A Multi-Source Fusion Navigation and Positioning Method for Unmanned Aerial Vehicles that Does Not Rely on Satellite Signals

Abstract

This invention discloses a multi-source fusion navigation and positioning method for unmanned aerial vehicles (UAVs) that does not rely on satellite signals, belonging to the field of UAV docking navigation technology. The method includes the following steps: Step 1: Construction of the fusion navigation and positioning system; Step 2: Time registration and outlier removal; Step 3: Solving for information allocation coefficients; Step 4: Time-varying measurement noise estimation; Step 5: Time and measurement update; Step 6: Master filter fusion and differentiated feedback. This solution addresses the problems of insufficient sensor redundancy, rigid information allocation in multi-source fusion navigation algorithms, and inflexible fault isolation mechanisms in existing docking navigation and positioning systems, thereby improving the accuracy and anti-interference capability of UAV navigation and positioning.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) docking and navigation technology, and in particular to a multi-source fusion navigation and positioning method for UAVs that does not rely on satellite signals. Background Technology

[0002] Precision docking missions such as aerial refueling of unmanned aerial vehicles (UAVs) place extremely high demands on the accuracy, reliability, and anti-interference capabilities of navigation systems, representing one of the key technological challenges in the aviation field. While existing technologies such as Global Navigation Satellite Systems (GNSS) and their enhancement technologies (e.g., real-time dynamic carrier phase differential technology) can provide UAVs with meter- to centimeter-level positioning services in outdoor environments, their signals are susceptible to obstruction, ionospheric interference, and malicious spoofing. During critical docking phases, signal instability or even loss may occur, leading to a sharp decline in positioning accuracy or insufficient update frequency, making it difficult to meet the precise docking requirements of cone-shaped docking. To reduce reliance on GNSS, single-sensor solutions such as vision or lidar have been introduced. Vision solutions can obtain relative pose through camera imaging and feature recognition (e.g., identifying cone-shaped targets), but their computational process involves image preprocessing, feature matching, and coordinate transformation, making it complex. Furthermore, it is extremely sensitive to changes in lighting, strong light, cloud cover, and scene texture (e.g., weakly textured environments), resulting in insufficient stability in complex aerial environments. While lidar can provide accurate point clouds, it is also susceptible to interference from atmospheric conditions such as clouds and haze, and has limitations such as large weight and power consumption.

[0003] For multi-source data fusion, the traditional Kalman filter method is typically used. This method employs a centralized processing architecture, which carries a high risk of single-point failures and lacks built-in isolation mechanisms. While classic federated filtering algorithms offer some improvement in fault isolation, they still suffer from the problem of inflexible information distribution and fault isolation.

[0004] Therefore, it is necessary to develop a navigation and positioning method and device that does not rely on satellite signals, can adaptively fuse multi-source information, and has strong fault tolerance, so as to improve the reliability of UAVs in completing high-precision aerial docking in complex environments. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-source fusion navigation and positioning method for UAVs that does not rely on satellite signals, addressing the aforementioned shortcomings. This method solves problems such as insufficient sensor redundancy, rigid information allocation and fault isolation mechanisms in existing docking navigation and positioning systems, thereby improving the accuracy and anti-interference capability of UAV navigation and positioning.

[0006] This invention is achieved through the following scheme:

[0007] A multi-source fusion navigation and positioning method for unmanned aerial vehicles (UAVs) that does not rely on satellite signals includes the following steps:

[0008] Step 1: Building an integrated navigation and positioning system;

[0009] Step 2: Time registration and outlier removal; In each filtering cycle, the time registration of sensor data is first performed to uniformly register the measurement data of each asynchronous sensor to the timestamp of the current filtering cycle. Then, outlier detection and removal are performed on the registered measurement data based on the chi-square test.

[0010] Step 3: Solve for the information allocation coefficients; build federated sub-filters for each sensor, and calculate the information allocation coefficients by using the trace of the covariance output of the sub-filters; calculate the singular value matrix of the covariance, and perform block reset on the information allocation coefficients, retaining only the variance of each state variable in the covariance.

[0011] Step 4: Time-varying measurement noise estimation; Construct a sliding window by comparing the actual innovation residual covariance with the theoretical residual covariance, use the residual sequence estimate within the window to determine the fault degree of the sub-filter, and adjust the measurement noise of each sub-filter accordingly.

[0012] Step 5: Time and Measurement Update; The relative position and velocity of the carrier reference point and the cone sleeve are taken as the common state of the system, and each sub-filter independently updates the time and measurement using the common state;

[0013] Step 6: Main filter fusion and differentiated feedback; The main filter receives the local optimal estimation results from all sub-filters, calculates the global optimal estimation according to the information fusion criterion, and uses the information allocation coefficients obtained in Step 3 to provide navigation state feedback to each sub-filter.

[0014] In step 1, vision and lidar are used as measurement sensors for the multi-source fusion navigation and positioning system, and the navigation and positioning fusion processing terminal is used as the navigation and positioning solution unit to build the multi-source fusion navigation and positioning system.

[0015] In step 2, specifically, within each filtering cycle, time registration is first performed. Using the system's unified clock as a reference, the asynchronously acquired data from each sensor are synchronized in time. Linear interpolation is then used to uniformly register the measurement data from each sensor to the current filtering cycle timestamp. Secondly, outlier removal is performed based on the chi-square test, and the normalized squared information statistic for each sensor is calculated.

[0016]

[0017] in To measure the residual, Let be the residual covariance matrix; if , If the value is the critical value of the chi-square distribution, then the measurement data of the sensor in this filtering cycle is determined to be an outlier and is discarded, and will not participate in the subsequent filtering calculations in this cycle.

[0018] In step 3, specifically, the system state vector is defined as the position and velocity relative to the relative cone sleeve, i.e. The position information is based on a conical coordinate system; an independent sub-filter is constructed for each sensor, and the state vector of each filter is initialized. Covariance Matrix ; Sub-filter in Covariance matrix at time step according to Perform singular value decomposition to obtain The singular value matrix.

[0019] The singular value matrix is:

[0020]

[0021] In the formula Let represent the singular value matrix corresponding to the state vector. Based on the singular value matrix and the trace calculation, the allocation coefficients under the block reset condition are obtained:

[0022] ,

[0023] In the formula, Indicates the first Calculate the information allocation matrix based on the allocation coefficients corresponding to each state vector. :

[0024] ,

[0025] In the formula, Its dimension is consistent with that of the state vector.

[0026] In step 4, specifically, a fixed length is defined. A sliding window used to store the most recent Measurement updates at each moment and its theoretical covariance , As the initial noise value, each time a new measurement is obtained, the latest ( , Store the data in the window to cover the original data; calculate the sample covariance matrix within the window as an estimate of the equivalent innovation covariance.

[0027] The estimation of the equivalent innovation covariance is as follows:

[0028]

[0029] Let the sample estimate equal to theoretical value Thus, the solution is obtained. One sensor in Time measurement noise covariance matrix The estimated value:

[0030] .

[0031] In step 5, specifically, after obtaining the allocation information, each sub-filter performs time updates using a local system model. Sub-filters, state prediction, and error covariance prediction:

[0032]

[0033] In the formula, Represents the state transition matrix. This represents the process noise driving matrix; when each sub-filter obtains sensor measurement data... At that time, perform independent Kalman measurement updates:

[0034]

[0035] In the formula, For Kalman gain, This step involves incorporating new sensor information into the respective state estimates to obtain updated local optimum estimates. and .

[0036] In step 6, specifically, the main filter receives the local optimal results from all sub-filters and calculates the global optimal estimate based on the following information fusion criteria;

[0037]

[0038] The information is reset based on the allocation matrix in step 3, and the information allocation rules are as follows:

[0039]

[0040] In the formula, Represents the global error covariance matrix. and These represent the process noise covariance matrix of the sub-filter and the global noise covariance matrix of the main filter, respectively.

[0041] In step 1, a visual camera and a lidar are specifically used.

[0042] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0043] 1. The fault tolerance and anti-interference capability of the navigation and positioning system are significantly enhanced: This invention replaces the original sensor combination with vision and lidar, uses a high-performance navigation and positioning fusion processing terminal as the navigation calculation unit, and adopts sliding window time-varying noise estimation as the fault judgment and processing mechanism of the navigation and positioning fusion method, which significantly enhances the fault tolerance and anti-interference capability of the navigation and positioning system.

[0044] 2. Significantly Improved Docking Accuracy of Navigation and Positioning Systems: The method of this invention uses lidar as a replacement for GNSS sensors, which offers superior measurement accuracy. This invention achieves more refined information allocation through solving the singular value matrix of the covariance and a block-based reset strategy. Experimental results show that in UAV cone docking missions, the navigation accuracy of this invention is less than 2 cm under normal operating conditions and less than 5 cm under strong interference conditions.

[0045] 3. Excellent real-time performance: The method of this invention has the characteristics of low computational complexity. The device of this invention uses a high-performance edge computing platform as the fusion navigation and positioning solution unit. The navigation system updates at a frequency of no less than 50Hz, which can ensure the real-time perception and compensation of sensor anomalies during the process of the UAV rapidly approaching the cone sleeve to achieve docking.

[0046] 4. This solution provides a vision / LiDAR fusion navigation and positioning method based on improved federated filtering, which is suitable for high-reliability autonomous navigation of UAVs that achieve precise docking with the drogue, such as during aerial refueling.

[0047] 5. This invention improves the information allocation coefficient calculation and noise estimation methods of the classical federated filtering algorithm by using the covariance singular value matrix and sliding window, thereby improving the fault tolerance and adaptability of the navigation system. Attached Figure Description

[0048] Figure 1 This is a complete flowchart of the multi-source fusion navigation and positioning method described in this invention;

[0049] Figure 2 This is a comparison chart of sensor data and fusion results in the X direction using the present invention and the traditional method under interference scenarios in the embodiment.

[0050] Figure 3 This is a comparison chart of sensor data and fusion results in the Y direction using the present invention and the traditional method under interference scenarios in the embodiment.

[0051] Figure 4This is a comparison chart of sensor data and fusion results in the Z direction using the present invention and the traditional method under interference scenarios in the embodiment.

[0052] Figure 5 This is a schematic diagram of the working scenario described in this invention. Detailed Implementation

[0053] All features disclosed in this specification, or all steps in all disclosed methods or processes, may be combined in any way, except for mutually exclusive features and / or steps.

[0054] Any feature disclosed in this specification (including any appended claims and abstract) may be replaced by other equivalent or similar features, unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is merely one example of a series of equivalent or similar features.

[0055] In the description of this invention, it should be understood that the terms "upper," "lower," "left," "right," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a predetermined orientation, or be constructed and operated in a predetermined orientation. Therefore, they should not be construed as limitations on this invention.

[0056] Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature.

[0057] Example 1

[0058] like Figures 1-4 As shown, the present invention provides a technical solution:

[0059] A multi-source fusion navigation and positioning method for unmanned aerial vehicles (UAVs) that does not rely on satellite signals includes the following steps:

[0060] Step 1, Building a Fusion Navigation and Positioning System: For close-range docking, relying solely on GNSS and vision is insufficient to guarantee reliable input of the fusion data source. This invention uses vision and lidar as measurement sensors for the multi-source fusion navigation and positioning system, and a navigation and positioning fusion processing terminal as the navigation and positioning solution unit to build a multi-source fusion navigation and positioning system.

[0061] Step 2, Time Registration and Outlier Removal: Within each filtering cycle, time registration of sensor data is first performed to uniformly register the measurement data of each asynchronous sensor to the current filtering cycle timestamp. Then, outlier detection and removal are performed on the registered measurement data based on the chi-square test to improve the data input quality.

[0062] Step 3, Solve for information allocation coefficients: Build federated sub-filters for each sensor. The original method calculates the information allocation coefficients by using the trace of the covariance output of the sub-filters. This method calculates the singular value matrix of the covariance based on the original method and performs block reset on the information allocation coefficients. The block reset only retains the variance of each state variable in the covariance.

[0063] Step 4, Time-varying measurement noise estimation: The original method uses discrete and switching control to isolate faults, which has poor adaptability. This method constructs a sliding window by comparing the actual information residual covariance with the theoretical residual covariance. It uses the residual sequence estimate within the window to determine the fault degree of the sub-filter and adjusts the measurement noise of each sub-filter accordingly, so that this method has local adaptive capability.

[0064] Step 5, Time and Measurement Update: This method uses the relative position and velocity of the carrier reference point and the cone sleeve as the common state of the system, and each sub-filter independently updates the time and measurement using the common state.

[0065] Step 6, main filter fusion and differentiated feedback: The main filter receives the local optimal estimation results from all sub-filters, calculates the global optimal estimation according to the information fusion criterion, and combines the information allocation coefficients obtained in step 3 to provide navigation state feedback to each sub-filter.

[0066] Example 2

[0067] like Figure 5 As shown, the present invention provides a technical solution:

[0068] Taking a drone performing a cone-shaped docking mission as an example, the implementation process of this method in the docking mission is as follows:

[0069] Step 1, Construction of the integrated navigation and positioning system: Specifically, it is equipped with a visual camera and a LiDAR for high-precision docking;

[0070] Step 2, time registration and outlier removal;

[0071] Within each filtering cycle, time registration is first performed. Using the system's unified clock as a reference, the asynchronously acquired data from each sensor are synchronized. Linear interpolation is then used to uniformly register the sensor measurements to the current filtering cycle timestamp. Secondly, outlier removal is performed based on the chi-square test, and the normalized squared information statistic for each sensor is calculated.

[0072]

[0073] in To measure the residual, Let be the residual covariance matrix. , If the value is the critical value of the chi-square distribution, the measurement data of the sensor in this filtering cycle is determined to be an outlier and is discarded, and will not participate in the subsequent filtering calculations in this cycle.

[0074] Step 3: System initialization and solution of information allocation coefficients;

[0075] Define the system state vector as the position and velocity relative to the relative cone sleeve, i.e. The position information is based on a conical coordinate system. An independent sub-filter is constructed for each sensor, and the state vector of each filter is initialized. Covariance Matrix Sub-filters in Covariance matrix at time step according to Perform singular value decomposition to obtain Singular value matrix:

[0076]

[0077] In the formula Let represent the singular value matrix corresponding to the state vector. Based on the singular value matrix and the trace calculation, the allocation coefficients under the block reset condition are obtained:

[0078] ,

[0079] In the formula, Indicates the first Calculate the information allocation matrix based on the allocation coefficients corresponding to each state vector. :

[0080] ,

[0081] In the formula, Its dimension is consistent with that of the state vector;

[0082] Step 4, measurement noise estimation;

[0083] Define a fixed length as A sliding window used to store the most recent Measurement updates at each moment and its theoretical covariance , As the initial noise value, each time a new measurement is obtained, the latest ( , The data is stored in the window, overwriting the existing data. The sample covariance matrix within the window is calculated as an estimate of the equivalent innovation covariance.

[0084]

[0085] Let the sample estimate equal to theoretical value Therefore, the solution can be found in the first... One sensor in Time measurement noise covariance matrix The estimated value:

[0086] .

[0087] Step 5, Time and Measurement Update;

[0088] After obtaining the allocation information, each sub-filter performs time updates using a local system model. For the first... Sub-filters, state prediction, and error covariance prediction:

[0089]

[0090] In the formula, Represents the state transition matrix. This represents the process noise driving matrix. When each sub-filter acquires sensor measurement data... At that time, perform independent Kalman measurement updates:

[0091]

[0092] In the formula, For Kalman gain, This is the measurement matrix. This step incorporates the new information from the sensors into their respective state estimates, resulting in an updated local optimum estimate. and .

[0093] Step 6: Information fusion and differentiated feedback;

[0094] The main filter receives the local optimal results from all sub-filters and calculates the global optimal estimate based on the following information fusion criteria.

[0095]

[0096] The information is reset based on the allocation matrix in step 3, and the information allocation rules are as follows:

[0097]

[0098] In the formula, Represents the global error covariance matrix. and These represent the process noise covariance matrix of the sub-filter and the global noise covariance matrix of the main filter, respectively.

[0099] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-source fusion navigation and positioning method for unmanned aerial vehicles (UAVs) that does not rely on satellite signals, characterized in that: Includes the following steps: Step 1: Building a fusion navigation and positioning system; specifically, using vision and lidar as measurement sensors for the multi-source fusion navigation and positioning system, and a navigation and positioning fusion processing terminal as the navigation and positioning solution unit, to build a multi-source fusion navigation and positioning system; Step 2: Time registration and outlier removal; Within each filtering cycle, the time registration of sensor data is first performed, and the measurement data of each asynchronous sensor are uniformly registered to the timestamp of the current filtering cycle. Then, outlier detection and removal are performed on the registered measurement data based on the chi-square test. Step 3: Solve for the information allocation coefficients; build federated sub-filters for each sensor, and calculate the information allocation coefficients by using the trace of the covariance output of the sub-filters; calculate the singular value matrix of the covariance, and perform block reset on the information allocation coefficients, retaining only the variance of each state variable in the covariance. Specific; Define the system state vector as the position and velocity relative to the relative cone sleeve, i.e. The position information is based on the cone-shaped coordinate system; Construct an independent sub-filter for each sensor and initialize the state vector of each filter. Covariance Matrix ; Sub-filter in Covariance matrix at time step according to Perform singular value decomposition to obtain The singular value matrix; The singular value matrix is: In the formula Let represent the singular value matrix corresponding to the state vector. Based on the singular value matrix and the trace calculation, the allocation coefficients under the block reset condition are obtained: ， In the formula, Indicates the first Calculate the information allocation matrix based on the allocation coefficients corresponding to each state vector. : ， In the formula, Its dimension is consistent with that of the state vector; Step 4: Time-varying measurement noise estimation; Construct a sliding window by comparing the actual innovation residual covariance with the theoretical residual covariance, use the residual sequence estimate within the window to determine the fault degree of the sub-filter, and adjust the measurement noise of each sub-filter accordingly. Step 5: Time and Measurement Update; The relative position and velocity of the carrier reference point and the cone sleeve are taken as the common state of the system, and each sub-filter independently updates the time and measurement using the common state; Step 6: Main filter fusion and differentiated feedback; The main filter receives the local optimal estimation results from all sub-filters, calculates the global optimal estimation according to the information fusion criterion, and uses the information allocation coefficients obtained in Step 3 to provide navigation state feedback to each sub-filter.

2. The UAV multi-source fusion navigation and positioning method that does not rely on satellite signals as described in claim 1, characterized in that: In step 2, specifically, within each filtering cycle, time registration is first performed. Using the system's unified clock as a reference, the asynchronously acquired data from each sensor are synchronized in time. Linear interpolation is then used to uniformly register the measurement data from each sensor to the timestamp of the current filtering cycle. Secondly, outlier removal is performed based on the chi-square test, and the normalized squared information statistic for each sensor is calculated: in To measure the residual, Let be the residual covariance matrix; if , If the value is the critical value of the chi-square distribution, then the measurement data of the sensor in this filtering cycle is determined to be an outlier and is discarded, and will not participate in the subsequent filtering calculations in this cycle.

3. The UAV multi-source fusion navigation and positioning method that does not rely on satellite signals as described in claim 1, characterized in that: In step 4, specifically, a fixed length is defined. A sliding window used to store the most recent Measurement updates at each moment and its theoretical covariance , As the initial noise value, each time a new measurement is obtained, the latest ( , Store the data in the window to cover the original data; calculate the sample covariance matrix within the window as an estimate of the equivalent innovation covariance.

4. The UAV multi-source fusion navigation and positioning method that does not rely on satellite signals as described in claim 3, characterized in that: The estimation of the equivalent innovation covariance is as follows: Let the sample estimate equal to theoretical value Thus, the solution is obtained. One sensor in Time measurement noise covariance matrix The estimated value: 。 5. The UAV multi-source fusion navigation and positioning method that does not rely on satellite signals as described in claim 1, characterized in that: In step 5, specifically, after obtaining the allocation information, each sub-filter performs time updates using a local system model. Sub-filters, state prediction, and error covariance prediction: In the formula, Represents the state transition matrix. Represents the process noise driving matrix; When each sub-filter obtains sensor measurement data At that time, perform independent Kalman measurement updates: In the formula, For Kalman gain, For measurement matrix; This step incorporates the new information from the sensors into their respective state estimates, resulting in updated local optimum estimates. and .

6. The UAV multi-source fusion navigation and positioning method that does not rely on satellite signals as described in claim 1, characterized in that: In step 6, specifically, the main filter receives the local optimal results from all sub-filters and calculates the global optimal estimate based on the following information fusion criteria; The information is reset based on the allocation matrix in step 3, and the information allocation rules are as follows: In the formula, Represents the global error covariance matrix. and These represent the process noise covariance matrix of the sub-filter and the global noise covariance matrix of the main filter, respectively.

7. The UAV multi-source fusion navigation and positioning method that does not rely on satellite signals as described in claim 1, characterized in that: In step 1, a visual camera and a lidar are specifically used.

Images