A vehicle-mounted Beidou positioning deviation self-calibration system integrating inertial navigation information

Abstract

The present invention discloses a vehicle-mounted Beidou positioning deviation self-calibration system that integrates inertial navigation information, and relates to the field of navigation and positioning technology. It is used to solve the problems of positioning deviation accumulation and signal failure in complex environments. The system realizes the spatiotemporal alignment of inertial navigation and Beidou observation data through timestamp interpolation and coordinate system conversion, and constructs a multi-source synchronous data stream. Based on the vehicle's incomplete constraint model, the difference between the inertial navigation-calculated speed and the Beidou Doppler speed is analyzed, and a weight factor is generated in combination with a confidence threshold to suppress abnormal observations. The geometric matching error is calculated by high-precision lane map curvature and vehicle steering angle, and the candidate trajectory is screened to generate lateral calibration parameters. The noise covariance is dynamically adjusted through a tightly coupled filtering algorithm, and the lane constraint calibration mode is switched when the Beidou signal is interrupted. The inertial navigation error is compensated through error boundary constraints to achieve stable output. Deep fusion of multi-source data and adaptive anti-error calibration are achieved to improve system robustness and positioning continuity.

Description

Technical Field

[0001] The present invention relates to the field of navigation and positioning technology, and in particular to a vehicle-mounted Beidou positioning deviation self-calibration system that integrates inertial navigation information. Background Art

[0002] With the rapid development of smart cars, vehicle-road collaboration, and intelligent transportation systems, high-precision positioning capabilities for vehicles while driving have become the fundamental support for achieving functions such as lane-level assisted driving, automatic parking, path planning, and active safety control. Traditional positioning methods struggle to guarantee continuity and accuracy, and are prone to problems such as signal loss, jumps, and error accumulation, seriously impacting vehicle operational safety and control effectiveness. Therefore, a technical solution is urgently needed that can integrate information from multiple sensors and achieve stable, reliable, and high-precision positioning to meet the urgent needs of intelligent transportation systems for positioning continuity, accuracy, and robustness.

[0003] Currently, commonly used vehicle positioning technologies rely primarily on the combined application of satellite navigation systems and inertial navigation systems. Common combinations include loose and tight combinations. In the loose combination structure, the position results provided by the satellite positioning system are used as external input to correct inertial navigation errors. This approach is highly dependent on satellite signals, and positioning errors accumulate rapidly if the signals are blocked. The tight combination structure combines raw observations such as satellite pseudoranges or carrier phases with inertial data for calculation. While this improves fusion accuracy to a certain extent, practical applications still suffer from issues such as insufficient sensor error modeling and underutilization of motion constraints. Furthermore, existing methods generally ignore the nonholonomic kinematic constraints imposed on vehicles during actual road driving and the lane geometry information contained in high-precision maps, failing to effectively incorporate these into the error analysis and compensation process. This results in a lack of effective positioning calibration mechanisms in scenarios where satellite signals are unavailable or severely degraded, resulting in technical bottlenecks such as poor stability and rapid error drift. A highly robust vehicle positioning system that integrates vehicle kinematic constraints with road geometry and possesses adaptive error control capabilities is urgently needed. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides a vehicle-mounted Beidou positioning deviation self-calibration system that integrates inertial navigation information, which solves the problems of the above-mentioned background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a vehicle-mounted Beidou positioning deviation self-calibration system that integrates inertial navigation information, including the following modules: a multi-source data synchronization and alignment module, a kinematic constraint consistency analysis module, a lane geometry constraint analysis module, and an adaptive fusion and calibration module; the multi-source data synchronization and alignment module is used to perform time series synchronization based on timestamp interpolation on the three-axis acceleration and three-axis angular velocity data output by the inertial navigation system, and the pseudo-range observation values ​​and carrier phase observation values ​​output by the Beidou satellite navigation system, and realize spatial reference alignment through coordinate system conversion, and output a multi-source data stream with a unified time and space reference; the kinematic constraint consistency analysis module is used to calculate the difference between the inertial navigation system-calculated speed and the Beidou system-solved speed based on Doppler frequency shift according to the vehicle's incomplete constraint model. , when the difference exceeds the dynamic confidence threshold, the corresponding data credibility weight factor is generated; the lane geometry constraint analysis module is used to call the lane curvature information in the high-precision lane-level map, combined with the real-time steering angle data of the vehicle, calculate the matching error between the lane geometry constraint and the vehicle kinematic model, screen the candidate trajectories that meet the matching error threshold, and output the lateral position calibration parameters and error boundaries; the adaptive fusion and calibration module is used to fuse multi-source data based on the tightly coupled filtering algorithm when the on-board Beidou signal is available, dynamically adjust the noise covariance weight of the Beidou observation value through the weight factor, and output the robust fusion positioning result. When the on-board Beidou signal is continuously lost for more than a preset threshold, it switches to the calibration mode driven by lane geometry constraints, suppresses the accumulated error of inertial navigation based on the candidate trajectory and error boundary, and outputs continuous and stable position information.

[0006] Furthermore, the timing synchronization of the multi-source data synchronization and alignment module includes the following: performing timestamp interpolation alignment on the high-frequency three-axis acceleration and angular velocity data output by the inertial navigation system, and the low-frequency pseudorange and carrier phase observation values ​​output by the Beidou satellite navigation system, to eliminate the timing mismatch caused by sampling rate differences; through the dynamic delay compensation mechanism, correcting the fixed delay and random jitter in the data transmission process to ensure the timing consistency of multi-source data streams; performing conflict detection on the interpolated timing sequence, identifying abnormal segments with reversed timestamps and exceeded intervals, and performing local resampling repairs through adjacent valid data segments.

[0007] Furthermore, the specific process of calculating the difference between the speed calculated by the inertial navigation system and the speed solved by the Beidou system based on Doppler frequency shift according to the vehicle's non-holonomic constraint model is as follows: based on the Ackermann steering geometry relationship, the theoretical constraint equation of the lateral speed is constructed, and the theoretical value of the lateral speed calculated by the inertial navigation system is derived; the speed solved by the Beidou Doppler frequency shift is rotated from the carrier coordinate system to the navigation coordinate system, and the antenna installation angle error is compensated; the vehicle acceleration confidence interval is introduced, and the confidence threshold of the speed difference is dynamically softened: the threshold is expanded in sudden acceleration or braking scenarios, and the threshold is tightened in uniform speed scenarios.

[0008] Furthermore, when the difference exceeds the dynamic confidence threshold, the specific process of generating the corresponding data credibility weight factor is as follows: a weight decay curve based on the S-type function is designed. When the difference exceeds the threshold, the weight factor decreases nonlinearly with the excess ratio to avoid weight jumps caused by hard judgments; a time decay factor is introduced to perform exponential decay weighting on historical abnormal events to ensure that short-term anomalies do not permanently reduce the credibility of the data source; the weight recovery rate is adaptively adjusted according to the road type, accelerating the recovery of Beidou data weight in curved scenarios and prioritizing trust in inertial data in straight scenarios.

[0009] Furthermore, the lane curvature information in the high-precision lane-level map is called upon, combined with the vehicle's real-time steering angle data, to calculate the matching error between the lane geometric constraints and the vehicle kinematic model. The specific process is as follows: discrete curvature points of the lane centerline are extracted from the high-precision map, a continuous curvature function is constructed through multiple spline interpolations, and its rate of change is calculated; based on the vehicle's real-time steering angle and longitudinal speed, the theoretical lane curvature is inferred through the kinematic model, and a tire cornering stiffness compensation term is introduced to improve model accuracy; the matching error between the actual curvature and the theoretical curvature is used as the main term of the error function, and the vehicle's lateral offset is mapped as a curvature error gain coefficient to perform a weighted correction on the error function; the penalty weight for trajectories close to the lane edge is increased to enhance the model's responsiveness to boundary constraints.

[0010] Furthermore, the specific process of screening candidate trajectories that meet the matching error threshold and outputting the lateral position calibration parameters and error boundaries is as follows: the tracks generated by inertial dead reckoning are segmented through a sliding window mechanism to construct candidate trajectory clusters; the weighted matching error is calculated for each candidate trajectory, and they are ranked based on the probability score, and the non-inferior trajectory clusters are screened through the Pareto front; the Gaussian mixture model is applied to the selected preferred trajectory clusters for cluster analysis, and the lateral position mean and covariance matrix of the cluster center trajectory are extracted as the calibration parameters and confidence intervals.

[0011] Furthermore, when the vehicle-borne Beidou signal is available, multi-source data is fused based on the tightly coupled filtering algorithm, and the noise covariance weight of the Beidou observation value is dynamically adjusted through the weight factor to output the robust fusion positioning result. The specific process is as follows: a joint state vector containing the inertial device zero bias, scale factor error and Beidou receiver clock error is constructed, the inertial device temperature drift term is explicitly modeled in the state vector, and a covariance correlation matrix is ​​established with the Beidou clock error term; the weight factor is mapped to the independent noise covariance component of the Beidou pseudorange and carrier phase observation values, and the low-confidence Beidou observation values ​​are down-weighted to achieve heterogeneous down-weighting of multiple observation sources; the cost function is constructed through the robust M-estimator, and the filter residual is reweighted iteratively to suppress the influence of multi-path abnormal observations.

[0012] Furthermore, when the continuous loss of the on-board Beidou signal exceeds a preset threshold, it switches to a calibration mode driven by lane geometry constraints. The specific process of suppressing the accumulated error of inertial navigation based on candidate trajectories and error boundaries is as follows: in the early stage of Beidou signal failure, a lane-binding corridor is generated based on historical positioning results and high-precision maps to constrain the lateral divergence of inertial dead reckoning; through an improved residual resampling particle filter, lane curvature constraints are injected into the proposal distribution to force particles to be distributed along a reasonable lane centerline; and the calibrated zero-bias estimate is reversely injected into the tightly coupled filter state vector to achieve smooth mode switching when the signal is restored and avoid positioning jumps.

[0013] The present invention has the following beneficial effects:

[0014] (1) A vehicle-mounted BeiDou positioning deviation self-calibration system that integrates inertial navigation information can achieve accurate alignment of the output data of the inertial navigation system and the BeiDou system in terms of time sequence and space reference by setting up a multi-source data synchronization and alignment module and a kinematic constraint consistency analysis module, thereby improving the basic consistency before multi-source information fusion; at the same time, based on the vehicle's incomplete constraint model, a dynamic analysis of the speed estimation difference is performed, and a weight factor is generated in combination with a confidence judgment mechanism, which effectively suppresses the interference of abnormal data on the fusion result and enhances the system's ability to judge the reliability of navigation data under abnormal conditions.

[0015] (2) A vehicle-mounted Beidou positioning deviation self-calibration system that integrates inertial navigation information. By setting up a lane geometry constraint analysis module and an adaptive fusion and calibration module, it uses high-precision lane-level maps and real-time steering angle data to establish a spatial matching constraint relationship, thereby achieving accurate estimation and constraint correction of the vehicle's lateral deviation. When the Beidou signal is available, the system fuses multi-source observation information based on a tightly coupled filtering algorithm and improves the anti-error performance through dynamic weight control. In the event of continuous loss of the Beidou signal, it automatically switches to a calibration mode driven by geometric constraints, which can effectively limit the accumulation of inertial navigation errors and ensure the continuity and stability of positioning results in complex environments.

[0016] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of a vehicle-mounted Beidou positioning deviation self-calibration system that integrates inertial navigation information according to the present invention.

[0018] Figure 2 This is a flow chart of the adaptive fusion and calibration module of the present invention. DETAILED DESCRIPTION

[0019] The embodiment of the present application solves the problems of the existing vehicle-mounted Beidou positioning system in complex environments, such as the decline in positioning accuracy, the inability to effectively suppress the accumulation of inertial navigation system errors, and the lack of a dynamic credibility assessment mechanism, through a vehicle-mounted Beidou positioning deviation self-calibration system that integrates inertial navigation information. The spatiotemporal benchmark consistency is achieved through a multi-source data synchronization and alignment module, strengthening the foundation for multi-source observation data fusion; the dynamic detection and confidence quantification of speed deviations are achieved through a kinematic constraint consistency analysis module, improving the ability to identify abnormal data; the lateral position correction is performed using high-precision maps and vehicle status data in combination with the lane geometry constraint analysis module, improving the accuracy of vehicle path constraints; and finally, the dynamic anti-error fusion and failure switching strategy of multi-source information are achieved through an adaptive fusion and calibration module, significantly enhancing the continuity and reliability of the vehicle-mounted positioning system in scenarios where satellite signals are weakened or lost.

[0020] The overall idea of ​​the solution in the embodiments of this application is as follows:

[0021] Firstly, the three-axis acceleration and three-axis angular velocity data output by the inertial navigation system (INS) and the pseudorange observations and carrier phase observations output by the BeiDou satellite navigation system (BDS) are synchronized based on timestamp interpolation through the timing synchronization module.

[0022] Then, the coordinate system conversion method is used to unify the inertial navigation data and Beidou satellite navigation data into the same spatial reference, thereby outputting a multi-source data stream with a unified time and space reference, preparing for subsequent processing.

[0023] Based on the nonholonomic constraint model of the vehicle, the difference between the speed calculated by the inertial navigation system and the speed calculated by the BeiDou system based on Doppler frequency shift is calculated.

[0024] When the speed difference between the two exceeds the dynamic confidence threshold, a corresponding data credibility weight factor is generated. This weight factor will dynamically adjust the weight of the observation value in the subsequent data fusion process to ensure that more reliable data has a greater impact on the system output.

[0025] The lane curvature information in the high-precision lane-level map is called and combined with the vehicle's real-time steering angle data to calculate the matching error between the lane geometric constraints and the vehicle kinematic model.

[0026] Based on the calculated matching error, candidate trajectories that meet the matching error threshold are screened out, and the corresponding lateral position calibration parameters and error boundaries are output to provide support for subsequent calibration.

[0027] When the vehicle-mounted BeiDou signal is available, the multi-source data is fused using a tightly coupled filtering algorithm. At this point, the noise covariance weights of the BeiDou observations are dynamically adjusted based on the calculated credibility weights, thereby outputting robust fusion positioning results.

[0028] If the onboard BeiDou signal loss exceeds a preset threshold, the system automatically switches to a calibration mode driven by lane geometry constraints. This mode uses candidate trajectories and error bounds to suppress the accumulated error of the inertial navigation system, ensuring the stability and continuity of positioning results.

[0029] See also Figure 1 , an embodiment of the present invention provides a technical solution: a vehicle-mounted Beidou positioning deviation self-calibration system that integrates inertial navigation information, including the following modules: a multi-source data synchronization and alignment module, a kinematic constraint consistency analysis module, a lane geometry constraint analysis module, and an adaptive fusion and calibration module; the multi-source data synchronization and alignment module is used to perform time series synchronization based on timestamp interpolation on the three-axis acceleration and three-axis angular velocity data output by the inertial navigation system, and the pseudo-range observation values ​​and carrier phase observation values ​​output by the Beidou satellite navigation system, and realize spatial reference alignment through coordinate system conversion, and output a multi-source data stream with a unified time and space reference; the kinematic constraint consistency analysis module is used to calculate the difference between the inertial navigation system-calculated speed and the Beidou system-calculated speed based on the Doppler frequency shift solution according to the vehicle's incomplete constraint model. When the difference When the dynamic confidence threshold is exceeded, the corresponding data credibility weight factor is generated; the lane geometry constraint analysis module is used to call the lane curvature information in the high-precision lane-level map, combined with the vehicle's real-time steering angle data, calculate the matching error between the lane geometry constraint and the vehicle kinematic model, screen the candidate trajectories that meet the matching error threshold, and output the lateral position calibration parameters and error boundaries; the adaptive fusion and calibration module is used to fuse multi-source data based on the tightly coupled filtering algorithm when the on-board Beidou signal is available, dynamically adjust the noise covariance weight of the Beidou observation value through the weight factor, and output the robust fusion positioning result. When the on-board Beidou signal is continuously lost for more than the preset threshold, it switches to the lane geometry constraint-driven calibration mode, suppresses the accumulated error of inertial navigation based on the candidate trajectory and error boundary, and outputs continuous and stable position information.

[0030] In this implementation, the multi-source data synchronization and alignment module: This module's core task is to unify the temporal and spatial processing of observation data from different sensor systems (inertial navigation systems and Beidou satellite navigation systems). Due to the different sampling frequencies and clock deviations of each sensor, time synchronization is achieved through timestamp-based interpolation methods to align all data based on the same time reference. Furthermore, since each system operates in a different coordinate system (for example, inertial navigation systems often use a vehicle coordinate system, while Beidou systems use a geographic coordinate system or a geocentric coordinate system), this module achieves spatial reference consistency through coordinate system conversion. Three-axis acceleration and three-axis angular velocity: These are the accelerometer and gyroscope data acquired by the inertial measurement unit (IMU) in the inertial navigation system, respectively, and serve as the basic input for vehicle attitude and velocity estimation; pseudorange observations: The propagation time between the Beidou receiver and the satellite is multiplied by the speed of light, which contains various error components; carrier phase observations: Compared to pseudorange, they offer higher measurement accuracy and are used for precise positioning. Kinematic Constraint Consistency Analysis Module: This module establishes a vehicle kinematic inference model based on the vehicle's physical characteristics, known as "nonholonomic constraints." It evaluates the consistency of the data from the two systems by comparing the difference between the inertial navigation system's inferred velocity and the Beidou system's velocity measured using Doppler shift. If the difference exceeds a preset dynamic confidence threshold, indicating a possible anomaly or error, the module generates a data confidence weighting factor for dynamic weighting in the subsequent fusion process. Nonholonomic constraint models are often used to describe practical kinematic constraints where a vehicle is restricted to forward motion and lateral movement is restricted. Doppler shift-determined velocity: Calculates the receiver's radial velocity relative to the satellite based on satellite signal frequency shift, reflecting the vehicle's actual speed. Weighting factors are used to adjust the influence of data reliability on the fusion results during multi-source information fusion. Lane Geometric Constraint Analysis Module: This module utilizes detailed road geometry information (such as curvature and width) from high-precision lane-level maps (HD maps) and combines it with real-time steering angle data to perform geometric constraint matching on the vehicle's trajectory. By constructing a geometric error model, the deviation between the vehicle's actual motion trajectory and the lane model is calculated, and candidate trajectories with an acceptable error range are screened. Lateral position calibration parameters and error bounds are then output to achieve lateral compensation for inertial navigation system position drift. High-precision lane-level maps: These maps have higher resolution than traditional navigation maps, achieving lane-level accuracy. Steering angle data: These are derived from steering wheel angle sensors or front wheel angle sensors. Error bounds: These represent an acceptable confidence interval for trajectory error, used for subsequent fusion judgment. The adaptive fusion and calibration module: This module is the fusion core of the entire system.When BeiDou signals are normal, a tight coupling filter algorithm is used to fuse and estimate the INS and BeiDou data. The noise covariance matrix of the BeiDou observations is dynamically adjusted based on the aforementioned weighting factors, achieving robust estimation and outputting high-precision positioning results. However, when BeiDou signals are continuously lost for a period exceeding a threshold (e.g., in occlusion scenarios), the system automatically switches to an error suppression mode based on lane geometry constraints. This mode uses candidate trajectories and error bounds to suppress the accumulated error of the INS, ensuring continuous and stable position output. The tight coupling filter algorithm performs a joint state space estimation between INS and GNSS observations, independent of external preliminary positioning. Noise covariance weighting adjusts the contribution of different observation sources to the estimated state based on their reliability. Robust fusion ensures that even if some observations are anomalous, the overall estimate remains stable, preventing it from being dominated by outliers. Accumulated error refers to the integral drift caused by long-term operation of the INS system, which often accumulates over time.

[0031] Specifically, the timing synchronization of the multi-source data synchronization and alignment module includes the following: performing timestamp interpolation alignment on the high-frequency three-axis acceleration and angular velocity data output by the inertial navigation system, and the low-frequency pseudorange and carrier phase observation values ​​output by the Beidou satellite navigation system, to eliminate the timing mismatch caused by sampling rate differences; through the dynamic delay compensation mechanism, correcting the fixed delay and random jitter in the data transmission process to ensure the timing consistency of multi-source data streams; performing conflict detection on the interpolated timing sequence, identifying abnormal segments with reversed timestamps and exceeded intervals, and performing local resampling repairs through adjacent valid data segments.

[0032] In this implementation, timestamp interpolation alignment is used: The three-axis acceleration and three-axis angular velocity data output by the inertial navigation system (INS) are sampled at a high frequency (typically hundreds of Hz or higher), while the pseudorange and carrier phase observations output by the Beidou Navigation Satellite System (BDS) are sampled at a lower frequency (typically a few Hz or lower). Due to this sampling rate difference, direct comparison of these two data types results in timing mismatches, preventing proper fusion. Operation: A timestamp interpolation algorithm is used to align the high-frequency inertial navigation data and the low-frequency Beidou satellite data to the same time base. This method uses interpolation methods (such as linear or spline interpolation) to fill in the gaps between the lower-frequency data to ensure that both have corresponding values ​​at the same time point, eliminating timing deviations caused by sampling frequency differences. Dynamic delay compensation: When data is transmitted from sensors (such as inertial navigation systems and satellite navigation systems) to the data fusion system, certain transmission delays occur. These delays can be fixed or affected by random jitter. Operation: A dynamic delay compensation mechanism is introduced to correct for transmission delays in data from different sources. Fixed latency refers to the regular delay caused by system architecture or communication links, while random jitter refers to latency variations due to inherent uncertainties in the communication network or sensors. This mechanism infers and compensates for these delays based on real-time measurements and historical data, ensuring the time consistency of multi-source data streams. Conflict Detection and Repair: Timestamps may become out of order or have abnormal time intervals due to various reasons, such as clock skew or sensor failure. For example, sensors may experience data loss or corruption at certain moments, resulting in out-of-order timestamps or excessively long time intervals. Action: Conflict detection is performed on the interpolated time series data to identify potential anomalies, such as timestamp inversion (i.e., timestamps not arranged in ascending order) or time interval violations (i.e., data time intervals are abnormally large). Once such a conflict is detected, the system resolves it through local resampling repair. This method uses the values ​​of adjacent valid data segments to compensate or resample the abnormal data segments, restoring them to a reasonable time series state. This mechanism ensures that the final output data stream is time-continuous and anomaly-free.

[0033] Specifically, the specific process of calculating the difference between the speed calculated by the inertial navigation system and the speed solved by the Beidou system based on Doppler frequency shift according to the vehicle's non-holonomic constraint model is as follows: based on the Ackermann steering geometry relationship, the theoretical constraint equation of the lateral speed is constructed, and the theoretical value of the lateral speed calculated by the inertial navigation system is derived; the speed solved by the Beidou Doppler frequency shift is rotated from the carrier coordinate system to the navigation coordinate system, and the antenna installation angle error is compensated; the vehicle acceleration confidence interval is introduced, and the confidence threshold of the speed difference is dynamically softened: the threshold is expanded in sudden acceleration or braking scenarios, and the threshold is tightened in uniform speed scenarios.

[0034] In this implementation, the theoretical constraint equation of lateral velocity is constructed: Based on the Ackerman steering geometry, the lateral velocity The longitudinal speed can be and steering angle For a vehicle with a steering system, the lateral speed is determined by the radius of the curve the vehicle is traveling and the steering angle of the vehicle. The formula is: Assuming the steering angle of the vehicle is and longitudinal speed It is known that the lateral velocity It can be calculated using the following more complex formula: ;in: : lateral velocity of the vehicle. : Longitudinal velocity of the vehicle. : Vehicle wheelbase (distance between wheels). : The turning radius of the vehicle. : The steering angle of the vehicle. Coordinate system conversion: The BeiDou satellite navigation system and the inertial navigation system use different coordinate systems. The speed calculated by BeiDou needs to be converted from the satellite coordinate system to the vehicle's navigation coordinate system. Usually, the speed calculated by BeiDou is based on the earth's inertial coordinate system, while the inertial navigation system is based on the vehicle's coordinate system. Therefore, a rotation matrix is ​​needed to convert the coordinate system. Formula: Through the rotation matrix Increase the speed of the BeiDou system Convert from satellite coordinate system to vehicle coordinate system: ;in: : The speed of the vehicle in the navigation coordinate system. : The speed of BeiDou Doppler solution. : Coordinate system rotation matrix, Antenna installation angle error compensation: Antenna installation angle error will affect the speed calculated by the BeiDou system. Therefore, it is necessary to compensate for the known antenna installation error angle. To compensate for the installation error of the antenna. The compensation process corrects the speed data solved by Beidou by rotating the matrix. The formula shows that: if the installation error angle of the antenna is taken into account , the speed after compensation It can be calculated by the following formula: ;in: : Speed ​​after compensation. : The speed of the vehicle in the navigation coordinate system. : Antenna installation angle error. : The velocity components of the vehicle on the x and y axes. Dynamic softening: Based on the acceleration of the vehicle , dynamically adjusts the confidence threshold of the speed difference. When the vehicle is in a state of rapid acceleration or sudden braking, a larger speed difference is allowed, so the threshold will be expanded; while in a constant speed state, the threshold will be tightened to enhance accuracy. This dynamic adjustment helps to adapt to different driving scenarios. The formula is: Based on the vehicle's acceleration and acceleration threshold , the confidence threshold can be dynamically adjusted by the following formula : ;in: : Dynamically adjusted speed difference threshold. : Basic speed difference threshold. : Threshold adjustment coefficient, used to determine the extent of threshold expansion. : The acceleration of the vehicle. : Acceleration threshold, used to determine whether the vehicle is in a state of sudden acceleration or sudden braking.

[0035] Specifically, when the difference exceeds the dynamic confidence threshold, the specific process of generating the corresponding data credibility weight factor is as follows: design a weight decay curve based on the S-type function. When the difference exceeds the threshold, the weight factor decreases nonlinearly with the excess ratio to avoid weight jumps caused by hard judgments; introduce a time decay factor to perform exponential decay weighting on historical abnormal events to ensure that short-term anomalies do not permanently reduce the credibility of the data source; adaptively adjust the weight recovery rate according to the road type, accelerate the recovery of Beidou data weight in curved scenarios, and give priority to trusting inertial data in straight scenarios.

[0036] In this implementation scheme, the S-type function is usually used to describe a smooth nonlinear change process. It is widely used in control systems, especially for scenarios that require smooth transitions, such as weight factors. Its output value range is usually between 0 and 1, which makes it suitable for adjusting weight factors and avoiding mutations caused by hard decisions. Design a weight attenuation curve based on the S-type function: when the speed difference exceeds the dynamic confidence threshold, the weight factor will decay nonlinearly according to the S-type function. Through the gradual nature of the S-type function, the mutation caused by hard decisions is avoided, ensuring a smooth change in data credibility. Formula: Weight factor As the difference exceeds the limit ratio The change of is attenuated using an S-type function. The specific formula is: ;in: : Current time The weight factor of . : The ratio of the current speed difference exceeding the threshold. : Dynamic confidence threshold, exceeding the difference will trigger decay. : A parameter that controls the decay rate; larger values ​​result in faster decay. Introducing a time decay factor: To prevent historical anomalies from permanently impacting the credibility of the data source, a time decay factor is designed. The weight of historical anomalies decays exponentially over time, making more recent anomalies have a greater impact on data credibility, while older anomalies have a smaller impact on credibility. Formula: The weighting of historical anomalies can be modified using the following formula: ;in: :Current time The historical weight factor. : The historical weight factor at the previous time point. : Exponential decay rate, which controls the decay speed of historical abnormal events. : The time difference from the occurrence of historical abnormal events to the current moment. Adaptively adjust the weight recovery rate according to the road type: Different road scenarios have different requirements for weight recovery. In the curve scene, the vehicle's inertial navigation may produce large errors, so it is necessary to accelerate the recovery of the Beidou data weight; in the straight road scene, the inertial navigation data is more reliable, and the inertial data should be trusted first. Therefore, the recovery rate is adjusted according to the road type. The formula shows: According to the different road types, the weight recovery rate Will be adjusted. Assuming the road type is Indicates that Indicates a curve, Represents a straight line. The weight recovery rate adjustment formula is: ;in: :Current time The weight recovery rate. : Basic recovery rate, used for straight-line scenarios. : Adjustment coefficient, which determines the acceleration of the recovery rate in the curve scene. : Road type indicator (curve or straight). The final weight factor is generated by combining the above factors: The final weight factor is the result of the combined effect of the above factors. First, the current weight decay factor is calculated based on the S-type function, then the decay is adjusted according to historical anomalies, and finally the recovery rate is adjusted according to the road type to obtain the final comprehensive weight factor. The formula is: The final weight factor It can be calculated by the following formula: ;in: : The final adjusted weight factor. : Attenuation factor based on the sigmoid function. : Decay factor based on historical anomalies. : Recovery rate based on road type.

[0037] Specifically, the lane curvature information in the high-precision lane-level map is called, combined with the vehicle's real-time steering angle data, to calculate the matching error between the lane geometric constraints and the vehicle kinematic model. The specific process is as follows: discrete curvature points of the lane centerline are extracted from the high-precision map, a continuous curvature function is constructed through multiple spline interpolations, and its rate of change is calculated; based on the vehicle's real-time steering angle and longitudinal speed, the theoretical lane curvature is inferred through the kinematic model, and a tire cornering stiffness compensation term is introduced to improve model accuracy; the matching error between the actual curvature and the theoretical curvature is used as the main term of the error function, and the vehicle's lateral offset is mapped as a curvature error gain coefficient to perform a weighted correction on the error function; the penalty weight for trajectories close to the lane edge is increased to enhance the model's responsiveness to boundary constraints.

[0038] In this implementation, the discrete curvature points of the lane centerline are extracted from the high-precision map. The high-precision map contains detailed geometric information of the lane, including the discrete curvature points of the lane centerline. We extract the curvature information from these curvature points and construct a continuous curvature function through spline interpolation. Construction of the curvature function: The curvature of the lane centerline is a measure of the curvature of the curve, where Represents the arc length parameter on the curve. Through multiple spline interpolation, a smooth and continuous curvature function can be constructed: ;in: is the weight coefficient of spline interpolation; is the number of nodes (discrete points), which indicates the total number of discrete curvature points used for interpolation, that is, the degree of refinement of the centerline of the entire lane; is the spline basis function; It is the arc length coordinate of the vehicle along the lane. The theoretical lane curvature is inferred from the vehicle's real-time steering angle and longitudinal speed. and longitudinal speed , we use the vehicle's kinematic model to calculate the vehicle's theoretical lane curvature The basic relationship in the vehicle kinematic model is: ;in: is the steering angle of the vehicle; is the wheelbase of the vehicle. This theoretical lane curvature assumes an ideal vehicle steering process, where Describes the desired curvature of the vehicle. In order to improve the accuracy of the vehicle kinematic model, the tire lateral stiffness effect needs to be taken into account. That is, the tire will deviate when turning, resulting in a deviation between the theoretical curvature and the actual situation. It can be expressed by the following formula: ;in: is the tire's cornering stiffness compensation coefficient; The lateral deviation angle of the vehicle indicates the angle at which the vehicle deviates from the ideal trajectory. This compensation can make the curvature calculation more consistent with the actual vehicle motion. The matching error between the actual curvature and the theoretical curvature is calculated by comparing the actual curvature and theoretical curvature By calculating the difference between the lane geometry constraint and the vehicle kinematic model, we can calculate the matching error between the lane geometry constraint and the vehicle kinematic model: ; This error represents the degree of match between the actual driving trajectory of the vehicle and the center line of the lane. The lateral offset of the vehicle is mapped to the curvature error gain coefficient. The relationship between the lateral offset and the curvature error can be expressed by the gain factor Adjust. The calculation formula of the gain coefficient is: ;in: is the adjustment coefficient; Is the lateral offset of the vehicle on the lane. When the vehicle deviates from the center of the lane, the gain coefficient Will decrease, thus increasing the influence of lateral offset on curvature error. Weighted correction error function According to the horizontal offset Make weighted corrections to dynamically adjust the response to the error: Penalty weight and boundary constraint To improve the lane geometry constraint's responsiveness to the boundary, we introduce penalty weights into the error function. Specifically, when the vehicle approaches the lane edge, the weight is increased, making the boundary constraint more stringent. Define boundary constraint weights for: ; As the vehicle approaches the edge of the lane, the weight increases, thereby increasing the response to the boundary. The final error function is: ; This error function integrates the influence of lane geometry constraints, vehicle kinematic model, tire cornering stiffness, lateral offset and boundary constraints, and is used to measure the matching error between the vehicle and the lane.

[0039] Specifically, the process of screening candidate trajectories that meet the matching error threshold and outputting the lateral position calibration parameters and error boundaries is as follows: the tracks generated by inertial dead reckoning are segmented through a sliding window mechanism to construct candidate trajectory clusters; the weighted matching error is calculated for each candidate trajectory, and they are ranked based on the probability score, and the non-inferior trajectory clusters are screened through the Pareto front; the Gaussian mixture model is applied to the selected preferred trajectory clusters for cluster analysis, and the lateral position mean and covariance matrix of the cluster center trajectory are extracted as the calibration parameters and confidence intervals.

[0040] In this implementation, the track is segmented using a sliding window mechanism. First, the track generated by the inertial navigation system is segmented using a sliding window mechanism. Set a window size and sliding step length , slide on the track according to the window size and step size to obtain several candidate track segments. Each candidate track segment It can be expressed as: ;in: is a discrete point on the trajectory, representing the time Horizontal and vertical position at the moment; Is the number of points in the window, which determines the length of each trajectory segment. Calculate the matching error and probability score for each candidate trajectory segment , calculate the matching error based on the lane geometry constraints and the vehicle's kinematic model The matching error can be calculated based on how much the vehicle deviates from the lane centerline: ;in: is the lane curvature in the lane geometry information; The theoretical lane curvature is calculated based on the vehicle kinematic model. Then, the probability score of each candidate trajectory is calculated based on the matching error. The probability score can be obtained by normalizing the matching error: ;in: is a positive constant used to adjust the impact of matching error on the score; It is a trajectory segment The matching error is 0.000. Based on the Pareto front, the non-inferior solution trajectory cluster is screened according to the probability score. All candidate trajectories are sorted and those that meet the Pareto front are selected. The Pareto front is the set of solutions that are not dominated by other solutions under multiple objective functions. In this step, the selection of trajectory clusters is based on two criteria: matching error and probability score. By sorting, the trajectory clusters that perform best in terms of matching error and probability score are selected to generate non-inferior solution trajectory clusters. : ; Apply Gaussian mixture model to perform cluster analysis on the non-inferior solution trajectory clusters after screening Gaussian mixture model (GMM) is applied for cluster analysis. Gaussian mixture model is a probabilistic model based on multiple Gaussian distributions, which is used to identify multiple potential clusters from the data. Assuming that the distribution of each candidate trajectory cluster is a Gaussian distribution, the EM algorithm (expectation maximization algorithm) is used to estimate the mean and covariance of each cluster. is the set of non-inferior solution trajectory clusters after screening, where each Represents the state vector of a trajectory. The probability density function of the Gaussian mixture model can be expressed as: ;in: It is The weights of the Gaussian distribution; It is Gaussian distribution with mean , the covariance is ; is the set of parameters of the model; is the number of Gaussian distributions. Through cluster analysis of the Gaussian mixture model, the cluster center (i.e., mean) of each candidate trajectory cluster can be obtained. and covariance matrix , thereby extracting the mean and covariance of the lateral position as calibration parameters and confidence intervals. Output the lateral position calibration parameters and error boundaries. The mean lateral position of the cluster center trajectory and covariance matrix can be used as calibration parameters for the lateral position, and the covariance matrix Can be used to express confidence intervals for the lateral position: ; ; These parameters provide a calibrated value for the lateral position for each trajectory and provide a confidence interval for the position calibration in subsequent steps.

[0041] See also Figure 2 Specifically, when the vehicle-borne Beidou signal is available, multi-source data is fused based on the tightly coupled filtering algorithm, and the noise covariance weight of the Beidou observation value is dynamically adjusted through the weight factor to output the robust fusion positioning result. The specific process is as follows: construct a joint state vector containing the inertial device zero bias, scale factor error and Beidou receiver clock error, explicitly model the inertial device temperature drift term in the state vector, and establish a covariance correlation matrix with the Beidou clock error term; map the weight factor to the independent noise covariance component of the Beidou pseudorange and carrier phase observation values, and down-weight the low-confidence Beidou observation values ​​to achieve heterogeneous down-weighting of multiple observation sources; construct a cost function through the robust M-estimator, re-weight the filter residual iteratively, and suppress the influence of multi-path abnormal observations.

[0042] In this implementation, a joint state vector is constructed and the temperature drift term is explicitly modeled. First, a joint state vector is constructed, which includes the bias of the inertial sensor, the scale factor error, and the clock error of the BeiDou receiver. The joint state vector x can be expressed as: ;in: Indicates the zero bias of the inertial sensor (gyroscope zero bias and accelerometer bias ); Indicates the clock error of the Beidou receiver; Represents the temperature drift term. Covariance correlation matrix Describes the error relationship between each state quantity: ; Mapping of weight factors and noise covariance adjustment Next, map the weight factors The noise covariance matrix of the BeiDou pseudorange and carrier phase observations is obtained. The weight factor is dynamically adjusted according to the credibility of the BeiDou observations, and the low-credibility observations will be weighted less. Assume that the noise covariance of the BeiDou pseudorange is , the noise covariance of the carrier phase is , then the covariance adjusted based on the weight factor is: ; ;in: and denote the raw noise covariance of pseudorange and carrier phase respectively; It is a dynamically adjusted weight factor that changes with the reliability of the BeiDou signal. The dynamic adjustment of the weight factor can be made according to the actual BeiDou signal quality. For example, when the signal is weak or the multipath effect is strong, the weight factor The weight of the low-confidence observation value will be reduced, thereby reducing its influence on the filtering result. The BeiDou observation value with low confidence will be dynamically downgraded. Specifically, the low-confidence observation value will be downgraded by the weight factor to reduce its influence on the final fusion result. At this time, the weight factor Adjustments will be made according to the following rules: ;in: is a regulating factor used to adjust the relationship between signal quality and weight; is the error of BeiDou observation value (bias estimated by multipath effect). Reweighted iteration is performed by robust M-estimator. In order to further suppress the influence of multipath effect and abnormal observation, the M-estimator is used to reweight the filter residual. The M-estimator is constructed by defining the cost function , for the residual Weighted to reduce the impact of abnormal observations. Cost function This can take the form of, for example, the Huber function or the Tukey function: ;in: is the filtering residual; is a constant that determines the saturation threshold of the cost function. At each iteration, the filter residual is reweighted to suppress the influence of large residual values. The final filter result is optimized using weighted least squares.

[0043] Specifically, when the continuous loss of the on-board Beidou signal exceeds a preset threshold, it switches to a calibration mode driven by lane geometry constraints. The specific process of suppressing the accumulated inertial navigation error based on candidate trajectories and error boundaries is as follows: in the early stage of Beidou signal failure, a lane-binding corridor is generated based on historical positioning results and high-precision maps to constrain the lateral divergence of inertial dead reckoning; through an improved residual resampling particle filter, lane curvature constraints are injected into the proposal distribution to force particles to be distributed along a reasonable lane centerline; the calibrated zero-bias estimate is reversely injected into the tightly coupled filter state vector to achieve smooth mode switching when the signal is restored and avoid positioning jumps.

[0044] In this implementation, lane-binding corridors are generated in the early stage of Beidou signal failure by combining historical positioning results with high-precision map data. A lane-binding corridor is a constraint area that limits lateral divergence and is usually defined by the lane centerline and lane width. Assume that the discrete points of the lane centerline are , where each point represents the coordinates of the lane center, and the lane width is , then the lane-bound corridor can be expressed as: ;in: It is the zth discrete point of the lane centerline, representing the coordinate point of the lane center on the two-dimensional plane. is the width of the lane. Represents the constructed lane-bound corridor area, which is a set of lateral constraints for the vehicle's inertial position. It represents the lateral offset of a point on the lane centerline, and its range is limited to [−w / 2, w / 2] to ensure that the offset point is still within the lane range. represents a set of ordered pairs of two real numbers, i.e., the Euclidean two-dimensional plane, indicating is a two-dimensional position point. At this time, the lateral divergence of the inertial navigation system will be constrained by the lane corridor to ensure that the vehicle travels along the lane as much as possible during the calculation process and reduce the lateral deviation. Residual resampling by improved particle filter In order to further suppress the error accumulation of the inertial navigation system, an improved residual resampling particle filter is adopted. This filtering method forces the particles to be distributed along a reasonable lane centerline by injecting lane curvature constraints into the proposed distribution. The core idea of ​​the particle filter algorithm is to estimate the state of the system based on the position and weight of each particle. Here, each particle represents a possible position of the vehicle, and the update process can be described by the following steps: Particle state: Particle Indicates the The state vector of a particle at the current moment k, including position and velocity information. Prediction step: Predict the particle state based on the vehicle's kinematic model: ;in: is the kinematic model of the vehicle; is the control input (such as steering angle, vehicle speed, etc.); Is the process noise. Inject lane curvature constraint: In the predicted particle state, inject lane curvature information to ensure that particles are distributed along the lane centerline. Assume that the lane curvature is , lane geometry constraints need to be satisfied when particles are updated: ; Resampling step: By weighted sampling, more likely particles are selected and particles that do not meet the constraints are eliminated. Eventually, after multiple resampling and updating steps, the particles will gather near reasonable lane positions, suppressing the cumulative error of the inertial navigation system. Back-injection of calibrated zero-bias estimate During the calibration process, the zero-bias estimate of the inertial navigation system is back-injected into the state vector of the tightly coupled filter to provide a corrected state estimate. Assume that the state vector of the tightly coupled filter is , the zero-bias estimate is , then the calibrated state vector is: ; In this process, the bias estimate Indicates the deviation in the inertial navigation system caused by sensor errors or other factors. Smooth Mode Switching When the Beidou signal is restored, the system needs to smoothly switch from lane geometry constraint mode back to inertial navigation system mode. At this time, the smooth switching mechanism of the state vector is used to avoid positioning jumps during the system switching. The smooth mode transition can be expressed by the following formula: ;in: represents the state vector after the slip, which ultimately combines the estimation of the inertial navigation system and the lane geometry constraints. α is the smoothing coefficient, which controls how smoothly the system transitions; is the calibrated inertial navigation state vector. When the BeiDou signal is restored and positioning accuracy is high, the value of α will gradually increase, transitioning to the original filtered result of the inertial navigation system. When the BeiDou signal quality is poor, the value of α is kept low, and positioning continues to rely on lane geometry constraints.

[0045] In summary, this application has at least the following effects:

[0046] A vehicle-mounted Beidou positioning deviation self-calibration system that integrates inertial navigation information effectively integrates inertial navigation and Beidou observation data through spatiotemporal reference alignment and a tightly coupled filtering algorithm, improving positioning accuracy and continuity in complex scenarios. Based on kinematic constraints and dynamic adjustment of weight factors, it suppresses interference from abnormal observation values ​​and improves the system's robustness under signal obstruction or multipath effects. Through lane geometry constraints and error boundary analysis, it uses high-precision maps to correct inertial navigation drift when Beidou signals are lost, significantly reducing long-term positioning deviations. Combining real-time signal quality with vehicle motion status, it autonomously switches fusion strategies and calibration modes to ensure stable output in extreme environments such as urban canyons and tunnels. Through a non-holonomic constraint model and candidate trajectory screening mechanism, it reduces redundant computation and enables efficient real-time processing on the vehicle embedded platform.

[0047] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may be implemented entirely in hardware, entirely in software, or in a combination of software and hardware. Furthermore, the present invention may be implemented as a computer program product embodied in one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0048] The present invention is described with reference to flowcharts and / or block diagrams of systems, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0049] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0050] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0051] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0052] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A vehicle-mounted Beidou positioning deviation self-calibration system integrating inertial navigation information, characterized in that: It includes the following modules: multi-source data synchronization and alignment module, kinematic constraint consistency analysis module, lane geometry constraint analysis module, and adaptive fusion and calibration module; The multi-source data synchronization and alignment module is used to synchronize the three-axis acceleration and three-axis angular velocity data output by the inertial navigation system, and the pseudo-range observation values ​​and carrier phase observation values ​​output by the Beidou satellite navigation system based on timestamp interpolation, and realize spatial reference alignment through coordinate system conversion, and output a multi-source data stream with a unified time and space reference; The kinematic constraint consistency analysis module is used to calculate the difference between the inertial navigation system-derived speed and the Beidou system-derived speed based on the Doppler shift solution based on the vehicle's nonholonomic constraint model, and generate a corresponding data credibility weight factor when the difference exceeds a dynamic confidence threshold; The lane geometry constraint analysis module is used to call the lane curvature information in the high-precision lane-level map, combine it with the vehicle's real-time steering angle data, calculate the matching error between the lane geometry constraint and the vehicle kinematic model, screen candidate trajectories that meet the matching error threshold, and output lateral position calibration parameters and error boundaries; The adaptive fusion and calibration module is used to fuse multi-source data based on a tightly coupled filtering algorithm when the on-board Beidou signal is available, dynamically adjust the noise covariance weight of the Beidou observation value through a weight factor, and output a robust fusion positioning result. When the on-board Beidou signal is continuously lost for more than a preset threshold, it switches to a calibration mode driven by lane geometry constraints, suppresses the accumulated error of inertial navigation based on candidate trajectories and error boundaries, and outputs continuous and stable position information; The timing synchronization of the multi-source data synchronization and alignment module includes the following: Perform timestamp interpolation alignment on the high-frequency three-axis acceleration and angular velocity data output by the inertial navigation system, and the low-frequency pseudorange and carrier phase observations output by the Beidou satellite navigation system, eliminating timing mismatches caused by sampling rate differences. Through the dynamic delay compensation mechanism, the fixed delay and random jitter in the data transmission process are corrected to ensure the timing consistency of multi-source data streams; Conflict detection is performed on the interpolated time series to identify abnormal segments with reversed timestamps and exceeded intervals, and local resampling and repair are performed through adjacent valid data segments; The specific process of calculating the difference between the inertial navigation system-derived velocity and the BeiDou system-derived velocity based on Doppler shift using the vehicle's nonholonomic constraint model is as follows: According to the Ackermann steering geometry, the theoretical constraint equation of the lateral velocity is constructed and the theoretical value of the lateral velocity calculated by the inertial navigation system is derived. Perform rotation transformation from the carrier coordinate system to the navigation coordinate system on the velocity calculated by BeiDou Doppler frequency shift, and compensate for the antenna installation angle error; The vehicle acceleration confidence interval is introduced to dynamically soften the confidence threshold of the speed difference: Expand the threshold in sudden acceleration or braking scenarios, and tighten the threshold in constant speed scenarios; When the difference exceeds the dynamic confidence threshold, the specific process of generating the corresponding data credibility weight factor is as follows: A weight decay curve based on an S-shaped function is designed. When the difference exceeds the threshold, the weight factor decreases nonlinearly with the excess ratio, avoiding weight jumps caused by hard decisions. Introducing a time decay factor to perform exponential decay weighting on historical anomalies to ensure that short-term anomalies do not permanently reduce the credibility of the data source; Adaptively adjust the weight recovery rate based on road type, accelerate the recovery of Beidou data weight in curved scenarios, and prioritize inertial data in straight road scenarios; The specific process of calling the lane curvature information in the high-precision lane-level map and combining it with the vehicle's real-time steering angle data to calculate the matching error between the lane geometry constraints and the vehicle kinematic model is as follows: Extract discrete curvature points of the lane centerline from a high-precision map, construct a continuous curvature function through multiple spline interpolations, and calculate its rate of change; Based on the vehicle's real-time steering angle and longitudinal speed, the theoretical lane curvature is inferred through a kinematic model, and a tire cornering stiffness compensation term is introduced to improve model accuracy. The matching error between the actual curvature and the theoretical curvature is used as the main term of the error function, and the vehicle lateral offset is mapped to the curvature error gain coefficient to perform weighted correction on the error function. Increase the penalty weight for trajectories close to the lane edge to enhance the model's responsiveness to boundary constraints.

2. The vehicle-mounted Beidou positioning deviation self-calibration system integrating inertial navigation information according to claim 1 is characterized in that: The specific process of screening candidate trajectories that meet the matching error threshold and outputting the lateral position calibration parameters and error boundaries is as follows: The trajectory generated by dead reckoning is segmented through a sliding window mechanism to construct candidate trajectory clusters. Calculate the weighted matching error for each candidate trajectory, sort them based on the probability score, and filter the non-inferior solution trajectory cluster through the Pareto front; The Gaussian mixture model was applied to the selected optimal trajectory clusters for cluster analysis, and the lateral position mean and covariance matrix of the cluster center trajectory were extracted as calibration parameters and confidence intervals.

3. The vehicle-mounted Beidou positioning deviation self-calibration system integrating inertial navigation information according to claim 2 is characterized in that: When the vehicle-mounted BeiDou signal is available, the multi-source data is fused based on the tightly coupled filtering algorithm, and the noise covariance weight of the BeiDou observation value is dynamically adjusted through the weight factor. The specific process of outputting the robust fusion positioning result is as follows: Construct a joint state vector that includes the inertial device zero bias, scale factor error, and BeiDou receiver clock error. Explicitly model the inertial device temperature drift term in the state vector and establish a covariance correlation matrix with the BeiDou clock error term. The weight factors are mapped to the independent noise covariance components of BeiDou pseudorange and carrier phase observations, and low-credibility BeiDou observations are down-weighted to achieve heterogeneous down-weighting of multiple observation sources. The cost function is constructed by using the robust M-estimator, and the filtering residual is iteratively reweighted to suppress the influence of multipath abnormal observations.

4. The vehicle-mounted Beidou positioning deviation self-calibration system integrating inertial navigation information according to claim 3 is characterized in that: When the vehicle-mounted BeiDou signal is continuously lost for more than a preset threshold, the system switches to the lane geometry constraint-driven calibration mode. The specific process of suppressing the accumulated inertial navigation error based on the candidate trajectory and error boundary is as follows: In the early stages of Beidou signal failure, lane binding corridors are generated based on historical positioning results and high-precision maps to constrain the lateral divergence of inertial dead reckoning. Through the improved residual resampling particle filter, the lane curvature constraint is injected into the proposal distribution to force the particles to be distributed along the reasonable lane centerline; The calibrated zero bias estimate is reversely injected into the tightly coupled filter state vector to achieve smooth mode switching when the signal is restored, thus avoiding positioning jumps.

Images