A vehicle trajectory data-driven vehicle lane-changing start and end point identification method
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GUANGZHOU UNIVERSITY
- Filing Date
- 2023-01-09
- Publication Date
- 2026-06-09
Smart Images
Figure CN116092033B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of trajectory data processing, and particularly relates to a method for identifying the start and end points of vehicle lane changes driven by vehicle trajectory data. Background Technology
[0002] When making lane-changing decisions, drivers need to comprehensively consider information about the current lane, the target lane, and adjacent vehicles. Compared to car-following, lane-changing behavior is more complex and riskier. Research on lane-changing behavior can significantly improve traffic safety and road capacity. Accurate identification and extraction of the start and end points of lane changes are of great importance for determining lane-changing time and ensuring lane-changing safety.
[0003] Lane changing is an important driving behavior during vehicle operation, and many studies have extracted and predicted it. Currently, a popular approach to extracting the start and end points of lane changes is to use machine learning to determine these points and predict the trajectory, providing technical support for autonomous driving. Other studies use the speeds of the vehicle ahead in the target lane and the current lane as criteria for determining the lane-changing intention of autonomous vehicles, employing vehicle dynamics to assess lane-changing safety.
[0004] Existing research on extracting the start and end points of lane changes from trajectory data is largely based on empirical judgments, lacking experimentally validated models for this purpose. Data collection on lane-changing behavior primarily focuses on simulated lane-changing driving by experimenters and data from autonomous vehicles, but the results are difficult to generalize to practical applications. However, using natural vehicle trajectory data can intuitively represent driver lane-changing behavior. Reflecting real traffic conditions through vehicle trajectory data helps extend lane-changing start and end point models to related research fields, providing theoretical support.
[0005] Existing research on extracting lane change start and end points from trajectory data mainly relies on empiricism, such as using empirically defined limits for lateral displacement changes, vehicle acceleration variations, and lane line and vehicle position conditions to determine lane change start and end points. However, in most trajectory datasets, lane line coordinates are difficult to obtain and have poor accuracy. Furthermore, changes in lane IDs in trajectory datasets do not necessarily indicate that a vehicle is crossing a lane line; lane change start and end points are located before and after lane ID changes. In addition, empiricist judgment limits may not be applicable to all scenarios, and the accuracy of their data collection needs further verification. Summary of the Invention
[0006] The purpose of this invention is to provide a vehicle trajectory data-driven method for identifying the start and end points of lane changes. Based on the natural driving trajectory data of vehicles, a model for extracting the start and end points of lane changes based on frame difference and lateral displacement change is established. The parameters of the model are calibrated and verified through experiments. This model can be used to collect the start and end points of lane changes in trajectory datasets. At the same time, this invention uses a massive vehicle trajectory dataset as the data source. The dataset can be obtained through UAV data collection and image recognition trajectory. More vehicle trajectory data can be incorporated in the future. This invention has broad applicability in the future.
[0007] To achieve the above objectives, the present invention provides the following technical solution:
[0008] A vehicle trajectory data-driven method for identifying the start and end points of lane changes includes the following steps:
[0009] S1. Process vehicle trajectory data;
[0010] S2. Construct a lane change start and end point extraction model based on frame difference and lateral displacement change.
[0011] S3. Use HighD visualization simulation to calibrate the parameters of the model;
[0012] S4. Compare, analyze, and verify the experimental results of the model.
[0013] S5. Analyze the characteristics of lane change time distribution.
[0014] Furthermore, the vehicle trajectory data is lane-changing vehicle trajectory data in the HighD dataset. The HighD dataset provides four types of files for each record. The four types of files include an image of the recorded highway segment, a CSV file describing the recorded location, a CSV file recording an overview of the vehicle trajectory, and a CSV file containing detailed vehicle trajectory data.
[0015] The HighD visualization simulation is a function repository provided by the HighD dataset. It can be run using Python or Matlab to visualize the trajectory data in the HighD dataset and reproduce the vehicle trajectory of road segments frame by frame. It is a publicly available visualization function repository.
[0016] The HighD dataset is a publicly available dataset of vehicle trajectory data collected by drones in Cologne, Germany.
[0017] The CSV file is a file that stores tabular data in plain text format.
[0018] Furthermore, the processing of vehicle trajectory data specifically involves using the four types of files as a basis and employing Python 3.8 to preprocess the trajectory data for each record. The preprocessing specifically includes:
[0019] S101. Summarize the recorded information in the HighD dataset and calculate the total traffic volume and vehicle type ratio information contained in the HighD dataset;
[0020] S102. Using the vehicle lane change frequency information provided in the CSV file that records the vehicle trajectory overview, filter out vehicles with single lane change behavior, and summarize the corresponding category files according to different number of lanes, different speed limits, different vehicle types, different vehicle type ratios, and different driving directions.
[0021] S103. Extract the CSV file of detailed trajectory data of the lane-changing vehicle, match the corresponding information by the vehicle ID information around each target vehicle, the information includes horizontal and vertical coordinates, horizontal and vertical velocity and acceleration, vehicle length and width, and merge the matched information with the data of each frame of the target vehicle.
[0022] S104. Remove data with incomplete lane change trajectories and abnormal start and end points of trajectories;
[0023] S105. Based on the characteristics of driving speed under different driving directions, determine the direction of vehicle lane change and classify the vehicle trajectory data for lane change to the left and lane change to the right.
[0024] S106. Collect traffic parameter information of vehicles changing lanes, including lateral displacement change, minimum speed, maximum speed and average speed.
[0025] Furthermore, the lane change start and end point extraction model is specifically as follows:
[0026]
[0027] Where t represents the number of frames from the start to the end of the lane change for each vehicle, F1 represents the first frame of each vehicle's trajectory data, F2 represents the number of frames when the lane ID changes in each vehicle's trajectory data, F3 represents the last frame of each vehicle's trajectory data, and X... t With X t-T These represent the lateral coordinates of the vehicle at times t and tT, respectively. T represents the frame difference, and S represents the limit value of the lateral displacement change. S and T need to be calibrated experimentally.
[0028] Furthermore, the lane change start and end point extraction model uses the frame difference T to calculate the lateral displacement change corresponding to each frame. Using the moment when the lane ID information changes, F2, as the boundary, it traces backward from F2 to extract the moment when the lateral displacement change of the vehicle within the time period F1-F2 is less than or equal to the limit value S, and marks this moment as the lane change start point. It traces backward from F2 to extract the moment when the lateral displacement change of the vehicle within the time period F2-F3 is less than or equal to the limit value S, and marks this moment as the lane change end point.
[0029] The accuracy of the lane change start and end point extraction model depends on the calibration of the frame difference T and the lateral displacement change S.
[0030] Furthermore, the parameter calibration of the model using HighD visualization simulation specifically involves:
[0031] S301. Set up an experimental vehicle and design different experimental schemes based on special circumstances, including different driving directions, different number of lanes, different vehicle types, and different lane-changing directions.
[0032] S302. The start and end points of lane changes for each experimental vehicle are recorded using HighD visualization simulation, which serves as a basis for judging the accuracy of each experimental scheme.
[0033] S303. Calculate the lateral displacement change of the test vehicle under different experimental schemes, and draw the distribution diagram of the lateral displacement change of each test vehicle.
[0034] S304. By analyzing the changes in the lateral displacement of the experimental vehicle, formulate rules for judging the start and end points of lane changes, and compare the differences in lane change times among different groups of experimental vehicles.
[0035] S305. Compare the accuracy of different experimental schemes and calibrate the start and end points of lane changes to extract model parameters.
[0036] S306. Verify different experimental schemes to check the accuracy of the lane change start and end point extraction model.
[0037] Furthermore, the comparative analysis and verification of the model experiment results specifically involves comparing the accuracy of the different experimental schemes, whereby the accuracy is the ratio of the lane change time collected by the simulation to the lane change time collected by the model.
[0038] Furthermore, the closer the ratio is to 1, the more accurate the model's acquisition of lane change start and end points.
[0039] Furthermore, the frame difference T includes 0.2s, 0.4s, 0.6s, 0.8s, and 1.0s. The optimal frame difference for the lane change start-end point extraction model is 0.2 seconds, and the optimal limit for lateral displacement change is 0.05 meters.
[0040] Furthermore, the dataset includes data on 6,506 vehicles with complete lane-changing trajectories for a single lane change.
[0041] The beneficial technical effects of the present invention are at least as follows:
[0042] 1. This invention uses vehicle natural driving trajectory data as a dataset, serving as the data basis for calibrating the model parameters for extracting lane change start and end points;
[0043] 2. A lane change start and end point extraction model based on frame difference and lateral displacement change was established. The parameters of the model were calibrated and verified through experiments. The optimal frame difference was determined to be 0.2 seconds and the optimal lateral displacement change limit was determined to be 0.05 meters.
[0044] 3. By identifying and extracting the lane change start and end point model, the duration of the lane change process (lane change time) of vehicles on the highway is determined to be in the range of 2.88-7.32 seconds. Attached Figure Description
[0045] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.
[0046] Figure 1 This is a flowchart of a vehicle lane change origin and destination identification method driven by vehicle trajectory data according to an embodiment of the present invention.
[0047] Figure 2 This is a diagram showing the lateral displacement changes of some experimental vehicles in an embodiment of the present invention.
[0048] Figure 3 This is a precision distribution chart for each experimental group in the embodiments of the present invention.
[0049] Figure 4 This is a scatter plot of lane-changing time distribution according to an embodiment of the present invention.
[0050] Figure 5 This is a diagram showing the distribution of lane-changing time boxes in an embodiment of the present invention. Detailed Implementation
[0051] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.
[0052] Example
[0053] like Figure 1As shown, the vehicle trajectory data-driven vehicle lane change start and end point identification method provided by the present invention includes the following steps:
[0054] S1. Process the vehicle trajectory data.
[0055] Constructing a lane change origin-end point extraction model requires supporting real-world road network data and real-world traffic flow data. This technique uses lane change vehicle trajectory data from the HighD dataset as a foundation to study methods for reasonably calibrating the parameters of the lane change origin-end point extraction model. The HighD dataset provides four types of files for each record: an image recording the road segment, a CSV file describing the record location, a CSV file summarizing the vehicle trajectory, and a CSV file containing detailed vehicle trajectory data. The HighD visualization simulation is a function repository provided by the HighD dataset. Run using Python or Matlab, it can visualize the trajectory data from the HighD dataset, reproducing the vehicle trajectory situation frame by frame. It is a publicly available visualization function repository.
[0056] This technique extracts lane change trajectory data. Based on these four types of files, Python 3.8 is used to preprocess the trajectory data for each record. The specific operations are as follows:
[0057] ① Summarize the information recorded in 60 datasets and statistically analyze the total traffic volume, vehicle type ratio, and other information contained in HighD;
[0058] ② By using the vehicle lane change frequency information provided in the vehicle trajectory overview file, vehicles with single lane change behavior are filtered out, and corresponding category files are summarized according to different number of lanes, different speed limits, different vehicle types, different vehicle type ratios, and different driving directions;
[0059] ③ Extract detailed trajectory data of vehicles changing lanes. By matching the vehicle ID information around each target vehicle with the corresponding horizontal and vertical coordinates, horizontal and vertical velocity and acceleration, vehicle length and width, etc., the matched information is merged with the data of each frame of the target vehicle.
[0060] ④ Remove data with incomplete lane change trajectories or abnormal start and end points;
[0061] ⑤ Based on the characteristics of driving speed under different driving directions, determine the direction of vehicle lane change and classify the vehicle trajectory data for lane change to the left and lane change to the right.
[0062] ⑥ Collect traffic parameter information such as the change in lateral displacement of vehicles changing lanes, minimum speed, maximum speed, and average speed.
[0063] The HighD dataset is a publicly available dataset of vehicle trajectory data collected by drones in Cologne, Germany.
[0064] HighD visualization simulation is a function repository provided by the HighD dataset. It can be run using Python or Matlab to visualize the trajectory data in the HighD dataset and reproduce the vehicle trajectory of road segments frame by frame. It is a publicly available visualization function repository.
[0065] CSV is a file format that stores tabular data in plain text.
[0066] S2. Construct a lane change start and end point extraction model based on frame difference and lateral displacement change.
[0067] During lane changing, the change in lateral displacement of a vehicle is often normally or skewed. The distribution of the change in lateral displacement also differs under different frame rate differences. Based on the frame rate difference and the change in lateral displacement, this invention proposes a lane change start and end point extraction model as shown in Formula 1:
[0068]
[0069] In Formula 1, t represents the number of frames from the start to the end of the lane change for each vehicle, F1 represents the first frame of the trajectory data for each vehicle, F2 represents the number of frames when the lane ID changes in the trajectory data for each vehicle, F3 represents the last frame of the trajectory data for each vehicle, and X... t With X t-T These represent the lateral coordinates of the vehicle at times t and tT, respectively. T represents the frame difference, and S represents the limit value of the lateral displacement change. S and T need to be calibrated experimentally.
[0070] The model calculates the lateral displacement change for each frame using the frame difference T. Taking the lane ID change at time F2 as the boundary, it traces backward from F2, extracting the moment when the lateral displacement change is less than or equal to the threshold value S within the time interval F1-F2, and marking this moment as the lane change start point. Tracing backward from F2, it extracts the moment when the lateral displacement change is less than or equal to the threshold value S within the time interval F2-F3, and marks this moment as the lane change end point. To ensure the accuracy of the lane change end point, after each extraction of the end point, it checks whether the lateral displacement change remains less than the threshold value S for the next 10 frames. If it is not consistently less than the threshold value, the lane change is not yet complete, and the maximum lateral displacement change within these 10 frames is used as the new lane change end point. This process is repeated until the lateral displacement change remains less than the threshold value for the next 10 frames after the lane change end point is extracted. This minimum frame number is then used as the minimum number of frames for the lane change end point.
[0071] S3. Use HighD visualization simulation to calibrate the parameters of the model.
[0072] The accuracy of the lane change start / end point extraction model based on frame difference and lateral displacement change depends on the calibration of the frame difference T and the lateral displacement change S. This paper uses HighD visualization simulation to calibrate the model parameters. The experimental steps are as follows:
[0073] ① The experimental vehicle was set up based on four main scenarios: different driving directions, different numbers of lanes, different vehicle types, and different lane-changing directions. Experimental vehicles were set up according to these four scenarios.
[0074] ② The start and end points of lane changes for each experimental vehicle were recorded using HighD visualization simulation, which served as the basis for judging the accuracy of each experimental scheme;
[0075] ③ Calculate the lateral displacement change of the test vehicle under different experimental schemes, and draw the distribution diagram of the lateral displacement change of each test vehicle;
[0076] ④ By analyzing the changes in the lateral displacement of the experimental vehicle, rules for judging the start and end points of lane changes were formulated, and the differences in lane change times among different groups of experimental vehicles were compared.
[0077] ⑤ Compare the accuracy of each scheme, calibrate the start and end points of lane changes, and extract model parameters;
[0078] ⑥ Verify the scheme and check the accuracy of the lane change start and end point extraction model.
[0079] Since the HighD frame rate is 25Hz, this invention presets the frame difference T to 0.2s, 0.4s, 0.6s, 0.8s, and 1.0s, and observes the distribution of lateral displacement changes in the experimental vehicle. The lateral displacement changes of some experimental vehicles are shown below. Figure 2 As shown.
[0080] The HighD dataset uses the top-left corner as the origin and divides the lanes into two parts representing different driving directions, hereinafter referred to as uphill and downhill. Lateral displacement change is calculated by the change in the vehicle's lateral coordinate within each frame difference. Due to the coordinate system, for uphill vehicles, the vehicle speed is positive, and the lateral displacement change is positive when the vehicle turns left and negative when it turns right; for downhill vehicles, the vehicle speed is negative, and the lateral displacement change is negative when the vehicle turns left and positive when it turns right. Figure 1 The distribution of lateral displacement changes of some experimental vehicles under different frame rate differences is shown. The distribution trend of lateral displacement changes is basically the same under different frame rate differences, but there are significant differences in the magnitude of the changes; the larger the frame rate difference, the greater the lateral displacement change. Based on the analysis of the distribution of lateral displacement changes under different frame rate differences, fifteen experimental schemes were developed, as shown in Table 1.
[0081] Table 1. Classification of Experimental Schemes
[0082]
[0083] According to the experimental scheme in Table 1, the present invention sets up a corresponding experimental vehicle in each scheme, extracts the start and end points of lane changing by the frame difference and the boundary of lateral displacement change in each group of experimental schemes, and calculates the lane changing time accordingly. The extracted lane changing time is compared with the lane changing time collected by the visualization simulation to determine the accuracy of each group of experimental schemes.
[0084] S4. Compare, analyze, and verify the experimental results of the model.
[0085] Figure 3 The accuracy of fifteen experimental schemes is shown. The accuracy is the ratio of the lane change time collected by simulation to the lane change time collected by model. The average of all accuracies in each group is used as the accuracy of the entire experimental group. The closer the ratio is to 1, the more accurate the start and end points of the lane change collected by the model are.
[0086] from Figure 3 It can be seen that the model accuracy decreases with the increase of the frame rate difference. Under the same frame rate difference, the change in the limit of the lateral displacement change also affects the model accuracy. Overall, the model accuracy first increases and then decreases with the increase of the lateral displacement change. To better compare the changes in accuracy of different experimental groups, the top seven groups in terms of accuracy were compared in detail, and the results are shown in Table 2.
[0087] Table 2 Comparison of Experimental Results
[0088]
[0089] Table 2 shows the accuracy of lane change start and end point acquisition for different groups under different conditions. The experimental groups with a frame difference of 0.2s and a threshold of 0.05m and a frame difference of 0.4s and a threshold of 0.10m have the closest average accuracy to 1. However, the experimental group with a frame difference of 0.4s and a threshold of 0.10m performs poorly in acquiring data for lane changes: cars changing left in a six-lane uphill lane, trucks changing left in a four-lane downhill lane, and trucks changing right in a four-lane downhill lane. The group with a frame difference of 0.2s and a threshold of 0.05m performs well in all situations, maintaining an acquisition error within 5% under different driving directions, vehicle types, lanes, and lane change directions. Therefore, experimental group 2, with a frame difference of 0.2s (5 frames) and a threshold of 0.05m, is selected as the calibration values for the lane change start and end point extraction model S and T. The lane change start and end point extraction update is shown in Formula 2.
[0090]
[0091] In the formula, t represents the number of frames for the start and end points of lane changes for each vehicle, F1 represents the first frame of the trajectory data for each vehicle, F2 represents the number of frames when the lane ID changes in the trajectory data for each vehicle, F3 represents the last frame of the trajectory data for each vehicle, and X t With X t-5 These represent the lateral coordinates of the vehicle at times t and t-5, respectively.
[0092] To avoid potential errors from the experimental vehicles during the experiment, the model needs to be validated. This invention applies the lane change start / end point extraction model to the lane change vehicle records of 60 HighD datasets, extracting the start / end points of all lane changes and calculating the corresponding lane change times. Relying on the visualization simulation provided by HighD, this invention extracts the actual lane change times of sample vehicles from the 60 datasets, compares them with the lane change times collected by the model, and calculates the average validation error for different road segments. The validation results are shown in Table 3 below.
[0093] Table 3 Model Validation Results
[0094]
[0095] As shown in Table 3, the average test error of the lane change start and end point extraction model is less than 4% in different road segments. This means that the lane change time collected by the model for each vehicle differs from the actual time by less than 4%. Taking a lane change time of 4 seconds for a vehicle as an example, the lane change time error is within 0.16 seconds, which is within the acceptable error range. This proves that the lane change start and end point extraction model is effective and can be used to collect lane change start and end points.
[0096] S5. Analyze the characteristics of lane change time distribution.
[0097] This invention extracts complete lane-change trajectory data for 6506 vehicles from 60 datasets, including 5789 cars and 717 trucks. 47% of the vehicles changed lanes to the left, and 53% changed lanes to the right. A lane-change start and end point extraction model based on frame difference and lateral displacement change is used to extract the start and end points for each vehicle, and the corresponding lane-change time is calculated. The lane-change time distribution is as follows: Figure 4 and Figure 5 As shown.
[0098] Figure 4 A scatter plot showing the distribution of lane-changing times for all lane-changing vehicles is presented. Different colored data points represent different datasets. The lane-changing time data points are relatively concentrated, but there are still a few outlier data points. Figure 5 It can be seen that the lane-changing time of vehicles on the highway ranges from 2.88s to 7.32s, and the lane-changing time and average speed show a normal distribution trend.
[0099] The above embodiments of the present invention establish a lane change start and end point extraction model based on the natural driving trajectory data of vehicles and the change in lateral displacement. The parameters of the model were calibrated and verified through experiments. This model can be used to collect lane change start and end points in trajectory datasets. At the same time, the present invention uses a massive vehicle trajectory dataset as the data source. The dataset can be obtained through UAV data collection and image recognition trajectory. More vehicle trajectory data can be incorporated in the future. Therefore, it has broad applicability and scalability in the future.
[0100] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
Claims
1. A vehicle trajectory data-driven method for identifying the start and end points of a lane change, characterized in that, Includes the following steps: S1. Process vehicle trajectory data; S2. Construct a lane change start and end point extraction model based on frame difference and lateral displacement change. S3. Use HighD visualization simulation to calibrate the parameters of the model; S4. Compare, analyze, and verify the experimental results of the model. S5. Analyze the characteristics of lane change time distribution; The vehicle trajectory data is lane-changing vehicle trajectory data in the HighD dataset. The HighD dataset provides four types of files for each record. The four types of files include an image of the recorded highway segment, a CSV file describing the recorded location, a CSV file recording an overview of the vehicle trajectory, and a CSV file containing detailed vehicle trajectory data. The HighD dataset is a publicly available dataset of vehicle trajectory data collected by drones in Cologne, Germany. The CSV file is a file that stores tabular data in plain text format; The processing of vehicle trajectory data specifically involves using the four types of files as a basis and employing Python 3.8 to preprocess the trajectory data for each record. The preprocessing specifically includes: S101. Summarize the recorded information in the HighD dataset and calculate the total traffic volume and vehicle type ratio information contained in the HighD dataset; S102. Using the vehicle lane change frequency information provided in the CSV file that records the vehicle trajectory overview, filter out vehicles with single lane change behavior, and summarize the corresponding category files according to different number of lanes, different speed limits, different vehicle types, different vehicle type ratios, and different driving directions. S103. Extract the CSV file containing detailed trajectory data of lane-changing vehicles. Match the corresponding information by using the vehicle ID information around each target vehicle. The information includes horizontal and vertical coordinates, horizontal and vertical velocity and acceleration, vehicle length and width. Merge the matched information with the data of each frame of the target vehicle. S104. Remove data with incomplete lane change trajectories and abnormal start and end points of trajectories; S105. Based on the characteristics of driving speed under different driving directions, determine the direction of vehicle lane change and classify the vehicle trajectory data for lane change to the left and lane change to the right. S106. Collect traffic parameter information for lane-changing vehicles, including lateral displacement change, minimum speed, maximum speed and average speed. The lane change start and end point extraction model is specifically as follows: in, t This represents the frame number of the start and end points of each lane change for each vehicle. This represents the first frame of the trajectory data for each vehicle. This represents the number of frames in the lane ID that change for each vehicle's trajectory data. This represents the last frame of each vehicle's trajectory data. and These respectively indicate that the vehicle is in t and tT The horizontal coordinate of time, T Represents the frame rate difference. S This represents the limit value of the lateral displacement change. S and T It needs to be calibrated through experiments.
2. The vehicle trajectory data-driven method for identifying the start and end points of lane changes according to claim 1, characterized in that, The lane change start / end point extraction model uses the frame difference T to calculate the lateral displacement change for each frame, and uses the lane ID information to determine the time of change. As the boundary, from Tracing back, extracting - The change in lateral displacement of the vehicle within the time period is less than or equal to the threshold value. S The maximum time is marked as the starting point for lane changing; from Tracing back, extracting - The change in lateral displacement of the vehicle within the time period is less than or equal to the threshold value. S The minimum time is marked as the end of the lane change; The accuracy of the lane change start and end point extraction model depends on the calibration of the frame difference T and the lateral displacement change S.
3. The vehicle trajectory data-driven method for identifying the start and end points of lane changes according to claim 2, characterized in that, The specific steps for using HighD visualization simulation to calibrate the model parameters are as follows: S301. Set up an experimental vehicle and design different experimental schemes based on special circumstances, including different driving directions, different number of lanes, different vehicle models, and different lane-changing directions. S302. The start and end points of lane changes for each experimental vehicle are recorded using HighD visualization simulation, which serves as a basis for judging the accuracy of each experimental scheme. S303. Calculate the lateral displacement change of the test vehicle under different experimental schemes, and draw the distribution diagram of the lateral displacement change of each test vehicle. S304. By analyzing the changes in the lateral displacement of the experimental vehicle, formulate rules for judging the start and end points of lane changes, and compare the differences in lane change times among different groups of experimental vehicles. S305. Compare the accuracy of different experimental schemes and calibrate the start and end points of lane changes to extract model parameters. S306. Verify different experimental schemes to check the accuracy of the lane change start and end point extraction model.
4. The vehicle trajectory data-driven method for identifying the start and end points of lane changes according to claim 3, characterized in that, The comparative analysis and verification of the model experiment results specifically involves comparing the accuracy of the different experimental schemes, whereby the accuracy is the ratio of the lane change time collected by the simulation to the lane change time collected by the model.
5. The vehicle trajectory data-driven method for identifying the start and end points of lane changes according to claim 4, characterized in that, The closer the ratio between the lane change time acquired by the simulation and the lane change time acquired by the model is to 1, the more accurate the start and end points of the lane change acquired by the model are.
6. The vehicle trajectory data-driven method for identifying the start and end points of lane changes according to claim 2, characterized in that, The frame difference T includes 0.2s, 0.4s, 0.6s, 0.8s, and 1.0s.
7. The vehicle trajectory data-driven method for identifying the start and end points of lane changes according to claim 1, characterized in that, The dataset includes data on 6,506 vehicles with complete lane-change trajectories for a single lane change.