A method and system for locating unmanned aerial vehicles (UAVs)

Abstract

This invention relates to the field of control or regulation systems for non-electrical variables, and more particularly to a UAV positioning method and system. The system includes an airborne positioning module integrating an ultra-wideband transceiver, an inertial measurement unit, a visual sensor module, and a global navigation satellite system (GNSS) receiver module. It also coordinates with an external UWB positioning beacon. Through core functional modules such as real-time multimodal data quality assessment, dynamic weighted fusion positioning, environmental perception and situational awareness, and robustness and integrity monitoring, it achieves high-precision, continuous, and reliable positioning of the UAV. This invention significantly improves positioning accuracy and system robustness, enhances adaptability to complex environments and anti-interference capabilities, and effectively reduces system deployment and maintenance costs.

Description

Technical Field

[0001] This invention relates to the field of control or regulation systems for non-electrical variables, and in particular to a method and system for locating unmanned aerial vehicles (UAVs). Background Technology

[0002] In the diverse missions performed by unmanned aerial vehicles (UAVs), high-precision and highly adaptable positioning capabilities are the cornerstone for autonomous flight, precise operations, collaborative control, and even obstacle avoidance. Their performance directly impacts mission success rates. Therefore, continuously improving the positioning accuracy, anti-interference capabilities, and multi-scenario compatibility of UAVs in complex environments remains a core focus for technical personnel.

[0003] Among the existing technologies, the Chinese invention patent with publication number CN110865337A and publication date of March 6, 2020, entitled "UAV Laser Positioning Device", is based on utilizing the narrow beam characteristics of laser and its high directionality of propagation. By deploying laser transmitters and receivers within a preset area, a positioning system based on the laser measurement principle is constructed. This system can achieve high-precision positioning of UAVs to a certain extent, especially in indoor or enclosed environments with stable lighting conditions and relatively regular spatial structures, where its positioning effect shows certain advantages.

[0004] However, the deployment of laser equipment and the precise control of the reflector group are highly dependent. This means that in order to ensure positioning accuracy, a large amount of dedicated hardware must be pre-deployed in the work area and extremely precise calibration and maintenance must be carried out. This undoubtedly greatly increases the deployment cost of the system and seriously restricts its large-scale application in vast, dynamic or unstructured outdoor environments. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for locating unmanned aerial vehicles (UAVs) to solve the deep-seated technical limitations of existing UAVs in terms of adaptability to complex environments and anti-interference capabilities.

[0006] One aspect of the present invention provides a method for locating a drone, comprising:

[0007] During flight, the UAV collects raw data related to its own motion in real time through the UWB transceiver, IMU, visual sensor module and GNSS receiver module of the onboard positioning module;

[0008] Real-time quality assessment, outlier detection, and filtering are performed on the collected raw data;

[0009] Using preprocessed visual image data, the current environment of the UAV is semantically segmented and classified through deep learning algorithms to identify the environment type; at the same time, by combining the abnormal fluctuations of UWB, GNSS and IMU data, potential interference sources are identified; based on the identified environment type and the real-time motion state of the UAV, the state space model and observation model parameters used for subsequent fusion positioning are dynamically selected.

[0010] The pre-processed and quality-assessed sensor data is input into the dynamic weighted fusion positioning engine;

[0011] Real-time monitoring and evaluation of the output fusion positioning results;

[0012] After robustness and integrity monitoring, the UAV's three-dimensional position, three-dimensional velocity, attitude, positioning accuracy, timestamp, and integrity indicators are transmitted to the UAV flight controller, mission planning system, or ground control station via a standard data interface in a low-latency manner for flight control, route planning, mission execution, and safety warning.

[0013] Another aspect of the present invention provides a drone positioning system, comprising:

[0014] The unmanned aerial vehicle (UAV) body; at least three external positioning beacons, wherein the external positioning beacons are UWB anchor points with ultra-wideband ranging capabilities, and their precise position coordinates are known and stored in a high-performance processing unit; an airborne positioning module mounted on the UAV body; the airborne positioning module includes at least:

[0015] An ultra-wideband transceiver is used to acquire ranging information between the UAV and the external positioning beacon;

[0016] An inertial measurement unit (IMU) is used to provide information on the motion status of the UAV over a short timescale.

[0017] The visual sensor module is used to provide the relative position and attitude information of the drone in the local environment;

[0018] Global Navigation Satellite System (GNSS) receiver module, used to provide the UAV's global absolute position information;

[0019] The high-performance processing unit interacts with the ultra-wideband transceiver, the inertial measurement unit, the visual sensor module, and the global navigation satellite system receiving module via a data interface.

[0020] In some embodiments, the high-performance processing unit is configured with at least:

[0021] An environment-adaptive data preprocessing unit is used to perform real-time quality assessment, outlier detection, and filtering on raw data from the ultra-wideband transceiver, the inertial measurement unit, the visual sensor module, and the global navigation satellite system receiving module.

[0022] The dynamic weighted fusion positioning engine is used to integrate preprocessed multimodal sensor data, estimate the UAV's three-dimensional position, three-dimensional velocity and attitude in real time, and dynamically adjust the weight of each modal sensor observation in the fusion algorithm based on the sensor data quality assessment results provided by the environment adaptive data preprocessing unit.

[0023] The environmental perception and situation recognition module is used to identify the current environment type and potential interference sources by utilizing data from the visual sensor module, the ultra-wideband transceiver, the global navigation satellite system receiving module, and the inertial measurement unit, and to dynamically adjust the motion model parameters used in the dynamic weight fusion positioning engine based on the identified environmental changes and the motion state of the UAV.

[0024] The robustness and integrity monitoring unit is used to monitor and evaluate the positioning results output by the dynamic weight fusion positioning engine in real time to ensure the reliability and availability of the positioning information.

[0025] In some embodiments, the inertial measurement unit includes at least:

[0026] The three-axis MEMS gyroscope has an angular rate measurement range of ±2000 degrees / second and a drift rate of less than 0.1 degrees / hour.

[0027] The triaxial MEMS accelerometer has a measurement range of ±16g and an bias stability of less than 10 microgravity accelerations.

[0028] The triaxial magnetometer has a measurement range of ±500 microtesla.

[0029] The inertial measurement unit outputs three-dimensional angular velocity, three-dimensional linear acceleration, and three-dimensional geomagnetic field strength data at a sampling rate of at least 200Hz. The three-dimensional angular velocity, three-dimensional linear acceleration, and three-dimensional geomagnetic field strength data are used to provide attitude changes, velocity increments, and position estimations of the UAV within a short time scale, and serve as prediction inputs for the dynamic weighted fusion positioning engine.

[0030] In some embodiments, the visual sensor module includes at least a high-resolution global shutter camera with an image resolution of 1280x1024 pixels, a frame rate of up to 200 frames per second, and a lens distortion coefficient of less than 0.005; the visual sensor module is connected to the high-performance processing unit through a high-speed digital interface to transmit image data in real time.

[0031] In some embodiments, the dynamic weighted fusion positioning engine is a state estimator based on nonlinear optimization; the state vector of the state estimator includes at least the position, velocity, and attitude of the UAV in the world coordinate system, the deviation term of the inertial measurement unit, and the position correction amount of the external ultra-wideband anchor point;

[0032] The prediction of the dynamic weighted fusion positioning engine is based on the data of the inertial measurement unit. By using the angular velocity and acceleration information of the inertial measurement unit and combining it with the motion model of the UAV, the state of the UAV at the next moment is predicted.

[0033] The dynamic weighted fusion positioning engine updates by using the ranging data from the ultra-wideband transceiver, the relative pose information output by the visual sensor module, and the absolute position information provided by the global navigation satellite system receiving module to correct the predicted state.

[0034] In some embodiments, the ultra-wideband observation update calculates the predicted ranging value based on the currently estimated UAV position and the known external ultra-wideband anchor point position, and uses the residual between the actual ranging value and the predicted ranging value as the observation update item;

[0035] The visual observation update utilizes at least the feature point reprojection error constructed from the relative pose increment of the visual odometry to correct the UAV's motion in the local coordinate system and converts it into a correction of the global state vector.

[0036] The Global Navigation Satellite System (GNSS) observation update uses the absolute position information of the GNSS as a global position observation to correct long-term drift and cumulative errors in the fused positioning results.

[0037] In some embodiments, the environmental perception and situation recognition module identifies the current environment type, including indoor, outdoor, urban canyon, open area, forest, or water surface, through image semantic segmentation and target detection algorithms; by analyzing the abnormal fluctuations of the ultra-wideband signal, the pseudorange or carrier phase jump of the global navigation satellite system, and the abnormal data of the inertial measurement unit, and in combination with the environmental information, it identifies potential interference sources; and dynamically adjusts the motion model parameters used in the dynamic weight fusion positioning engine according to the motion state of the UAV and the identified environmental changes.

[0038] In some embodiments, the global navigation satellite system receiving module includes at least a GNSS receiver chipset that supports L1 and L2 frequency bands and supports GPS, GLONASS, Galileo and BeiDou BDS constellations, as well as a high-precision multipath-suppressed GNSS antenna.

[0039] In some embodiments, the high-performance processing unit includes an embedded central processing unit, a graphics processing unit, and high-speed memory and non-volatile memory.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] This invention achieves complementary advantages among different modalities by fusing sensor data from multiple physical principles, including UWB, IMU, vision, and GNSS. Its UWB ranging is unaffected by optical conditions, possesses obstacle-penetrating capabilities, and provides stable data in adverse weather or non-line-of-sight environments. This invention requires only a relatively small number of easily deployable UWB anchor points, which can be small, battery-powered portable devices, significantly reducing infrastructure deployment costs, maintenance difficulties, and initial calibration accuracy requirements. Furthermore, this invention, through an environment-adaptive data preprocessing unit and a dynamic weighted fusion positioning engine, can evaluate the performance of each sensor in real time. The invention focuses on data quality, identifying environmental changes and potential interference sources, and intelligently adjusting sensor weights. It utilizes GNSS for global coarse positioning and UWB for fast, robust short-range acquisition, combined with the high-frequency update capabilities of IMU and visual sensors. This enables the UAV to quickly acquire positioning in any initial state. During high-dynamic maneuvers (such as sharp turns and rapid ascents and descents), continuous fusion and prediction of multimodal data effectively prevents positioning loss, achieving seamless and accurate dynamic tracking. The presence of a robustness and integrity monitoring unit allows the system to evaluate the reliability of positioning results in real time, providing integrity indicators. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart of the UAV positioning method of the present invention;

[0044] Figure 2 This is a schematic diagram of the unmanned aerial vehicle (UAV) positioning system of the present invention. Detailed Implementation

[0045] The following will refer to the appendices in the embodiments of the present invention. Figure 1-2 The technical solutions in the embodiments of the present invention will be clearly and completely described together. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0046] Application Overview:

[0047] This invention provides a UAV positioning method and system, which aims to overcome the deep-seated limitations of existing UAV positioning schemes in terms of adaptability to complex environments, anti-interference ability, real-time performance of multi-sensor data fusion, system robustness, and cost control. In particular, it can provide continuous, stable, and high-precision positioning information in environments with strong interference, dynamic changes, and unstructured environments. Example

[0048] Figure 1 The present invention provides a flowchart of a UAV positioning method. The execution flow of the method covers the complete process from raw data acquisition to final positioning result output, and the various steps are closely coordinated.

[0049] Multimodal sensor data acquisition: In this step, during flight, the UAV uses its onboard positioning module (UWB transceiver, IMU22, vision sensor module, and GNSS receiver module) to acquire raw data related to its own motion in real time and at high frequency. Specifically, the UWB transceiver continuously and actively engages in bidirectional ranging communication with pre-deployed external UWB anchor points to obtain precise ranging values ​​between the UAV and each anchor point, along with their associated signal quality parameters, such as Received Signal Strength Indication (RSSI) and Signal-to-Noise Ratio (SNR). Simultaneously, it acquires raw UWB pulse waveforms for subsequent multipath analysis. The IMU, with its inherent high frequency (e.g., 200Hz or higher), continuously acquires data on the UAV's angular velocity, linear acceleration, and geomagnetic field strength in three-dimensional space, reflecting the UAV's instantaneous motion state. The vision sensor module captures image sequences of the UAV's surrounding environment at a high frame rate (e.g., 100-200 frames / second), capturing rich visual features. Meanwhile, the GNSS receiver module continuously receives signals from multiple satellite constellations and calculates the UAV's global latitude and longitude, altitude, three-dimensional velocity, and precise GPS time information. It also provides integrity auxiliary data such as position accuracy factor (PDOP), number of available satellites, and RTK status. All acquired raw data is time-synchronized and transmitted to the high-performance processing unit for further processing.

[0050] Environmental adaptive data preprocessing, performed by the environmental adaptive data preprocessing unit within the high-performance processing unit after raw data acquisition, conducts real-time quality assessment, outlier detection, and filtering for various types of raw data. Specifically, for UWB ranging data, the preprocessing unit deeply analyzes the RSSI and SNR of each ranging value and combines this with characteristic analysis of the UWB pulse waveform to determine whether severe multipath effects exist. If the signal strength is below a preset threshold or the multipath effect index is too high, the ranging value will be marked as low-quality data, and its weight in subsequent fusion may be reduced or temporarily removed in extreme cases. For IMU raw data, the preprocessing unit uses attitude estimation algorithms (such as quaternion-based complementary filtering or extended Kalman filtering) to perform attitude calculation and real-time correction and compensation for sensor biases (such as gyroscope bias and accelerometer bias). Simultaneously, it suppresses high-frequency noise through low-pass filtering and other methods, thereby improving the accuracy of trajectory estimation.

[0051] For visual image data, rigorous distortion correction is first performed to eliminate inherent lens errors, followed by brightness equalization and noise suppression to optimize image quality. Crucially, the unit assesses the robustness and reliability of the current visual data in real time by evaluating the number of feature points extracted from the image (e.g., whether it meets the minimum number required for effective visual odometry), the uniformity of feature point distribution in the image, and the tracking success rate and reprojection error between consecutive frames. Visual positioning information is considered low-confidence when the number of feature points is too low, the distribution is uneven, or the tracking quality is poor. For GNSS data, the preprocessing unit monitors its output PDOP value, the number of available satellites, and RTK fixation status in real time. If the PDOP value is too high or the number of available satellites is too low, the confidence level of the GNSS data is reduced. Furthermore, after initial quality assessment, all sensor data undergoes final outlier identification and filtering using statistical principles (e.g., three standard deviations, i.e., the 3σ criterion) or geometric consistency methods (e.g., the RANSAC algorithm) to ensure that the data input to the fusion engine is clean and reliable. For example, for a set of GNSS location observations, if an observation point deviates from most observation points by more than 3 standard deviations, it is considered an outlier and removed.

[0052] Environmental perception and situational awareness, this step is performed by the environmental perception and situational awareness module in the high-performance processing unit, which aims to provide contextual information for fusion localization and achieve intelligent adaptation. The module uses pre-processed visual image data and advanced deep learning algorithms (e.g., semantic segmentation models based on convolutional neural networks) to perform pixel-level semantic classification of the UAV's current environment, thereby accurately identifying various environmental types such as indoor corridors, outdoor open areas, urban canyons, dense forests, water areas, or underground spaces.

[0053] Simultaneously, by combining analysis of abnormal fluctuations from UWB, GNSS, and IMU data—such as sudden and drastic jumps in UWB ranging values, abnormal interruptions in GNSS signal strength or carrier phase, and abnormal high-frequency vibrations or bias drift in IMU data—the system comprehensively identifies potential sources of electromagnetic interference, GNSS signal obstruction or spoofing, multipath effects, and IMU malfunctions, along with their severity. Based on the identified environmental type and the UAV's real-time motion state (e.g., determining whether the UAV is hovering, flying at a constant speed, or performing high-dynamic maneuvers by analyzing instantaneous velocity, acceleration, and angular velocity calculated from IMU data), the module dynamically selects or adjusts the state-space model and observation model parameters used in the subsequent dynamic weighted fusion positioning engine. For example, in open areas where GNSS data is typically highly reliable, the system can prioritize using a high-precision GNSS-assisted motion model; while in urban canyons or indoor environments where GNSS signals are limited, the system will focus on the fusion of UWB ranging and visual odometry, and may adopt a motion model more suitable for unstructured environments. This environmental awareness capability enables the system to predict and respond to various complex scenarios.

[0054] Dynamic weighted fusion localization, the core of this invention's localization method, is executed by the dynamic weighted fusion localization engine in the high-performance processing unit. Preprocessed and quality-assessed multimodal sensor data, combined with identified environmental situational information, is input into the engine. The engine first predicts the UAV's state at the next moment based on the current UAV state (including position, velocity, and attitude) and high-frequency data from the IMU, using a pre-set UAV motion model (dynamically adjusted according to environmental perception results). This prediction step is performed at a high frequency, providing continuity and smoothness in localization.

[0055] Subsequently, the engine updates its observations, revising the predicted state using ranging data from the UWB transceiver, relative pose increments output from the visual sensor module (visual odometry or SLAM results), and absolute position information provided by the GNSS receiver module. Crucially, the fusion positioning engine dynamically adjusts the weights of each modality's observations in the state update process in real time, based on the data quality assessment results of each sensor modality and the identified environmental situation. Specifically, a pre-trained fuzzy logic controller or deep neural network (e.g., a Long Short-Term Memory network, LSTM) is employed. The controller's input parameters include UWB's RSSI, SNR, and multipath effect index; the number of visual feature points; feature point tracking quality; image texture richness; ambient lighting conditions; GNSS's PDOP value; the number of available satellites; and RTK fixed status. Based on these inputs, the controller outputs dynamic weight coefficients for each sensor modality in real time. For example, when UWB signal quality is excellent and anchor point geometry is well-distributed, the weight of UWB will be increased; when the UAV enters a textured and well-lit environment, the weight of visual odometry will increase; while when GNSS signals are blocked or interfered with, their weight will be significantly reduced (even to near zero), while the weights of UWB and visual / IMU will be increased accordingly to ensure the continuity and accuracy of positioning. This dynamic weight adjustment mechanism enables the system to intelligently select and prioritize the use of the most reliable sensor data sources in any complex or changing scenario, thereby significantly improving the robustness and accuracy of positioning results.

[0056] Robustness and integrity monitoring, performed by the robustness and integrity monitoring unit within the high-performance processing unit, aims to monitor and evaluate the output fused positioning results in real time, ensuring the reliability and availability of the positioning information. This step includes multiple layers: First, cross-comparison of independent solution results from different sensor modes is performed to check multi-source data redundancy. For example, the independently calculated GNSS position is compared with the independently calculated UWB position and the output of the fused positioning engine to calculate the consistency error between them. Second, consistency checks are performed on the internal residuals of the fused positioning engine, such as through chi-square tests or statistical analysis of the residual sequence, to determine whether the residuals conform to the expected random distribution. If the residuals show systematic deviations or abnormal jumps, it indicates that there may be error sources not covered by the model or that a sensor has malfunctioned. When a sensor data source is detected to be continuously abnormal or does not conform to expectations, the unit will trigger a fault diagnosis and isolation mechanism. For example, if the ranging value of a UWB anchor point does not match the fusion result for a continuous long period of time, its weight will be reduced to zero, and its data will be excluded from the fusion process to prevent it from contaminating the final positioning result. Simultaneously, based on parameters such as the confidence ellipsoid size and estimated error covariance matrix of the fused positioning results, the system calculates and outputs integrity indicators in real time, including positioning accuracy (e.g., root mean square error RMSE), horizontal protection limit (HPL), and vertical protection limit (VPL). These indicators quantify the reliability and availability of positioning information. When the positioning integrity indicator falls below a preset safety threshold, the system can trigger an early warning and send a signal to the UAV flight controller, suggesting switching to a safe mode or performing specific emergency operations.

[0057] In the final step of positioning output, the high-performance processing unit's data transmission and output interface transmits the UAV's real-time 3D position (including latitude / longitude / elevation or XYZ coordinates), 3D velocity, attitude (quaternion or Euler angles), positioning accuracy, timestamp, and integrity indicators—all after robustness and integrity monitoring—to the UAV flight controller, mission planning system, or ground control station via a high-speed, standardized data interface (such as Ethernet, CAN, or a custom protocol) with low latency. This information forms the basis for precise flight control, autonomous route planning, efficient mission execution, and timely safety warnings, ensuring the safe, stable, and efficient operation of the UAV in various complex application scenarios. Example

[0058] Based on the same inventive concept as the UAV positioning method in Embodiment 1 above, such as Figure 2As shown, this invention also provides a UAV positioning system, the core of which is an onboard positioning module mounted on the UAV itself. This module works in conjunction with at least three pre-deployed external positioning beacons. The external positioning beacons, acting as UWB anchor points, have their precise position coordinates determined before system startup using high-precision measurement or self-calibration technology and stored in a high-performance processing unit within the onboard positioning module. These UWB anchor points provide accurate distance or time difference information, forming the basis for the UAV's local high-precision positioning.

[0059] The airborne positioning module integrates multiple key sensor units, including an ultra-wideband (UWB) transceiver, an inertial measurement unit (IMU), a visual sensor module, and a Global Navigation Satellite System (GNSS) receiver module. All these sensor units interact with the high-performance processing unit in real time via their respective high-speed data interfaces, ensuring the synchronous acquisition and efficient processing of massive amounts of sensor data. The high-performance processing unit, as the computational core of the entire system, is equipped with software functional modules such as an environment-adaptive data preprocessing unit, a dynamic weighted fusion positioning engine, an environmental perception and situational awareness module, a robustness and integrity monitoring unit, and data transmission and output interfaces. These modules work together to achieve accurate and robust positioning of the UAV.

[0060] Specifically, the ultra-wideband transceiver, a key component of the positioning system of this invention, operates within a carefully designed frequency range of 3.1 GHz to 10.6 GHz, fully utilizing the ultra-wideband characteristics of UWB signals to achieve centimeter-level ranging accuracy. The transceiver integrates a clock synchronization and data processing unit for implementing Two-Way Ranging (TWR) or Time Difference of Arrival (TDOA) protocols. In TWR mode, the UWB transceiver precisely measures the round-trip time of the signal by exchanging a series of pulse packets, thereby calculating the distance between the UAV and the external UWB anchor point. In TDoA mode, the precise time difference between the arrival times of signals from different anchor points is used to calculate the UAV's position. The ultra-wideband transceiver communicates wirelessly with the external UWB anchor point through a high-gain wideband omnidirectional antenna array. This antenna array is optimized to ensure effective reception and transmission of UWB signals in complex three-dimensional space, even in the presence of non-line-of-sight obstructions, by utilizing the signal diffraction and penetration capabilities to obtain usable ranging information. Its ranging accuracy is typically better than 0.1 meters, providing high-precision distance observation for subsequent fusion positioning. The ranging data output by the ultra-wideband transceiver contains several key parameters, specifically the accurate ranging value, the timestamp synchronized with the UAV system time, and real-time evaluated signal quality parameters, such as Received Signal Strength Indication (RSSI) and Signal-to-Noise Ratio (SNR). It also outputs demodulated and analyzed UWB pulse waveform characteristics and multipath effect index. These parameters are crucial for the subsequent environmental adaptive data preprocessing unit to perform data quality assessment.

[0061] The Inertial Measurement Unit (IMU) is fundamental for providing motion state information of a UAV. Manufactured using high-performance MEMS (Micro-Electro-Mechanical Systems) technology, the IMU features a high sampling rate and low drift characteristics. Internally, it includes a three-axis MEMS gyroscope to measure the UAV's angular rates in pitch, roll, and yaw axes, with a measurement range of ±2000 degrees / second and a drift rate strictly controlled to less than 0.1 degrees / hour, ensuring accurate attitude estimation over short periods. Simultaneously, the IMU integrates a three-axis MEMS accelerometer to measure the UAV's linear acceleration in the X, Y, and Z directions, with a measurement range of ±16g and bias stability better than 10 microgravity accelerations, accurately reflecting the UAV's translational motion. Furthermore, the IMU contains a three-axis magnetometer with a measurement range of ±500 microtesla, used to assist in heading correction, especially in environments where GNSS signals are interfered with or unavailable. The IMU continuously outputs three-dimensional angular velocity, three-dimensional linear acceleration, and three-dimensional geomagnetic field strength data at a sampling rate of at least 200Hz. This high-frequency data provides the dynamic weighted fusion positioning engine within the high-performance processing unit with information on the UAV's motion status over a short timescale, including attitude changes, velocity increments, and position estimation. It serves as the core input for the prediction steps of the fusion positioning engine, continuously estimating the UAV's position through inertial navigation principles, effectively compensating for the deficiencies of other sensors in terms of update rate or continuity.

[0062] The visual sensor module is key to this invention for environmental perception and local localization. This module is equipped with a high-resolution global shutter camera with an image resolution of up to 1280x1024 pixels, ensuring excellent image detail capture. The global shutter feature avoids the rolling shutter effect, ensuring clear and distortion-free images even during high-speed motion, which is crucial for visual odometry (VO) or Simultaneous Localization and Mapping (SLAM) algorithms. Its frame rate can reach 200 frames per second, providing rich, continuous image sequences and ample data flow for visual tracking and pose estimation in high-dynamic scenes. The lens distortion coefficient of this camera is strictly controlled to less than 0.005, ensuring geometric accuracy of the images and reducing the impact of image distortion on pose estimation. The visual sensor module connects to a high-performance processing unit via a high-speed digital interface (e.g., MIPI CSI-2 or LVDS) to transmit uncompressed or efficiently compressed image data in real time. After receiving image data, the high-performance processing unit runs preset visual odometry or SLAM algorithms, such as those based on ORB (Oriented Fast and Rotated BRIEF) features or SIFT (Scale-Invariant Feature Transform) features. These algorithms first extract a large number of salient and distinguishable feature points from the image, then perform feature matching between consecutive frames, and accurately estimate the UAV's relative position and attitude changes in the local environment through triangulation or least-squares-based, nonlinear optimization methods (such as Bundle Adjustment). This relative pose information serves as an important observation input for the dynamic weight fusion positioning engine, providing crucial local positioning capabilities, especially in indoor or urban canyon environments where GNSS signals are limited.

[0063] The Global Navigation Satellite System (GNSS) receiver module provides a global absolute position reference for this invention. At its core is a GNSS receiver chipset that supports multiple frequencies (e.g., L1 / L2, L5) and multiple constellations (e.g., the US GPS, Russia's GLONASS, the EU's Galileo, and China's BeiDou BDS). By receiving signals from multiple satellite systems, the GNSS receiver module significantly improves the number of available satellites and positioning accuracy in complex, obstructed environments. It is paired with a high-precision multipath suppression GNSS antenna, which, through sophisticated structural design and signal processing technology, effectively suppresses multipath effects generated in complex environments such as urban high-rise buildings and surface water, thereby improving the accuracy of pseudorange and carrier phase measurements. The GNSS receiver module outputs the UAV's global latitude and longitude, altitude, three-dimensional velocity, and precise time information. Furthermore, it provides several integrity parameters, such as the Position Dilution of Precision (PDOP), the number of available satellites, and the carrier phase differential (RTK) status (Fixed, Float, or Single Point Positioning). These parameters are crucial for the environment-adaptive data preprocessing unit and the robustness and integrity monitoring unit to assess the reliability and confidence of GNSS data. GNSS data, as the global absolute position observation of the fusion positioning engine, is mainly used to correct the drift accumulated during long-term trajectory extrapolation from other sensors (especially IMUs), ensuring high-precision positioning of the UAV in the global coordinate system.

[0064] The high-performance processing unit (CPU) is the computational and decision-making hub of the UAV positioning system. Its hardware configuration is designed to meet the demands of real-time processing of multimodal sensor data and execution of complex algorithms, typically including one or more high-performance embedded central processing units (CPUs), graphics processing units (GPUs), or field-programmable gate arrays (FPGAs). The CPU is responsible for scheduling system tasks, running the operating system, and executing general-purpose algorithms; the GPU excels in parallel computing and is suitable for image processing, feature extraction, and deep learning inference; while the FPGA provides highly customizable hardware acceleration capabilities, particularly suitable for high-concurrency signal processing and low-latency control logic. These processors work together, communicating with all sensor units via high-speed bus interfaces (e.g., PCIe or Gigabit Ethernet). The CPU is also equipped with large-capacity high-speed memory, such as DDR4 synchronous dynamic random access memory, for caching raw sensor data and intermediate algorithm results; and non-volatile memory (eMMC or SSD) for storing system firmware, preset parameters, map data, and historical logs. The software functional modules running on the CPU constitute the core intelligence of the positioning scheme of this invention.

[0065] The function of the environment adaptive data preprocessing unit is to perform real-time quality assessment, outlier detection and filtering on raw data from UWB transceivers, IMUs, visual sensor modules and GNSS receiver modules, so as to provide a clean and reliable data source for the subsequent fusion positioning engine.

[0066] Specifically, for UWB ranging data, the preprocessing unit evaluates the reliability of the ranging value based on the real-time received RSSI, SNR, UWB pulse waveform characteristics (e.g., pulse width, peak shape), and multipath effect index (quantifying multipath intensity by analyzing the characteristics of the UWB pulse response signal, such as energy attenuation and delay distribution). If the RSSI is lower than a preset threshold (e.g., -85 dBm) or the SNR is too low (e.g., below 5 dB), the ranging value is marked as of poor quality, possibly caused by excessive distance or occlusion. If the multipath effect index is too high, it indicates severe multipath interference. In this case, the unit will perform corresponding corrections on the ranging value based on a multipath model (e.g., reflection path attenuation model), or reduce its weight in fusion positioning, or even temporarily exclude it.

[0067] Furthermore, for IMU data, the preprocessing unit performs attitude estimation and bias correction on the raw outputs of the gyroscope and accelerometer by integrating advanced attitude estimation algorithms (e.g., quaternion-based complementary filtering or extended Kalman filtering). This unit also performs real-time estimation and compensation for sensor bias and noise. For example, through statistical analysis of data under long-term static or low-dynamic conditions, the constant bias of the gyroscope and accelerometer is estimated and eliminated; simultaneously, adaptive noise estimation techniques (e.g., based on Allan variance analysis or power spectral density analysis) are used to assess the sensor noise level in real time, and high-frequency random noise is suppressed through filtering algorithms (such as Butterworth filters or Wiener filters) to effectively suppress cumulative drift during trajectory extrapolation.

[0068] For visual sensor data, the preprocessing unit performs rigorous distortion correction, brightness equalization, and noise suppression on the captured raw images. Distortion correction removes geometric distortion from the image using pre-calibrated camera parameters (e.g., radial and tangential distortion coefficients). Brightness equalization improves the visual quality of the image under different lighting conditions through histogram equalization or adaptive contrast enhancement algorithms. Noise suppression typically employs Gaussian filtering, median filtering, or nonlocal mean filtering to smooth the image and remove random noise. More importantly, the unit assesses the robustness of the current visual data by evaluating the number of feature points in the image (e.g., requiring more than 50 ORB feature points), their uniformity of distribution (e.g., feature points should be evenly distributed across the four quadrants of the image, avoiding concentration in a certain area), and the tracking quality of feature points across consecutive frames (e.g., based on a reprojection error of less than 2 pixels or optical flow consistency checks). When the number of feature points is less than a preset threshold or the tracking failure rate is too high (e.g., a tracking success rate of less than 70% for five consecutive frames), the system marks the visual localization information as low confidence and notifies the dynamic weight fusion localization engine to adjust its weights accordingly.

[0069] Furthermore, for GNSS data, the preprocessing unit monitors in real time the PDOP value, number of available satellites, carrier phase residual, and RTK fixed / floating / single-point status output by the GNSS receiver module. If the PDOP value exceeds a preset threshold (e.g., 3.0) or the number of available satellites is too low (e.g., less than 4 satellites), the confidence level of the GNSS data is reduced. The preprocessing unit also employs methods based on statistical principles (e.g., the 3σ criterion) or geometric consistency (e.g., the RANSAC algorithm, random sampling consistency) to identify and filter out outliers in the raw data output by various sensors. For example, for UWB ranging data, if a ranging value deviates from the expected distance obtained through other sensors (such as IMU estimation or visual estimation) by more than 3 standard deviations, it is considered an outlier and removed. The RANSAC algorithm can be used to identify and remove outlier matching pairs during visual feature point matching, ensuring the accuracy of pose estimation. This multi-level preprocessing mechanism ensures high reliability and consistency of the data input to the fusion positioning engine, effectively preventing erroneous data from contaminating the final positioning results.

[0070] The dynamic weighted fusion positioning engine is the core of this invention for achieving high-precision and robust positioning. Its function is to integrate the quality-assessed sensor data from various modalities provided by the environment adaptive data preprocessing unit to estimate the UAV's 3D position, 3D velocity, and attitude in real time. The core algorithm of the fusion positioning engine is either a state estimator based on nonlinear optimization (e.g., factor graph optimization) or a model-based state estimator (e.g., Extended Kalman Filter (E-KF), Unscented Kalman Filter (U-KF), or Particle Filter).

[0071] In most application scenarios, the Extended Kalman Filter (EKF) is the preferred choice due to its computational efficiency and excellent performance. Its state vector is carefully designed to include not only the UAV's position (X, Y, Z), three-dimensional velocities (Vx, Vy, Vz), and attitude (typically represented by quaternions such as roll, pitch, yaw) in the world coordinate system, but also extended to include IMU bias terms (gyroscope bias, accelerometer bias) and position corrections for external UWB anchors that may occur during long-term operation. The inclusion of these error states allows the system to calibrate sensor errors and environmental parameters online, further improving positioning accuracy and robustness.

[0072] The prediction step of the fusion positioning engine is based on IMU data. Utilizing the angular velocity and linear acceleration information output from the IMU at high frequencies, combined with a pre-defined UAV motion model (e.g., a uniform velocity model for slow, stable flight, a uniform acceleration model for high-speed maneuvers, or a more complex dynamic model), the UAV's state at the next moment is accurately predicted. This prediction process is continuously performed at a high frequency (e.g., 200Hz) to ensure the continuity and smoothness of the positioning trajectory.

[0073] The update step of the fusion positioning engine then uses UWB ranging data, relative pose information output by visual odometry, and absolute position information provided by GNSS to correct the predicted state. Specifically:

[0074] Upon receiving UWB ranging data, the engine calculates the predicted ranging value based on the currently estimated UAV position and the precise positions of the UWB anchor points known and stored in the high-performance processing unit. Subsequently, the residual between the actual and predicted ranging values ​​(the difference between the measured and estimated values) is used as an observation update term, and the UAV's state estimate is corrected through Kalman gain weighting. This process considers the noise characteristics and multipath effects of UWB ranging.

[0075] Furthermore, when receiving relative pose increments from visual odometry (e.g., relative translation and rotation of the UAV between two frames) or feature point reprojection errors constructed by SLAM (the deviation between the observed feature point positions and the reprojected feature point positions estimated through the current pose), the engine utilizes these observations to correct the UAV's motion in the local coordinate system. These local updates are then translated into corrections to the global state vector, particularly addressing position and attitude errors accumulated during IMU extrapolation. Visual observations provide high-precision relative pose information, especially in environments with GPS denial or severe multipath propagation.

[0076] Simultaneously, upon receiving GNSS absolute position information, the engine treats it as a global position observation. The latitude, longitude, altitude, and velocity information provided by GNSS are aligned with the world coordinate system within the fusion positioning engine after coordinate transformation (e.g., conversion to a geocentric Earth-fixed coordinate system or a local northeast-sky coordinate system), and are used to correct long-term drift and accumulated errors in the fusion positioning results. Even if the GNSS signal is interrupted for a short period, the fusion of IMU and visual / UWB can maintain positioning, and the absolute position information can be used to recalibrate the system after the GNSS signal is restored.

[0077] The key to the fusion positioning engine lies in its dynamic weight allocation mechanism. This mechanism adjusts the weights of each modal sensor observation in the fusion algorithm in real time based on the data quality assessment results provided by the environment adaptive data preprocessing unit and the environmental situation identified by the environment perception and situation recognition module. Specifically, the dynamic weight allocation mechanism is implemented through an adaptive module based on a fuzzy logic control system or a deep learning model (e.g., a pre-trained Long Short-Term Memory network LSTM). Taking the fuzzy logic control system as an example, the input parameters of this module include: RSSI, SNR, and multipath effect exponent of the UWB signal; noise levels and bias stability of the IMU's gyroscope and accelerometer; the number of feature points captured by the visual sensor, the success rate of feature point tracking, image texture richness (e.g., quantified by calculating the gray-level co-occurrence matrix features or the frequency domain characteristics of the Fourier transform), and ambient lighting conditions (e.g., determined by the average brightness or histogram distribution of the image); GNSS PDOP value, number of available satellites, carrier phase residual, and RTK fixed state. These input parameters are fuzzified into fuzzy sets such as "low," "medium," and "high," and inference is performed using a pre-defined fuzzy rule base (e.g., "If the UWB signal quality is high and the multipath effect is low, then the UWB weight is high"; "If the GNSS signal is occluded and the visual texture is rich, then the GNSS weight is reduced and the visual weight is increased"). The inference results are defuzzified, and the weight coefficients of each sensor mode at the current time are output. The dynamic adjustment range of these weight coefficients is from 0 to 1.

[0078] For example, when a drone flies in an open area with strong GNSS signals, low PDOP values, and RTK in FIX mode, the weight of GNSS is significantly increased (e.g., close to 1.0) to ensure global absolute accuracy. When the drone enters an urban canyon, GNSS signals are affected by multipath and obstruction, and the PDOP value increases, the weight of GNSS is automatically decreased (e.g., reduced to 0.3). Simultaneously, the weight of UWB is increased based on its ranging signal quality (e.g., RSSI and SNR) and anchor point geometry (e.g., DOP value), and the weight of visual odometry may also increase if the current environment has rich texture and sufficient lighting. Conversely, if the drone flies indoors, the weight of GNSS may drop to 0, and UWB and vision become the primary positioning methods. In this case, the geometrical factor of accuracy (GDOP) of UWB ranging and the tracking stability of visual features will determine their weights. If the UWB signal suffers from severe multipath due to obstacles or some anchor points are completely obstructed, its weight may decrease, while the weights of vision and IMU are increased accordingly. If the drone suddenly enters an environment with insufficient lighting or lack of texture (e.g., flying near a monochrome wall), the visual weight will be reduced, while the weights of IMU and UWB will be increased accordingly. This dynamic weight adjustment mechanism ensures that the system always prioritizes the most reliable sensor data under different environmental conditions and interference scenarios, achieving a smooth transition and robust improvement in positioning performance, and preventing the failure of the entire positioning system due to the failure or performance degradation of a single sensor.

[0079] The environmental perception and situational awareness module utilizes visual sensors to perform real-time analysis of the UAV's surrounding environment. Combined with data from other sensors, it identifies the current environment type and potential sources of interference, thus providing environmental context information for the dynamic weighted fusion localization engine. Specifically, this module employs image semantic segmentation and object detection algorithms, typically using deep learning models based on convolutional neural networks (CNNs) (e.g., SegNet, U-Net, or YOLO series). These models are pre-trained on large datasets of labeled environmental images, enabling them to perform pixel-level classification of captured images and identify specific environmental types such as indoors, outdoors, urban canyons, open areas, forests, water surfaces, and tunnels. For example, it determines the current environment by recognizing features such as buildings, roads, trees, and water bodies in the image. Simultaneously, this module analyzes abnormal fluctuations in UWB signals (e.g., sudden drops in RSSI or jumps in ranging values, which may indicate that the UWB anchor point is blocked or subject to local interference), GNSS pseudorange or carrier phase jumps (which may indicate GNSS signal spoofing, denial, or strong multipath effects), and IMU data anomalies (e.g., sudden high-frequency vibrations or unexpected offset drift, which may indicate sensor failure or UAV structural resonance), combined with identified environmental information, to comprehensively determine the specific nature and extent of potential electromagnetic interference, multipath effects, GNSS spoofing, or denial-of-service interference sources. The module will also dynamically adjust the motion model parameters used in the fusion positioning engine based on the UAV's motion state (e.g., instantaneous velocity, acceleration, and angular velocity calculated from IMU data to determine whether the UAV is hovering, flying at a constant speed, accelerating rapidly, making a sharp turn, or in free fall) and identified environmental changes. For example, in hovering, the motion model can be simplified to a random walk model with small position changes; in high-speed level flight, a constant speed model is used; and in sharp turns, a motion model that considers centripetal acceleration is activated to more accurately describe the real-time dynamics of the UAV and improve the accuracy of the prediction step.

[0080] The robustness and integrity monitoring unit functions to monitor and evaluate the positioning results output by the dynamic weighted fusion positioning engine in real time, ensuring the reliability and availability of positioning information and providing safety assurance for UAV flight. This monitoring unit is implemented through technologies such as multi-source data redundancy verification, consistency checking, fault diagnosis and isolation, and positioning integrity assessment.

[0081] Specifically, in terms of multi-source data redundancy verification, the unit cross-compares positioning results from different sensor modalities (e.g., positioning results independently calculated by UWB, pose independently output by visual odometry, and absolute position independently calculated by GNSS). By calculating the consistency error between them (e.g., the Euclidean distance or angle difference between each independent calculation result and the fused result), it determines whether there are significant deviations in the data from each sensor. If the deviation exceeds a preset threshold, there may be sensor malfunction or environmental anomaly.

[0082] Regarding consistency checks, the unit continuously monitors the statistical characteristics of the residuals within the fusion positioning engine. For example, it uses the chi-square test or standard deviation analysis to determine whether the residuals conform to the expected random distribution (i.e., whether the mean of the residuals is close to zero and whether the variance is consistent with the estimated noise covariance). If the residuals exhibit systematic deviations (e.g., consistently positive or negative, or abnormally increased fluctuations) or anomalous spikes, it indicates the possible existence of error sources not covered by the model, sensor malfunctions, or environmental model mismatches.

[0083] When such anomalies are detected, the unit will initiate a fault diagnosis and isolation mechanism. When the data quality of a sensor consistently falls below the threshold set by the environment adaptive data preprocessing unit, or its residuals during the fusion process consistently exceed the expected range, the unit can promptly diagnose the sensor's fault. For example, if UWB ranging values ​​deviate significantly from expected values ​​for multiple consecutive frames, or visual feature point tracking fails continuously, the sensor is considered potentially faulty. In this case, the unit sends a signal to the dynamic weight allocation mechanism to reduce the weight of the faulty sensor to near zero (e.g., 0.01), thereby isolating its data from the fusion process and preventing it from contaminating the final positioning result. This mechanism ensures that the system can still provide usable positioning services even when some sensors fail.

[0084] Simultaneously, the robustness and integrity monitoring unit generates and outputs positioning integrity indices in real time based on the confidence ellipsoid size of the fused positioning results, the position accuracy level (e.g., horizontal and vertical accuracy at 95% confidence), and the system health status. Integrity indices include the Horizontal Protection Level (HPL) and the Vertical Protection Level (VPL), which represent the maximum boundaries of horizontal and vertical position errors at a specified confidence level, respectively. These indices quantify the reliability of the positioning results. When the positioning integrity indices (e.g., HPL or VPL) fall below a preset safety threshold, the system can trigger an early warning signal to the flight controller or automatically switch to a safer flight mode (e.g., hovering, return to home, or forced landing) to ensure the safe operation of the UAV.

[0085] The data transmission and output interface functions to achieve high-speed transmission and standardized output of raw sensor data, preprocessed data, and fused positioning results, ensuring that UAV positioning information can be acquired in real time and accurately by the flight controller, mission planning system, or other external devices. The interface includes, but is not limited to: one or more high-speed serial communication interfaces (e.g., LVDS or MIPI CSI-2) dedicated to the raw transmission of visual image data to meet the high bandwidth requirements of the visual sensor module. It also includes one or more SPI (Serial Peripheral Interface) or I2C (Inter-Integrated Circuit) interfaces for the transmission of control commands to the IMU and UWB transceivers and the transmission of low-speed data (such as IMU attitude data and UWB ranging values). Furthermore, a high-speed Ethernet interface or a custom bus (e.g., PCIe-based) is configured for high-speed data exchange between software modules within the high-performance processing unit and for external communication with the UAV flight controller, mission planning system, or ground control station. The interface outputs the UAV's real-time 3D position (selectable latitude / longitude or XYZ coordinates), 3D velocity, attitude (typically quaternions or Euler angles to reduce singularity), positioning accuracy (e.g., horizontal and vertical root mean square errors), timestamp (synchronized with system time), and positioning integrity metrics (HPL / VPL) using standard protocols (e.g., NMEA protocol, ROS message format, or custom binary protocol). All data transmissions are optimized for low latency, ensuring timely response from the flight control system and maintaining flight stability and mission execution accuracy.

[0086] In one specific embodiment, the UAV positioning system of the present invention was deployed for testing in a simulated urban canyon environment. The test area was approximately 200 meters long and 50 meters wide, flanked by buildings approximately 50 meters high, with a few trees and lampposts in the middle. Within this area, five UWB anchor points were pre-deployed, their coordinates measured via RTK-GNSS and stored in a high-performance processing unit, with an anchor point deployment density of one every 20 meters. The UWB transceiver, IMU, visual sensor module, and GNSS receiver module in the UAV's onboard positioning module were all configured according to the aforementioned specifications.

[0087] The test drone took off from an open area, entered an urban canyon, flew an "S"-shaped route within the canyon, and then flew out of the canyon back to the takeoff point. Throughout the flight, the drone transmitted its location data in real time.

[0088] During takeoff, the UAV is located in an open area with strong GNSS signal, a stable PDOP value of around 1.2, and more than 15 available satellites. RTK remains in FIX mode throughout. At this time, the environmental perception and situational awareness module identifies the area as "open." The dynamic weighted fusion positioning engine, through a fuzzy logic controller, adjusts the GNSS weight to 0.95, while the UWB and vision weights are 0.03 and 0.02, respectively. The IMU weight is fixed at 0.9, primarily used for high-frequency prediction and short-term attitude estimation. The horizontal root mean square error (RMSE) of positioning accuracy is approximately 0.02 meters, and the vertical RMSE is approximately 0.03 meters.

[0089] When the UAV entered the urban canyon area, tall buildings severely obstructed GNSS signals and caused multipath effects. The environmental perception and situational awareness module quickly identified the "urban canyon" environment. At this time, the GNSS PDOP value soared to 4.5, the number of available satellites plummeted to 5, and the RTK status frequently switched from FIX to FLOAT and even SPP. The environmental adaptive data preprocessing unit assessed that the GNSS data quality had significantly deteriorated. Simultaneously, the line-of-sight between the UWB transceiver and some UWB anchor points was obstructed, but the UWB signal, due to its penetrating ability, could still acquire ranging information through non-line-of-sight paths. Some ranging values ​​showed slight multipath effects, and RSSI and SNR decreased slightly. In the urban canyon, the visual sensor module, due to the rich texture and feature points provided by the tall buildings and roads (with the number of feature points consistently above 200), maintained a tracking success rate above 95%.

[0090] In this complex environment, the dynamic weight allocation mechanism of the dynamic weight fusion positioning engine plays a crucial role. The fuzzy logic controller rapidly adjusts the weights based on real-time sensor quality parameters: the GNSS weight is significantly reduced to 0.2; the UWB weight is adjusted to 0.4 (slightly decreased due to non-line-of-sight) based on the number of available anchor points, RSSI, SNR, and multipath effect index; while the visual odometry weight is significantly increased to 0.35 due to rich environmental texture. The IMU weight remains at 0.9, continuing to provide short-term, high-frequency updates. The robustness and integrity monitoring unit continuously monitors the fusion positioning residuals. When an abnormal jump occurs in a GNSS pseudorange, it is immediately marked as an anomalous observation, and its impact is reduced or temporarily removed during the fusion process. At this stage, the positioning accuracy of the system remains at 0.15 meters in the horizontal direction, 0.20 meters in the vertical direction, 0.5 meters in the HPL, and 0.8 meters in the VPL, ensuring stable flight of the UAV in the complex canyon environment.

[0091] When the drone flew out of the canyon and re-entered the open area, the GNSS signal was restored, and the number of PDOPs and satellites returned to normal. The system weights automatically and smoothly adjusted back to a GNSS-dominant configuration, and the positioning accuracy quickly recovered to within 0.03 meters. Throughout the entire process, the continuity and accuracy of the positioning results were effectively guaranteed.

[0092] In contrast, the inventors presented a drone positioning system based on LiDAR SLAM (LiDAR-SLAM) and tested it in the same urban canyon environment as the embodiments described above. This LiDAR SLAM system employs a high-precision multi-line LiDAR that estimates the drone's pose by scanning the surrounding environment and performing feature matching and point cloud registration.

[0093] During takeoff in open areas, LiDAR SLAM systems can provide relatively high positioning accuracy, with a horizontal RMSE of approximately 0.05 meters and a vertical RMSE of approximately 0.08 meters. However, when drones enter urban canyons, especially in areas with direct sunlight, glass curtain walls, or road surface water reflections, the LiDAR echo signal is subject to strong interference. The laser beam may experience specular reflection, leading to missing point cloud data or numerous false features; strong light exposure may saturate the LiDAR receiver, preventing it from acquiring effective depth information. In these situations, LiDAR SLAM systems are prone to the following problems:

[0094] Sparse or distorted point clouds: In areas with strong light or many reflective surfaces, the quality of the acquired point cloud data deteriorates, making feature point extraction difficult and leading to matching failures or errors.

[0095] Localization loss: Due to insufficient point cloud features or incorrect matching, the LIDAR SLAM algorithm may be unable to continuously track the drone's pose, causing the localization system to lose lock and fail to provide continuous localization information.

[0096] Poor robustness: LiDAR systems are sensitive to ambient light and surface materials. In the aforementioned complex environments, their positioning accuracy drops sharply or even fails completely. In our tests, when the drone flew into a highly reflective area in an urban canyon, the LiDAR SLAM system experienced three positioning losses, each lasting more than 5 seconds. These losses required manual intervention or switching to a backup mode (such as a barometer / IMU for rough estimation), severely impacting flight stability and safety.

[0097] Based on the foregoing, the following table details the performance indicators of the system of the present invention and the comparative system under different flight environments:

[0098]

[0099] The comparative data clearly shows that in open areas, both systems can provide relatively accurate positioning, but the system of this invention is slightly superior. However, in challenging urban canyon environments, especially under conditions of strong reflection or intense sunlight, the performance of a single-modal LiDAR SLAM system deteriorates sharply, positioning errors increase significantly, and positioning availability is extremely low, with frequent positioning loss, making it completely unsuitable for UAV missions. In contrast, the dynamic weighted multimodal fusion positioning system of this invention intelligently fuses data from multiple sensors such as UWB, IMU, vision, and GNSS, and dynamically adjusts the weights of each sensor according to the environment. Even in extremely complex urban canyon environments, it can maintain high positioning accuracy and excellent positioning availability, significantly improving the system's robustness and environmental adaptability. This provides UAVs with continuous, stable, and high-precision positioning information, greatly expanding the application scenarios and flight safety boundaries of UAVs.

[0100] In addition, the system of the present invention has a significant advantage in the initial acquisition time of the system, and can provide positioning for the UAV more quickly.

[0101] In summary, the present invention significantly surpasses existing positioning schemes based on a single mode.

[0102] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0103] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for locating unmanned aerial vehicles (UAVs), characterized in that, include: During flight, the UAV collects raw data related to its own motion in real time through the UWB transceiver, IMU, visual sensor module and GNSS receiver module of the onboard positioning module; Real-time quality assessment, outlier detection, and filtering are performed on the collected raw data; Using preprocessed visual image data, the current environment of the UAV is semantically segmented and classified through deep learning algorithms to identify the environment type; at the same time, by combining the abnormal fluctuations of UWB, GNSS and IMU data, potential interference sources are identified; based on the identified environment type and the real-time motion state of the UAV, the state space model and observation model parameters used for subsequent fusion positioning are dynamically selected. The preprocessed and quality-assessed sensor data is input into the dynamic weighted fusion localization engine. The fusion localization engine dynamically adjusts the weight of each modality observation in the state update process in real time based on the data quality assessment results of each sensor modality and the identified environmental situation. Real-time monitoring and evaluation of the output fusion positioning results; The UAV's three-dimensional position, three-dimensional velocity, attitude, positioning accuracy, timestamp, and integrity indicators, after robustness and integrity monitoring, will be transmitted to the UAV flight controller, mission planning system, or ground control station via a standard data interface in a low-latency manner for flight control, route planning, mission execution, and safety warning. Among them, the integrity indicators include the horizontal protection limit (HPL) and the vertical protection limit (VPL).

2. A drone positioning system, characterized in that, include: The drone itself; At least three external positioning beacons, which are UWB anchor points with ultra-wideband ranging capabilities, and their precise location coordinates are known and stored in a high-performance processing unit; An onboard positioning module mounted on the drone body; The airborne positioning module includes at least: An ultra-wideband transceiver is used to acquire ranging information between the UAV and the external positioning beacon; An inertial measurement unit (IMU) is used to provide information on the motion status of the UAV over a short timescale. The visual sensor module is used to provide the relative position and attitude information of the drone in the local environment; Global Navigation Satellite System (GNSS) receiver module, used to provide the UAV's global absolute position information; The high-performance processing unit interacts with the ultra-wideband transceiver, the inertial measurement unit, the visual sensor module, and the global navigation satellite system receiving module through a data interface. The high-performance processing unit is configured with at least: An environment-adaptive data preprocessing unit is used to perform real-time quality assessment, outlier detection, and filtering on raw data from the ultra-wideband transceiver, the inertial measurement unit, the visual sensor module, and the global navigation satellite system receiving module. The dynamic weighted fusion positioning engine is used to integrate preprocessed multimodal sensor data, estimate the UAV's three-dimensional position, three-dimensional velocity and attitude in real time, and dynamically adjust the weight of each modal sensor observation in the fusion algorithm based on the sensor data quality assessment results provided by the environment adaptive data preprocessing unit. The environmental perception and situation recognition module is used to identify the current environment type and potential interference sources by utilizing data from the visual sensor module, the ultra-wideband transceiver, the global navigation satellite system receiving module, and the inertial measurement unit, and to dynamically adjust the motion model parameters used in the dynamic weight fusion positioning engine based on the identified environmental changes and the motion state of the UAV. The robustness and integrity monitoring unit is used to monitor and evaluate the positioning results output by the dynamic weight fusion positioning engine in real time to ensure the reliability and availability of the positioning information.

3. The system according to claim 2, characterized in that, The inertial measurement unit includes at least: The three-axis MEMS gyroscope has an angular rate measurement range of ±2000 degrees / second and a drift rate of less than 0.1 degrees / hour. The triaxial MEMS accelerometer has a measurement range of ±16g and an bias stability of less than 10 microgravity accelerations. The triaxial magnetometer has a measurement range of ±500 microtesla. The inertial measurement unit outputs three-dimensional angular velocity, three-dimensional linear acceleration, and three-dimensional geomagnetic field strength data at a sampling rate of at least 200Hz. The three-dimensional angular velocity, three-dimensional linear acceleration, and three-dimensional geomagnetic field strength data are used to provide attitude changes, velocity increments, and position estimations of the UAV within a short time scale, and serve as prediction inputs for the dynamic weighted fusion positioning engine.

4. The system according to claim 2, characterized in that, The visual sensor module includes at least a high-resolution global shutter camera with an image resolution of 1280x1024 pixels, a frame rate of up to 200 frames per second, and a lens distortion coefficient of less than 0.

005. The visual sensor module is connected to the high-performance processing unit through a high-speed digital interface to transmit image data in real time.

5. The system according to claim 2, characterized in that, The dynamic weighted fusion positioning engine is a state estimator based on nonlinear optimization; the state vector of the state estimator includes at least the position, velocity, and attitude of the UAV in the world coordinate system, the deviation term of the inertial measurement unit, and the position correction amount of the external ultra-wideband anchor point; The prediction of the dynamic weighted fusion positioning engine is based on the data of the inertial measurement unit. By using the angular velocity and acceleration information of the inertial measurement unit and combining it with the motion model of the UAV, the state of the UAV at the next moment is predicted. The dynamic weighted fusion positioning engine updates by using the ranging data from the ultra-wideband transceiver, the relative pose information output by the visual sensor module, and the absolute position information provided by the global navigation satellite system receiving module to correct the predicted state.

6. The system according to claim 5, characterized in that, The update of the ultra-wideband transceiver ranging data is to calculate the predicted ranging value based on the currently estimated UAV position and the known position of the external ultra-wideband anchor point, and use the residual between the actual ranging value and the predicted ranging value as the observation update item. The relative pose information update output by the visual sensor module uses at least the feature point reprojection error constructed from the relative pose increment of the visual odometry to correct the motion of the UAV in the local coordinate system and convert it into a correction of the global state vector. The Global Navigation Satellite System (GNSS) observation update uses the absolute position information of the GNSS as a global position observation to correct long-term drift and cumulative errors in the fused positioning results.

7. The system according to claim 2, characterized in that, The environmental perception and situation recognition module identifies the current environment type, including indoor, outdoor, urban canyon, open area, forest or water surface, through image semantic segmentation and target detection algorithms; it identifies potential interference sources by analyzing abnormal fluctuations in ultra-wideband signals, pseudorange or carrier phase jumps of the global navigation satellite system, and abnormal data from the inertial measurement unit, combined with environmental information; and it dynamically adjusts the motion model parameters used in the dynamic weight fusion positioning engine based on the UAV's motion state and the identified environmental changes.

8. The system according to claim 2, characterized in that, The global navigation satellite system receiver module includes at least a GNSS receiver chipset that supports L1 and L2 frequency bands and supports GPS, GLONASS, Galileo and BeiDou BDS constellations, as well as a high-precision multipath suppression GNSS antenna.

9. The system according to claim 2, characterized in that, The high-performance processing unit includes an embedded central processing unit, a graphics processing unit, and high-speed and non-volatile memory.

Images