A vehicle positioning data processing method and system based on V2I communication and a medium

Abstract

This application discloses a vehicle positioning data processing method, system, and medium based on V2I communication, relating to autonomous driving: It performs time synchronization and spatial alignment processing on multi-source sensor data; it uses an extended Kalman filter algorithm to fuse lidar point cloud data, inertial measurement data, and vehicle odometer data to obtain the vehicle's local prior position information; based on the vehicle's local prior position information, UWB ranging data, and roadside unit position data, it constructs an objective function based on V2I ranging observation, and obtains the optimized vehicle position using the least squares method; it matches the optimized vehicle position with a pre-stored map, calculates the positioning correction based on road geometric constraints, and outputs the final vehicle positioning result. This application constructs a three-layer progressive positioning framework of local prior estimation – V2I ranging optimization – map geometric constraints, solving the problem of continuous positioning accuracy for autonomous vehicles in closed environments with no / weak satellite signals.

Description

Technical Field

[0001] This application relates to the field of autonomous driving, and in particular to a vehicle positioning data processing method, system and medium based on V2I communication. Background Technology

[0002] With the rapid development of intelligent transportation and autonomous driving technologies, accurate and reliable vehicle positioning has become a key foundational technology for realizing driverless driving. Especially in special application scenarios such as underground mines and tunnels, driverless technology is of great significance for improving operational efficiency and ensuring personnel safety. However, these enclosed environments are precisely the scenarios where traditional positioning technologies face severe challenges.

[0003] In open outdoor environments, most vehicle positioning systems currently employ Global Navigation Satellite Systems (GNSS), such as GPS and BeiDou, directly receiving satellite signals for positioning. Some high-precision systems also integrate Real-Time Kinematic (RTK) or Differential GNSS (DGPS) technology, achieving sub-meter or centimeter-level positioning in areas with good signal coverage. To improve usability in scenarios where GNSS is unavailable, such as tunnels and mines, many vehicle systems integrate Inertial Measurement Units (IMUs), wheel speed / odometers, and LiDAR, employing methods such as Kalman filtering and particle filtering to achieve multi-sensor fusion positioning.

[0004] However, existing technologies have the following problems in practical applications:

[0005] First, GNSS signals are easily blocked in environments such as urban high-rise buildings (urban canyons), tunnels, and underground parking lots, resulting in a serious decrease in positioning accuracy or even complete failure. Although high-precision solutions such as RTK can provide centimeter-level positioning, they rely heavily on satellite differential signals and external reference stations, resulting in high network deployment costs and making them unusable in enclosed environments such as mines and tunnels.

[0006] Secondly, while fusion positioning schemes based on local sensors (IMU / odometer / LiDAR) do not rely on external signals, they have inherent drawbacks: IMUs and odometers accumulate errors, leading to severe positioning drift after prolonged operation, making it impossible to achieve high-precision positioning independently; LiDAR point cloud registration methods (such as ICP / NDT) are highly dependent on the richness of environmental features and the quality of the point cloud, and are prone to degradation and failure in scenarios with simple structures and sparse features, such as tunnels and mines. These methods still cannot completely overcome the problems of error accumulation and drift under environmental occlusion and complex road conditions.

[0007] Therefore, how to establish a robust positioning framework that integrates multi-source information without relying on a single positioning method in complex environments such as mines and tunnels where GNSS signals are lacking, while ensuring positioning continuity and effectively suppressing cumulative errors to achieve long-term high-precision positioning, is a technical problem that urgently needs to be solved. Summary of the Invention

[0008] To address the failure of vehicle-mounted positioning systems in environments without GNSS signals, such as mines and tunnels, this application provides a vehicle positioning data processing method, system, and medium based on V2I communication. By fusing vehicle-mounted multi-sensor data with roadside unit UWB ranging observation data, a three-layer progressive positioning framework of local prior estimation, V2I ranging optimization, and map geometric constraints is constructed to improve the continuous positioning accuracy of unmanned vehicles in closed environments with no or weak satellite signals.

[0009] One aspect of this application provides a vehicle positioning data processing method based on V2I communication, comprising: S1, collecting multi-source sensor data, including: lidar point cloud data, inertial measurement data, vehicle odometer data, roadside unit position data received via V2X communication, and UWB ranging data; S2, performing time synchronization and spatial alignment processing on the multi-source sensor data; S3, using an extended Kalman filter algorithm to fuse the lidar point cloud data, inertial measurement data, and vehicle odometer data to obtain the vehicle's local prior position information; S4, constructing an objective function based on V2I ranging observation based on the vehicle's local prior position information, UWB ranging data, and roadside unit position data, solving the objective function using the least squares method to obtain the optimized vehicle position; S5, matching the optimized vehicle position with a pre-stored map, calculating the positioning correction based on road geometric constraints, and outputting the final vehicle positioning result.

[0010] Furthermore, in S3, the extended Kalman filter algorithm is used to fuse LiDAR point cloud data, inertial measurement data, and vehicle odometer data to obtain the vehicle's local prior position information, including: defining the system state vector. ,in, For the vehicle's position coordinates, For heading angle, For speed, To achieve zero bias in the gyroscope, To achieve zero bias in the accelerometer, state prediction is performed based on inertial measurement data and vehicle odometer data in each sampling period to obtain the predicted state vector. Based on the state transition Jacobian matrix process noise covariance matrix Calculate the prediction error covariance matrix When new LiDAR point cloud data is received, the NDT algorithm is used to register the current frame point cloud with the prior map or the previous frame point cloud to obtain registration transformation parameters; based on the registration transformation parameters, the vehicle's observation pose in the map coordinate system is extracted. Based on the observed pose Constructing lidar observation equations Based on the predicted state vector Prediction error covariance matrix and radar observation equations The vehicle pose information is obtained by performing state and covariance updates through Kalman filtering. , as the vehicle's local prior location information;

[0011] Specifically, this application achieves online identification and dynamic compensation of systematic errors in the sensor system by constructing a six-dimensional extended state vector containing the IMU's zero-bias parameters. Traditional solutions often use gyroscope zero-bias parameters... and accelerometer zero bias Treating these zero biases as fixed parameters or ignoring them directly leads to severe cumulative drift after long-term operation. This solution incorporates these zero biases into the state estimation framework and continuously corrects them through recursive filtering, fundamentally suppressing the long-term drift problem of the IMU, which is crucial for continuous positioning in GNSS-free environments.

[0012] Furthermore, the introduction of the NDT algorithm for point cloud registration demonstrates stronger robustness in feature-sparse environments such as tunnels and mines compared to traditional methods like ICP. By transforming the point cloud into a continuously differentiable probability distribution, NDT can provide reliable relative pose estimation even in long, straight tunnels with degraded geometric features. This robust observation is adaptively weighted in the final estimation using Kalman gain, ensuring that the system outputs stable and reliable local prior positions under various complex environments.

[0013] Furthermore, based on the observed pose Constructing lidar observation equations ,in, To predict the state vector The vehicle pose prediction value extracted from it. These are the predicted vehicle position coordinates. For the predicted vehicle heading angle, To observe noise;

[0014] Furthermore, in S4, based on the vehicle's local prior location information, UWB ranging data, and roadside unit location data, an objective function based on V2I ranging observation is constructed. The objective function is solved using the least squares method to obtain the optimized vehicle position, including: acquiring the vehicle's local prior location information output in S3. As an initial estimate of the vehicle's position; the position is received via V2X communication from the roadside unit broadcasting the location. Constructing the UWB ranging observation equation: ,in, This indicates the UWB distance measurement value between the vehicle and the roadside unit; Represents the vehicle's position coordinates to be optimized; Indicates the location coordinates of the roadside unit; This represents UWB ranging noise; based on the vehicle's local prior location information. Construct the vehicle's local prior observation equation: ,in, This indicates the vehicle's pose that needs optimization. This represents the local prior position estimation error; based on the UWB ranging observation equation and the vehicle's local prior observation equation, a joint optimization objective function is constructed. The Gauss-Newton method is used to evaluate the objective function. Perform iterative optimization to calculate the vehicle position parameters that minimize the objective function. As an optimized location for vehicles;

[0015] In particular, while traditional inertial navigation and lidar fusion can provide continuous relative positioning, it inevitably produces cumulative drift after long-term operation, which is especially serious in enclosed tunnel and mine environments. This application introduces UWB ranging observations from roadside units and constructs ranging observation equations. By establishing a geometric constraint relationship between the vehicle's position and the roadside base station with known precise coordinates, the system can obtain periodic absolute position correction.

[0016] Furthermore, this application does not simply replace local estimation with UWB ranging results, but rather achieves this by jointly optimizing the objective function. This approach organically integrates two information sources. On one hand, UWB ranging observations provide anchor points to prevent drift; on the other hand, local prior information... This ensures the continuity and smoothness of the positioning results, avoiding position jumps caused by single UWB ranging errors. By iteratively solving this nonlinear optimization problem using the Gauss-Newton method, the system can effectively eliminate accumulated errors while maintaining trajectory continuity, achieving an optimal balance between relative and absolute positioning.

[0017] Finally, this application not only solves the problem of absolute positioning in GNSS-free environments, but more importantly, it establishes a scalable framework. When multiple roadside units are deployed, the system can form stronger geometric constraints through multi-point ranging, further improving positioning accuracy and robustness.

[0018] Furthermore, based on the UWB ranging observation equation and the vehicle's local prior observation equation, a joint optimization objective function is constructed. : in, , , This represents the local prior estimate of the covariance matrix; This represents the variance of UWB ranging noise.

[0019] Specifically, the first item This represents the weighted sum of squares of the UWB ranging residuals, with weights determined by the variance of the ranging noise. The reciprocal of the value determines the position. This design ensures that when UWB ranging accuracy is high (small variance), this factor dominates the optimization, effectively guiding the vehicle position towards the true position and solving the problem of cumulative error divergence.

[0020] Second item It is a prior constraint term based on Mahalanobis distance, where the covariance matrix... As a weight matrix, it accurately characterizes the reliability of the local estimation in different directions. When the local sensor fusion quality is high, If the value is relatively small, the weight of this constraint term increases, ensuring that the positioning result does not drastically change due to a single UWB ranging anomaly; conversely, if the accumulated error increases, it will lead to... As the size increases, the system will rely more on UWB observations for correction.

[0021] When the UWB signal is good, its absolute positioning capability is fully utilized to eliminate accumulated errors. When the UWB signal is blocked or there is multipath interference, local priors are used to maintain trajectory continuity. By minimizing this objective function, the system can maintain trajectory smoothness while ensuring positioning accuracy.

[0022] Furthermore, S5 matches the optimized vehicle location with the pre-stored map, calculates the positioning correction based on road geometric constraints, and outputs the final vehicle positioning result, including: obtaining the optimized vehicle location output by S4. The system extracts road geometry information within a pre-defined range for vehicle optimization from a pre-stored map. This road geometry information includes the coordinate sequence of the road centerline, the coordinate sequence of the lane boundary lines, and the road heading angle corresponding to each coordinate point. The coordinate points on the road centerline and lane boundary lines are used as map constraint points. Based on a Hidden Markov Model, the system optimizes the vehicle's position... Within a preset range, candidate matching points are searched from map constraint points to form a matching candidate set. Where j is the candidate point index; calculate the Euclidean distance between each candidate point and the vehicle's optimized position. ; Obtain the vehicle heading angle output by S3 Calculate the road heading angle corresponding to each candidate point. The heading angle residual between: ,in, Let the road heading angle be the one corresponding to the j-th candidate point extracted from the road geometry information; construct the total loss function: ,in, For positional weights, The heading angle is used as the weight; the candidate point that minimizes the total loss function L_j is selected as the optimal map matching point. Based on the optimal map matching point and the vehicle's optimized location, the positioning correction is calculated, and the final vehicle positioning result is output. ;

[0023] In particular, in enclosed environments such as mine shafts and tunnels, road networks are highly structured—vehicles can only travel along predetermined lanes or tunnels. This strong geometric constraint provides an additional source of information for positioning. This application constructs a total loss function that integrates location distance and heading angle residuals. It not only considers the spatial distance relationship between vehicles and roads, but more importantly, it introduces a heading angle consistency constraint. This dual constraint mechanism can effectively distinguish easily confused scenarios such as parallel roads and intersections.

[0024] Furthermore, this application employs a Hidden Markov Model (HMM) framework for candidate point search and matching, fully considering the temporal continuity of vehicle motion. Traditional nearest neighbor matching may cause matching points to jump between different roads due to local noise, while HMM ensures the smoothness and rationality of the matching results over time by modeling state transition probabilities. This is of great significance for solving the problem of positioning continuity in complex scenarios such as tunnel bifurcation and alleyway intersections.

[0025] This application utilizes the hard constraints of the road to eliminate lateral positioning errors, ensuring that the final output positioning result not only meets lane-level requirements in absolute accuracy, but more importantly, guarantees the rationality of the positioning result—the vehicle is always within the drivable area.

[0026] Furthermore, based on the Hidden Markov Model, the optimal location for the vehicle is determined. Within a preset range, candidate matching points are searched from map constraint points to form a matching candidate set. This includes: optimizing vehicle location Centered on a target location, a search radius R is set to define the search area. All road segments within the search area are identified. For each road segment, the vertical projection point from the vehicle's optimized position to that road segment is calculated. These vertical projection points, along with discrete points along the road centerline and lane boundary lines within the search area, are used as initial candidate points. Based on a Hidden Markov Model, the transition probability from the vehicle's previous position to each initial candidate point is calculated using the vehicle's historical trajectory. Candidate points with transition probabilities below a threshold are eliminated, forming the final matching candidate set. , where j is the candidate point index.

[0027] Furthermore, based on the optimal map matching point and the vehicle's optimized position, the positioning correction is calculated, and the final vehicle positioning result is output. ,include: ; Where α is the confidence weighting parameter, and its value ranges from 0 to 1;

[0028] In particular, in practical applications, pre-stored maps may have problems such as surveying errors, update lags, or local distortions. Forcing the vehicle's location directly onto the map could actually reduce positioning accuracy. This application achieves optimized positioning by introducing a confidence-weighted parameter α. Matching points on the map Adaptive fusion.

[0029] First, when the sensor fusion quality is high and the positioning uncertainty is low, a larger α value can be set to retain more sensor measurement results and avoid introducing additional deviations due to map errors. Second, in areas where sensor performance degrades but map features are obvious, the α value can be reduced to make full use of road geometric constraints to improve positioning accuracy. Finally, this continuously adjustable fusion strategy ensures the smoothness of the positioning trajectory and avoids position jumps caused by sudden switching to map matching mode.

[0030] Another aspect of this application provides a vehicle positioning data processing system based on V2I communication, comprising: a data acquisition module for acquiring multi-source sensor data, including: lidar point cloud data, inertial measurement data, vehicle odometer data, roadside unit position data received via V2X communication, and UWB ranging data; a data preprocessing module connected to the data acquisition module for performing time synchronization and spatial alignment processing on the multi-source sensor data; and a local positioning module connected to the data preprocessing module for fusing lidar point cloud data, inertial measurement data, and vehicle odometer data using an extended Kalman filter algorithm to obtain the vehicle's local prior position information; the local positioning module includes: a state definition unit for defining the system state vector. The system includes: a state prediction unit for predicting the state based on inertial measurement data and vehicle odometer data; a point cloud registration unit for performing point cloud registration using the NDT algorithm to extract the observed pose; and a filtering and updating unit for updating the state using Kalman filtering and outputting vehicle pose information. The V2I optimization module, connected to the local positioning module and the data acquisition module, is used to construct a joint optimization objective function based on V2I ranging observations, using the vehicle's local prior location information, UWB ranging data, and roadside unit location data. The optimal vehicle position is obtained by solving the Gauss-Newton method. The map matching module, connected to the V2I optimization module and the local positioning module, is used to match the optimized vehicle position with the pre-stored map, calculate the positioning correction based on road geometric constraints, and output the final vehicle positioning result. The map matching module includes: a map extraction unit for extracting road geometric information from the pre-stored map; a candidate point search unit for searching candidate matching points based on a Hidden Markov Model; and a matching evaluation unit for constructing the total loss function. Select the best map matching point; the fusion output unit is used to calculate the final positioning result through confidence-weighted fusion. .

[0031] Compared to existing technologies, the advantages of this application are:

[0032] By constructing a three-layer progressive positioning framework—local prior estimation, V2I ranging optimization, and map geometric constraints—this method fundamentally solves the problem of high-precision vehicle positioning in environments without GNSS signals. First, it employs extended Kalman filtering to fuse LiDAR, IMU, and odometer data, achieving deep fusion of multi-source sensors through six-dimensional state vectors and NDT point cloud registration, effectively suppressing accumulated drift errors. Then, it innovatively constructs a joint optimization objective function by combining UWB ranging observations with local priors, utilizing the absolute position reference provided by roadside units and a weighted least squares framework to achieve optimal vehicle-road information fusion, reducing positioning errors from the meter level to the decimeter level. Finally, it introduces a map matching mechanism based on a hidden Markov model, using adaptive adjustments of the total loss function of position distance and heading angle residuals, as well as confidence weighting parameters, to fully utilize structured road geometric constraints for fine-tuning. Through the organic combination of three progressive optimizations, this invention achieves a seamless transition from local relative positioning to global absolute positioning, effectively solving the technical problem of cumulative error divergence in GNSS-free environments, and providing centimeter-level precision continuous positioning capability for unmanned driving in special scenarios such as mines and tunnels. Attached Figure Description

[0033] This application will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein:

[0034] Figure 1 This is an exemplary flowchart illustrating a vehicle positioning data processing method based on V2I communication according to some embodiments of this application;

[0035] Figure 2 This is a schematic diagram illustrating the positioning correction according to some embodiments of this application. Detailed Implementation

[0036] The methods and systems provided in the embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0037] like Figure 1 As shown, multi-source sensor data is collected, including: LiDAR point cloud data, inertial measurement data, vehicle odometer data, roadside unit position data received via V2X communication, and UWB ranging data. The multi-source sensor data undergoes time synchronization and spatial alignment processing. An extended Kalman filter algorithm is used to fuse the LiDAR point cloud data, inertial measurement data, and vehicle odometer data to obtain the vehicle's local prior position information. Based on the vehicle's local prior position information, UWB ranging data, and roadside unit position data, a target function based on V2I ranging observation is constructed. The target function is solved using the least squares method to obtain the vehicle's optimized position. The optimized vehicle position is matched with a pre-stored map, and the positioning correction is calculated based on road geometric constraints to output the final vehicle positioning result.

[0038] First, multi-sensor spatial calibration is performed to determine the rigid body transformation relationship between the LiDAR, IMU, and vehicle coordinate system. Specifically, this includes: LiDAR-vehicle calibration: calculating the rotation matrix from the LiDAR coordinate system to the vehicle coordinate system using a calibration board or feature point matching method. Translation vector IMU-vehicle calibration: Determine the transformation parameters from the IMU coordinate system to the vehicle coordinate system. and LiDAR-IMU joint calibration: Determines the relative pose relationship between the two sensors for subsequent data fusion. Extended Kalman filter (EKF) is used to fuse LiDAR, IMU, and wheel speed data.

[0039] The PTP (Precision Time Protocol) time synchronization strategy is adopted to achieve microsecond-level time synchronization: the domain controller acts as the PTP master clock and broadcasts synchronization signals to the lidar and IMU; each sensor is configured with a PTP slave clock and dynamically adjusts its local clock to keep synchronized with the master clock; during data acquisition, a unified timestamp is added to each sensor data frame.

[0040] The vehicle, equipped with sensors, traverses the target area, storing LiDAR point clouds, IMU data, and odometer information in ROSbag format. The collected multi-source data is input into the Fast-LIO algorithm, which constructs a 3D point cloud map in real time by tightly coupling LiDAR and IMU data, and outputs a dense point cloud in PCD format.

[0041] By integrating local prior location information, UWB ranging, and RSU coordinates into the V2I ranging observation equation and optimization target, higher-precision location information is output.

[0042] The fused positioning results are matched with a high-precision map (lane lines, road geometry features), unreasonable positioning points are eliminated, and drift is corrected by methods such as least squares or particle filtering.

[0043] Specifically, a point cloud map is constructed: the LiDAR and vehicle body, and the LiDAR and IMU are calibrated to obtain their spatial coordinate relationships. The IMU and LiDAR are synchronized using a PTP time synchronization strategy, as illustrated in the diagram below. This invention only uses a domain controller, LiDAR, and IMU. Sensor data is collected and stored in bag format. Each data point is input into the Fast-Lio algorithm for SLAM mapping, resulting in a PCD-formatted point cloud map. The point cloud map undergoes downsampling, radius filtering, ground point cloud filtering, and noise reduction to obtain a high-precision point cloud map, which serves as the prior map for localization in this system.

[0044] The vehicle-side data acquisition process includes raw sensor data: lidar point cloud information, IMU (accelerometer, gyroscope) data, odometer information, real-time positioning and dynamic information of neighboring vehicles and roadside units (RSUs) obtained through V2X communication, and high-precision map constraint information. Noise removal is then performed on the raw point cloud data.

[0045] For initial vehicle-mounted fusion positioning, an extended Kalman filter (EKF) is used to fuse GNSS, IMU, and wheel speed data.

[0046] Set system status variables: ,in, For location, For heading angle, For speed, and Zero bias for IMU.

[0047] High-frequency IMU / odometer prediction uses IMU and wheel speed data to predict the vehicle's new state in each sampling period.

[0048] ;

[0049] ;

[0050] ;

[0051] ;

[0052] ; ;in, and These are the acceleration and angular velocity measured by the IMU, respectively, and Δt is the sampling interval.

[0053] Covariance prediction: ;

[0054] LiDAR point cloud observation matching: When a new frame of LiDAR point cloud arrives, NDT is used to register the current frame with the map or the previous frame of point cloud to obtain the vehicle's observation pose in the map coordinate system. ;

[0055] LiDAR observation equations: ;in, To observe noise.

[0056] Kalman filter fusion update, calculation of observation matrix (LiDAR observations correspond to x, y, θ): .

[0057] Calculate the Kalman gain: ;

[0058] State quantity update: ;

[0059] Covariance update: ;

[0060] Through the above process, high-precision pose (x, y, θ) information is output in real time.

[0061] Cooperative V2X positioning enhancement uses V2X communication to periodically broadcast and receive the following data: current fused positioning results for this vehicle / nearby vehicles (e.g., based on IMU, LiDAR, or odometer). RSU and UWB periodically broadcast their precise coordinates. and Based on the prior vehicle positioning, a V2I ranging observation equation is established.

[0062] UWB ranging observation equation: ;in: This indicates the actual distance measurement result from UWB. Indicates the vehicle's current actual location (to be estimated); Indicates the known location of the RSU; This indicates ranging noise.

[0063] Local prior observation equation: ; This is the local prior error, typically Gaussian noise;

[0064] Joint optimization objective function:

[0065] The least-squares objective that integrates UWB ranging and local prior is: in, This represents the variance of UWB ranging noise. To estimate the covariance matrix for local priors, the Gauss-Newton method is used. Minimize the position to obtain the optimal position estimate (x, y) of the vehicle.

[0066] like Figure 2 As shown, map matching and positioning correction combine the vehicle's original positioning trajectory (which may contain drift or jumps) obtained through sensor fusion with a high-precision road map (such as lane lines, road center lines, intersection topology, etc.) to correct the vehicle's actual road position, achieving high-precision and robust positioning. The high-precision map contains structured data such as road center lines, lane boundaries, intersection topology, and speed limit information.

[0067] A Hidden Markov Model (HMM) is used to search for all possible map candidate points (such as the nearest road centerline point, the nearest lane line, landmark points, etc.) near the current location point to form a matching candidate set. Calculate the Euclidean distance between each candidate point and the vehicle fusion solution: ;

[0068] The heading angle residual is: ;

[0069] The total loss function can be weighted and summed: in, As weight.

[0070] Select the matching point with the minimum total loss so that Minimize, thus obtaining the optimal map matching location. .

[0071] The matching point is used as the new observation input, and the pose obtained in step (3) is smoothed by weighted averaging. The weighting parameter is selected as (0-1) confidence level.

[0072] The foregoing illustrative description of the present application and its embodiments is not restrictive and can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. The accompanying drawings are only one embodiment of the present application, and the actual structure is not limited thereto. Therefore, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the present application, such designs should fall within the scope of protection of this application. Furthermore, the word "comprising" does not exclude other elements or steps, and the word "a" preceding an element does not exclude the inclusion of "a plurality" of that element. Terms such as "first," "second," etc., are used to indicate names and do not indicate any specific order.

Claims

1. A vehicle positioning data processing method based on V2I communication, characterized in that, include: S1 collects multi-source sensor data, including: lidar point cloud data, inertial measurement data, vehicle odometer data, roadside unit position data received via V2X communication, and UWB ranging data; S2 performs time synchronization and spatial alignment processing on multi-source sensor data; S3 uses the extended Kalman filter algorithm to fuse lidar point cloud data, inertial measurement data and vehicle odometer data to obtain the vehicle's local prior position information. S4. Based on the vehicle's local prior location information, UWB ranging data, and roadside unit location data, construct an objective function based on V2I ranging observation, and solve the objective function using the least squares method to obtain the vehicle's optimal position. S5 matches the optimized vehicle location with the pre-stored map, calculates the positioning correction based on road geometric constraints, and outputs the final vehicle positioning result.

2. The vehicle positioning data processing method based on V2I communication according to claim 1, characterized in that: S3, obtain the vehicle's local prior location information, including: Define the system state vector ,in, For the vehicle's position coordinates, For heading angle, For speed, To achieve zero bias in the gyroscope, To achieve zero bias in the accelerometer; In each sampling period, state prediction is performed based on inertial measurement data and vehicle odometer data to obtain a predicted state vector. ; Based on the state transition Jacobian matrix process noise covariance matrix Calculate the prediction error covariance matrix ; When new lidar point cloud data is received, the NDT algorithm is used to register the current frame point cloud with the prior map or the previous frame point cloud to obtain the registration transformation parameters. Based on the registration transformation parameters, the observed pose of the vehicle in the map coordinate system is extracted. ; Based on the observed pose Constructing lidar observation equations ; Based on the predicted state vector Prediction error covariance matrix and radar observation equations The vehicle pose information is obtained by performing state and covariance updates through Kalman filtering. This serves as the vehicle's local prior location information.

3. The vehicle positioning data processing method based on V2I communication according to claim 2, characterized in that: Based on the observed pose Constructing lidar observation equations : ,in, To predict the state vector The vehicle pose prediction value extracted from it. These are the predicted vehicle position coordinates. For the predicted vehicle heading angle, To observe noise.

4. The vehicle positioning data processing method based on V2I communication according to claim 2, characterized in that: S4 yields the optimized vehicle location, including: Obtain the vehicle's local prior location information output by S3. As an initial estimate of the vehicle's position; Location broadcast by roadside unit via V2X communication ; Constructing the UWB ranging observation equation: ,in, This indicates the UWB distance measurement value between the vehicle and the roadside unit; Represents the vehicle's position coordinates to be optimized; Indicates the location coordinates of the roadside unit; This indicates UWB ranging noise; Based on the vehicle's local prior location information Construct the vehicle's local prior observation equation: ,in, This indicates the vehicle pose that needs to be optimized. This indicates the local prior location estimation error; Based on the UWB ranging observation equation and the vehicle's local prior observation equation, a joint optimization objective function is constructed. ; The Gauss-Newton method is used to evaluate the objective function. Perform iterative optimization to calculate the vehicle position parameters that minimize the objective function. This serves as an optimized location for the vehicle.

5. The vehicle positioning data processing method based on V2I communication according to claim 4, characterized in that: Constructing a joint optimization objective function : in, , , This represents the local prior estimate of the covariance matrix; This represents the variance of UWB ranging noise.

6. The vehicle positioning data processing method based on V2I communication according to claim 4, characterized in that: S5, Calculates the positioning correction based on road geometric constraints, including: Obtain the optimized vehicle location from S4 output. ; Extract road geometry information within a preset range for the vehicle's optimized location from a pre-stored map. The road geometry information includes: the coordinate sequence of the road centerline, the coordinate sequence of the lane boundary line, and the road heading angle corresponding to each coordinate point. Use the coordinates of the road centerline and lane boundary lines as map constraint points; Based on the Hidden Markov Model, in the vehicle optimization position Within a preset range, candidate matching points are searched from map constraint points to form a matching candidate set. Where j is the candidate point index; Calculate the Euclidean distance between each candidate point and the vehicle's optimized position. ; Obtain the vehicle heading angle output by S3 Calculate the road heading angle corresponding to each candidate point. The heading angle residual between: ,in, Let be the road heading angle corresponding to the j-th candidate point extracted from the road geometry information; Construct the total loss function: ,in, For positional weights, Weighted by heading angle; The candidate point that minimizes the total loss function L_j is selected as the optimal map matching point. ; Based on the optimal map matching point and the optimized vehicle location, the positioning correction is calculated, and the final vehicle positioning result is output. .

7. The vehicle positioning data processing method based on V2I communication according to claim 6, characterized in that: Form a matching candidate set ,include: Vehicle-optimized positioning Centered on a target location, set a search radius R to define the search area; Identify all road segments within the search area; For each road segment, calculate the vertical projection point of the vehicle's optimized position onto the corresponding road segment; The vertical projection points, along with the discrete points of the road centerline and lane boundary lines within the search area, are used together as initial candidate points. Based on the Hidden Markov Model, the transition probability from the previous position to each initial candidate point is calculated according to the vehicle's historical trajectory. Candidate points with transition probabilities below a threshold are removed to form the final matching candidate set. , where j is the candidate point index.

8. The vehicle positioning data processing method based on V2I communication according to claim 6, characterized in that: Output the final vehicle location result. ,include: ; ; Where α is the confidence weighting parameter, and its value ranges from 0 to 1.

9. A vehicle positioning data processing system based on V2I communication, characterized in that, include: The data acquisition module is used to collect data from multiple sources of sensors, including: lidar point cloud data, inertial measurement data, vehicle odometer data, roadside unit position data received via V2X communication, and UWB ranging data. The data preprocessing module is used to perform time synchronization and spatial alignment processing on multi-source sensor data; The local positioning module is used to fuse lidar point cloud data, inertial measurement data and vehicle odometer data using the extended Kalman filter algorithm to obtain the vehicle's local prior position information. The V2I optimization module is used to construct a joint optimization objective function based on V2I ranging observations, using the vehicle's local prior location information, UWB ranging data, and roadside unit location data. The optimal vehicle position is obtained by solving the Gauss-Newton method. ; The map matching module is used to match the optimized vehicle location with the pre-stored map, calculate the positioning correction based on road geometric constraints, and output the final vehicle positioning result.

Images