Unmanned aerial vehicle pose estimation method based on credibility weighted strong tracking filtering

Abstract

The invention provides an unmanned aerial vehicle pose estimation method based on credibility weighted strong tracking filtering. According to the method, the posterior error covariance after actual strong tracking filtering iteration and the theoretical true posterior covariance of algorithm iteration are utilized to define a trust factor used for describing filtering deviation under noise uncertainty, the trust factor acts on self-adaptive calculation of a forgetting factor, then a time-varying fading factor is corrected through calculation of an innovation error covariance, and therefore the filtering deviation under noise uncertainty is corrected. And tracking estimation under the condition of noise uncertainty is realized. Compared with traditional strong tracking series filtering, the method can quantify the uncertainty of noise and relieve experience dependence of forgetting factors. Meanwhile, the time-varying fading factor acts on the process noise covariance, and compared with a traditional action mode, the method has better interpretability. When the residual error and the fading factor are calculated, compared with the traditional fixed forgetting factor and the trust factor, the self-adaptive adjustment of the forgetting factor can be realized, the performance error is reduced, and meanwhile, the artificial experience assignment is reduced.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of unmanned aerial vehicle pose estimation, and particularly relates to an unmanned aerial vehicle pose estimation method suitable for strong tracking series filtering combined with credibility evaluation. BACKGROUND

[0002] Based on the data collected by the ship-borne radar sensor, the pose of the unmanned aerial vehicle is estimated. When filtering iteratively processes the target signal, filter divergence, steady-state oscillation, and insufficient filter adaptability often occur, which has a great impact on the solution of the position and velocity of the unmanned aerial vehicle. Therefore, the problem of quantifying and correcting the signal solution result needs to be solved urgently.

[0003] As an advanced state estimation technology for nonlinear and non-Gaussian systems, strong tracking filtering can effectively suppress filter divergence compared with traditional Kalman series filtering, and has better estimation advantages in model mismatch and parameter mutation scenarios. At present, it is widely used in target tracking, fault diagnosis and other fields. However, in the iteration process of strong tracking filtering, the selection of some parameters in the calculation of the fading factor still depends on experience setting, and lacks adaptive ability under uncertain conditions. At the same time, when suppressing filter divergence, the threshold value is generally set to be small, and the probability of triggering the fading factor is very high, which may cause excessive adjustment of the error covariance and filter gain.

[0004] In actuarial science, credibility theory is often used to balance the credibility of individual risk data and group data, but its prototype actually comes from the Bayesian theorem of probability theory. The essence of the credibility theory in actuarial science is the specific application and extension of linear Bayesian in actuarial science. However, the Kalman series filtering is also a technology that applies linear Bayesian method in signal processing field. Therefore, there is a certain relationship between the credibility theory and the Kalman filtering algorithm.

[0005] Nowadays, with the in-depth research in various fields, the theoretical system development of them is relatively perfect. The integration of mature methods in different fields and different disciplines can promote the development of the whole scientific research. The credibility theory of actuarial science is applied to the field of unmanned aerial vehicle pose estimation as a new evaluation method to improve the accuracy of pose estimation. SUMMARY

[0006] The present application proposes a new credibility evaluation method suitable for strong tracking series filtering, which is applied to the unmanned aerial vehicle pose estimation scene in order to obtain better filtering results.

[0007] To achieve the above purpose, the first aspect of the present application provides a method for unmanned aerial vehicle pose estimation based on credibility weighted strong tracking filtering, comprising the following steps:

[0008] Step 1: establish the unmanned aerial vehicle state space equation and iterative recursion formula with mismatched noise, and modify the mechanism of the time-varying fading factor;

[0009] Step 2: calculate the actual time-varying fading factor changing the mechanism based on the orthogonality principle;

[0010] Step 3: define and calculate the trust factor based on the real value of the posterior error covariance matrix;

[0011] Step 4: adaptively adjust the forgetting factor at k+1 time based on the trust factor at k time, and act on the innovation covariance calculation at k+1 time;

[0012] Step 5: cyclically iterate steps 2 to 4, and output the relative position and relative velocity of the unmanned aerial vehicle to the radar sensor at any time after each filtering iteration.

[0013] The second aspect of the application provides an electronic device for unmanned aerial vehicle pose estimation based on credibility weighted strong tracking filtering, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned unmanned aerial vehicle pose estimation method based on credibility weighted strong tracking filtering when executing the program.

[0014] The third aspect of the application provides a computer readable storage medium, the storage medium stores a computer program, and the computer program is used for executing the above-mentioned unmanned aerial vehicle pose estimation method based on credibility weighted strong tracking filtering.

[0015] Based on the above technical solution, the application has the following beneficial effects:

[0016] The application defines a trust factor for describing the filtering deviation under noise uncertainty by using the posterior error covariance after actual strong tracking filtering iteration and the theoretical real posterior covariance of algorithm iteration, and acts on the adaptive calculation of the forgetting factor, and then modifies the time-varying fading factor through the calculation of the innovation error covariance, so as to realize the tracking estimation under noise uncertainty. Compared with the traditional strong tracking series filter, the method can quantify the uncertainty of noise and relieve the experience dependence of the forgetting factor.

[0017] The application applies the time-varying fading factor to the process noise covariance, which has better interpretability compared with the traditional action mode. Compared with the traditional fixed forgetting factor, the trust factor can realize the adaptive adjustment of the forgetting factor when calculating the residual and the fading factor, reduce the performance error, and reduce the artificial experience assignment. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1A flowchart of a UAV pose estimation method based on credibility weighted strong tracking filtering of an embodiment of the present application;

[0019] Figure 2 A flowchart of an embodiment of the present application for updating the fading factor based on the trust factor;

[0020] Figure 3 A comparison chart of the simulation filtering result position state quantity and the velocity state quantity of an embodiment of the present application;

[0021] Figure 4 A comparison chart of the simulation filtering result error details of an embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, the drawings described in the following are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0023] As shown in Figure 1 , the present application provides a UAV pose estimation method based on credibility weighted strong tracking filtering: first, based on the mismatched process noise and measurement noise, the time-varying fading factor of the actual filtering process is calculated. Second, based on the state error and innovation error, the trust factor is introduced in the filtering process, which is used to evaluate the error and the credibility of one strong tracking filtering iteration. Then, based on the time-varying fading factor of the actual filtering process, the real time-varying fading factor calculation problem is converted into a bias problem. Finally, the calculation of the process noise, the measurement noise and the trust factor under three different situations is discussed, and the trust factor is used for the correction of the residual error, and the adjustment of the time-varying fading factor in the actual process is further realized. Through the orthogonal constraint and the trust factor correction feedback evaluation system of the fading factor, the target state at any time is solved. Specifically, it includes the following steps:

[0024] Step 1, strong tracking filtering prediction of noise mismatch: the state space equation and the iterative recursive formula of the UAV with mismatched noise are established, and the mechanism of the time-varying fading factor is modified. It includes:

[0025] Step 1.1, combining the noise error transmission relationship of the UAV and the strong tracking filtering method, taking the parameter to be estimated as the state quantity of the biased filter in the strong tracking filtering, establishing the linear Gaussian discrete system of the UAV, and initializing the filtering parameters;

[0026] Step 1.2, in the strong tracking filtering recursive calculation framework, the mechanism of the time-varying fading factor is changed, and the time-varying fading factor is applied to the process noise covariance matrix, so as to improve the theoretical interpretability.

[0027] Further, the step 1.1 is specifically:

[0028] When the environment appears unknown mutation interference affecting the flight of the unmanned aerial vehicle, the parameters to be estimated are the state quantities of the biased filter in the strong tracking filter, a linear Gaussian discrete system of the unmanned aerial vehicle is established, and a state space model is:

[0029]

[0030] wherein, is a state quantity, including two state quantities of position and speed. is a state transition matrix, is a system observation quantity, is a system observation matrix, is a mismatched process noise, is a mismatched measurement noise, and a subscript represents an arbitrary sampling time. The actual mismatched noise covariance satisfies:

[0031]

[0032]

[0033] wherein, is an actual mismatched process noise covariance matrix, is an actual mismatched measurement noise covariance matrix, is a real matched process noise covariance matrix, is a mismatched and real matched process noise covariance deviation matrix, is a real matched measurement noise covariance matrix, is a mismatched and real matched measurement noise covariance deviation matrix.

[0034] Further, the step 1.2 is specifically:

[0035] According to the basic framework of the strong tracking filter, in the time update stage: according to the last time posterior error covariance, the target state and the prediction error covariance at the current time are predicted, the time-varying fading factor is dynamically generated and the action mechanism is changed, the time-varying fading factor is applied to the process noise covariance matrix, rather than to the term related to the traditional state transition matrix, and the theoretical explainability is improved.

[0036] In the measurement update part: based on the innovation and the prediction error covariance, the filter gain is calculated, and the target state and the posterior error covariance are updated. The recursive formula of the strong tracking filter with the modified time-varying fading factor action mechanism is:

[0037] The time update of the strong tracking filter considering the mismatched noise covariance matrix and modifying the mechanism of time-varying fading factor is:

[0038]

[0039]

[0040] wherein, The superscript T of the matrix denotes the transpose.

[0041] The measurement update of the strong tracking filter considering the mismatched noise covariance matrix and modifying the mechanism of time-varying fading factor is:

[0042]

[0043]

[0044]

[0045] wherein, is the state prediction value of the filter, is the prediction error covariance matrix of the filter, is the time-varying fading factor, is the filter gain, is the posterior error covariance matrix of the filter, is the state estimate, is the residual (innovation) of the filter.

[0046] Step 2, calculation of the time-varying fading factor: the actual time-varying fading factor changing the mechanism is calculated based on the orthogonality principle. At any sampling time, the prediction error covariance matrix of the filter and the gain matrix are still dynamically adjusted through the time-varying fading factor changing the mechanism, and strictly satisfy the orthogonality principle constraint; based on the constraint condition, the orthogonality of the residual sequence is taken as the optimization objective, and the time-varying fading factor is solved online, so that the actual innovation covariance matches the theoretical value, thereby realizing the adaptive calculation of the time-varying fading factor.

[0047] The actual time-varying fading factor is determined through the orthogonality constraint of the residual sequence, that is, the residual sequence satisfies the orthogonality principle:

[0048]

[0049] The gain matrix of the filter is brought into the orthogonality principle, and the common non-zero matrix of the two matrices is extracted to make the equation hold, the innovation error covariance matrix is equal to the actual innovation error covariance , and the actual time-varying fading factor at any time is calculated through the moving term and the likeand set the weakening factor , reduce the possibility of over-regulation, the equation is finally simplified as:

[0050]

[0051] where, is the posterior error covariance matrix at the previous time, that is, the prior error covariance matrix at the current time.

[0052] For ease of expression, let and is:

[0053]

[0054]

[0055] where, is the weakening factor, generally taking the value of 0.95.

[0056] As a variant of the standard Kalman filter, the strong tracking filter is designed to amplify uncertainty in one direction to deal with sudden states. The actual time-varying fading factor is calculated by online optimization to match the actual innovation covariance with the theoretical value, so the actual time-varying fading factor cannot be less than 1, and the calculation formula is:

[0057]

[0058] where, is the maximum function, is the trace of the matrix.

[0059] Step 3, trust factor calculation: compare the posterior error covariance matrix true value to define and calculate the trust factor. According to the analysis of the posterior error covariance deviation caused by the mismatch of the actual filtering process noise covariance, based on the real radar observation data and the filtering estimation result, the actual posterior error covariance and the real posterior error covariance deviation and the actual innovation error covariance and the real innovation error covariance deviation are calculated, the posterior error covariance deviation degree of measurement based on spectral norm is constructed, and the reciprocal of the deviation degree of measurement is defined as the trust factor , in order to prevent the deviation from being zero and make the fraction meaningless, a constant term is added to the denominator, and finally the noise and trust factor are estimated in three scenarios.

[0060] Because of the mismatch of the noise in the above filtering process, the posterior error covariance matrix obtained by filtering cannot accurately describe the real error, because there is an error between the actual posterior error covariance and the real value, so the real state error is obtained by subtracting the actual state quantity from the posterior prediction state quantity:

[0061]

[0062] wherein, represents the estimated state of the actual filtering, is a unit matrix.

[0063] The covariance of the real state quantity error is calculated, i.e. the real posterior error covariance matrix is:

[0064]

[0065] wherein, is the state estimation quantity deviation, is the filtering state prediction deviation, is the filtering real posterior error covariance matrix, is the filtering real prediction error covariance matrix.

[0066] The filtering actual innovation error is reflected by the actual prediction observation quantity, and the actual innovation error covariance is the mathematical expectation of the product of the actual prediction observation quantity and its corresponding transpose, and the calculation result is:

[0067]

[0068] The filtering real innovation error is reflected by the residual (innovation), and the corresponding real innovation error covariance is the mathematical expectation of the product of the real innovation error and its corresponding transpose, and the calculation result is represented by the following formula:

[0069]

[0070] wherein, is the filtering actual innovation error covariance matrix, is the filtering real innovation error covariance matrix, is the real time-varying fading factor, is the actual prior error covariance matrix at k moment, is the real prior error covariance matrix at k moment, in order to prevent the deviation of error accumulation from quantifying the mismatch noise, the actual prior error covariance matrix at k moment and the real prior error covariance matrix at k moment are considered to be equal here.

[0071] The actual posterior error covariance and the real posterior error covariance are subtracted to obtain the first deviation:

[0072]

[0073] The actual innovation error covariance and the real innovation error covariance are subtracted, and the error of the process noise covariance matrix of the two is defined as , and the second deviation is:

[0074]

[0075]

[0076] where, is the difference between the actual filtered and the true value of the process noise covariance. Through analysis, the true time-varying fading factor is close to the actual filtered time-varying fading factor , and the calculation can be further simplified:

[0077]

[0078] The difference between the actual posterior error covariance and the true posterior error covariance, the actual innovation error covariance and the true innovation error covariance is analyzed, and the following conclusions are obtained:

[0079] 1) If the difference between the actual filtered and the true value of the process noise covariance and the difference between the actual filtered and the true value of the measurement noise covariance are both positive semi-definite, then the difference between the actual filtered and the true value of the posterior error covariance matrix and the difference between the actual filtered and the true value of the innovation error covariance matrix are also positive semi-definite.

[0080] 2) If the difference between the actual filtered and the true value of the process noise covariance and the difference between the actual filtered and the true value of the measurement noise covariance are both negative semi-definite, then the difference between the actual filtered and the true value of the posterior error covariance matrix and the difference between the actual filtered and the true value of the innovation error covariance matrix are also negative semi-definite.

[0081] Based on the above conclusions, the posterior error covariance bias degree of measurement based on the spectral norm is constructed. In order to prevent the bias from being zero and make the fraction meaningless, a constant term is added to the denominator, and the trust factor is defined. The calculation formula of

[0082]

[0083] where, is the two norm.

[0084] In order to facilitate calculation and expression, the first bias is defined as , and the long matrix in the formula is defined as and , and the defined matrix is as follows:

[0085]

[0086] The difference between the actual and true values ​​of the posterior error covariance matrix and the difference between the actual and true values ​​of the innovation error covariance matrix can be simplified using the defined matrices as follows:

[0087]

[0088]

[0089] The calculation of noise and credibility (trust factor) is divided into the following three cases:

[0090] 1) When the measurement noise is known, that is... Substitute the first deviation With the second deviation Estimate the amplification factor Deviation from the first Then, further estimate the inaccurate process noise. :

[0091]

[0092]

[0093]

[0094] 2) When the process noise is known, i.e. ,at this time Inaccurate estimation of measurement noise and first deviation :

[0095]

[0096]

[0097] 3) When both process noise and measurement noise are unknown, i.e. and All are not 0

[0098] Approximate estimation is achieved by combining variational Bayesian or EM algorithms with optimization algorithms. and Then compared with the actual assumptions and By comparison, we can conclude that and Further conclusions Then calculate the first deviation. :

[0099]

[0100] The above three situations Incorporating the trust factor The reliability of the filtering can be obtained in the formula.

[0101] Step 4, adaptive forgetting factor calculation and innovation covariance update: based on the trust factor at time k, the forgetting factor at time k+1 is adaptively adjusted, and is used for the innovation covariance calculation at time k+1.

[0102] As shown in Figure 2 , the forgetting factor at time k+1 is dynamically updated based on the trust factor at time k, the adaptive updating formula of the forgetting factor is designed, on the basis of weighted summation of the initial forgetting factor, the adaptive exponential term is designed, and the adjustment parameter is designed. The adaptive weighted summation formula of the forgetting factor is:

[0103]

[0104] wherein, is the adaptive forgetting factor, is the initial forgetting factor, and the experience value is 0.95; is the trust factor at time k, and the value is ; is the adjustment parameter.

[0105] The adaptive forgetting factor is used to replace the originally experienced forgetting factor, and is used for the calculation of the historical innovation covariance related term of the innovation covariance matrix at time k+1. The calculation formula is:

[0106]

[0107] wherein, is the residual (innovation) at time k+1.

[0108] Based on the trust factor at time k, the adaptive forgetting factor at time k+1 is designed, the calculation formula of the residual at time k+1 is corrected, the time-varying fading factor at time k+1 is further corrected, and finally the subsequent iteration calculation at the next sampling time is substituted, so as to improve the judgment and tracking of the noise deviation.

[0109] The adaptive forgetting factor is used to control the update weight of the innovation covariance, and the proportion of the historical information and the current observation data is balanced. When the trust factor is lower than the designed adjustment parameter, it is indicated that the historical innovation error is larger, the adaptive forgetting factor should be reduced, the adaptive forgetting factor is closer to the initial forgetting factor, the weight of the current observation data is increased, and the current observation data is more relied on. When the trust factor is higher, it is indicated that the historical innovation error is smaller, the historical innovation has a certain reference value, the adaptive forgetting factor can be appropriately increased, and the current observation data and the historical data need to be considered.

[0110] Step 5, loop iteration and output: loop iteration steps 2~4, after each filtering iteration, the relative position and relative velocity of the UAV to the radar sensor at any time can be output.

[0111] Based on the same concept as the above method, the embodiment of the application further provides an electronic device for unmanned aerial vehicle pose estimation based on credibility weighted strong tracking filtering, comprising:

[0112] a memory for storing computer programs and data;

[0113] a processor for executing the computer program to realize the electronic method for unmanned aerial vehicle pose estimation based on credibility weighted strong tracking filtering.

[0114] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0115] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the device by running or executing the computer program and / or modules stored in the memory, and calling the data stored in the memory.

[0116] Based on the same concept as the above method, the embodiment of the application further provides a computer readable storage medium, the storage medium stores a computer program, and the computer program is used to execute the method for unmanned aerial vehicle pose estimation based on credibility weighted strong tracking filtering.

[0117] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments of each method. Wherein, any reference to memory, storage, database or other medium used in each embodiment provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM) and the like.

[0118] Figure 3 and Figure 4 To prove the effectiveness of the method, the pose estimation results of a certain time-varying linear motion unmanned aerial vehicle flight experiment are selected for verification. Compared with the traditional strong tracking filter, the method improves the performance by 7.19% and 7.01% in the simulation environment under the position and velocity observation, and the error reduction ratio in the mutation area is 4.76%. Compared with the traditional method, the method has the following advantages: the time-varying fading factor acts on the process noise covariance, which has better interpretability compared with the traditional action mode. Compared with the traditional fixed forgetting factor, the trust factor can realize the adaptive adjustment of the forgetting factor, reduce the performance error, and reduce the artificial experience assignment. 、 、 、 The method has the following advantages compared with the traditional method: the time-varying fading factor acts on the process noise covariance, which has better interpretability compared with the traditional action mode. Compared with the traditional fixed forgetting factor, the trust factor can realize the adaptive adjustment of the forgetting factor, reduce the performance error, and reduce the artificial experience assignment.

[0119] Obviously, the above-mentioned embodiments of the present application are only examples for clearly illustrating the present application, and are not a limitation on the embodiments of the present application. For ordinary skilled in the art, on the basis of the above description, other different forms of changes or variations can be made, and here it is impossible to enumerate all the embodiments. Any obvious changes or variations derived from the technical solutions of the present application are still within the protection scope of the present application.

Claims

1. A UAV pose estimation method based on credibility-weighted strong tracking filter, characterized in that, The method comprises the following steps: Step 1: establishing a UAV state space equation and an iterative recursion formula in the presence of mismatched noise, and modifying the action mechanism of a time-varying fading factor; Step 2: calculating the actual time-varying fading factor changing the action mechanism based on the orthogonality principle; Step 3: defining and calculating a trust factor by comparing a real value of a posterior error covariance matrix; Step 4: adaptively adjusting a forgetting factor at k+1 time based on the trust factor at k time, and applying the forgetting factor to the calculation of a new information covariance at k+1 time; Step 5: iteratively performing steps 2 to 4, and outputting a relative position and a relative velocity of a UAV with respect to a radar sensor at any time after each filtering iteration.

2. The method of claim 1, wherein, The step 1 is specifically: Step 1.1: combining noise error transmission relationships of the UAV and a strong tracking filtering method, taking estimated parameters as state quantities of a biased filter in the strong tracking filtering method, establishing a linear Gaussian discrete system of the UAV, and initializing filtering parameters; Step 1.2: changing the action mechanism of the time-varying fading factor under a strong tracking filtering recursion calculation framework, and applying the time-varying fading factor to a process noise covariance matrix to improve theoretical interpretability.

3. The method of claim 2, wherein, The step 1.2 is specifically: According to the basic framework of the strong tracking filtering method, in a time update stage: generating a time-varying fading factor dynamically and changing the action mechanism of the time-varying fading factor by applying the time-varying fading factor to the process noise covariance matrix based on a target state and a prediction error covariance at a current time predicted from a posterior error covariance at a previous time, so as to improve theoretical interpretability. In a measurement update part: calculating a filtering gain based on new information and the prediction error covariance, and updating the target state and the posterior error covariance.

4. The method according to claim 1, 2 or 3, characterized in that, The step 2 is specifically: Determining the actual time-varying fading factor through orthogonality constraints of residual sequences; Bringing a gain matrix of filtering into the orthogonality principle, and extracting a common non-zero matrix of the two matrices; calculating the actual time-varying fading factor at any instant of time and setting the weakening factor to reduce the likelihood of over-regulation; The strong tracking filtering method is a variant of the standard Kalman filtering method, and the actual time-varying fading factor is calculated through online optimization to make the actual new information covariance match a theoretical value.

5. The method of claim 4, wherein, The step 3 is specifically: Obtaining a real state quantity error by subtracting a posterior prediction state quantity from an actual state quantity; Obtaining a real posterior error covariance matrix by calculating a covariance of the real state quantity error; Obtaining an actual new information error covariance by calculating an expectation of a product of an actual prediction observation and a transpose corresponding to the actual prediction observation; Obtaining a real new information error covariance corresponding to a real new information error by calculating a mathematical expectation of a product of the real new information error and a transpose corresponding to the real new information error; Obtaining a first deviation by subtracting the actual posterior error covariance from the real posterior error covariance; differencing the actual innovation error covariance from the true innovation error covariance and defining an error in the process noise covariance matrix of both , resulting in a second bias; Analyzing the first deviation and a second deviation, constructing a posterior error covariance deviation degree based on a spectral norm, and calculating a trust factor.

6. The method of claim 5, wherein, The calculation manner of the trust factor is specifically: When the equivalent measurement noise is known: estimate the fading factor amplification With the first deviation, further estimate the inaccurate process noise, bring in the trust factor calculation formula, and obtain the credibility of the filter When the process noise is known: estimating an inaccurate measurement noise and the first deviation, bringing the inaccurate measurement noise and the first deviation into a trust factor calculation formula, and obtaining a filtering credibility; When both process noise and measurement noise are unknown: approximate estimation of real matching process noise covariance matrix and real matching measurement noise covariance matrix, compared with actual non-matching process noise covariance matrix and actual non-matching measurement noise covariance matrix actually assumed, it is concluded that ; the first deviation is calculated, and the trust factor calculation formula is brought in to obtain the credibility of filtering.

7. The method of claim 1, wherein, The step 4 is specifically: The forgetting factor at k+1 moment is dynamically updated based on the trust factor at k moment, an adaptive updating formula of forgetting factor is designed, an adaptive exponential term is designed based on the weighted sum of the initial forgetting factor, and an adjusting parameter is designed; The adaptive forgetting factor is used to replace the original experience value forgetting factor, and is applied to the calculation of the innovation covariance matrix at k+1 moment; Based on the trust factor at k moment, the adaptive forgetting factor at k+1 moment is designed, the calculation formula of the residual error at k+1 moment is corrected, and the time-varying fading factor at k+1 moment is further corrected.

8. An electronic device for unmanned aerial vehicle pose estimation based on credibility-weighted strong tracking filter, the device comprising: Comprise: A memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the program to realize the unmanned aerial vehicle pose estimation method based on the trust weighted strong tracking filtering in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, and the computer program is used to execute the unmanned aerial vehicle pose estimation method based on the trust weighted strong tracking filtering in any one of claims 1-7.

Images