Abnormal positioning data detection method and system based on mileage calculation

Abstract

The invention discloses an abnormal positioning data detection method and system based on mileage reckoning, and the method comprises the steps: obtaining the satellite positioning data and multi-dimensional odometer observation data of a target carrier in each epoch; obtaining odometer environment error data, a carrier attitude estimation error, a point cloud position estimation error and an odometer estimation error of the target carrier in each epoch according to the carrier position data and the point cloud feature position data, and obtaining a position error confidence threshold value of the target carrier in each epoch in combination with a pre-obtained odometer inherent distance measurement error; and finally, obtaining the position deviation between the mileage positioning data of each epoch and the satellite positioning data, so as to obtain an abnormal data detection result of the satellite positioning data according to the position deviation and a position error confidence threshold, thereby improving the accuracy of abnormal detection.

Description

Technical Field

[0001] This invention relates to the field of positioning, and specifically to a method and system for detecting abnormal positioning data based on mileage estimation. Background Technology

[0002] Currently, positioning using a combined navigation method based on satellite signal receivers, odometry IMUs (inertial measurement units), and lidar is the mainstream solution for fusion positioning technology. This solution can achieve high positioning accuracy and stability under normal conditions.

[0003] However, when the satellite signal receiver is in a scenario with deception or interference (such as malicious signal tampering, electromagnetic interference suppression, etc.), the positioning results will be severely inaccurate or even fail. Among them, although the odometry IMU can provide position estimation data by sensing motion to maintain positioning output in the short term, it will be affected by the accumulation of errors over a long period of time, and the positioning accuracy will continue to decline. While lidar has the ability to perceive three-dimensional environmental features and is not easily affected by electromagnetic interference, and can assist in positioning through scene feature matching, it is difficult to achieve long-distance high-precision positioning on its own. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention discloses an anomaly detection method and system based on mileage estimation, which improves the accuracy of anomaly detection in satellite positioning data.

[0005] To achieve the above objectives, this invention discloses an anomaly location data detection method based on mileage estimation, comprising: The satellite positioning data and multidimensional odometry observation data of the target vehicle at each epoch are acquired; wherein, the multidimensional odometry observation data includes odometer positioning data, vehicle position data, point cloud feature position data and posterior estimation error covariance matrix; Based on the carrier location data and the point cloud feature location data, obtain the odometry environmental error data of the target carrier at each epoch; The carrier attitude estimation error of the target carrier in each epoch is obtained based on the carrier position data corresponding to two adjacent epochs and the posterior estimation error covariance matrix. The point cloud position estimation error of the target carrier in each epoch is obtained based on the carrier attitude estimation error and the point cloud feature position data corresponding to two adjacent epochs. The odometer estimation error of the target vehicle in each epoch is obtained based on the posterior estimation error covariance matrix. The position error confidence threshold of the target carrier in each epoch is obtained based on the point cloud position estimation error, the odometer estimation error, the odometer environmental error data, and the pre-acquired odometer inherent ranging error. The positional deviation between the mileage positioning data and the satellite positioning data at each epoch is obtained, and the abnormal data detection result of the satellite positioning data is obtained based on the positional deviation and the position error confidence threshold.

[0006] This invention discloses an anomaly positioning data detection method based on odometer estimation. It acquires satellite positioning data and multi-dimensional odometry observation data of the target vehicle at each epoch, laying the foundation for subsequent error quantification based on the multi-dimensional odometry observation data. Next, it obtains odometry environmental error data based on the vehicle's position data and point cloud feature position data, quantifying the direct impact of environmental geometry on odometry accuracy by comparing the target position with environmental feature positions. It then obtains the vehicle attitude estimation error based on the vehicle position data of two adjacent epochs and the posterior estimation error covariance matrix, estimating the cumulative attitude error to address short-term odometer drift by utilizing the position changes and error statistical characteristics of adjacent time points. Finally, it calculates the odometry environmental error based on the vehicle attitude estimation error and the point cloud feature position data of two adjacent epochs. The system obtains point cloud position estimation errors from cloud feature location data, and combines attitude errors and dynamic changes in environmental features to improve the accuracy of point cloud matching and reduce environmental interference. It obtains odometer estimation errors based on the posterior estimation error covariance matrix, directly using the error matrix to simplify the odometer error extraction process. Based on point cloud position estimation errors, odometer estimation errors, odometer environmental error data, and pre-acquired inherent odometer ranging errors, it obtains position error confidence thresholds, integrates multiple error sources and environmental factors, and dynamically sets reliable thresholds to enhance detection robustness. It acquires the position deviation between odometer positioning data and satellite positioning data at each epoch, and obtains abnormal data detection results based on the position error confidence threshold. Real-time comparison enables accurate identification of satellite positioning anomalies.

[0007] As a preferred example, obtaining the odometry environmental error data of the target carrier at each epoch based on the carrier location data and the point cloud feature location data includes: For any given epoch, obtain the position data corresponding to each point cloud feature from the point cloud feature position data corresponding to that epoch; For any point cloud feature, the direction vector of the point cloud feature relative to the target carrier is obtained based on the location data corresponding to the point cloud feature and the carrier location data corresponding to the epoch. Construct the direction cosine matrix of the target carrier in the epoch based on several direction vectors corresponding to the point cloud feature location data; Obtain the covariance matrix of the direction cosine matrix, and perform matrix inversion on the covariance matrix to obtain the weight coefficient matrix; The sum of several matrix elements on the main diagonal of the weight coefficient matrix is ​​obtained, and the square root operation is performed on the sum to obtain the odometer environmental error data of the target vehicle in the epoch.

[0008] The above scheme provides fundamental information about environmental features by acquiring the location data of each point cloud feature, ensuring that calculations are based on actual observation data and avoiding error estimation bias due to missing information. Secondly, it calculates direction vectors based on the point cloud feature location data and the carrier location data, quantifying the direction of each feature relative to the carrier, directly reflecting the local influence of environmental geometry, and enabling direction information to accurately capture environmental structure. Next, it constructs a direction cosine matrix based on multiple direction vectors, integrating direction distribution characteristics to achieve statistical modeling of the overall environmental geometry, enhancing adaptability to environmental complexity. Then, it obtains the covariance matrix of the direction cosine matrix and inverts it to obtain the weight coefficient matrix. Statistical methods are used to evaluate the importance of different directions; the weight coefficient matrix can more accurately represent the uncertainty and noise impact of environmental geometry, improving the robustness of error estimation. Finally, it calculates the sum of the diagonal elements of the weight coefficient matrix and takes the square root, integrating the influence of all directions to generate a single environmental error data value, ensuring comprehensive and accurate quantification results, thus directly addressing the challenge of accurately measuring the influence of environmental geometry.

[0009] As a preferred example, the step of obtaining the carrier attitude estimation error of the target carrier in each epoch based on the carrier position data corresponding to two adjacent epochs and the posterior estimation error covariance matrix includes: For any given epoch, acquire the historical epochs adjacent to the given epoch and the historical carrier location data of the target carrier in the historical epoch; Based on the target carrier's position data in the epoch and the historical carrier position data, the displacement of the target carrier in the epoch is obtained; The product of the displacement and the pre-acquired empirical constant coefficient is used as the scaling factor for the carrier attitude estimation error. The original attitude error of the target carrier in the epoch is obtained from the posterior estimation error covariance matrix corresponding to the epoch, and the original attitude error is scaled into a real-time attitude error according to the carrier attitude estimation error scaling factor and a preset spherical interpolation method. When it is determined that the real-time attitude error has a vertical observation constraint based on the posterior estimation error covariance matrix of the epoch and the historical posterior estimation error covariance matrix of the historical epoch, the real-time attitude error is used as the carrier attitude estimation error of the target carrier in each epoch.

[0010] The above scheme obtains historical epoch and historical carrier position data, allowing for the quantification of motion trends based on historical changes, thereby reducing error accumulation. Displacement data obtained from current and historical carrier position data accurately reflects the actual motion amplitude of the carrier, providing an objective basis for error adjustment. Multiplying the displacement by empirical constants generates scaling factors, taking into account the impact of motion intensity on error; these empirical factors, based on prior knowledge, ensure the reasonableness of scaling. After obtaining the original attitude error from the posterior estimation error covariance matrix, it is scaled to the real-time attitude error using scaling factors and spherical interpolation. Spherical interpolation achieves smooth adjustment in attitude space, avoiding linear distortion. Finally, when the real-time attitude error is subject to vertical observation constraints based on the posterior estimation error covariance matrix and the historical matrix, this error is adopted. The vertical observation constraints verify the reliability of the error, ensuring more accurate attitude estimation and thus improving the robustness of subsequent positioning and detection.

[0011] As a preferred example, the step of obtaining the carrier attitude estimation error of the target carrier in each epoch based on the carrier position data corresponding to two adjacent epochs and the posterior estimation error covariance matrix further includes: When it is determined that the real-time attitude error does not have the vertical observation constraint, the historical attitude estimation error of the target vehicle in the historical epoch is obtained. The product of the historical carrier attitude estimation error and the real-time attitude error is taken as the carrier attitude estimation error of the target carrier in the given epoch.

[0012] The above scheme identifies conditions requiring additional processing by determining that the real-time attitude error lacks vertical observation constraints, thus avoiding estimation bias caused by directly ignoring unconstrained scenarios. Next, it obtains the historical carrier attitude estimation error at a historical epoch. This utilizes historical data from the time series, compensating for the deficiencies of real-time data under unconstrained conditions and ensuring the continuity of error estimation. Finally, the product of the historical carrier attitude estimation error and the real-time attitude error is used as the carrier attitude estimation error for the current epoch. This method integrates historical cumulative error and real-time state error through a product operation, enhancing the comprehensiveness of error estimation and improving the stability and accuracy of attitude estimation even without vertical constraints. Specifically, it introduces past error information based on the historical carrier attitude estimation error to prevent errors from being ignored; it reflects the current dynamics based on the real-time attitude error; and it achieves synergistic optimization of both through product, simplifying the calculation process and enhancing robustness.

[0013] As a preferred example, the step of obtaining the point cloud position estimation error of the target carrier in each epoch based on the carrier attitude estimation error and the point cloud feature position data corresponding to two adjacent epochs includes: For any given epoch, acquire the historical point cloud feature location data of the historical epochs adjacent to the given epoch and the target carrier in the historical epoch. The position data of each point cloud feature is obtained from the point cloud feature position data corresponding to the epoch, and the historical position data of each historical point cloud feature is obtained from the historical point cloud feature position data. New point cloud features are obtained from a plurality of point cloud features based on the comparison results between the historical location data and the location data. Obtain the direction vector corresponding to the newly added point cloud feature and the historical direction vector corresponding to each of the historical point cloud features, so as to match the target historical point cloud feature corresponding to the newly added point cloud feature from multiple historical point cloud features according to the angle between the direction vector and the historical direction vector. The displacement vector between the newly added point cloud feature and the target historical point cloud feature is obtained based on the position data corresponding to the newly added point cloud feature and the historical position data corresponding to the target historical point cloud feature. Based on the displacement vector and the carrier attitude estimation error corresponding to the epoch, the real-time error corresponding to the newly added point cloud feature is obtained; The historical position error confidence threshold of the target carrier at the historical epoch is obtained, and the sum of the historical position error confidence threshold and the real-time error is used as the cumulative error corresponding to the newly added point cloud feature. A number of cumulative errors are obtained for the epoch, and these cumulative errors are used as the point cloud position estimation error of the target carrier in the epoch.

[0014] The above scheme provides a benchmark reference for obtaining historical epoch and historical point cloud feature location data, ensuring the reliability of change detection; extracting information from current and historical location data allows for the identification of newly added point cloud features, avoiding the omission of dynamic changes; obtaining newly added point cloud features based on the comparison results of historical and location data, directly filtering for feature additions and subtractions, reducing misjudgments; obtaining direction vectors and matching target historical point cloud features based on the included angle, using geometric relationships to ensure accurate feature correspondence and reduce matching errors; obtaining displacement vectors based on location data, quantifying feature movement, and providing a basis for error calculation; combining displacement vectors and carrier attitude estimation errors to obtain real-time errors, incorporating attitude influences, and more accurately reflecting the uncertainty of new features; obtaining historical location error confidence thresholds and calculating cumulative errors, inheriting historical error data to achieve error accumulation control; finally, obtaining the cumulative error as the point cloud location estimation error, outputting a comprehensive value, and enhancing robustness in dynamic environments.

[0015] As a preferred example, obtaining the odometry estimation error of the target vehicle in each epoch based on the posterior estimation error covariance matrix includes: For any given epoch, the position estimation variance of the target carrier in each coordinate axis direction and the square root of each position estimation variance are obtained from the posterior estimation error covariance matrix corresponding to that epoch. The multiple square roots are combined into a multidimensional vector, which is then used as the odometry estimation error of the target carrier in the epoch.

[0016] The above scheme obtains the position estimation variance of the target vehicle in each coordinate axis direction from the posterior estimation error covariance matrix. This utilizes the existing variance data in the covariance matrix to directly extract the uncertainty information of the position estimation, avoiding the accumulation of errors caused by additional calculations. Next, the square root of each position estimation variance is calculated, converting the variance into a standard deviation. This more intuitively represents the actual magnitude of the error, as the standard deviation is the square root of the variance and more accurately reflects the error range. Then, multiple square roots are combined into a multidimensional vector. This preserves the distribution characteristics of the error in different coordinate axis directions, making the odometry estimation error multidimensional and able to comprehensively capture the anisotropic characteristics of spatial errors. Finally, using the multidimensional vector as the odometry estimation error provides a standardized mathematical representation, facilitating its direct use in the calculation of the position error confidence threshold, thereby improving the accuracy and reliability of anomaly location data detection.

[0017] As a preferred example, the inherent ranging error of the odometer includes radar ranging error and observation noise corresponding to each point cloud feature; The step of obtaining the position error confidence threshold of the target carrier in each epoch based on the point cloud position estimation error, the odometer estimation error, the odometer environmental error data, and the pre-acquired inherent odometer ranging error includes: For any epoch, obtain the first product of the odometer environmental error data and the radar ranging error for that epoch; Obtain the sum of the observation noise of all newly added point cloud features within the epoch; For any newly added point cloud feature in the epoch, obtain the ratio of the observation noise of the newly added point cloud feature to the sum of the noise, and obtain the second product of the ratio and the accumulated error of the newly added point cloud feature. The sum of all the second products within the epoch is obtained to obtain the feature observation error corresponding to the epoch. Obtain the modulus of the odometry estimation error at the epoch, and obtain the position error confidence threshold of the target vehicle at the epoch based on the sum of the modulus, the first product, and the feature observation error.

[0018] The above scheme obtains the first product of the odometer environmental error data and the radar ranging error, taking into account the amplification effect of the environment on radar error, making the calculation more in line with the actual scenario. The sum of the observation noise of all newly added point cloud features within the epoch is obtained, aggregating feature-related random noise and providing a basis for subsequent weight allocation. For any newly added point cloud feature in the epoch, the ratio of the observation noise of the newly added point cloud feature to the sum is obtained, and the second product of this ratio and the accumulated error of the newly added point cloud feature is obtained. This assigns a weight to each feature based on its noise contribution, multiplied by the accumulated error, highlighting the error impact of high-noise features and ensuring that threshold calculation is more targeted. The sum of all second products within the epoch is obtained to obtain the feature observation error corresponding to that epoch, which quantifies the overall uncertainty of feature matching and integrates the error contributions of all features. The modulus of the odometry estimation error at that epoch is obtained. Based on the sum of the modulus, the first product, and the characteristic observation error, the position error confidence threshold is obtained. This integrates odometry motion error and environmental geometric error to generate a more reliable confidence boundary for accurately judging anomalies in satellite positioning data.

[0019] As a preferred example, the step of obtaining the position deviation between the mileage positioning data and the satellite positioning data for each epoch, and obtaining the abnormal data detection result of the satellite positioning data based on the position deviation and the position error confidence threshold, includes: For any given epoch, obtain the positional deviation between the mileage positioning data and the satellite positioning data for that epoch; When the position deviation is greater than the position error confidence threshold, the satellite positioning data corresponding to that epoch is determined to be abnormal positioning data.

[0020] The above scheme, for any epoch, acquires the positional deviation between the odometer data and the satellite positioning data for that epoch. This step calculates the deviation in real time at each time point, enabling targeted detection across all epochs and avoiding inaccuracies caused by the accumulation of overall errors, thus enhancing the timeliness and specificity of the detection. Then, when the positional deviation exceeds a positional error confidence threshold, the satellite positioning data corresponding to that epoch is determined to be abnormal positioning data. This determination logic uses a pre-calculated positional error confidence threshold as an objective standard, directly comparing the deviation with the threshold, ensuring the quantification of the determination criteria and avoiding misjudgments caused by subjective or vague judgments. Furthermore, considering that the confidence threshold originates from error estimation, it improves the robustness of detection in deception or interference scenarios. Overall, by implementing deviation acquisition and threshold comparison step by step, the operability and accuracy of the detection mechanism are enhanced.

[0021] On the other hand, the present invention discloses an abnormal positioning data detection system based on mileage estimation, including a data acquisition module, an environmental error module, a carrier error module, a point cloud error module, a mileage error module, a position error module, and a data detection module; The data acquisition module is used to acquire satellite positioning data and multidimensional odometry observation data of the target carrier at each epoch; wherein, the multidimensional odometry observation data includes odometer positioning data, carrier position data, point cloud feature position data, and posterior estimation error covariance matrix; The environmental error module is used to obtain the odometry environmental error data of the target carrier at each epoch based on the carrier location data and the point cloud feature location data. The carrier error module is used to obtain the carrier attitude estimation error of the target carrier in each epoch based on the carrier position data corresponding to two adjacent epochs and the posterior estimation error covariance matrix. The point cloud error module is used to obtain the point cloud position estimation error of the target carrier in each epoch based on the carrier attitude estimation error and the point cloud feature position data corresponding to two adjacent epochs. The odometer error module is used to obtain the odometer estimation error of the target vehicle in each epoch based on the posterior estimation error covariance matrix. The position error module is used to obtain the position error confidence threshold of the target carrier in each epoch based on the point cloud position estimation error, the odometer estimation error, the odometer environmental error data, and the pre-acquired odometer inherent ranging error. The data detection module is used to obtain the position deviation between the mileage positioning data and the satellite positioning data at each epoch, so as to obtain the abnormal data detection result of the satellite positioning data based on the position deviation and the position error confidence threshold.

[0022] This invention discloses an anomaly positioning data detection system based on odometer estimation. It acquires satellite positioning data and multi-dimensional odometry observation data of the target vehicle at each epoch, laying the foundation for subsequent error quantification based on the multi-dimensional odometry observation data. Next, it obtains odometry environmental error data based on the vehicle's position data and point cloud feature position data, quantifying the direct impact of environmental geometry on odometry accuracy by comparing the target position with environmental feature positions. It then obtains the vehicle attitude estimation error based on the vehicle position data of two adjacent epochs and the posterior estimation error covariance matrix, estimating the cumulative attitude error to address short-term odometer drift by utilizing the position changes and error statistical characteristics of adjacent time points. Finally, it calculates the odometry environmental error based on the vehicle attitude estimation error and the point cloud feature position data of two adjacent epochs. The system obtains point cloud position estimation errors from cloud feature location data, and combines attitude errors and dynamic changes in environmental features to improve the accuracy of point cloud matching and reduce environmental interference. It obtains odometer estimation errors based on the posterior estimation error covariance matrix, directly using the error matrix to simplify the odometer error extraction process. Based on point cloud position estimation errors, odometer estimation errors, odometer environmental error data, and pre-acquired inherent odometer ranging errors, it obtains position error confidence thresholds, integrates multiple error sources and environmental factors, and dynamically sets reliable thresholds to enhance detection robustness. It acquires the position deviation between odometer positioning data and satellite positioning data at each epoch, and obtains abnormal data detection results based on the position error confidence threshold. Real-time comparison enables accurate identification of satellite positioning anomalies.

[0023] As a preferred example, the environmental error module includes a direction recognition unit and an environmental influence unit; The direction recognition unit is used to, for any epoch, obtain the position data corresponding to each point cloud feature from the point cloud feature position data corresponding to the epoch; and for any point cloud feature, obtain the direction vector of the point cloud feature relative to the target carrier based on the position data corresponding to the point cloud feature and the carrier position data corresponding to the epoch. The environmental impact unit is used to construct a direction cosine matrix of the target vehicle in the epoch based on several direction vectors corresponding to the point cloud feature location data; obtain the covariance matrix of the direction cosine matrix, and perform matrix inversion on the covariance matrix to obtain the weight coefficient matrix; obtain the sum of several matrix elements on the main diagonal of the weight coefficient matrix, and perform square root operation on the sum to obtain the odometry environmental error data of the target vehicle in the epoch.

[0024] The above scheme provides fundamental information about environmental features by acquiring the location data of each point cloud feature, ensuring that calculations are based on actual observation data and avoiding error estimation bias due to missing information. Secondly, it calculates direction vectors based on the point cloud feature location data and the carrier location data, quantifying the direction of each feature relative to the carrier, directly reflecting the local influence of environmental geometry, and enabling direction information to accurately capture environmental structure. Next, it constructs a direction cosine matrix based on multiple direction vectors, integrating direction distribution characteristics to achieve statistical modeling of the overall environmental geometry, enhancing adaptability to environmental complexity. Then, it obtains the covariance matrix of the direction cosine matrix and inverts it to obtain the weight coefficient matrix. Statistical methods are used to evaluate the importance of different directions; the weight coefficient matrix can more accurately represent the uncertainty and noise impact of environmental geometry, improving the robustness of error estimation. Finally, it calculates the sum of the diagonal elements of the weight coefficient matrix and takes the square root, integrating the influence of all directions to generate a single environmental error data value, ensuring comprehensive and accurate quantification results, thus directly addressing the challenge of accurately measuring the influence of environmental geometry. Attached Figure Description

[0025] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0026] Figure 1 This is a flowchart illustrating an abnormal location data detection method based on mileage estimation disclosed in an embodiment of the present invention. Figure 2 This is a flowchart illustrating an anomaly location data detection system based on mileage estimation disclosed in an embodiment of the present invention. Figure 3 This is a flowchart illustrating an abnormal location data detection method based on mileage estimation, as disclosed in another embodiment of the present invention. Figure 4 This is a side view diagram of the trajectory of a simulation experiment disclosed in another embodiment of the present invention; Figure 5 This is a schematic diagram of the position deviation curve of a simulation experiment disclosed in another embodiment of the present invention; Figure 6 This is a schematic diagram of the post-optimization curve of a simulation experiment disclosed in another embodiment of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Example 1 Reference Figure 1 To improve the accuracy of anomaly detection in satellite positioning data, this embodiment discloses an anomaly detection method based on mileage estimation, which mainly includes: Step 101: Obtain satellite positioning data and multidimensional odometry observation data of the target vehicle at each epoch; wherein, the multidimensional odometry observation data includes mileage positioning data, vehicle position data, point cloud feature position data, and posterior estimation error covariance matrix.

[0029] Step 102: Obtain the odometry environmental error data of the target carrier at each epoch based on the carrier location data and the point cloud feature location data.

[0030] In this embodiment, the steps mainly include: for any epoch, obtaining the position data corresponding to each point cloud feature from the point cloud feature position data corresponding to the epoch; for any point cloud feature, obtaining the direction vector of the point cloud feature relative to the target carrier based on the position data corresponding to the point cloud feature and the carrier position data corresponding to the epoch; constructing the direction cosine matrix of the target carrier in the epoch based on the several direction vectors corresponding to the point cloud feature position data; obtaining the covariance matrix of the direction cosine matrix, and performing matrix inversion on the covariance matrix to obtain the weight coefficient matrix; obtaining the sum of several matrix elements on the main diagonal of the weight coefficient matrix, and performing square root operation on the sum to obtain the odometry environmental error data of the target carrier in the epoch.

[0031] In this embodiment, the above steps provide basic information about environmental features by acquiring the location data of each point cloud feature, ensuring that the calculation is based on actual observation data and avoiding error estimation bias due to missing information. Secondly, direction vectors are calculated based on the point cloud feature location data and the carrier location data, quantifying the direction of each feature relative to the carrier and directly reflecting the local influence of environmental geometry, enabling the direction information to accurately capture the environmental structure. Next, a direction cosine matrix is ​​constructed based on multiple direction vectors, integrating the direction distribution characteristics to achieve statistical modeling of the overall environmental geometry and enhancing adaptability to environmental complexity. Then, the covariance matrix of the direction cosine matrix is ​​obtained and inverted to obtain the weight coefficient matrix. The importance of different directions is evaluated using statistical methods. The weight coefficient matrix can more accurately represent the uncertainty and noise influence of environmental geometry, improving the robustness of error estimation. Finally, the sum of the main diagonal elements of the weight coefficient matrix is ​​calculated and squared to integrate the influence of all directions, generating a single environmental error data value, ensuring that the quantification result is comprehensive and accurate, thereby directly solving the challenge of accurately measuring the influence of environmental geometry.

[0032] Step 103: Obtain the carrier attitude estimation error of the target carrier in each epoch based on the carrier position data corresponding to two adjacent epochs and the posterior estimation error covariance matrix.

[0033] In this embodiment, the step mainly includes: for any given epoch, acquiring historical epochs adjacent to the given epoch and historical carrier position data of the target carrier in the historical epoch; acquiring the displacement of the target carrier in the given epoch based on the carrier position data of the target carrier in the given epoch and the historical carrier position data; using the product of the displacement and a pre-acquired empirical constant coefficient as a carrier attitude estimation error scaling factor; acquiring the original attitude error of the target carrier in the given epoch from the posterior estimation error covariance matrix corresponding to the given epoch, and scaling the original attitude error to a real-time attitude error based on the carrier attitude estimation error scaling factor and a preset spherical interpolation method; when it is determined that the real-time attitude error has a vertical observation constraint based on the posterior estimation error covariance matrix of the given epoch and the historical posterior estimation error covariance matrix of the historical epochs, using the real-time attitude error as the carrier attitude estimation error of the target carrier in each given epoch.

[0034] When it is determined that the real-time attitude error does not have the vertical observation constraint, the historical carrier attitude estimation error of the target carrier in the historical epoch is obtained; the product of the historical carrier attitude estimation error and the real-time attitude error is taken as the carrier attitude estimation error of the target carrier in the epoch.

[0035] In this embodiment, the steps described above for obtaining historical epochs and historical carrier position data allow for the quantification of motion trends based on historical changes, thereby reducing error accumulation. Obtaining displacement based on current and historical carrier position data accurately reflects the actual motion amplitude of the carrier, providing an objective basis for error adjustment. Multiplying the displacement by an empirical constant coefficient generates a scaling factor, taking into account the impact of motion intensity on the error; the empirical coefficient, based on prior knowledge, ensures the reasonableness of the scaling. After obtaining the original attitude error from the posterior estimation error covariance matrix, it is scaled to the real-time attitude error using the scaling factor and spherical interpolation method. Spherical interpolation achieves smooth adjustment in the attitude space, avoiding linear distortion. Finally, when the real-time attitude error is determined to have a vertical observation constraint based on the posterior estimation error covariance matrix and the historical matrix, this error is used. The vertical observation constraint verifies the reliability of the error, ensuring more accurate attitude estimation and thus improving the robustness of subsequent positioning and detection.

[0036] Step 104: Obtain the point cloud position estimation error of the target carrier in each epoch based on the carrier attitude estimation error and the point cloud feature position data corresponding to two adjacent epochs.

[0037] In this embodiment, the step mainly includes: for any given epoch, acquiring historical point cloud feature location data of adjacent historical epochs and the target carrier at the historical epoch; acquiring location data of each point cloud feature from the point cloud feature location data corresponding to the epoch and historical location data of each historical point cloud feature from the historical point cloud feature location data; acquiring new point cloud features from a plurality of point cloud features based on the comparison result between the historical location data and the location data; acquiring the direction vector corresponding to the new point cloud feature and the historical direction vector corresponding to each historical point cloud feature, so as to match the target carrier from a plurality of historical point cloud features based on the angle between the direction vector and the historical direction vector. The process involves: adding a new point cloud feature corresponding to a target historical point cloud feature; obtaining a displacement vector between the new point cloud feature and the target historical point cloud feature based on the location data corresponding to the new point cloud feature and the historical location data corresponding to the target historical point cloud feature; obtaining a real-time error corresponding to the new point cloud feature based on the displacement vector and the carrier attitude estimation error corresponding to the epoch; obtaining a confidence threshold for the historical position error of the target carrier at the historical epoch, and summing the historical position error confidence threshold and the real-time error as the cumulative error corresponding to the new point cloud feature; and obtaining several cumulative errors at the epoch, and using these cumulative errors as the point cloud position estimation error of the target carrier at the epoch.

[0038] In this embodiment, the above steps provide a benchmark reference for obtaining historical epochs and historical point cloud feature location data, ensuring the reliability of change detection; extracting information from current and historical location data allows for the identification of newly added point cloud features, avoiding the omission of dynamic changes; obtaining newly added point cloud features based on the comparison results of historical and location data, directly filtering for feature additions and subtractions, reducing misjudgments; obtaining direction vectors and matching target historical point cloud features based on the included angle, using geometric relationships to ensure accurate feature correspondence and reduce matching errors; obtaining displacement vectors based on location data, quantifying feature movement, and providing a basis for error calculation; combining displacement vectors and carrier attitude estimation errors to obtain real-time errors, fusing attitude influences, and more accurately reflecting the uncertainty of new features; obtaining historical location error confidence thresholds and calculating cumulative errors, inheriting historical error data to achieve error accumulation control; finally, obtaining the cumulative error as the point cloud location estimation error, outputting a comprehensive value, and enhancing robustness in dynamic environments.

[0039] Step 105: Obtain the odometry estimation error of the target vehicle in each epoch based on the posterior estimation error covariance matrix.

[0040] In this embodiment, the step mainly includes: for any epoch, obtaining the position estimation variance of the target vehicle in each coordinate axis direction and the square root of each position estimation variance from the posterior estimation error covariance matrix corresponding to the epoch; combining multiple square roots into a multidimensional vector, so as to use the multidimensional vector as the odometry estimation error of the target vehicle in the epoch.

[0041] In this embodiment, the above steps obtain the position estimation variance of the target vehicle in each coordinate axis direction from the posterior estimation error covariance matrix. This utilizes the existing variance data in the covariance matrix to directly extract the uncertainty information of the position estimation, avoiding the accumulation of errors caused by additional calculations. Next, the square root of each position estimation variance is calculated, and the variance is converted into standard deviation. This more intuitively represents the actual magnitude of the error because the standard deviation is the square root of the variance, which can more realistically reflect the error range. Then, multiple square roots are combined into a multidimensional vector. This preserves the distribution characteristics of the error in different coordinate axis directions, making the odometry estimation error multidimensional and able to comprehensively capture the anisotropic characteristics of spatial errors. Finally, using the multidimensional vector as the odometry estimation error provides a standardized mathematical representation, which is convenient for subsequent direct use in the calculation of the position error confidence threshold, thereby improving the accuracy and reliability of abnormal positioning data detection.

[0042] Step 106: Obtain the position error confidence threshold of the target carrier in each epoch based on the point cloud position estimation error, the odometer estimation error, the odometer environmental error data, and the pre-acquired odometer inherent ranging error.

[0043] In this embodiment, the steps mainly include: the inherent ranging error of the odometer includes the radar ranging error and the observation noise corresponding to each point cloud feature; for any epoch, obtaining the first product of the odometer environmental error data and the radar ranging error of the epoch; obtaining the sum of the observation noise of all newly added point cloud features in the epoch; for any newly added point cloud feature in the epoch, obtaining the ratio of the observation noise of the newly added point cloud feature to the sum of the noise, and obtaining the second product of the ratio and the cumulative error of the newly added point cloud feature; obtaining the sum of all the second products in the epoch to obtain the feature observation error corresponding to the epoch; obtaining the magnitude of the odometer estimation error of the epoch, so as to obtain the position error confidence threshold of the target carrier in the epoch based on the magnitude, the first product and the sum of the feature observation errors.

[0044] In this embodiment, the above steps obtain the first product of the odometer environmental error data and the radar ranging error, taking into account the amplification effect of the environment on radar error, making the calculation more in line with the actual scenario. The sum of the observation noise of all newly added point cloud features within this epoch is obtained, aggregating feature-related random noise and providing a basis for subsequent weight allocation. For any newly added point cloud feature in this epoch, the ratio of the observation noise of the newly added point cloud feature to the sum is obtained, and the second product of this ratio and the accumulated error of the newly added point cloud feature is obtained. This assigns a weight based on its noise contribution to each feature, multiplied by the accumulated error, highlighting the error impact of high-noise features and ensuring that threshold calculation is more targeted. The sum of all second products within this epoch is obtained to obtain the feature observation error corresponding to this epoch, which quantifies the overall uncertainty of feature matching and integrates the error contributions of all features. The modulus of the odometry estimation error at that epoch is obtained. Based on the sum of the modulus, the first product, and the characteristic observation error, the position error confidence threshold is obtained. This integrates odometry motion error and environmental geometric error to generate a more reliable confidence boundary for accurately judging anomalies in satellite positioning data.

[0045] Step 107: Obtain the position deviation between the mileage positioning data and the satellite positioning data for each epoch, so as to obtain the abnormal data detection result of the satellite positioning data based on the position deviation and the position error confidence threshold.

[0046] In this embodiment, the step mainly includes: for any epoch, obtaining the position deviation between the mileage positioning data and the satellite positioning data of the epoch; when the position deviation is greater than the position error confidence threshold, determining that the satellite positioning data corresponding to the epoch is abnormal positioning data.

[0047] In this embodiment, the above steps, for any epoch, acquire the positional deviation between the odometer positioning data and the satellite positioning data for that epoch. This step calculates the deviation in real time at each time point, enabling the detection to specifically cover all epochs and avoiding inaccuracies caused by the accumulation of overall errors, thereby enhancing the timeliness and specificity of the detection. Then, when the positional deviation is greater than the positional error confidence threshold, the satellite positioning data corresponding to that epoch is determined to be abnormal positioning data. This determination logic uses the pre-calculated positional error confidence threshold as an objective standard, directly comparing the deviation with the threshold, ensuring the quantification of the determination basis, avoiding misjudgments caused by subjective or vague judgments, and improving the robustness of detection in deception or interference scenarios by combining the characteristic that the confidence threshold comes from error estimation. Overall, by implementing deviation acquisition and threshold comparison step by step, the operability and accuracy of the detection mechanism are enhanced.

[0048] On the other hand, refer to Figure 2 This embodiment also discloses an abnormal positioning data detection system based on mileage estimation, which mainly includes a data acquisition module 201, an environmental error module 202, a carrier error module 203, a point cloud error module 204, a mileage error module 205, a position error module 206, and a data detection module 207.

[0049] The data acquisition module 201 is used to acquire satellite positioning data and multidimensional odometry observation data of the target carrier at each epoch; wherein, the multidimensional odometry observation data includes mileage positioning data, carrier position data, point cloud feature position data and posterior estimation error covariance matrix.

[0050] The environmental error module 202 is used to obtain the odometry environmental error data of the target carrier at each epoch based on the carrier location data and the point cloud feature location data.

[0051] The carrier error module 203 is used to obtain the carrier attitude estimation error of the target carrier in each epoch based on the carrier position data corresponding to two adjacent epochs and the posterior estimation error covariance matrix.

[0052] The point cloud error module 204 is used to obtain the point cloud position estimation error of the target carrier in each epoch based on the carrier attitude estimation error and the point cloud feature position data corresponding to two adjacent epochs.

[0053] The mileage error module 205 is used to obtain the mileage estimation error of the target vehicle in each epoch based on the posterior estimation error covariance matrix.

[0054] The position error module 206 is used to obtain the position error confidence threshold of the target carrier in each epoch based on the point cloud position estimation error, the odometer estimation error, the odometer environmental error data, and the pre-acquired odometer inherent ranging error.

[0055] The data detection module 207 is used to obtain the position deviation between the mileage positioning data and the satellite positioning data at each epoch, so as to obtain the abnormal data detection result of the satellite positioning data based on the position deviation and the position error confidence threshold.

[0056] In this embodiment, the environmental error module 202 includes a direction recognition unit and an environmental influence unit; The direction recognition unit is used to, for any epoch, obtain the position data corresponding to each point cloud feature from the point cloud feature position data corresponding to the epoch; and for any point cloud feature, obtain the direction vector of the point cloud feature relative to the target carrier based on the position data corresponding to the point cloud feature and the carrier position data corresponding to the epoch. The environmental impact unit is used to construct a direction cosine matrix of the target vehicle in the epoch based on several direction vectors corresponding to the point cloud feature location data; obtain the covariance matrix of the direction cosine matrix, and perform matrix inversion on the covariance matrix to obtain the weight coefficient matrix; obtain the sum of several matrix elements on the main diagonal of the weight coefficient matrix, and perform square root operation on the sum to obtain the odometry environmental error data of the target vehicle in the epoch.

[0057] This embodiment discloses an anomaly positioning data detection method and system based on odometer estimation. It acquires satellite positioning data and multi-dimensional odometry observation data of the target vehicle at each epoch, laying the foundation for subsequent error quantification based on the multi-dimensional odometry observation data. Next, it acquires odometry environmental error data based on vehicle position data and point cloud feature position data, quantifying the direct impact of environmental geometry on odometry accuracy by comparing the target position with environmental feature positions. It then acquires the vehicle attitude estimation error based on the vehicle position data of two adjacent epochs and the posterior estimation error covariance matrix, estimating the cumulative attitude error to address short-term odometer drift by utilizing the position changes and error statistical characteristics of adjacent time points. Finally, it calculates the vehicle attitude estimation error based on the vehicle attitude estimation error and the data from two adjacent epochs. The system obtains point cloud position estimation errors from point cloud feature location data, and combines attitude errors and dynamic changes in environmental features to improve the accuracy of point cloud matching and reduce environmental interference. It obtains odometer estimation errors based on the posterior estimation error covariance matrix, directly using the error matrix to simplify the odometer error extraction process. It obtains position error confidence thresholds based on point cloud position estimation errors, odometer estimation errors, odometer environmental error data, and pre-acquired inherent odometer ranging errors. It integrates multiple error sources and environmental factors, dynamically setting reliable thresholds to enhance detection robustness. It obtains the position deviation between odometer positioning data and satellite positioning data at each epoch, and obtains abnormal data detection results based on the position error confidence threshold. Through real-time comparison, it achieves accurate identification of satellite positioning anomalies.

[0058] Example 2 With the development of satellite positioning technology, high-precision satellite positioning data has become a key technical feature of positioning. However, satellite signal receivers used for positioning may generate inaccurate or even erroneous positioning data when faced with deception or interference. Therefore, anomaly detection of satellite positioning data has become a problem that needs to be solved in the positioning field.

[0059] To solve the above technical problems, refer to Figure 3 This embodiment discloses an anomaly location data detection method based on mileage estimation, which mainly includes: Step 301: Acquire multi-dimensional odometry observation data and satellite positioning data of the target carrier in each epoch in real time, and acquire odometry environmental error data of the target carrier in each epoch based on the multi-dimensional odometry observation data.

[0060] In this embodiment, the step mainly includes: when locating the target carrier, firstly, a satellite signal receiver fixed on the target carrier can be used to receive satellite signals in real time for each epoch, so as to output the satellite positioning data of the target carrier in each epoch based on the satellite signals. Preferably, the satellite signal receiver, such as a BeiDou receiver, installed on the target carrier outputs the BeiDou satellite positioning data of the target carrier in each epoch.

[0061] Next, to detect anomalies in the satellite positioning data, a tightly coupled odometer mounted on the target vehicle can be used to monitor the state of the target vehicle in real time at each epoch, outputting multi-dimensional odometer observation data of the target vehicle at each epoch. Preferably, to ensure the accuracy of anomaly detection in the satellite positioning data, the tightly coupled odometer can be a LiDAR-IMU odometer coupled with a lidar and an inertial measurement unit, so that the multi-dimensional odometer observation data of the target vehicle at each epoch can be output through the LiDAR-IMU odometer.

[0062] Furthermore, to ensure the accuracy of anomaly detection, the following data are obtained from the multi-dimensional odometry observation data: mileage positioning data calculated after locating the target vehicle using the LiDAR-IMU odometry; vehicle position data estimated by the LiDAR-IMU odometry based on the three-dimensional position of the target vehicle; point cloud feature position data obtained after detecting the environment of the target vehicle; and the posterior estimation error covariance matrix output by the LiDAR-IMU odometry at each epoch. Specifically, the vehicle position data refers to the coordinate position of the target vehicle in three-dimensional space at each epoch; the point cloud feature position data refers to the coordinate position of point clouds with characteristic significance in three-dimensional space; the point cloud is a collection of a large number of discrete points obtained after the LiDAR scans the surrounding environment of the target vehicle, and these points represent different object surfaces in the environment; the point cloud features refer to those points in the point cloud that have special geometric or statistical characteristics; and the posterior estimation error covariance matrix in the LiDAR-IMU odometry is used to describe the statistical characteristics of the vehicle state estimation error.

[0063] Secondly, after acquiring the satellite positioning data and the multi-dimensional odometry observation data, in order to obtain the influence of the surrounding environment on the odometry positioning error, the influence can be quantified using the point cloud feature location data and carrier location data in the multi-dimensional odometry observation data, so as to obtain odometry environmental error data that measures the degree of influence of single-frame point cloud feature distribution on positioning error.

[0064] Specifically, when acquiring the odometer environmental error data, the point cloud feature location data and carrier location data of the k-th epoch are extracted: from multiple point cloud features corresponding to the point cloud feature location data of the k-th epoch, the i-th point cloud feature is selected, and the three-dimensional position of the i-th point cloud feature is obtained. The three-dimensional position of the target carrier in the k-th epoch is... Next, the direction vector of the i-th point cloud feature in the k-th epoch is calculated relative to the target carrier, so as to construct the direction cosine matrix of the k-th epoch based on the direction cosine matrix, and then the odometer environmental error data of the k-th epoch is calculated based on the direction cosine matrix.

[0065] Specifically, the formula for calculating the direction vector is: in, Let be the unit direction vector of the i-th point cloud feature relative to the target carrier in the k-th epoch, and denominator be . The straight-line distance between the i-th point cloud feature and the target carrier is used for vector normalization; Let be the distance between the i-th point cloud feature and the target carrier.

[0066] The formula for constructing the directional cosine matrix is: in, for A 3×3 matrix (n is the total number of point cloud features in the k-th epoch), with each row corresponding to the unit direction vector of a point cloud feature; the matrix transpose symbol... This is used to arrange the feature direction vectors in rows to form a direction cosine matrix.

[0067] The formula for calculating the odometer environmental error data is as follows: in, Let be the covariance matrix of the direction cosine matrix (3×3), and (-1) denotes matrix inversion; the... The weight coefficient matrix; These are the weight coefficient matrices. The elements on the main diagonal correspond to the three dimensions of x, y, and z; where x, y, and z correspond to the x-axis, y-axis, and z-axis in three-dimensional space, respectively. The square root operation is used to integrate the weighted contributions of the three-dimensional position into a single geometric precision factor (LDOP). The smaller the LDOP value (i.e., the value of the odometry environmental error data), the more uniform the point cloud feature distribution, and the less the positioning error is affected by the feature distribution.

[0068] Step 302: Obtain the newly added point cloud features corresponding to each epoch based on the multidimensional odometry observation data between two adjacent epochs, and obtain the point cloud position estimation error corresponding to the newly added point cloud features of each epoch based on the multidimensional odometry observation data.

[0069] In this embodiment, this step mainly establishes the correlation between attitude error and position error, specifically including: for any epoch, taking the previous epoch as the historical epoch corresponding to the epoch, and obtaining the historical carrier position data and historical point cloud feature position data of the target carrier in the historical epoch.

[0070] The historical point cloud feature location data and the point cloud feature location data of the target carrier in the specified epoch are compared to determine the location data of several newly added point cloud features of the target carrier in the specified epoch. Next, the historical location data of each existing point cloud feature in the map is obtained from the historical point cloud feature location data. Next, based on the direction vector calculation method in the aforementioned steps, the direction vector corresponding to the newly added point cloud feature and the historical direction vector corresponding to each of the historical point cloud features are obtained. Then, the target historical point cloud feature corresponding to the newly added point cloud feature is matched from multiple historical point cloud features based on the angle between the direction vector and the historical direction vector. Specifically, the unit direction vector of the j-th newly added point cloud feature relative to the target carrier in the k-th epoch is calculated as follows: Next, the nearest target historical point cloud feature is matched to the newly added point cloud feature using the principle of minimum vector angle. Specifically, the matching process is as follows: In the formula, This refers to the location data corresponding to the j-th newly added point cloud feature in the k-th epoch. The unit direction vector of the newly added point cloud features relative to the target carrier; This is the unit direction vector of historical point cloud features in the map relative to the target carrier; It is negatively correlated with the cosine of the angle between the two vectors, and has a minimum value. The target historical point cloud features with the smallest included angle are used to achieve accurate matching between newly added features and existing features.

[0071] Next, after matching the target historical point cloud features, the displacement vector between the newly added point cloud features and the target historical point cloud features is obtained based on the position data corresponding to the newly added point cloud features and the historical position data corresponding to the target historical point cloud features. The spatial association between the newly added point cloud features and the target historical point cloud features is realized through the displacement vector, thus building a bridge for the conversion of attitude error to position error.

[0072] Then, the carrier position data of the target carrier in the k-th epoch and the historical carrier position data of the historical epoch are obtained, that is, the carrier position data of the target carrier in the k-th epoch are obtained. With the (k-1)th epoch Next, based on the target carrier's position data in that epoch and the historical carrier position data, the displacement of the target carrier in that epoch is obtained, and then the displacement is combined with an empirical constant factor. A scaling factor for the carrier attitude estimation error is calculated to avoid the problem of repeated accumulation of attitude error with increasing epochs, which could far exceed the actual error. Specifically, the calculation process for the scaling factor for the carrier attitude estimation error is as follows: In the formula, This represents the displacement of the target carrier from the (k-1)th epoch to the kth epoch. This is an empirical constant factor (for Horizon type LiDAR, the value range is 0.015-0.02). This is the scaling factor for the carrier attitude estimation error, used to avoid excessive amplification of attitude error due to epoch accumulation.

[0073] Next, the original attitude error of the target vehicle at that epoch is obtained from the posterior estimation error covariance matrix corresponding to that epoch, and then scaled to a real-time attitude error according to the vehicle attitude estimation error scaling factor and a preset spherical interpolation method. Specifically, the real-time attitude error is calculated as follows: In the formula, The unit quaternion (representing the error-free attitude) extracted from the posterior estimation error covariance matrix is ​​the... The original attitude error (quaternion form) extracted from the posterior estimated error covariance matrix, the for and The angle between the vectors; These are the interpolation weighting coefficients, which are used to uniformly scale the original attitude error to a certain value using spherical interpolation. The corresponding degree avoids sudden changes in attitude error; the The real-time attitude error is denoted as .

[0074] Then, based on the posterior estimation error covariance matrix of the epoch and the historical posterior estimation error covariance matrix of the historical epoch, it is determined whether the odometer has a vertical observation constraint at the epoch, so as to classify and correct the attitude error according to whether there is a continuous vertical observation constraint.

[0075] Specifically, when constraints are present, the real-time attitude error is directly used as the carrier attitude estimation error for the current epoch. When there are no constraints, the optimized real-time attitude error is accumulated with the historical attitude error from the previous epoch to serve as the carrier attitude estimation error for the target carrier in that epoch. Specifically, the process of obtaining the carrier attitude estimation error is as follows: in, Let be the carrier attitude estimation error at the k-th epoch. If there are continuous vertical observation constraints at the current epoch (mean gravity direction estimation is normal), then the attitude error is corrected through global constraints, and the corrected error is directly obtained using the real-time attitude error calculation method. If there is no vertical constraint, then the scaled attitude error will be... attitude error at epoch k-1 Multiplication allows for the reasonable accumulation of attitude errors.

[0076] Finally, based on the carrier attitude estimation error corresponding to each epoch and the displacement vector corresponding to each newly added point cloud feature, the real-time error of the newly added point cloud feature at that epoch is obtained. Then, the historical position error confidence threshold of the target carrier at that historical epoch is obtained, and the sum of the historical position error confidence threshold and the real-time error is taken as the cumulative error corresponding to the newly added point cloud feature. Specifically, the formula for calculating the cumulative error is as follows: In the formula, It is the cumulative error at the j-th newly added feature in the k-th epoch; Represents the confidence threshold for the historical position error calculated at the (k-1)th epoch; This is the carrier attitude estimation error at the k-th epoch; This represents the displacement vector between the newly added point cloud feature j and the corresponding target historical point cloud feature. Since attitude error is reflected in the observation of feature position through a "smaller near, larger far" phenomenon, the formula utilizes... In order to convert minute attitude errors into changes in three-dimensional position error from the carrier to the feature ( (A rotation matrix is ​​used in the calculation).

[0077] Step 303: Obtain the odometer estimation error of the target vehicle in each epoch based on the multidimensional odometer observation data, and obtain the position error confidence threshold of the target vehicle in each epoch based on the point cloud position estimation error, the odometer estimation error, the odometer environmental error data and the pre-acquired odometer inherent ranging error.

[0078] In this embodiment, the step mainly includes: first estimating the error covariance matrix from the posterior perspective. The relevant parts of the three-dimensional position information of the target carrier are extracted, and the standard deviation of the position information is obtained through specific calculations, which is defined as the position uncertainty. The location uncertainty is used as the odometry estimation error of the target carrier in each epoch. This uncertainty can directly characterize the location uncertainty of the odometry when performing frame-local map registration at the current epoch, providing core basic location error data support for subsequent error accumulation calculation.

[0079] Specifically, the position uncertainty The calculation formula is as follows: in, This is a 3×1 dimension uncertainty vector, corresponding to the three dimensions x, y, and z respectively; For the posterior estimation error covariance matrix, , , These are the variances of the estimated positions in the covariance matrix corresponding to the x, y, and z dimensions, respectively.

[0080] Secondly, the ranging error of the radar equipment is obtained from the pre-acquired inherent ranging error of the odometer. and the adaptive observation noise corresponding to each point cloud feature. The process involves combining odometer environmental error data with radar ranging error to calculate the combined contribution of feature distribution and hardware ranging to the position error. Secondly, the cumulative error at newly added features is weighted and averaged based on adaptive observation noise for each feature, highlighting the reference value of high-reliability features for error estimation. Finally, the position uncertainty, the aforementioned two error contribution results, and information based on the cumulative error propagation from the previous epoch are integrated to iteratively calculate the confidence threshold for the target carrier's position error in the k-th epoch. Through this epoch-based iterative process, multi-dimensional data such as position, attitude, feature distribution, and hardware error are continuously integrated to ultimately construct an odometry error accumulation model that can iteratively calculate the degree of error accumulation.

[0081] Specifically, the position error confidence threshold The calculation formula is: In the formula, The position error confidence threshold of the target carrier in the k-th epoch; The modulus of the positional uncertainty reflects the LIO's own registration error; This refers to the ranging error of LiDAR (with fixed equipment parameters). The positioning error is caused by the combined effect of point cloud feature distribution and ranging error. The cumulative error at the i-th feature in the k-th epoch; This is the adaptive observation noise for the i-th feature (reflecting the reliability of feature observation). These are the feature weight coefficients, and a weighted average is used to achieve a reasonable contribution of different reliability features to the position error. Step 304: Obtain the position deviation for each epoch based on the multidimensional odometry observation data and the satellite positioning data, and obtain the abnormal data detection result of the satellite positioning data based on the position deviation and the position error confidence threshold.

[0082] In this embodiment, the main step is as follows: Let the satellite positioning data of the k-th epoch be... The odometer positioning data extrapolated from the multidimensional odometry observation data output by LIO in the k-th epoch is: The positional deviation between satellite positioning data and odometer positioning data is: ;like Then determine This is an outlier in the BeiDou system; filter out this result to avoid interfering with multi-source fusion. Then determine To ensure the effectiveness of BeiDou data, global constraints are retained and used for LiDAR / IMU / BeiDou fusion positioning. It is important to note that point cloud data from multi-dimensional odometry observations undergo distortion correction based on IMU motion compensation, filtering redundant ground points to preserve planar and edge features. BeiDou data is also screened, removing invalid data with a "single-point solution" positioning status and retaining valid positioning results.

[0083] This embodiment discloses an anomaly positioning data detection method based on odometer estimation. By combining odometry environmental error data with dynamic attitude error correction using point cloud features, it can accurately estimate the cumulative error range of odometry positioning, providing an objective and reliable reference for anomaly judgment of satellite behavior data and avoiding misjudgments caused by relying on satellite's own positioning quality information (which has limited accuracy). For abnormal jump values ​​of satellites in occlusion and multipath scenarios, it achieves accurate filtering through "error range comparison", effectively suppressing the interference of outliers on LiDAR / IMU / satellite multi-source fusion positioning and improving the robustness of the fusion system in complex scenarios. The scheme introduces vertical constraint correction and feature adaptive weighting, taking into account the error estimation accuracy under different scenarios (sufficient / missing vertical features), making the outlier filtering results more in line with actual positioning needs, and ultimately improving the long-term positioning accuracy and stability of the multi-source fusion positioning system.

[0084] In this first embodiment, to verify as Figure 1 or Figure 3To assess the accuracy of an anomaly location data detection method based on mileage estimation, the experiment first involved selecting an experimental scenario and preparing the equipment. Specifically, a typical test section was chosen, including a turnaround section obstructed by an overpass, to verify the effectiveness of the position error accumulation estimation method under the target environment. Next, the test equipment included a LiDAR, IMU, BeiDou receiver, and data acquisition host. A ground truth device was used to provide a reference benchmark. Before the experiment, the external parameters of the LIO system composed of the LiDAR and IMU needed to be calibrated to determine the translation and rotation parameters, ensuring that the coordinate systems of the two systems were consistent, thus providing an accurate foundation for subsequent data acquisition and error estimation.

[0085] Secondly, experimental data acquisition and preprocessing were conducted. Driving along a pre-set test route, the data acquisition system was activated to simultaneously acquire multi-dimensional odometer observation data output by the LIO and BeiDou data. The LIO system output the carrier's 3D position, point cloud feature 3D position, and posterior estimation error covariance matrix per epoch at a fixed frequency, while the BeiDou receiver output the positioning result per epoch at a fixed frequency. During the data preprocessing stage, the point cloud data acquired by the LIO underwent distortion correction based on IMU motion compensation, filtering redundant ground points to retain planar and edge features that reflect the environmental structure. The BeiDou data was screened, removing invalid data with a "single-point solution" positioning status, retaining valid positioning results, and eliminating invalid interference to ensure the accuracy of subsequent error estimation. During the acquisition process, points of different colors in the trajectory represent areas where vertical features were detected or areas lacking vertical features; the distribution of these areas needs to be recorded synchronously for subsequent error analysis.

[0086] Next, odometry environmental error data is calculated, which measures the impact of single-frame point cloud feature distribution on positioning error. First, for the preprocessed LIO point cloud data of each epoch, a RANSAC algorithm combined with curvature analysis is used to extract planar and edge features. When the field of view is wide, the feature distribution is more uniform, resulting in higher registration accuracy; when the field of view is narrow, the feature distribution is concentrated, leading to greater positioning error. Then, the unit direction vector of each point cloud feature relative to the current epoch carrier position is calculated using the following formula: In the formula, Let be the 3D position of the i-th point cloud feature in the k-th epoch. The vector represents the carrier position data in the k-th epoch, with the denominator being the straight-line distance between the feature and the carrier, used for vector normalization. The direction vectors of all features in the same epoch are arranged in matrix form to construct a direction cosine matrix: In the formula, n represents the total number of point cloud features in the k-th epoch. Matrix transpose arranges the feature direction vectors in rows to integrate feature distribution information. The weight coefficient matrix is ​​solved by transposing and inversely operating the direction cosine matrix. Finally, the odometer environmental error data LDOP is calculated using the main diagonal elements of the weighting coefficient matrix, using the following formula: In the formula, These are the weight coefficient matrices. The elements on the main diagonal corresponding to the x, y, and z dimensions are combined by the square root operation to integrate the weighted contributions of the three-dimensional position into a single geometric precision factor. The larger the LDOP value, the more significant the influence of the feature distribution on the positioning error.

[0087] Furthermore, the positional uncertainty is first calculated from the posterior estimation error covariance matrix output by the LIO system. In the process, the variance of the corresponding x, y, and z dimension positional information is extracted using the formula: Converting to standard deviation yields the location uncertainty vector. This vector represents the LIO's own registration error. Next, attitude error association and correction are performed. First, the nearest existing feature is matched for the newly added feature in the current epoch, and the unit vector of the newly added feature relative to the carrier is calculated: Then through vector dot product operation Find the existing feature with the smallest included angle; calculate the carrier attitude estimation error scaling factor based on the LIO system displacement and empirical coefficients: In the formula, This is an empirical constant, with a value ranging from 0.015 to 0.02 for Horizon-type LiDAR. Let be the displacement modulus of the carrier in adjacent epochs; the attitude error is uniformly scaled to the calculated scaling level using a spherical interpolation method, with the following formula: In the formula, For unit quaternions, This represents the attitude error at the current epoch. for and The angle between the vectors; combined with vertical constraints to correct attitude error, if there are continuous vertical observation constraints in the current epoch and the gravity direction estimation is normal, the attitude error... ,otherwise Finally, the position error is iteratively accumulated using the formula: Integrating error contributions, where, For the magnitude of the positional uncertainty, For LiDAR ranging error, The cumulative error at the i-th feature in the k-th epoch. For feature-adaptive observation noise, The feature weight coefficients ultimately yield the cumulative range of carrier position error per epoch. .

[0088] Finally, BeiDou outlier verification was conducted based on the error accumulation results. The spatial deviation between the BeiDou positioning result and the LIO carrier position at each epoch was calculated, and this deviation was compared with the cumulative position error range of the corresponding epoch. Perform a comparison; if the deviation exceeds... If the result is within the range, it is considered a BeiDou anomaly; otherwise, it is considered a valid positioning result.

[0089] Based on the above method, the effects of different methods are compared by drawing trajectory side views. Specifically, a schematic diagram showing the effects is provided by drawing trajectory side views as follows. Figure 4 As shown. Among them, Figure 4 This includes a first motion trajectory generated based on optimized RTK (Real-Time Differential Positioning) technology for positioning the target vehicle; a second motion trajectory generated based on positioning quality information weighting for positioning the target vehicle; a third motion trajectory generated based on mileage estimation and filtering outliers for positioning the target vehicle; and the true trajectory of the target vehicle. Figure 4 It can be seen that when the target carrier moves along Figure 4 When moving in the indicated direction, as Figures 1 to 3 The method shown can effectively suppress the interference of abnormal jumps on the trajectory. Among them, Figure 4 There are signal obstruction areas; the true trajectory is the actual motion trajectory of the target carrier.

[0090] Next, draw as follows Figure 5 The elevation difference curves shown are for different methods, which are used to illustrate the positional deviation between the trajectory drawn by different methods and the actual trajectory. Figure 5 This includes a first elevation difference curve representing elevation difference curves generated based on traditional RTK technology, a second elevation difference curve representing elevation difference curves generated based on optimized RTK technology, a third elevation difference curve representing elevation difference curves generated based on positioning quality information weighting, and a fourth elevation difference curve representing elevation difference curves generated based on mileage estimation and outlier filtering. The horizontal axis of each elevation difference curve represents the operating mileage of the target vehicle, measured in meters (m). The vertical axis of each elevation difference curve represents the elevation difference between the positioning result and the actual trajectory, also measured in meters (m). Figure 5 It can be seen that the elevation differences between the positioning results based on different methods and the actual trajectory are small in regions 1 and 3, and the differences between the positioning results of different methods are also small; however, when the target vehicle is traveling in a signal-blocked area, i.e., traveling in areas such as... Figure 5Within region 2 shown, the original RTK-based positioning method exhibits a large positive error (35 to 40 m), while the other three optimized methods all produce negative errors. The method presented in this paper maintains a stable error within -18 to -20 m, significantly outperforming the other methods. Specifically, at the first and second turnaround points, due to LiDAR registration with the old map, the error curve of this embodiment shows a "broken line" pattern with no abrupt changes, demonstrating a smooth transition. Quantitative calculations show that the fusion positioning RMSE of the proposed method is 46.8% better than the original RTK optimization method and 29.3% better than the positioning quality weighting method.

[0091] Finally, experimental results were conducted to verify the reasonableness of the cumulative positional error range under different scenarios. Specifically, in urban road sections with numerous vertical features, the average deviation between the calculated elevation difference curve and the actual error curve was 22.45%-27.75%, while in suburban road sections with fewer vertical features, the average deviation was 13.63%-24.11%. This indicates that the error range is larger in occluded scenarios and smaller in open scenarios, demonstrating the method's scenario adaptability. Preferably, to further verify the accuracy of the proposed method, pre-acquired error data was used to... Figure 5 The elevation difference curve shown is used for trajectory optimization to obtain the following result: Figure 6 The post-optimization trajectory is shown. Among them, Figure 6 The post-optimization trajectory includes a first optimized curve representing the elevation difference curve generated by positioning based on traditional RTK technology, optimized based on errors; a second optimized curve representing the elevation difference curve generated by positioning based on optimized RTK technology, optimized based on errors; a third optimized curve representing the elevation difference curve generated by positioning based on positioning quality information weighting, optimized based on errors; and a fourth optimized curve representing the elevation difference curve generated by positioning based on mileage estimation after filtering outliers, optimized based on errors. Figure 6 It can be seen that the trajectory of the method provided in this embodiment has the smallest deviation from the true value. The RMSE in the signal-blocked area is improved by 31.3% compared with the original RTK optimization method and by 68.3% compared with the positioning quality weighting method. This verifies that the position error accumulation estimation method can effectively support BeiDou outlier filtering and improve the robustness and accuracy of multi-source fusion positioning.

[0092] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for detecting abnormal location data based on mileage estimation, characterized in that, include: The satellite positioning data and multidimensional odometry observation data of the target vehicle at each epoch are acquired; wherein, the multidimensional odometry observation data includes odometer positioning data, vehicle position data, point cloud feature position data and posterior estimation error covariance matrix; Based on the carrier location data and the point cloud feature location data, obtain the odometry environmental error data of the target carrier at each epoch; The carrier attitude estimation error of the target carrier in each epoch is obtained based on the carrier position data corresponding to two adjacent epochs and the posterior estimation error covariance matrix. The point cloud position estimation error of the target carrier in each epoch is obtained based on the carrier attitude estimation error and the point cloud feature position data corresponding to two adjacent epochs. The odometer estimation error of the target vehicle in each epoch is obtained based on the posterior estimation error covariance matrix. The position error confidence threshold of the target carrier in each epoch is obtained based on the point cloud position estimation error, the odometer estimation error, the odometer environmental error data, and the pre-acquired odometer inherent ranging error. The positional deviation between the mileage positioning data and the satellite positioning data at each epoch is obtained, and the abnormal data detection result of the satellite positioning data is obtained based on the positional deviation and the position error confidence threshold.

2. The method for detecting abnormal location data based on mileage estimation according to claim 1, characterized in that, The step of obtaining the odometry environmental error data of the target carrier at each epoch based on the carrier location data and the point cloud feature location data includes: For any given epoch, obtain the position data corresponding to each point cloud feature from the point cloud feature position data corresponding to that epoch; For any point cloud feature, the direction vector of the point cloud feature relative to the target carrier is obtained based on the location data corresponding to the point cloud feature and the carrier location data corresponding to the epoch. Construct the direction cosine matrix of the target carrier in the epoch based on several direction vectors corresponding to the point cloud feature location data; Obtain the covariance matrix of the direction cosine matrix, and perform matrix inversion on the covariance matrix to obtain the weight coefficient matrix; The sum of several matrix elements on the main diagonal of the weight coefficient matrix is ​​obtained, and the square root operation is performed on the sum to obtain the odometer environmental error data of the target vehicle in the epoch.

3. The method for detecting abnormal location data based on mileage estimation according to claim 2, characterized in that, The step of obtaining the carrier attitude estimation error of the target carrier in each epoch based on the carrier position data corresponding to two adjacent epochs and the posterior estimation error covariance matrix includes: For any given epoch, acquire the historical epochs adjacent to the given epoch and the historical carrier location data of the target carrier in the historical epoch; Based on the target carrier's position data in the epoch and the historical carrier position data, the displacement of the target carrier in the epoch is obtained; The product of the displacement and the pre-acquired empirical constant coefficient is used as the scaling factor for the carrier attitude estimation error. The original attitude error of the target carrier in the epoch is obtained from the posterior estimation error covariance matrix corresponding to the epoch, and the original attitude error is scaled into a real-time attitude error according to the carrier attitude estimation error scaling factor and a preset spherical interpolation method. When it is determined that the real-time attitude error has a vertical observation constraint based on the posterior estimation error covariance matrix of the epoch and the historical posterior estimation error covariance matrix of the historical epoch, the real-time attitude error is used as the carrier attitude estimation error of the target carrier in each epoch.

4. The method for detecting abnormal location data based on mileage estimation according to claim 3, characterized in that, The step of obtaining the carrier attitude estimation error of the target carrier in each epoch based on the carrier position data corresponding to two adjacent epochs and the posterior estimation error covariance matrix further includes: When it is determined that the real-time attitude error does not have the vertical observation constraint, the historical attitude estimation error of the target vehicle in the historical epoch is obtained. The product of the historical carrier attitude estimation error and the real-time attitude error is taken as the carrier attitude estimation error of the target carrier in the given epoch.

5. The method for detecting abnormal location data based on mileage estimation according to any one of claims 2-4, characterized in that, The step of obtaining the point cloud position estimation error of the target carrier in each epoch based on the carrier attitude estimation error and the point cloud feature position data corresponding to two adjacent epochs includes: For any given epoch, acquire the historical point cloud feature location data of the historical epochs adjacent to the given epoch and the target carrier in the historical epoch. The position data of each point cloud feature is obtained from the point cloud feature position data corresponding to the epoch, and the historical position data of each historical point cloud feature is obtained from the historical point cloud feature position data. New point cloud features are obtained from a plurality of point cloud features based on the comparison results between the historical location data and the location data. Obtain the direction vector corresponding to the newly added point cloud feature and the historical direction vector corresponding to each of the historical point cloud features, so as to match the target historical point cloud feature corresponding to the newly added point cloud feature from multiple historical point cloud features according to the angle between the direction vector and the historical direction vector. The displacement vector between the newly added point cloud feature and the target historical point cloud feature is obtained based on the position data corresponding to the newly added point cloud feature and the historical position data corresponding to the target historical point cloud feature. Based on the displacement vector and the carrier attitude estimation error corresponding to the epoch, the real-time error corresponding to the newly added point cloud feature is obtained; The historical position error confidence threshold of the target carrier at the historical epoch is obtained, and the sum of the historical position error confidence threshold and the real-time error is used as the cumulative error corresponding to the newly added point cloud feature. A number of cumulative errors are obtained for the epoch, and these cumulative errors are used as the point cloud position estimation error of the target carrier in the epoch.

6. The method for detecting abnormal location data based on mileage estimation according to claim 1, characterized in that, The step of obtaining the odometry estimation error of the target vehicle in each epoch based on the posterior estimation error covariance matrix includes: For any given epoch, the position estimation variance of the target carrier in each coordinate axis direction and the square root of each position estimation variance are obtained from the posterior estimation error covariance matrix corresponding to that epoch. The multiple square roots are combined into a multidimensional vector, which is then used as the odometry estimation error of the target carrier in the epoch.

7. The method for detecting abnormal location data based on mileage estimation according to claim 6, characterized in that, The inherent ranging error of the odometer includes radar ranging error and the observation noise corresponding to each point cloud feature. The step of obtaining the position error confidence threshold of the target carrier in each epoch based on the point cloud position estimation error, the odometer estimation error, the odometer environmental error data, and the pre-acquired inherent odometer ranging error includes: For any epoch, obtain the first product of the odometer environmental error data and the radar ranging error for that epoch; Obtain the sum of the observation noise of all newly added point cloud features within the epoch; For any newly added point cloud feature in the epoch, obtain the ratio of the observation noise of the newly added point cloud feature to the sum of the noise, and obtain the second product of the ratio and the accumulated error of the newly added point cloud feature. The sum of all the second products within the epoch is obtained to obtain the feature observation error corresponding to the epoch. Obtain the modulus of the odometry estimation error at the epoch, and obtain the position error confidence threshold of the target vehicle at the epoch based on the sum of the modulus, the first product, and the feature observation error.

8. The method for detecting abnormal location data based on mileage estimation according to claim 1, characterized in that, The step of obtaining the position deviation between the mileage positioning data and the satellite positioning data for each epoch, and obtaining the abnormal data detection result of the satellite positioning data based on the position deviation and the position error confidence threshold, includes: For any given epoch, obtain the positional deviation between the mileage positioning data and the satellite positioning data for that epoch; When the position deviation is greater than the position error confidence threshold, the satellite positioning data corresponding to that epoch is determined to be abnormal positioning data.

9. An anomaly location data detection system based on mileage estimation, characterized in that, It includes a data acquisition module, an environmental error module, a carrier error module, a point cloud error module, a mileage error module, a position error module, and a data detection module; The data acquisition module is used to acquire satellite positioning data and multidimensional odometry observation data of the target carrier at each epoch; wherein, the multidimensional odometry observation data includes odometer positioning data, carrier position data, point cloud feature position data, and posterior estimation error covariance matrix; The environmental error module is used to obtain the odometry environmental error data of the target carrier at each epoch based on the carrier location data and the point cloud feature location data. The carrier error module is used to obtain the carrier attitude estimation error of the target carrier in each epoch based on the carrier position data corresponding to two adjacent epochs and the posterior estimation error covariance matrix. The point cloud error module is used to obtain the point cloud position estimation error of the target carrier in each epoch based on the carrier attitude estimation error and the point cloud feature position data corresponding to two adjacent epochs. The odometer error module is used to obtain the odometer estimation error of the target vehicle in each epoch based on the posterior estimation error covariance matrix. The position error module is used to obtain the position error confidence threshold of the target carrier in each epoch based on the point cloud position estimation error, the odometer estimation error, the odometer environmental error data, and the pre-acquired odometer inherent ranging error. The data detection module is used to obtain the position deviation between the mileage positioning data and the satellite positioning data at each epoch, so as to obtain the abnormal data detection result of the satellite positioning data based on the position deviation and the position error confidence threshold.

10. The anomaly location data detection system based on mileage estimation according to claim 9, characterized in that, The environmental error module includes a direction recognition unit and an environmental impact unit; The direction recognition unit is used to, for any epoch, obtain the position data corresponding to each point cloud feature from the point cloud feature position data corresponding to the epoch; and for any point cloud feature, obtain the direction vector of the point cloud feature relative to the target carrier based on the position data corresponding to the point cloud feature and the carrier position data corresponding to the epoch. The environmental impact unit is used to construct a direction cosine matrix of the target vehicle in the epoch based on several direction vectors corresponding to the point cloud feature location data; obtain the covariance matrix of the direction cosine matrix, and perform matrix inversion on the covariance matrix to obtain the weight coefficient matrix; obtain the sum of several matrix elements on the main diagonal of the weight coefficient matrix, and perform square root operation on the sum to obtain the odometry environmental error data of the target vehicle in the epoch.

Images