Bayesian estimation-based vehicle sensor bias calibration method and system, electronic equipment and storage medium

Abstract

The invention discloses a Bayesian estimation-based vehicle sensor offset calibration method and system, electronic equipment and a storage medium. The method comprises: identifying a vehicle static or moving state based on a vehicle speed; in a static state, estimating the yaw rate and the static bias of the longitudinal / transverse acceleration sensor through Kalman filtering by utilizing the priori knowledge that the vehicle acceleration truth value is zero; during movement, static bias is used as an initial value, and dynamic bias is estimated online through extended Kalman filtering based on a nonlinear dynamic model containing transverse velocity, yaw rate change rate and acceleration bias; and finally, according to the vehicle state, selecting a corresponding offset value to carry out real-time compensation on a sensor signal. The system comprises the functional modules for realizing the method. The electronic equipment and the storage medium are used for storing and executing corresponding programs. According to the invention, full-working-condition self-adaptive online calibration of the sensor bias is realized, the signal precision and the system reliability are improved fundamentally, and the calibration cost is reduced.

Description

Technical Field

[0001] This invention belongs to the field of vehicle electronic control and sensor signal processing technology, specifically relating to a vehicle sensor bias calibration method, system, electronic device and storage medium based on Bayesian estimation. Background Technology

[0002] In automotive active safety systems (such as Automatic Emergency Braking (AEB), Lane Departure Warning (LDW), and Forward Collision Warning (FCW), yaw rate and acceleration sensors are key components for acquiring vehicle dynamics. Currently, most vehicles use low-cost tuning fork gyroscopes to measure yaw rate; however, these sensors inevitably suffer from static bias due to installation errors and long-term vehicle vibrations, and this bias drifts over time. Existing acceleration sensors are mostly inertial measurement units, which are significantly affected by the longitudinal and lateral slopes of the road. They also exhibit static bias, and when the vehicle is traveling on an incline, the gravitational component introduces a dynamic bias.

[0003] In existing technologies, there are two main approaches to handling sensor errors: one is to calculate the reference yaw rate using information such as wheel speed and steering angle, and then identify sensor output anomalies through cross-validation. However, this method cannot fundamentally estimate and compensate for the sensor's own bias. The other approach is to construct a Kalman filter based on a vehicle dynamics model and a complex observer (such as a Romberg observer) to estimate the bias of the accelerometer. However, this type of method is complex to model, difficult to tune parameters, and highly dependent on model accuracy. Furthermore, because the prediction equation contains a yaw rate term, the prediction matrix is ​​time-varying, making it difficult to guarantee the stability of the filter. In turn, the bias of the yaw rate sensor itself will affect the accuracy of the estimation, leading to errors in the prediction equation.

[0004] Therefore, there is an urgent need for a method that can directly, efficiently, and adaptively calibrate sensor bias online to fundamentally improve the accuracy and robustness of the sensor's original signal. Summary of the Invention

[0005] This invention aims to overcome the shortcomings of the prior art and provide a vehicle sensor bias calibration method, system, electronic device and storage medium based on Bayesian estimation. It can adaptively estimate and compensate the static and dynamic biases of the yaw rate sensor and the longitudinal and lateral acceleration sensors in both stationary and moving vehicle states, thereby improving the accuracy of sensor signals, ensuring the reliability of active safety functions of automobiles that rely on such signals, and reducing the development cost of vehicle model matching.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a vehicle sensor bias calibration method based on Bayesian estimation, comprising the following steps: Vehicle status recognition steps: Determine whether the vehicle is stationary or in motion based on its speed; Static bias estimation steps: When the vehicle is stationary, enable the static bias estimator, construct a state space model based on the prior knowledge that the true value of acceleration is zero when the vehicle is stationary, and use Kalman filtering to estimate the static biases of the yaw rate sensor, longitudinal acceleration sensor and lateral acceleration sensor respectively. Dynamic bias estimation initialization steps: When the vehicle enters a moving state, the dynamic bias estimator is enabled, and the static bias estimates of yaw rate, longitudinal acceleration and lateral acceleration output by the static bias estimator are used as the initial values ​​of yaw rate, longitudinal acceleration bias and lateral acceleration bias in the dynamic bias estimator, respectively. Dynamic bias estimation steps: Based on the extended Kalman filter (EKF) framework, a nonlinear state-space model is established that includes lateral velocity, longitudinal velocity, yaw rate, yaw rate change rate, lateral acceleration bias, and longitudinal acceleration bias. Combined with sensor measurements and auxiliary estimators, the dynamic bias of the longitudinal and lateral acceleration sensors under motion conditions is estimated online iteratively. Offset output and compensation steps: Based on the current vehicle status, select the corresponding static or dynamic offset estimate, and use the selected offset estimate to compensate the original measurement signal of the sensor corresponding to the offset estimate in real time.

[0007] Secondly, the present invention provides a vehicle sensor bias calibration system based on Bayesian estimation for implementing the above method, the system comprising: The status recognition module is used to determine the vehicle status based on the vehicle speed signal; The static bias estimation module is used to perform Kalman filtering when the vehicle is stationary to estimate the static bias of the sensors. The dynamic bias estimation module is used to perform extended Kalman filtering while the vehicle is moving to estimate the dynamic bias of the sensor. Its initialization depends on the output of the static bias estimation module. The auxiliary estimation module provides estimates of lateral velocity and yaw rate change. The compensation output module is used to compensate the sensor signal based on the current effective bias value and then output it.

[0008] Thirdly, the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0009] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the above-described method.

[0010] Beneficial effects Compared with the prior art, the present invention has at least the following beneficial effects: This invention directly estimates the bias error source from the sensor signal and compensates for it, rather than indirectly identifying anomalies or avoiding errors, thus improving signal accuracy from the root.

[0011] This invention designs static and dynamic bias estimators for two typical operating conditions: vehicle stationary and vehicle in motion. It also designs intelligent switching and initialization strategies to achieve adaptive online calibration under all operating conditions.

[0012] This invention abandons complex designs such as Romberg observers and adopts a simplified transverse velocity estimation and standard Kalman filter / extended Kalman filter (EKF) framework, which reduces the complexity of modeling and parameter tuning and improves practicality.

[0013] This invention enables low-cost sensors such as tuning fork gyroscopes to meet the accuracy requirements of advanced active safety systems. At the same time, the "adaptive calibration" feature eliminates the need for a dedicated static calibration process when a vehicle rolls off the production line, saving development costs and time. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the overall process of the vehicle sensor bias calibration method based on Bayesian estimation in an embodiment of the present invention. Figure 2 This is a schematic diagram of the convergence curve for static yaw rate bias estimation in an embodiment of the present invention; Figure 3 This is a schematic diagram of the dynamic longitudinal acceleration bias estimation curve in an embodiment of the present invention; Figure 4 This is a schematic diagram of the dynamic lateral acceleration bias estimation curve in an embodiment of the present invention. Detailed Implementation

[0016] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be noted that the Kalman filter and its nonlinear generalization, the Extended Kalman Filter (EKF), used in this invention are specific implementations of Bayesian estimation theory under the assumption of linearity or linearizable Gaussianism. Therefore, any estimation based on Kalman filter or EKF mentioned below falls within the scope of the Bayesian estimation framework. The following embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0017] Example 1: like Figure 1 As shown, this embodiment of the invention provides a vehicle sensor bias calibration method based on Bayesian estimation, including the following steps: S1: Vehicle state recognition and estimator enable judgment.

[0018] Vehicle speed is calculated by acquiring wheel speed signals through the vehicle's CAN bus; specifically, the longitudinal speed of the vehicle can be calculated using a longitudinal speed filter. When the vehicle speed remains below a first threshold for a first time duration, the vehicle is determined to be stationary. The first threshold is 0.3~0.5 m / s, and its setting is generally the minimum resolution of the axle speed sensor. The first time duration threshold is set to 0.5 seconds. For example, when... If the speed is less than 0.5 m / s for more than 0.5 seconds, the vehicle is considered to be stationary; at this time, the static offset estimator for acceleration and the static offset estimator for yaw rate are enabled and reset.

[0019] When the vehicle speed remains above the second threshold for a second time duration, the vehicle is determined to be in motion. In this embodiment of the invention, the second threshold is set to be the same as the first threshold, and the second time duration threshold is set to be the same as the first time duration threshold; for example, when If the speed is >0.5 m / s for more than 0.5 seconds, the vehicle is considered to have started moving. At this point, the dynamic bias estimator (EKF) is enabled and reset, and the output estimate of the static bias estimator is fixed. Simultaneously, the static bias estimates of yaw rate, longitudinal acceleration, and lateral acceleration output from the static bias estimator are used as the initial values ​​for the yaw rate, longitudinal acceleration bias, and lateral acceleration bias states in the dynamic bias estimator, respectively.

[0020] S2: Static bias estimation of the accelerometer.

[0021] When the vehicle is stationary, the theoretical true values ​​of longitudinal and lateral accelerations are 0 m / s². 2 The state vector for constructing the static bias estimation process is as follows: The measurement input is ;in , For acceleration measurement related items, , The bias term to be estimated is, i.e. For longitudinal acceleration, For lateral acceleration, For longitudinal acceleration bias, This is a lateral acceleration bias.

[0022] Then, construct the state-space model, including the state transition matrix and the measurement matrix:

[0023] in, Ψ k|k-1 Let k be the state transition matrix from time k-1 to time k. H k Let k be the measurement output matrix. X k Let k be the state vector at time k. Z k For the measurement at time k, P k Let be the estimated covariance matrix at time k; P alagt , P algtbisa , P alat , P alatbisa These are the estimated variances of longitudinal acceleration, longitudinal acceleration bias, lateral acceleration, and lateral acceleration bias, respectively.

[0024] Initialization: Since the vehicle is stationary, the true value of the acceleration is 0 m / s² during the initialization of the static bias estimator. 2 Therefore, in the initial posterior state, when initializing the posterior estimation covariance matrix, the variance of the acceleration term can be set to a relatively small value, i.e. P alagt and P alat The initial covariance is set to a relatively small first value range, for example, 1e-8 to 1e-6; this indicates that the initial error of the acceleration term is very small, close to the true value, and does not require excessive correction in subsequent Kalman filtering. The initial covariance of the bias is set to a larger value, i.e. P algtbisa , P alatbisa The second numerical range is set to a large range, for example, 1e-1 to 9e-1, to characterize its large initial uncertainty. The initial posterior estimate of acceleration is set to 0, which allows the estimator to converge quickly and maintain the stability of the estimated output. Once the filter stabilizes, the bias estimate of the output hardly changes.

[0025] The system noise covariance matrix Q and the measurement noise covariance R are preset based on the sensor noise characteristics.

[0026] Subsequently, standard Kalman filtering iterations are performed, and static bias estimation gain is calculated iteratively. The specific iterative process includes: One-step prior estimation:

[0027] One-step prior estimation of covariance:

[0028] One-step Kalman gain calculation:

[0029] One-step posterior state estimation:

[0030] One-step posterior covariance calculation:

[0031] in, X - k|k-1 State variables X k Prior estimates, P - k|k-1 To estimate the covariance matrix a priori, Z k Let k be the sensor measurement value at time k. Z - k|k-1 For measurement and prediction, P k Let K be the posterior estimated covariance matrix at time k. For posterior estimation; Q k-1 Let R be the system noise covariance matrix at time k-1, and let R be the accelerometer sensor measurement error covariance matrix. K k Let K be the Kalman filter gain at time k; I It is the identity matrix. H k For the measurement matrix, K k For Kalman gain.

[0032] Through the above iterative process, the static bias estimator can quickly converge to obtain a stable static bias estimate of the accelerometer. The yaw rate static bias estimation adopts an acceleration analogy principle, ultimately yielding a yaw rate static bias estimate, which will not be elaborated upon here.

[0033] S3: Initialization of dynamic bias estimation.

[0034] Once the vehicle enters motion, considering that the tuning fork gyroscope design of the vehicle's yaw rate sensor can inherently offset the additional acceleration caused by the longitudinal and lateral slopes of the road under dynamic conditions, the yaw rate only requires compensation through static bias estimation. However, for longitudinal and lateral acceleration, bias estimation output is necessary under dynamic conditions. Specifically, During the initialization phase, in the state vector of the dynamic EKF estimator, the initial values ​​of the lateral acceleration bias and the longitudinal acceleration bias are taken from the static estimates of the lateral acceleration bias and the longitudinal acceleration bias output in step S2, respectively. Their corresponding initial estimation covariance values ​​are also inherited from the corresponding terms in the posterior covariance matrix finally output by the static estimator. That is, the initial covariance of the bias states in the dynamic bias estimator is set to be equal to the variance of the bias terms in the posterior estimation covariance matrix finally output by the static bias estimator that correspond physically to each bias state.

[0035] S4: Auxiliary quantity estimation.

[0036] Lateral velocity estimation: based on dynamic formula To make an estimate, among which The yaw rate, This is the distance from the center of the rear axle to the center of gravity of the vehicle. Finally, the estimated... It will be used as an observation of EKF.

[0037] Estimation of the rate of change of yaw rate: First, the state equation and measurement equation for the yaw rate related terms are constructed as follows:

[0038]

[0039] in, The system white noise at time k-1 The noise is measured for the yaw rate at time k. T The sampling period is It is a measurement and estimation of the yaw rate; and These are the rates of change of the yaw rate at time k and time k-1, respectively. and These are the yaw rates at time k and time k-1, respectively. and These are the timestamps for times k and k-1, respectively.

[0040] Then, based on the Kalman filter iterative formula and the state equation and measurement equation related to the yaw rate mentioned above, the rate of change of the yaw rate can be recursively estimated, and the posterior yaw rate change estimate and the corresponding posterior covariance matrix can be estimated. The estimated yaw rate change will be used as a state variable of the EKF.

[0041] S5: Dynamic bias estimation based on EKF.

[0042] This step is based on the Extended Kalman Filter (EKF) framework, which establishes a nonlinear state-space model that includes lateral velocity, longitudinal velocity, yaw rate, yaw rate change rate, lateral acceleration bias, and longitudinal acceleration bias. Combined with sensor measurements and auxiliary estimators, the dynamic bias of the longitudinal and lateral acceleration sensors under motion conditions is estimated online iteratively.

[0043] Specifically, firstly, the state vector of the dynamic EKF is defined as: .

[0044] Then, a nonlinear state-space model is established, incorporating lateral velocity, longitudinal velocity, yaw rate, rate of change of yaw rate, lateral acceleration bias, and longitudinal acceleration bias. This model includes nonlinear state transition equations and measurement equations. The state equations are: (1) In formula (1), For lateral speed, For longitudinal velocity, For yaw rate, For longitudinal acceleration measurements, For lateral acceleration measurement values, For lateral acceleration bias, For longitudinal acceleration bias, The rate of change of yaw rate This represents the white noise of the system corresponding to the derivative component of the transverse velocity. This represents the system white noise corresponding to the derivative component of the lateral acceleration bias. This represents the white noise of the system corresponding to the derivative component of the longitudinal velocity. This represents the system white noise corresponding to the longitudinal acceleration bias derivative component. This represents the white noise of the system corresponding to the derivative component of the rate of change of yaw rate; Represents the derivative of the transverse velocity. This represents the derivative of the lateral acceleration bias. Represents the derivative of the longitudinal velocity. This represents the longitudinal acceleration bias derivative; The derivative of the rate of change of yaw rate. Represents the state vector. This represents the input vector.

[0045] The measurement equation is:

[0046] Where the superscript or subscript k is the time series index. v zy k The lateral velocity measurement at time k. v yk Let k be the lateral velocity state variable at time k. v zx k For the longitudinal velocity measurement at time k, v x k Let k be the longitudinal velocity state quantity at time k. γ z k The yaw rate is measured at time k. Let k be the lateral acceleration bias at time k. The longitudinal acceleration bias at time k; Let k be the measurement white noise vector. White noise for the lateral velocity measurement component. White noise for the longitudinal velocity measurement component, White noise for the yaw rate measurement component.

[0047] Next, the aforementioned nonlinear state transition equations are discretized based on the relationship between vehicle dynamics and kinematics. Specifically, discretization is performed using first-order difference. Let the state function and input function be: (2) In formula (2), v y k and v y k-1 , respectively, are the lateral velocities at times k and k-1; Let be the lateral acceleration at time k-1; Let be the longitudinal acceleration at time k-1; and These represent the lateral acceleration biases at times k and k-1, respectively. and These represent the longitudinal acceleration biases at times k and k-1, respectively. and These represent the rates of change of the yaw rate at time k and time k-1, respectively. and These are the yaw rates at time k and time k-1, respectively. and These are the timestamps for times k and k-1, respectively. , , , , These are the components at time k-1; T The sampling period.

[0048] The noise driving matrix G(k) is set as follows: (3) In formula (3), where w ( k () represents the measurement white noise vector at time k; White noise representing the lateral velocity component. White noise representing the lateral acceleration bias component. White noise representing the longitudinal velocity component, White noise representing the longitudinal acceleration bias component. White noise representing the rate of change of yaw rate.

[0049] Therefore, formulas (1), (2), and (3) can be written in the following nonlinear discrete form, where B(k) is the input matrix and u(k) is the input vector: (4) Finally, the state transition function is linearized. Using the aforementioned posterior state estimate as the nominal point, a first-order multivariate Taylor expansion is performed, yielding the corresponding Jacobian matrix: (5) In formula (5), and These are the posterior estimates of the yaw rate at times k and k-1, respectively. and These are the posterior estimates of the lateral acceleration bias at times k and k-1, respectively. and These are the posterior estimates of the longitudinal acceleration bias of the centroid at times k and k-1, respectively. and These are the posterior estimates of the transverse velocity of the centroid at times k and k-1, respectively. and The state vectors at time k and k-1 are respectively. and The posterior estimate; T is the sampling time interval.

[0050] Substituting equation (5) into equation (4) yields the linearized state transition equation, as shown in equation (6): (6) At this point, the nonlinear state equation is transformed into a linear discrete form of the state variables. Subsequently, the Extended Kalman Filter (EKF) iteration is performed: Prediction: Based on the posterior estimate of the previous time step Given the state equation, calculate the prior state estimate and prior covariance. The calculation process combines formulas (5) and (6) to obtain the one-step predictive prior estimate:

[0051] In each EKF iteration, the Jacobian matrix is ​​calculated based on the posterior state estimate of the previous time step.

[0052] And prior estimates of covariance:

[0053] Update: Calculate the Kalman filter gain:

[0054] Then combine the measured values The update yields the posterior state estimate:

[0055] And the posterior estimate of covariance:

[0056] In each iteration, the system noise covariance matrix Q is calculated by using the lateral and longitudinal acceleration biases predicted by the EKF prior to calculate the prior lateral and longitudinal velocities. Then, the calculated prior lateral and longitudinal velocities are subtracted from the corresponding acceleration sensor measurements and multiplied by the EKF gain. Finally, the product is used to correct the prior bias estimation and obtain the posterior bias estimation. Therefore, in formula (3) ω 1 k , ω 3 k , ω 5 k The corresponding system noise covariance needs to be set to a relatively large value to increase the weight of measurement updates in state correction and ensure the filter's ability to track dynamic biases. For example, its value range is [1e-1~9e-1].

[0057] Repeat the EKF iterative steps described above, and linearize the state transition equation based on the new posterior estimate. According to the EKF iterative algorithm: combine the measured values ​​of longitudinal acceleration, lateral acceleration, and yaw rate of change with the prior estimate of the vehicle's lateral dynamics to repeatedly correct the bias estimate. Finally, at the end of each iteration, output the bias estimates of longitudinal acceleration and lateral acceleration, and finally output the optimal bias estimates of lateral acceleration and longitudinal acceleration.

[0058] S6: Bias output and compensation.

[0059] This step selects the corresponding static or dynamic bias estimate based on the current vehicle status, and uses the selected bias estimate to compensate the original measurement signal of the sensor corresponding to the bias estimate in real time.

[0060] Specifically, the controller outputs a valid bias value based on the currently enabled estimator. If the vehicle is stationary, it outputs the bias estimate from the static bias estimator; if the vehicle is moving, it initializes the dynamic EKF filter using the bias estimate from the previous static bias estimator and outputs the dynamic bias estimate. Finally, the bias estimates from the static or dynamic bias estimators are sent to upper-layer applications (such as AEB and ESC) for real-time compensation of the original measurement signals from the corresponding sensors. For example, yaw rate, longitudinal acceleration, and lateral acceleration signals are compensated based on the output yaw rate, longitudinal acceleration, and lateral acceleration bias estimates, respectively (e.g., subtracting the bias estimate from the original signal) to obtain corrected signal values, thus improving signal accuracy.

[0061] Figure 2 The figure shows the convergence process of the static bias estimation of the yaw rate when the vehicle is stationary multiple times. As can be seen from the figure, it can quickly converge to a stable value each time. Figure 3 and Figure 4 The simulation test from a stationary state to a moving state demonstrates the estimated curves of the longitudinal and lateral acceleration dynamic biases, which can effectively track the bias changes introduced by factors such as slope.

[0062] Example 2 Based on the same technical principles as Embodiment 1, this embodiment of the invention provides a vehicle sensor bias calibration system based on Bayesian estimation. This system can be integrated into the vehicle's electronic control unit (ECU) as a software module, or implemented using a combination of hardware and software. The system includes: Status recognition module: configured to receive vehicle speed signals in real time, determine the vehicle status based on the vehicle speed signals, and output vehicle status flags (stationary / moving).

[0063] Static bias estimation module: Communicates with the state recognition module. It is activated when a stationary flag is received, and internally implements the Kalman filter algorithm as described in step S2 of Embodiment 1 to estimate the static bias of the yaw rate, longitudinal and lateral acceleration sensors.

[0064] Auxiliary estimation module: continuously running, internally implementing the lateral velocity estimator and yaw rate change rate estimator as described in step S4 of Example 1, providing input to the dynamic bias estimation module, that is, providing lateral velocity estimates and yaw rate change rate estimates.

[0065] The dynamic bias estimation module communicates with the state recognition module, the static bias estimation module, and the auxiliary estimation module. It is activated upon receiving a motion flag and internally implements the extended Kalman filter algorithm as described in step S5 of Embodiment 1 to estimate the dynamic bias of the sensor. The initialization data (initial bias state value and variance) for this module comes from the static bias estimation module.

[0066] Compensation Output Module: Communicates with the static bias estimation module and the dynamic bias estimation module. Based on the currently valid vehicle state flag, it selects the bias estimate value from the static or dynamic bias estimation module, performs compensation on the corresponding original sensor input signal as described in step S6 of Example 1, and outputs the corrected signal.

[0067] The modules communicate with each other through designated data interfaces to exchange status, parameters and estimated values, and work together to complete the entire calibration process.

[0068] Example 3: This invention provides an electronic device for performing the vehicle sensor bias calibration method based on Bayesian estimation. This electronic device may be a vehicle domain controller, a dedicated sensor processing unit, or an onboard computer with corresponding computing capabilities. The electronic device includes at least: Memory: Used to store executable computer programs and data generated during operation (such as state estimates, covariance matrices, threshold parameters, etc.). The memory can be a combination of non-volatile memory (such as Flash) and volatile memory (such as RAM).

[0069] Processor: Coupled to the memory, configured to read and execute a computer program stored in the memory. When the computer program is executed, the processor is controlled to perform all the steps of the vehicle sensor bias calibration method as described in Embodiment 1, or to control the functional units within the electronic device to construct and operate the vehicle sensor bias calibration system based on Bayesian estimation as described in Embodiment 2.

[0070] Furthermore, embodiments of the present invention also provide a computer-readable storage medium, such as a USB flash drive, a solid-state drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk. This storage medium stores a computer program. When the computer program in the storage medium is loaded into the memory of the aforementioned electronic device and executed by its processor, the electronic device can implement the vehicle sensor bias calibration method based on Bayesian estimation as described in Embodiment 1, or implement the function of the vehicle sensor bias calibration system based on Bayesian estimation as described in Embodiment 2.

[0071] In summary, this invention provides a complete online calibration solution for vehicle sensors. Its core innovation lies in the creative application of a Bayesian estimation-based Kalman filter / EKF framework to the direct estimation of sensor biases, and the design of an intelligent architecture featuring "static-dynamic separation and collaborative initialization." The static bias estimator utilizes strong prior knowledge of a stationary vehicle to achieve rapid and stable bias acquisition; the dynamic bias estimator, starting from the static results and combining vehicle dynamics, achieves accurate tracking of time-varying biases during driving. The two are seamlessly connected through intelligent switching logic based on vehicle speed, forming an adaptive, all-condition closed-loop calibration system. This method not only fundamentally improves the signal quality of yaw rate and acceleration sensors, enhancing the reliability and safety of advanced driver assistance systems, but its "adaptive" characteristic also effectively reduces the calibration cost and development cycle of mass-produced vehicles, demonstrating significant industrial application value.

[0072] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A vehicle sensor bias calibration method based on Bayesian estimation, characterized in that, Includes the following steps: Vehicle status recognition steps: Determine whether the vehicle is stationary or in motion based on its speed; Static bias estimation steps: When the vehicle is stationary, enable the static bias estimator, construct a state space model based on the prior knowledge that the true value of acceleration is zero when the vehicle is stationary, and use Kalman filtering to estimate the static biases of the yaw rate sensor, longitudinal acceleration sensor and lateral acceleration sensor respectively. Dynamic bias estimation initialization steps: When the vehicle enters a moving state, the dynamic bias estimator is enabled, and the static bias estimates of yaw rate, longitudinal acceleration and lateral acceleration output by the static bias estimator are used as the initial values ​​of yaw rate, longitudinal acceleration bias and lateral acceleration bias in the dynamic bias estimator, respectively. Dynamic bias estimation steps: Based on the extended Kalman filter (EKF) framework, a nonlinear state-space model is established that includes lateral velocity, longitudinal velocity, yaw rate, yaw rate change rate, lateral acceleration bias, and longitudinal acceleration bias. Combined with sensor measurements and auxiliary estimators, the dynamic bias of the longitudinal and lateral acceleration sensors under motion conditions is estimated online iteratively. Offset output and compensation steps: Based on the current vehicle status, select the corresponding static or dynamic offset estimate, and use the selected offset estimate to compensate the original measurement signal of the sensor corresponding to the offset estimate in real time.

2. The method according to claim 1, characterized in that, The vehicle status recognition step specifically includes: When the vehicle speed remains below the first threshold for a first time duration, the vehicle is determined to be stationary, triggering the static bias estimator and resetting its state. When the vehicle speed continues to exceed the second threshold for a second time period, it is determined that the vehicle is in motion, triggering the dynamic bias estimator and resetting its state, while locking the output of the static bias estimator.

3. The method according to claim 1, characterized in that, In the static bias estimation step, for the static bias estimation of the accelerometer, the state vector of this estimation process is: ,in , For acceleration measurement related items, , The bias term to be estimated; When initializing the posterior estimation covariance matrix, the variance of the acceleration measurement related terms is set to a first numerical range, and the variance of the bias term to be estimated is set to a second numerical range greater than the first numerical range.

4. The method according to claim 1, characterized in that, The dynamic bias estimation step is preceded by an auxiliary estimation step: Based on formula Estimate the lateral speed of the vehicle ,in The yaw rate, Let the distance from the center of the rear axle of the vehicle to the center of gravity be the estimated distance. As an observation of the EKF; The yaw rate change rate is estimated using a Kalman filter, and the estimated yaw rate change rate is used as a state variable of the EKF.

5. The method according to claim 1 or 4, characterized in that, In the dynamic bias estimation step, the state equation of the nonlinear state-space model is obtained by discretization based on the relationship between vehicle dynamics and kinematics, and its Jacobian matrix is ​​calculated in each EKF iteration based on the posterior state estimate of the previous time step.

6. The method according to claim 1, characterized in that, In the dynamic bias estimation step, the noise variance terms in the system noise covariance matrix of the EKF corresponding to the lateral velocity, yaw rate, and yaw rate change rate of the state variables are set to be greater than the values ​​of the noise variance terms corresponding to other state variables.

7. The method according to claim 1, characterized in that, In the dynamic bias estimation initialization step, the initial covariance of the bias state in the dynamic bias estimator is set to be equal to the variance of the bias term corresponding to each bias state in a physical sense in the posterior estimation covariance matrix finally output by the static bias estimator.

8. A vehicle sensor bias calibration system based on Bayesian estimation, characterized in that, The system for implementing the method as described in any one of claims 1-7 comprises: The status recognition module is used to determine the vehicle status based on the vehicle speed signal; The static bias estimation module is used to perform Kalman filtering when the vehicle is stationary to estimate the static bias of the sensors. The dynamic bias estimation module is used to perform extended Kalman filtering while the vehicle is moving to estimate the dynamic bias of the sensor. Its initialization depends on the output of the static bias estimation module. The auxiliary estimation module provides estimates of lateral velocity and yaw rate change. The compensation output module is used to compensate the sensor signal based on the current effective bias value and then output it.

9. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-7.

Images