CN113029173A - Vehicle navigation method and device - Google Patents
Vehicle navigation method and device Download PDFInfo
- Publication number
- CN113029173A CN113029173A CN202110255357.4A CN202110255357A CN113029173A CN 113029173 A CN113029173 A CN 113029173A CN 202110255357 A CN202110255357 A CN 202110255357A CN 113029173 A CN113029173 A CN 113029173A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- inertial navigation
- covariance
- data
- output state
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000005259 measurement Methods 0.000 claims description 30
- 230000001133 acceleration Effects 0.000 claims description 28
- 238000013528 artificial neural network Methods 0.000 claims description 22
- 238000012549 training Methods 0.000 claims description 21
- 230000000306 recurrent effect Effects 0.000 claims description 18
- 238000012545 processing Methods 0.000 claims description 15
- 238000007781 pre-processing Methods 0.000 claims description 8
- 238000005516 engineering process Methods 0.000 abstract description 7
- 239000011159 matrix material Substances 0.000 description 30
- 230000008569 process Effects 0.000 description 10
- 238000010586 diagram Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000005070 sampling Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 230000007704 transition Effects 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000013434 data augmentation Methods 0.000 description 2
- 238000005034 decoration Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000004880 explosion Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000013519 translation Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009194 climbing Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000005295 random walk Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
- G01C21/34—Route searching; Route guidance
- G01C21/3407—Route searching; Route guidance specially adapted for specific applications
- G01C21/343—Calculating itineraries, i.e. routes leading from a starting point to a series of categorical destinations using a global route restraint, round trips, touristic trips
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
- G01C21/34—Route searching; Route guidance
- G01C21/3446—Details of route searching algorithms, e.g. Dijkstra, A*, arc-flags, using precalculated routes
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C25/00—Manufacturing, calibrating, cleaning, or repairing instruments or devices referred to in the other groups of this subclass
- G01C25/005—Manufacturing, calibrating, cleaning, or repairing instruments or devices referred to in the other groups of this subclass initial alignment, calibration or starting-up of inertial devices
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Manufacturing & Machinery (AREA)
- Navigation (AREA)
Abstract
The invention discloses a vehicle navigation method and device. Wherein, the method comprises the following steps: acquiring inertial navigation data of a vehicle motion path; determining covariance used in filter updating at each moment through a noise parameter adapter according to inertial navigation data; updating output state parameters determined by the filter according to inertial navigation data according to the covariance, wherein the output state parameters comprise deviation estimated values of the inertial navigation data and the covariance corresponding to the deviation estimated values; and navigating the vehicle according to the updated output state parameters. The invention solves the technical problem of poor precision of the inertial navigation positioning mode of the vehicle in the related technology.
Description
Technical Field
The invention relates to the field of navigation, in particular to a vehicle navigation method and device.
Background
The intelligent vehicle is a comprehensive system integrating functions of environmental perception, planning decision, multi-level auxiliary driving and the like, intensively applies technologies such as computer, modern sensing, information fusion, communication, artificial intelligence, automatic control and the like, and is a typical high and new technology complex. It is necessary to know where the smart vehicle is in the environment and how the vehicle is passing through the environment. The information measured by imaging sensors such as laser radar and ultrasonic systems can be verified through accurate vehicle dynamic estimation values. This ensures that the entire movement is safe. The global navigation satellite system can provide global navigation positioning services, but signals may be lost when passing through some dense building areas or tunnels, so that phase tracking is lost, and the global navigation satellite system has the disadvantages of poor real-time performance, insufficient navigation positioning accuracy and the like. An inertial measurement unit is a device that measures the three-axis attitude angle (or angular rate) and acceleration of an object. Generally, an IMU includes three single-axis accelerometers and three single-axis gyroscopes, the accelerometers detect acceleration signals of an object in three independent axes of a carrier coordinate system, and the gyroscopes detect angular velocity signals of the carrier relative to a navigation coordinate system, and measure angular velocity and acceleration of the object in three-dimensional space, and then solve the attitude of the object. For these reasons, inertial measurement units have become a key component of intelligent vehicles. An Inertial Navigation System (INS), also called an Inertial reference System, is an autonomous Navigation System that does not rely on external information, nor radiates energy to the outside (as in radio Navigation). The working environment of the device not only comprises the air and the ground, but also can be underwater. Conventional inertial navigation systems use an IMU to perform an integration operation over time. The problems of large error and poor accuracy exist.
In view of the above problems, no effective solution has been proposed.
Disclosure of Invention
The embodiment of the invention provides a vehicle navigation method and device, which at least solve the technical problems of poor precision of inertial navigation positioning modes of vehicles in the related technology.
According to an aspect of an embodiment of the present invention, there is provided a vehicle navigation method including: acquiring inertial navigation data of a vehicle motion path; determining covariance used in filter updating at each moment through a noise parameter adapter according to the inertial navigation data; updating output state parameters determined by the filter according to the inertial navigation data according to the covariance, wherein the output state parameters comprise deviation estimated values of the inertial navigation data and covariance corresponding to the deviation estimated values; and navigating the vehicle according to the updated output state parameters.
Optionally, the obtaining inertial navigation data of the vehicle motion path includes: acquiring three-axis acceleration and three-axis angular velocity of the vehicle through an accelerometer and a gyroscope; and preprocessing the acquired triaxial acceleration and triaxial angular velocity, wherein the preprocessing comprises error compensation processing and temperature compensation processing.
Optionally, determining, by the noise parameter adaptor according to the inertial navigation data, a covariance used in updating the filter at each time includes: inputting the acquired three-axis acceleration and three-axis angular velocity into the noise parameter adapter; the noise parameter adapter calculates the covariance used in the filter update at each time instant by a recurrent neural network.
Optionally, before inputting the acquired three-axis acceleration and three-axis angular velocity into the noise parameter adapter, the method further includes: sampling a part of data from the collected triaxial acceleration data and triaxial angular velocity data as training data; calculating an error between an output state parameter obtained through the filter according to the training data and a real track corresponding to the training data; carrying out weight adjustment on a recurrent neural network of the noise parameter adapter according to the error; and updating the adjusted recurrent neural network through an optimizer.
Optionally, before updating the output state parameter determined by the filter according to the inertial navigation data according to the covariance, the method further includes: inputting the inertial navigation data into the filter; and calculating the inertial navigation data through the filter according to the iterative extended Kalman filter IEKF to obtain corresponding output state parameters.
Optionally, updating the output state parameter determined by the filter according to the inertial navigation data according to the covariance includes: updating a virtual observation value used for determining an output state according to the covariance; and obtaining updated output state parameters through Extended Kalman Filter (EKF) according to the updated virtual observed value.
Optionally, navigating the vehicle according to the updated output state parameter includes: determining the relative positions of the accelerometer and the gyroscope and the position origin of the vehicle according to the positions of the accelerometer and the gyroscope on the vehicle; determining a pseudo-measurement value from the relative position; inputting the pseudo measurement value into the filter to obtain a coordinate of a position origin of the vehicle; and determining whether the navigation route of the vehicle is established or not according to the coordinates of the position origin of the vehicle and the output state parameter.
According to another aspect of the embodiments of the present invention, there is also provided a vehicular navigation apparatus including: the acquisition module is used for acquiring inertial navigation data of a vehicle motion path; the determining module is used for determining covariance used in filter updating at each moment through a noise parameter adapter according to the inertial navigation data; the updating module is used for updating output state parameters determined by the filter according to the inertial navigation data according to the covariance, wherein the output state parameters comprise deviation estimated values of the inertial navigation data and the covariance corresponding to the deviation estimated values; and the navigation module is used for navigating the vehicle according to the updated output state parameters.
According to another aspect of the embodiments of the present invention, there is also provided a processor for executing a program, wherein the program executes to perform the vehicle navigation method according to any one of the above.
According to another aspect of the embodiments of the present invention, there is also provided a computer storage medium including a stored program, wherein when the program runs, an apparatus in which the computer storage medium is located is controlled to execute the vehicle navigation method according to any one of the above.
In the embodiment of the invention, inertial navigation data of a vehicle motion path is acquired; determining covariance used in filter updating at each moment through a noise parameter adapter according to inertial navigation data; updating output state parameters determined by the filter according to inertial navigation data according to the covariance, wherein the output state parameters comprise deviation estimated values of the inertial navigation data and the covariance corresponding to the deviation estimated values; according to the updated output state parameters, the vehicle is navigated, and the purpose of accurately navigating the vehicle according to the updated output state parameters is achieved, so that the technical effect of improving the accuracy of vehicle navigation is achieved, and the technical problem of poor precision of an inertial navigation positioning mode of the vehicle in the related technology is solved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the invention without limiting the invention. In the drawings:
FIG. 1 is a flow chart of a method of navigating a vehicle according to an embodiment of the invention;
FIG. 2 is a schematic diagram of a navigational positioning system configuration according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a flow of a navigation positioning method according to an embodiment of the present invention;
fig. 4 is a schematic diagram of a vehicular navigation apparatus according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
In accordance with an embodiment of the present invention, there is provided a method embodiment of a vehicle navigation method, it should be noted that the steps illustrated in the flowchart of the accompanying drawings may be executed in a computer system, such as a set of computer executable instructions, and that while a logical order is illustrated in the flowchart, in some cases the steps illustrated or described may be executed in an order different than that presented herein.
Fig. 1 is a flowchart of a vehicle navigation method according to an embodiment of the present invention, as shown in fig. 1, the method including the steps of:
step S102, obtaining inertial navigation data of a vehicle motion path;
step S104, determining covariance used in filter updating at each moment through a noise parameter adapter according to inertial navigation data;
step S106, updating output state parameters determined by the filter according to inertial navigation data according to the covariance, wherein the output state parameters comprise deviation estimated values of the inertial navigation data and the covariance corresponding to the deviation estimated values;
and S108, navigating the vehicle according to the updated output state parameters.
Through the steps, receiving an instruction, wherein the instruction is used for requesting industrial data; under the condition that the account corresponding to the instruction is a common account, accessing the database according to the instruction to obtain first corpus data; when the account corresponding to the instruction is an abnormal account and the instruction comprises an opening instruction, opening a corpus corresponding to the opening instruction according to the opening instruction to access to obtain second corpus data, wherein the corpus knowledge base comprises a plurality of corpuses which are in a normally closed state; under the condition that an account corresponding to the instruction is an uncommon account and the instruction does not include a starting command, the method for accessing the database to obtain the first corpus data according to the instruction achieves the purpose of performing different data queries through different user accounts by performing ordinary responses through the first corpus and performing unusual responses through the second corpus, thereby achieving the technical effect of ensuring that different users can perform different data queries while reducing the storage pressure of the system, and further solving the technical problems of poor precision and inertia navigation positioning mode of vehicles in the related technology.
The above-mentioned inertial navigation data of obtaining the vehicle movement path includes: acquiring three-axis acceleration and three-axis angular velocity of the vehicle through an accelerometer and a gyroscope; and preprocessing the acquired triaxial acceleration and triaxial angular velocity, wherein the preprocessing comprises error compensation processing and temperature compensation processing.
Acquiring inertial navigation data information of an Inertial Measurement Unit (IMU) fixed on the trolley by using a central controller, wherein the inertial navigation data information comprises data acquired by a gyroscope and an accelerometer; and (4) carrying out respective error compensation on the data acquired by the gyroscope and the accelerometer obtained in the step one through a central controller to obtain an observed value. And acceleration and angular velocity information acquired by the inertial measurement unit IMU is utilized, wherein the acceleration and angular velocity information comprises triaxial accelerometer data and triaxial gyroscope data measured by the inertial measurement unit IMU. The collected data are respectively sent to an acceleration signal processing unit and an angular rate signal processing unit and then sent to an ARM chip through an AD converter. Various assumptions about the motion state may be automatically detected in real time based on measurements of the IMU.
The accelerometer and the gyroscope respectively obtain three-axis acceleration information and three-axis angular velocity information, and the data is preprocessed through the data processing module, wherein the preprocessing comprises temperature compensation of devices. Therefore, the acquired inertial navigation data is corrected, and the accuracy of data processing is improved.
Optionally, determining, by the noise parameter adaptor according to the inertial navigation data, a covariance used in updating the filter at each time includes: inputting the acquired three-axis acceleration and three-axis angular velocity into a noise parameter adapter; the noise parameter adapter calculates the covariance used in the filter update at each instant of time through a recurrent neural network.
Establishing a noise parameter adapter, measuring the covariance N used by the noise parameter adapter in calculating filter updates at each time kk+1The basic core of the noise parameter adapter is a recurrent neural network. The network takes a window of N inertial measurements as input and performs the calculations.
The complete structure of the noise parameter adapter consists of a set of recurrent neural network RNN layers and then an output vector
The complete layer of (2). Covariance
The calculation is as follows:
the introduction of the hyper-parameter β serves to control the upper and lower bounds of the covariance expansion, so that the dynamic range of the covariance is 10-β～10β. When beta epsilon R is more than 0, and sigma in the formulalatAnd σupCorresponding to an initial guess of the noise parameters. Thus, the network may expand the covariance by a factor of 10βAnd compresses it to a factor of 10 with respect to its original value-β。
For process noise parameter QnIt was fixed to Q.
Optionally, before inputting the acquired three-axis acceleration and three-axis angular velocity into the noise parameter adapter, the method further includes: sampling a part of data from the collected triaxial acceleration data and triaxial angular velocity data as training data; calculating an error between an output state parameter obtained through a filter according to the training data and a real track corresponding to the training data; carrying out weight adjustment on a recurrent neural network of the noise parameter adapter according to the error; and updating the adjusted recurrent neural network through an optimizer.
In particular, the goal is to optimize the filter estimation
Calculated relative translational error trel. It is the average incremental error of all possible subsequences of length 100m, …,800 m.
Then, a learning rate of 10 is selected-4The Adam optimizer of (1) to update the parameters. Training is to repeat the following iterations for the truncated time period:
iteration step one: selecting data: sampling a portion of a data set;
and (5) iteration step two: calculating the error between the output of the filter and the true trajectoryεThe parameters of the filter, including the covariance matrix, etc., are then updated according to a formula. For the measurement covariance R in the measurement equation, the parameter in the filter is generally set by a person, and is considered as a variable here. The error epsilon is then transferred according to the chain ruleReturning, sending to the network for weight adjustment;
and a third iteration step: network parameters are updated with gradients and optimizers.
The number of epochs is very large, even infinite, where 600 was chosen. Dropout and data augmentation are used in training to avoid the occurrence of overfitting. Dropout refers to randomly dropping network elements during the training process, setting the probability p of any RNN network element set during the sequence iteration to zero at 0.5. 9 1-minute sequences were sampled as one batch, where each sequence started randomly from any time. Adding a standard deviation of 10 to the data-4Small gaussian noise, i.e. data expansion techniques. Step two is iterated using standard back-propagation computation, and finally the gradient norm is reduced to a maximum value of 1 to avoid gradient explosion in step iteration three.
Optionally, before updating the output state parameter determined by the filter according to the inertial navigation data according to the covariance, the method further includes: inputting inertial navigation data into a filter; and calculating the inertial navigation data through a filter according to the iterative extended Kalman filter IEKF to obtain corresponding output state parameters.
Optionally, updating the output state parameter determined by the filter according to the inertial navigation data according to the covariance includes: updating a virtual observation value used for determining an output state according to the covariance; according to the updated virtual observed value, an updated output state parameter is obtained through Extended Kalman Filter (EKF).
Output state x of IEKF (Iterative extended Kalman Filter, Iterative Kalman Filter, also known as Iterative generalized Kalman Filter)nIncluding the position, velocity, attitude deviation and its covariance of the IMU.
Wherein the input to the system is a measurement of the IMU, wherein
The attitude, the speed and the position of the carrier at the k step (or at the time t) are all in a navigation coordinate system. Each variable being a transformation of the carrier system to the navigation system, in which
Is the zero offset of the IMU.
Is an external reference matrix.
Including vehicle attitude, speed and IMU bias.
xk+1＝f(xk,uk)+wk；
Wherein xkRepresenting the state to be estimated, ukIs an input, wkIs process noise, assuming the noise is Gaussian, the mean is zero, and the covariance matrix is Qk. Assuming that the side information takes the form of a loose equality constraint, then h (x)k) 0 is available. It is then customary to generate virtual observations from the constraint function:
yk＝h(xk)+nk；
assuming that the noise is a central Gaussian nk～N(0，Nk) Wherein the covariance matrix NkSetting and reflecting the validity of the information by the user: n is a radical ofkThe larger the confidence in the information is.
Starting with an initial gaussian confidence in the state,
wherein
Representing the initial estimate, the covariance matrix P0With the uncertainty associated with it, the EKF alternates between the two steps. In the prediction step, the value is estimated
In case of noise off, i.e.:
wherein FkIs a state transition matrix, Fk TIs a transpose of the state transition matrix. G iskIs f (-) with respect to xkAnd ukJacobian matrix of GT kIs GkThe transposing of (1). In the updating step, taking into account the pseudo-measurements, the Kalman equation gives the estimate updated accordingly
And its covariance matrix Pk+1. To implement EKF, the designer needs to determine the functions f (-) and h (-) and the associated process noise matrix QkAnd Nk. The embodiment adopts the neural network to carry out noise parameter QkAnd NkAnd performing integral learning.
Optionally, navigating the vehicle according to the updated output state parameter includes: determining the relative positions of the accelerometer and the gyroscope and the position origin of the vehicle according to the positions of the accelerometer and the gyroscope on the vehicle; determining a pseudo-measurement value from the relative position; inputting the pseudo measured value into a filter to obtain the coordinate of the position origin of the vehicle; and determining whether the navigation route of the vehicle is established or not according to the coordinates of the position origin of the vehicle and the output state parameter.
Since the Inertial Measurement Unit (IMU) is also rigidly attached to the car,
and
representing the misalignment between the IMU and the frame, so they are approximately constant.
The velocity of the origin of the car, expressed in the reference coordinate system, is:
From the basic helix theory, where Pn is the vehicle to IMU horizontal arm. The lateral and vertical velocities of the car are substantially zero, that is, two scalar false observations are generated:
wherein the noise is
Being a central Gaussian noise, covariance matrix Nk∈R2×2Is gaussian noise. Then inputting the pseudo-measured value to the filter
Determining whether the navigation route of the vehicle is established based on the coordinates of the position origin of the vehicle and the output state parameter may be determining a measured value of the IMU of the vehicle based on the output state parameter to automatically detect whether various assumptions about the motion state are established in conjunction with the coordinates of the position origin of the vehicle.
It should be noted that the present application also provides an alternative implementation, and the details of the implementation are described below.
The embodiment provides a deep learning navigation positioning method, which comprises the following steps: acquiring navigation data of a vehicle movement path, and processing abnormal data of the movement path data; normalizing the corrected data; determining input and output data of a neural network, determining the number of neurons of an optimal hidden layer, and establishing a recurrent neural network; and (4) predicting by using the trained recurrent neural network. The embodiment has high prediction precision and can be widely applied to navigation of other wheeled vehicles.
In the embodiment, the vehicle motion data is acquired by the three-axis accelerometer and the three gyroscopes for training, and the trained recurrent neural network is used for prediction. A covariance matrix of the invariant extended kalman filter is estimated. The method has the advantages of high precision, low cost, small volume and strong real-time property.
Fig. 2 is a schematic diagram of a structure of a navigation positioning system according to an embodiment of the present invention, fig. 3 is a schematic diagram of a flow of a navigation positioning method according to an embodiment of the present invention, as shown in fig. 2 and fig. 3, the present embodiment discloses a navigation positioning method for deep learning, which aims to solve a disadvantage that a gradient is easy to disappear during training of an indoor inertial navigation method improved based on a bp neural network in the related art, and a disadvantage that a precision is deteriorated due to an increase in covariance matrix parameters when a navigation method and a system based on extended kalman particle filtering encounter a climbing turn.
In order to solve the above technical problem, the present embodiment adopts the following technical solutions: the method comprises the following steps:
the method comprises the following steps: acquiring data information of an IMU (inertial measurement Unit) fixed on the trolley by using a central controller, wherein the data information comprises data acquired by a gyroscope and an accelerometer;
step two: the central controller carries out respective error compensation on the data acquired by the gyroscope and the accelerometer obtained in the step one to obtain an observed value;
step three: training the recurrent neural network model to determine the optimal hidden layer. And normalizing the load data into [0, 1] by using a normalization formula so that the load data are in the same number level, thereby accelerating the convergence speed of the neural network.
Step four: inputting the observed values of the gyroscope and the accelerometer into a designed RNN model for training; on the other hand, the observation values of the gyroscope and the accelerometer are used as the input of the invariant extended Kalman filtering;
step five: additional data sets are used as tests of the network and compared to the real results to derive accuracy information.
Acceleration and angular velocity information acquired by an Inertial Measurement Unit (IMU) is provided, wherein the Inertial Measurement Unit (IMU) measures data including three-axis accelerometer data and three-axis gyroscope data. The collected data are respectively sent to an acceleration signal processing unit and an angular rate signal processing unit and then sent to an ARM chip through an AD converter. The algorithm may automatically detect in real time whether various assumptions about the motion state hold based on measurements of the IMU. The detector is built on the basis of a recurrent neural network and is trained by using real data of the IMU ground. A state-of-the-art invariant extended kalman filter is implemented, which uses the output of the detector as a pseudo measurement value, statistically combining it with the IMU output. It produces accurate estimates of position, attitude, velocity and sensor bias and associated covariance; the method is not limited to IMU-based navigation, and the trained detector may be a module of any positioning algorithm.
The embodiment discloses a vehicle dead reckoning method based on combination of an invariant extended Kalman filter and a recurrent neural network. The method comprises the steps of data acquisition of an inertial navigation system, data preprocessing of the inertial navigation system, establishment of an IMU (inertial measurement Unit) model, establishment of a Kalman filter model, establishment of a noise parameter adapter (network model design) and test and evaluation of the network model.
Data acquisition of an inertial navigation system: the accelerometer and the gyroscope respectively obtain three-axis acceleration information and three-axis angular velocity information, and the data is preprocessed through the data processing module, wherein the preprocessing comprises temperature compensation of devices.
The method comprises the following steps: establishing an IMU model: with CkE SO (3) represents the IMU direction, i.e. the mapping of the carrier coordinate system to the rotation matrix of the navigation coordinate system, by
Indicating its speed in the navigation system, by
Representing its position in the navigation system, the kinetic model is as follows:
wherein samples are taken between two discrete times at dt, wherein the IMU speed is
Its position in the navigation coordinate system
Is a 3 x 3 directional cosine matrix representing the IMU direction. Finally (y) x represents and y ∈ R3Cross product of (d) is related to a skew symmetric matrix. True angular velocity ωk∈R3And true specific acceleration ak∈R3Is an input to the system. In the application scenario of the embodiment, the influence of earth rotation and coriolis acceleration is ignored, the earth is considered to be flat, and the gravity vector g epsilon R3Is a known constant.
The IMU provides ω as followsn∈R3And an∈R3Noise and biased measurements of (a).
Wherein
Is a quasi-constant deviation of the phase of the signal,
is zero mean gaussian noise. The deviation is a random walk.
Wherein
Is zero mean gaussian noise.
Step two: establishing an invariant extended Kalman filter:
output state x of IEKF (iterative extended Kalman Filter)nIncluding the position, velocity, attitude deviation and its covariance of the IMU.
Wherein the input to the system is a measurement of the IMU, wherein
The attitude, the speed and the position of the carrier at the k step (or at the time t) are all in a navigation coordinate system. Each variable being a transformation of the carrier system to the navigation system, in which
Is the zero offset of the IMU.
Is an external reference matrix.
Including vehicle attitude, speed and IMU bias.
xk+1＝f(xk,uk)+wk；
Wherein xkRepresenting the state to be estimated, ukIs an input, wkIs process noise, assuming the noise is Gaussian, the mean is zero, and the covariance matrix is Qk. Assuming that the side information takes the form of a loose equality constraint, then h (x)k) 0 is available. It is then customary to generate virtual observations from the constraint function:
yk＝h(xk)+nk；
assuming that the noise is a central Gaussian nk～N(0，Nk) Wherein the covariance matrix NkSetting and reflecting the validity of the information by the user: n is a radical ofkThe larger the confidence in the information is.
Starting with an initial gaussian confidence in the state,
wherein
Representing the initial estimate, the covariance matrix P0With the uncertainty associated with it, the EKF alternates between the two steps. In the prediction step, the value is estimated
In case of noise off, i.e.:
wherein FkIs a state transition matrix, Fk TIs a transpose of the state transition matrix. GkIs f (-) with respect to xkAnd ukJacobian matrix of GT kIs GkThe transposing of (1). In the updating step, taking into account the pseudo-measurements, the Kalman equation gives the estimate updated accordingly
And its covariance matrix Pk+1. To implement EKF, the designer needs to determine the functions f (-) and h (-) and the associated process noise matrix QkAnd Nk. The invention adopts a neural network to carry out noise parameter QkAnd NkAnd performing integral learning.
Step four: a kinetic model is defined.
Since the Inertial Measurement Unit (IMU) is also rigidly attached to the car,
and
representing the misalignment between the IMU and the frame, so that they are approximately constant
The velocity of the origin of the car, expressed in the reference coordinate system, is:
From the basic helix theory, where Pn is the vehicle to IMU horizontal arm. The lateral and vertical velocities of the car are substantially zero, that is, two scalar false observations are generated:
wherein the noise is
Being a central Gaussian noise, covariance matrix Nk∈R2×2Is gaussian noise. Then inputting the pseudo-measured value to the filter
Step five: establishing a noise parameter adapter:
the measurement noise parameter adapter calculates the covariance N used in the filter update at each time kk+1The basic core of the adapter is the recurrent neural network. The network takes a window of N inertial measurements as input and performs the calculations.
The complete structure of the adapter consists of a set of RNN layers, followed by an output vector
The complete layer of (2). Covariance
The calculation is as follows:
the introduction of the hyper-parameter β serves to control the upper and lower bounds of the covariance expansion, so that the dynamic range of the covariance is 10-β-10β. When beta epsilon R is more than 0, and sigma in the formulalatAnd σupCorresponding to an initial guess of the noise parameters. Thus, the network may expand the covariance by a factor of 10βAnd compresses it to a factor of 10 with respect to its original value-β。
For process noise parameter QnIt was fixed to Q.
The goal is to optimize the filter estimation
Calculated relative translational error trel. It is the average incremental error of all possible subsequences of length 100m, …,800 m.
Then, a learning rate of 10 is selected-4The Adam optimizer of (1) to update the parameters. Training is to repeat the following iterations for the truncated time period:
iteration step one: selecting data (sampling a portion of a data set);
and (5) iteration step two: the error e between the output of the filter and the real trajectory is calculated, after which the parameters of the filter are updated according to a formula, including covariance matrices, etc. For the measurement covariance R in the measurement equation, the parameter in the filter is generally set by a person, and is considered as a variable here. Then the error is calculated according to the chain ruleεTransmitting the data back to the network for weight adjustment;
and a third iteration step: network parameters are updated with gradients and optimizers.
The number of epochs is very large, even infinite, where 600 was chosen. Dropout and data augmentation are used in training to avoid the occurrence of overfitting. Dropout refers to the probability of randomly dropping a network element during the training process, setting any RNN network elements during sequence iterationp is set to zero 0.5. 9 1-minute sequences were sampled as one batch, where each sequence started randomly from any time. Adding a standard deviation of 10 to the data-4Small gaussian noise, i.e. data expansion techniques. Step two is iterated using standard back-propagation computation, and finally the gradient norm is reduced to a maximum value of 1 to avoid gradient explosion in step iteration three.
Step six: testing and evaluating the network model:
the network was tested on a KITTI (KITTI data set, created by the German Carlsuhe institute of technology and Toyota American institute of technology, is currently the computer vision algorithm evaluation data set in the internationally largest autonomous driving scenario) data set: firstly, introducing evaluation indexes:
mean translation error (t)rel): the average relative translational incremental error for all possible subsequences of length 100m, …, …,800m is expressed as a percentage of the distance traveled.
Two sequences are selected as a test set, and compared with a real value obtained by a laser radar, the average translation error is 10m, but the traditional extended Kalman filtering cannot achieve the precision.
Fig. 4 is a schematic diagram of a vehicular navigation apparatus according to an embodiment of the present invention, and as shown in fig. 4, according to another aspect of the embodiment of the present invention, there is also provided a vehicular navigation apparatus including: an acquisition module 42, a determination module 44, an update module 46, and a navigation module 48, which are described in detail below.
An acquisition module 42 for acquiring inertial navigation data of a vehicle movement path; a determining module 44, connected to the obtaining module 42, for determining the covariance used in the filter update at each moment through the noise parameter adaptor according to the inertial navigation data; an updating module 46, connected to the determining module 44, for updating an output state parameter determined by the filter according to the inertial navigation data according to the covariance, where the output state parameter includes a deviation estimation value of the inertial navigation data and a covariance corresponding to the deviation estimation value; and a navigation module 48, connected to the update module 46, for navigating the vehicle according to the updated output state parameters.
By the device, inertial navigation data of the vehicle motion path is acquired by the acquisition module 42; the determination module 44 determines the covariance used in the filter update at each moment in time by the noise parameter adapter based on the inertial navigation data; the updating module 46 updates the output state parameters determined by the filter according to the inertial navigation data according to the covariance, wherein the output state parameters include the deviation estimation value of the inertial navigation data and the covariance corresponding to the deviation estimation value; the navigation module 48 performs a vehicle navigation mode according to the updated output state parameter, so as to achieve the purpose of accurately navigating the vehicle according to the updated output state parameter, thereby achieving the technical effect of improving the accuracy of vehicle navigation, and further solving the technical problem of poor precision of an inertial navigation positioning mode of the vehicle in the related art.
According to another aspect of the embodiments of the present invention, there is also provided a processor for executing a program, wherein the program executes to perform the vehicle navigation method according to any one of the above.
According to another aspect of the embodiments of the present invention, there is also provided a computer storage medium including a stored program, wherein when the program runs, an apparatus in which the computer storage medium is located is controlled to execute the vehicle navigation method according to any one of the above.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
In the above embodiments of the present invention, the descriptions of the respective embodiments have respective emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments.
In the embodiments provided in the present application, it should be understood that the disclosed technology can be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units may be a logical division, and in actual implementation, there may be another division, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, units or modules, and may be in an electrical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic or optical disk, and other various media capable of storing program codes.
The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.
Claims (10)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110255357.4A CN113029173A (en) | 2021-03-09 | 2021-03-09 | Vehicle navigation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110255357.4A CN113029173A (en) | 2021-03-09 | 2021-03-09 | Vehicle navigation method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113029173A true CN113029173A (en) | 2021-06-25 |
Family
ID=76467301
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110255357.4A Pending CN113029173A (en) | 2021-03-09 | 2021-03-09 | Vehicle navigation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113029173A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113587916A (en) * | 2021-07-27 | 2021-11-02 | 北京信息科技大学 | Real-time sparse visual odometer, navigation method and system |
| CN114018250A (en) * | 2021-10-18 | 2022-02-08 | 杭州鸿泉物联网技术股份有限公司 | Inertial navigation method, electronic device, storage medium, and computer program product |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8417490B1 (en) * | 2009-05-11 | 2013-04-09 | Eagle Harbor Holdings, Llc | System and method for the configuration of an automotive vehicle with modeled sensors |
| CN109029435A (en) * | 2018-06-22 | 2018-12-18 | 常州大学 | Improve inertia-earth magnetism combination dynamic accuracy of attitude determination method |
| CN109059912A (en) * | 2018-07-31 | 2018-12-21 | 太原理工大学 | A kind of GPS/INS integrated positioning method based on wavelet neural network |
| CN109781099A (en) * | 2019-03-08 | 2019-05-21 | 兰州交通大学 | A Navigation Method and System for Adaptive UKF Algorithm |
-
2021
- 2021-03-09 CN CN202110255357.4A patent/CN113029173A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8417490B1 (en) * | 2009-05-11 | 2013-04-09 | Eagle Harbor Holdings, Llc | System and method for the configuration of an automotive vehicle with modeled sensors |
| CN109029435A (en) * | 2018-06-22 | 2018-12-18 | 常州大学 | Improve inertia-earth magnetism combination dynamic accuracy of attitude determination method |
| CN109059912A (en) * | 2018-07-31 | 2018-12-21 | 太原理工大学 | A kind of GPS/INS integrated positioning method based on wavelet neural network |
| CN109781099A (en) * | 2019-03-08 | 2019-05-21 | 兰州交通大学 | A Navigation Method and System for Adaptive UKF Algorithm |
Non-Patent Citations (1)
| Title |
|---|
| MARTIN BROSSARD 等，: ""AI-IMU Dead-Reckoning"", 《IEEE TRANSACTIONS ON INTELLIGENT VEHICLES》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113587916A (en) * | 2021-07-27 | 2021-11-02 | 北京信息科技大学 | Real-time sparse visual odometer, navigation method and system |
| CN113587916B (en) * | 2021-07-27 | 2023-10-03 | 北京信息科技大学 | Real-time sparse vision odometer, navigation method and system |
| CN114018250A (en) * | 2021-10-18 | 2022-02-08 | 杭州鸿泉物联网技术股份有限公司 | Inertial navigation method, electronic device, storage medium, and computer program product |
| CN114018250B (en) * | 2021-10-18 | 2024-05-03 | 杭州鸿泉物联网技术股份有限公司 | Inertial navigation method, electronic device, storage medium and computer program product |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Song et al. | Neural-network-based AUV navigation for fast-changing environments | |
| CN107289933B (en) | Double Kalman filter navigation device and method based on MEMS sensor and VLC positioning fusion | |
| Aggarwal et al. | MEMS-based integrated navigation | |
| CN100462682C (en) | A self-calibration method for spacecraft based on predictive filtering and UPF | |
| CN109991636A (en) | Map constructing method and system based on GPS, IMU and binocular vision | |
| CN109000642A (en) | A kind of improved strong tracking volume Kalman filtering Combinated navigation method | |
| CN102096086B (en) | Self-adaptive filtering method based on different measuring characteristics of GPS (Global Positioning System)/INS (Inertial Navigation System) integrated navigation system | |
| EP2856273B1 (en) | Pose estimation | |
| CN113074739A (en) | UWB/INS fusion positioning method based on dynamic robust volume Kalman | |
| CN109682375A (en) | A kind of UWB supplementary inertial localization method based on fault-tolerant decision tree | |
| CN105571578B (en) | A North-finding Method Using In-situ Rotation Modulation Using Pseudo-Observation Instead of Precision Turntable | |
| CN109931955A (en) | An Initial Alignment Method for Strapdown Inertial Navigation System Based on State Dependent Lie Group Filtering | |
| CN107883965A (en) | Based on optical information Interactive Multiple-Model strong tracking volume Kalman filtering air navigation aid | |
| Qin et al. | Accuracy improvement of GPS/MEMS-INS integrated navigation system during GPS signal outage for land vehicle navigation | |
| CN108387236B (en) | A Polarized Light SLAM Method Based on Extended Kalman Filtering | |
| CN112967392A (en) | Large-scale park mapping and positioning method based on multi-sensor contact | |
| CN116295511B (en) | Robust initial alignment method and system for pipeline submerged robot | |
| CN107576327A (en) | Varistructure integrated navigation system design method based on Observable degree analysis of Beidou double | |
| CN111750854A (en) | Vehicle positioning method, device, system and storage medium | |
| CN110375731A (en) | A kind of mixing interacting multiple model filters method | |
| CN116224407B (en) | A kind of GNSS and INS integrated navigation positioning method and system | |
| CN113029173A (en) | Vehicle navigation method and device | |
| Wang et al. | A robust backtracking CKF based on Krein space theory for in-motion alignment process | |
| Zhang et al. | An integrated navigation method for small-sized AUV in shallow-sea applications | |
| CN114018250B (en) | Inertial navigation method, electronic device, storage medium and computer program product |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210625 |