Method and device for improving vehicle satellite positioning accuracy using six-axis inertial sensor

Abstract

The application discloses a method and equipment for improving vehicle-mounted satellite positioning precision by using a six-axis inertial sensor, and the method comprises the following steps: converting a satellite positioning problem into a position correction problem of pseudo-range residual error and line-of-sight vector; converting the position correction problem from a geocentric geodetic coordinate system into a north-east-ground coordinate system; extracting features of satellite data; extracting features of inertial data; connecting the features of satellite data and the features of inertial data to obtain fusion features, and taking the fusion features as a backbone network; adopting four prediction heads of the backbone network to output a position offset correction value, a displacement vector from a previous satellite guidance moment to a current satellite guidance moment, an attitude angle change amount and a speed change amount; selecting a loss function to learn the position offset correction value, the displacement vector, the attitude angle change amount and the speed change amount; and converting the displacement vector, the attitude angle change amount and the speed change amount into position, speed and angle. The application can adapt to more complex scenes, and the precision of satellite positioning is higher.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault time prediction, and in particular to a method and device for improving vehicle satellite positioning accuracy using a six-axis inertial sensor. BACKGROUND

[0002] Satellite positioning technology is a high-tech technology with deep technical background and wide application prospect. Satellite positioning technology has been widely used in military, civilian and commercial fields. In the military field, it can provide real-time, all-weather and global navigation services for the army, navy and air force; in the civilian field, it can provide accurate position information for car navigation, automatic driving, atmospheric observation, geographic survey, marine rescue, etc. However, satellite positioning is susceptible to multipath and non-line-of-sight effects, resulting in a decrease in positioning accuracy. The six-axis inertial sensor is a sensor that can measure the force in three orthogonal directions and the moment in three orthogonal directions. It is a kind of sensor that uses the principle of inertia to measure and record the motion state of an object, and can obtain its acceleration and angular velocity. Deep learning has the ability to model complex models and learn more representative features, and has achieved great success in vision and natural language processing scenarios. Using deep learning methods, using six-axis sensor data and satellite navigation data to improve satellite navigation positioning is of great significance.

[0003] Currently, satellite positioning schemes can be divided into traditional methods or deep learning methods. The most important one in the traditional method is the combined Kalman filter method. This method makes certain assumptions about the data and state, and then derives the prediction equation and update equation to fuse multiple sensor device data and achieve positioning. The deep learning method uses collected satellite data to predict multipath or non-line-of-sight to classify and exclude problematic satellite data to optimize positioning results.

[0004] When modeling, the traditional method makes assumptions and simplifications about the model. The established model is often suboptimal, and manual parameter tuning is required, such as using the Kalman filter method, which requires designing the noise covariance matrix Q and the measurement covariance matrix R, as well as the initial state covariance matrix P. The tuning process is extremely tedious and time-consuming. The deep learning method uses the satellite data collected at the moment to determine whether the satellite data is problematic, but does not consider the local information of the carrier's own motion. SUMMARY

[0005] In order to overcome the above technical defects, the present application provides a method and device for improving vehicle satellite positioning accuracy using a six-axis inertial sensor, which can improve the accuracy of positioning.

[0006] In order to solve the above problems, the present application is implemented according to the following technical solutions:

[0007] The application discloses a method for improving the positioning accuracy of a vehicle-mounted satellite, and has the characteristics that the method comprises the following steps:

[0008] Converting the satellite positioning problem into a position correction problem of a pseudo-range residual and a line-of-sight vector;

[0009] Converting the position correction problem from a geocentric geodetic coordinate system into a north-east-ground coordinate system;

[0010] Extracting features of satellite data;

[0011] Extracting features of inertial data;

[0012] Connecting the features of the satellite data and the features of the inertial data to obtain fusion features, and taking the fusion features as a backbone network;

[0013] Using four prediction heads of the backbone network to output a position offset correction value, a displacement vector from a previous satellite guidance time to a current satellite guidance time, an attitude angle change amount and a speed change amount;

[0014] Selecting a loss function to learn the position offset correction value, the displacement vector, the attitude angle change amount and the speed change amount;

[0015] Converting the displacement vector, the attitude angle change amount and the speed change amount into a position, a speed and an angle.

[0016] As a further improvement of the application, the step of extracting features of satellite data comprises:

[0017] Performing 0 padding on the satellite data to complete the satellite data to the same length;

[0018] Mapping the satellite data to a higher-dimensional space;

[0019] Using an aggregation method to aggregate the features of the satellite data to obtain the features of the satellite data.

[0020] As a further improvement of the application, the step of extracting features of inertial data comprises:

[0021] Performing equal-time-interval interpolation processing on the collected inertial device data set;

[0022] Taking the inertial device data set after the equal-time-interval interpolation processing as input, taking the angle, the speed, the position information and the time interval at a previous time as initial state, and using a long short-term memory artificial neural network to extract features of the inertial data.

[0023] As a further improvement of the present application, in the step of learning the position offset correction value, the displacement vector, the attitude angle change amount and the speed change amount by the selected loss function, the position offset correction value, the displacement vector and the speed change amount are learned by using the least mean square error, and the attitude angle change amount is learned by using the loss function of cosine.

[0024] As a further improvement of the present application, the step of converting the satellite positioning problem into the position correction problem of the pseudo-range residual and the line-of-sight vector includes:

[0025] The pseudo-range residual and the line-of-sight vector are constructed by using the pseudo-range residual and the line-of-sight vector from the vehicle position to the satellite, wherein the vehicle position is the position obtained by using the traditional satellite positioning method, and the pseudo-range residual is the distance calculated by measuring the pseudo-range of the mth satellite.

[0026] The present application also provides an electronic device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete the communication among each other through the communication bus;

[0027] The memory is used for storing a computer program.

[0028] The processor is used for executing the program stored on the memory to realize the above-mentioned method.

[0029] The present application also provides one or more computer readable media having instructions stored thereon, which, when executed by one or more processors, cause the processors to perform the above-mentioned method.

[0030] Compared with the prior art, the present application has the following beneficial effects: a deep learning method is used to construct a backbone network for satellite positioning, which does not need to make assumptions and simplifications on the model, does not need time-consuming and laborious parameter tuning, and can adapt to more complex scenarios; the motion information of the carrier itself, i.e., the local positioning information, is obtained by using a six-axis sensor, and the motion information of the carrier itself is involved in the satellite positioning calculation to assist the satellite positioning to improve the positioning, so that the accuracy of satellite positioning is higher. BRIEF DESCRIPTION OF DRAWINGS

[0031] The specific embodiments of the present application will be further described in detail below with reference to the accompanying drawings, in which:

[0032] Fig. 1 A flowchart of the method for improving the accuracy of vehicle-mounted satellite positioning by using a six-axis inertial sensor according to the present application;

[0033] Fig. 2 A partial flowchart of the method for improving the accuracy of vehicle-mounted satellite positioning by using a six-axis inertial sensor according to the present application;

[0034] Fig. 3 Another partial flowchart of a method for improving vehicle-mounted satellite positioning accuracy using a six-axis inertial sensor, as described in this invention. Detailed Implementation

[0035] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0036] The inventive concept of this invention is as follows: Based on satellite observation data and inertial navigation device (INS) observation data, a deep learning method is used to fuse the two types of measurement data, reducing the influence of non-line-of-sight (NLOS) or multipath satellite data and utilizing the high short-time accuracy of inertial navigation to improve satellite positioning performance. The satellite navigation positioning problem is transformed into a position correction problem using residuals and line-of-sight vectors. Based on the characteristics of satellite navigation data, an embedding method and aggregation are designed. Since INS data has temporal characteristics, a long short-term memory (LSTM) artificial neural network is used for embedding. The embeddings from the two different data sources are fused, and four prediction heads are used to learn the position correction, displacement offset, angle change, and velocity change at different times. This provides a mechanism for multi-sensor fusion and facilitates the fusion of new sensors.

[0037] Based on the above-mentioned inventive concept, this invention provides a method for improving the accuracy of vehicle-mounted satellite positioning using a six-axis inertial sensor, such as... Figs. 1 to 3 As shown, the steps include:

[0038] S1. The satellite positioning problem is transformed into a position correction problem using pseudorange residuals and line-of-sight vectors.

[0039] Considering that pseudo-moments are generally in the millions, which can easily lead to model non-convergence, the satellite positioning problem is converted into pseudo-moment residuals, which can reduce the value to a much smaller level, around the hundreds.

[0040] Specifically, in the geocentric low-fixed ECEF coordinate system, the vehicle's position at time i is: The pseudo-moment measurement of the satellite observed at that time and location is M i The number of observed satellites, To represent the position of the corresponding satellite, the pseudo-moment and the set of satellite positions are set as follows: Due to the real vehicles The location is unknown. Let's assume the location is obtained using traditional satellite positioning methods. Let the position estimated by deep learning be... and The offset is Real vehicle location and The offset is Then there are:

[0041]

[0042] Since the values ​​of the pseudomoments are very large, subsequent training is difficult to converge, so they need to be transformed. The formula for calculating pseudomoments is as follows:

[0043]

[0044] in This represents the various errors that occur during the measurement process. Due to... Much smaller than Will See as For small perturbations in position, linearizing (4-3) yields:

[0045]

[0046] but:

[0047]

[0048] in The pseudo-moment and the actual position of the m-th satellite are measured. The calculated distance difference is the pseudo-moment residual. From The line-of-sight (LOS) vector pointing towards the satellite is determined by the position offset pseudo-moment residual and the LOS vector. The set of pseudo-moment residuals and the LOS vector is defined as follows:

[0049]

[0050] It can also be equivalently represented as M i =(R i ,I i ),

[0051] S2. The position correction problem is transformed from the geocentric coordinate system to the northeast coordinate system.

[0052] Due to the geographical limitations of the collected data, using a geocentric coordinate system results in poor generalization of the learned data. Converting to a northeast-northeast coordinate system improves the learning effect. The conversion method is as follows:

[0053]

[0054] The reason for converting the ecef coordinate system to the ned coordinate system is that the collected data is limited to a certain city, and using the ecef coordinate system is a global coordinate related to the earth as a sphere. The model trained by the data collected in Guangzhou is not suitable for northeast China, and the ned coordinate system is a local coordinate and has little to do with the sphere, so it has stronger generalization. Therefore, the purposes of steps S1 and S2 are to train a better model.

[0055] S3, extract features of satellite data:

[0056] S31, since the number of satellites observed at different times or different locations is different, when batch samples are used for training, 0 needs to be filled to make up the same length.

[0057] S32, map satellite data to a higher dimensional space.

[0058] S33, since the order of satellites observed at different times is changing and the number is also changing, an aggregation method can be used to aggregate the features of satellite data to obtain the features of satellite data.

[0059] S4, extract features of inertial data, assuming that the coordinate system of the inertial device and the coordinate system of the vehicle remain the same, the arm of the inertial device and the satellite antenna is 0, and the zero offset of the inertial device has been corrected. Let the position of the previous satellite guidance time be The velocity is The pitch angle is θ i-1 , the roll angle is γ i-1 , and the heading angle is The collected inertial device data set between two satellite guidance times is:

[0060]

[0061] Where a tn represents the three-axis accelerometer data collected at time tn, g tn represents the three-axis gyroscope data collected at time tn. N i represents the number of inertial device data collected between two satellite guidance times.

[0062] The specific steps are as follows:

[0063] S41, let the inertial data collected between two satellite guidance times be a fixed value N, and the equal-time interval interpolation result of Imu i data is recorded as The time interval is recorded as Δt i ;

[0064] S42, the inertial device collects inertia data, which is time sequence information, the time instants before and after are related to each other, the inertia data set after interpolation processing at equal time intervals is taken as input, the angle, speed and position information at the previous time instant and the time interval are taken as initial state, and a long short-term memory artificial neural network is used to extract the characteristics of the inertia data.

[0065] S5, the characteristics of the satellite data obtained in step S3 and the characteristics of the inertia data obtained in step S4 are connected, a full connection layer is used to realize fusion of the characteristics to obtain fusion characteristics, and the fusion characteristics are taken as the backbone network.

[0066] S6, taking the fusion characteristics as input, the four prediction heads of the backbone network are used to output position offset correction values displacement vector from the previous satellite guidance time instant to the current satellite guidance time instant attitude angle change speed change

[0067] S7, selecting a loss function to learn the position offset correction value, the displacement vector, the attitude angle change and the speed change.

[0068] Specifically, the position offset correction value, the displacement vector and the speed change are learned by using the least mean square error, and the specific calculation is as follows:

[0069]

[0070] Where N b is the number of samples in the subset, Δp ned,j is the label value of the offset correction value of the jth sample, i.e. Δps ned,j is the label value of the displacement vector of the jth sample, i.e. Δvs ned,j is the label value of the speed change of the jth sample is the output value of the model.

[0071] Considering that the angle is periodic with a period of 2π, the attitude angle change is learned by using the cosine loss function, and the following loss function is used:

[0072]

[0073] Where ΔΦ j represents the label value of the angle change of the jth sample in the sample set, i.e. j ΔΦ (i-1,i) . is the predicted angle change value of the model.

[0074] S8, converting the displacement vector, the attitude angle change and the speed change into position, speed and angle.

[0075]

[0076] The predicted position and velocity are converted to the geocentric and geodetic coordinates, and the angle adopts the Euler angle of the vehicle coordinate relative to the north east ground.

[0077] In summary, the present application has the following technical effects:

[0078] 1. Different feature extraction methods are adopted for satellite navigation data and inertial navigation data, solving the defect of data difference and thus solving the problem of data difference leading to model training divergence;

[0079] 2. Multi-head prediction is used to realize learning of different physical quantities, and the trained model is robust and provides more comprehensive positioning information relative to satellite navigation positioning;

[0080] 3. A cosine-based loss function is adopted, which can solve the problem of model training divergence caused by the periodicity of angles;

[0081] 4. An end-to-end training framework is realized, which can better model the entire positioning system.

[0082] Compared with the Kalman method of the prior art, the present application has fewer assumptions and derivation processes, and can model a more accurate and complex model; compared with the prior art of using deep learning to remove non-line-of-sight satellites or multi-path methods, the present application combines the global information of satellite navigation and the local information of inertial navigation, and can better remove abnormal satellite data to achieve higher positioning accuracy and provide attitude and speed information.

[0083] Based on the same inventive concept, another embodiment of the present application provides a computer device comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to load the method as described above.

[0084] Specifically, the computer device comprises a memory and a processor, the memory and the processor are communicatively connected through a bus, and the memory stores a computer program which can run on the processor and further load the clearing method in the present application.

[0085] The memory can include a random access memory (RAM) and can also include a non-volatile memory, such as at least one disk memory. Optionally, the memory can also be at least one storage device located away from the aforementioned processor.

[0086] The processor described above can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0087] Based on the same inventive concept, another embodiment of the present application provides a computer readable storage medium, which stores a computer program / instruction, and the computer program / instruction is executed by a processor to load the method described above in the present application.

[0088] Each of the embodiments in the present specification is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same and similar parts of each embodiment can be referred to each other.

[0089] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, device, electronic equipment, storage medium or computer program product. Therefore, the embodiments of the present application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present application can be in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CDROM, optical storage, etc.) containing computer usable program code.

[0090] The above is only the preferred embodiment of the present application, and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for improving vehicle-mounted satellite positioning accuracy using a six-axis inertial sensor, characterized in that, Including the following steps: The satellite positioning problem is transformed into a position correction problem using pseudorange residuals and line-of-sight vectors. The position correction problem is transformed from the geocentric-fixed coordinate system to the northeast geocentric coordinate system; Extracting features from satellite data: pad the satellite data with zeros to make it the same length, map the satellite data to a higher-dimensional space, and use aggregation methods to aggregate the features of the satellite data to obtain the features of the satellite data; Features of inertial data are extracted by interpolating the collected inertial device dataset at equal time intervals. The inertial device dataset after equal time interval interpolation is used as input, and the angle, velocity, position information and time interval of the previous moment are used as the initial state. Long short-term memory artificial neural network is used to extract features of inertial data. The features of satellite data and inertial data are connected to obtain fused features, which are then used as the backbone network. The four prediction heads of the backbone network output the position offset correction value, the displacement vector from the previous satellite navigation time to the current satellite navigation time, the attitude angle change, and the velocity change. The loss function is selected to learn the position offset correction value, displacement vector, attitude angle change, and velocity change; The displacement vector, attitude angle change, and velocity change are converted into position, velocity, and angle.

2. The method according to claim 1, characterized in that, In the step of selecting a loss function to learn the position offset correction value, displacement vector, attitude angle change, and velocity change, the position offset correction value, displacement vector, and velocity change are learned using the minimum mean square error, while the attitude angle change is learned using a cosine loss function.

3. The method according to claim 1, characterized in that, The step of transforming the satellite positioning problem into a position correction problem using pseudorange residuals and line-of-sight vectors includes: A set of pseudorange residuals and line-of-sight vectors is constructed using pseudorange residuals and the line-of-sight vectors from the vehicle's position to the satellite. The vehicle's position is the position obtained using traditional satellite positioning methods, and the pseudorange residual is the distance calculated from the actual position to the m-th satellite using pseudorange.

4. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; The memory is used to store computer programs; When the processor executes a program stored in the memory, it implements the method as described in any one of claims 1-3.

5. One or more computer-readable media having instructions stored thereon that, when executed by one or more processors, cause the processors to perform the method as described in any one of claims 1-3.

Images