An automated construction method for a multi-vehicle interaction lane-changing scenario library for an autonomous vehicle

Abstract

The application discloses a kind of multi-vehicle interaction lane change scene library automation construction methods for automatic driving vehicle, which proposes the semantic representation mode of vehicle driving scene and scene, and realizes the automation extraction of multi-vehicle interaction lane change scene based on the representation mode.Simultaneously, it also proposes the risk scene clustering method considering the position relationship of lane changing vehicle and surrounding vehicle, which can realize the risk level classification of multi-vehicle interaction lane change scene.Finally, the automation extraction and construction of multi-vehicle interaction lane change scene library with risk attribute label and reflecting risk evolution process are realized.The extraction of scene data in the application has the advantages of comprehensive scene elements, good method portability, and scene risk grading.

Description

Technical Field

[0001] This invention relates to the field of safety testing for autonomous vehicles, and in particular to an automated method for constructing a multi-vehicle interaction lane-changing scenario library for autonomous vehicles. Background Technology

[0002] The training and testing of autonomous vehicle (EV) algorithm models rely heavily on a large amount of driving scenario and situational data. Improving the quality of test data will help enhance the accuracy of EV algorithms. However, multi-vehicle lane-changing scenarios often involve complex behavioral interactions, making it challenging for current EV algorithms to make socially compatible lane-changing decisions in such scenarios. Currently, both aerial trajectory datasets and EV datasets can obtain information about the vehicle itself, surrounding vehicles, and the road environment. These datasets contain various driving scenarios and both valid and invalid data, thus they cannot be directly applied to specific scenario training and testing of models. Current research on scenario data extraction suffers from incomplete extraction of scenario elements, poor method portability, and a lack of scenario risk grading. Therefore, designing an automated method for constructing a multi-vehicle lane-changing scenario library for EVs is crucial for improving the safety of EV lane-changing behavior in complex environments. Summary of the Invention

[0003] This invention proposes an automated method for constructing a multi-vehicle lane-changing scenario library for autonomous vehicles. The method proposes a semantic representation of vehicle driving scenarios and contexts, and based on this representation, achieves automated extraction of multi-vehicle lane-changing scenarios. Simultaneously, this invention also proposes a risk scenario clustering method considering the positional relationship between the lane-changing vehicle and surrounding vehicles, enabling risk level classification of multi-vehicle lane-changing scenarios. Ultimately, this invention achieves automated extraction and construction of a multi-vehicle lane-changing scenario library with risk attribute labels and capable of reflecting the risk evolution process.

[0004] This invention proposes an automated construction method for a multi-vehicle lane-changing scenario library for autonomous vehicles, which includes a scenario and data structuring processing module, a scenario representation module, an automated extraction module for multi-vehicle lane-changing interaction scenarios, and a scenario risk classification module.

[0005] A method for automatically constructing a multi-vehicle interaction lane-changing scenario library for autonomous vehicles, characterized by the following steps:

[0006] a. The data structuring module includes a data resampling submodule, a scenario and situation data extraction submodule, and a vehicle type classification submodule. The inputs to this module are the data sampling frequency and the dataset. Based on actual needs, this module extracts data from datasets such as aerial trajectories or automated connected driving datasets at a specified data collection frequency, and inputs the processed data into the scenario representation module and the multi-vehicle lane-changing interaction scenario automated extraction module.

[0007] b. The scenario representation module digitally represents multi-vehicle lane-changing interaction scenarios through semantics, thereby achieving a clear representation of the number of scenario elements and the relationships between them. A scenario is defined as a seven-tuple containing the vehicle ID, vehicle type, lane-changing type, weather, travel time, and scene. The lane-changing type is further defined as a three-tuple containing three types of information: lane-changing area, lane-changing direction, and whether the lane change is continuous. A scene is defined as a four-tuple containing the vehicle ID, scene elements, element attributes, and element relationships.

[0008] c. The automated extraction module for multi-vehicle lane-changing interaction scenarios includes a sub-module for determining the type of lane-changing interaction scenario, a sub-module for obtaining the start and end points of the lane-changing interaction scenario, and a sub-module for extracting the relationship between scenario elements. This module further processes the output data of the data structuring module to extract the data of the vehicle lane-changing process from the complete trajectory data, as well as the values ​​corresponding to various scenario elements defined in the scenario representation module. The output data of this module will be used as the input data of the scenario representation module to update the data in the lane-changing type triplet, scenario quadruplet, and scenario seventuple.

[0009] d. The scenario risk classification module includes a scenario risk calculation submodule, a scenario risk clustering submodule, and a risk assessment submodule. The input data of this module is the semantic representation structure data of the multi-vehicle interactive lane-changing scenario output by the scenario representation module. The scenario risk calculation submodule calculates the risk of each scenario in the scenario and generates a risk matrix. The risk matrices of all scenarios are input into the scenario risk clustering submodule to classify the scenarios into risk categories. Finally, the classified results are input into the risk assessment submodule to obtain the risk level of each scenario. The output data of this module will be used as the input data of the scenario representation module to update the data in the scenario seven-tuple.

[0010] The data resampling submodule described in step a uses a time-interval-based linear interpolation method to resample the data to a specified time interval.

[0011] The information extracted by the scenario data extraction submodule in step a includes three major categories of data: traffic participant information, driving environment information, and road environment information.

[0012] The traffic participant information includes the vehicle's speed, acceleration, heading angle, lane ID, coordinates, and coordinates of the four vertices of the vehicle body.

[0013] The driving environment information includes weather information and travel time periods, wherein the weather information includes sunny days, rainy days, foggy days, snowy days, sand and dust, and strong wind; the travel time periods include: morning peak (6:00-9:00), upper off-peak (9:00-12:00), lower off-peak (12:00-16:00), evening peak (16:00-20:00), and other periods (20:00-24:00; 24:00-6:00).

[0014] The road environment information includes four major categories: dynamic traffic facilities, static traffic facilities, road channelization, and lane types, wherein the dynamic traffic facilities include traffic lights, tidal lane lights, etc.; the static traffic facilities include traffic signs and plates, road isolation facilities, etc.; the road channelization includes diverging and merging ramps, road edges; the lane types include dashed lines, solid lines, and curbstones, and this type of information extracts the names and coordinates of various elements existing in the current driving environment.

[0015] The input value of the vehicle type division sub-module described in step a is the coordinates of four vertices of the vehicle body, the vehicle body length is calculated according to the coordinates of the vertices, the judgment condition for a small-sized vehicle is that the vehicle body length L<5.83(m), the judgment condition for a medium-sized vehicle is that the vehicle body length 5.83(m)<L<12.58(m), and the judgment condition for a large-sized vehicle is that the vehicle body length L>12.58(m).

[0016] The definition formula of the scenario heptad described in step b is as follows:

[0017] SC t (Ego,Weather,Time,CarType,LCType,AScene t ,RiskLevel)t∈T

[0018] Wherein, T represents the duration of the original lane change scenario, t represents a certain moment in the scenario, Ego represents the lane-changing vehicle, Weather represents the weather, Time is the travel time period, CarType is the vehicle type, LCType is a parameter containing a specific set of lane change behaviors, AScene t represents the scene at a certain moment, and RiskLevel is the scenario risk level.

[0019] The definition formula of the lane change type triad described in step b is as follows:

[0020] Type={Area,Dir,Con}

[0021] Among them, Area indicates the area where the vehicle lane-changing behavior occurs, including the straight sections of the divergence area, merging area, and weaving area; Dir indicates the direction of the vehicle lane-changing behavior, including left lane changing and right lane changing; Con is used to indicate whether the vehicle is performing continuous lane-changing behavior.

[0022] The definition formula for the scenario quadruple mentioned in step b is as follows;

[0023] AScene t (AS) = {Ego, E t (AS),R t (AS),S t (AS)}

[0024] Where Ego represents the lane-changing vehicle; t represents a specific moment in the scenario; E t (AS) is the set of scene elements at time t, including traffic participants and road environment parameters; R t (AS) is a set of positional relationships between the vehicle and other scene elements, including front, rear, left, right, left front, left rear, right front, and right rear; S t (AS) is a collection of states of scene elements. When the element type is vehicle, it includes the vehicle's speed, position, vehicle type, heading angle, and driving lane ID. When the element type is road environment, it includes the element category, name, and range coordinates.

[0025] The lane change interaction scenario type judgment submodule described in step c determines whether the vehicle is continuously changing lanes based on the number of different lane IDs contained in the vehicle trajectory data. If the number of IDs is greater than 2, it is considered a continuous lane change. The direction of the lane change is determined based on the ID values ​​of the target lane and the original lane, as well as the vehicle's driving direction. For example, if the vehicle is traveling along the x-axis and the driving lane ID is greater than the target lane ID, the vehicle is changing lanes to the left; otherwise, it is changing lanes to the right. The module matches the coordinates of the lane change point with the position coordinates of road environment elements to obtain the area where the lane change scenario occurs, as well as the name and coordinates of the road environment elements contained in the scenario.

[0026] The submodule for obtaining the start and end points of the lane change interaction scenario in step c employs a search method based on the timestamps of the lane change points. Considering that continuous lane changes may involve multiple change points, the search starts from the first change point and searches backward from the last change point. During the search, if the vehicle's lateral acceleration is the same as the previous moment, and the difference between its lateral position and the change point position is greater than 0.9 meters, the search is considered complete. The current timestamp is recorded as the start and end marker of the scenario, and this marker is used to segment and extract the lane change interaction scenario from the original dataset.

[0027] The scenario element relationship extraction submodule described in step c allows setting the extraction range for surrounding elements, typically a 100m radius area centered on the lane-changing vehicle. Within this range, it acquires the vehicles in front and behind in the driving lane, as well as the closest vehicles in the adjacent lanes to the vehicle. Then, it obtains the positional relationship and operational status information between the vehicle and the vehicle from the data structuring module. If the lane of the lane-changing vehicle or the target lane is the outermost lane of the road, it is necessary to obtain the positional relationship between the lane and the vehicle, as well as the lane coordinates.

[0028] The scenario risk calculation submodule described in step d calculates the risk values ​​of lane-changing vehicles and surrounding vehicles respectively, assigns position index values ​​to vehicles at specific locations based on the position information provided by surrounding vehicles, and forms a risk matrix as the model output. The output of this module is input into the scenario risk clustering submodule.

[0029] The risk between vehicles can be calculated using the following formula:

[0030]

[0031] Where t represents a specific moment in the scenario. and These represent vehicles v. i and v j The set of four boundary points and The two points that are closest to each other. and These represent vehicles v. i and v j The velocity vector.

[0032] The location indices for different vehicles in different scenarios are shown in the table below:

[0033]

[0034] The following is an example of a risk matrix for a scenario:

[0035]

[0036] The scenario risk clustering submodule described in step d first uses a Gaussian mixture model (GMM) to transform the input scenario risk matrix two-dimensional time series data into discrete clusters, and then uses the bag-of-words model with hidden Dirichlet distribution (LDA) to achieve clustering of the time series data. This clustering process can preserve the number of elements, risks, and spatial location relationships in the scenario.

[0037] The risk assessment submodule described in step d constructs a scenario-based integrated risk sequence based on scenario-specific risk indicators. Where 1≤t≤N, and based on this risk sequence, four indicators for assessing scenario risk are proposed. The calculation methods and meanings of the indicators are shown in the table below.

[0038]

[0039] r t Let be the minimum risk value of surrounding vehicles and lane-changing vehicles in the scenario at time t. The calculation method is as follows:

[0040]

[0041] R Mean and R Max These represent the average and maximum values ​​of the lane-change scenario risk sequence, respectively, reflecting the distribution of risk scenarios.

[0042] SLE C and SLE Max The calculation is based on time-varying derivative SLE C This is a risk probability estimate, representing the likelihood of a risk occurring. When... A sustained value greater than 0 over a given period in the sequence indicates a continuous accumulation of risk; the longer the accumulation period, the greater the likelihood of the risk occurring. Δs represents the longest cumulative risk time series. Risk Intensity Estimate (SLE) Max The value is used to measure the intensity of risk occurrence. The larger the value, the more drastic the risk change between time t and t+1. Attached Figure Description

[0043] Figure 1 This is a structural diagram of a method for automatically constructing a library of multi-vehicle interaction lane-changing scenarios for autonomous vehicles.

[0044] Figure 2 This is a diagram illustrating lane-changing types.

[0045] Figure 3 This is a screenshot of a lane-changing scenario. Detailed Implementation

[0046] The specific embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this disclosure.

[0047] As an embodiment of the present invention, such as Figure 1 The method for automatically constructing a multi-vehicle lane-changing scenario library for autonomous vehicles includes: a data structuring processing module 101, a scenario representation module 105, an automated extraction module for multi-vehicle lane-changing interaction scenarios 109, and a scenario risk classification module 113.

[0048] The data structuring processing module includes a data resampling submodule 102, a scene and scenario data extraction submodule 103, and a vehicle type classification submodule 104. The inputs to this module are the data sampling frequency and the dataset. Based on actual needs, this module extracts data from datasets such as aerial trajectories or automated connected driving datasets at a specified data collection frequency, and inputs the processed data into the scenario representation module and the automated extraction module for multi-vehicle lane-changing interaction scenarios.

[0049] The scenario representation module digitally represents multi-vehicle lane-changing interaction scenarios through semantics, thereby achieving a clear representation of the number of scenario elements and the relationships between them. A scenario is defined as a seven-tuple (108) containing the vehicle ID, vehicle type, lane-changing type, weather, travel time, and scene. The lane-changing type is further defined as a three-tuple (106) containing three types of information: lane-changing area, lane-changing direction, and whether the lane change is continuous. A scene is defined as a four-tuple (107) containing the vehicle ID, scene elements, element attributes, and element relationships.

[0050] The multi-vehicle lane-changing interaction scenario automatic extraction module 109 includes a lane-changing interaction scenario type judgment submodule 110, a lane-changing interaction scenario start and end point acquisition submodule 111, and a scenario element relationship extraction submodule 112. This module further processes the output data of the data structuring processing module to extract the vehicle lane-changing process data from the complete trajectory data, as well as the values ​​corresponding to various scenario elements defined in the scenario representation module. The output data of this module will be used as the input data of the scenario representation module to update the data in the lane-changing type triplet, scenario quadruplet, and scenario seventuple.

[0051] d. The scenario risk classification module 113 includes a scenario risk calculation submodule 114, a scenario risk clustering submodule 115, and a risk assessment submodule 116. The input data of this module is the semantic representation structure data of the multi-vehicle interactive lane-changing scenario output by the scenario representation module. The scenario risk calculation submodule calculates the risk of each scenario in the scenario and generates a risk matrix. The risk matrices of all scenarios are input into the scenario risk clustering submodule to classify the scenarios into risk categories. Finally, the classified results are input into the risk assessment submodule to obtain the risk level of each scenario. The output data of this module will be used as the input data of the scenario representation module to update the data in the scenario seven-tuple.

[0052] The data resampling submodule described in step a uses a time-interval-based linear interpolation method to resample the data to a specified time interval.

[0053] The information extracted by the scenario data extraction submodule in step a includes three major categories of data: traffic participant information, driving environment information, and road environment information.

[0054] Wherein the traffic participant information includes the speed, acceleration, heading angle, driving lane ID, coordinates, and coordinates of four vertices of the vehicle body.

[0055] The driving environment information includes weather information and travel time periods, wherein the weather information includes sunny days, rainy days, foggy days, snowy days, sand and dust, and strong wind; the travel time periods include: morning peak (6:00-9:00), upper off-peak (9:00-12:00), lower off-peak (12:00-16:00), evening peak (16:00-20:00), and others (20:00-24:00; 24:00-6:00).

[0056] The road environment information includes four major categories: dynamic traffic facilities, static traffic facilities, road channelization, and lane types, wherein the dynamic traffic facilities include traffic lights, tidal lane lights, etc.; the static traffic facilities include traffic signs and plates, road isolation facilities, etc.; the road channelization includes merging and diverging ramps, road edges; the lane types include dashed lines, solid lines, and curbs, and this type of information extracts the names and coordinates of various elements existing in the current driving environment.

[0057] The input value of the vehicle type classification sub-module described in step a is the coordinates of the four vertices of the vehicle body, the vehicle body length is calculated according to the coordinates of the vertices, the judgment condition for a small vehicle is that the vehicle body length L<5.83(m), the judgment condition for a medium vehicle is that the vehicle body length 5.83(m)<L<12.58(m), and the judgment condition for a large vehicle is that the vehicle body length L>12.58(m).

[0058] The definition formula of the scenario heptad described in step b is as follows:

[0059] SC t (Egp,Weather,Time,CarType,LCType,AScene t ,RiskLevel)t∈T

[0060] Wherein, T represents the duration of the original lane change scenario, t represents a certain moment in the scenario, Ego represents the lane change vehicle, Weather represents the weather, Time is the travel time period, CarType is the vehicle type, LCType is a parameter including a set of specific lane change behaviors, AScene t represents the scenario at a certain moment, and RiskLevel is the scenario risk level.

[0061] The definition formula of the lane change type triplet described in step b is as follows:

[0062] Type={Area,Dir,Con}

[0063] Among them, Area indicates the area where the vehicle lane-changing behavior occurs, including the straight sections of the divergence area, merging area, and weaving area; Dir indicates the direction of the vehicle lane-changing behavior, including left lane changing and right lane changing; Con is used to indicate whether the vehicle is performing continuous lane-changing behavior.

[0064] The definition formula for the scenario quadruple mentioned in step b is as follows;

[0065] AScene t (AS) = {Ego, E t (AS),R t (AS),S t (AS)}

[0066] Where Ego represents the lane-changing vehicle; t represents a specific moment in the scenario; E t (AS) is the set of scene elements at time t, including traffic participants and road environment parameters; R t (AS) is a set of positional relationships between the vehicle and other scene elements, including front, rear, left, right, left front, left rear, right front, and right rear; S t (AS) is a collection of states of scene elements. When the element type is vehicle, it includes the vehicle's speed, position, vehicle type, heading angle, and driving lane ID. When the element type is road environment, it includes the element category, name, and range coordinates.

[0067] The lane change interaction scenario type judgment submodule in step c determines whether the vehicle is making continuous lane changes based on the number of different lane IDs contained in the vehicle trajectory data. If the number of IDs is greater than 2, it is considered a continuous lane change (e.g., ...). Figure 2 (As shown in Figures 1, 4, and 5); the direction of lane changing is determined based on the ID values ​​of the target lane and the original lane, as well as the vehicle's travel direction. For example: if the vehicle is traveling along the x-axis (e.g., ... Figure 2 As shown in Figure 6), and the driving lane ID is greater than the target lane ID, then the vehicle is changing lanes to the left (e.g., ...). Figure 2 (As shown in Figure 2), otherwise it is a right lane change (as shown in Figure 2). Figure 2 As shown in Figure 3); based on the lane change point (such as... Figure 2 The coordinates of the road environment elements (as shown in Figure 8) are matched to obtain the area where the lane change scene occurs, as well as the names and coordinates of the road environment elements contained in the scene.

[0068] The submodule for obtaining the start and end points of the lane change interaction scenario in step c employs a method based on searching before and after the timestamp of the lane change point. Considering that continuous lane changes may involve multiple change points, the search starts from the first change point and searches backwards from the last change point. During the search, if the vehicle's lateral acceleration is the same as the previous moment, and the difference between its lateral position and the change point position is greater than 0.9 meters (e.g., ...), the search will continue. Figure 3 As shown in Figure 10), this indicates that the search is complete, and the current timestamp is recorded as the start and end marker of the scenario (e.g., ...). Figure 3 As shown in Figures 10 and 14), and this identifier is used to segment and extract lane-changing interaction scenarios from the original dataset (e.g., Figure 3 (As shown in Figure 13).

[0069] The scenario element relationship extraction submodule described in step c allows setting the extraction range for surrounding elements, typically a 100m radius area centered on the lane-changing vehicle. Within this range, it acquires the vehicles in front and behind in the driving lane, as well as the closest vehicles in the adjacent lanes to the vehicle. Then, it obtains the positional relationship and operational status information between the vehicle and the vehicle from the data structuring module. If the lane of the lane-changing vehicle or the target lane is the outermost lane of the road (e.g., ...), ... Figure 2 As shown in Figure 9, it is necessary to obtain the positional relationship between the lane and the vehicle, as well as the coordinates of the lane.

[0070] The scenario risk calculation submodule described in step d calculates the risk values ​​of lane-changing vehicles and surrounding vehicles respectively, assigns position index values ​​to vehicles at specific locations based on the position information provided by surrounding vehicles, and forms a risk matrix as the model output. The output of this module is input into the scenario risk clustering submodule.

[0071] The risk between vehicles can be calculated using the following formula:

[0072]

[0073] Where t represents a specific moment in the scenario. and These represent vehicles v. i and v j The set of four boundary points and The two points that are closest to each other. and These represent vehicles v. i and v j The velocity vector.

[0074] The location indices for different vehicles in different scenarios are shown in the table below:

[0075]

[0076] The following is an example of a risk matrix for a scenario:

[0077]

[0078] The scenario risk clustering submodule described in step d first uses a Gaussian mixture model (GMM) to transform the input scenario risk matrix two-dimensional time series data into discrete clusters, and then uses the bag-of-words model with hidden Dirichlet distribution (LDA) to achieve clustering of the time series data. This clustering process can preserve the number of elements, risks, and spatial location relationships in the scenario.

[0079] The risk assessment submodule described in step d constructs a scenario-based integrated risk sequence based on scenario-specific risk indicators. Where 1≤t≤N, and based on this risk sequence, four indicators for assessing scenario risk are proposed. The calculation methods and meanings of the indicators are shown in the table below.

[0080]

[0081] r t Let be the minimum risk value of surrounding vehicles and lane-changing vehicles in the scenario at time t. The calculation method is as follows:

[0082]

[0083] R Mean and R Max These represent the average and maximum values ​​of the lane-change scenario risk sequence, respectively, reflecting the distribution of risk scenarios.

[0084] SLE C and SLE Max The calculation is based on time-varying derivative SLE C This is a risk probability estimate, representing the likelihood of a risk occurring. When... A sustained value greater than 0 over a given period in the sequence indicates a continuous accumulation of risk; the longer the accumulation period, the greater the likelihood of the risk occurring. Δs represents the longest cumulative risk time series. Risk Intensity Estimate (SLE) Max The value is used to measure the intensity of risk occurrence. The larger the value, the more drastic the risk change between time t and t+1.

Claims

1. An automated construction method for a multi-vehicle lane-changing scenario library for autonomous vehicles, wherein the system implementing the method includes a scenario and data structuring processing module, a scenario representation module, an automated extraction module for multi-vehicle lane-changing interaction scenarios, and a scenario risk classification module; Its features are, Includes the following steps: a. The data structuring processing module includes a data resampling submodule, a scenario and situation data extraction submodule, and a vehicle type classification submodule; the input of this module is the data sampling frequency and the dataset; this module extracts data from aerial photography trajectories or automated connected driving datasets according to actual needs and specified data collection frequencies, and inputs the processed data into the scenario representation module and the multi-vehicle lane-changing interaction scenario automated extraction module; b. The scenario representation module digitally represents multi-vehicle lane-changing interaction scenarios in a semantic way, thereby achieving a clear representation of the number of scenario elements and the relationships between elements; the scenario is defined as a seven-tuple containing the vehicle number, vehicle type, lane-changing type, weather, travel time, and scene; the lane-changing type is defined as a three-tuple containing three types of information: lane-changing area, lane-changing direction, and whether the lane change is continuous; and the scene is defined as a four-tuple containing the vehicle number, scene elements, element attributes, and element relationships. c. The multi-vehicle lane-changing interaction scenario automatic extraction module includes a lane-changing interaction scenario type judgment submodule, a lane-changing interaction scenario start and end point acquisition submodule, and a scenario element relationship extraction submodule. This module further processes the output data of the data structuring processing module to extract the vehicle lane-changing process data from the complete trajectory data, as well as the corresponding values ​​of various scenario elements defined in the scenario representation module. The output data of this module will be used as the input data of the scenario representation module to update the data in the lane-changing type triplet, scenario quadruplet, and scenario seventuple. d. The scenario risk classification module includes a scenario risk calculation submodule, a scenario risk clustering submodule, and a risk assessment submodule. The input data of this module is the semantic representation structure data of the multi-vehicle interactive lane-changing scenario output by the scenario representation module. The scenario risk calculation submodule calculates the risk of each scenario in the scenario and generates a risk matrix. The risk matrices of all scenarios are input into the scenario risk clustering submodule to classify the scenarios into risk categories. Finally, the classified results are input into the risk assessment submodule to obtain the risk level of each scenario. The output data of this module will be used as the input data of the scenario representation module to update the data in the scenario seven-tuple. The data resampling submodule described in step a uses a time-interval-based linear interpolation method to resample the data to a specified time interval.

2. The method for automatically constructing a multi-vehicle interaction lane-changing scenario library for autonomous vehicles according to claim 1, characterized in that, The scenario data extraction submodule described in step a extracts three main categories of data: traffic participant information, driving environment information, and road environment information. Among them, the traffic participant information includes the vehicle's speed, acceleration, heading angle, lane ID, coordinates, and coordinates of the four vertices of the vehicle body; the driving environment information includes weather information and travel time. Road environment information includes four main categories: dynamic traffic facilities, static traffic facilities, road channelization, and lane types. Dynamic traffic facilities include traffic lights and tidal lane lights; static traffic facilities include traffic signs and road barriers; road channelization includes merging and diverging ramps and road edges; and lane types include dashed lines, solid lines, and curbs. This type of information extracts the names and coordinates of various elements present in the current driving environment.

3. The method for automatically constructing a multi-vehicle interaction lane-changing scenario library for autonomous vehicles according to claim 1, characterized in that, The input value of the vehicle type classification submodule in step a is the coordinates of the four vertices of the vehicle body, and the length of the vehicle body is calculated based on the left side of the vertices.

4. The method for automatically constructing a multi-vehicle interaction lane-changing scenario library for autonomous vehicles according to claim 1, characterized in that, The definition formula for the scenario seven-tuple mentioned in step b is as follows: Where T represents the duration of the original lane-changing scenario, and t represents a specific moment in the scenario. Represents vehicles changing lanes. Represents the weather. For travel time, It refers to the vehicle type, LC. It is a parameter that contains a set of specific lane-changing behaviors. Describes a scene at a specific moment. Scenario risk level.

5. The method for automatically constructing a multi-vehicle interaction lane-changing scenario library for autonomous vehicles according to claim 1, characterized in that, The formula for defining the lane change type triplet mentioned in step b is as follows: Among them, Area indicates the area where the vehicle lane-changing behavior occurs, including the straight sections of the divergence area, merging area, and weaving area; Dir indicates the direction of the vehicle lane-changing behavior, including left lane changing and right lane changing; Con is used to indicate whether the vehicle is performing continuous lane-changing behavior.

6. The method for automatically constructing a multi-vehicle interaction lane-changing scenario library for autonomous vehicles according to claim 1, characterized in that, The definition formula for the scenario quadruple mentioned in step b is as follows; in, For vehicles changing lanes; t represents a specific moment in the scenario; It is a collection of scene elements at time t, including traffic participants and road environment parameters; This is a set of positional relationships between the vehicle and other scene elements, including front, back, left, right, left front, left rear, right front, and right rear; It is a collection of the states of scene elements. When the element type is vehicle, it includes the vehicle's speed, position, vehicle type, heading angle, and driving lane ID; when the element type is road environment, it includes the element category, name, and range coordinates.

7. The method for automatically constructing a multi-vehicle interaction lane-changing scenario library for autonomous vehicles according to claim 1, characterized in that, The lane change interaction scenario type judgment submodule in step c determines whether the vehicle is continuously changing lanes based on the number of different lane IDs contained in the vehicle trajectory data. If the number of IDs is greater than 2, it is considered continuous lane change. The direction of lane change is determined based on the ID size of the target lane and the original lane, as well as the vehicle's driving direction. Matching the coordinates of the lane change point with the location coordinates of road environment elements yields the area where the lane change scenario occurs, as well as the names and coordinates of the road environment elements contained within the scenario.

8. The method for automatically constructing a multi-vehicle interaction lane-changing scenario library for autonomous vehicles according to claim 1, characterized in that, The submodule for obtaining the start and end points of the lane change interaction scenario in step c adopts a method based on the search before and after the timestamp of the lane change point. Considering that there may be multiple change points for continuous lane changes, the starting point is searched forward from the first change point, and the ending point is searched backward from the last change point. During the search process, if the lateral acceleration of the vehicle is the same as the previous moment, and the difference between the lateral position and the change point position is greater than 0.9 meters, it indicates that the search is complete, and the timestamp of the current moment is recorded as the start and end identifier of the scenario. This identifier is then used to segment and extract the lane change interaction scenario from the original dataset.

9. The method for automatically constructing a multi-vehicle interaction lane-changing scenario library for autonomous vehicles according to claim 1, characterized in that, The scenario element relationship extraction submodule described in step c sets the extraction range of surrounding elements to an area with a radius of 100m centered on the lane-changing vehicle; within this range, the vehicles in front and behind in the driving lane, as well as the vehicles in front and behind that are closest to the vehicle in the adjacent lane, are obtained. Then, the location relationship and operating status information between the vehicle and the self-driving vehicle are obtained from the data structuring module; If the lane in which the vehicle is traveling or the target lane is the outermost lane of the road, then the positional relationship between the lane and the vehicle, as well as the coordinates of the lane, are obtained.

Images