GNSS RTK and MEMS IMU fusion positioning method based on vehicle-mounted smart phone terminal

Abstract

A GNSS (Global Navigation Satellite System) RTK (Real Time Kinematic) and MEMS (Micro Electro Mechanical System) IMU (Inertial Measurement Unit) fusion positioning method based on a vehicle-mounted intelligent mobile phone terminal comprises the following steps: acquiring original measurement values of a GNSS and an MEMS IMU, converting the GNSS into an international general format RINEX, downloading nearby public base station satellite data to carry out RTK positioning of a base station and a mobile phone terminal, and carrying out speed smoothing filtering to obtain position calculation; smooth transition when the satellite quality is poor is completed through a GNSS RTK and MEMS IMU loose combination; and finally, completing threshold judgment of parking detection on IMU data through a tree structure Parzen estimation algorithm (TPE), calculating a precision difference with a data truth value, and iteratively obtaining an optimal threshold and a positioning result. The method is simple and accurate in positioning based on the low-precision sensor and the satellite and inertial navigation sensor data of the vehicle-mounted mobile phone by combining the parking characteristics of the vehicle during running.

Description

Technical Field

[0001] The present invention relates to a technology in the field of high-precision positioning, specifically a method for fusion positioning of a global navigation satellite system real-time dynamic (GNSS RTK) and a micro-electromechanical system inertial measurement unit (MEMS IMU) based on a vehicle-mounted smart phone terminal. Background Art

[0002] Accurate smartphone positioning services have made many navigation features used today possible. Since 2016, raw measurements from GNSS chipsets of Android phones have become more publicly accessible through Android APIs, serving as a bridge between advanced positioning technology and positioning applications for billions of Android users in the mass market. High-precision positioning using GNSS and IMU raw data has attracted scholars around the world. However, current mobile phones can only provide positioning accuracy of 3-5 meters, which is not accurate enough and unreliable for vehicle navigation, and lane-level positioning based on mobile phone data has not been achieved. Summary of the invention

[0003] In view of the problems of existing methods for processing vehicle-mounted mobile phone positioning, such as poor real-time performance, high calculation complexity, inability to avoid GNSS RTK erroneous positioning resulting in position flying points and the like, susceptibility to environmental interference, and inability to ensure continuity and completeness of positioning, the present invention proposes a method for fusion positioning of GNSS RTK and MEMS IMU based on a vehicle-mounted smart phone terminal. Combined with the parking characteristics of the vehicle during driving, the method is based on low-precision sensors and satellite and inertial navigation sensor data of the vehicle-mounted mobile phone, and the positioning is simple and accurate.

[0004] The present invention is achieved through the following technical solutions:

[0005] The present invention relates to a method for GNSS RTK and MEMS IMU fusion positioning based on a vehicle-mounted smart phone terminal, comprising:

[0006] S1. First, obtain the raw measurement values ​​of GNSS and MEMS IMU through the smartphone API, convert GNSS into the international common format RINEX, and then download the nearby public base station satellite data;

[0007] S2, complete the RTK positioning of the base station and the mobile phone, and use velocity smoothing filtering to obtain the position solution;

[0008] S3, through the combination of GNSS RTK and MEMS IMU to achieve smooth transition when satellite quality is poor;

[0009] S4. The threshold value of parking detection is determined by the IMU data through the tree structured Parzen estimation algorithm (TPE), and the accuracy difference is calculated with the true value of the data, and the optimal threshold value and positioning result are obtained iteratively.

[0010] The GNSS RTK and MEMS IMU combination specifically includes:

[0011] a) Taking advantage of the fact that vehicles do not suddenly accelerate or decelerate at the same time in a short period of time, the results of satellite positioning that may have large errors are eliminated;

[0012] b) Using high-precision inertial navigation sensor data to obtain the error random model parameters of the mobile phone MEMS during vehicle motion, where the random model parameters are unknown;

[0013] c) Use the satellite solution speed at the initial moment to complete the initial attitude alignment of the MEMS sensor;

[0014] d) Through the method of linear interpolation, the high-frequency MEMS timestamp and the low-frequency satellite timestamp are completely aligned, and the incremental values ​​of the aligned accelerometer and gyroscope are obtained to complete the mechanical arrangement algorithm.

[0015] e) The processed satellite-solved velocity and position information is loosely combined with the MEMS data to obtain the initial values ​​and random model parameters without considering the arm distance.

[0016] The tree structure Parzen estimation algorithm specifically includes:

[0017] i) Setting a wide range of reasonable thresholds for the accelerometer and gyroscope data to determine the parking time, and determining that the time period is the vehicle stopping time period when the gyroscope and accelerometer measurement values ​​are less than the threshold for more than 5 consecutive seconds;

[0018] ii) Replace the location data of the parking interval with the median of the input location, compare the location result obtained by this threshold with the ground truth result, and calculate the score;

[0019] iii) It is beneficial for the TPE method to iteratively explore the search space of these two thresholds and find the optimal parameters that achieve the highest position accuracy under parking detection;

[0020] iv) Save the trained model so that the parameter model can be reused on the test set data without ground track truth.

[0021] The present invention relates to a system for implementing the above method, comprising: a GNSS RTK positioning unit, a speed smoothing filter unit, a loose combination unit and a parking detection unit, wherein: the GNSS RTK positioning unit performs filtering and solving processing according to the GNSS signals of a vehicle-mounted mobile phone and a nearby public base station to obtain the position and speed information of the vehicle; the speed smoothing filter unit performs smoothing filtering on the position result in the form of a kinematic equation after screening and interpolating the obtained vehicle celestial speed to obtain a position solution; the loose combination unit performs loose combination processing on the obtained position solution after removing outliers and the position speed attitude information obtained and processed by a MEMS IMU sensor of the vehicle-mounted mobile phone to obtain a smooth position solution; the parking detection unit determines a threshold value through a TPE method according to the gyroscope and accelerometer information of the MEMS IMU, completes the detection of the vehicle stop period, and obtains a more accurate position result. Technical Effects

[0022] The present invention utilizes satellite data of a ground public base station. When the satellite quality is good, a relative positioning method of pseudorange and carrier phase is adopted between a car as a mobile station and a reference station. When the satellite quality is poor, such as no carrier phase or a fixed ambiguity ratio value less than 3, a pseudorange differential positioning method is adopted. The speed information selected according to the characteristic of a small celestial speed of the car is used to smooth the position results forward and backward. The original data of a mobile phone MEMS IMU with unknown IMU random model parameters is used to evaluate the IMU random model parameters in a vehicle-mounted dynamic scene, and a loose combination solution with satellite positioning results is completed. The TPE optimizer is used to optimize the threshold judgment of parking detection, and no manual input or adjustment of the threshold parameters is required, thereby greatly improving the accuracy of the parking detection period.

[0023] Compared with the traditional RTK method, which will have a huge impact on the positioning result when the ambiguity cannot be fixed or is fixed incorrectly, the present invention objectively selects the GNSS positioning output result according to the satellite quality and solution quality indicators to ensure the reliability and completeness of positioning, which is more conducive to real-time vehicle navigation positioning; both speed smoothing filtering and loose combination adopt outlier judgment processing on the solution results to further ensure the accuracy and reliability of the results; and proposes a parameter estimation method for the MEMS IMU sensor of the vehicle-mounted mobile phone when the parameters are unknown, to ensure that the IMU and GNSS results with unknown parameters can still be correctly combined; compared with the traditional parking detection method, the present invention introduces the tree structure Parzen algorithm, which makes the detection intelligent and de-subjective, better adapts to a variety of data, and improves the positioning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a flow chart of the present invention;

[0025] Figure 2 This is a flow chart of GNSS RTK and velocity smoothing filtering of the present invention;

[0026] Figure 3 This is a schematic diagram of the interpolation situation of the loose combination corresponding to the time of the present invention;

[0027] Figure 4 It is a loose combination flow chart of the present invention;

[0028] Figure 5 This is a schematic diagram of the parking detection process completed by the TPE method of the present invention;

[0029] Figure 6-Figure 8 It is a schematic diagram of the effect of the embodiment. DETAILED DESCRIPTION

[0030] like Figure 1 As shown, this embodiment relates to a method for GNSS RTK and MEMS IMU fusion positioning based on a vehicle-mounted smartphone terminal, including:

[0031] The first step, such as Figure 2 As shown, according to the pseudorange, carrier phase and base station position, the position and covariance are obtained by positioning through GNSS RTK, specifically: the original satellite and inertial device data measurement values ​​are obtained through the API of the vehicle-mounted smart phone, and the satellite measurement values ​​are converted into RINEX format, and then the public base station information near the road where the car is traveling is found, and the satellite observation files and navigation messages of the corresponding time are downloaded; when the satellite signal quality is poor or the satellite status of the current epoch does not meet the requirements, pseudorange differential positioning is used instead, and the position and speed information of each moment in the driving process of the car is obtained by solving the present invention.

[0032] The present invention specifically comprises:

[0033] a) When the reference station B and the mobile station R observe satellite i and satellite j at the same time, the pseudorange and carrier phase observation equations of the reference station are specifically:

[0034] b) For the base station, the pseudorange and carrier phase observation results are similar, specifically: Where: subscript f is a specific carrier band, subscript B represents a base station, subscript R represents a mobile station, and superscript i represents a base station. G represents GNSS satellite i, ρ is the geometric distance between the satellite and the receiver, dt is the clock difference, I is the ionospheric delay, T is the tropospheric delay, λ is the wavelength of the f band, dρ is the orbit error, and v and ε are other errors.

[0035] c) The double difference combination of short baseline observations is: in: is the double difference operator, The double difference between stations and satellites can eliminate the errors related to the receiver (satellite clock error, satellite hardware delay error, receiver clock error and receiver hardware delay). The ionospheric delay error I, tropospheric delay error T and satellite orbit error dρ have strong spatial correlation. After the double difference of the short baseline, these error terms are greatly weakened and are generally regarded as noise terms and included in other errors.

[0036] d) Because the coordinates of the base station are known precisely, is a constant value; the position of the mobile station is unknown, so Linearization yields: Among them, the subscript R0 represents the calculation using the reference position of the mobile station. The observation equation using the Kalman filter method is specifically: Where: Observation matrix Measurement error covariance matrix C n*6 =A, D n*n =λI n×n , σ is the standard deviation of the pseudorange and carrier phase measurement errors of each satellite, I represents the identity matrix, and n+1 represents the number of available satellites observed by the base station and the mobile phone.

[0037] e) When satellite j has the best quality (better satellite elevation angle and signal-to-noise ratio) and is set as the reference satellite, the state vector Where: r xyz is the receiver position, v xyz is the velocity, the double difference of the carrier phase ambiguity between the base station and the mobile station and the satellite i and the reference satellite j The subscript gpsL5 represents that the satellite system is GPS and the frequency band is L5. According to the state equation and observation equation, the Kalman filter process is: Where: transfer matrix Covariance matrix

[0038] After filtering, a floating-point solution of the carrier phase ambiguity is obtained. The lambda method is used to fix the ambiguity. When the quality is good, a fixed solution can be obtained. When the quality is poor, it may not be fixed or fixed incorrectly (if the Ratio value obtained by the lambda method is less than 3, it is considered to be fixed incorrectly). At this time, only pseudo-range differential positioning is used to obtain the output result, and the carrier phase measurement is discarded.

[0039] Step 2: Figure 2 As shown, the position and covariance obtained in the first step are smoothed and filtered, including:

[0040] 2.1 For the real-time positioning of the mobile phone in the vehicle, the location information r at time k+1 k+1 , the position r obtained by RTK solution at time k k and speed v k And the kinematic equation obtains the position prediction value r at time k+1 k+1,k and the position information obtained by RTK solution at time k+1 Through the combination of Kalman filters, a more robust and accurate position result is obtained. k+1 , specifically: Among them: subscript k+1 represents k+1 moment, subscript k+1,k represents the transition value from k to k+1 moment, r k+1 =r k+1,k +K k+1 (z k+1 -H k+1 r k+1,k ), r k+1,k =r k +v k *dt, H k+1 =I, in, are the covariance matrices of position and velocity at time k solved by RTK respectively.

[0041] 2.2 For the post-processable vehicle-mounted mobile phone positioning, first perform threshold-based outlier judgment and elimination on all speed information obtained from the overall trajectory solution, and then obtain the complete trajectory speed through linear interpolation. Through the forward smoothing method, the positioning under the time series can be reversely smoothed. Finally, the forward and reverse smoothing results are combined, that is, for data of length n, specifically: r fb =P fb (P f r f +P b r b ), P fb =((P f ) -1+(P b ) -1 ) -1 , where: the positive position and covariance results are r f , P f , the reverse location and covariance result is r b , P b .

[0042] Step 3: Figure 3 and Figure 4 As shown in the figure, according to the position and speed after the second step of smooth filtering, loose combined positioning based on GNSSRTK and MEMS IMU is performed to obtain the longitude, latitude and speed, including:

[0043] 3.1 Taking advantage of the fact that vehicles will not suddenly accelerate or decelerate at the same time in a short period of time, that is, the front and rear position difference sequence will not have large outliers, the position information obtained by RTK and speed smoothing is processed in time series, and the speed covariance matrix is ​​combined to determine the moment of large deviation of the position result. For example, a sliding window with a window length w of 5 and a threshold th of 2 is used for judgment. The moments outside the threshold are marked with a flag to eliminate these possible erroneous position information moments. Specifically, flag = n, mean + th × std > x n >mean-th×std, where: x n Represents the difference in longitude and latitude between adjacent epochs at time n.

[0044] 3.2MEMS IMU initialization uses the RTK solution result of the first epoch for coarse alignment. The position and velocity are the RTK results, and its attitude is: The subscript RTK,k,E indicates the speed calculated by RTK in direction E (east) at time k.

[0045] In this embodiment, a mobile phone MEMS IMU and a high-precision IMU device are placed together to collect data, specifically I LowP and I HighP Finally, the motion information of MEMS data is removed by using the high-precision IMU data during the motion process to obtain the MEMS approximate static data. MEMS =I LowP -I HighP Then, the Allan variance is used to calculate the near-static data I MEMS The random model parameters of the MEMS IMU during the motion process are analyzed. The advantage of this method is that the unknown parameters or the error random model parameters of the vehicle-mounted dynamic mobile phone MEMS that are more accurate than the factory parameters can be obtained with only one acquisition.

[0046] The frequency of MEMS IMU is much higher than the satellite frequency, and in most cases they are not completely aligned at a certain moment. Therefore, it is necessary to compare the timestamp information (i) of the GNSS RTK to be combined with the timestamp information (k, k+1) of the MEMS IMU three-axis accelerometer and gyroscope at the most recent moment before and after (k<i<k+1), and interpolate the data of the MEMS IMU three-axis accelerometer and gyroscope to the corresponding time of RTK solution through linear interpolation.

[0047] If the original measurement value of the gyroscope or accelerometer from time k to time k+1 is a k,k+1 , the measurement data of the gyroscope or accelerometer from time k to time i is specifically a k,i , i to k+1 is a i,k+1 , like Figure 3 shown.

[0048] At the time of non-low-frequency satellite solution, the high-frequency IMU obtains continuous position, velocity and attitude information through mechanical arrangement: it is updated in the order of velocity, position and attitude: in, represents the speed at the previous moment, represents the specific force integral term, Represents the Columbia integral term: in is the rotation matrix from system A at time a to system B at time b, f b (t) is the IMU measurement value in the carrier coordinate system b, is the value of gravity acceleration, Depends on latitude and speed R M , R N Represents the major and minor axes of the Earth, is the latitude value.

[0049] The position update can be further completed based on the updated speed. The differential equation of the position changing with time is described as: in, v D,k represents the velocity value of the ground direction in the NED coordinate system at time k. Note that the above iteration order needs to be based on the altitude h, latitude The longitude λ is updated in order.

[0050] The attitude update is implemented using the cosine matrix: The attitude transformation matrix of b system and n system is and Calculated as: φ k , k is the equivalent rotation vector, and the continuous position, velocity and attitude information of the high-frequency IMU can be obtained.

[0051] When the MEMS timestamp and RTK timestamp are perfectly aligned, they are combined using the following Kalman filter equation:

[0052] Derivation of the state equation: Among them: position r, velocity v, attitude φ, gyro bias b g and accelerometer bias b a , the continuous time differential equation is, from which the transfer matrix Φ can be obtained k,k-1 , The state equation is: k =Φ k,k-1 x k-1 +G k-1 w k-1 , observation equation: D R The conversion matrix from the curve representation method representing the position (latitude, longitude and height) to the rectangular coordinate representation is as follows:

[0053] The fourth step is Figure 5 As shown, parking detection is performed based on the latitude and longitude obtained in the third step, the measured acceleration signal and the gyroscope data to obtain the accurate position, specifically including:

[0054] 4.1 After setting a reasonable range for the estimated accelerometer threshold and gyroscope threshold (e.g., the accelerometer threshold can be set in the range of [0.0, 0.5] m / s / s), the TPE method will randomly select a value as the initial value. When the gyroscope and accelerometer measurement values ​​are less than the threshold for more than 5 consecutive seconds, the time period is judged as the vehicle stop time period;

[0055] 4.2 During the vehicle stop period, the median of the input position is used to replace the position data of the parking interval. The position result obtained by this threshold is compared with the ground truth result to calculate the score. For example, taking the longitude and latitude data of length n as an example: for time j, the error of this time is calculated using the Haversine function: in: λ represents latitude and longitude respectively, and the subscript GT represents ground truth.

[0056] 4.3 Summarize the errors at all times and calculate the average of the first 50% and 95% errors Where: D j ={d1...d n By setting the number of iterations, TPE can effectively explore the search space of these two thresholds, find the optimal parameters that minimize the location score (maximum accuracy) under parking detection, and save this training model for use on test set data without true values.

[0057] Preferably, when the amount of on-board mobile phone data with true position values ​​is sufficient, the thresholds of the MEMS IMU gyroscope and accelerometer under different conditions are subdivided according to the mobile phone model, etc.

[0058] After specific practical experiments, based on the publicly collected in-vehicle mobile phone GNSS and IMU data provided by the third Google Mobile Positioning Challenge, after specific practical experiments, the positioning scenario of the Xiaomi 8 "mi8" mobile phone on the test set section "2022-04-01-18-22-us-ca-lax-t" is used as an example. The path is collected by a car mobile phone near Los Angeles International Airport in California, the United States at 18:00 on April 1, 2022. The positioning result obtained by the present invention is different from the true value result error as shown in the figure below. Figure 6 As shown in the figure, rtk score represents the error score between the GNSS RTK positioning test and the true result (95th percentile error and 50th percentile error average), v_ks score represents the error score between the velocity smoothing filter positioning test and the true result, lc score represents the error score between the loose combination positioning test and the true result, pd score represents the error score between the parking detection judgment test result and the true result, bl_wls score in the table represents the error score between the least squares result of the baseline provided by the dataset and the true value, and rtklib_py score represents the error score obtained by the python version of the open source code rtklib.

[0059] The 95th percentile error, 50th percentile error, score, maximum error and RMSE error of the path data are listed in Table 1.

[0060] Table 1 Mi8 path positioning result error of 2022-04-01-18-22-us-ca-lax-t (unit: m)

[0061] According to Table 1, the error accuracy of the present invention is improved compared with the baseline least squares result provided in the data set. The parking detection score completed on the basis of RTK and loose combination is the best. The 50th percentile error is improved by about 2 meters compared with the baseline result error, and the 95th percentile error is improved by about 5.4 meters. The mean of the two (i.e., the evaluation score index of the present invention) is improved by about 3.7 meters. Due to the large error of the pseudo-range signal, the result of the single-point positioning using only pseudo-range performs very poorly in terms of the maximum error, up to 7444 meters. Partially erroneous carrier phase fixation will also cause the position error of rtklib to fluctuate, with a maximum error of 51 meters, but the maximum error of the present invention does not exceed 15 meters. Compared with the score positioning accuracy of the open source software rtklib, it is also greatly improved by about 32%, 0.8 meters.

[0062] Based on the GNSS RTK and pine combination results in the present invention, the TPE method is used to autonomously find the optimal threshold for parking detection. In Table 2, the true category is positive examples representing the moment when the true value speed is zero, and the predicted category is positive examples representing the moment when the speed determined by parking detection is zero.

[0063] Table 2 Confusion matrix of parking detection judgment at 1462 moments of mi8 path of 2022-04-01-18-22-us-ca-lax-t The number of times the detection speed is zero The number of times the detection speed is non-zero The number of times the true velocity is zero 50 28 The number of moments when the true velocity is non-zero 5 1379

[0064] By comparing the speed with the true value result, it is determined that when the true value speed is zero, the accuracy of the method can be detected The solution process is as follows: Figure 5 As shown, Figure 7 Showing the difference between the solution speed and the true value speed, the figure is at 8.2e 12 The period of time milliseconds ago is the parking period under the viaduct. The specific positioning scenario is as follows Figure 8 As shown on the surface of the earth, during this period, the present invention successfully detected the parking situation, and the error was greatly reduced compared with other methods.

[0065] The present invention is used to process 156 path data of 20 types of smartphones in the training set of all ground truth results in the public data set provided by the 3rd Mobile Positioning Challenge, and the obtained average scores are shown in Table 3.

[0066] Table 3 Mean error of positioning results of 156 paths in the training set (unit: m)

[0067] As shown in Table 3, the accuracy of the present invention is improved by 2.7 meters, about 65.38%, compared with the least squares baseline reference result, and the maximum error accuracy is improved by 1886.8854 meters, about 99.698%. The mean of the 50th percentile error results of the present invention is less than 1m, and the mean error is below 1.5m, which is more than 2.6m higher than the baseline result. Based on the results of GNSS RTK and loose combination, the various percentile errors and maximum errors obtained by using the TPE method to autonomously find the optimal threshold method for parking detection are still the smallest.

[0068] Compared with the baseline positioning provided by the competition, the relative positioning of the traditional open source software rtklib has an improvement of 0.85 and 2.6 meters in scoring accuracy, respectively. When the maximum values ​​of 156 paths are averaged, the maximum values ​​of the two are 100742 meters and 1892 meters, respectively. It can be seen that these two algorithms cannot guarantee the reliability and stability of the results. The maximum error mean of the present invention is 6.8 meters, and both accuracy and reliability are guaranteed. Based on the RTK method, speed smoothing filtering and loose combination methods can further improve the scoring accuracy by 0.7 centimeters and 4 centimeters. The multiple positioning accuracy obtained after parking detection judgment is the highest.

[0069] The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principle and purpose of the present invention. The protection scope of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. Each implementation scheme within its scope shall be subject to the constraints of the present invention.

Claims

1. A method for GNSS RTK and MEMSIMU fusion positioning based on a vehicle-mounted smartphone terminal, characterized in that: By obtaining the original measurement values ​​of GNSS and MEMSIMU, GNSS is converted into the international common format RINEX, and nearby public base station satellite data is downloaded for RTK positioning of the base station and mobile phone, and the position solution is obtained by velocity smoothing filtering; and then the smooth transition when the satellite quality is poor is completed through the loose combination of GNSS RTK and MEMSIMU; Finally, the tree-structured Parzen estimation algorithm is used to complete the threshold judgment of parking detection on the IMU data, and the accuracy difference is calculated with the true value of the data, and the optimal threshold and positioning result are iteratively obtained.

2. The method for GNSS RTK and MEMSIMU fusion positioning based on a vehicle-mounted smartphone terminal according to claim 1 is characterized in that: The GNSS RTK and MEMSIMUS combination specifically includes: a) Taking advantage of the fact that vehicles do not suddenly accelerate or decelerate at the same time in a short period of time, the results of satellite positioning that may have large errors are eliminated; b) Using high-precision inertial navigation sensor data to obtain the error random model parameters of the mobile phone MEMS during vehicle motion, where the random model parameters are unknown; c) Use the satellite solution speed at the initial moment to complete the initial attitude alignment of the MEMS sensor; d) By linear interpolation, the high-frequency MEMS timestamp and the low-frequency satellite timestamp are completely aligned, and the increment values ​​of the aligned accelerometer and gyroscope are obtained to complete the mechanical arrangement algorithm; e) The processed satellite-solved velocity and position information is loosely combined with the MEMS data to obtain the initial values ​​and random model parameters without considering the arm distance.

3. The method for GNSS RTK and MEMSIMU fusion positioning based on a vehicle-mounted smartphone terminal according to claim 1 is characterized in that: The tree structure Parzen estimation algorithm specifically includes: i) Setting a wide range of reasonable thresholds for the accelerometer and gyroscope data to determine the parking time, and determining that the time period is the vehicle stopping time period when the gyroscope and accelerometer measurement values ​​are less than the threshold for more than 5 consecutive seconds; ii) Replace the location data of the parking interval with the median of the input location, compare the location result obtained by this threshold with the ground truth result, and calculate the score; iii) It is beneficial for the TPE method to iteratively explore the search space of these two thresholds and find the optimal parameters that achieve the highest position accuracy under parking detection; iv) Save the trained model so that the parameter model can be reused on the test set data without ground track truth.

4. The method for GNSS RTK and MEMSIMU fusion positioning based on a vehicle-mounted smartphone terminal according to claim 1 is characterized in that: The RTK positioning of the reference station and the mobile phone, that is, positioning by GNSS RTK according to pseudorange, carrier phase and base station position to obtain position and covariance, is specifically: obtaining the original satellite and inertial device data measurement values ​​through the API of the vehicle-mounted smart phone, and converting the satellite measurement values ​​into RINEX format, searching for public reference station information near the road where the car is traveling, downloading the satellite observation file and navigation message of the corresponding time; when the satellite signal quality is poor or the satellite status of the current epoch does not meet the requirements, pseudorange differential positioning or single-point positioning is used instead, and the position and speed information of each moment in the driving process of the car is obtained by solving the present invention.

5. The method for GNSS RTK and MEMSIMU fusion positioning based on a vehicle-mounted smartphone terminal according to claim 4 is characterized in that: The present invention specifically comprises: a) When the reference station B and the mobile station R observe satellite i and satellite j at the same time, the pseudorange and carrier phase observation equations of the reference station are specifically: b) For base stations and other satellites, the pseudorange and carrier phase observation results are similar, specifically: Where: subscript f is a specific carrier band, subscript B represents a base station, subscript R represents a mobile station, and superscript i represents a base station. G represents GNSS satellite i, ρ is the geometric distance between the satellite and the receiver, dt is the clock difference, I is the ionospheric delay, T is the tropospheric delay, λ is the wavelength of the f band, dρ is the orbit error, and v and ε are other errors; c) The double difference combination of short baseline observations is: in: is the double difference operator, d) Because the coordinates of the base station are known precisely, is a constant value; the position of the mobile station is unknown, so Linearization yields: Among them, the subscript R0 represents the calculation using the reference position of the mobile station. The observation equation using the Kalman filter method is specifically: Where: Observation matrix Measurement error covariance matrix C n*6 =A, D n*n =λI n×n , σ is the standard deviation of the pseudorange and carrier phase measurement errors of each satellite, and I represents the identity matrix; e) When satellite j has a better satellite elevation angle and signal-to-noise ratio, it is set as the reference satellite, and the state vector Where: r xyz is the position of the receiver, v xyz is the velocity, the double difference of the carrier phase ambiguity between the base station and the mobile station and the satellite i and the reference satellite j The subscript gpsL5 represents that the satellite system is GPS and the frequency band is L5. According to the state equation and observation equation, the Kalman filter process is: Where: transfer matrix Covariance matrix 6. The method for GNSS RTK and MEMSIMU fusion positioning based on a vehicle-mounted smartphone terminal according to claim 1 is characterized in that: The smoothing filtering specifically includes: 2.1 For the real-time positioning of the mobile phone in the vehicle, the location information r at time k+1 k+1 , the position r obtained by RTK solution at time k k and speed v k And the kinematic equation obtains the predicted position value r at time k+1 k+1,k and the position information obtained by RTK solution at time k+1 Through the combination of Kalman filters, a more robust and accurate position result is obtained. k+1 , specifically: Among them: subscript k+1 represents k+1 moment, subscript k+1,k represents the transition value from k to k+1 moment, r k+1 =r k+1,k +K k+1 (z k+1 -H k+ 1r k+1,k ), r k+1,k =r k +v k *dt, H k+1 =I, in, are the covariance matrices of position and velocity at time k obtained by RTK solution; 2.2 For the post-processable vehicle-mounted mobile phone positioning, first perform threshold-based outlier judgment and elimination on all speed information obtained from the overall trajectory solution, and then obtain the complete trajectory speed through linear interpolation. Through the forward smoothing method, the positioning under the time series can be reversely smoothed. Finally, the forward and reverse smoothing results are combined, that is, for data of length n, specifically: P fb =((P f ) -1 +(P b ) -1 ) -1 , r fb =P fb (P f r f +P b r b ), where: the positive position and covariance results are r f , P f , the reverse location and covariance result is r b , P b .

7. The method for GNSS RTK and MEMSIMU fusion positioning based on a vehicle-mounted smartphone terminal according to claim 1 is characterized in that: The loose combination positioning specifically includes: 3.1 Taking advantage of the fact that vehicles will not suddenly accelerate or decelerate at the same time in a short period of time, that is, the front and rear position difference sequence will not have large outliers, the position information obtained by RTK and speed smoothing is processed in time series, and the speed covariance matrix is ​​combined to determine the moment of large deviation of the position result. For example, a sliding window with a window length w of 5 and a threshold th of 2 is used for judgment. The moments outside the threshold are marked with a flag to eliminate these possible erroneous position information moments. Specifically, flag = n, mean + th × std > x n >mean-th×std, where: x n Represents the difference in longitude and latitude between adjacent epochs at time n; 3.2 Initialization of MEMS IMU uses the RTK solution result of the first epoch for coarse alignment. The position and velocity are the RTK results, and its attitude is:

8. The method for GNSS RTK and MEMSIMU fusion positioning based on a vehicle-mounted smartphone terminal according to claim 1 is characterized in that: The parking detection specifically includes: 4.1 After setting the range of the estimated accelerometer threshold and gyroscope threshold, the TPE method will randomly select a value as the initial value. When the gyroscope and accelerometer measurement values ​​are less than the threshold for more than 5 consecutive seconds, the time period is judged as the vehicle stop time period; 4.2 During the vehicle stop period, the median of the input position is used to replace the position data of the parking interval. The position result obtained by this threshold is compared with the ground truth result to calculate the score: For time j, the error at time j is calculated using the Haversine function: in: λ represents latitude and longitude respectively, and the subscript GT represents groundtruth. 4.3 Summarize the errors at all times and calculate the average of the first 50% and 95% errors Where: D j ={d1...d n By setting the number of iterations, TPE can effectively explore the search space of these two thresholds, find the optimal parameters that minimize the location score under parking detection, that is, the highest accuracy, and save this training model for use on test set data without true values.

9. The method for GNSS RTK and MEMSIMU fusion positioning based on a vehicle-mounted smartphone terminal according to claim 8 is characterized in that: When the amount of data from the on-board mobile phone with true position value is sufficient, the thresholds of the MEMSIMU gyroscope and accelerometer under different conditions are subdivided according to the mobile phone model, etc.

10. A system for GNSS RTK and MEMSIMU fusion positioning based on a vehicle-mounted smartphone terminal to implement any of the methods described in claims 1-9, characterized in that: include: GNSS RTK positioning unit, speed smoothing filter unit, loose combination unit and parking detection unit, wherein: the GNSS RTK positioning unit performs filtering and solving processing according to the GNSS signals of the vehicle-mounted mobile phone and the nearby public base station to obtain the position and speed information of the vehicle; the speed smoothing filter unit performs smoothing filtering on the position result in the form of kinematic equations after screening and interpolating the obtained vehicle celestial speed to obtain the position solution; the loose combination unit removes outliers from the obtained position solution and performs loose combination processing with the position, speed and attitude information obtained and processed by the vehicle-mounted mobile phone MEMSIMU sensor to obtain a smooth position solution; The parking detection unit determines the threshold value through the TPE method based on the gyro and accelerometer information of the MEMSIMU, completes the detection of the vehicle stop period, and obtains a more accurate position solution.

Images