Motor vehicle track measurement and optimization method based on Beidou satellite

Abstract

The invention discloses a motor vehicle track measurement and optimization method based on a Beidou satellite, and relates to the technical field of Beidou positioning measurement and track optimization, and the method comprises the steps: measuring a vehicle track based on an IMU and Beidou tight coupling fusion vehicle track measurement method, and obtaining an original vehicle track; based on a trajectory enhancement strategy in a CL-TSim model, performing potential path recovery on an original vehicle trajectory to obtain an enhanced interaction trajectory, calculating a distance between the original vehicle trajectory and a movement feature representation vector, screening out a vehicle optimization trajectory segment, and outputting a high-quality trajectory segment which can best represent a real vehicle driving mode. The problem of insufficient vehicle trajectory data measurement precision is solved, a space-time co-occurrence mode of vehicle movement is realized, and missed road network matching in a complex remote environment is ensured and intelligent screening and optimization of sub-trajectory segments are realized according to vehicle start and end point positioning and estimation of vehicle midway positioning which is not detected.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of Beidou positioning measurement and trajectory optimization, and particularly relates to a Beidou satellite-based motor vehicle trajectory measurement and optimization method. BACKGROUND

[0002] As a time and space infrastructure independently constructed by China, the Beidou satellite navigation system plays an increasingly important role in the field of motor vehicle trajectory measurement. With the popularization of Beidou vehicle-mounted terminals and the continuous expansion of industry application scenarios, vehicle trajectory data presents the characteristics of explosive growth of information, diversification of dimensions and uneven quality. The unpredictable problems existing in the data acquisition, calculation and storage process lead to the inability to guarantee the data quality and the complexity of the query method, which seriously restricts the in-depth application of trajectory data in the fields of intelligent transportation, vehicle monitoring and driving behavior analysis. In addition, when vehicles drive in complex environments such as urban canyons, viaducts and dense forests, satellite signal shielding and interference often lead to a decrease in positioning accuracy or even a discontinuity of the trajectory, further increasing the challenge of trajectory measurement.

[0003] Traditional single satellite positioning technology or simple Kalman filtering algorithm cannot maintain stable and reliable high-precision positioning in such dynamically changing environments, which seriously affects the realization of high-level applications such as lane-level positioning and fine trajectory analysis. In addition, the original Beidou trajectory data contains noise points and abnormal points caused by device errors and environmental interference. Vehicle inertial measurement and control multi-source data and Beidou positioning data are usually simply superimposed or independently processed, without deep fusion. The lack of adaptive fusion model that can dynamically adjust the weight according to the scene leads to a rapid decline in the overall positioning accuracy and robustness of the system when the satellite signal is poor, and cannot form complementary advantages. The current processing method focuses on the geometric shape of the trajectory, but lacks deep mining of the rich semantic information behind the trajectory. SUMMARY

[0004] The technical problem solved by the present application is that the existing technology cannot solve the problems of positioning drift, trajectory interruption and noise interference caused by serious satellite signal shielding, lacks continuity compensation for broken or missing points in the original trajectory, and cannot effectively capture the spatiotemporal co-occurrence pattern of vehicle movement.

[0005] To solve the above technical problems, the present application provides the following technical solutions: The Beidou satellite-based motor vehicle trajectory measurement and optimization method comprises the following steps: Step S1, measuring the vehicle trajectory based on an IMU and Beidou tightly coupled fusion vehicle trajectory measurement method to obtain an original vehicle trajectory; Step S2, performing potential path recovery on the original vehicle trajectory based on a trajectory enhancement strategy in the CL-TSim model to obtain an enhanced interactive trajectory; Step S3, using a double-channel trajectory encoder, spatiotemporal co-occurrence fusion is performed on the enhanced interaction trajectory to obtain a mobile feature representation vector. Step S4, a road network trajectory similarity measurement model is used to calculate the distance between the original vehicle trajectory and the mobile feature representation vector, and an optimized vehicle trajectory segment is screened out.

[0006] Preferably, the step S1 comprises the following sub-steps: Step S11, original Beidou observation data, three-axis acceleration and three-axis angular velocity are collected through a Beidou receiver and an IMU inertial measurement unit; Step S12, the original Beidou observation data, three-axis acceleration and three-axis angular velocity are subjected to precision optimization based on an IMU and Beidou tightly coupled fusion vehicle trajectory measurement method to obtain high-precision state data; Step S13, the high-precision state data is subjected to error correction using a Kalman filter to obtain an original vehicle trajectory; The original vehicle trajectory comprises a vehicle ID, a timestamp, latitude and longitude coordinates, instantaneous speed, vehicle driving direction, magnetic declination angle and elevation.

[0007] Preferably, the processing logic for precision optimization of the original Beidou observation data, three-axis acceleration and three-axis angular velocity based on the IMU and Beidou tightly coupled fusion vehicle trajectory measurement method is as follows: Preferably, the processing logic for precision optimization of the original Beidou observation data, three-axis acceleration and three-axis angular velocity based on the IMU and Beidou tightly coupled fusion vehicle trajectory measurement method is as follows: A unified time reference is established, the original Beidou observation data, three-axis acceleration and three-axis angular velocity are timestamped, and based on the fusion state at the previous time, state prediction is obtained by using the three-axis acceleration and three-axis angular velocity at the current time, the state prediction comprising a potential velocity, an attitude prediction sequence and a latitude and longitude prediction coordinate.

[0008] Preferably, the processing logic for error correction of the high-precision state data using the Kalman filter is as follows: The initial state vector is constructed by using the state prediction quantity, the pseudo-range prediction values to each Beidou satellite are calculated based on the INS predicted position in the state prediction quantity, the pseudo-range residual sequence is obtained by calculating the difference between the pseudo-range prediction values and the pseudo-range in the original Beidou observation data, the initial state vector is constructed by using the state prediction quantity, the pseudo-range prediction values to each Beidou satellite are calculated based on the INS predicted position in the state prediction quantity, the pseudo-range residual sequence is obtained by calculating the difference between the pseudo-range prediction values and the pseudo-range in the original Beidou observation data, and the pseudo-range residual sequence is taken as the measurement update value; the state update estimation processing is completed by using the Kalman filter based on the measurement update value, the error state is obtained, the error state is fed back to the INS, and the original vehicle trajectory is output.

[0009] Preferably, the step S2 specifically comprises: According to the latitude and longitude space range of the original vehicle trajectory, a vehicle grid is constructed, and the latitude and longitude coordinates of the continuous time stamps are mapped into discrete grid numbers to obtain a vehicle trajectory sequence. Based on the trajectory enhancement strategy in the CL-TSim model, potential enhancement processing is performed on the vehicle trajectory sequence to obtain an enhanced interactive trajectory, the enhanced interactive trajectory comprising the vehicle trajectory sequence, a first trajectory and a second trajectory, the first trajectory and the second trajectory being used to represent the same driving behavior of the vehicle on the same road but different positioning coordinates obtained.

[0010] Preferably, the processing logic of constructing the vehicle grid according to the latitude and longitude space range of the original vehicle trajectory and mapping the latitude and longitude coordinates of the continuous time stamps into discrete grid numbers is as follows: The maximum latitude and longitude values of the original vehicle trajectory are obtained, the current geographical area is divided into vehicle grids in a fixed threshold rule, the latitude and longitude coordinates of the original vehicle trajectory are mapped into the vehicle grid to obtain Beidou positioning points , the row index and the column index of the Beidou positioning points in the vehicle grid are calculated, and the calculation expression of the row index and the column index is as follows: , wherein, is the longitude number, is the latitude number, is the longitude value of the Beidou positioning point, is the accuracy span of one vehicle grid, is the minimum longitude value in the original vehicle trajectory, is the latitude value of the Beidou positioning point, is the minimum latitude value in the original vehicle trajectory, ​​​For a vehicle grid spanning latitudinal distances, latitude and longitude coordinates are sequentially converted into BeiDou positioning points according to the vehicle's direction of travel to obtain the vehicle trajectory sequence. .

[0011] Preferably, the logic for potential enhancement processing of vehicle trajectory sequences based on the trajectory enhancement strategy in the CL-TSim model is as follows: The vehicle trajectory sequence is downsampled uniformly and randomly to obtain the first enhanced trajectory. Second Enhanced Trajectory The trajectory points corresponding to the first and second enhanced trajectories are mapped to Mercator coordinates and random noise is added. The random noise is sampled from a normal distribution N(0,1) to obtain the first trajectory. Second trajectory The calculation expressions for the first trajectory and the second trajectory are: ; ; in, The scaling factor is set to 50 meters.

[0012] Preferably, step S3 specifically includes: The enhanced interactive trajectory and the original vehicle trajectory are spatiotemporally fused using a dual-channel trajectory encoder: The first and second trajectories are input into the grid channel of a dual-channel trajectory encoder. A lightweight grid encoder is used to sequentially encode the enhanced interactive trajectory to obtain the first grid vector. Second grid vector ; The vehicle trajectory sequence is input into the BeiDou channel of the dual-channel trajectory encoder. The BDS trajectory encoder based on the attention mechanism encodes the original vehicle trajectory into points to obtain the BDS vector. ; The first grid vector, the second grid vector, and the BDS vector are concatenated and fused through a fully connected layer to obtain the moving feature representation vector. The calculation expression for the moving feature representation vector is as follows: ; in, For the moving feature representation vector, This is the weight matrix of the fully connected layer. This is the concatenation result of the first grid vector, the second grid vector, and the BDS vector, where b is the bias term.

[0013] Preferably, the processing logic of the lightweight mesh encoder and the BDS trajectory encoder is as follows: The processing logic of the lightweight mesh encoder is as follows: The longitude and latitude numbers of the first and second trajectories are mapped to longitude embedding vectors and latitude embedding vectors, respectively. The longitude embedding vectors and latitude embedding vectors are then added together to obtain the embedding vector. To obtain the grid embedding vector, set its dimension to d, and add sine and cosine position encoding vectors to each grid embedding vector to obtain the updated grid embedding vector. The calculation expression for the updated grid embedding vector is as follows: ; ; ; Where k is the dimension index and i is the location index. For the updated grid embedding vector, This is a sinusoidal position encoding vector. This is the cosine position encoding vector; The grid embedding vectors are aggregated into a grid vector using mean pooling, and the grid vector includes a first grid vector and a second grid vector. The processing logic of the BDS trajectory encoder is as follows: The vehicle trajectory sequence is normalized and feature mapped. Sine and cosine position encoding vectors are added, and the outputs of all heads are concatenated using a multi-head attention mechanism to finally obtain the BDS vector.

[0014] Preferably, step S4 specifically includes: The original vehicle trajectory is divided into several sub-trajectories using a trajectory segmentation algorithm. A dual-channel encoder is used to perform vector encoding on the sub-trajectories to obtain sub-trajectory codes. The cosine similarity between each sub-trajectory code and the moving feature vector is calculated. The cosine similarity is used as the trajectory comprehensive score. The top s sub-trajectories with the highest trajectory comprehensive scores are selected as optimized trajectory segments to generate optimized vehicle trajectory segments.

[0015] The beneficial effects of this invention are as follows: By constructing a complete technical system from data acquisition to trajectory screening, it effectively solves three core problems in complex urban environments: unreliable vehicle trajectory data quality, missing semantic information, and difficulty in unified evaluation of multi-source heterogeneous data. This invention first adopts a tightly coupled fusion technology of IMU and BeiDou, and transforms the original sensor data into high-precision state data containing instantaneous velocity and attitude prediction sequences by establishing a unified time reference, coordinate system transformation, and inertial navigation calculation. Then, a pseudorange residual observation model is constructed using a Kalman filter to correct closed-loop errors, and finally, the original vehicle trajectory with both absolute accuracy and continuity is output. This basic processing step lays a reliable data foundation for subsequent in-depth optimization.

[0016] In the trajectory optimization stage, this invention introduces an innovative grid representation and contrastive learning mechanism. By discretizing continuous latitude and longitude coordinates into a grid number sequence, it achieves intelligent conversion from coordinate space to semantic space, effectively unifying the spatial scale differences between different devices. The trajectory enhancement strategy based on the CL-TSim model simulates real-world data variation through uniform downsampling and random downsampling. Combined with random noise injection in the Mercator coordinate system, it generates dual trajectory views that represent the same driving behavior but with different observation details. This enhancement process not only enriches the data diversity but also allows the model to focus on the essential movement semantics of the trajectory through the contrastive learning mechanism.

[0017] This method further designs a heterogeneous dual-channel trajectory encoder architecture. The grid channel enhances the trajectory through embedding representation and positional encoding, extracting robust semantic features. The BeiDou channel directly analyzes the original trajectory sequence using an attention mechanism, preserving fine spatiotemporal details. The features from the two channels are deeply fused through a fully connected layer to generate feature vectors that comprehensively represent vehicle movement patterns. Finally, through trajectory segmentation and semantic similarity measurement, intelligent filtering and optimization of sub-trajectory segments are achieved, outputting high-quality trajectory segments that best represent the vehicle's actual driving mode. This solves the problem of insufficient measurement accuracy of vehicle trajectory data, realizes the spatiotemporal co-occurrence pattern of vehicle movement, and estimates the mid-route location of undetected vehicles based on the vehicle's origin and destination points, ensuring road network matching in complex and remote environments. This technical solution significantly enhances the application value of trajectory data in intelligent transportation, autonomous driving, and other fields, providing reliable technical support for high-precision location services. Attached Figure Description

[0018] Figure 1 A flowchart illustrating the steps of a method for measuring and optimizing the trajectory of a motor vehicle based on the BeiDou satellite, provided as an embodiment of the present invention. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Example, refer to Figure 1 This paper provides a method for measuring and optimizing vehicle trajectories based on the BeiDou satellite system, including the following steps: Step S1: Measure the vehicle trajectory based on the vehicle trajectory measurement method of tight coupling and fusion of IMU and BeiDou to obtain the original vehicle trajectory; Step S2: Based on the trajectory enhancement strategy in the CL-TSim model, perform potential path recovery on the original vehicle trajectory to obtain the enhanced interactive trajectory. Step S3: Using a dual-channel trajectory encoder, the enhanced interactive trajectory is spatiotemporally co-occurred and fused to obtain a motion feature representation vector; Step S4: Calculate the distance between the original vehicle trajectory and the motion feature representation vector using the road network trajectory similarity measurement model, and select the optimized vehicle trajectory segment.

[0021] This invention further designs a heterogeneous dual-channel trajectory encoder architecture. The grid channel enhances the trajectory through embedding representation and position encoding, extracting robust semantic features. The Beidou channel directly analyzes the original trajectory sequence using an attention mechanism, preserving fine spatiotemporal details. The features of the two channels are deeply fused through a fully connected layer to generate a feature vector that comprehensively represents the vehicle's movement pattern. Finally, through trajectory segmentation and semantic similarity measurement, intelligent filtering and optimization of sub-trajectory segments are achieved, outputting the high-quality trajectory segments that best represent the vehicle's actual driving pattern. This solves the problem of insufficient measurement accuracy of vehicle trajectory data, realizes the spatiotemporal co-occurrence mode of vehicle movement, and estimates the mid-route location of undetected vehicles based on the vehicle's origin and destination points, ensuring road network matching in complex and remote environments. This embodiment significantly enhances the application value of trajectory data in intelligent transportation, autonomous driving, and other fields, providing reliable technical support for high-precision location services.

[0022] Step S1 includes the following sub-steps: Step S11: Collect raw BeiDou observation data, three-axis acceleration, and three-axis angular velocity through the BeiDou receiver and IMU inertial measurement unit; Step S12: Based on the vehicle trajectory measurement method of tight coupling and fusion of IMU and Beidou, the accuracy of the original Beidou observation data, three-axis acceleration and three-axis angular velocity are optimized to obtain high-precision state data; Step S13: Use a Kalman filter to correct errors in the high-precision state data to obtain the original vehicle trajectory; The original vehicle trajectory includes vehicle ID, timestamp, latitude and longitude coordinates, instantaneous speed, vehicle direction of travel, magnetic declination, and elevation.

[0023] This method acquires raw data synchronously via an onboard BeiDou receiver and an IMU (Integrated Measurement Unit). It utilizes heterogeneous and complementary data sources: BeiDou provides absolute, but potentially discontinuous or interfered-with, position and time references (latitude and longitude, UTC time), while the IMU provides relative, but high-frequency, motion increment information (three-axis acceleration, three-axis angular velocity) unaffected by external signals. The principle is to use the IMU's high integration accuracy over short periods to compensate for instantaneous gaps and jumps in the BeiDou signal. For example, when a vehicle enters a tunnel, the BeiDou signal completely fails, but the IMU can still continuously infer the vehicle's displacement and attitude changes through integrated acceleration and angular velocity, ensuring uninterrupted trajectory tracking.

[0024] In step S12, under a unified navigation coordinate system, the dynamic model of the IMU is deeply fused with the original observations of BeiDou (such as pseudorange). The underlying purpose of this step is to perform pre-filtering and state prediction at the data level to form a preliminary optimized high-precision state data. Through the mechanical arrangement of the inertial navigation system INS, the high-frequency measurement values ​​of the IMU are integrated into a continuous sequence of position, velocity and attitude predictions. In terms of top-level design, this approach binds a low-frequency absolute reference (BeiDou) with a high-frequency relative motion source (IMU), providing a high-frequency, continuous state prediction sequence for the next step of optimal estimation. A state vector containing error terms is constructed, and the pseudorange residuals generated by BeiDou observations are used as measurement updates. Optimal fusion is performed through Kalman gain. This step realizes a closed-loop error correction mechanism. The principle is that the filter continuously estimates the sensor zero bias and other error states of the IMU and feeds them back to the system model, thereby correcting the inertial calculation process in S12 in real time. This effectively suppresses the accumulation of IMU errors over time, which makes the output original vehicle trajectory not only highly accurate when the signal is good, but also greatly reduces the divergence speed of its calculated trajectory when the signal is briefly lost.

[0025] The processing logic for optimizing the accuracy of raw BeiDou observation data, three-axis acceleration, and three-axis angular velocity based on the tightly coupled fusion of IMU and BeiDou in vehicle trajectory measurement is as follows: The processing logic for optimizing the accuracy of raw BeiDou observation data, three-axis acceleration, and three-axis angular velocity based on the tightly coupled fusion of IMU and BeiDou in vehicle trajectory measurement is as follows: A unified time reference is established, and the original BeiDou observation data, three-axis acceleration, and three-axis angular velocity are timestamped. Based on the fusion state of the previous moment, the state prediction is obtained by using the three-axis acceleration and three-axis angular velocity of the current moment for inertial navigation calculation. The state prediction includes the following velocity, attitude prediction sequence, and latitude and longitude prediction coordinates.

[0026] By deeply collaborating with the inertial measurement unit (IMU) and the BeiDou system, a self-predictive and self-correcting perception system is established. Its advantage lies in organically combining the absolute positioning capability of satellites with the relative motion sensing capability of inertial devices, forming a complementary advantage. When the satellite signal is good, the inertial data can smooth and optimize the trajectory and provide rich attitude information. When the vehicle briefly enters a tunnel or under an overpass, causing the satellite signal to be interrupted, the system can immediately switch to pure inertial navigation mode and maintain the continuous output of the trajectory through integral motion sensor data to avoid positioning interruption. In addition, through a tightly coupled fusion algorithm, the system can finally output high-frequency trajectory points with the same frequency as the IMU, thereby more precisely depicting the instantaneous motion state of the vehicle.

[0027] The logic for correcting errors in high-precision state data using a Kalman filter is as follows: An initial state vector is constructed using state predictions. Based on the INS predicted positions in the state predictions, pseudorange predictions to each BeiDou satellite are calculated. The pseudorange predictions are subtracted from the pseudoranges in the original BeiDou observation data to obtain a pseudorange residual sequence. This pseudorange residual sequence is then used as the measurement update value. Based on the measurement update value, a Kalman filter is used to complete the state update estimation process to obtain the error state. The error state is then fed back into the INS, and the original vehicle trajectory is output.

[0028] By comparing the residual sequences generated from predicted pseudoranges and actual observations, the error trend of the inertial navigation system is accurately identified. These residual data are then weighted by the state covariance matrix and transformed into the optimal error state estimate. Employing a tightly coupled architecture, the system directly processes raw satellite observation data, enabling effective correction even in traditional positioning challenges where the number of visible satellites is insufficient.

[0029] Step S2 specifically includes: Based on the latitude and longitude spatial range of the original vehicle trajectory, a vehicle grid is constructed, and the latitude and longitude coordinates of continuous timestamps are mapped to discrete grid numbers to obtain the vehicle trajectory sequence. Based on the trajectory enhancement strategy in the CL-TSim model, the vehicle trajectory sequence is subjected to potential enhancement processing to obtain an enhanced interactive trajectory. The enhanced interactive trajectory includes a first trajectory and a second trajectory, which are used to represent the same driving behavior of the vehicle on the same road, but the obtained positioning coordinates are different.

[0030] Gridding maps continuous latitude and longitude coordinates to a standardized discrete grid number sequence, effectively unifying the spatial scale differences between devices with different precision. This method filters out minor fluctuations in positioning data, enabling the model to focus on the macroscopic semantic features of the movement path rather than getting bogged down in minor deviations at the coordinate level. The grid-encoded trajectory sequence retains the spatial topological relationship of the original path and has better noise resistance. By comparing and learning from dual trajectory views, the model can spontaneously identify trajectory patterns that differ in coordinates but are consistent in semantics. The resulting enhanced interactive trajectory provides an ideal data foundation for subsequent trajectory similarity calculation and semantic analysis.

[0031] Based on the latitude and longitude spatial range of the original vehicle trajectory, a vehicle grid is constructed, and the processing logic of mapping the latitude and longitude coordinates of continuous timestamps to discrete grid numbers is as follows: Obtain the maximum and minimum latitude and longitude values ​​of the original vehicle trajectory, divide the current geographical area into vehicle grids with fixed threshold rules, and then obtain the latitude and longitude coordinates of the original vehicle trajectory. Mapped onto the vehicle grid, BeiDou positioning points are obtained. Calculate the row and column indices of the BeiDou positioning point in the vehicle grid. The expressions for calculating the row and column indices are as follows: ; ; in, Longitude numbering Latitude numbering The longitude value of the BeiDou positioning point. For the precision span of a vehicle grid, This represents the minimum longitude value in the original vehicle trajectory. The latitude value of the BeiDou positioning point. This represents the minimum latitude value in the original vehicle trajectory. For a vehicle grid spanning latitudinal distances, latitude and longitude coordinates are sequentially converted into BeiDou positioning points according to the vehicle's direction of travel to obtain the vehicle trajectory sequence. .

[0032] By calculating the extreme values ​​of latitude and longitude of the trajectory coverage area and dynamically dividing it into grids, the system can adapt to different spatial scales of different journeys, ensuring that the grid number sequence accurately reflects the actual movement path of the vehicle. Each grid becomes the basic semantic unit of spatial analysis, and the vehicle trajectory is elegantly expressed as the temporal arrangement of these grid units. Trajectory data from devices with different sampling frequencies and different precisions are unified into the same grid space for comparison and analysis, effectively eliminating the data heterogeneity problem caused by device differences. At the same time, the discrete grid number sequence is easier for machine learning models to process than continuous floating-point coordinates.

[0033] Based on the trajectory enhancement strategy in the CL-TSim model, the logic for potential enhancement processing of vehicle trajectory sequences is as follows: The vehicle trajectory sequence is downsampled uniformly and randomly to obtain the first enhanced trajectory. Second Enhanced Trajectory The trajectory points corresponding to the first and second enhanced trajectories are mapped to Mercator coordinates and random noise is added. The random noise is sampled from a normal distribution N(0,1) to obtain the first trajectory. Second trajectory The expressions for calculating the first and second trajectories are: ; ; in, The scaling factor is set to 50 meters.

[0034] Uniform downsampling simulates the fixed sampling rate difference between different devices, which enables the model to understand the behavior of the same path under different sampling frequencies. Random downsampling restores the data packet loss phenomenon caused by signal blockage or equipment failure in the actual system, and trains the model to have the reasoning ability to deal with incomplete trajectory data. The two downsampling strategies work together to ensure the high compatibility of the model with different sampling conditions. Adding normally distributed random noise to Mercator coordinates accurately reproduces the random error characteristics present in the positioning system. Choosing 50 meters as the scaling factor for the noise preserves the overall morphological characteristics of the trajectory while introducing reasonable coordinate-level perturbations. This approach forces the model to learn to distinguish between the essential motion pattern of the trajectory and surface coordinate fluctuations during training, thereby cultivating a robust feature extraction capability that is insensitive to noise. The most valuable innovation of this method lies in constructing a dual-track view that represents the same trip but with different observation details. Although the first and second tracks originate from the same movement behavior, they have reasonable differences in sampling patterns and noise performance. This allows the model to spontaneously understand the stable and unchanging semantic core in the trajectory data by comparing the two enhanced views. Through this enhancement process, the model no longer relies excessively on specific coordinate values ​​or sampling point density, but learns to focus on the macroscopic motion patterns and topological features of the trajectory. This capability enables the method to maintain stable analytical performance when facing different brands of positioning devices, different urban environments, and different signal conditions, providing a solid technical guarantee for large-scale intelligent trajectory analysis applications.

[0035] Step S3 specifically includes: The enhanced interactive trajectory and the original vehicle trajectory are spatiotemporally fused using a dual-channel trajectory encoder: The first and second trajectories are input into the grid channel of a dual-channel trajectory encoder. A lightweight grid encoder is used to sequentially encode the enhanced interactive trajectory to obtain the first grid vector. Second grid vector ; The vehicle trajectory sequence is input into the BeiDou channel of the dual-channel trajectory encoder. The BDS trajectory encoder based on the attention mechanism encodes the original vehicle trajectory into points to obtain the BDS vector. ; The first grid vector, the second grid vector, and the BDS vector are concatenated and fused through a fully connected layer to obtain the moving feature representation vector. The expression for calculating the moving feature representation vector is as follows: ; in, For the moving feature representation vector, This is the weight matrix of the fully connected layer. This is the concatenation result of the first grid vector, the second grid vector, and the BDS vector, where b is the bias term.

[0036] Through the synergistic effect of dual-channel trajectory encoders, multi-dimensional feature extraction and deep fusion of vehicle movement behavior are achieved. Its advantage lies in its ability to process the enhanced semantic trajectory after gridding and the original BeiDou positioning trajectory in parallel, thereby simultaneously capturing macroscopic semantic information and microscopic positioning features of the movement pattern. The grid channel is specifically designed to process the enhanced first and second trajectories. A lightweight encoder transforms the grid sequence into semantically representative feature vectors. The gridded and enhanced trajectory sequence can effectively filter out localization noise, allowing the encoder to focus on the path's topology and movement patterns. The BeiDou channel directly processes the original vehicle trajectory sequence. Its attention-based encoder autonomously identifies the differences in importance between different points in the trajectory. This design allows the model to focus on representative key trajectory points, such as turns or path intersections, while downplaying the role of redundant points on straight sections. This channel preserves the fine spatiotemporal information of the original trajectory, supplementing valuable detailed data for motion features. The purpose of the dual-channel structure is to achieve a complementary effect of features. The semantic features provided by the grid channel and the detailed features extracted by the Beidou channel are spliced ​​together and fused with the fully connected layer to form a comprehensive and robust motion feature representation vector. The representation vector contains both the macroscopic motion intention of the trajectory and retains important microscopic motion details, providing a rich and accurate feature foundation for subsequent similarity calculation.

[0037] The processing logic for the lightweight mesh encoder and the BDS track encoder is as follows: The processing logic of the lightweight mesh encoder is as follows: The longitude and latitude numbers of the first and second trajectories are mapped to longitude embedding vectors and latitude embedding vectors, respectively. The longitude embedding vectors and latitude embedding vectors are then added together to obtain the embedding vector. To obtain the grid embedding vector, set its dimension to d, and add sine and cosine position encoding vectors to each grid embedding vector to obtain the updated grid embedding vector. The expression for calculating the updated grid embedding vector is: ; ; ; Where k is the dimension index and i is the location index. For the updated grid embedding vector, This is a sinusoidal position encoding vector. This is the cosine position encoding vector; The grid embedding vectors are aggregated into a grid vector using mean pooling. The grid vector includes a first grid vector and a second grid vector. The processing logic of the BDS trajectory encoder is as follows: The vehicle trajectory sequence is normalized and feature mapped. Sine and cosine position encoding vectors are added, and the outputs of all heads are concatenated using a multi-head attention mechanism to finally obtain the BDS vector.

[0038] The Beidou trajectory encoder uses an advanced attention mechanism to process the raw trajectory data. After normalizing the vehicle trajectory sequence, it introduces position encoding and then uses a multi-head attention mechanism to capture the long-range dependencies between trajectory points. This method enables the model to autonomously identify key points in the trajectory and pay more attention to important trajectory segments, thereby extracting fine local motion features. The collaborative work of the two encoders constitutes the core advantage of this method. The semantic features provided by the grid encoder complement the detailed features extracted by the Beidou encoder, which not only grasps the overall trend of the movement behavior, but also retains the key motion details. This dual representation capability enables the system to understand both the macroscopic path selection and microscopic motion changes of the vehicle.

[0039] Step S4 specifically includes: The original vehicle trajectory is divided into several sub-trajectories using a trajectory segmentation algorithm. A dual-channel encoder is used to perform vector encoding on the sub-trajectories to obtain sub-trajectory codes. The cosine similarity between each sub-trajectory code and the moving feature vector is calculated. The cosine similarity is used as the trajectory comprehensive score. The top s sub-trajectories with the highest trajectory comprehensive scores are selected as optimized trajectory segments to generate optimized vehicle trajectory segments.

[0040] By using a dual-channel encoder to vectorize each sub-trajectory, the system obtains a sub-trajectory representation that includes both grid semantic information and retains the original point features, ensuring a comprehensive characterization of the sub-trajectory motion pattern. Calculating the cosine similarity between the sub-trajectory code and the overall movement feature vector is essentially an assessment of the consistency between each local trajectory segment and the global movement pattern. The semantic similarity-based assessment mechanism can effectively identify high-quality trajectory segments that are highly consistent with the overall driving pattern, while filtering out low-quality data segments caused by signal interference or positioning anomalies. By selecting the top few sub-trajectories with the highest similarity as optimized trajectory segments, a data-driven trajectory self-purification process is essentially completed. This method not only retains the trajectory segments that best represent the vehicle's actual driving path, but also automatically eliminates unreliable data segments that are contaminated by noise or have abnormal fluctuations. The optimized trajectory segments generated by this technical approach can more accurately reflect the vehicle's actual driving behavior, opening up a new technical path for achieving precise trajectory analysis in intelligent transportation systems.

[0041] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0042] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for measuring and optimizing vehicle trajectories based on BeiDou satellites, characterized in that, Includes the following steps: Step S1: Measure the vehicle trajectory based on the vehicle trajectory measurement method of tight coupling and fusion of IMU and BeiDou to obtain the original vehicle trajectory; Step S2: Based on the trajectory enhancement strategy in the CL-TSim model, perform potential path recovery on the original vehicle trajectory to obtain the enhanced interactive trajectory. Step S3: Using a dual-channel trajectory encoder, the enhanced interactive trajectory is spatiotemporally co-occurred and fused to obtain a motion feature representation vector; Step S4: Calculate the distance between the original vehicle trajectory and the motion feature representation vector using the road network trajectory similarity measurement model, and select the optimized vehicle trajectory segment.

2. The method for measuring and optimizing vehicle trajectory based on BeiDou satellite as described in claim 1, characterized in that, Step S1 includes the following sub-steps: Step S11: Collect raw BeiDou observation data, three-axis acceleration, and three-axis angular velocity through the BeiDou receiver and IMU inertial measurement unit; The raw BeiDou observation data includes pseudorange, carrier phase, and Doppler shift. Step S12: Based on the vehicle trajectory measurement method of tight coupling and fusion of IMU and Beidou, the accuracy of the original Beidou observation data, three-axis acceleration and three-axis angular velocity are optimized to obtain high-precision state data; Step S13: Use a Kalman filter to correct errors in the high-precision state data to obtain the original vehicle trajectory; The original vehicle trajectory includes vehicle ID, timestamp, latitude and longitude coordinates, instantaneous speed, vehicle direction of travel, magnetic declination, and elevation.

3. The method for measuring and optimizing vehicle trajectory based on BeiDou satellite as described in claim 1, characterized in that, The processing logic for optimizing the accuracy of raw BeiDou observation data, three-axis acceleration, and three-axis angular velocity based on the tightly coupled fusion of IMU and BeiDou in vehicle trajectory measurement is as follows: A unified time reference is established, and the original BeiDou observation data, three-axis acceleration, and three-axis angular velocity are timestamped. Based on the fusion state of the previous moment, the state prediction is obtained by inertial navigation calculation using the three-axis acceleration and three-axis angular velocity of the current moment. The state prediction includes the following velocity, attitude prediction sequence, and latitude and longitude prediction coordinates.

4. The method for measuring and optimizing vehicle trajectory based on BeiDou satellite as described in claim 1, characterized in that, The logic for correcting errors in high-precision state data using a Kalman filter is as follows: An initial state vector is constructed using state predictions. Based on the INS predicted positions in the state predictions, pseudorange predictions to each BeiDou satellite are calculated. The pseudorange predictions are subtracted from the pseudoranges in the original BeiDou observation data to obtain a pseudorange residual sequence. This pseudorange residual sequence is then used as the measurement update value. Based on the measurement update value, a Kalman filter is used to complete the state update estimation process to obtain the error state. The error state is then fed back into the INS, and the original vehicle trajectory is output.

5. The method for measuring and optimizing vehicle trajectory based on BeiDou satellite as described in claim 2, characterized in that, Step S2 specifically includes: Based on the latitude and longitude spatial range of the original vehicle trajectory, a vehicle grid is constructed, and the latitude and longitude coordinates of continuous timestamps are mapped to discrete grid numbers to obtain the vehicle trajectory sequence. Based on the trajectory enhancement strategy in the CL-TSim model, the vehicle trajectory sequence is subjected to potential enhancement processing to obtain an enhanced interactive trajectory. The enhanced interactive trajectory includes the vehicle trajectory sequence, a first trajectory, and a second trajectory. The first trajectory and the second trajectory are used to represent the same driving behavior of the vehicle on the same road, but the obtained positioning coordinates are different.

6. The method for measuring and optimizing vehicle trajectory based on BeiDou satellite as described in claim 3, characterized in that, Based on the latitude and longitude spatial range of the original vehicle trajectory, a vehicle grid is constructed, and the processing logic of mapping the latitude and longitude coordinates of continuous timestamps to discrete grid numbers is as follows: Obtain the maximum and minimum latitude and longitude values ​​of the original vehicle trajectory, divide the current geographical area into vehicle grids with fixed threshold rules, and then obtain the latitude and longitude coordinates of the original vehicle trajectory. Mapped onto the vehicle grid, BeiDou positioning points are obtained. Calculate the row and column indices of the BeiDou positioning point in the vehicle grid. The expressions for calculating the row and column indices are as follows: ； ； in, Longitude numbering Latitude numbering The longitude value of the BeiDou positioning point. For the precision span of a vehicle grid, This represents the minimum longitude value in the original vehicle trajectory. The latitude value of the BeiDou positioning point. This represents the minimum latitude value in the original vehicle trajectory. For a vehicle grid spanning latitudinal distances, latitude and longitude coordinates are sequentially converted into BeiDou positioning points according to the vehicle's direction of travel to obtain the vehicle trajectory sequence. .

7. The method for measuring and optimizing vehicle trajectory based on BeiDou satellite as described in claim 4, characterized in that, Based on the trajectory enhancement strategy in the CL-TSim model, the logic for potential enhancement processing of vehicle trajectory sequences is as follows: The vehicle trajectory sequence is downsampled uniformly and randomly to obtain the first enhanced trajectory. Second Enhanced Trajectory The trajectory points corresponding to the first and second enhanced trajectories are mapped to Mercator coordinates and random noise is added. The random noise is sampled from a normal distribution N(0,1) to obtain the first trajectory. Second trajectory The calculation expressions for the first trajectory and the second trajectory are as follows: ； ； in, The scaling factor is set to 50 meters.

8. The method for measuring and optimizing vehicle trajectory based on BeiDou satellite as described in claim 5, characterized in that, Step S3 specifically includes: A dual-channel trajectory encoder is used to perform spatiotemporal co-occurrence fusion of the enhanced interactive trajectory and the original vehicle trajectory: The first and second trajectories are input into the grid channel of a dual-channel trajectory encoder. A lightweight grid encoder is used to sequentially encode the enhanced interactive trajectory to obtain the first grid vector. Second grid vector ; The vehicle trajectory sequence is input into the BeiDou channel of the dual-channel trajectory encoder. The BDS trajectory encoder based on the attention mechanism encodes the original vehicle trajectory into points to obtain the BDS vector. ; The first grid vector, the second grid vector, and the BDS vector are concatenated and fused through a fully connected layer to obtain the moving feature representation vector. The calculation expression for the moving feature representation vector is as follows: ； in, For the moving feature representation vector, This is the weight matrix of the fully connected layer. This is the concatenation result of the first grid vector, the second grid vector, and the BDS vector, where b is the bias term.

9. The method for measuring and optimizing vehicle trajectory based on BeiDou satellite as described in claim 8, characterized in that, The processing logic of the lightweight mesh encoder and the BDS trajectory encoder is as follows: The processing logic of the lightweight mesh encoder is as follows: The longitude and latitude numbers of the first and second trajectories are mapped to longitude embedding vectors and latitude embedding vectors, respectively. The longitude embedding vectors and latitude embedding vectors are then added together to obtain the embedding vector. To obtain the grid embedding vector, set its dimension to d, and add sine and cosine position encoding vectors to each grid embedding vector to obtain the updated grid embedding vector. The calculation expression for the updated grid embedding vector is as follows: ； ； ； Where k is the dimension index and i is the location index. For the updated grid embedding vector, This is a sinusoidal position encoding vector. This is the cosine position encoding vector; The grid embedding vectors are aggregated into a grid vector using mean pooling, and the grid vector includes a first grid vector and a second grid vector. The processing logic of the BDS trajectory encoder is as follows: The vehicle trajectory sequence is normalized and feature mapped. Sine and cosine position encoding vectors are added, and the outputs of all heads are concatenated using a multi-head attention mechanism to finally obtain the BDS vector.

10. The method for measuring and optimizing vehicle trajectory based on BeiDou satellite as described in claim 1, characterized in that, Step S4 specifically includes: The original vehicle trajectory is divided into several sub-trajectories using a trajectory segmentation algorithm. A dual-channel encoder is used to perform vector encoding on the sub-trajectories to obtain sub-trajectory codes. The cosine similarity between each sub-trajectory code and the moving feature vector is calculated. The cosine similarity is used as the trajectory comprehensive score. The top s sub-trajectories with the highest trajectory comprehensive scores are selected as optimized trajectory segments to generate optimized vehicle trajectory segments.

Images