A method for compensating the installation angle error of vehicle-mounted inertial navigation based on pattern recognition

Abstract

The present invention provides a method for compensating the installation angle error of an in-vehicle inertial navigation based on pattern recognition. First, an inertial navigation device is installed on a vehicle, six-axis data of an inertial measurement unit (IMU) is read, the output value of the z-axis of the gyroscope is projected onto the vehicle body system to obtain the vehicle body yaw angular velocity, and strapdown inertial navigation is updated using the IMU data to identify the vehicle motion mode. Then, GNSS is used for observation fusion. Finally, the vehicle speed is fused through a Kalman filter to estimate the installation angle error in real time. Compared with the existing methods, this method reduces the usage threshold of GNSS / INS integrated navigation. Only by fixing the inertial navigation device on the vehicle, the installation angle compensation can be automatically completed during driving. Moreover, it provides an accurate and real-time method for identifying the vehicle motion state, and different observation fusions can be performed according to different vehicle motion states to ensure that the estimation of the installation angle of the inertial device is not affected by incorrect observation quantities, improving the robustness and estimation accuracy of the installation angle estimation of the inertial device.

Description

Technical Field

[0001] The present invention relates to the field of positioning and navigation, and particularly to a vehicle-mounted inertial navigation installation angle error compensation method based on pattern recognition. Background Art

[0002] In the field of positioning and navigation, the Global Navigation Satellite System (GNSS) and the Inertial Navigation System (INS) are highly complementary. The combination of the two can make up for each other's weaknesses and obtain more accurate positioning information.

[0003] When integrated navigation is applied in a vehicle environment, the inertial navigation device cannot be completely aligned with the vehicle body coordinate system during installation, which will affect the calculation results of the final attitude, speed, and position. Currently, common installation angle estimation methods include the acceleration vector method, the magnetometer observation method, etc. However, complex vehicle conditions such as engine vibration and strong magnetic interference will affect the estimation effect of the existing methods. Therefore, the present invention proposes an inertial navigation installation angle error compensation method based on vehicle pattern recognition. This method introduces the vehicle speed information provided by the Controller Area Network (CAN) to provide additional observations for integrated navigation. However, in working conditions such as turning and low speed, the observation effect of the vehicle speed is not ideal. Therefore, the present invention proposes to estimate the installation angle of the inertial device by using the CAN vehicle speed and GNSS observations based on the recognition of the vehicle motion state. Summary of the Invention

[0004] In order to compensate for the inertial navigation installation angle error and improve the positioning accuracy of integrated navigation, the present invention provides a vehicle-mounted inertial navigation installation angle error compensation method based on pattern recognition. First, the vehicle motion mode is recognized; secondly, GNSS is used for observation fusion; finally, when the vehicle motion mode is in the high-speed straight state, the vehicle speed is fused through a Kalman filter to estimate the installation angle error in real time. Compared with the existing methods, this method has stronger robustness and faster estimation speed.

[0005] The present invention adopts the following technical solutions:

[0006] Step S1: Install the inertial navigation device on a test vehicle and perform system initialization.

[0007] Step S2: Read the six-axis data of the Inertial Measurement Unit (IMU) in real time, including the x, y, and z axis data of the gyroscope and accelerometer, and perform sliding window mean preprocessing on the original data.

[0008] Step S3: When the vehicle is stationary, calculate the roll mounting angle θ and pitch mounting angle θ between the inertial navigation device and the vehicle body using the accelerometer data of the IMU. Additionally, assume the heading angle θ is 0. Calculate the transformation matrix between the inertial navigation device system and the vehicle body system through the Euler angle to attitude matrix formula using θ, θ, and θ. r and pitch mounting angle θ p , and additionally, assume the heading angle θ y is 0. Let θ r , θ p and θ y be calculated to obtain the transformation matrix between the inertial navigation device system and the vehicle body system through the Euler angle to attitude matrix formula.

[0009] Step S4: Use to project the output value wbz of the gyroscope z-axis onto the vehicle body system to obtain the yaw angular velocity wbz of the vehicle body: b : v

[0010]

[0011] Step S5: Use the IMU data for strapdown inertial navigation update to obtain the updated attitude, position p, position p IMU and velocity v IMU information.

[0012] Step S6: When a GNSS signal is received, perform information fusion through a Kalman filter. The Kalman filtering process includes filter initialization, one-step prediction, and measurement update. The initialization process includes initializing the error state quantity, system covariance matrix, and posterior covariance matrix. The one-step prediction is obtained from the strapdown inertial navigation error propagation equation. In the measurement update, the differences between the three-dimensional velocity output by GNSS and the velocity updated by the strapdown inertial navigation in step S5, and the differences between the position output by GNSS and the position updated by the strapdown inertial navigation in step S5 are used as observations to finally obtain the corrected error state quantity. Then, subtract the error state quantity estimated by the Kalman from the attitude, velocity, and position output by the strapdown inertial navigation update in step S5 to obtain more accurate attitude, velocity, and position information.

[0013] Step S6 includes the following steps:

[0014] Step S6-1: Initialize the Kalman filter, including initializing the error state vector X, system noise covariance Q, and posterior covariance P.

[0015] Take the error state vector X as an 18-dimensional vector:

[0016]

[0017] where, is the estimated error of the attitude in the navigation coordinate system, δv n is the estimated error of the velocity in the navigation system, δp​n is the estimated error of the position under the navigation system, b ω and b a are the dynamic zero biases of the gyroscope and accelerometer respectively, δk is the scale factor error of the CAN vehicle speed, δθ p is the pitch angle error of the installation angle of the inertial navigation device, δθ y is the heading angle error of the installation angle of the inertial navigation device.

[0018] The system covariance matrix Q is:

[0019] Q = diag([IMU arw IMU vrw IMU gpsd IMU apsd 2 )

[0020] where, IMU arw and IMU vrw are the random walk coefficients of the gyroscope and the random walk coefficients of the accelerometer, IMU gpsd and IMU apsd are the zero bias instability coefficients of the gyroscope and accelerometer respectively.

[0021] The posterior covariance matrix P is:

[0022] P = diag([IMU install GNSS stdv GNSS stdp IMU gpsd IMU apsd IMU err 2 )

[0023] where, IMU install is the installation error included angle, GNSS stdv is the GNSS velocity error variance, GNSS stdp is the GNSS position error variance, IMU err is composed of the installation angle error and the vehicle speed scale factor.

[0024] Step S6-2, Kalman filter one-step prediction, use the strapdown inertial navigation error propagation equation for one-step prediction to obtain the prior state quantity X prior .

[0025] Step S6-3, Kalman filter measurement update, construct the GNSS measurement quantity Z gnss , the GNSS measurement matrix H gnss , perform measurement update on the prior state quantity to obtain the posterior state quantity X post .

[0026] ​​GNSS Observation Quantity Z gnss :

[0027] Z gnss = [v gnss - v IMU p gnss - p IMU T

[0028] where, v gnss is the three - dimensional velocity output by GNSS, v IMU is the three - dimensional velocity output by the strapdown inertial navigation update in step S5, p gnss is the position output by GNSS (including longitude, latitude and altitude information), p IMU is the position output by the strapdown inertial navigation update in step S5 (including longitude, latitude and altitude information).

[0029] GNSS Observation Matrix H gnss :

[0030]

[0031] where, O 3 is a 3 - order zero matrix, I 3 is a 3 - order identity matrix, R m is the meridian radius of curvature, R n is the transverse radius of curvature, L is the latitude, and h is the altitude.

[0032] After measurement update, the posterior error state vector X post :

[0033]

[0034] In the formula, the variables with "post" in the subscript are the posterior errors of the corresponding variables in the error state vector X.

[0035] Step S6 - 4. Subtract the attitude position p IMU and velocity v IMU obtained by the strapdown inertial navigation update in step S5 from the attitude error post position error δp n_post and velocity error δv n_post in the posterior state quantity X.

[0036]

[0037] p ins = p IMU - δp n_post

[0038] v​ins = v IMU -δv n_post

[0039] Wherein, p ins , v ins are respectively the attitude, position and velocity estimates obtained by integrated navigation.

[0040] Step S7: Receive vehicle speed information through CAN. When the vehicle speed information is received, use wbz v and the vehicle speed information to identify the basic motion state of the vehicle. The motion state can be divided into high-speed straight state, turning state, parking state and normal state.

[0041] The said step S7 includes the following steps:

[0042] Step S7-1: Perform sliding window mean processing on wbz v and the CAN vehicle speed to obtain the body yaw rate wbz v_filtered after mean processing and the CAN vehicle speed v filtered .

[0043] Step S7-2: When v filtered is greater than the first vehicle speed threshold, wbz v_filtered is less than the first angular velocity threshold and greater than the second angular velocity threshold, it is determined as the high-speed straight state; when wbz v_filtered is greater than the first angular velocity threshold or less than the second angular velocity threshold, it is determined as the turning state; when v filtered is less than the second vehicle speed threshold, it is determined as the parking state; in other cases, it is determined as the normal motion state;

[0044] The said first vehicle speed threshold, second vehicle speed threshold, first angular velocity threshold and second angular velocity threshold are all determined according to the empirical values of the vehicle speed and angular velocity in each operating state commonly used in this field.

[0045] Step S8: In the high-speed straight state, the CAN vehicle speed observation model selects the observation model with installation angle error added, and estimates the inertial navigation installation angle error and CAN vehicle speed scale error through the Kalman filter. In other motion states, the CAN vehicle speed observation model selects the ordinary vehicle speed observation model, and corrects other state variables through the Kalman filter without affecting the installation angle error and CAN vehicle speed scale factor error.

[0046] The said step S8 includes the following steps:

[0047] In the high-speed straight motion state, select the difference between the speed estimation result of integrated navigation and the CAN vehicle speed with scale factor added as the observable quantity Z CAN_k, select the observation matrix H that incorporates the installation angle error CAN_install Perform observation fusion.

[0048] The CAN vehicle speed observation value Z with the scale factor incorporated CAN_k :

[0049]

[0050] Among them, v ins is the combined navigation speed estimation result in step S6, is a three-dimensional vector, and its first, second, and third dimensions are respectively; is the combined navigation attitude estimation result, is the transformation matrix from the vehicle body coordinate system to the inertial navigation device system, v D is the CAN vehicle speed value, and k is the vehicle speed scale factor;

[0051] Because the result obtained by Kalman is the attitude error, after correction, the correct attitude is obtained, which is represented as Euler angles, that is Corresponding to the rotation matrix is Substantially the same as the combined navigation attitude estimation result

[0052] The observation matrix H that incorporates the installation angle error CAN_install :

[0053] H CAN_install = [M 1 M 2 O 3 O 3 O 3 M 3 T

[0054] Among them,

[0055] () × represents the skew-symmetric matrix, and O 3 is a 3-order zero matrix.

[0056] In other motion states, due to inaccurate speed observations provided by GNSS during turning or low-speed motion, the estimated speed v ins is also inaccurate. Therefore, at this time, the estimation of the installation angle of the inertial navigation device is not performed. Select the difference between the combined navigation speed estimation result and the CAN vehicle speed without the scale factor as the observation value Z CAN , select the observation matrix H without the installation angle error CAN Perform observation fusion. ​

[0057] Observed quantity Z CAN :

[0058]

[0059] Observation matrix H CAN :

[0060] H CAN =[M 1 M 2 O 3 O 3 O 3 O 3 T

[0061] Step S9. Correct the current CAN vehicle speed scale and the installation angle transformation matrix by using the CAN vehicle speed scale error, the pitch angle installation angle error of the inertial navigation device, and the heading angle installation angle error estimated in step S8.

[0062] The said step S9 includes the following steps:

[0063] Step S9-1. Correct the CAN vehicle speed scale factor of the previous step by using the estimated CAN vehicle speed scale error δk:

[0064] k=k (-) +δk

[0065] Step S9-2. Correct the installation angle coordinate transformation matrix of the previous step by using the pitch angle installation angle error of the inertial navigation device and the heading angle installation angle error of the inertial navigation device As shown in the following formula:

[0066]

[0067] Among them, is the attitude conversion matrix between the inertial navigation device system and the vehicle system, that is, the matrix representation form of the installation angle of the inertial navigation device. What is obtained in step S3 is the preliminary estimated value of the conversion matrix, and it is continuously corrected in step S9, so that this value is getting closer and closer to the actual installation angle between the inertial navigation device and the vehicle.

[0068] Compared with the prior art, the beneficial effects of the present invention are:

[0069] 1. The use threshold of GNSS / INS integrated navigation is reduced. The user does not need to care about the placement angle of the inertial navigation device. Just fix the inertial navigation device on the vehicle, and the installation angle compensation can be automatically completed during driving. This benefits from the CAN vehicle speed observation method provided by the present invention. By using the CAN vehicle speed that always points to the positive direction of the vehicle head, the installation angle of the inertial device can be estimated in real time and compensated.​

[0070] 2. Improve the robustness and calibration accuracy of the installation angle calibration of inertial devices. During the installation angle calibration of inertial devices, users do not need to care about the driving route of the vehicle, and the installation angle compensation can be completed under any driving conditions. This benefits from the accurate and real-time vehicle motion state recognition method provided by the present invention, which can perform different observation fusions for different vehicle motion states to ensure that the estimation of the installation angle of inertial devices is not affected by incorrect observation quantities, thereby improving the robustness and estimation accuracy of the installation angle estimation of inertial devices. Description of the Drawings

[0071] Figure 1 It is the flowchart of the vehicle-mounted inertial navigation installation angle error compensation in the embodiment of the present invention.

[0072] Figure 2 It is the heading angle estimation error diagram in the embodiment of the present invention. Detailed Embodiments

[0073] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer and more understandable, the spirit of the content disclosed by the present invention will be clearly described below with reference to the drawings and in detail. After any person skilled in the art in the relevant technical field understands the embodiments of the content of the present invention, the techniques taught by the content of the present invention can be changed and modified, which does not deviate from the spirit and scope of the content of the present invention. The illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0074] In this embodiment, the feasibility and superiority of the present invention are verified through actual vehicles, and the specific steps are as follows:

[0075] Step S1: Install the inertial navigation device on the test vehicle, and the x-axis of the inertial navigation device is approximately flush with the head direction of the vehicle.

[0076] Step S2: Read the six-axis data of the Inertial Measurement Unit (IMU) at a frequency of 100 hz, including the x, y, and z-axis data of the gyroscope and accelerometer. In this embodiment, the original gyroscope data wb xyz and the accelerometer velocity ab xyz are preprocessed by a sliding window mean:

[0077]

[0078]

[0079] Step S3: When the vehicle is stationary, use the accelerometer data of the IMU to calculate the roll installation angle θ r and the pitch installation angle θ p, in addition, it is considered that the heading angle θ y is 0. θ r , θ p and θ y are used to calculate the attitude transformation matrix between the inertial navigation device system and the vehicle body system through the Euler angle to attitude matrix formula In this embodiment, the installation angle θ install is as follows:

[0080] θ install =[θ r θ p θ y T =[0.021 -0.039 0] T

[0081] Step S4, use to project the output value wbz of the gyroscope z-axis onto the vehicle body system to obtain wbz b : v :

[0082]

[0083] Step S5, use the strapdown inertial navigation algorithm for update to obtain the updated attitude, position, and velocity information.

[0084] Step S6, when a GNSS signal is received, use the three-dimensional velocity and longitude and latitude information output by GNSS as observations, and perform information fusion with the attitude, position, and velocity obtained in the previous step through a Kalman filter to obtain a more reliable integrated navigation result of attitude, velocity, and position.

[0085] The said step S6 includes the following steps:

[0086] Step S6-1, initialize the Kalman filter, including initializing the error state vector X, the system noise covariance Q, and the posterior covariance P.

[0087] Take the error state vector X as an 18-dimensional vector:

[0088]

[0089] Among them, is the estimated error of the attitude in the navigation coordinate system, δv n is the estimated error of the velocity in the navigation system, δp n is the estimated error of the position in the navigation system, b ω and b a are the dynamic zero biases of the gyroscope and the accelerometer respectively, δk is the scale factor error of the CAN vehicle speed, δθ p is the pitch angle error of the installation angle of the inertial navigation device, δθ y ​It is the heading angle error of the installation angle of the inertial navigation device.

[0090] In this embodiment, the system covariance matrix Q is:

[0091] Q = diag([0.001I 3 , 0.003I 3 , 1×10 -4 I 3 , 0.003I 3 ) 2 )

[0092] The posterior covariance matrix P is:

[0093] P = diag([0.1I 3 , I 3 , 4×10 -7 I 3 , O 3 , O 3 , 0.1, 0.01, 0.01] 2 )

[0094] Step S6-2: Perform one-step prediction of Kalman filtering using the strapdown inertial navigation error propagation equation.

[0095] Step S6-3: Kalman filter measurement update, construct the GNSS measurement Z gnss , GNSS observation matrix H gnss .

[0096] The GNSS measurement Z gnss :

[0097] Z gnss = [v gnss - v IMU p gnss - p IMU T

[0098] The GNSS observation matrix H gnss :

[0099]

[0100] where O 3 is a 3×3 zero matrix, I 3 is a 3×3 identity matrix, R m is the radius of meridian curvature, R n is the radius of transverse curvature, L is the latitude, and h is the altitude.

[0101] Step S6-4: Use the error estimated by the Kalman filter to correct the system state vector, and finally output the positioning and attitude estimation results.

[0102] Step S7: Receive CAN vehicle speed information at a frequency of 10 hz. When the vehicle speed information is received, use wbz v and the vehicle speed information to identify the basic motion state of the vehicle. The motion state can be divided into a high-speed straight state, a turning state, a parking state, and a normal state.

[0103] The said step S7 includes the following steps:

[0104] Step S7-1: Perform a sliding window mean process on wbz v and the CAN vehicle speed.

[0105] Step S7-2: When the vehicle speed is greater than the first vehicle speed threshold, the angular velocity of the gyro z-axis is less than the first angular velocity threshold and greater than the second angular velocity threshold, it is determined as the high-speed straight state; when the angular velocity of the gyro z-axis is greater than the third angular velocity threshold or less than the fourth angular velocity threshold, it is determined as the turning state; when the vehicle speed is less than the second vehicle speed threshold, it is determined as the parking state; in other cases, it is determined as the normal motion state.

[0106] In this embodiment, the following thresholds are adopted. The first vehicle speed threshold is 5 m / s, and the second vehicle speed threshold is 0.01 m / s; the first angular velocity threshold is 0.05 rad / s, the second angular velocity threshold is -0.05 rad / s, the third angular velocity threshold is 0.1 rad / s, and the fourth angular velocity threshold is -0.1 rad / s.

[0107] Step S8: In the high-speed straight state, the CAN vehicle speed observation model selects the observation model with the installation angle error added, and estimates the inertial navigation installation angle error and the CAN vehicle speed scale error through the Kalman filter. In other motion states, the CAN vehicle speed observation model selects the normal vehicle speed observation model, and corrects other state variables through the Kalman filter without affecting the installation angle error and the CAN vehicle speed scale error.

[0108] The said step S8 includes the following steps:

[0109] In the high-speed straight motion state, select the difference between the speed estimation result of the integrated navigation and the CAN vehicle speed as the observation quantity Z, and select the observation matrix H with the installation angle error added for observation fusion.

[0110] Observation quantity Z CAN_k :

[0111]

[0112] Among them, v n is the speed estimation result of the integrated navigation in step S6, is the attitude estimation result of the integrated navigation, is the transformation matrix from the vehicle system to the inertial navigation equipment system, v D is the CAN vehicle speed value, and k is the CAN vehicle speed scale factor.

[0113] Observation matrix H CAN_install :

[0114] H CAN_install =[M 1 M 2 O 3 O 3 O 3 M 3 T

[0115] wherein, () × represents an anti-symmetric matrix, and O 3 is a 3rd-order zero matrix.

[0116] In other motion states, due to inaccurate speed observations provided by GNSS during turning or low-speed motion, the estimated speed v ins is also inaccurate. Therefore, the installation angle of the inertial navigation equipment is not estimated at this time. Select the following observation quantity Z and observation matrix H for observation fusion.

[0117] Observation quantity Z CAN :

[0118]

[0119] Observation matrix H CAN :

[0120] H CAN =[M 1 M 2 O 3 O 3 O 3 O 3 T

[0121] Step S9. Use the CAN vehicle speed scale error, the pitch angle installation angle error of the inertial navigation equipment, and the heading angle installation angle error of the inertial navigation equipment estimated in step S7 to correct the current CAN vehicle speed scale and the installation angle transformation matrix.

[0122] The said step S9 includes the following steps:

[0123] Step S9-1. Use the estimated CAN vehicle speed scale error to correct the CAN vehicle speed scale of the previous step:

[0124] k = k (-) + δk​​

[0125] Step S9-2: Correct the installation angle coordinate transformation matrix of the previous step by using the pitch angle installation angle error and the heading angle installation angle error of the inertial navigation device As shown in the following formula:

[0126]

[0127] Wherein, is the attitude transformation matrix between the inertial navigation device system and the vehicle system, that is, the matrix representation form of the installation angle of the inertial navigation device. As time goes by, The estimated value of will get closer and closer to the actual installation angle between the inertial navigation device and the vehicle.

[0128] Figure 2 This is the heading installation angle estimation error graph of this embodiment. The initial heading installation angle has an error of 30°. Within 100 s of the algorithm running, the heading installation angle can converge to near the true value, and the error angle is less than 3°. Compared with the method of calculating the installation angle of the inertial device by using the accelerometer and geomagnetic observation, the installation angle compensation method provided by the present invention has stronger robustness and estimation accuracy.

[0129] The above embodiments are the preferred embodiments of the present invention, but the embodiments of the present invention are not limited by the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement methods and are all included in the protection scope of the present invention.

Claims

1. A method for compensating the installation angle error of vehicle-mounted inertial navigation based on pattern recognition, characterized in that: The method includes the following steps: Step S1: Install the inertial navigation device on the vehicle and perform system initialization; Step S2: Read the six-axis data of the inertial measurement unit (IMU) in real time, including the x, y, and z-axis data of the gyroscope and accelerometer, and perform sliding window mean preprocessing on the original data; Step S3: When the vehicle is stationary, calculate the roll mounting angle θ between the inertial navigation device and the vehicle body using the accelerometer data of the IMU r and the pitch mounting angle θ p . Additionally, assume the heading mounting angle θ y to be 0; Calculate the transformation matrix between the inertial navigation device system and the vehicle body coordinate system by the Euler angle to attitude matrix formula using θ r , θ p and θ y . Step S4: Use the transformation matrix between the inertial navigation device system and the vehicle body coordinate system to project the output value of the gyroscope z-axis onto the vehicle body system to obtain the vehicle body yaw angular velocity; Step S5: Use the IMU data to perform strapdown inertial navigation update to obtain the updated attitude, position p , and velocity v IMU ; position p IMU and velocity v IMU ; Step S6: When receiving the GNSS signal, use the three-dimensional velocity and longitude and latitude information output by the GNSS as observations, and perform information fusion with the updated attitude, position, and velocity obtained in the previous step through the Kalman filter to obtain the estimation results of the integrated navigation attitude, position, and velocity; Step S7: Receive the vehicle speed information through CAN, and use the vehicle body yaw angular velocity and CAN vehicle speed information to identify the basic motion state of the vehicle. The motion states are divided into high-speed straight state, turning state, parking state, and normal motion state; Step S8: In the high-speed straight state, the CAN vehicle speed observation model selects the observation matrix with the installation angle error added, and estimates the installation angle error of the inertial navigation device and the CAN vehicle speed scale error through the Kalman filter; in other motion states, the CAN vehicle speed observation model selects the observation matrix without the installation angle error added, and corrects other state variables through the Kalman filter without affecting the installation angle error and the CAN vehicle speed scale error; Step S9: Use the CAN vehicle speed scale error, the pitch installation angle error of the inertial navigation device, and the heading installation angle error of the inertial navigation device obtained in Step S8 to correct the current CAN vehicle speed scale and the installation angle transformation matrix.

2. The method for compensating the installation angle error of vehicle-mounted inertial navigation based on pattern recognition according to claim 1, characterized in that, In step S4, using the transformation matrix between the inertial navigation device system and the vehicle body coordinate system project the output value wbz of the z-axis of the gyroscope b onto the vehicle body system to obtain the vehicle body yaw angular velocity wbz v , specifically:

3. The method for compensating the installation angle error of vehicle-mounted inertial navigation based on pattern recognition according to claim 1, characterized in that, The specific steps of Step S6 include the following steps: Step S6-1: Initialize the Kalman filter, including the initialization of the error state vector X, the system noise covariance Q, and the posterior covariance P; Take the error state vector X as an 18-dimensional vector: Among them, is the estimated error of the attitude in the navigation coordinate system, δv n is the estimated error of the velocity in the navigation coordinate system, δp n is the estimated error of the position in the navigation coordinate system, b ω and b a are the dynamic zero biases of the gyroscope and accelerometer respectively, δk is the CAN vehicle speed scale factor error, δθ p is the pitch mounting angle error of the inertial navigation device, δθ y is the heading mounting angle error of the inertial navigation device; The system covariance matrix Q is: Q = diag([IMU arw IMU vrw IMU gpsd IMU apsd 2 )​ Among them, IMU arw and IMU vrw are the random walk coefficients of the gyroscope and the random walk coefficients of the accelerometer, and IMU gpsd and IMU apsd are the bias instability coefficients of the gyroscope and the accelerometer respectively; The posterior covariance matrix P is: P = diag([IMU install GNSS stdv GNSS stdp IMU gpsd IMU apsd IMU err 2 )​ where, IMU install is the installation error included angle, GNSS stdv is the GNSS velocity error variance, GNSS stdp is the GNSS position error variance, IMU err is composed of the installation angle error angle and the vehicle speed scale factor; Step S6-2: Perform a Kalman one-step prediction using the strapdown inertial navigation error propagation equation to obtain the prior state quantity X pr i or ; Step S6-3, Kalman filter measurement update, construct GNSS measurement Z gnss , GNSS observation matrix H gnss , perform measurement update on the prior state quantity X prior to obtain the posterior error state vector X post ; The GNSS observation quantity Z described above gnss : Z gnss = [v gnss - v IMU p gnss - p IMu T ​ Among them, v gnss is the three-dimensional velocity output by GNSS, and v IMU is the velocity after the strapdown inertial navigation update in step S5. p gnss is the position output by GNSS, and p IMU is the position after the strapdown inertial navigation update in step S5; The GNSS observation matrix H described above gnss : Among them, O 3 is a 3rd-order zero matrix, I 3 is a 3rd-order identity matrix, R m is the meridian curvature radius, R n is the transverse curvature radius, L is the latitude, and h is the altitude; After measurement update, the posterior error state vector X is obtained post : In the formula, the variables with "post" in the subscript are the posterior errors of the corresponding variables in the error state vector X; Step S6-4: Subtract the attitude error position p IMU and velocity v IMU in the posterior error state vector X post from the attitude position error δp n_post and velocity error δv n_post obtained by updating the strapdown inertial navigation in Step S5: p ins = p IMU - δp n_post v ins = v IMU - δv n_post wherein p ins and v ins are respectively the estimated results of the combined navigation attitude, position and velocity.

4. The method for compensating the installation angle error of vehicle-mounted inertial navigation based on pattern recognition according to claim 1, characterized in that: The specific steps of Step S7 include the following steps: Step S7-1: Perform sliding window mean processing on wbz v and the CAN vehicle speed to obtain the body yaw rate wbz v_filtered after mean processing and the CAN vehicle speed v filtered ; Step S7-2, when v filtered is greater than the first vehicle speed threshold, and wbz v_filtered is less than the first angular velocity threshold and greater than the second angular velocity threshold, it is determined to be in a high-speed straight state; when wbz v_filtered is greater than the first angular velocity threshold or less than the second angular velocity threshold, it is determined to be in a turning state; when v filtered is less than the second vehicle speed threshold, it is determined to be in a parking state; in other cases, it is determined to be in a normal motion state.

5. The method for compensating the installation angle error of vehicle-mounted inertial navigation based on pattern recognition according to claim 1, characterized in that, The specific steps of Step S8 are as follows: Under the high-speed straight-line motion state, the difference between the speed estimation result of the integrated navigation and the CAN vehicle speed with the scale factor added is selected as the observation quantity Z CAN_k , and the observation matrix H with the installation angle error added is selected CAN_install for observation fusion; The observed quantity Z CAN_k : Among them, v ins is the combined navigation speed estimation result in step S6, is a three-dimensional vector, and are its first dimension, second dimension, and third dimension respectively; is the combined navigation attitude estimation result, is the transformation matrix between the inertial navigation device system and the vehicle body coordinate system, v D is the CAN vehicle speed value, and k is the CAN vehicle speed scale factor; Observation matrix \(H\) with installation angle error added CAN_install : H CAN_install = [M 1 M 2 O 3 O 3 O 3 M 3 T ​ Among them, () × represents a skew-symmetric matrix, and O 3 is a 3rd-order zero matrix; In other motion states, the difference between the speed estimation result of the integrated navigation and the CAN vehicle speed without adding the scale factor is selected as the observation quantity Z CAN , and the observation matrix H without adding the installation angle error is selected CAN Perform observation fusion: The observed quantity Z CAN : The observation matrix H mentioned above CAN : H CAN = [M 1 M 2 O 3 O 3 O 3 O 3 T 。​ 6. The method for compensating the installation angle error of vehicle-mounted inertial navigation based on pattern recognition according to claim 1, characterized in that, The specific steps of Step S9 include the following steps: Step S9-1: Use the CAN vehicle speed scale factor error to correct the CAN vehicle speed scale factor k: k = k (-) + δk Among them, k (-) represents the CAN vehicle speed scale factor at the previous moment, k is the CAN vehicle speed scale factor, and δk is the CAN vehicle speed scale factor error; Step S9-2: Use the pitch installation angle error δθ of the inertial navigation device p and the heading installation angle error δθ of the inertial navigation device y to correct the transformation matrix between the inertial navigation device system and the vehicle body coordinate system in the previous step The formula is as follows: Among them, I 3 is a 3-order identity matrix, is the transformation matrix between the inertial navigation equipment system and the vehicle body coordinate system at the previous moment.

Images