[0028] Figure 4 This is the third flowchart of the UAV positioning and annotation method based on total station and RTK GNSS provided by this invention.

[0029] Figure 5 This is the fourth flowchart of the UAV positioning and annotation method based on total station and RTK GNSS provided by this invention.

[0030] Figure 6 This is the fifth flowchart illustrating the UAV positioning and annotation method based on total station and RTK GNSS provided by this invention.

[0031] Figure 7 This is a schematic diagram of data insertion and smoothing provided by the present invention.

[0032] Figure 8 This is the sixth flowchart of the UAV positioning and annotation method based on total station and RTK GNSS provided by this invention.

[0033] Figure 9 This is the seventh flowchart of the UAV positioning and annotation method based on total station and RTK GNSS provided by the present invention.

[0034] Figure 10 This is a schematic diagram of the UAV positioning and marking device based on total station and RTK GNSS provided by the present invention.

[0035] Figure 11 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0037] As a primary airborne positioning method, the Global Navigation Satellite System (GNSS) can reliably provide centimeter-level solutions in open areas, but its performance degrades significantly in urban areas. The sharp reduction in visible satellites and geometric degradation caused by high-rise buildings in cities increases the reliance on single or a small number of satellites for solution processing. Strong reflections from building surfaces trigger multipath effects, directly contaminating carrier phase observations and leading to frequent failures in integer ambiguity resolution (AR). The solution state jumps between fixed solutions (centimeter-level) and floating-point solutions (sub-meter-level), and even exhibits persistent position drift. Furthermore, under meteorological conditions such as cloudy skies, thick clouds, rain, and snowfall, L-band signals are attenuated, leading to a decrease in the carrier-to-noise density ratio (C / N0) observed by the receiver. This makes the already vulnerable signal at the edge of obstruction more prone to loss of lock, resulting in a deterioration of the dilution of precision (DOP) and inducing cycle slips, further compromising the stability of real-time kinematic (RTK) solutions. In more extreme scenarios (such as under bridges, semi-enclosed overpasses, or tunnel entrances), GNSS signals may be momentarily interrupted, forcing the system to rely on the inertial measurement unit (IMU) for short-term recursion. However, inertial drift can rapidly accumulate to meter-level errors within seconds, disrupting trajectory continuity and compressing the mission safety margin.

[0038] As a complementary ground-based positioning method, robotic total stations (TS) can provide millimeter- to sub-millimeter-level relative coordinate and velocity information under unobstructed line-of-sight conditions, significantly enhancing GNSS. However, their strict dependence on line-of-sight (LOS) makes them highly susceptible to occlusion and loss of lock-on in dynamic urban environments. Occlusion sources include static structures such as buildings, trees, and billboards, as well as dynamic targets such as vehicles and pedestrians. Furthermore, TSs are typically deployed at a single point, limiting their coverage and viewing angle, and blind spots are unavoidable. Their electromechanical and communication cycles limit the update frequency (e.g., 1–10 Hz), making it difficult to fully match the dynamic bandwidth of high-speed mobile platforms. In wide-area and multi-target application scenarios, deployment and maintenance costs rise rapidly, and system scalability is limited.

[0039] Therefore, fusing the wide-area coverage characteristics of GNSS with the local high-precision characteristics of TS is an inevitable path to achieve robust positioning across all scenarios, weather conditions, and attitudes. However, in practice, common loosely coupled or tightly coupled implementations often directly incorporate both types of observations into the Extended Kalman Filter (EKF), revealing key systemic bottlenecks during engineering deployment. Firstly, there is inconsistency in cross-sensor timescales: TS observations are affected by the "request-execution-response-dead zone" measurement process, resulting in Equivalent Center Time (ECT) deviation; the 1PPS (hardware pulse), network, or PTP (Precise Time Protocol) synchronization links on the GNSS side introduce sampling jitter and slow drift. Without unified modeling and online calibration of time delay and drift, the fused solution will produce centimeter-level systemic biases and exhibit non-stationary errors that vary with dynamic operating conditions and observation rhythms. Secondly, there is a lack of robustness in identifying non-line-of-sight (NLOS) and multipath observations: relying solely on GNSS intrinsic quality indicators such as CN0 and DOP for gating makes it difficult to reliably remove contaminated observations in complex occlusion environments; TS information, which can provide high-precision geometric priors and consistency constraints, is often not fully used for exogenous consistency checks and NLOS discrimination of GNSS observations, resulting in filter bias, slow convergence, or even divergence when abnormal data is injected.

[0040] In summary, a single positioning source cannot simultaneously meet the requirements of coverage, accuracy, and continuity in real-world urban scenarios, while simple observation stitching cannot resolve the fundamental challenges of cross-sensor spatiotemporal alignment and insufficient environmental perception. A fusion framework for engineering applications requires system-level redesign in areas such as time-scale unification and latency compensation, observation consistency modeling and quality assessment, robust NLOS recognition and multipath suppression, and enhanced observability of tightly coupled IMUs. Only then can it stably support centimeter-level autonomous positioning of UAVs in complex urban low-altitude environments. Furthermore, to support the training and evaluation of computer vision models, a key requirement in practical engineering is how to convert the aforementioned centimeter-level positioning results into pixel-level automatic annotations (including YOLO formats) without increasing manual annotation costs, ensuring consistency and auditability across time and space, and forming a data closed loop.

[0041] Therefore, this application focuses on the generation of centimeter-level location truth and continuous operation assurance for unmanned aerial vehicles (UAVs) in complex urban scenarios. The core idea is to use time reference (TS) and carrier RTK GNSS as "two types of metrological anchors." Before observations are fused, the time reference, reference frame, installation extrinsic parameters, and uncertainty characterization are unified. A geometric evidence-first gating strategy suppresses NLOS and multipath propagation. Covariance consistency handover is implemented in the information domain to achieve a smooth switch of multi-source dominance and auditable traceability. To clarify the scope and methodological boundaries, this invention does not use end-to-end deep neural networks as the main fusion engine, does not introduce visual sensing-related collaborative timing, rolling shutter modeling, and PnP / BA visual geometry solution processes, and does not rely on black-box AI adjustment modules (such as learning-driven robust regression or adaptive weighting). Its robustness comes from analytical methods such as geometric mapping of time errors, geometric gating, and information domain regularization. Based on this, the present invention further projects the prior three-dimensional position and shape of the UAV in the world coordinate system onto the pixel plane through camera extrinsic and intrinsic parameters to generate ground truth labels in multiple formats, including YOLO, for target detection model training and evaluation.

[0042] The following is combined with Figures 1 to 11 This invention describes the UAV positioning and annotation method based on total station and RTK GNSS provided by the present invention.

[0043] Figure 1 This is one of the flowcharts illustrating the UAV positioning and annotation method based on total station and RTK GNSS provided by this invention, such as... Figure 1 As shown, the method includes the following: Step 101: Based on the time synchronization logs of the Real-Time Dynamic Global Navigation Satellite System (RTK GNSS) and the total station (TS), perform time alignment processing on RTK GNSS and TS.

[0044] Step 102: Based on the position information and line-of-sight direction of the TS in time-aligned state, and the satellite ephemeris and line-of-sight occlusion parameters of the RTK GNSS in time-aligned state, remove the non-line-of-sight propagation (NLOS) data from the first observation data of the UAV by the RTK GNSS to obtain the optimized first observation data.

[0045] Step 103: Transform the optimized first observation data from the first coordinate system to the target coordinate system, and transform the TS second observation data of the UAV from the second coordinate system to the target coordinate system.

[0046] The first coordinate system is the coordinate system corresponding to RTK GNSS, and the second coordinate system is the coordinate system corresponding to TS.

[0047] Step 104: In the target coordinate system, the optimized first and second observation data are fused based on dynamic weights to obtain fused data.

[0048] The dynamic weights are determined based on the data quality of the optimized first and second observation data, and the continuity parameter of the covariance of the fused data satisfies the preset continuity parameter.

[0049] Step 105: Smooth the fused data based on the out-of-order or delayed data of the UAV to determine the real-time positioning data of the UAV.

[0050] Step 106: Based on the real-time positioning data of the UAV, mark the location of the UAV in the images captured by the UAV's onboard camera.

[0051] In one possible implementation, during the time alignment process between RTK GNSS and TS, a unified time reference needs to be established. This is achieved by recording the 1PPS / ToD (time message) / PTP time synchronization logs between the RTK GNSS and the total station, and then analyzing and estimating the equivalent center time correction and time uncertainty of the total station's measurement window. ), and GNSS sampling time drift and time uncertainty ( ), and map both to location-level equivalent noise respectively.

[0052] Furthermore, based on the instantaneous position and line-of-sight direction of the total station at a unified reference time, combined with satellite ephemeris and obstruction visibility, the geometric consistency and residual consistency of GNSS satellite observations are judged on a satellite-by-satellite basis, and NLOS observations are identified, weighted, or eliminated.

[0053] Then, the GNSS solution is aligned from the ITRF / WGS84 reference frame (the global reference frame in which the GNSS is located) to the ENU (East-North-Up) coordinate system through a seven-parameter transformation, and the uncertainty of this transformation is explicitly propagated to the position layer. The total station polar coordinate observations are then combined with the station pose uncertainty through Jacobi mapping and incorporated into the ENU.

[0054] The reference frame alignment employs a seven-parameter (Bursa–Wolf) model, with parameter covariance... Uncertainty is explicitly propagated to the location layer to avoid systematic biases introduced by frame inconsistencies.

[0055] In one possible implementation, the uncertainties of the lever arm and mounting angle from the total station prism APC and GNSS antenna APC to the UAV's center of mass are uniformly propagated to the center of mass position covariance, thus achieving end-to-end error budgeting.

[0056] Thus, by emphasizing the metrological principle of "caliber unification preceding fusion," GNSS global frameworks (such as ITRF) map to the local ENU through seven-parameter alignment, and also incorporate the covariance of the aligned solution. The position covariance is explicitly propagated via Jacobi. The polar coordinates of the TS are transformed via the station-tower-ENU link, retaining the time-space mapping term in addition to the angular distance error. The installation angle and lever arm from the GNSS antenna phase center and the TS prism center to the UAV centroid are propagated together with the same aperture. Thus, coordinates from different sources all have covariances that are separable in origin, traceable in path, and consistent in form, making subsequent arithmetic in the information domain an auditable quantitative synthesis, rather than a black-box parameter tuning based on empirical weights.

[0057] In one possible implementation, the information matrices and information vectors of the total station and GNSS are weighted and fused in the information domain using a dynamic weight α driven by a quality score, and a handover factor constraint on the second-order difference of the state is introduced during master-slave switching to maintain covariance continuity.

[0058] In one possible implementation, out-of-order or late observations are inserted historically and smoothed consistently to output UAV position, uncertainty, and quality audit records. In the GNSS integer ambiguity solution stage, a time consistency gating is set, and fixing is prohibited when the equivalent time deviation between the total station and GNSS exceeds a threshold to avoid spurious convergence.

[0059] In one possible implementation, for late or asynchronous observations, historical insertion is used and the window depth and insertion frequency are limited to control computational complexity, and a uniform r95 (2D / 3D) and HDOP / PDOP index is output after smoothing.

[0060] Thus, when real-world links experience delays and out-of-order delivery, by maintaining a fixed-depth information matrix cache, late observations falling into the window are recalculated using small-window information and RTS backtransmission. This corrects the state and covariance only within the local domain, maintaining global consistency on the timeline without sacrificing real-time performance. Whether to trigger recalculation is scheduled using a dual-threshold mechanism of "information gain predictability" and "spectral radius stability," with batch smoothing delayed if necessary. The core idea is to treat consistency as a local conservation law: information is conserved within the window, and real-time performance is conserved outside the window.

[0061] In this embodiment, the time alignment of RTK GNSS and TS is first performed based on their time logs. Then, under time alignment, non-line-of-sight (NLOS) data from the first observation data of the UAV from RTK GNSS is removed based on the TS's position information and line-of-sight direction, as well as the RTK GNSS's satellite ephemeris and line-of-sight occlusion parameters, resulting in optimized first observation data. Further, the optimized first observation data is transformed from the first coordinate system corresponding to RTK GNSS to the target coordinate system, and the second observation data of the UAV from TS is transformed from the second coordinate system corresponding to TS to the target coordinate system. In the target coordinate system, the optimized first and second observation data are fused based on dynamic weights to obtain fused data. Finally, the fused data is smoothed based on the UAV's out-of-order or delayed data to determine the UAV's real-time positioning data. Thus, this application achieves time consistency between RTK GNSS and TS by performing time alignment. Then, the RTKGNSS and TS observation data are converted to the same target coordinate system and fused to accurately determine the real-time positioning data of the UAV. This improves the accuracy of UAV positioning in urban environments.

[0062] Figure 2 This is the second flowchart illustrating the UAV positioning and annotation method based on total station and RTK GNSS provided by this invention. Figure 2 As shown, the above "Step 101, based on the time synchronization logs of the Real-Time Dynamic Global Navigation Satellite System (RTK GNSS) and the total station TS, performs time alignment processing on the RTK GNSS and TS" specifically includes the following: Step 201: Based on the RTK GNSS time log, determine the sampling time drift and first time uncertainty of the RTK GNSS.

[0063] Step 202: Based on the time log of TS, determine the equivalent center time correction and second time uncertainty of the measurement window of TS.

[0064] Step 203: Based on the sampling time drift, the first time uncertainty, the equivalent center time correction amount, and the second time uncertainty, perform time alignment processing on RTK GNSS and TS.

[0065] In one possible implementation, Figure 3 This is a schematic diagram of the mapping relationship of the dual time scales on the time axis provided by the present invention, as shown below. Figure 3 As shown, the timeline includes the request phase, execution (integration window) phase, response phase, and dead zone. The request phase is... to The interval, i.e., the period from when the system initiates a time request to when it obtains the raw time data; the execution (integration window) phase is... to The interval, or Event Capture Time (ECT) core processing window, generates and calibrates timestamps through integration calculations; the response phase is... to The interval, i.e., the period after the ECT timestamp is output, is the time waiting for GNSS timescale alignment; the dead zone is... to The interval refers to the delay interval of GNSS time-stamp processing, which includes the time overhead of signal reception, processing, etc.

[0066] It should be noted that the ECT timescale ( ) represents the timestamp after the event was captured and processed by the integration window. GNSS timestamp ( This represents the time stamp output by the GNSS system, including processing delays. Time uncertainty ( This reflects the transmission relationship from time synchronization error to position error, where... Indicates speed, This represents the standard deviation of the time error, where This represents the equivalent noise covariance mapped from the time uncertainty to the location layer.

[0067] Thus, by introducing the concept of "dual-timescale conformality," the uncertainty of the time link is treated as a directed perturbation in the attitude space, and explicit mapping and suppression are completed before fusion. Specifically, the measurement window request-execution-response-dead-time process of the time series (TS) is compressed into an equivalent central time. The sampling-baseband-output delay of GNSS is modeled as follows: Their respective time variances are expressed through the unit vector of the velocity direction. Mapped to This "geometry of time error" allows time inconsistencies to be presented in a structured way within the spatial covariance, avoiding bias and covariance mismatch caused by passive absorption during filtering. Considering the extreme sensitivity of AR fixation to time consistency, cross-source time gating is proposed: when At this time, AR is temporarily disabled to fix and amplify the time mapping item, thus blocking the centimeter-level jump caused by time misalignment from a mechanism perspective. This is a "time-space conformal" idea based on causal links, rather than the traditional approach of simply incorporating clock drift into the state. This indicates the time correction amount for TS. Indicates the time correction amount for GNSS. This indicates the set time threshold.

[0068] In this embodiment, by determining the sampling time drift and first time uncertainty of the RTK GNSS, and the equivalent center time correction and second time uncertainty of the TS measurement window, time alignment processing can be performed on the RTK GNSS and TS based on this information. This ensures that the time of the data collected by the RTK GNSS and TS when locating the UAV is consistent, thereby enabling the subsequent real-time determination of the accurate UAV positioning data.

[0069] In one possible implementation, the core of "step 103, transforming the optimized first observation data from the first coordinate system to the target coordinate system, and transforming the second observation data of the UAV from the second coordinate system to the target coordinate system" is to uniformly transform the geocentric coordinate system (ITRF) of GNSS and the polar / local coordinate system (t) of total station (TS) to the local ENU coordinate system, and complete the strict propagation of covariance through Jacobian matrix, and finally output the position and covariance of UAV in ENU.

[0070] Specifically, GNSS outputs the position in the ITRF geocentric coordinate system. The model needs to be transformed into ENU using a 7-parameter transformation (translation, rotation, scaling) to obtain its position in the ENU coordinate system. Simultaneously propagating covariance .in, This represents a 7-parameter conversion function. The Jacobian matrix representing the 7-parameter transformation. The covariance of the 7 parameters is obtained through coordinate transformation calibration.

[0071] Furthermore, since TS measures polar coordinates (azimuth, elevation, and slant range), it needs to be converted to rectangular coordinates first, then to ENU, while simultaneously propagating the covariance of multi-source errors, thereby obtaining the rectangular coordinates from the TS station to the target. ,as well as Where R is the slant distance. Indicates the angle of elevation. Indicates azimuth. The Jacobian matrix representing the polar coordinate to rectangular coordinate conversion. Represents the covariance of polar coordinate measurements. This represents the rotation matrix from the local polar coordinate system to ENU. The Jacobian matrix representing the station position error → ENU position error. This represents the covariance of the TS station location. This represents the sum of the covariances of two independent error sources. This represents the total covariance of the TS location under ENU.

[0072] Ultimately, the UAV's position is calculated using APCs (Auxiliary Positioning Points) and local vehicle offsets, requiring conversion to ENU and propagation of multi-source errors. This yields the final UAV position. Covariance of UAV final position .in, This indicates the position of point APC below ENU. This represents the rotation matrix from the UAV body coordinate system to the ENU. This represents the offset vector of the UAV centroid relative to the APC point in the body coordinate system. This represents the covariance of the APC point locations. The Jacobian matrix represents the position relative to the rotation matrix. Denotes the covariance of the rotation matrix. express The measured covariance, through After rotating to the ENU frame, stack them.

[0073] Finally, the positions of GNSS, TS, and UAV are uniformly transformed into the ENU frame, and the covariance of each source is propagated through the Jacobian matrix to ensure that the positions and covariances are "within the same frame and can be fused". This allows the optimized first observation data to be transformed from the first coordinate system to the target coordinate system, and the second observation data of TS on the UAV to be transformed from the second coordinate system to the target coordinate system.

[0074] Figure 4 This is the third flowchart of the UAV positioning and annotation method based on total station and RTK GNSS provided by this invention, as shown below. Figure 4 As shown, the above "Step 102, based on the position information and line-of-sight direction of the TS in time-aligned state, and the satellite ephemeris and line-of-sight occlusion parameters of the RTK GNSS in time-aligned state, removes non-line-of-sight (NLOS) propagation data from the first observation data of the UAV by the RTK GNSS, and obtains optimized first observation data" specifically includes the following: Step 401: Based on location information and line-of-sight direction, as well as satellite ephemeris and line-of-sight occlusion parameters, determine the sub-observation data for each satellite in the first observation data.

[0075] The sub-observation data includes at least one of the following: residual, residual change rate, satellite carrier-to-noise ratio, and elevation angle threshold.

[0076] Step 402: Remove NLOS data from the first observation data based on the target parameter set to obtain optimized first observation data.

[0077] The target parameter set includes at least one of the following: location information and line-of-sight occlusion parameters, residuals and residual change rate, satellite carrier-to-noise ratio and elevation angle threshold, line-of-sight direction and UAV velocity direction.

[0078] In this embodiment, the position information of TS in time-aligned state is the prior position of the total station, and the line-of-sight occlusion parameter of RTKGNSS in time-aligned state is the visibility consistency of the satellite ephemeris.

[0079] In one possible implementation, in urban low-altitude scenarios, endogenous indicators such as C / N0 and DOP cannot provide reliable NLOS discrimination. Therefore, this application proposes a "geometric evidence-first" gating approach, using the instantaneous geometric prior provided by TS. Using environmental occlusion models as the source of evidence, a joint judgment is performed on visibility, residual morphology, and orientation consistency for each satellite. The core principle is not a specific threshold, but rather the logical priority and auditability of the evidence. Visibility evidence is rejected over residual evidence, which in turn takes precedence over orientation evidence. When the evidence chain provides a consistent signal of "slow positive bias in multipaths—visibility conflict—velocity-orientation mismatch," the measurement covariance of the corresponding satellite is monotonically expanded according to its degree of suspicion, and removed if necessary. This strategy represents a paradigm shift from "intensity gating" to "geometric consistency gating," offering advantages such as interpretability, replayability, auditability, and direct contribution to AR stability.

[0080] Specifically, the GNSS NLOS gating process based on TS mapping geometric priors is essentially a combination of high-precision position priors and satellite geometric features (ephemeris, visibility, elevation angle) provided by TS. Through satellite-by-satellite residual analysis and dynamic weight adjustment, NLOS satellite observations are precisely eliminated to improve GNSS positioning accuracy.

[0081] In one possible implementation, the prior position of the satellite track (TS) (with short-range, centimeter-level accuracy, unaffected by satellite obstruction) is used as a "true reference." Combined with satellite geometric features and residual dynamic analysis, line-of-sight (LOS) and non-line-of-sight (NLOS) satellites are distinguished. Then, the influence of NLOS satellites is weakened / removed through weighted expansion / removal, rather than direct hard removal, preserving some usable information and improving positioning robustness.

[0082] Specifically, first input the TS prior (the prior location measured by TS). TS sets satellite filtering thresholds The satellite's geometric characteristics (ephemeris, visibility, elevation threshold) are then used to calculate the satellite-by-satellite residuals. With residual rate of change Furthermore, the residuals, residual change rate, and satellite geometric features are integrated into a comprehensive score. The probability of satellite j being an NLOS satellite is quantified. Finally, based on the comprehensive score... Dynamically adjust the weight of satellite j ,Right now , where k is the expansion coefficient.

[0083] Thus, based on TS position information and line-of-sight direction, as well as RTK GNSS satellite ephemeris and line-of-sight occlusion parameters, this application determines the sub-observation data for each satellite in the first observation data. Then, based on the target parameter set, NLOS data is removed from the first observation data to obtain optimized first observation data. By using the TS prior position as a ground truth reference, combined with satellite geometric characteristics and residual dynamic analysis, it is possible to distinguish between line-of-sight satellites and non-line-of-sight satellites.

[0084] Figure 5 This is the fourth flowchart illustrating the UAV positioning and annotation method based on total station and RTK GNSS provided by this invention. Figure 5 As shown, the dynamic weights are determined in the following way: Step 501: Input the optimized data quality of the first observation data and the data quality of the second observation data into a preset function to obtain the initial weights.

[0085] Step 502: Perform low-pass filtering on the initial weights to obtain dynamic weights.

[0086] Among them, the rate of change of the dynamic weight is less than the preset rate of change.

[0087] In one possible implementation, the weight α (i.e., dynamic weight) of the information domain fusion is gradually varied through a hysteresis Logistic function and low-pass filtering. When the weight α changes rapidly, a weak constraint on the second-order difference of the state is added to limit the trajectory, ensuring the continuity and numerical stability of the covariance.

[0088] Thus, within the information domain, the information matrices and information vectors of TS and GNSS are convexly synthesized using smoothing weights α formed by quality scores, and a weak constraint on the second-order difference of the state is introduced as a smoothing prior to limit jerk impacts. At the numerical level, adaptive regularization and backoff using condition numbers and NIS / NEES ensure the controllability of the information matrix spectral properties. This method transforms the "dominance switching" into a differentiable optimization process with boundary protection, ensuring that the trajectory maintains a consistent covariance and interpretable energy profile during the handover transient.

[0089] Specifically, the weights are initially calculated using the Logistic function, based on the quality index of TS. and GNSS quality indicators As input, the quality index is mapped to an initial weight of 0 to 1 using the Logistic function. Then, as shown in Equation 1, the weights are smoothed using a low-pass filter (LPF) to smooth the initial weights. Low-pass filtering is used to avoid jitter in the fusion result caused by abrupt changes in weights. Quality metrics include positioning accuracy, residuals, and DOP value.

[0090] Formula 1 Furthermore, information domain fusion calculation is performed. The advantage of information domain fusion (rather than direct location fusion) is that it can directly combine covariance, ensuring the consistency of covariance after fusion, as shown in Formula 2.

[0091] Formula 2 in, This represents the fused information vector. The information vector representing TS, Represents the information vector of GNSS. This indicates the jerk compensation term. For compensation coefficient, Let x be the jerk information vector, and let x represent the state estimate. By introducing kinematic constraints (jerk), we can avoid the fusion result from deviating from the laws of physical motion (e.g., when a drone is flying at a constant speed, the jerk is close to 0).

[0092] Then, the observations are fused using Formula 3. ).

[0093] Formula 3 in, This represents the observed values ​​of TS (e.g., position, velocity). This represents GNSS observations. By directly fusing the raw observations with the information vector, a complementary effect is achieved, improving the stability of the fusion process.

[0094] Finally, state and covariance calculations are performed, and the calculated state estimate is derived from the fused information vector. Covariance After solution and Furthermore, r95 must be continuous without any jumps, meaning the difference between adjacent epochs is less than a preset threshold.

[0095] Figure 6 This is the fifth flowchart illustrating the UAV positioning and annotation method based on total station and RTK GNSS provided by this invention. Figure 6 As shown, the above "Step 105: Smooth the fused data based on the out-of-order or delayed data of the UAV to determine the real-time positioning data of the UAV" specifically includes the following: Step 601: Based on the preset sliding window depth and preset insertion frequency, insert disordered or delayed data into the fused data to obtain the fused data after data insertion.

[0096] Step 602: Smooth the fused data after data insertion to determine the real-time positioning data of the UAV.

[0097] In one possible implementation, Figure 7 This is a schematic diagram of data insertion and smoothing provided by the present invention, such as... Figure 7 As shown, in time series data, out-of-order or delayed observations are inserted into their correct time positions through historical insertion, and windowing smoothing is used to improve data quality. The dots represent observations generated in normal chronological order, arranged from left to right along the time axis, representing the default ordered input. The arrows indicating "late observation insertion" refer to observations that should have appeared at an earlier time point but arrived later than subsequent data due to transmission delays, processing queuing, etc. Therefore, they need to be back-inserted to their original time positions, rather than appended to the latest data, to ensure the integrity of the time series.

[0098] Furthermore, after inserting out-of-order or delayed observation data, batch processing or the Rauch-Tung-Striebel (RTS) smoothing algorithm is used to re-filter all data (original observation data and inserted out-of-order or delayed observation data) within a time interval (i.e., a "smoothing window") that includes the insertion point. This process corrects the historical state estimation and improves the consistency and accuracy of the overall data.

[0099] Figure 8 This is the sixth flowchart of the UAV positioning and annotation method based on total station and RTK GNSS provided by this invention, as shown below. Figure 8 As shown, UAV positioning and annotation methods based on total station and RTK GNSS also include: Step 801: During the process of fusing the optimized first and second observation data, monitor the fusion parameters.

[0100] The fusion parameters include at least one of the following: the condition number of the information matrix, the normalized innovation square (NIS), and the normalized estimation error square (NEES).

[0101] Step 802: If the condition number of the information matrix is ​​greater than the preset condition number, and / or the consistency parameter of NIS and NEES does not match the preset consistency parameter, determine the real-time positioning data of the UAV based on the first observation data or the second observation data.

[0102] In one possible implementation, when the information matrix condition number exceeds the limit or the NIS / NEES consistency mismatch occurs, a fallback to single-source dominance (total station or GNSS) is triggered and the handover event and threshold are recorded in the evidence chain.

[0103] In other words, during information domain fusion computation, anomaly detection and robust rollback mechanisms are required. Fusion anomalies are detected using two indicators, triggering rollback and alarms. These two indicators are: condition number and NIS (Normalized Innovative Square). If the condition number / NIS exceeds the limit, a rollback to single-source computation is immediately initiated (TS takes precedence; if no TS exists, GNSS is rolled back). Simultaneously, an alarm is triggered, and the anomaly type, epoch, and indicator value are recorded in the evidence chain for subsequent tracing.

[0104] It should be noted that the "jitter and step" in multi-source switching is not an inherent characteristic of filters, but rather a product of discontinuous weights and asymmetric priors. This application proposes the idea of ​​"covariance-consistent handover": within the information domain, the information matrices and information vectors of TS and GNSS are convexly synthesized using smooth weights α formed by quality scores, and a weak constraint on the second-order difference of the state is introduced as a smooth prior to limit jerk impacts. At the numerical level, adaptive regularization and backoff based on condition numbers and NIS / NEES ensure the controllability of the spectral properties of the information matrix. Thus, by transforming "dominance switching" into a differentiable optimization process with boundary protection, the trajectory maintains a consistent covariance profile and interpretable energy form during the handover transient.

[0105] Figure 9 This is the seventh flowchart of the UAV positioning and annotation method based on total station and RTK GNSS provided by this invention, as shown below. Figure 9 As shown, the above "Step 106, marking the drone's position in the image captured by the drone's onboard camera based on the drone's real-time positioning data" specifically includes: Step 901: Determine the first positioning coordinates of the real-time positioning data in the world coordinate system.

[0106] Step 902: Based on the camera extrinsic parameters, transform the first positioning coordinates to the camera coordinate system to obtain the second positioning coordinates of the real-time positioning data in the camera coordinate system.

[0107] Step 903: Based on the camera intrinsic parameters and distortion parameters, perform pinhole projection and distortion processing on the second positioning coordinates to obtain a set of pixels.

[0108] The set of pixels is used to construct the drone.

[0109] Step 904: Generate the bounding box of the UAV based on the set of pixels.

[0110] The annotation box is used to mark the location of the drone in the image.

[0111] In one possible implementation, the world coordinate system W can be defined as the local ENU, and the camera coordinate system C as the optical coordinate system. The camera extrinsic parameters are transformed from the world coordinate system to the camera coordinate system to achieve a rigid body transformation between the coordinate systems, as shown in Equation 4.

[0112] Formula 4 in, This indicates the coordinates of the drone in the camera coordinate system. This represents the coordinates of the drone in the world coordinate system. This represents the coordinates (translation vector) of the origin of the camera coordinate system in the world coordinate system. This represents the rotation matrix from the world coordinate system to the camera coordinate system. Denotes a special orthogonal group. Represents a three-dimensional vector.

[0113] In one possible implementation, if the camera is mounted at a station or on a fixed base, its pose... The calculation can be performed using total station control points or collinear targets at multiple points and then permanently fixed; if the camera is mounted on a moving platform, the calculation can be performed using camera extrinsic parameters. UAV posture The camera pose is obtained by chaining. And its uncertainty is co-propagated to the pixel domain.

[0114] Thus, the origin deviation is eliminated by translation (i.e., calculation) ), points in the world coordinate system Translate to a position relative to the origin of the camera coordinate system. Then, rotate the coordinate system orientation using a rotation matrix. Multiplying the translated vector by the left transforms the vector's pose from the world coordinate system to the camera coordinate system, ultimately yielding the UAV's coordinates in the camera coordinate system. .

[0115] In one possible implementation, it is also necessary to determine the coordinates of the drone in the camera coordinate system. ,exist Under the premise of visibility, normalized projection is performed to obtain normalized coordinates. As shown in Formula 5.

[0116] Formula 5 It is understandable that normalized projection is the core of perspective projection, which can represent the three-dimensional coordinates of the drone in the camera coordinate system. Projecting onto the normalized plane at z=1 yields two-dimensional normalized coordinates. . The constraint condition ensures that the point is located in front of the camera (within the visible range); r represents the radial distance from the normalized point to the image origin, which is the core input of the subsequent distortion model. The normalized coordinates are dimensionless, independent of the camera hardware parameters, and serve as an intermediate quantity connecting 3D space and the pixel plane.

[0117] Furthermore, the normalized coordinates were adjusted using the Brown–Conrady distortion model. Distortion correction is performed to obtain the distorted normalized coordinates. As shown in Formula 6.

[0118] Formula Six It should be noted that the Brown-Conrady distortion model corrects lens distortion because actual camera lenses have radial and tangential distortion, which can cause the imaging point to deviate from the ideal position, thus requiring distortion compensation.

[0119] In one possible implementation, the distortion model uses the Brown-Conrady model, or an equidistant / equifixed angle fisheye model, with the version and hash of the model parameters incorporated into the chain of evidence.

[0120] Radial distortion, caused by lens shape defects, manifests as "barrel distortion" or "pincushion distortion" at the image edges, and the corresponding formula includes... Item, Radial distortion coefficient, low-order coefficient Dominates most distortions, higher-order coefficients Used to correct severe distortion in wide-field-of-view lenses. Tangential distortion is caused by lens mounting misalignment (the lens is not parallel to the image plane), and the corresponding formula contains... Item, It is the tangential distortion coefficient.

[0121] Thus, by inputting ideal normalized coordinates This will output the distorted normalized coordinates. If used for image correction, the coordinates of the distorted points are input, and the coordinates of the ideal points are solved (inverse distortion calculation).

[0122] Then, based on the distorted normalized coordinates ( By using Formula 7 to perform intrinsic parameter projection, the pixel coordinates of the UAV can be obtained. (i.e., a set of pixels).

[0123] Formula 7 It is understandable that the distorted normalized coordinates ( Mapping to the pixel plane, the core is the role of the camera intrinsic parameter matrix K, which is in the form of Equation 8.

[0124] Formula 8 in, This represents the pixel focal length, measured in pixels, and is the ratio of the physical focal length f to the pixel size. The ratio, i.e. , . The coordinates of the principal point are shown in pixels, corresponding to the intersection of the camera's optical axis and the imaging plane, usually the center of the image. 's' represents the tilt factor, caused by the x-axis and y-axis of the imaging plane not being perpendicular; for most cameras, 's' = 0.

[0125] It should be noted that for large field-of-view cameras such as fisheye cameras, models such as equidistant / equifixed angle cameras can be used instead, and the method is similar.

[0126] In one possible implementation, when generating the bounding box of the UAV based on the set of pixels, it is necessary to obtain the 2D bounding box / mask through the 3D shape prior, with the centroid of the UAV in the world coordinate system as the reference. Using prior shape information (such as a 3D bounding box / polyhedron / sphere approximation) as input, sample its set of visible vertices. The pixels are then projected sequentially onto the pixel plane to obtain the result. Therefore: 2D bounding boxes (YOLO, etc.) take the AABB of the visible point set. 2D (2D axis-aligned bounding box) yields (u min v min u max v max ), and normalized to (c x c y (w, h). Then segment the mask / polygon, perform convex hull or visibility clipping based on surface connectivity on the visible projection, generate 2D polygons and rasterize them into a mask. Specifically, visibility and clipping are performed on Z... C Points ≤0, out of view, or occluded (in conjunction with building / terrain models or depth maps) are removed. Cross-boundary targets are clipped and marked "ignore". Physical scale constraints are applied: if the UAV's external dimensions are known, distance-pixel scale consistency is gated and can be used as an audit item.

[0127] Thus, the core of the automatic generation process of 2D visual annotation based on 3D shape prior is to combine the three-dimensional geometric model of the UAV with the pose of the world coordinate system, and output the ground truth values ​​of 2D detection boxes, segmentation masks and other annotations in YOLO format through camera projection and post-processing.

[0128] Furthermore, to ensure that the annotations are consistent with the image exposure time, it is necessary to use the equivalent center time of camera exposure under a unified time base. As shown in Formula Nine.

[0129] Formula Nine in, This indicates the initial moment (start of time reference) when the camera triggers the signal. This represents the hardware delay from when the sensor receives the trigger signal to when exposure begins. This indicates the signal processing delay of the camera's internal circuitry. Indicates rolling shutter speed for line r The line delay. The pose p(t) output by the positioning module is used in... Interpolation / extrapolation, time uncertainty according to Mapped to the pixel domain as a labeled confidence interval or boundary buffer.

[0130] Thus, defining the Equivalent Center Time (ECT) and combining it with spatiotemporal mapping to ensure accurate matching of UAV 3D pose and 2D image annotation in the time dimension is a key step in automatic annotation in highly dynamic scenarios (such as UAV flight).

[0131] In one possible implementation, the camera uses a rolling shutter, and the camera's line exposure time parameter... Included Calculate and map its time uncertainty to a pixel domain boundary buffer.

[0132] In one possible implementation, the UAV shape prior can be a three-dimensional bounding box, ellipsoid, or polyhedron, and the projection differences between different priors are included in the annotation uncertainty as a systematic term.

[0133] In one possible implementation, multi-camera consistent output is supported, for each camera C j The annotations output the camera ID, extrinsic parameter version hash, and pixel domain uncertainty, and perform cross-camera consistency auditing.

[0134] Specifically, for multiple cameras C j , respectively with (K j D j , Projection generates annotations, outputting consistent annotation entries across cameras, pixel-level uncertainty, time audit, and visibility and clipping justifications. Annotation entries include: target ID, camera ID, frame number, YOLO box, mask, and polygon; pixel-level uncertainty... The pose covariance matrix of the UAV in the world coordinate system W Obtained through covariance propagation with external / internal references; time audit includes: , Rolling shutter parameters, interpolation order and residuals; visibility and cropping reasons include: FOV exceeding limits, Z... C ≤0, occlusion model hit rate, etc.

[0135] It should be noted that due to differences in intrinsic, distortion, and extrinsic parameters, the projection annotations of multiple cameras on the same UAV will vary. Consistency verification needs to achieve spatial consistency, temporal consistency, and attribute consistency. Spatial consistency means that the 2D bounding boxes / masks annotated by different cameras must overlap in the 3D regions corresponding to the same world coordinate system (within a threshold error). Temporal consistency means that the annotations of all cameras are aligned to a unified time reference. This avoids labeling deviations caused by differences in exposure times of different cameras; attribute consistency means that the target ID, visibility determination, error range, and other attributes of the same UAV are consistent across multiple cameras.

[0136] In this way, a unified "traceable, verifiable, and alignable" annotation system is established for UAV annotation from multiple camera perspectives. This ensures the spatial / temporal consistency of annotation results from different cameras and records the entire process of annotation generation through a complete chain of evidence, thus solving problems such as ambiguity in multi-camera annotation, error tracing, and verification of result credibility.

[0137] Finally, output YOLO tag format (one target per line): (cls, c x / W, c y / H, w / W, h / H). Where cls represents the target category ID (e.g., 0 for drones), which is an integer; c x / W, c y / H represents the center coordinates of the 2D bounding box, normalized to the image width and height (0~1), and is a floating-point number; w / W and h / H represent the width and height of the 2D bounding box, normalized to the image width and height (0~1), and are floating-point numbers; W and H are the image width and height in pixels. It can also output the segmentation, bounding box, and category_id of the COCO JSON, along with evidence chain fields (time, extrinsic parameter version hash, propagation covariance, etc.).

[0138] In this way, by encapsulating the results of 3D-to-2D projection, temporal alignment, and multi-camera consistency into a labeling format compatible with mainstream visual algorithms such as YOLO / COCO, and embedding custom evidence chain fields in the format, the standard requirements for algorithm training are met, while retaining auditable information throughout the labeling process.

[0139] In one possible implementation, this application considers the "chain of evidence" as a primary product equivalent to the location result: each gating, alignment, weight scheduling, backtracking, and smoothing operation is stored as structured metadata, accompanied by version and parameter signature hashes. The output r95, HDOP / PDOP, covariance spectral radius, NIS / NEES over-limit rate, AR / cycle slip status, NLOS list and weight trajectory, seven-parameter version and signature, OOSM insertion record, and impact assessment together constitute replayable, verifiable, and traceable engineering evidence. Its value lies in elevating "credibility" from a result attribute to a process attribute, meeting the essential needs of supervision, evidence collection, and safety assessment.

[0140] In one possible implementation, this application does not simply involve paralleling or cascading GNSS and TS, but proposes a methodological paradigm of "spatiotemporal consistency—geometric evidence priority—auditability in the information domain": First, before observations enter filtering, the temporal, spatial, installation, and caliber unification across modes is completed, making all source information comparable on the same uncertainty metrology; then, geometric consistency, rather than signal strength, is used as the primary criterion for GNSS observations; finally, in the information domain, "covariance consistency handover" achieves a smooth transfer of dominance. This paradigm moves the system from "using multiple sources in combination" to "using interpretable evidence to dominate multiple sources," with the core objective of moving the three persistent engineering problems of pseudo-convergence, timing misalignment, and multipath to a measurable, controllable, and auditable level.

[0141] This application utilizes total stations and RTK GNSS as two types of metrological anchors. Addressing the challenges of GNSS multipath / NLOS and line-of-sight interruptions in complex urban environments, it proposes the following: dual-timescale consistency modeling and time uncertainty mapping; GNSS satellite-by-satellite NLOS gating based on mapping geometric priors; seven-parameter alignment and covariance propagation from ITRF / WGS84 to ENU; weighted fusion of the information domain and "covariance-consistent master-slave handover"; and OOSM consistency smoothing and versioned evidence chain output for out-of-order observations. Building upon this, it further proposes an automatic visual data annotation module: projecting centimeter-level world coordinates and shape priors onto the pixel plane via camera extrinsic and intrinsic parameters to generate 2D bounding box / mask ground truth values ​​in YOLO / COCO formats, and propagating time / extrinsic / intrinsic parameters and positioning uncertainties consistently to the pixel domain, forming an auditable annotation evidence chain. This technology is mutually exclusive yet complementary to "end-to-end AI fusion," "unified metrological closed loop of total stations (excluding GNSS)," and "visual collaborative four-dimensional positioning," forming a core component of the "satellite-ground-visual annotation" joint protection system.

[0142] The following describes the UAV positioning and labeling device based on total station and RTK GNSS provided by the present invention. The UAV positioning and labeling device based on total station and RTK GNSS described below can be referred to in correspondence with the UAV positioning and labeling method based on total station and RTK GNSS described above.

[0143] Figure 10 This is a schematic diagram of the UAV positioning and marking device based on a total station and RTK GNSS provided by the present invention, as shown below. Figure 10 As shown, the UAV positioning and labeling device based on total station and RTK GNSS includes the following modules: processing module 1001, optimization module 1002, conversion module 1003, fusion module 1004, and determination module 1005; The processing module 1001 is used to perform time alignment processing on RTK GNSS and TS based on the time logs of the real-time dynamic global navigation satellite system RTK GNSS and the time logs of the total station TS. The optimization module 1002 is used to remove non-line-of-sight (NLOS) data from the first observation data of the UAV by RTK GNSS based on the position information and line-of-sight direction of TS in time-aligned state, as well as the satellite ephemeris and line-of-sight occlusion parameters of RTKGNSS in time-aligned state, so as to obtain optimized first observation data. The conversion module 1003 is used to convert the optimized first observation data from the first coordinate system to the target coordinate system, and to convert the second observation data of the UAV by the TS from the second coordinate system to the target coordinate system. The first coordinate system is the coordinate system corresponding to RTK GNSS, and the second coordinate system is the coordinate system corresponding to TS. The fusion module 1004 is used to fuse the optimized first observation data and the second observation data in the target coordinate system based on dynamic weights to obtain fused data. The dynamic weights are determined based on the data quality of the optimized first observation data and the data quality of the second observation data. The continuity parameter of the covariance of the fused data satisfies the preset continuity parameter. The determination module 1005 is used to smooth the fused data based on the out-of-order or delayed data of the UAV and determine the real-time positioning data of the UAV. The processing module 1001 is also used to mark the location of the UAV in the images captured by the UAV's onboard camera based on the UAV's real-time positioning data.

[0144] According to the present invention, a UAV positioning and labeling device based on a total station and RTK GNSS is provided, wherein the processing module 1001 is specifically used for: Based on the time logs of RTK GNSS, the sampling time drift and first time uncertainty of RTK GNSS are determined; Based on the time log of TS, the equivalent center time correction and second time uncertainty of the measurement window of TS are determined; Time alignment is performed on RTK GNSS and TS based on sampling time drift, first time uncertainty, equivalent center time correction, and second time uncertainty.

[0145] According to the present invention, an unmanned aerial vehicle (UAV) positioning and labeling device based on a total station and RTK GNSS is provided, wherein the optimization module 1002 is specifically used for: Based on location information and line-of-sight direction, as well as satellite ephemeris and line-of-sight occlusion parameters, sub-observation data for each satellite in the first observation data are determined. The sub-observation data includes at least one of the following: residual, residual change rate, satellite carrier-to-noise ratio, and elevation angle threshold. The optimized first observation data is obtained by removing NLOS data from the first observation data based on the target parameter set. The target parameter set includes at least one of the following: location information and line-of-sight occlusion parameters, residuals and residual change rate, satellite carrier-to-noise ratio and elevation angle threshold, line-of-sight direction and UAV velocity direction.

[0146] According to the present invention, a UAV positioning and labeling device based on a total station and RTK GNSS, the processing module 1001 is further used for: The optimized data quality of the first observation data and the data quality of the second observation data are input into a preset function to obtain the initial weights. The initial weights are low-pass filtered to obtain dynamic weights, and the rate of change of the dynamic weights is less than the preset rate of change.

[0147] According to the present invention, a UAV positioning and labeling device based on a total station and RTK GNSS is provided, wherein the determining module 1005 is specifically used for: Based on a preset sliding window depth and a preset insertion frequency, out-of-order or delayed data is inserted into the fused data to obtain the fused data after data insertion; The fused data after data insertion is smoothed to determine the real-time positioning data of the UAV.

[0148] According to the present invention, a UAV positioning and labeling device based on a total station and RTK GNSS is provided, wherein the processing module 1001 is further used for: During the fusion of the optimized first and second observation data, the fusion parameters are monitored. The fusion parameters include at least one of the following: the condition number of the information matrix, the normalized innovation square (NIS), and the normalized estimation error square (NEES). If the condition number of the information matrix is ​​greater than the preset condition number, and / or the consistency parameter between NIS and NEES does not match the preset consistency parameter, the real-time positioning data of the UAV is determined based on the first observation data or the second observation data.

[0149] According to the present invention, a UAV positioning and labeling device based on a total station and RTK GNSS is provided, wherein the processing module 1001 is further used for: Determine the first positioning coordinates of the real-time positioning data in the world coordinate system; Based on the camera extrinsic parameters, the first positioning coordinates are transformed into the camera coordinate system to obtain the second positioning coordinates of the real-time positioning data in the camera coordinate system; Based on camera intrinsic parameters and distortion parameters, pinhole projection and distortion processing are performed on the second positioning coordinates to obtain a set of pixels, which is used to construct the UAV. Based on the set of pixels, a bounding box for the drone is generated. The bounding box is used to mark the location of the drone in the image.

[0150] Figure 11 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 11As shown, the electronic device may include: a processor 1110, a communications interface 1120, a memory 1130, and a communications bus 1140, wherein the processor 1110, the communications interface 1120, and the memory 1130 communicate with each other through the communications bus 1140. The processor 1110 can call logical instructions in the memory 1130 to execute a UAV positioning and annotation method based on a total station and RTK GNSS. This method includes: performing time alignment processing on the RTK GNSS and TS based on the time logs of the real-time dynamic global navigation satellite system RTK GNSS and the total station TS; removing non-line-of-sight (NLOS) data from the first observation data of the UAV from the RTK GNSS based on the position information and line-of-sight direction of the TS in the time-aligned state, and the satellite ephemeris and line-of-sight occlusion parameters of the RTK GNSS in the time-aligned state, to obtain optimized first observation data; transforming the optimized first observation data from a first coordinate system to a target coordinate system, and transforming the second observation data of the UAV from the TS from a second coordinate system to the target coordinate system, where the first coordinate system is the RTK GNSS. The coordinate system is GNSS, and the second coordinate system is the coordinate system corresponding to TS. In the target coordinate system, the optimized first and second observation data are fused based on dynamic weights to obtain fused data. The dynamic weights are determined based on the data quality of the optimized first and second observation data. The continuity parameter of the covariance of the fused data satisfies the preset continuity parameter. The fused data is smoothed based on the out-of-order or delayed data of the UAV to determine the real-time positioning data of the UAV. Based on the real-time positioning data of the UAV, the position of the UAV is marked in the image acquired by the UAV's onboard camera.

[0151] Furthermore, the logical instructions in the aforementioned memory 1130 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0152] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the UAV positioning and labeling methods based on total station and RTK GNSS provided by the above methods. The method includes: performing time alignment processing on RTK GNSS and TS based on the time alignment logs of the real-time dynamic global navigation satellite system RTK GNSS and the total station TS; removing non-line-of-sight (NLOS) data from the first observation data of the UAV by RTK GNSS based on the position information and line-of-sight direction of TS in the time-aligned state, as well as the satellite ephemeris and line-of-sight occlusion parameters of RTK GNSS in the time-aligned state, to obtain optimized first observation data; transforming the optimized first observation data from a first coordinate system to a target coordinate system, and transforming the second observation data of the UAV by TS from a second coordinate system to the target coordinate system, wherein the first coordinate system is RTK GNSS. The coordinate system is GNSS, and the second coordinate system is the coordinate system corresponding to TS. In the target coordinate system, the optimized first and second observation data are fused based on dynamic weights to obtain fused data. The dynamic weights are determined based on the data quality of the optimized first and second observation data. The continuity parameter of the covariance of the fused data satisfies the preset continuity parameter. The fused data is smoothed based on the out-of-order or delayed data of the UAV to determine the real-time positioning data of the UAV. Based on the real-time positioning data of the UAV, the position of the UAV is marked in the image acquired by the UAV's onboard camera.

[0153] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program performs the above-described methods for UAV positioning and labeling based on a total station and RTK GNSS. This method includes: performing time alignment processing on the RTK GNSS and TS based on the time logs of the real-time dynamic global navigation satellite system RTK GNSS and the total station TS; removing non-line-of-sight (NLOS) propagation data from the first observation data of the UAV from the RTK GNSS based on the position information and line-of-sight direction of the TS in the time-aligned state, and the satellite ephemeris and line-of-sight occlusion parameters of the RTK GNSS in the time-aligned state, to obtain optimized first observation data; transforming the optimized first observation data from a first coordinate system to a target coordinate system, and transforming the second observation data of the UAV from the TS from a second coordinate system to the target coordinate system, wherein the first coordinate system is the RTK GNSS. The coordinate system is GNSS, and the second coordinate system is the coordinate system corresponding to TS. In the target coordinate system, the optimized first and second observation data are fused based on dynamic weights to obtain fused data. The dynamic weights are determined based on the data quality of the optimized first and second observation data. The continuity parameter of the covariance of the fused data satisfies the preset continuity parameter. The fused data is smoothed based on the out-of-order or delayed data of the UAV to determine the real-time positioning data of the UAV. Based on the real-time positioning data of the UAV, the position of the UAV is marked in the image acquired by the UAV's onboard camera.

[0154] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0155] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for UAV positioning and annotation based on total station and RTK GNSS, characterized in that, include: Based on the time synchronization logs of the real-time dynamic global navigation satellite system RTK GNSS and the total station TS, time alignment processing is performed on the RTK GNSS and the TS; Based on the position information and line-of-sight direction of the TS in time-aligned state, and the satellite ephemeris and line-of-sight occlusion parameters of the RTK GNSS in time-aligned state, the non-line-of-sight propagation (NLOS) data in the first observation data of the UAV by the RTK GNSS is removed to obtain the optimized first observation data. The optimized first observation data is transformed from the first coordinate system to the target coordinate system, and the second observation data of the UAV by the TS is transformed from the second coordinate system to the target coordinate system. The first coordinate system is the coordinate system corresponding to the RTKGNSS, and the second coordinate system is the coordinate system corresponding to the TS. In the target coordinate system, the optimized first observation data and the second observation data are fused based on dynamic weights to obtain fused data. The dynamic weights are determined based on the data quality of the optimized first observation data and the data quality of the second observation data. The continuity parameter of the covariance of the fused data satisfies a preset continuity parameter. The fused data is smoothed based on the disordered or delayed data of the UAV to determine the real-time positioning data of the UAV. Based on the real-time positioning data of the UAV, the location of the UAV is marked in the image captured by the UAV's onboard camera.

2. The UAV positioning and labeling method based on total station and RTK GNSS according to claim 1, characterized in that, The time alignment process based on the time logs of the Real-Time Kinematic GNSS (RTK GNSS) and the total station time synchronization (TS) includes: Based on the time log of the RTK GNSS, the sampling time drift and first time uncertainty of the RTK GNSS are determined; Based on the time log of the TS, determine the equivalent center time correction and the second time uncertainty of the measurement window of the TS; Based on the sampling time drift, the first time uncertainty, the equivalent center time correction, and the second time uncertainty, time alignment processing is performed on the RTK GNSS and the TS.

3. The UAV positioning and annotation method based on total station and RTK GNSS according to claim 1 or 2, characterized in that, Based on the position information and line-of-sight direction of the TS in time-aligned state, and the satellite ephemeris and line-of-sight occlusion parameters of the RTK GNSS in time-aligned state, the non-line-of-sight (NLOS) propagation data in the first observation data of the UAV by the RTK GNSS is removed to obtain optimized first observation data, including: Based on the location information and the line-of-sight direction, as well as the satellite ephemeris and the line-of-sight occlusion parameters, sub-observation data for each satellite in the first observation data are determined. The sub-observation data includes at least one of the following: residual, residual change rate, satellite carrier-to-noise ratio, and elevation angle threshold. The optimized first observation data is obtained by removing the NLOS data from the first observation data based on the target parameter set. The target parameter set includes at least one of the following: the location information and the line-of-sight occlusion parameter, the residual and the residual change rate, the satellite carrier-to-noise ratio and the elevation angle threshold, the line-of-sight direction and the speed direction of the UAV.

4. The UAV positioning and annotation method based on total station and RTK GNSS according to claim 1 or 2, characterized in that, The dynamic weights are determined in the following way: The optimized data quality of the first observation data and the data quality of the second observation data are input into a preset function to obtain the initial weights. The initial weights are subjected to low-pass filtering to obtain the dynamic weights, and the rate of change of the dynamic weights is less than a preset rate of change.

5. The UAV positioning and labeling method based on total station and RTK GNSS according to claim 1 or 2, characterized in that, The step of smoothing the fused data based on the disordered or delayed data of the UAV to determine the real-time positioning data of the UAV includes: Based on a preset sliding window depth and a preset insertion frequency, the disordered or delayed data is inserted into the fused data to obtain the fused data after data insertion; The smoothing process is applied to the fused data after data insertion to determine the real-time positioning data of the UAV.

6. The UAV positioning and annotation method based on total station and RTK GNSS according to claim 1 or 2, characterized in that, The method further includes: During the process of fusing the optimized first observation data and the second observation data, the fusion parameters are monitored, and the fusion parameters include at least one of the following: the condition number of the information matrix, the normalized innovation square (NIS), and the normalized estimation error square (NEES). If the condition number of the information matrix is ​​greater than the preset condition number, and / or the consistency parameter of the NIS and NEES does not match the preset consistency parameter, the real-time positioning data of the UAV is determined based on the first observation data or the second observation data.

7. The UAV positioning and annotation method based on total station and RTK GNSS according to claim 1 or 2, characterized in that, The step of marking the location of the UAV in the images captured by the UAV's onboard camera based on the UAV's real-time positioning data includes: Determine the first positioning coordinates of the real-time positioning data in the world coordinate system; Based on the camera extrinsic parameters, the first positioning coordinates are transformed to the camera coordinate system to obtain the second positioning coordinates of the real-time positioning data in the camera coordinate system; Based on camera intrinsic parameters and distortion parameters, pinhole projection and distortion processing are performed on the second positioning coordinates to obtain a set of pixels, which is used to constitute the UAV. Based on the set of pixels, a bounding box for the drone is generated, which is used to mark the location of the drone in the image.

8. A UAV positioning and marking device based on a total station and RTK GNSS, characterized in that, include: The module comprises a processing module, an optimization module, a conversion module, a fusion module, and a determination module. The processing module is used to perform time alignment processing on the RTK GNSS and the TS based on the time synchronization log of the real-time dynamic global navigation satellite system RTK GNSS and the time synchronization log of the total station TS; The optimization module is used to remove non-line-of-sight (NLOS) data from the first observation data of the UAV by the RTK GNSS based on the position information and line-of-sight direction of the TS in the time-aligned state, as well as the satellite ephemeris and line-of-sight occlusion parameters of the RTK GNSS in the time-aligned state, to obtain optimized first observation data. The conversion module is used to convert the optimized first observation data from the first coordinate system to the target coordinate system, and to convert the second observation data of the UAV by the TS from the second coordinate system to the target coordinate system. The first coordinate system is the coordinate system corresponding to the RTK GNSS, and the second coordinate system is the coordinate system corresponding to the TS. The fusion module is used to fuse the optimized first observation data and the second observation data in the target coordinate system based on dynamic weights to obtain fused data. The dynamic weights are determined based on the data quality of the optimized first observation data and the data quality of the second observation data. The continuity parameter of the covariance of the fused data satisfies a preset continuity parameter. The determining module is used to smooth the fused data based on the disordered or delayed data of the UAV, and determine the real-time positioning data of the UAV. The processing module is also used to mark the location of the UAV in the images captured by the UAV's onboard camera based on the UAV's real-time positioning data.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the UAV positioning and labeling method based on total station and RTK GNSS as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the UAV positioning and labeling method based on total station and RTK GNSS as described in any one of claims 1 to 7.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the UAV positioning and labeling method based on total station and RTK GNSS as described in any one of claims 1 to 7.

Images