Method and equipment for improving vehicle-mounted satellite positioning precision by using six-axis inertial sensor

Abstract

The invention discloses a method and equipment for improving vehicle-mounted satellite positioning precision by using a six-axis inertial sensor. The method comprises the following steps: converting a satellite positioning problem into a position correction problem of a pseudo-range residual error and a sight distance vector; converting a position correction problem from an earth-centered earth-fixed coordinate system to a north-east earth coordinate system; extracting features of the satellite data; extracting features of the inertial data; 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; outputting a position offset correction value, a displacement vector from the previous satellite navigation moment to the current satellite navigation moment, an attitude angle variable quantity and a speed variable quantity by adopting four prediction heads of the backbone network; selecting a loss function to learn a position offset correction value, a displacement vector, an attitude angle variation and a speed variation; and converting the displacement vector, the attitude angle variable quantity and the speed variable quantity into a position, a speed and an angle. The method can adapt to more complex scenes, and the satellite positioning precision is higher.

Description

Technical Field

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

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

[0003] At present, satellite navigation positioning solutions can be divided into traditional methods or deep learning methods. The most important traditional method is the combined Kalman filter method, which makes certain assumptions about data and state, and then derives prediction equations and update equations, integrating data from multiple sensor devices to achieve positioning. The deep learning method uses the collected satellite data to predict multipath or non-line-of-sight for classification, and eliminates problematic satellite data to optimize the positioning results.

[0004] Traditional methods make assumptions and simplify the model when building it. The model established is often suboptimal and requires manual parameter tuning. For example, the Kalman filter method requires the design of a noise covariance matrix. and the covariance matrix of the measurements And the covariance matrix of the initial state ,The tuning process is extremely cumbersome and time-consuming.,The deep learning method uses the satellite data collected at that moment to determine whether there is a problem with the satellite data, but does not take into account the local information of the carrier's own movement. Summary of the invention

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

[0006] In order to solve the above problems, the present invention is implemented according to the following technical solutions: A method for improving vehicle-mounted satellite positioning accuracy, characterized in that it comprises the steps of: Convert the satellite positioning problem into the position correction problem of pseudo-range residual and line-of-sight vector; Convert the position correction problem from the Earth-centered Earth-fixed coordinate system to the North-East Earth coordinate system; Extract features from satellite data; Extract features from inertial data; Connect the features of satellite data and inertial data to obtain fused features, and use the fused features as the backbone network; The four prediction heads of the backbone network are used to output the position offset correction value, the displacement vector from the previous satellite guidance moment to the current satellite guidance moment, the attitude angle change, and the speed change; Select the loss function to learn the position offset correction value, displacement vector, attitude angle change, and speed change; Convert displacement vector, attitude angle change, and velocity change into position, velocity, and angle.

[0007] As a further improvement of the present invention, the step of extracting the features of satellite data includes: Fill the satellite data with 0 to make it the same length; Map satellite data to a higher dimensional space; The aggregation method is used to aggregate the features of the satellite data to obtain the features of the satellite data.

[0008] As a further improvement of the present invention, the step of extracting the features of the inertial data includes: Perform equal time interval interpolation processing on the collected inertial device data set; The inertial device data set processed by equal time interval interpolation is taken as input, the angle, velocity, position information and time interval of the previous moment are taken as the initial state, and the long short-term memory artificial neural network is used to extract the features of the inertial data.

[0009] As a further improvement of the present invention, in the step of selecting a loss function to learn the position offset correction value, displacement vector, attitude angle change, and speed change, the position offset correction value, displacement vector, and speed change are learned using the minimum mean square error, and the attitude angle change is learned using the cosine loss function.

[0010] As a further improvement of the present invention, the step of converting the satellite positioning problem into a position correction problem of pseudorange residuals and line-of-sight vectors includes: A collection of pseudorange residuals and line-of-sight vectors is constructed using pseudorange residuals and line-of-sight vectors from a vehicle position to a satellite, wherein the vehicle position is a position obtained by a traditional satellite positioning method, and the pseudorange residual is a distance calculated from the actual position to the pseudorange of the mth satellite.

[0011] The present invention 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 communicate with each other via the communication bus; The memory is used to store computer programs; The processor is used to implement the above method when executing the program stored in the memory.

[0012] The present invention also provides one or more computer-readable media having instructions stored thereon, which, when executed by one or more processors, enable the processors to perform the above method.

[0013] Compared with the prior art, the present invention has the following beneficial effects: a deep learning method is used to construct a backbone network for satellite positioning, which does not require assumptions and simplifications of the model, does not require time-consuming and labor-intensive parameter adjustment, and can adapt to more complex scenarios; a six-axis sensor is used to obtain the carrier's own motion information, namely local positioning information, and the carrier's own motion information is involved in the satellite navigation positioning calculation, assisting the satellite navigation to improve positioning, thereby making the satellite positioning more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The specific embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings, wherein: Figure 1 This is a flow chart of a method for improving vehicle-mounted satellite positioning accuracy using a six-axis inertial sensor according to the present invention; Figure 2 It is a partial flow chart of a method for improving vehicle-mounted satellite positioning accuracy by using a six-axis inertial sensor according to the present invention; Figure 3 Another partial flow chart of a method for improving vehicle-mounted satellite positioning accuracy using a six-axis inertial sensor described in the present invention. DETAILED DESCRIPTION

[0015] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0016] The inventive concept of the present invention is: based on the observation data of the satellite and the observation data of its own inertial navigation device, the two types of measurement data are merged by a deep learning method to reduce the impact of non-line-of-sight, i.e. NLOS or multi-path satellite data and utilize the high short-term accuracy of inertial to improve the performance of satellite positioning. The satellite navigation positioning problem is converted into a problem of position correction using residuals and line-of-sight vectors. According to the data characteristics of satellite navigation, embeddings and aggregation are designed. Inertial navigation data has time series characteristics, and long short-term memory artificial neural networks are used for embeddings. The embeddings of data from two different sources are fused, and four prediction heads are used to learn the position correction amount, the displacement offset before and after, the angle change amount and the speed change amount. A mechanism for the fusion of multiple sensors is provided and it is easy to fuse new sensors.

[0017] Based on the above invention concept, the present invention provides a method for improving the vehicle-mounted satellite positioning accuracy by using a six-axis inertial sensor, such as Figures 1 to 3 As shown, the steps include: S1. Convert the satellite positioning problem into the position correction problem of pseudo-range residual and line-of-sight vector.

[0018] Considering that the pseudo-moment is usually in the millions, it is easy to cause the model to not converge. Therefore, the satellite positioning problem is converted into a pseudo-moment residual, and its converted value can be much smaller, about a hundred levels.

[0019] Specifically, in the Earth-centered low-solid ECEF coordinate system, the vehicle The position at time is , the pseudo-moment measurement of the satellite observed at this time and position is , is the number of observed satellites, is the position of the corresponding satellite, the pseudo-moment and satellite position set are set to . Due to the real The location is unknown. Assume that the traditional satellite positioning method is used to obtain the location . Let the position estimated by deep learning be ,and The offset is . Actual vehicle position and The offset is And then: = -

[0020] = -

[0021] Since the values ​​of pseudo-moments are large, subsequent training is difficult to converge and needs to be converted. The pseudo-moment calculation formula is as follows:

[0022] in Indicates the various errors produced during the measurement process. Much smaller than ,Will See For a small perturbation of the position, (4-3) is linearized to:

[0023]

[0024] but:

[0025]

[0026] in The actual position is measured to The pseudo moments and The calculated distance difference is the pseudo-moment residual. For The line of sight vector looking toward the satellite is LOS. The position offset is determined by the pseudo-moment residual and the line of sight vector. The set of pseudo-moment residual and line of sight vector is set as:

[0027] It can also be equivalently expressed as , , .

[0028] S2. Convert the position correction problem from the Earth-centered Earth-fixed coordinate system to the North-East Earth coordinate system.

[0029] Due to the geographical location of the data collection, the use of the Earth-centered Earth-fixed coordinate system will result in poor generalization of the learning results. Converting to the North-East Earth coordinate system will help improve the learning effect. The conversion method is as follows:

[0030]

[0031] The reason for converting the ecef coordinate system to the ned coordinate system is that the collected data is limited to a certain city. The ecef coordinate system is a global coordinate system related to the earth. The model trained with the data collected in Guangzhou is not suitable for the northeast. The ned coordinate system is a local coordinate system, which has little to do with the sphere, so it is more generalizable. Therefore, the purpose of steps S1 and S2 is to train a better model.

[0032] S3. Extract features of satellite data: S31. Since the number of satellites observed at different times or locations is different, when using batch sample training, it is necessary to fill in 0 to make it the same length.

[0033] S32. Map satellite data to a higher dimensional space.

[0034] S33. Since the order and number of satellites observed at different times are changing, an aggregation method can be used to aggregate the features of the satellite data to obtain the features of the satellite data.

[0035] S4. Extract the features of inertial data. Assume that the coordinate system of the inertial device is consistent with the coordinate system of the vehicle, the arm of the inertial device and the satellite antenna is 0, and the zero bias of the inertial device has been corrected. Assume that the position of the previous satellite navigation moment is , the speed is , the pitch angle is , the roll angle is , the heading angle is , The inertial device data set collected during the two satellite navigation periods is:

[0036] in Indicated in The three-axis accelerometer data collected at all times, Indicated in The data of the three-axis gyroscope collected at all times. Indicates the amount of inertial device data collected during the time interval between two satellite navigation systems.

[0037] The specific steps are as follows: S41. Assume that the inertial data collected between the two satellites is a fixed value. ,right The result of data interpolation at equal time intervals is recorded as The time interval is recorded as ; S42. The inertial data collected by the inertial device is time series information. The previous and next moments are interrelated. The inertial device data set after equal time interval interpolation is used as input, and the angle, speed, position information and time interval of the previous moment are used as the initial state. The long short-term memory artificial neural network is used to extract the characteristics of the inertial data.

[0038] S5. Connect the features of the satellite data obtained in step S3 and the features of the inertial data obtained in step S4, use a fully connected layer to achieve feature fusion to obtain fused features, and use the fused features as the backbone network.

[0039] S6, using the fused features as input, the four prediction heads of the backbone network output the position offset correction value , the displacement vector from the previous satellite guidance time to the current satellite guidance time , attitude angle change , speed change .

[0040] S7. Select a loss function to learn the position offset correction value, displacement vector, attitude angle change, and speed change.

[0041] Specifically, the position offset correction value, displacement vector, and speed change are learned using the minimum mean square error, and the specific calculation is as follows:

[0042]

[0043]

[0044] in is the number of samples in the subset, For the The label value of the offset correction value of the sample is = . For the The displacement vector label value of each sample is = . For the The speed change label value of each sample = . , , is the output value of the model.

[0045] Considering that the angle is For periodicity, the attitude angle change is learned using the cosine loss function, using the following loss function:

[0046] in Indicates the first The label value of the angle change of each sample is = . = is the angle change value predicted by the model.

[0047] S8. Convert the displacement vector, attitude angle change, and velocity change into position, velocity, and angle.

[0048]

[0049] = +

[0050]

[0051] = +

[0052] = +

[0053] The predicted position and velocity are converted to Earth-centered Earth-fixed coordinates, and the angles are expressed in Euler angles relative to the north-east of the Earth in vehicle coordinates.

[0054] In summary, the present invention has the following technical effects: 1. Different feature extraction methods are used for satellite navigation data and inertial navigation data to solve the defects of data differences, and then solve the problem of data differences that lead to non-convergence of model training; 2. Multi-head prediction is used to achieve learning of different physical quantities. The trained model is highly robust and provides more comprehensive positioning information compared to satellite navigation positioning. 3. The cosine-based loss function is used to solve the problem of non-convergence of model training caused by angle periodicity; 4. An end-to-end training framework is implemented, which can better model the entire positioning system.

[0055] Compared with the Kalman method in the prior art, the present invention has fewer assumptions and derivation processes, and can build a more accurate and complex model; compared with the prior art method of using deep learning to eliminate non-line-of-sight satellites or multi-path methods, the present invention combines the global information of satellite navigation and the local information of inertial navigation, which can better eliminate abnormal satellite data to achieve higher positioning accuracy and provide attitude and speed information.

[0056] Based on the same inventive concept, another embodiment of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to load the method of the present invention as described above.

[0057] Specifically, the computer device includes: a memory and a processor. The memory and the processor are connected via a bus communication. A computer program is stored in the memory. The computer program can be run on the processor to load the liquidation method of the present invention.

[0058] The memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0059] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0060] Based on the same inventive concept, another embodiment of the present invention provides a computer-readable storage medium on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the method described above in the present invention is loaded.

[0061] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0062] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, devices, electronic devices, storage media or computer program products. Therefore, embodiments of the present invention may take the form of complete hardware embodiments, complete software embodiments, or embodiments combining software and hardware. Moreover, embodiments of the present invention may take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CDROM, optical storage, etc.) containing computer-usable program codes.

[0063] The above are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should 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: Includes steps: Convert the satellite positioning problem into the position correction problem of pseudo-range residual and line-of-sight vector; Convert the position correction problem from the Earth-centered Earth-fixed coordinate system to the North-East Earth coordinate system; Extract features from satellite data; Extract features from inertial data; Connect the features of satellite data and inertial data to obtain fused features, and use the fused features as the backbone network; The four prediction heads of the backbone network are used to output the position offset correction value, the displacement vector from the previous satellite guidance moment to the current satellite guidance moment, the attitude angle change, and the speed change; Select the loss function to learn the position offset correction value, displacement vector, attitude angle change, and speed change; Convert displacement vector, attitude angle change, and velocity change into position, velocity, and angle.

2. The method according to claim 1, characterized in that: The step of extracting features of satellite data comprises: Fill the satellite data with 0 to make it the same length; Map satellite data to a higher dimensional space; The aggregation method is used to aggregate the features of the satellite data to obtain the features of the satellite data.

3. The method according to claim 1, characterized in that The step of extracting the features of the inertial data comprises: Perform equal time interval interpolation processing on the collected inertial device data set; The inertial device data set processed by equal time interval interpolation is taken as input, the angle, velocity, position information and time interval of the previous moment are taken as the initial state, and the long short-term memory artificial neural network is used to extract the features of the inertial data.

4. 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 speed change, the position offset correction value, displacement vector, and speed change are learned using the minimum mean square error, and the attitude angle change is learned using the cosine loss function.

5. The method according to claim 1, characterized in that The step of converting the satellite positioning problem into a position correction problem of pseudorange residual and line-of-sight vector comprises: A collection of pseudorange residuals and line-of-sight vectors is constructed using pseudorange residuals and line-of-sight vectors from a vehicle position to a satellite, wherein the vehicle position is a position obtained by a traditional satellite positioning method, and the pseudorange residual is a distance calculated from the actual position to the pseudorange of the mth satellite.

6. 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; The processor is used to implement the method according to any one of claims 1 to 5 when executing the program stored in the memory.

7. One or more computer-readable media having instructions stored thereon, which, when executed by one or more processors, cause the processors to perform the method according to any one of claims 1 to 5.

Images