Adaptive multi-source fusion navigation method and system based on dynamic scene perception

Abstract

The application provides a kind of adaptive multi-source fusion navigation method and system based on dynamic scene perception.The method comprises: correcting error of inertial measurement data, obtaining corrected inertial data, and inputting the corrected inertial data into an inertial navigation mechanical arrangement module to generate a navigation reference benchmark;Based on the observation data of the reference station and the mobile station, the double-insert observation value is obtained, the double-difference geometric distance between the stations and the stars is calculated using the position information output by the inertial navigation system, and the residual data between the double-insert observation value and the double-difference geometric distance is obtained;The residual data is input into a Kalman filter, and the optimal estimate value is output by the Kalman filter;The optimal estimate value is fed back to the inertial measurement unit for compensation, and the navigation reference benchmark is corrected to output the final navigation data.Under the harsh conditions of GNSS long-term or frequent interruption, and vision or laser radar failure, the application can still realize the fusion of high-precision, low-divergence and strong-robust positioning and safe obstacle avoidance for low-altitude aircraft.

Description

Technical Field

[0001] This application belongs to the field of navigation and positioning technology, and in particular relates to the field of unmanned aerial vehicle (UAV) navigation and positioning technology. Background Technology

[0002] One existing multi-source fusion navigation architecture utilizes the high data output of the Inertial Measurement Unit (IMU) to compensate for the low update frequency of GNSS (Global Navigation Satellite System) and performs track estimation when GNSS signals are briefly lost. Through optimal estimation algorithms such as Kalman filtering, GNSS and IMU data are tightly coupled at the raw data level to improve positioning accuracy and anti-interference capability. This scheme essentially relies on GNSS to provide absolute position information to correct the accumulated errors of the IMU. In complex low-altitude environments (such as under overpasses or dense building clusters), if GNSS experiences prolonged or frequent interruptions, the error of pure inertial navigation will accumulate rapidly over time (commonly known as time divergence), causing the positioning error to exceed the safe flight threshold within seconds. Furthermore, this scheme only provides position and attitude, lacking the ability to perceive the surrounding environment and cannot effectively handle complex decision-making requirements such as obstacle avoidance.

[0003] Furthermore, existing methods typically utilize visual or LiDAR (Light Detection and Ranging) sensors to perceive environmental features and achieve localization by matching them with an environmental map. Based on the dominant sensor, these methods can be further subdivided into two categories:

[0004] 1) Multi-line LiDAR-based solution: This solution utilizes high-precision point cloud data from LiDAR for matching and positioning. While highly accurate, it is extremely expensive, consumes a lot of power, and is bulky, making it difficult to widely adopt in lightweight low-altitude UAVs. More importantly, it has poor tolerance to extreme weather conditions (such as rain, fog, and dust) and is prone to failure in scenarios with sparse geographical features (such as wide bodies of water or snowfields). Once the LiDAR malfunctions, the entire system will be paralyzed due to over-reliance on a single main sensor, and other auxiliary sensors will be unable to take over.

[0005] 2) Vision-led fusion solution: This solution utilizes cameras to collect rich texture information and performs localization through tight coupling with an IMU (i.e., Visual-Inertial Odometry, VIO). While this solution is cost- and size-controllable, it has a heavy computational burden and requires extremely high onboard computing power. Furthermore, visual SLAM (Simultaneous Localization and Mapping) has inherent flaws: it is extremely sensitive to changes in lighting (such as sudden changes in lighting at tunnel entrances and exits) and prone to tracking loss in low-texture environments (such as white walls and smooth surfaces), leading to system output interruptions and making stability difficult to guarantee. Summary of the Invention

[0006] This application provides an adaptive multi-source fusion navigation method and system based on dynamic scene perception, which can still achieve high-precision, low-divergence, robust positioning and safe obstacle avoidance for low-altitude aircraft under harsh conditions such as long-term or frequent GNSS interruptions and visual or lidar failures.

[0007] In a first aspect, embodiments of this application provide an adaptive multi-source fusion navigation method based on dynamic scene perception, comprising: performing error correction on inertial measurement data output by an inertial measurement unit to obtain corrected inertial data, and inputting the corrected inertial data into an inertial navigation mechanical orchestration module to generate a navigation reference reference; obtaining double-interpolated observations based on observation data from a base station and a rover station, calculating the double-difference geometric distance between the station and the satellite using position information output by the inertial navigation system, and obtaining residual data between the double-interpolated observations and the double-difference geometric distance; inputting the residual data into a Kalman filter, and having the Kalman filter output an optimal estimate; feeding the optimal estimate back to the inertial measurement unit for compensation, and correcting the navigation reference reference, and outputting the final navigation data.

[0008] In one implementation of the first aspect, the step of error correction of the inertial measurement data output by the inertial measurement unit and obtaining corrected inertial data includes: estimating the error of the inertial measurement data in real time using the Kalman filter; and performing online compensation on the original angular increment and velocity increment output by the inertial measurement unit based on the error to obtain corrected inertial data.

[0009] In one implementation of the first aspect, the error of the inertial measurement data includes gyroscope bias, accelerometer bias, and scaling factor error.

[0010] In one implementation of the first aspect, the step of inputting the corrected inertial data into the inertial navigation mechanical orchestration module to generate a navigation reference reference includes: the inertial navigation mechanical orchestration module performing an integral operation on the input angular velocity information from the gyroscope, calculating the attitude transformation matrix of the carrier coordinate system relative to the navigation coordinate system in real time, and generating an attitude reference reference for the navigation carrier; the inertial navigation mechanical orchestration module performing an integral operation on the input specific force information from the accelerometer, updating the velocity components of the navigation carrier in the navigation coordinate system in real time, and generating a velocity reference reference for the navigation carrier; and the inertial navigation mechanical orchestration module performing a second integral operation on the updated velocity components to generate a position reference reference for the navigation carrier.

[0011] In one implementation of the first aspect, the acquisition of double-interpolated observations based on observation data from the base station and the rover includes: selecting a reference satellite based on the original observation data from the base station and the rover, and constructing double-difference observations that include carrier phase double-difference observations and / or pseudorange double-difference observations; wherein the double-difference observations are used to eliminate satellite clock errors, receiver clock errors, and most atmospheric delay errors.

[0012] In one implementation of the first aspect, the calculation of the double-difference geometric distance between the station and the satellite using the position information output by the inertial navigation system includes: using the position information of the rover station output by the inertial navigation system, combined with the known position information of the reference station and the satellite spatial position provided by the satellite ephemeris, to calculate the difference between the geometric distances between the rover station and the reference station to the same satellite, and then constructing the double-difference geometric distances corresponding to the reference satellite and the non-reference satellite.

[0013] In one implementation of the first aspect, the residual data includes at least residual atmospheric delay error, satellite orbit error, and carrier phase integer ambiguity; the step of inputting the residual data into a Kalman filter and having the Kalman filter output the optimal estimate includes: constructing the residual data into a measurement vector of the Kalman filter; establishing the measurement equation of the Kalman filter, the measurement equation being used to describe the mapping relationship between the measurement vector and the system state vector; using the output of the inertial navigation system's inertial navigation mechanical orchestration module as the basis for state transition, injecting the measurement vector into the Kalman filter, performing measurement updates, and obtaining the optimal estimate of the state vector; the optimal estimate includes at least: the optimal estimate of position error, the optimal estimate of velocity error, the optimal estimate of attitude error, the optimal estimate of inertial sensor zero bias, and the optimal estimate of the floating-point solution of carrier phase ambiguity.

[0014] In one implementation of the first aspect, the optimal estimate of the floating-point solution of the carrier phase ambiguity is input into a preset ambiguity search module, and the ambiguity search module outputs integer ambiguity based on an integer least squares search algorithm.

[0015] In one implementation of the first aspect, feeding back the optimal estimate to the inertial measurement unit for compensation and correcting the navigation reference datum includes: feeding back the optimal zero-bias estimate of the inertial sensor to the inertial measurement unit for compensation, and correcting the navigation reference datum generated by the inertial navigation mechanical orchestration module with the optimal position error estimate, the optimal velocity error estimate, and the optimal attitude error estimate.

[0016] Secondly, embodiments of this application provide an adaptive multi-source fusion navigation system based on dynamic scene perception, comprising: a sensing layer, an information transmission layer, and a decision processing layer; the sensing layer includes various types of sensors and an inertial measurement unit; the various types of sensors collect sensing information about the environment in which the navigation vehicle is located, and the inertial measurement unit senses the acceleration and angular velocity of the navigation vehicle through accelerometers and gyroscopes, thereby calculating the attitude angle and relative displacement changes; the information transmission layer transmits the data acquired by the sensing layer to the decision processing layer; the decision processing layer performs navigation decision processing using the adaptive multi-source fusion navigation method based on dynamic scene perception as described above.

[0017] The adaptive multi-source fusion navigation method and system based on dynamic scene perception provided in this application have the following beneficial effects:

[0018] 1. This application uses the error correction of the inertial measurement unit (IMU) and the navigation reference base generated by the inertial navigation system's mechanical arrangement as a foundation. It combines double-difference observations and residual data from the base station and rover station, and uses a Kalman filter to output the optimal estimate. This optimal estimate is then fed back to the IMU for compensation and to correct the navigation reference base, forming a closed-loop control system. This technical solution effectively suppresses the cumulative divergence of inertial navigation errors and eliminates satellite common errors using double-difference observations, achieving centimeter-level positioning. Even under harsh conditions such as long-term or frequent GNSS interruptions and visual or lidar failures, it can still achieve high-precision, low-divergence, robust positioning and obstacle avoidance for low-altitude aircraft.

[0019] 2. This application establishes a dual closed-loop correction mechanism by simultaneously feeding back the optimal estimate to the inertial measurement unit for compensation and correcting the navigation reference. When the estimate in any part fluctuates, the feedback compensation can dynamically adjust the input of the inertial measurement unit and the mechanically arranged reference output, preventing the unidirectional accumulation of errors from causing system failure, thereby improving the overall reliability of the navigation system. Attached Figure Description

[0020] Figure 1 The diagram shown is an overall flowchart of an adaptive multi-source fusion navigation method based on dynamic scene perception, according to an embodiment of this application.

[0021] Figure 2 The flowchart shown is a process for acquiring corrected inertial data in an adaptive multi-source fusion navigation method based on dynamic scene perception according to an embodiment of this application.

[0022] Figure 3 The diagram shown illustrates the principle of generating a navigation reference base in an adaptive multi-source fusion navigation method based on dynamic scene awareness, according to an embodiment of this application.

[0023] Figure 4 The diagram shown illustrates the principle of obtaining dual-interpolation observations in an adaptive multi-source fusion navigation method based on dynamic scene perception, according to an embodiment of this application.

[0024] Figure 5 The diagram shown illustrates the principle of obtaining double-difference geometric distance in an adaptive multi-source fusion navigation method based on dynamic scene perception, according to an embodiment of this application.

[0025] Figure 6 The flowchart shown is a process for determining the optimal estimate in an adaptive multi-source fusion navigation method based on dynamic scene perception, according to an embodiment of this application.

[0026] Figure 7 The diagram shown illustrates the overall implementation principle of an adaptive multi-source fusion navigation method based on dynamic scene perception, according to an embodiment of this application.

[0027] Figure 8 The diagram shown is a principle block diagram of an adaptive multi-source fusion navigation system based on dynamic scene perception, according to an embodiment of this application.

[0028] Figure 9 The diagram shown illustrates the implementation principle of an adaptive multi-source fusion navigation system based on dynamic scene perception, according to an embodiment of this application.

[0029] Component designation explanation

[0030] Adaptive multi-source fusion navigation based on dynamic scene perception -100; Sensing layer -110; Information transmission layer -120; Decision processing layer -130; S100~S400 - steps; S110~S120 - steps; S310~S330 - steps. Detailed Implementation

[0031] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0032] Before providing a detailed description of this embodiment, the nouns and terms used in this embodiment are explained, and the nouns and terms used in this embodiment are subject to the following interpretations:

[0033] RTK: Real-Time Kinematic, real-time dynamic differential;

[0034] INS: Inertial Navigation System;

[0035] IMU, Inertial Measurement Unit;

[0036] GNSS, Global Navigation Satellite System;

[0037] LAMBDA, Least-squares AMBiguity Decorrelation Adjustment, is a method for adjusting ambiguity by least squares.

[0038] To address the technical problems of low positioning accuracy and poor robustness of a single navigation source in complex dynamic scenarios in existing technologies, and the lack of adaptability of traditional multi-source fusion methods to dynamic scenarios, resulting in insufficient fusion efficiency and stability, this application provides an adaptive multi-source fusion navigation method and system based on dynamic scene perception. By constructing a closed-loop feedback architecture with tight coupling between inertial and satellite systems, it effectively solves the problems of low positioning accuracy and poor robustness of a single navigation source in complex dynamic scenarios in existing technologies, as well as the lack of scene adaptability in traditional multi-source fusion.

[0039] The following will refer to the appendices in the embodiments of this application. Figure 1 To be continued Figure 9 This application provides a detailed description of the technical solutions in its embodiments. This allows those skilled in the art to understand and implement the adaptive multi-source fusion navigation method and system based on dynamic scene perception, without requiring creative effort.

[0040] In the low-altitude economy, the precise flight and control of unmanned aerial vehicles (UAVs) rely on stable and reliable high-precision positioning technology. Satellite navigation RTK technology, as a key means to achieve centimeter-level positioning, effectively eliminates common errors such as satellite orbital errors, ionospheric and tropospheric delays through real-time differential data transmission between the base station and the rover. This provides a crucial position reference for UAVs' autonomous take-off and landing, precise route tracking, and close formation flight.

[0041] However, pure RTK technology still faces significant challenges in practical applications. In complex environments such as urban canyons and forests, signals are easily affected by obstruction and multipath interference, making it difficult for RTK algorithms to quickly and reliably determine the integer ambiguity of the carrier phase. Furthermore, in highly dynamic scenarios, the RTK update frequency (typically 1-20Hz) is insufficient to meet the control requirements for agile drone flight.

[0042] To overcome these limitations, this embodiment first provides an adaptive multi-source fusion navigation method based on dynamic scene awareness. This method employs a deep RTK / INS integrated navigation algorithm based on a centralized Kalman filter. Inertial Navigation Systems (INS), with their high output frequency (typically above 100Hz), short-term high accuracy, and fully autonomous operation, compensate for the shortcomings of RTK technology. The continuous attitude and position change information provided by INS can effectively assist in the rapid resolution of RTK integer ambiguities and provide high-precision dead reckoning capabilities during brief satellite signal interruptions. The core of this adaptive multi-source fusion navigation method based on dynamic scene awareness is a centralized Kalman filter, which achieves a significant improvement in navigation performance through the optimal fusion of multi-source observation information.

[0043] Figure 1 The flowchart shown is an example of an adaptive multi-source fusion navigation method based on dynamic scene awareness in this application. Figure 1 As shown, the adaptive multi-source fusion navigation method based on dynamic scene perception provided in this application includes the following steps S100 to S400.

[0044] Step S100: Perform error correction on the inertial measurement data output by the inertial measurement unit, obtain corrected inertial data, and input the corrected inertial data into the inertial navigation mechanical arrangement module to generate a navigation reference reference.

[0045] Step S200: Obtain double-interpolated observations based on the observation data of the base station and the rover station; calculate the double-difference geometric distance between the station and the satellite using the position information output by the inertial navigation system; and obtain the residual data between the double-interpolated observations and the double-difference geometric distance.

[0046] Step S300: Input the residual data into the Kalman filter, and the Kalman filter outputs the optimal estimate.

[0047] In step S400, the optimal estimated value is fed back to the inertial measurement unit for compensation, and the navigation reference benchmark is corrected to output the final navigation data.

[0048] The adaptive multi-source fusion navigation method based on dynamic scene perception provided in this embodiment firstly corrects the error of the raw data of the inertial measurement unit (IMU) and uses the inertial navigation mechanical orchestration module to generate a high-frequency continuous navigation reference reference, ensuring stable state recursion under GNSS signal interruption or high dynamic conditions. Secondly, based on the double-difference observations of the base station and the rover, the double-difference geometric distance is calculated and the residual is extracted by combining the position information output by the inertial navigation system. This residual directly reflects the deviation between satellite observation and inertial navigation calculation, thereby eliminating satellite clock error, receiver clock error and most atmospheric delay errors. Then, the residual data is input into a centralized Kalman filter, and the optimal estimate containing position, velocity, attitude error and ambiguity floating-point solution is output through time update and measurement update. Finally, the optimal estimate is fed back to the IMU for compensation and to correct the navigation reference reference, forming a full closed-loop control. This mechanism can adjust the fusion weights in real time according to the dynamic scene, continuously suppress the cumulative drift of the inertial navigation and make up for the temporary loss of lock of the GNSS, achieving centimeter-level high-precision positioning. It significantly improves the robustness and adaptability of the system in complex low-altitude environments (such as urban canyons, signal blockage, and high-dynamic flight), while relying only on low-cost GNSS / IMU sensors, thus balancing accuracy and cost.

[0049] The following is in conjunction with the appendix Figure 2 To be continued Figure 7 The steps S100 to S400 of the adaptive multi-source fusion navigation method based on dynamic scene perception in this embodiment will be described in detail.

[0050] Step S100: Perform error correction on the inertial measurement data output by the inertial measurement unit, obtain corrected inertial data, and input the corrected inertial data into the inertial navigation mechanical orchestration module to generate a navigation reference base.

[0051] In this embodiment, the inertial sensor errors (including gyroscope bias, accelerometer bias, and scaling factor errors) estimated in real time using a Kalman filter are first used to compensate online for the raw angular and velocity increments output by the IMU. The error-corrected IMU data is then sent to the inertial navigation mechanical orchestration module, where integration calculations are performed to update the carrier's position, velocity, and attitude in real time, forming a high-frequency navigation reference.

[0052] Figure 2The flowchart shown is a process for acquiring corrected inertial data in an adaptive multi-source fusion navigation method based on dynamic scene awareness, according to an embodiment of this application. Figure 2 As shown, in one implementation of this embodiment, the step of performing error correction on the inertial measurement data output by the inertial measurement unit and obtaining corrected inertial data includes:

[0053] Step S110: The error of the inertial measurement data is estimated in real time using the Kalman filter;

[0054] Step S120: Based on the error, perform online compensation on the original angular increment and velocity increment output by the inertial measurement unit to obtain corrected inertial data.

[0055] In one implementation of this embodiment, the error of the inertial measurement data includes gyroscope bias, accelerometer bias, and scaling factor error.

[0056] In this embodiment, a Kalman filter is used to estimate the output error of the inertial measurement unit in real time. The Kalman filter uses the inertial sensor error term in the system state vector as the parameter to be estimated, and outputs the optimal estimate of the inertial measurement data error through recursive calculation of time update and measurement update. The inertial measurement data error includes at least the zero bias error of the gyroscope, the zero bias error of the accelerometer, and the scaling factor error of the gyroscope and the accelerometer.

[0057] In this embodiment, based on the inertial measurement data error estimated by the Kalman filter, the original angular increment and velocity increment output by the inertial measurement unit are compensated online to obtain corrected inertial data. The online compensation includes: subtracting the estimated gyroscope zero bias from the original angular increment, subtracting the estimated accelerometer zero bias from the original velocity increment, and scaling the original measurement value according to the scaling factor error. The compensated corrected inertial data is used as the input of the inertial navigation mechanical orchestration module to generate a navigation reference benchmark.

[0058] The error estimation principle of the inertial measurement data is as follows:

[0059] First, the zero-bias error of the gyroscope is estimated: gyroscope zero-bias refers to the constant or slowly changing offset in the gyroscope output under static or dynamic conditions. This error is modeled as a first-order Markov process, meaning the zero-bias is assumed to approach its mean value exponentially over time, driven by random noise. This modeling method reflects the characteristics of the gyroscope's zero-bias drift over time; its rate of change is determined by the relevant time constant, with a larger time constant resulting in a slower change in zero-bias.

[0060] Secondly, the zero-bias error of the accelerometer is estimated: the accelerometer zero bias is a constant deviation in the accelerometer output, which undergoes slow random changes due to factors such as temperature and time. This error is modeled as a random walk process, meaning the derivative of the zero bias is a white noise process. This implies that the current zero bias is the zero bias of the previous moment plus a random increment, and the variance of the increment determines the random fluctuation amplitude of the zero bias. The random walk model is suitable for describing the characteristics of long-term accumulated, irregularly drifting errors.

[0061] Next, the scaling factor error is estimated: the scaling factor error refers to the inaccuracy of the sensor's scaling factor, causing a linear scaling deviation between the measured value and the true value. This error is characterized by the scaling factor error coefficient. The correction principle for the original measured value is as follows: the estimated scaling factor error is multiplied element-wise with the original measured value, and then the product term is subtracted from the original measured value to obtain the corrected measured value. The scaling factor error is usually modeled as a random constant or a first-order Markov process, and its estimate is adaptively adjusted by the Kalman filter during operation.

[0062] The aforementioned error estimation provides the Kalman filter with accurate system state equations and noise statistics. By estimating these error state variables in real time through the filter and feeding them back to the original measurement data for online compensation, the long-term cumulative error of inertial navigation can be effectively suppressed, thus improving navigation accuracy.

[0063] The correction angle increment and correction velocity increment obtained after online compensation are aligned by timestamp to form a continuous correction inertial data stream; the correction inertial data is then sent to the inertial navigation mechanical orchestration module to update the attitude, velocity and position reference of the carrier.

[0064] In one implementation of this embodiment, the step of inputting the corrected inertial data into the inertial navigation mechanical orchestration module to generate a navigation reference reference includes:

[0065] 1) The inertial navigation mechanical orchestration module integrates the angular velocity information from the input gyroscope, calculates the attitude transformation matrix of the carrier coordinate system relative to the navigation coordinate system in real time, and generates the attitude reference benchmark of the navigation carrier.

[0066] The inertial navigation mechanical orchestration module receives the error-corrected gyroscope angular velocity information and performs attitude update calculations using the navigation coordinate system (e.g., East-North-Earth coordinate system) as a reference. Specifically, it constructs an attitude transformation matrix of the carrier coordinate system relative to the navigation coordinate system using the angular velocity information. This matrix is ​​typically represented in quaternion or equivalent rotation vector form. By performing a one-time integration operation on the angular velocity information, the angular change of the carrier in each sampling period is calculated in real time, and the attitude transformation matrix is ​​recursively updated. Based on the updated attitude transformation matrix, the pitch, roll, and yaw angles of the carrier are extracted to form a continuous, high-frequency attitude reference output. This attitude reference is used to describe the real-time orientation of the navigation carrier in space, providing a coordinate transformation basis for subsequent velocity and position calculations.

[0067] 2) The inertial navigation mechanical orchestration module performs integral calculations on the specific force information input from the accelerometer, updates the velocity components of the navigation vehicle in the navigation coordinate system in real time, and generates a velocity reference benchmark for the navigation vehicle.

[0068] The inertial navigation mechanical orchestration module receives the accelerometer specific force information after error correction and, in conjunction with the generated attitude transformation matrix, performs velocity update calculations. First, the specific force information output by the accelerometers (the specific force vector in the vehicle coordinate system) is projected onto the navigation coordinate system through the attitude transformation matrix to obtain the specific force components in the navigation coordinate system. Then, the gravitational acceleration vector (and optional Coriolis correction terms) is subtracted from the specific force components in the navigation coordinate system to obtain the vehicle's true acceleration in the navigation coordinate system. This acceleration is then integrated in a single operation, and the north, east, and ground velocity components of the vehicle in the navigation coordinate system are updated in real time, forming a continuous, high-frequency velocity reference output. This velocity reference is used to describe the linear motion state of the navigation vehicle.

[0069] 3) The inertial navigation mechanical orchestration module performs a second integral operation on the updated velocity components to generate a position reference benchmark for the navigation carrier.

[0070] The inertial navigation mechanical orchestration module uses the generated velocity reference base as input to perform position update calculations. Specifically, it performs a double integration operation on the velocity components in the navigation coordinate system: first, it integrates the velocity to obtain the displacement increment; then, it accumulates the displacement increment based on the initial position, updating the carrier's geodetic coordinates, including latitude, longitude, and elevation, in real time. During the position update process, the influence of the Earth's curvature and coordinate system rotation on the position calculation must be considered, and corresponding differential equations or approximate algorithms are used for compensation. Finally, a continuous, high-frequency position reference base output is formed, which, together with the attitude reference base and velocity reference base, constitutes a complete navigation reference base.

[0071] Finally, the inertial navigation mechanical orchestration module aligns the generated attitude reference reference, velocity reference reference, and position reference reference with the same timestamp, forming a navigation reference reference data package containing the three elements of position, velocity, and attitude. This navigation reference reference is continuously output at a preset high-frequency update rate (e.g., 100Hz to 1000Hz) to fill the data gaps between adjacent epochs of external absolute positioning sensors (such as GNSS) and to provide high-frequency state recursion for the prediction phase of the combined navigation filter.

[0072] Figure 3 This diagram illustrates the principle of generating a navigation reference base in an adaptive multi-source fusion navigation method based on dynamic scene awareness, according to an embodiment of this application. Figure 3 As shown, firstly, the inertial navigation mechanical orchestration module uses the corrected gyroscope angular velocity information to calculate the attitude transformation matrix of the vehicle coordinate system relative to the navigation coordinate system in real time through a single integration operation, generating an attitude reference base including pitch, roll, and heading angles. Secondly, using the corrected accelerometer specific force information, combined with the attitude transformation matrix, it projects the information onto the navigation coordinate system and subtracts gravitational acceleration, then updates the north, east, and ground velocity components of the vehicle in the navigation coordinate system in real time through a single integration operation, forming a velocity reference base. Then, a second integration operation is performed on the velocity components to update the latitude, longitude, and elevation of the vehicle in real time, generating a position reference base. Finally, the attitude, velocity, and position reference bases are aligned with the same timestamp and continuously output at a preset high-frequency update rate (e.g., 100Hz to 1000Hz) to form a complete navigation reference base, used to fill the data gaps between adjacent epochs of GNSS and other absolute positioning sensors and to provide high-frequency state prediction for the combined navigation filter.

[0073] Step S200: Obtain double-input observations based on the observation data of the base station and the rover station, calculate the double-difference geometric distance between the station and the satellite using the position information output by the inertial navigation system, and obtain the residual data between the double-input observations and the double-difference geometric distance.

[0074] In the RTK processing chain, this embodiment utilizes the raw observation data from the base station and rover receivers to construct double-difference pseudorange and double-difference carrier phase observations. Simultaneously, using precise position information derived from INS combined with broadcast ephemeris data, the double-difference geometric distance between the station and satellite is calculated. The residual obtained by subtracting the double-difference observation from the double-difference geometric distance serves as the measurement input for the Kalman filter, containing key information such as satellite orbital error, atmospheric delay error, and integer ambiguity.

[0075] Specifically, in one implementation of this embodiment, the acquisition of double-interpolated observations based on observation data from the base station and the rover includes: selecting a reference satellite based on the original observation data from the base station and the rover, and constructing double-difference observations that include carrier phase double-difference observations and / or pseudorange double-difference observations; wherein the double-difference observations are used to eliminate satellite clock errors, receiver clock errors, and most atmospheric delay errors.

[0076] In this embodiment, raw observation data output from the base station receiver and the rover receiver are acquired respectively. The raw observation data includes at least carrier phase observations, pseudorange observations, and corresponding satellite ephemeris information. The base station is located at a fixed point with known coordinates, and the rover is located on a mobile platform to be positioned. Both observe the same satellite constellation synchronously.

[0077] A reference satellite is selected from the set of satellites jointly observed by the rover and the base station. The selection criteria for the reference satellite include, but are not limited to, satellite elevation angle, carrier-to-noise ratio (CNR), or signal quality indicator factor. Typically, the satellite with the highest elevation angle and the best signal quality is selected as the reference satellite to ensure the stability and reliability of the differential observations.

[0078] For each non-reference satellite, single-difference observations are constructed between the base station and the rover. Specifically, the rover's observations for a particular satellite are subtracted from the base station's observations for the same satellite to obtain single-difference carrier phase observations and / or single-difference pseudorange observations. This single-difference operation eliminates the common components of satellite clock errors and most atmospheric delay errors.

[0079] Using a selected reference satellite as a benchmark, the single-difference observation of the reference satellite is subtracted from the single-difference observation of each non-reference satellite to obtain double-difference observations. These double-difference observations include double-difference carrier phase observations and / or double-difference pseudorange observations. Double-difference operation further eliminates receiver clock errors and residual atmospheric delay errors, retaining only the geometric distance difference between the satellite and receiver, integer ambiguity (for carrier phase), and some unmodeled noise.

[0080] The constructed carrier phase double-difference observations and / or pseudorange double-difference observations are used as input data for subsequent processing modules. The carrier phase double-difference observations are used for high-precision positioning and integer ambiguity resolution, while the pseudorange double-difference observations are used for auxiliary positioning or initial ambiguity value determination. These double-difference observations simultaneously eliminate satellite clock errors, receiver clock errors, and most tropospheric and ionospheric delay errors, significantly improving positioning accuracy and robustness.

[0081] Figure 4 This diagram illustrates the principle of obtaining dual-interpolation observations in an adaptive multi-source fusion navigation method based on dynamic scene awareness, according to an embodiment of this application. Figure 4As shown, the simplified process of obtaining double-difference observations based on observation data from the base station and rover station is as follows: First, the raw carrier phase and pseudorange observation data of the base station and rover station are obtained respectively, and the satellite with the highest elevation angle and the best signal quality among the common-view satellites is selected as the reference satellite; then, for each non-reference satellite, the rover station observation value is subtracted from the base station observation value to obtain the inter-station single-difference observation value, and then the inter-station single-difference observation value of the reference satellite is subtracted from the single-difference observation value, thereby constructing the carrier phase double-difference observation value and / or pseudorange double-difference observation value; through the above double-difference operation, satellite clock error, receiver clock error and most of the tropospheric and ionospheric delay errors are effectively eliminated, and high-precision double-difference observation values ​​are output for subsequent residual calculation and ambiguity resolution.

[0082] In one implementation of this embodiment, the step of calculating the double-difference geometric distance between the station and the satellite using the position information output by the inertial navigation system includes: using the position information of the rover station output by the inertial navigation system, combined with the known position information of the reference station and the satellite spatial position provided by the satellite ephemeris, to calculate the difference between the geometric distances between the rover station and the reference station to the same satellite, and then constructing the double-difference geometric distances corresponding to the reference satellite and the non-reference satellite.

[0083] The system acquires the rover position information output in real time by the inertial navigation system, the pre-calibrated and stored reference station position information, and the satellite spatial position information provided by the satellite ephemeris. The rover position information consists of coordinate values ​​(e.g., latitude, longitude, and elevation) in the navigation coordinate system, the reference station position information consists of known fixed coordinates, and the satellite spatial position information is calculated based on the satellite ephemeris and signal transmission time.

[0084] The inter-station geometric distances to each commonly observed satellite are calculated separately for both the rover station and the base station. Specifically, for a given satellite, the geometric distance from the rover station to the satellite is calculated using the satellite's spatial coordinates and the rover station's coordinates, according to the Euclidean distance formula. Similarly, the geometric distance from the base station to the same satellite is calculated using the satellite's spatial coordinates and the base station's coordinates. For each non-reference satellite, the geometric distance from the rover station to the satellite is subtracted from the geometric distance from the base station to the satellite to obtain the corresponding inter-station single-difference geometric distance. This single-difference operation eliminates the common error term between the satellite and the station, while preserving the relative geometric relationships between the stations.

[0085] Using a pre-selected reference satellite as a benchmark, the inter-station single-difference geometric distance of the reference satellite is subtracted from the inter-station single-difference geometric distance of the non-reference satellite to obtain the double-difference geometric distance corresponding to that non-reference satellite. The double-difference geometric distance is expressed by the following formula: subtracting again from the inter-station single-difference distance eliminates residual common errors such as receiver clock errors, retaining only the double difference between the geometric distances from the two stations to the two satellites. The calculated double-difference geometric distances of each non-reference satellite are sorted by satellite number to form a double-difference geometric distance vector, which serves as the input to the subsequent residual calculation module. This vector has the same data structure as the double-difference observation vector and is used to subtract it to obtain the double-difference residual data.

[0086] Figure 5 This diagram illustrates the principle of obtaining double-difference geometric distance in an adaptive multi-source fusion navigation method based on dynamic scene awareness, according to an embodiment of this application. Figure 5 As shown, the simplified process of calculating the inter-station double-difference geometric distance using the position information output by the inertial navigation system is as follows: First, obtain the rover's position output in real time by the inertial navigation system, the known fixed position of the reference station, and the satellite's spatial position provided by the satellite ephemeris; calculate the inter-station geometric distance from the rover and the reference station to each common-view satellite respectively; then, subtract the geometric distance from the reference station to the same satellite from the geometric distance from the rover to the satellite to obtain the inter-station single-difference geometric distance; finally, using the selected reference satellite as the reference, subtract the single-difference geometric distance of the reference satellite from the single-difference geometric distance of the non-reference satellite to construct the double-difference geometric distance. This double-difference geometric distance has the same data structure as the aforementioned double-difference observation value and is used for subsequent residual calculation.

[0087] In one implementation of this embodiment, the residual data includes at least residual atmospheric delay error, satellite orbit error, and carrier phase integer ambiguity.

[0088] The aforementioned constructed carrier phase double-difference observations (or pseudorange double-difference observations) are aligned with the calculated double-difference geometric distances according to the satellite number and reference satellite correspondence, ensuring that both point to the same satellite pair (a combination of reference and non-reference satellites) and the same observation epoch. For each non-reference satellite, the corresponding double-difference geometric distance is subtracted from the double-difference observation to obtain the double-difference residual data; this operation is expressed as: Residual = Double-difference observation - Double-difference geometric distance. The double-difference residual data characterizes the degree of deviation between the actual observation and the theoretical geometric distance.

[0089] The double-difference residual data contains at least the following error components that are not modeled or cannot be completely eliminated by double-difference: residual atmospheric delay error (including tropospheric delay residual and ionospheric delay residual), residual satellite orbit error components, and carrier phase integer ambiguity terms (for carrier phase double-difference observations). Integer ambiguity is represented by an integer multiple of the wavelength constant and is a parameter that needs to be determined for subsequent ambiguity fixing; residual atmospheric delay and orbit error are slowly varying low-frequency components. The double-difference residual data for each epoch and for each non-reference satellite are arranged in chronological order to form a multidimensional residual vector sequence. This sequence is used as the observation information input to the measurement update module of the centralized Kalman filter to drive the filter to perform optimal estimation of system state quantities (including position error, velocity error, attitude error, sensor error, and ambiguity floating-point solution).

[0090] Step S300: Input the residual data into the Kalman filter, and the Kalman filter outputs the optimal estimate.

[0091] Figure 6 The flowchart shown is a process for determining the optimal estimate in an adaptive multi-source fusion navigation method based on dynamic scene awareness, according to an embodiment of this application. Figure 6 As shown, in this embodiment, the step of inputting the residual data into the Kalman filter and having the Kalman filter output the optimal estimate includes the following steps S310 to S330.

[0092] Step S310: Construct the residual data into a measurement vector for a Kalman filter;

[0093] Step S320: Establish the measurement equation of the Kalman filter, which is used to describe the mapping relationship between the measurement vector and the system state vector;

[0094] Step S330: Using the output of the inertial navigation system's inertial navigation mechanical orchestration module as the basis for state transition, the measurement vector is injected into the Kalman filter to perform measurement update and obtain the optimal estimate of the state vector.

[0095] The optimal estimates include at least: the optimal estimate of position error, the optimal estimate of velocity error, the optimal estimate of attitude error, the optimal estimate of inertial sensor zero bias, and the optimal estimate of carrier phase ambiguity floating-point solution.

[0096] The process of inputting the residual data into a Kalman filter and having the Kalman filter output the optimal estimate is as follows: First, a system state vector is constructed, including at least position error, velocity error, attitude error, inertial sensor bias, and carrier phase ambiguity floating-point solution. A state transition matrix and a process noise covariance matrix are also established. Using the high-frequency navigation reference output from the inertial navigation system's mechanical orchestration as the basis for state prediction, time updates are performed, and the one-step state prediction value and its covariance matrix are recursively obtained. Then, using the residual data as a measurement vector, a measurement matrix is ​​established (reflecting the line-of-sight vector mapping of position error and the integer coefficient relationship of the ambiguity term). Measurement updates are performed: the Kalman gain is calculated, and the one-step prediction value is weighted and fused with the measurement residual (actual measurement minus predicted measurement) to obtain the optimal state estimate; simultaneously, the covariance matrix is ​​updated. Finally, position, velocity, and attitude errors are extracted from the optimal estimate to correct the navigation reference, sensor bias is extracted to compensate for the IMU raw data, and the carrier phase ambiguity floating-point solution is output for subsequent integer fixing by the LAMBDA module. The entire filtering process is recursively applied epochally with the sampling period as the step size, achieving continuous optimal estimation of the navigation state.

[0097] In one implementation of this embodiment, the optimal estimate of the floating-point solution of the carrier phase ambiguity is input into a preset ambiguity search module, and the ambiguity search module outputs integer ambiguity based on an integer least squares search algorithm.

[0098] In one implementation of this embodiment, the carrier phase ambiguity floating-point solution and its covariance matrix output by the Kalman filter are input into a preset ambiguity search module. The ambiguity search module first performs a decorrelation transformation: using an integer transformation matrix to perform a linear transformation on the floating-point solution and its covariance matrix, compressing the correlation between ambiguity parameters and transforming the original search space from a flat, elongated shape into a roughly spherical, regular space, thereby significantly reducing the range of integer combinations to be searched. Then, an integer least squares search is performed in the decorrelated integer space, aiming to minimize the sum of squared residuals, to determine the integer vector that minimizes the objective function. Finally, the searched integer vector is mapped back to the original ambiguity space through an inverse transformation, outputting a fixed integer solution for integer ambiguity. This integer solution is used, on the one hand, to substitute back into the double-difference observation equation to constrain the position calculation for centimeter-level positioning; on the other hand, it can be fed back to the Kalman filter as a virtual observation to further improve filtering accuracy.

[0099] In this embodiment, the Kalman filter effectively suppresses the cumulative divergence of inertial navigation errors by continuously and optimally estimating system state variables (including position, velocity, attitude error, sensor error, and carrier phase ambiguity). The floating-point ambiguity estimate output by the filter is then fed into an ambiguity search module (such as the LAMBDA algorithm), where it is fixed to the correct integer ambiguity through integer least squares search.

[0100] In step S400, the optimal estimated value is fed back to the inertial measurement unit for compensation, and the navigation reference benchmark is corrected to output the final navigation data.

[0101] In one implementation of this embodiment, feeding back the optimal estimate to the inertial measurement unit for compensation and correcting the navigation reference benchmark includes: feeding back the optimal zero-bias estimate of the inertial sensor to the inertial measurement unit for compensation, and correcting the navigation reference benchmark generated by the inertial navigation mechanical orchestration module with the optimal estimates of the position error, the optimal estimates of the velocity error, and the optimal estimates of the attitude error.

[0102] The optimal zero-bias estimate of the inertial sensor output from the Kalman filter is fed back to the error compensation module of the inertial measurement unit. This optimal zero-bias estimate includes at least the gyroscope zero-bias estimate and the accelerometer zero-bias estimate. The error compensation module receives the original angular increment and velocity increment, and subtracts the corresponding zero-bias estimate from them, i.e.:

[0103] The corrected angle increment = original angle increment - gyroscope zero bias estimate; the corrected velocity increment = original velocity increment - accelerometer zero bias estimate. The compensated inertial data is then fed back into the inertial navigation mechanical orchestration module for the generation of the next navigation reference base, thus forming closed-loop control and effectively suppressing the long-term accumulation of inertial sensor errors.

[0104] The optimal estimates of position, velocity, and attitude errors output from the Kalman filter are applied to the navigation reference bases generated by the inertial navigation mechanical orchestration module. Specifically: for position correction, the corresponding position error estimates are subtracted from the latitude, longitude, and elevation output by the mechanical orchestration module to obtain the corrected position reference; for velocity correction, the corresponding velocity error estimates are subtracted from the north, east, and ground velocity components to obtain the corrected velocity reference; for attitude correction, the corresponding attitude error estimates are subtracted from the pitch, roll, and yaw angles (or compensated for in the form of quaternion errors) to obtain the corrected attitude reference. The corrected position, velocity, and attitude reference bases are then re-aligned according to timestamps to form the final high-precision navigation data output.

[0105] The aforementioned feedback compensation and correction operations are performed epoch-by-epoch with the output period of the Kalman filter as the step size. That is, after each epoch measurement update, the estimated error is immediately fed back to the IMU compensation module and the mechanical orchestration module to ensure that the inertial navigation calculation and filtering prediction in the next epoch are based on the corrected state. This epoch-by-epoch feedback mechanism guarantees the real-time performance and convergence of the system.

[0106] In this embodiment, a closed-loop feedback correction mechanism is employed. The sensor error parameters estimated by the Kalman filter are fed back to the IMU's raw observation end in real time for compensation. Simultaneously, the estimated position and velocity errors are used to correct the inertial navigation mechanical arrangement results. This feedback structure ensures that the INS maintains high navigation accuracy even during short-term GNSS signal interruptions. Finally, the system outputs optimally fused high-precision position, velocity, and attitude navigation parameters, providing stable and reliable state awareness for the UAV flight control system. The INS output frequency (above 100Hz) is much higher than that of RTK, which can assist RTK in quickly resolving ambiguities and meet the real-time control requirements of agile UAV flight; therefore, the solution has strong real-time performance.

[0107] Figure 7 This diagram illustrates the overall implementation principle of an adaptive multi-source fusion navigation method based on dynamic scene awareness, according to an embodiment of this application. Figure 7 As shown in the figure, the overall implementation principle of the adaptive multi-source fusion navigation method based on dynamic scene awareness provided in this embodiment is as follows:

[0108] First, online error correction is performed on the raw angular and velocity increments output by the inertial measurement unit (IMU). A Kalman filter is used to estimate the gyroscope bias, accelerometer bias, and scaling factor errors in real time, and the raw data is compensated based on these estimates to obtain corrected inertial data. This corrected data is then input into the inertial navigation mechanical orchestration module. An attitude reference reference is generated by integrating the angular velocity once, a velocity reference reference is generated by integrating the force once, and a position reference reference is generated by integrating the velocity twice. These three references together constitute a high-frequency continuous navigation reference reference. Simultaneously, based on the raw observation data from the base station and rover, a reference satellite is selected to construct carrier phase double-difference observations and / or pseudorange double-difference observations. Using the rover position output by the inertial navigation system, the known base station position, and satellite ephemeris, the inter-station geometric distances between each satellite are calculated, and double-difference geometric distances corresponding to the reference satellite and non-reference satellites are constructed. Subtracting the double-difference geometric distances from the double-difference observations yields double-difference residual data that includes residual atmospheric delay, orbital error, and integer ambiguity.

[0109] Then, using position, velocity, attitude error, sensor zero bias, and ambiguity floating-point solution as the system state vector, the output of the inertial navigation mechanical orchestration is used as the prediction basis for time updates, and the double-difference residual data is used as the measurement vector for measurement updates. The optimal estimates for each state are output through centralized Kalman filtering. The ambiguity floating-point solution and its covariance matrix are fed into the LAMBDA module for downcorrelation transformation and integer least squares search, and are fixed as integer ambiguity solutions.

[0110] Finally, the estimated sensor bias is fed back to the IMU data compensation stage, and the position, velocity, and attitude errors are fed back to the mechanical orchestration module to correct the navigation reference. The fixed integer ambiguity is then substituted back into the double-difference observation equation, thus forming a closed-loop control system. This process is recursively applied epochally with the filtering period as the step size, enabling adaptive fusion of GNSS and IMU information in dynamic scenarios to achieve centimeter-level high-precision positioning and effectively suppressing the accuracy attenuation and divergence issues of a single navigation source in signal obstruction or highly dynamic environments.

[0111] In summary, the adaptive multi-source fusion navigation method based on dynamic scene perception proposed in this embodiment employs a deep RTK / INS combination algorithm. RTK technology can eliminate common satellite errors, achieve centimeter-level positioning, and solve the problem of insufficient accuracy of pure GNSS. INS has the characteristics of high-frequency output and short-term high accuracy, which can make up for the shortcomings of RTK signal blockage and high dynamic scenes. At the same time, the deep fusion of the two is achieved through centralized Kalman filtering, and the weighted complementary effects suppress their respective errors (INS cumulative drift, RTK signal interruption). Furthermore, the preprocessing stage eliminates sensor and environmental interference errors, thus achieving high-precision and high-stability positioning.

[0112] The scope of protection of the adaptive multi-source fusion navigation method based on dynamic scene perception described in this application is not limited to the execution order of the steps listed in this embodiment. Any solution implemented by adding, deleting, or replacing steps in the prior art based on the principles of this application is included within the scope of protection of this application.

[0113] This application provides an adaptive multi-source fusion navigation system based on dynamic scene perception. Figure 8 The diagram shown is a principle block diagram of an adaptive multi-source fusion navigation system based on dynamic scene awareness, according to an embodiment of this application. Figure 8 As shown, the adaptive multi-source fusion navigation system based on dynamic scene perception includes: a sensing layer, an information transmission layer, and a decision processing layer.

[0114] The sensing layer includes various types of sensors and an inertial measurement unit (IMU). The sensors collect information about the environment in which the navigation vehicle is located. The IMU uses accelerometers and gyroscopes to sense the acceleration and angular velocity of the navigation vehicle, and then calculates the attitude angle and relative displacement changes. The information transmission layer transmits the data acquired by the sensing layer to the decision processing layer. The decision processing layer uses the adaptive multi-source fusion navigation method based on dynamic scene perception as described above to perform navigation decision processing.

[0115] In this embodiment, the sensing layer, acting as the five senses of the UAV system, is composed of various types of key sensors and is responsible for collecting perception information about the low-altitude environment in which the UAV is located. Satellite navigation systems can provide absolute position information for UAVs in open airspace, but in complex environments or environments with strong electromagnetic interference, satellite signals are easily blocked, reflected, or abnormally received and interpreted, leading to weakened signal strength, aggravated multipath effects, and impact on the reliability of positioning.

[0116] In this embodiment, the inertial measurement unit (IMU) senses the acceleration and angular velocity of the UAV through accelerometers and gyroscopes, and then calculates the attitude angles (pitch angle, roll angle, yaw angle) and relative displacement changes, exhibiting high-frequency output characteristics. However, as flight time increases, the IMU's measurement data, due to the superposition of factors such as angle drift caused by gyroscope measurement errors and zero-bias errors of accelerometers, leads to a decrease in navigation and positioning accuracy. Without the assistance of an external satellite navigation system to correct the information, the high-precision navigation requirements cannot be met. These two sensors operate based on their unique physical principles, collecting sensing data from different dimensions, providing rich material for subsequent information fusion.

[0117] In this embodiment, the information transmission layer serves as the neural network of the UAV system, responsible for rapidly and stably transmitting the massive amounts of raw data acquired by the sensing layer to the decision processing layer. This embodiment employs a low-latency, reliable wired connection between the sensors and the information transmission layer to ensure the immediacy and accuracy of data transmission, avoiding data loss or errors caused by electromagnetic interference or other factors.

[0118] In this embodiment, the decision processing layer undertakes the core tasks of fusing multi-source data, accurate positioning, and attitude determination. The decision processing layer is based on a high-performance embedded processor or edge computing unit and runs advanced multi-source information fusion algorithms. The decision processing layer includes a data preprocessing unit and a fusion algorithm unit. First, the data preprocessing unit performs noise reduction, calibration, and time synchronization operations on the sensor data from the transmission layer to eliminate errors introduced by sensor characteristics and environmental interference, ensuring data consistency and usability. Next, the fusion algorithm unit performs deep fusion of the preprocessed satellite navigation and inertial navigation multi-source data according to the system's preset fusion strategy. During the fusion process, based on the characteristics of different sensor data, algorithms such as Kalman filtering and its variants are used for linear system state estimation and data fusion, while particle filtering is used to handle strongly nonlinear and non-Gaussian noise scenarios. The algorithms weight and complement each sensor data, fully leveraging their advantages, suppressing error accumulation, and achieving high-precision navigation, positioning, and attitude determination.

[0119] In this embodiment, the sensing layer can integrate dual-core sensors: GNSS and IMU. GNSS is responsible for absolute positioning in open airspace, while IMU is responsible for continuous dead reckoning in signal interruption / complex environments. The information transmission layer uses a low-latency wired connection to ensure that the sensed data is quickly delivered to the decision processing layer, avoiding data loss due to electromagnetic interference. The decision processing layer is equipped with a high-performance embedded processor, which can complete data preprocessing and multi-source fusion calculation in real time. The decision processing layer can use algorithms such as particle filtering to deal with nonlinear and non-Gaussian noise scenarios and adapt to complex environments such as urban canyons, strong electromagnetic interference, and weak textures. Therefore, the environmental adaptability of this solution is better than that of a single-sensor navigation solution.

[0120] Figure 9 The diagram shown illustrates the implementation principle of an adaptive multi-source fusion navigation system based on dynamic scene awareness, according to an embodiment of this application. Figure 9 As shown, the adaptive multi-source fusion navigation system based on dynamic scene perception provided in this embodiment is divided into three layers in terms of architecture: the sensing layer, the information transmission layer, and the decision processing layer. In the sensing layer, positioning and navigation chips (various types of sensors, such as GNSS receivers) and IMU chips (inertial measurement units) are integrated.

[0121] The positioning and navigation core is responsible for acquiring absolute position awareness information (such as satellite signals) of the environment in which the navigation vehicle is located. The IMU core senses the vehicle's acceleration and angular velocity in real time through its internal accelerometer and gyroscope, and then calculates the attitude angle (attitude awareness) and relative displacement change (position awareness). The multi-source data acquired by the perception layer enters the information transmission layer through a stable transmission link. This layer uses the computing core as its hardware core to ensure that the perception data is transmitted to the decision processing layer with low latency and high reliability. After receiving the raw observation data and inertial data sent by the transmission layer, the decision processing layer calls the computing core to run the adaptive multi-source fusion navigation method described in this application to sequentially complete IMU error correction, mechanical arrangement to generate a high-frequency navigation reference benchmark, GNSS double-difference observation and residual extraction, centralized Kalman filter optimal estimation, LAMBDA ambiguity fixing, and error feedback compensation, and finally outputs a high-precision and highly robust fusion navigation result (fusion computation). The entire system works collaboratively through three layers, dynamically sensing scene changes and adaptively adjusting the fusion strategy, effectively improving the positioning accuracy and reliability of UAVs and other vehicles in complex environments.

[0122] The layered distributed architecture adopted by the adaptive multi-source fusion navigation system based on dynamic scene perception in this embodiment has many advantages. On the one hand, each layer has a clear function, is independent yet closely collaborative, which facilitates the modular design, development and maintenance of the system. When a fault occurs in a certain layer, it can be quickly located and repaired without affecting the basic operation of the entire system. On the other hand, through layered processing, the complexity of the system is effectively reduced, and the data processing efficiency and real-time performance are improved, enabling UAVs to respond quickly to external changes in complex and ever-changing low-altitude environments and accurately complete navigation tasks, providing solid and reliable technical support for various applications in the low-altitude economy. In addition, the modular design means that when adding sensors or optimizing algorithms, the entire system does not need to be reconstructed. At the same time, the processing layer uses high-performance computing units to support algorithm iteration, so the system has high reliability, convenient maintenance and strong scalability.

[0123] In summary, this embodiment firstly achieves centimeter-level high-frequency navigation reference by generating a high-frequency navigation reference through online IMU error correction and inertial navigation mechanical orchestration. Combined with GNSS double-difference observation and residual extraction, and utilizing centralized Kalman filtering and LAMBDA ambiguity fixing to form a closed-loop feedback, centimeter-level high-precision positioning is achieved, effectively suppressing inertial navigation cumulative drift. Simultaneously, the system can dynamically perceive environmental changes and maintain continuous and reliable navigation output even in GNSS signal obstruction, highly dynamic, or complex electromagnetic scenarios, demonstrating strong environmental adaptability. Secondly, the system adopts a hierarchical modular architecture with high real-time performance (100Hz-level inertial navigation output, epoch-by-epoch filtering feedback), meeting the agile control requirements of platforms such as UAVs. Furthermore, it relies only on low-cost GNSS / IMU sensors, eliminating the need for expensive equipment such as lidar, thus keeping hardware costs under control and facilitating lightweight integration and large-scale application. Therefore, this application effectively overcomes the various shortcomings of existing technologies and possesses high industrial applicability.

[0124] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. An adaptive multi-source fusion navigation method based on dynamic scene perception, characterized in that, include: Error correction is performed on the inertial measurement data output by the inertial measurement unit to obtain corrected inertial data, and the corrected inertial data is input into the inertial navigation mechanical orchestration module to generate a navigation reference reference. Based on the observation data of the base station and the rover station, double-input observations are obtained. The double-difference geometric distance between the station and the satellite is calculated using the position information output by the inertial navigation system. The residual data between the double-input observations and the double-difference geometric distance is obtained. The residual data is input into a Kalman filter, and the Kalman filter outputs the optimal estimate. The optimal estimate is fed back to the inertial measurement unit for compensation, and the navigation reference is corrected to output the final navigation data.

2. The adaptive multi-source fusion navigation method based on dynamic scene perception according to claim 1, characterized in that, The step of performing error correction on the inertial measurement data output by the inertial measurement unit and obtaining corrected inertial data includes: The error of the inertial measurement data is estimated in real time using the Kalman filter; Based on the error, the original angular increment and velocity increment output by the inertial measurement unit are compensated online to obtain corrected inertial data.

3. The adaptive multi-source fusion navigation method based on dynamic scene perception according to claim 2, characterized in that, The errors in the inertial measurement data include gyroscope bias, accelerometer bias, and scaling factor error.

4. The adaptive multi-source fusion navigation method based on dynamic scene perception according to claim 1 or 3, characterized in that, The step of inputting the corrected inertial data into the inertial navigation mechanical orchestration module to generate a navigation reference reference includes: The inertial navigation mechanical orchestration module integrates the angular velocity information from the input gyroscope, calculates the attitude transformation matrix of the carrier coordinate system relative to the navigation coordinate system in real time, and generates the attitude reference benchmark of the navigation carrier. The inertial navigation mechanical orchestration module integrates the specific force information from the input accelerometer, updates the velocity components of the navigation vehicle in the navigation coordinate system in real time, and generates a velocity reference benchmark for the navigation vehicle. The inertial navigation mechanical orchestration module performs a second integral operation on the updated velocity components to generate a position reference benchmark for the navigation vehicle.

5. The adaptive multi-source fusion navigation method based on dynamic scene perception according to claim 1, characterized in that, The acquisition of double-interpolated observations based on observation data from the base station and the rover includes: Based on the raw observation data from the base station and the rover, a reference satellite is selected to construct a double-difference observation value that includes carrier phase double-difference observation values ​​and / or pseudorange double-difference observation values; wherein, the double-difference observation values ​​are used to eliminate satellite clock errors, receiver clock errors, and most atmospheric delay errors.

6. The adaptive multi-source fusion navigation method based on dynamic scene perception according to claim 5, characterized in that, The calculation of the double-difference geometric distance between stations and satellites using the position information output by the inertial navigation system includes: Using the rover position information output by the inertial navigation system, combined with the known reference station position information and the satellite spatial position provided by the satellite ephemeris, the difference between the station-satellite geometric distances of the rover and the reference station to the same satellite is calculated, and then the double difference geometric distances corresponding to the reference satellite and non-reference satellites are constructed.

7. The adaptive multi-source fusion navigation method based on dynamic scene perception according to claim 1, characterized in that, The residual data includes at least the residual atmospheric delay error, satellite orbit error, and carrier phase integer ambiguity; The step of inputting the residual data into a Kalman filter and having the Kalman filter output the optimal estimate includes: The residual data is used to construct a measurement vector for a Kalman filter; Establish the measurement equations of the Kalman filter, which are used to describe the mapping relationship between the measurement vector and the system state vector; Using the output of the inertial navigation system's mechanical orchestration module as the basis for state transition, the measurement vector is injected into the Kalman filter to perform measurement updates and obtain the optimal estimate of the state vector. The optimal estimate includes at least: the optimal estimate of position error, the optimal estimate of velocity error, the optimal estimate of attitude error, the optimal estimate of inertial sensor zero bias, and the optimal estimate of carrier phase ambiguity floating-point solution.

8. The adaptive multi-source fusion navigation method based on dynamic scene perception according to claim 7, characterized in that, Also includes: The optimal estimate of the floating-point solution of the carrier phase ambiguity is input into a preset ambiguity search module, which outputs integer ambiguity based on an integer least squares search algorithm.

9. The adaptive multi-source fusion navigation method based on dynamic scene perception according to claim 7, characterized in that, The step of feeding back the optimal estimate to the inertial measurement unit for compensation and correcting the navigation reference reference includes: The optimal zero-bias estimate of the inertial sensor is fed back to the inertial measurement unit for compensation. The optimal estimates of the position error, velocity error, and attitude error are used to correct the navigation reference reference generated by the inertial navigation mechanical orchestration module.

10. An adaptive multi-source fusion navigation system based on dynamic scene perception, characterized in that, include: The sensory layer, information transmission layer, and decision processing layer; The sensing layer includes various types of sensors and inertial measurement units; The various types of sensors collect perception information about the environment in which the navigation vehicle is located. The inertial measurement unit senses the acceleration and angular velocity of the navigation vehicle through accelerometers and gyroscopes, and then calculates the attitude angle and relative displacement changes. The information transmission layer transmits the data acquired by the sensing layer to the decision processing layer; The decision processing layer uses the adaptive multi-source fusion navigation method based on dynamic scene perception as described in any one of claims 1 to 9 for navigation decision processing.

Images