A bionic tight combination of heading and attitude measurement method based on sequential cubature Kalman filter
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HARBIN INST OF TECH
- Filing Date
- 2023-12-19
- Publication Date
- 2026-07-03
Smart Images
Figure CN117824644B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of integrated navigation technology, and more specifically, relates to a biomimetic compact combination attitude measurement method based on sequential capacitive Kalman filtering. Background Technology
[0002] Although polarization vision sensors can provide reliable heading information for navigation systems in scenarios such as GNSS rejection and magnetic interference, they are easily affected by factors such as obstacle obstruction, weather changes, and lighting conditions. A single sensor is often insufficient to meet the requirements of full autonomy. Therefore, in recent years, scholars at home and abroad have carried out research on combined attitude measurement technology based on biomimetic polarization vision.
[0003] In 2003, the University of Zurich in Switzerland built a combined positioning system based on an odometer, polarization compass, and panoramic camera. It used a six-channel point-source polarization sensor to detect heading, correcting the accumulated error of the odometer, and a panoramic camera to detect landmarks. Even after the robot had traveled hundreds of meters, it could still achieve homing accuracy within 1 meter. In 2008, the University of Rome Biomedical University in Italy designed a combined system based on an inertial sensor, magnetometer, and polarization compass, and used complementary filters for data fusion and attitude estimation. In 2015, Beijing Institute of Technology used polarization heading and odometer velocity as measurement information and used the unscented Kalman (UKF) method to estimate the system state, effectively correcting the accumulated error of the MEMS-IMU. In 2017, Beijing University of Aeronautics and Astronautics used a point-source polarization sensor to assist the SINS / GPS / geomagnetic system, achieving redundancy in heading information, and adopted real-time fault detection and isolation technology based on Kalman filtering to improve the system's fault tolerance. Also in 2017, the German Aerospace Center (DLR) combined an image-based polarization sensor with... In 2020, Beijing University of Aeronautics and Astronautics (BUAA) used three point-source polarization sensors to determine the polarization heading and achieved dual-vector attitude determination using the gravity vector of the inertial navigation system and the solar vector of the polarization sensor. In 2021, the National University of Defense Technology (NUDT) built a biomimetic multi-source navigation system using a binocular camera, MIMU, three-axis magnetometer, and polarization sensor. A graph optimization method was used to fuse the dead reckoning attitude, magnetometer heading, and polarization heading from the visual-inertial system, improving the robustness of the navigation system. In 2022, North China University of Technology (NCT) used a multi-rate strong tracking square root capacitive Kalman filter to fuse MEMS-IMU / GPS / polarization compass, enabling adaptive estimation of system covariance and overcoming the problem of inconsistent sampling frequencies among different sensors. Also in 2022, BUAA proposed a multi-modal switching variational Bayesian adaptive Kalman filter algorithm for the SINS / polarization sensor / geomagnetic combined system, designing different filtering modes for different interference conditions to improve system robustness. Furthermore, in 2023, under full moonlight conditions, the team used an image-based polarization sensor to assist an inertial / star-sensor hybrid system. Even when the star sensor malfunctioned, it still provided usable heading information, offering a reference for autonomous navigation in complex nighttime environments. While the aforementioned loosely coupled method can combine the advantages of polarization sensors and other navigation sensors to achieve better navigation performance, and heading accuracy when polarization information is interfered with can be improved by studying fusion algorithms, further research is needed.
[0004] However, traditional methods still have the following drawbacks: (1) The polarization heading used in loosely coupled methods has undergone a series of nonlinear transformations or complex image feature extraction operations, and it is difficult to guarantee the reliability of effective sky region extraction; (2) Loosely coupled methods are difficult to adapt to slowly changing polarization heading errors caused by occlusion interference; (3) Traditional methods do not consider the effect of acceleration information on the carrier attitude correction; (4) There is no suitable information fusion mechanism for nonlinear tightly coupled systems. Summary of the Invention
[0005] The purpose of this invention is to provide a biomimetic tightly coupled attitude measurement method based on sequential capacitive Kalman filtering, which aims to solve the technical problem that loosely coupled methods in the prior art are easily affected by occlusion interference and atmospheric depolarization effect.
[0006] To achieve the above objectives, the technical solution adopted by this invention is: to provide a biomimetic compact combination attitude measurement method based on sequential capacitive Kalman filtering, comprising:
[0007] Step 1: Construct the system state equations for the attitude measurement system;
[0008] Step 2: Obtain the error AOP value based on the system state equation and adjust the attitude measurement system state after error correction;
[0009] Step 3: Construct attitude measurement equations to obtain attitude parameters;
[0010] Step 4: Fuse the attitude and attitude parameters and the adjusted AOP values of the attitude measurement system to obtain the fused attitude and attitude measurement system parameters;
[0011] Step 5: Filter the polarization visual measurement channels to obtain the filtered polarization visual measurement channels;
[0012] Step 6: Construct a new polarization vision sensor measurement equation based on the selected polarization vision measurement channels and the fused attitude measurement system parameters;
[0013] Step 7: Obtain new attitude parameters based on the new polarization vision sensor measurement equation, and obtain attitude measurement results based on the new attitude parameters and the state of the attitude measurement system after error adjustment of the AOP value.
[0014] Optionally, step 1 includes:
[0015] Step 1.1: Adjust the solar zenith deviation for n frames. and azimuth deviation As a state vector;
[0016] Step 1.2: Construct the system state equations of the attitude measurement system based on the state vectors;
[0017] The state vector is:
[0018]
[0019] Where, φ=[φ E ,φ N ,φ U ] T For the platform misalignment angle of the IMU, G b =[G bx G by G bz ] T This is the scaling factor error vector of the gyroscope. This is due to the solar zenith deviation. This refers to the azimuth deviation;
[0020] The system state equation of the attitude measurement system is:
[0021]
[0022]
[0023]
[0024]
[0025] in, Let Φ represent the system state of the attitude measurement system, Φ be the error differential equations for the solar azimuth and zenith angles, and w be the Gaussian noise. Let be the projection of the rotational angular rate of the n-frame relative to the i-frame onto the n-frame.
[0026] Optionally, the attitude parameters obtained in step 3 include:
[0027] The attitude measurement parameters include the compensated observation vector, the corrected solar vector, the AOP estimate, and the error-adjusted gravity vector estimate and measurement.
[0028] Optionally, the steps in step 3 to obtain the attitude parameters include:
[0029] Step 3.1: Construct an AOP compensation model and obtain the compensated observation vector based on the AOP compensation model;
[0030] Step 3.2: Obtain AOP and DOP estimates based on the current attitude and heading of the INS, and construct a multi-channel polarization visual measurement equation based on the AOP and DOP estimates;
[0031] Step 3.3: Construct the acceleration-based gravity vector measurement equation, and obtain the error-adjusted gravity vector estimate and measurement value based on the acceleration-based gravity vector measurement equation.
[0032] Optionally, step 3.1, which involves obtaining the compensated observation vector, includes:
[0033] Step 3.1.1: Obtain the calibration parameters and distortion parameters of the fisheye camera;
[0034] Step 3.1.2: Establish an AOP compensation model under tilted conditions;
[0035] Step 3.1.3: Obtain the compensated observation vector based on the calibration parameters and distortion parameters of the fisheye camera and the AOP compensation model under tilt conditions;
[0036] The compensated observation vector is:
[0037]
[0038]
[0039]
[0040] Where, γ n α is the zenith angle of the compensated observation vector in the n-system. n The azimuth angle of the compensated observation vector in the n-system. Represents a three-dimensional vector in the n-system. Let ro be a three-dimensional vector in the b-system, ro be the roll angle of the polarization vision sensor, and p be the yaw rate. i y is the pitch angle of the polarization vision sensor, ya is the yaw angle of the polarization sensor, (3) is the third element, (2) is the second element, and (1) is the first element.
[0041] Optionally, step 3.2 includes:
[0042] Step 3.2.1: Obtain the deviation between the direction cosine matrix and the true value, and define the deviation as... in,
[0043] Step 3.2.2: Obtain the observation vector in the n-system corresponding to the m-th pixel after misalignment angle correction based on the AOP compensation model under bias and tilt conditions.
[0044] Step 3.2.3: Obtain the corrected solar vector based on the solar angle azimuth error and zenith angle error in the state vector;
[0045] Step 3.2.4: Obtain the estimated value of AOP based on the corrected solar vector and Rayleigh scattering model;
[0046] Step 3.2.5: Obtain the DOP estimate based on the scattering angle between the observed vector and the solar vector;
[0047] Step 3.2.6: Construct a multi-channel polarization visual measurement equation based on the DOP and AOP estimates;
[0048] The estimated value of AOP is:
[0049]
[0050] in, and This represents the azimuth and zenith angle correction values for the observed vector in the n-system. This is an estimate of AOP;
[0051] The scattering angle between the observed vector and the solar vector is:
[0052]
[0053] Where, θ m The scattering angle between the observed vector and the solar vector. This is the observation vector in the n-system corresponding to the m-th pixel after misalignment correction. The observation direction corresponding to pixel S in the polarization image;
[0054] The estimated value of DOP is:
[0055]
[0056] Among them, DOP max This represents the maximum value measured by the polarization sensor.
[0057] The multi-channel polarization vision measurement equation is:
[0058]
[0059] in, Measurement noise for a single polarization channel. These are measurements from a polarization vision sensor. This is an estimate of AOP.
[0060] Optionally, step 3.3 includes:
[0061] Step 3.3.1: Calculate the residual of the gravity vector;
[0062] Step 3.3.2: Obtain the acceleration and gravity vector measurement equation based on the residual of the gravity vector;
[0063] Step 3.3.3: Obtain the error-adjusted estimated and measured values of the gravity vector based on the acceleration-based gravity vector measurement equation;
[0064] The residual of the gravity vector is:
[0065]
[0066] Among them, g b Let g be the accelerometer gravity vector in the b-frame. n This represents the gravitational vector in the n-frame at the current location. The platform is out of alignment.
[0067] The equation for the acceleration-to-gravity vector measurement is:
[0068] Z g =δg=H g X+V g
[0069]
[0070] Among them, Z g To record the acceleration as a measurement of the gravitational vector, V g X represents the measurement noise of the gravity vector, and X represents the state vector.
[0071] Optionally, step 4 includes:
[0072] Based on AOP and DOP measurements, a reliable polarization measurement channel is selected. Combined with a robust fusion mechanism based on sequential filtering and gravity vector measurement equations, the system attitude is corrected to obtain the fused attitude measurement system parameters.
[0073] Optionally, step 5 includes:
[0074] Step 5.1: Set the AOP threshold and DOP threshold, where the AOP threshold is... The DOP threshold is
[0075] Step 5.2: Obtain the discrimination criteria for the measurement channel based on the AOP threshold and DOP threshold;
[0076] Step 5.3: Based on the discrimination criteria of the measurement channel, the polarization channel is filtered to obtain the filtered polarization channel;
[0077] The criteria for determining the measurement channel are:
[0078]
[0079] Where, ε a To measure the polarization angle threshold of the channel, ε d The polarization threshold of the measurement channel.
[0080] The new polarization vision sensor measurement equation in step 5 is:
[0081]
[0082] Among them, Z pol For the new polarization vision sensor measurement model, Measurement noise for a single polarization channel. is the AOP threshold, and h(X) is the measurement parameter of the polarization vision sensor.
[0083] The beneficial effects of the biomimetic tightly coupled attitude measurement method based on sequential capacitive Kalman filtering provided by this invention are as follows: Compared with existing technologies, this invention proposes a tightly coupled model based on all-sky polarization information, fusing attitude information from image-based polarization sensors and MEMS, thus improving the robustness of the system and avoiding nonlinear transformations or complex feature extractions of heading solutions in loosely coupled methods. This invention also proposes a robust fusion mechanism based on sequential capacitive Kalman filtering, improving the accuracy of the effective sky observation area. Robust information fusion from gyroscopes, accelerometers, and polarization sensors enhances the robustness of the combined system under complex weather conditions and obstruction interference. This invention constructs a polarization / MEMS tightly coupled attitude measurement device based on sequential capacitive Kalman filtering, and verifies the method through outdoor dynamic rotation tests and vehicle-mounted tests. Furthermore, the sensors used in this invention are all autonomous sensors, capable of providing high-precision attitude reference information without relying on satellites. Attached Figure Description
[0084] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0085] Figure 1 This is a schematic diagram of the AOP compensation model under tilted conditions provided in an embodiment of the present invention.
[0086] Figure 2 This is a schematic diagram of multi-channel polarization measurement of a tightly coupled model provided in an embodiment of the present invention.
[0087] Figure 3 A flowchart of a robust fusion algorithm based on a tightly coupled PS / MIMU system provided for embodiments of the present invention.
[0088] Figure 4 A diagram of the transposition test equipment provided in an embodiment of the present invention.
[0089] Figure 5 The heading measurement results of the outdoor dynamic rotation test provided in the embodiment of the present invention.
[0090] Figure 6 The unmanned platform vehicle-mounted test trajectory diagram and the slightly obscured polarization diagram are provided for embodiments of the present invention.
[0091] Figure 7 The diagram shows the course measurement results of an unmanned platform vehicle-mounted test under slight disturbances, as provided in an embodiment of the present invention.
[0092] Figure 8 The unmanned platform vehicle-mounted test trajectory diagram and the severely obscured polarization diagram are provided for embodiments of the present invention.
[0093] Figure 9 The diagram shows the course measurement results of an unmanned platform vehicle-mounted test under severe interference, as provided in an embodiment of the present invention. Detailed Implementation
[0094] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.
[0095] Attitude and attitude measurement systems composed of polarization sensor (PS)-assisted inertial measurement units can provide reliable attitude and heading information for a vehicle in GNSS-denied or magnetically interfered environments. However, existing loosely coupled methods are susceptible to obstruction interference and atmospheric depolarization effects. Therefore, this paper proposes a tightly coupled attitude and attitude measurement method based on an all-sky polarization mode. First, a tightly coupled system model is established, adding solar azimuth error and zenith error to the state vector, and using the polarization angle deviation of multiple polarization channels as the measurement value. Based on this, this invention proposes a robust fusion mechanism based on sequential filters. Linear gravity vector measurement updates and nonlinear polarization measurement updates are performed sequentially to improve the robustness of the system.
[0096] according to Figure 1-9 This implementation will be described first by constructing a tightly coupled state equation based on solar vector estimation. The AOP estimate determined by INS differs somewhat from the actual measured value, which will lead to errors in the solar elevation angle calculation. Therefore, to improve the accuracy of AOP estimation, the solar zenith deviation under n frames is... and azimuth deviation Choosing it as a state vector, it can be represented as:
[0097]
[0098] Where φ=[φ E ,φ N ,φ U ] TIndicates the platform misalignment angle of the IMU, G b =[G bx G by G bz ] T This represents the scaling factor error vector of the gyroscope. The system's state equation can be expressed as:
[0099]
[0100] w represents Gaussian white noise, defined as follows:
[0101]
[0102] Each term in w represents Gaussian noise for the corresponding state variable. To simplify the state equations, the error differential equations for the solar azimuth and zenith angles can be expressed as follows:
[0103]
[0104] in Let A represent the projection of the rotational angular velocity of the n-frame relative to the i-frame onto the n-frame.
[0105]
[0106] Then, a multi-channel polarization visual measurement equation is constructed. The attitude angle of the polarization sensor affects the sky area observed by the sensor, thus affecting the AOP measurement value. Therefore, an AOP compensation model under tilted conditions is established, such as... Figure 1 As shown in b):
[0107] P represents the observer's position, and S represents the direction of the Sun on the celestial sphere, expressed as the zenith angle γ. s and azimuth α s express. This represents the observation direction corresponding to a single pixel in a polarization image. In the b-system, γ b and α b yes The zenith angle and azimuth angle, therefore It can be represented as (6). The transformation relationship from 2D pixels to 3D observation vectors can be obtained based on the intrinsic parameters and distortion parameters calibrated by the fisheye camera.
[0108]
[0109] γ b and α b This represents the zenith angle and azimuth angle of the observed vector in the b-system. The process of transforming to n can be represented as:
[0110]
[0111] in
[0112]
[0113] ro, pi, and ya represent the roll, pitch, and yaw angles of the polarization vision sensor, respectively. Therefore, the zenith and azimuth angles of the observation vector in the n-frame can be calculated.
[0114]
[0115] Traditional loosely coupled methods use heading or solar vectors as observations to achieve optimal attitude and heading estimates, which are easily affected by occlusion or other nonlinear factors. However, the proposed tightly coupled method can directly utilize the raw AOP values output by the polarization vision sensor as the observation vector, such as... Figure 2 As shown:
[0116] The red circle indicates the tightly coupled observation area, and the white pixels correspond to the measurement channels of the corresponding polarization unit. The AOP value is estimated in real time based on the current attitude and heading of the INS, thereby improving the robustness of the integrated navigation system. M uniformly distributed polarization channels are selected from the fixed field of view (FOV) of the biomimetic polarization vision sensor. The m-th polarization channel (u... m ,v m The observation vector in the corresponding carrier system is set to This observation vector can be determined based on the direction cosine matrix under the current state. However, due to the misalignment angle of the carrier, there is a certain deviation between the direction cosine matrix and the true value. Therefore, we define... in
[0117]
[0118] Therefore, Substituting into formula (6), we can obtain the observation vector in the n-system corresponding to the m-th pixel after misalignment correction. Then, according to the Rayleigh scattering model, the direction of the E vector should be perpendicular to the direction of the observed vector. and solar vector The resulting scattering surface. In addition, the solar vector is corrected by incorporating the solar azimuth and zenith angle errors in the state vector to obtain... and Where α s and γ s The solar azimuth and zenith angles are calculated based on the local geographical location and time. Therefore, the estimated value of AOP can be obtained based on the current INS solution attitude.
[0119]
[0120] in, and This represents the azimuth and zenith angle corrections for the observed vector in the n-system. Meanwhile, the INS-estimated DOP value... The scattering angle between the observed vector and the solar vector can be used as a basis. Find:
[0121]
[0122] DOP max This can be obtained from the maximum value measured by the polarization sensor. This is then combined with the measurements from the AOP's polarization vision sensor. Constructing observation equations
[0123]
[0124] in, This represents the measurement noise for a single polarization channel. Compared to loosely coupled methods, tightly coupled methods do not require least-squares fitting of the solar vector or complex feature extraction of the solar meridian. Instead, the AOP value output by the polarization sensor is directly used as the measurement vector, thus reducing the dependence on the minimum number of polarization channels and the integrity of the polarization mode. Furthermore, the measurement noise of tightly coupled methods directly reflects the observation noise of the original polarization vision sensor. Complex nonlinear transformations of the noise probability density function can be avoided.
[0125] Secondly, the accelerometer gravity vector measurement equation was constructed. In addition to multi-channel polarization visual measurements, this project also used the gravity vector measured by the accelerometer as a measurement to improve the accuracy of attitude measurement. The accelerometer gravity vector in the b-frame can be expressed as g. b =[a x ,a y ,a z ] T The gravitational vector in the n-frame at the current position can be represented as g. n =[0,0,g] T , where g represents the magnitude of gravitational acceleration. Theoretically, g b and g n The relationship is Due to platform misalignment Due to the existence of this, the estimated value of the gravity vector calculated using the rotation matrix differs from the measured value to some extent. The residual of the gravity vector can be expressed as...
[0126]
[0127] Therefore, the measurement equations for a tightly coupled system can be expressed as follows:
[0128] Z g =δg=Hg X+V g (15)
[0129] Among them, V a H represents the measurement noise of the gravity vector. a It can be represented as
[0130]
[0131] Finally, multi-source information fusion based on sequential capacitive Kalman filtering is performed. In practical applications of PS / MIMU combined systems, polarization vision sensors are easily affected by complex environments and motion states such as buildings, tree canopies, and carrier tilt, causing serious interference to the original AOP and DOP measurements. To address this challenge, a robust fusion mechanism based on sequential filtering is proposed, which selects a reliable polarization measurement channel by combining AOP and DOP measurements. Furthermore, the gravity vector measurement equation is incorporated to correct the system attitude, avoiding the influence of attitude errors on heading measurements.
[0132] According to formula (13), the reliability of the measurement channel is directly related to the AOP measurement value of the polarization sensor and the AOP estimate value of the INS. In addition, the DOP measurement value can also determine whether the polarization channel is blocked by obstacles or interfered with by clouds. Previous studies have typically used complex feature extraction methods to improve the robustness of heading measurements. However, these methods still have significant limitations in selecting the effective measurement area in the sky. Therefore, this paper proposes a polarization visual measurement channel selection method that combines AOP and DOP reliability. Setting... and This allows us to obtain the discrimination criteria for the measurement channel.
[0133]
[0134] ε a and ε d These are the polarization angle threshold and polarization degree threshold for measuring the channel. In this paper, ε is selected respectively. a For 5° and ε a Each is set to 0.2. After screening, if there are M2 polarization vision measurement channels that meet the conditions, a new polarization vision sensor measurement equation can be obtained:
[0135]
[0136] Equations (18) and (15) are the measurement equations for the robust fusion mechanism. However, the gravity measurement model Z... g It is linear, while the polarization measurement model Z polIt is nonlinear. If a nonlinear filtering algorithm is applied to both, the efficiency of the algorithm will be greatly reduced. Therefore, this study adopts a sequential filtering method. Gravity measurement is updated after the time update of gyroscope data. When the polarization vision sensor inputs, the AOP estimate of the polarization channel is updated based on the current covariance matrix and the one-step prediction state. Due to the nonlinearity of the multi-channel polarization measurement equation, capacitive Kalman filtering is used to update the polarization measurement, and the covariance matrix and gain matrix are calculated through the capacitive points. The robust fusion algorithm based on sequential capacitive Kalman filtering proposed in this patent is as follows: Figure 3 As shown.
[0137] The following describes an outdoor dynamic repositioning test without obstruction interference, using a polarized light / MEMS tightly coupled attitude measurement transposition based on sequential capacitive Kalman filtering. The test equipment includes... Figure 4 As shown, the polarization light sensor, MEMS inertial measurement unit, and fiber optic inertial navigation system are all fixed on an aluminum alloy bracket to avoid relative displacement between the sensors and the measurement reference during the experiment. Lithium-ion battery modules power each sensor, while a large lithium-ion battery module powers the computer. The aluminum alloy bracket is then fixed to an electric rotating platform. After power-on, the system first undergoes preheating and initial alignment, then the turntable is controlled by a remote controller to rotate periodically. During this process, polarization light images, MEMS gyroscope and accelerometer output data, and the attitude information of the fiber optic inertial navigation system are acquired. Finally, the computer (CPU model: Laptop i7-11800) is used for attitude calculation. Figure 5 The heading measurements are shown for MIMU, polarization sensor, loosely coupled (LC) method, and tightly coupled (TC) method. For the image-based polarization sensor, the least squares method is used to fit the solar azimuth angle, and then the polarization heading is calculated. The loosely coupled method directly uses the heading output of the polarization sensor as the measurement. The TC method uses the AOP value obtained from 255 polarization units uniformly distributed in the field of view as the measurement. The AOP threshold ε is also shown. a Set to 5°, DOP threshold ε d Set to 0.2.
[0138] Due to accumulated errors in the inertial solution, the heading result of the MIMU drifts over time, making it difficult to meet navigation requirements. While the direction measured by the polarization sensor has no accumulated error, it is severely affected by stray light from atmospheric multiple scattering and ground reflection, resulting in significant random errors. In contrast, LC and TC methods can better combine the advantages of biomimetic sensors and inertial units to achieve higher heading accuracy. The heading errors of different methods in rotation testing are shown in the figure, and the root mean square error (RMSE) of heading is shown in the table. The TC method has a 32.5% lower RMSE than the LC method. Although the random error of polarization heading and the accumulated error of the MIMU are suppressed to some extent under the loose coupling process of the MIMU and polarization sensor, the LC method lacks a robust polarization channel filtering mechanism. Some polarization channels that experience depolarization effects due to atmospheric multiple scattering and ground reflection also participate in the heading calculation, reducing heading accuracy. The tightly coupled method can estimate the polarization information of the polarization vision sensor in real time based on the current attitude. Combined with a robust fusion mechanism, it eliminates measurement channels with large differences between AOP and DOP, improving heading accuracy.
[0139] The following is a vehicle-mounted test of the unmanned platform under slight disturbance. During the test, the bionic polarization sensor, MIMU, FINS, GNSS receiver, GNSS antenna, and power module were fixed to the unmanned platform. The polarization sensor and MIMU constituted the attitude measurement system, and the FINS and GNSS integrated navigation served as the reference benchmark. The unmanned vehicle's trajectory near the Harbin Institute of Technology Aerospace Museum is shown below. Figure 6 As shown, the green marker indicates the starting point, and the red marker indicates the ending point. Some of the collected polarization images were slightly obscured by tree canopies, streetlights, and buildings.
[0140] The heading results of the unmanned platform aircraft test are as follows Figure 7 As shown, the cumulative error of the MIMU and the random interference from the polarization sensor still exist under dynamic conditions. However, the combined approach can leverage the advantages of both types of sensors to obtain heading results closer to the true values. The heading RMSE of the tightly coupled method is 32.3% lower than that of the loosely coupled method. This is because the road segments in which the unmanned platform moves are obstructed by buildings and tree canopies, causing deviations in the heading calculated by the polarization vision sensor, which in turn leads to poor Kalman filtering results in the combined system. In contrast, the tightly coupled method estimates polarization information through the system's current attitude and heading, and uses measurement bias and DOP bias to filter the measurement channels. This makes the accuracy of the AOP channels involved in the fusion more accurate, verifying the advantages of the proposed method in practical applications.
[0141] Finally, a vehicle-mounted test of the unmanned platform under severe interference was conducted. The trajectory of the unmanned platform during the test is shown in the figure. In the latter half of the test, the collected all-sky polarization pattern was severely disrupted due to the influence of buildings, tree canopies, and buses.
[0142] Vehicle test heading results under severe obstruction interference are as follows Figure 9 As shown in the diagram, in the latter half of the experiment, the heading error of the polarization sensor was very large due to the severe destruction of the all-sky polarization information contained in the polarization image. Furthermore, the decrease in heading measurement accuracy was not a sudden process, but rather occurred gradually as the number of effective channels involved in the calculation decreased. Therefore, it was difficult to eliminate the slowly changing measurement error in the Kalman filtering process using heading error discrimination methods. When the loosely coupled method fused unreliable navigation information, the heading error gradually increased. Conversely, the tightly coupled method did not use the polarization image alone to solve for the heading, but estimated the theoretical AOP and DOP with the help of a MEMS-IMU. Moreover, the robust fusion mechanism eliminated unreliable measurement channels, greatly improving the robustness of the integrated system and the heading measurement accuracy. Compared with the loosely coupled method, the root mean square error of the heading was reduced by 70.2%, verifying the advantages of this method in practical applications.
[0143] The above description is only 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 biomimetic compact combination attitude measurement method based on sequential capacitive Kalman filtering, characterized in that, The steps of the biomimetic compact combination attitude measurement method based on sequential capacitive Kalman filtering include: Step 1: Construct the system state equations for the attitude measurement system; The steps in step 1 to construct the system state equations include: Step 1.1: Adjust the solar zenith deviation for n frames. and azimuth deviation As a state vector; Step 1.2: Construct the system state equation of the attitude measurement system based on the state vector; The state vector is: in, For the platform misalignment angle of the IMU, This is the scaling factor error vector of the gyroscope. This is due to the solar zenith deviation. This refers to the azimuth deviation; The system state equation of the attitude measurement system is: in, The system state of the attitude measurement system. The differential equations for the errors of the solar azimuth and zenith angles are given. w It is Gaussian noise. for n System relative to i The rotational angular rate of the system is n Projection under the system; Step 2: Obtain the attitude measurement system state after error adjustment based on the system state equation; Step 3: Construct attitude measurement equations to obtain attitude parameters; The attitude parameters obtained in step 3 include: Compensated observation vector, corrected solar vector, AOP estimate, and error-adjusted gravity vector estimate and measurement; Step 3, which involves obtaining attitude parameters, includes: Step 3.1: Construct an AOP compensation model and obtain the compensated observation vector based on the AOP compensation model; Step 3.1, which involves obtaining the compensated observation vector, includes: Step 3.1.1: Obtain the calibration parameters and distortion parameters of the fisheye camera; Step 3.1.2: Establish an AOP compensation model under tilted conditions; Step 3.1.3: Obtain the compensated observation vector based on the calibration parameters and distortion parameters of the fisheye camera and the AOP compensation model under tilt conditions; The compensated observation vector is: in, For the compensated observation vector in n The zenith corner of the tether, For the compensated observation vector in n The azimuth of the system, express n Three-dimensional vectors under the system, for b Three-dimensional vectors under the system, The roll angle of the polarization vision sensor. The pitch angle of the polarization vision sensor. (3) is the yaw angle of the polarization sensor, (2) is the third element, (1) is the second element, and (2) is the first element. Step 3.2: Obtain AOP and DOP estimates based on the current attitude and heading of the INS, and construct a multi-channel polarization visual measurement equation based on the AOP and DOP estimates; Step 3.3: Construct the acceleration-to-gravity vector measurement equation, and obtain the error-adjusted gravity vector estimate and measurement value based on the acceleration-to-gravity vector measurement equation; Step 4: Fuse the attitude and attitude parameters and the attitude measurement system state after AOP value error adjustment to obtain the fused attitude and attitude measurement system parameters; Step 4, which involves obtaining the fused attitude measurement system parameters, includes the following steps: Based on AOP and DOP measurements, a reliable polarization measurement channel is selected. Combined with a robust fusion mechanism based on sequential filtering and gravity vector measurement equation, the system attitude is corrected to obtain the fused attitude measurement system parameters. Step 5: Filter the polarization visual measurement channels to obtain the filtered polarization visual measurement channels; Step 5, which involves obtaining the filtered polarization visual measurement channels, includes: Step 5.1: Set the AOP threshold and DOP threshold, where the AOP threshold is... The DOP threshold is ; Step 5.2: Obtain the discrimination conditions for the measurement channel based on the AOP threshold and DOP threshold; Step 5.3: Based on the discrimination criteria of the measurement channel, the polarization channel is filtered to obtain the filtered polarization channel; The criteria for determining the measurement channel are: in, To measure the polarization angle threshold of the channel, The polarization threshold of the measurement channel; Step 6: Construct a new polarization vision sensor measurement equation based on the selected polarization vision measurement channels and the fused attitude measurement system parameters; Step 7: Obtain new attitude parameters based on the new polarization vision sensor measurement equation, and obtain attitude measurement results based on the new attitude parameters and the state of the attitude measurement system after error adjustment of the AOP value.
2. The biomimetic compact combination attitude measurement method based on sequential capacitive Kalman filtering according to claim 1, characterized in that, Step 3.2 involves constructing the multi-channel polarization visual measurement equation, including: Step 3.2.1: Obtain the deviation between the direction cosine matrix and the true value, and define the deviation as... ,in, ; Step 3.2.2: Based on the AOP compensation model under the aforementioned deviation and tilt states, obtain the first [missing information - likely a specific step or step] after misalignment angle correction. The corresponding pixel The observed vector under the system; ; Step 3.2.3: Obtain the corrected solar vector based on the solar azimuth error and zenith angle error in the state vector; Step 3.2.4: Obtain the estimated value of AOP based on the corrected solar vector and Rayleigh scattering model; Step 3.2.5: Obtain the DOP estimate based on the scattering angle between the observed vector and the solar vector; Step 3.2.6: Construct a multi-channel polarization visual measurement equation based on the DOP estimate and AOP estimate; The estimated value of AOP is: in, and express Correction values for the azimuth and zenith angle of the observed vector under the system. This is an estimate of AOP; The scattering angle between the observed vector and the solar vector is: in, The scattering angle between the observed vector and the solar vector. For the first time after misalignment correction The corresponding pixel The observed vector under the system, Polarization image S The viewing direction corresponding to each pixel; The estimated value of DOP is: in, This represents the maximum value measured by the polarization sensor. The multi-channel polarization vision measurement equation is: in, Measurement noise for a single polarization channel. These are measurements from a polarization vision sensor. This is an estimate of AOP. These are the measurement parameters for the polarization vision sensor.
3. The biomimetic compact combination attitude measurement method based on sequential capacitive Kalman filtering according to claim 1, characterized in that, Step 3.3, which involves obtaining the error-adjusted gravity vector estimate and measurement, includes the following steps: Step 3.3.1: Calculate the residual of the gravity vector; Step 3.3.2: Obtain the acceleration-gravity vector measurement equation based on the residual of the gravity vector; Step 3.3.3: Based on the acceleration-to-gravity vector measurement equation, obtain the error-adjusted estimated value and measured value of the gravity vector; The residual of the gravity vector is: ； in, for The accelerometer gravity vector under the system, For the current position The gravitational vector in the system, The platform is out of alignment. The equation for the acceleration-to-gravity vector measurement is: in, The acceleration is recorded as the measured value of the gravity vector. X represents the measurement noise of the gravity vector, and X represents the state vector.
4. The biomimetic compact combination attitude measurement method based on sequential capacitive Kalman filtering according to claim 1, characterized in that, The new polarization vision sensor measurement equation in step 5 is: in, For the new polarization vision sensor measurement model, Measurement noise for a single polarization channel. The AOP threshold, These are the measurement parameters for the polarization vision sensor.