A vehicle lane-changing risk prediction method, device, equipment and storage medium

Abstract

The application discloses a vehicle lane-changing risk prediction method and device, equipment and a storage medium, and relates to the technical field of vehicle intelligent driving. The method comprises the following steps: determining a target driving image sequence from a natural driving data set; inputting a preset target prompt word and the target driving image sequence into a preset video large language model to determine a lane-changing risk cause parameter; determining a feasibility determination signal for representing the possibility of lane-changing of a lane-changing vehicle based on a forward and backward time interval and the speed of each vehicle, planning a displacement trajectory for the lane-changing vehicle, obtaining an initial vehicle lane-changing trajectory, verifying the initial vehicle lane-changing trajectory based on a preset vehicle kinematics constraint to obtain a vehicle feasible lane-changing trajectory; deducing the longitudinal interaction process between the lane-changing vehicle and the front and rear vehicles based on the lane-changing start and end time and the vehicle feasible lane-changing trajectory, obtaining a lane-changing risk scene sample, and determining a vehicle lane-changing risk prediction result. In this way, rich and reliable data support can be provided for intelligent driving system testing and risk assessment.

Description

Technical Field

[0001] This invention relates to the field of intelligent driving technology for vehicles, and in particular to a method, device, equipment, and storage medium for predicting lane change risks for vehicles. Background Technology

[0002] With the rapid development of intelligent driving technology, building a test scenario library covering diverse risk levels has become an important foundation for the verification and evaluation of intelligent driving systems. Natural driving data acquisition systems have been widely used in driving behavior analysis, and their videos and vehicle motion parameters provide real data sources for entry scenario research. Currently, the generation of neighboring vehicle entry scenarios mainly relies on manual annotation and rule-driven methods based on natural driving data. This involves manually screening entry events frame by frame and recording key parameters, using a fixed-threshold gap acceptance model to determine lane change feasibility, and generating lane change trajectories using straight-line or circular interpolation. However, this method has several shortcomings: First, manual annotation is time-consuming, labor-intensive, and subject to subjective differences; second, a constant THW threshold cannot describe the forced lane change behavior of drivers at intersections who are forced to accept smaller gaps due to the shortening distance to the stop line, which does not conform to the real conflict mechanism; third, simple interpolation cannot guarantee the continuous smoothness of lateral velocity and acceleration during lane change, does not consider vehicle kinematic constraints, and the authenticity and executability of the generated samples are insufficient; fourth, it relies only on limited data fragments, making it difficult to cover high-risk, sparsely occurring forced entry scenarios, thus limiting sample diversity.

[0003] Therefore, it is necessary to solve the problem of how to automatically identify risky behaviors in traffic scenarios and issue early warnings in order to provide rich and reliable data support for the testing and risk assessment of intelligent driving systems. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a method, apparatus, device, and storage medium for predicting vehicle lane-changing risks, which can provide rich and reliable data support for intelligent driving system testing and risk assessment. The specific solution is as follows: Firstly, this application discloses a method for predicting vehicle lane-changing risks, including: Based on vehicle operating parameters, driving video frames containing vehicle lane-changing behavior are extracted from the natural driving dataset to obtain the target driving image sequence. The preset target prompt words and the target driving image sequence are input into the preset video large language model, and the lane change risk causal parameters are determined based on the structured risk causal analysis results output by the preset video large language model. The lane change risk causal parameters include the start and end times of the lane change, the duration of the lane change, the time distance between the front and rear directions, and the speed of each vehicle. The speed of each vehicle includes the speed of the lane-changing vehicle, the speed of the vehicle in front of the target lane, and the speed of the vehicle behind the target lane. Based on the forward and backward time distances and the speeds of each vehicle, a feasibility judgment signal is determined to characterize allowing the lane-changing vehicle to change lanes. Based on the feasibility determination signal and the lane change duration, the displacement trajectory of the lane-changing vehicle is planned to obtain the initial vehicle lane-changing trajectory, and the initial vehicle lane-changing trajectory is verified based on the preset vehicle kinematic constraints to obtain the vehicle feasible lane-changing trajectory. Based on the start and end times of the lane change and the feasible lane change trajectory of the vehicle, the longitudinal interaction process between the lane-changing vehicle and the vehicle in front of and behind the target lane is deduced to obtain lane change risk scenario samples, and the vehicle lane change risk prediction result is determined based on the lane change risk scenario samples.

[0005] Optionally, the step of extracting driving segment video frames containing vehicle lane-changing behavior from the natural driving dataset based on vehicle operating parameters to obtain the target driving image sequence includes: Natural driving data is collected within a preset distance range before the stop line of the intersection approach lane to obtain a natural driving dataset; the natural driving data includes driving videos and vehicle operating parameters synchronized with the driving videos in time; Based on the vehicle operating parameters, driving segments containing vehicle lane-changing behavior are filtered from the natural driving dataset. Based on the time axis, video frames are uniformly extracted from the driving video corresponding to the driving segment at preset time intervals, and the extracted video frames are arranged according to the extraction time sequence to obtain a target driving image sequence that retains the temporal context.

[0006] Optionally, the step of inputting the preset target prompt words and the target driving image sequence into a preset video large language model, and determining the lane change risk causation parameters based on the structured risk causation analysis results output by the preset video large language model, includes: Construct preset target prompts that include role definition fields, task planning fields, and output constraint fields; The preset target prompt words and the target driving image sequence are input into the preset video big language model to obtain the structured risk causation analysis results output by the preset video big language model; Based on the structured risk causation analysis results, key risk events are located, and based on the distribution of lane-changing behavior parameters corresponding to the key risk events in the natural driving dataset, the lane-changing risk causation parameters corresponding to the key risk events are determined.

[0007] Optionally, the step of determining the feasibility judgment signal for allowing the lane-changing vehicle to change lanes based on the forward and backward time distances and the speeds of each vehicle includes: The forward and backward time distances are determined as follows: the forward time distance is the ratio of the forward clearance between the lane-changing vehicle and the vehicle in front of the target lane to the speed of the lane-changing vehicle among the vehicle speeds; the forward clearance is the value obtained by subtracting the vehicle length of the vehicle in front of the target lane from the difference between the longitudinal position of the vehicle in front of the target lane and the longitudinal position of the lane-changing vehicle; the backward time distance is the ratio of the backward clearance between the lane-changing vehicle and the vehicle behind the vehicle in the target lane to the speed of the vehicle behind the vehicle among the vehicle speeds; the backward clearance is the value obtained by subtracting the vehicle length of the lane-changing vehicle from the difference between the longitudinal position of the lane-changing vehicle and the longitudinal position of the vehicle behind the vehicle in the target lane. Determine the target distance between the lane-changing vehicle and the intersection stop line, and construct the forward time distance threshold and the backward time distance threshold as a three-segment linear function of the target distance based on the preset far-end boundary distance and the preset near-end critical distance; If the forward time distance is not less than the forward time distance threshold and the backward time distance is not less than the backward time distance threshold, a feasibility determination signal representing that lane changing is allowed is generated.

[0008] Optionally, the step of planning the displacement trajectory of the lane-changing vehicle based on the feasibility determination signal and the lane-changing duration to obtain the initial vehicle lane-changing trajectory includes: Construct a vehicle kinematic model of the lane-changing vehicle; the vehicle kinematic model uses the center of mass position, heading angle, longitudinal velocity and equivalent front wheel steering angle of the lane-changing vehicle as state variables, and the rate of change of the heading angle is the ratio of the product of the longitudinal velocity and the tangent of the equivalent front wheel steering angle to the wheelbase of the lane-changing vehicle. Based on the feasibility determination signal, the lane-changing duration is taken as the initial lane-changing duration. Using the initial lane-changing duration as the time domain, a fifth-order polynomial is used to construct the lateral displacement trajectory of the lane-changing vehicle from the center line of the current lane to the center line of the target lane according to the lateral boundary conditions. The lateral boundary conditions are that the lateral position at the start of the lane change is the lateral position of the lane-changing starting point, and the lateral position at the end of the lane change is the lateral position of the center of the target lane. The lateral velocity and lateral acceleration at the start and end of the lane change are both zero. The longitudinal acceleration response of the lane-changing vehicle to the vehicle in front in the target lane is determined based on a preset car-following model, and the longitudinal acceleration response is integrated sequentially to obtain the longitudinal displacement trajectory of the lane-changing vehicle. The lateral displacement trajectory and the longitudinal displacement trajectory are fused based on the vehicle kinematic model to obtain the initial vehicle lane-changing trajectory.

[0009] Optionally, the step of verifying the initial vehicle lane-changing trajectory based on preset vehicle kinematic constraints to obtain a feasible vehicle lane-changing trajectory includes: Preset vehicle kinematic constraints are constructed based on the vehicle kinematic model; the preset vehicle kinematic constraints include steering angle constraints, steering angular velocity constraints, and lateral acceleration constraints; Based on the initial vehicle lane-changing trajectory, the equivalent front wheel angle, equivalent front wheel angle change rate, and lateral acceleration at each moment are determined, and the equivalent front wheel angle, equivalent front wheel angle change rate, and lateral acceleration are verified based on the preset vehicle kinematic constraints. If the verification is successful, the initial vehicle lane-changing trajectory will be taken as the feasible lane-changing trajectory of the vehicle. If the verification fails, the lateral boundary conditions remain unchanged, the initial lane-changing duration is increased by a preset time step, and the initial vehicle lane-changing trajectory is regenerated until the verification passes, so that the initial vehicle lane-changing trajectory when the verification passes is taken as the vehicle's feasible lane-changing trajectory.

[0010] Optionally, the step of deriving the longitudinal interaction process between the lane-changing vehicle and the vehicle in front and behind in the target lane based on the start and end times of the lane change and the feasible lane-changing trajectory of the vehicle to obtain lane-changing risk scenario samples includes: The range of values ​​for the risk propensity stratification parameters is divided to obtain each risk propensity layer, and random sampling is performed within each risk propensity layer to obtain the corresponding lane change scenario parameter combination; the risk propensity stratification parameters include the lane change duration, and the target forward time distance and target backward time distance corresponding to the lane change end time in the forward and backward time distances; Based on the combination of lane-changing scenario parameters and the speed of each vehicle, the longitudinal gap between the lane-changing vehicle and the vehicle in front of the target lane and the vehicle behind the target lane at the end of the lane change is determined, and the initial longitudinal position of each vehicle at the initial time is obtained by reverse calculation based on the longitudinal gap. Based on the initial longitudinal position and the vehicle's feasible lane-changing trajectory, the longitudinal interaction process is iteratively simulated within a preset simulation time using a preset distance step length, so as to obtain lane-changing risk scenario samples carrying corresponding risk tendency layer identifiers.

[0011] Secondly, this application discloses a vehicle lane-changing risk prediction device, comprising: The image sequence acquisition module is used to extract driving segment video frames containing vehicle lane-changing behavior from the natural driving dataset based on vehicle operating parameters to obtain the target driving image sequence. The parameter determination module is used to input the preset target prompt words and the target driving image sequence into the preset video large language model, and determine the lane change risk causation parameters based on the structured risk causation analysis results output by the preset video large language model; the lane change risk causation parameters include the start and end times of the lane change, the duration of the lane change, the time distance between the front and rear directions, and the speed of each vehicle; the speed of each vehicle includes the speed of the lane-changing vehicle, the speed of the vehicle in front of the target lane, and the speed of the vehicle behind the target lane; The signal generation module is used to determine a feasibility judgment signal that characterizes allowing the lane-changing vehicle to change lanes, based on the forward and backward time distances and the speeds of each vehicle. The trajectory planning module is used to plan the displacement trajectory of the lane-changing vehicle based on the feasibility judgment signal and the lane-changing duration to obtain the initial vehicle lane-changing trajectory, and to verify the initial vehicle lane-changing trajectory based on the preset vehicle kinematic constraints to obtain the feasible vehicle lane-changing trajectory. The risk prediction module is used to deduce the longitudinal interaction process between the lane-changing vehicle and the vehicle in front of and behind the target lane based on the start and end times of the lane change and the feasible lane-changing trajectory of the vehicle, so as to obtain lane-changing risk scenario samples, and determine the vehicle lane-changing risk prediction result based on the lane-changing risk scenario samples.

[0012] Thirdly, this application discloses an electronic device, comprising: Memory, used to store computer programs; A processor is used to execute the computer program to implement the aforementioned vehicle lane change risk prediction method.

[0013] Fourthly, this application discloses a computer-readable storage medium for storing a computer program, which, when executed by a processor, implements the aforementioned vehicle lane-changing risk prediction method.

[0014] As can be seen, in this embodiment, driving segment video frames containing lane-changing behavior are extracted from the natural driving dataset based on vehicle operating parameters to obtain a target driving image sequence; preset target prompt words and the target driving image sequence are input into a preset video large language model, and lane-changing risk causation parameters are determined based on the structured risk causation analysis results output by the preset video large language model; the lane-changing risk causation parameters include lane-changing start and end times, lane-changing duration, forward and backward time distances, and vehicle speeds; the vehicle speeds include the speeds of the lane-changing vehicle, the vehicle in front of the target lane, and the vehicle behind the target lane; based on the forward and backward time distances and the vehicle speeds, the lane-changing risk causation parameters are determined. Speed ​​determination is used to characterize the feasibility judgment signal for allowing the lane-changing vehicle to change lanes; based on the feasibility judgment signal and the lane-changing duration, the displacement trajectory of the lane-changing vehicle is planned to obtain the initial vehicle lane-changing trajectory, and the initial vehicle lane-changing trajectory is verified based on preset vehicle kinematic constraints to obtain the vehicle's feasible lane-changing trajectory; based on the lane-changing start and end times and the vehicle's feasible lane-changing trajectory, the longitudinal interaction process between the lane-changing vehicle and the vehicle in front of and behind in the target lane is deduced to obtain lane-changing risk scenario samples, and the vehicle lane-changing risk prediction result is determined based on the lane-changing risk scenario samples. Specifically, by automatically extracting lane-change risk causal parameters through a large video language model, the subjective differences and high costs of manual frame-by-frame annotation are avoided; by designing the gap acceptance threshold as a dynamic function of the distance from the stop line, the forced lane-change decision mechanism of drivers in intersection approach lanes compressing acceptable gaps due to limited remaining distance is accurately characterized; by verifying the lateral trajectory and kinematic constraints using a five-order polynomial, the smoothness and physical feasibility of the lane-change trajectory are ensured; and by employing a risk-prone hierarchical sampling strategy, entry scenarios covering high, medium, and low risk levels are generated in batches, providing rich and reliable data support for intelligent driving system testing and risk assessment. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0016] Figure 1 This is a flowchart of a vehicle lane-changing risk prediction method disclosed in this application; Figure 2 This application discloses a specific flowchart of risk causation analysis; Figure 3 This is a schematic diagram illustrating a specific example of forced lane entry by a neighboring vehicle disclosed in this application; Figure 4This is a schematic diagram of a specific vehicle kinematic model disclosed in this application; Figure 5 This application discloses a specific method for predicting vehicle lane-changing risks. Figure 6 This is a schematic diagram of a vehicle lane-changing risk prediction device disclosed in this application. Figure 7 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] In existing technologies, the generation of adjacent vehicle lane-changing scenarios relies on manual annotation, fixed thresholds, and simple interpolation, which suffers from low efficiency, inaccuracy with actual conflict mechanisms, lack of kinematic constraints, and insufficient high-risk samples. Therefore, this application will specifically introduce a vehicle lane-changing risk prediction method that can solve the above problems.

[0019] See Figure 1 As shown in the figure, this application discloses a method for predicting vehicle lane-changing risks, including: Step S11: Extract driving segment video frames containing vehicle lane-changing behavior from the natural driving dataset based on vehicle operating parameters to obtain the target driving image sequence.

[0020] In this embodiment, the step of extracting driving video frames containing lane-changing behavior from a natural driving dataset based on vehicle operating parameters to obtain a target driving image sequence includes: collecting natural driving data within a preset distance range before the stop line of the intersection approach lane to obtain a natural driving dataset; the natural driving data includes driving videos and vehicle operating parameters synchronized with the driving videos in time; filtering driving segments containing lane-changing behavior from the natural driving dataset based on the vehicle operating parameters; uniformly extracting video frames from the driving videos corresponding to the driving segments at preset time intervals based on a time axis, and arranging the extracted video frames according to the extraction time sequence to obtain a target driving image sequence that retains the temporal context. Specifically, the preset distance range is set to 200m before the intersection stop line to cover the main occurrence range of lane-changing events at the approach lane. The driving videos and vehicle operating parameters (including vehicle speed, acceleration, lateral displacement, etc.) are collected synchronously in time to ensure the accuracy of subsequent parameter extraction. The video frame extraction adopts a fixed time interval uniform frame extraction method. The extracted frame sequence is arranged in chronological order, which fully preserves the temporal context of the lane-changing process. This ensures that the video large language model can perceive the dynamic evolution of the lane-changing behavior and provide effective image input for subsequent risk causation analysis.

[0021] Step S12: Input the preset target prompt words and the target driving image sequence into the preset video large language model, and determine the lane change risk causation parameters based on the structured risk causation analysis results output by the preset video large language model; the lane change risk causation parameters include the start and end times of the lane change, the duration of the lane change, the forward and backward time distances, and the speeds of each vehicle; the speeds of each vehicle include the speeds of the lane-changing vehicle, the vehicle in front of the target lane, and the vehicle behind the target lane.

[0022] In this embodiment, the step of inputting preset target prompt words and the target driving image sequence into a preset video large language model, and determining lane-change risk causal parameters based on the structured risk causal analysis results output by the preset video large language model, includes: constructing preset target prompt words including role definition fields, task planning fields, and output constraint fields; inputting the preset target prompt words and the target driving image sequence into the preset video large language model to obtain the structured risk causal analysis results output by the preset video large language model; locating key risk events based on the structured risk causal analysis results, and determining the lane-change risk causal parameters corresponding to the key risk events based on the distribution of lane-change behavior parameters corresponding to the key risk events in the natural driving dataset. That is, the image sequence is input into a pre-trained video large language model, and the model performs risk causal analysis through standardized prompt words, automatically identifying the forced entry behavior of neighboring vehicles and outputting a structured causal chain. Based on the causal chain, key risk events are located, and combined with the lane-change behavior parameter distribution obtained from the statistical analysis of natural driving data, the key parameters are confirmed for subsequent lane-change feasibility judgment and trajectory generation. Figure 2 As shown, firstly, a standardized prompt word template is constructed, with its core logic revolving around three roles: as a role definer to establish the expert's identity, as a task planner to decompose and analyze steps and logical processes, and as an output constraint to standardize the format and content scope. The design follows three principles: chain-thinking guidance to force explicit reasoning within the model, limited option constraints to limit the scope of key judgment terms, and templated output. This applies to situations where the vehicle is driving normally within the lane (e.g., ...). Figure 3As shown in the diagram, a vehicle in an adjacent lane forcibly cuts in when there is insufficient safety clearance. The large language model can output the following risky behavior: On a sunny, dry, multi-lane straight road, the vehicle behind in the target lane (the own vehicle) does not maintain sufficient longitudinal safety distance and emergency braking buffer space. When a vehicle in an adjacent lane suddenly cuts in and occupies the clearance in front, the own vehicle cannot decelerate in time to avoid it, resulting in a rear-end collision. Two causal chains can be deduced from this: First, controlling the unintended behavior manifests as delayed longitudinal braking, the preceding decision being delayed longitudinal acceleration / deceleration, which leads to a misjudgment of intent at the prediction level, ultimately attributing to external environmental influences—another vehicle suddenly changing lanes; Second, controlling the unintended behavior manifests as insufficient longitudinal deceleration, the preceding decision being insufficient longitudinal acceleration / deceleration, which leads to trajectory deviation at the prediction level, ultimately attributing to its own functional limitations—the lack of interactive influence prediction. Specifically, the preset target prompts adopt a three-part structure: the role definition field sets the model as a risk assessment expert with a background in traffic accident analysis; the task planning field requires the model to identify the forced entry behavior of neighboring vehicles, gradually infer the causal chain, and locate key risk events, following three principles: chain-like thinking guidance, limited option constraints, and templated output; the output constraint field specifies a structured output format to ensure the consistency and parsability of the output. Taking the typical scenario of forced entry by a neighboring vehicle as an example, the video big language model can deduce two causal chains: first, "unexpected control - late longitudinal braking" traces back to "late longitudinal acceleration / deceleration decision," then to "misjudgment of intent" in the perception layer, and finally to the influence of the external environment - "sudden lane change by another vehicle"; second, "unexpected control - insufficient longitudinal deceleration" traces back to "insufficient longitudinal acceleration / deceleration decision," then to "trajectory deviation," and finally to its own functional limitations - "lack of prediction of interaction impact." Based on the above causal chain, the key risk causal parameters are determined as follows: lane change duration Tlc (characterizing the suddenness of the lane change), lane change start time tlc,start, forward target distance THWfront target and backward target distance THWrear target when the lane change is completed (characterizing the safety margin after the lane change), and initial speed of the lane-changing vehicle ve(0), initial speed of the vehicle in front of the target lane vl(0) ​​and initial speed of the vehicle behind the target lane vr(0) (characterizing the collision risk caused by the relative speed difference).

[0023] Step S13: Based on the forward and backward time distances and the speeds of each vehicle, determine a feasibility judgment signal to characterize whether the lane-changing vehicle is allowed to change lanes.

[0024] In this embodiment, determining the feasibility judgment signal for allowing the lane-changing vehicle to change lanes based on the forward and backward time distances and the speeds of each vehicle includes: determining the forward time distance and the backward time distance in the forward and backward time distances; the forward time distance is the ratio of the forward clearance between the lane-changing vehicle and the vehicle in front of the target lane to the speed of the lane-changing vehicle in the speeds of each vehicle; the forward clearance is the value obtained by subtracting the vehicle length of the vehicle in front of the target lane from the difference between the longitudinal position of the vehicle in front of the target lane and the longitudinal position of the lane-changing vehicle; the backward time distance is the backward clearance between the lane-changing vehicle and the vehicle behind the vehicle in the target lane. The rear clearance is the ratio of the target lane following vehicle speed to the speed of each vehicle. The rear clearance is the difference between the longitudinal position of the lane-changing vehicle and the longitudinal position of the target lane following vehicle, minus the vehicle length of the lane-changing vehicle. The target distance between the lane-changing vehicle and the intersection stop line is determined, and the forward time distance threshold and the rear time distance threshold are constructed as a three-segment linear function of the target distance based on the preset far-end boundary distance and the preset near-end critical distance. When the forward time distance is not less than the forward time distance threshold and the rear time distance is not less than the rear time distance threshold, a feasibility judgment signal representing that lane changing is allowed is generated.

[0025] Specifically, to ensure forward and backward safety after lane changing, this patent considers the clearance constraints between the lane-changing vehicle and the vehicles in front and behind in the target lane. The forward clearance of the target lane is defined as: ; The target lane rear clearance is defined as: ; Where, x e x f x r The table shows the longitudinal positions L of the lane-changing vehicle, the vehicle in front of the target lane, and the vehicle behind the target lane. e L f Given the vehicle length, calculate the time headway (THW) corresponding to the front and rear clearances. ; ; Among them, v e For the speed of vehicles changing lanes, v r The speed of the vehicle behind in the target lane, To prevent extremely small values ​​where the denominator is zero, a value of 0.1s is set. Considering the urgency of forced lane changes at intersection entrances, this patent does not use a fixed time-distance threshold, but instead designs the front and rear time-distance thresholds as the distance d from the stop line. e The function that makes the threshold change with d e The contraction gradually decreases. To facilitate implementation and maintain model simplicity, a three-stage linear contraction approach is adopted. Let d...r d represents the distance to the far boundary. c The proximal critical distance (satisfying d) r d c Then, the forward time distance threshold is defined as: ; The backward time interval threshold is defined as: ; Where, d r d c To control the distance parameters of the threshold contraction interval, they were set to 80m and 40m respectively; , The forward conservative thresholds for the proximal and distal ends are set to 2.2s and 1.0s, respectively. , The backward conservative thresholds for the proximal and distal ends are set to 2.5s and 1.2s, respectively. The condition is met if and only if... and The system allows lane-changing to generate a feasibility determination signal to characterize whether the lane-changing vehicle is allowed to change lanes; otherwise, if the forced lane-changing condition is not met, the parameters are readjusted or the sample is discarded.

[0026] Step S14: Based on the feasibility determination signal and the lane-changing duration, perform displacement trajectory planning for the lane-changing vehicle to obtain an initial vehicle lane-changing trajectory, and verify the initial vehicle lane-changing trajectory based on preset vehicle kinematic constraints to obtain a feasible vehicle lane-changing trajectory.

[0027] In this embodiment, the step of planning the displacement trajectory of the lane-changing vehicle based on the feasibility determination signal and the lane-changing duration to obtain the initial vehicle lane-changing trajectory includes: constructing a vehicle kinematic model of the lane-changing vehicle; the vehicle kinematic model uses the center of mass position, heading angle, longitudinal velocity, and equivalent front wheel steering angle of the lane-changing vehicle as state variables, and the rate of change of the heading angle is the ratio of the product of the longitudinal velocity and the tangent of the equivalent front wheel steering angle to the wheelbase of the lane-changing vehicle; based on the feasibility determination signal, the lane-changing duration is used as the initial lane-changing duration, and the initial lane-changing duration is used as the time domain, and a fifth-order polynomial is used to construct the... The lateral displacement trajectory of the lane-changing vehicle from the centerline of the current lane to the centerline of the target lane; the lateral boundary conditions are that the lateral position at the start of the lane change is the starting lateral position of the lane change, and the lateral position at the end of the lane change is the center lateral position of the target lane, and the lateral velocity and lateral acceleration at the start and end of the lane change are both zero; the longitudinal acceleration response of the lane-changing vehicle to the vehicle in front in the target lane is determined based on a preset car-following model, and the longitudinal acceleration response is integrated sequentially to obtain the longitudinal displacement trajectory of the lane-changing vehicle; the lateral displacement trajectory and the longitudinal displacement trajectory are fused based on the vehicle kinematics model to obtain the initial vehicle lane-changing trajectory.

[0028] In this embodiment, the step of verifying the initial vehicle lane-changing trajectory based on preset vehicle kinematic constraints to obtain a feasible lane-changing trajectory includes: constructing preset vehicle kinematic constraints based on the vehicle kinematic model; the preset vehicle kinematic constraints include steering angle constraints, steering angular velocity constraints, and lateral acceleration constraints; determining the equivalent front wheel angle, equivalent front wheel angle change rate, and lateral acceleration corresponding to each moment based on the initial vehicle lane-changing trajectory, and verifying the equivalent front wheel angle, equivalent front wheel angle change rate, and lateral acceleration based on the preset vehicle kinematic constraints; if the verification passes, the initial vehicle lane-changing trajectory is taken as a feasible lane-changing trajectory; if the verification fails, the lateral boundary conditions remain unchanged, the initial lane-changing duration is increased by a preset time step, and the initial vehicle lane-changing trajectory is regenerated until the verification passes, so that the initial vehicle lane-changing trajectory when the verification passes is taken as a feasible lane-changing trajectory.

[0029] In summary, this application uses a classic kinematic bicycle model to describe the motion of lane-changing vehicles in a plane. This bicycle model is based on the following assumptions: 1. The vehicle's linear motion is ignored, meaning that the vehicle described in this application is a moving object on a two-dimensional plane; 2. Assuming that the vehicle's structure is like a bicycle, that is, the two front tires of the vehicle have the same angle and rotation speed, and the same is true for the two rear tires, then the front and rear tires can each be described by a single tire; 3. Assuming that the vehicle's motion is also like that of a bicycle, this means that the front tires control the vehicle's turning angle.

[0030] Under these assumptions, such as Figure 4 As shown, the planar kinematics of the vehicle can be characterized by the following state variables and control quantities: the position (x, y) of the vehicle's center of mass in the global coordinate system, and the vehicle's heading angle. Vehicle longitudinal speed v, and equivalent front wheel steering angle Let the vehicle distance be L, then the continuous-time kinematic equations of the bicycle model can be written as: ; in, and v is the vehicle's speed in the forward and backward, and left and right directions; v is the vehicle's longitudinal speed. This refers to the vehicle's heading angle; The table shows the angle of rotation of the front wheel; L represents the slant distance (the distance between the front and rear wheels, designed to be 2.9m in this invention).

[0031] Furthermore, to ensure the trajectory is physically executable, it must satisfy the vehicle's basic kinematic constraints.

[0032] Steering angle constraint: The steering wheel has a maximum steering angle limit. To ensure that the steering angle generated by the model does not exceed the physical limits of a real vehicle, the following equation must be satisfied: ; in, The maximum steering angle is designed to be 15° in this invention.

[0033] Steering angular velocity constraint: The steering wheel cannot instantly turn from 0 to its maximum angle; steering requires time (due to rate limitations imposed by the steering motor or driver operation). To limit the rate of change of the steering angle and ensure the continuity of steering actions, further requirements are made: ; in, The maximum rate of change of steering angle is designed to be 50° / s in this invention.

[0034] Curvature lateral acceleration constraint: Excessive lateral acceleration can lead to tire slippage, loss of vehicle control, or discomfort. To limit the maximum permissible value of lateral acceleration, the following equation must be satisfied: ; The corresponding lateral acceleration is approximately: Therefore: ; in For curvature, a y Centripetal acceleration; The maximum centripetal acceleration is set to 2.0 m / s² in this invention.

[0035] It should be noted that lane changing is essentially the process by which a vehicle transitions from the center line of the current lane to the center line of the target lane within a limited distance. Therefore, the problem of generating lane changing trajectories can be transformed into a trajectory construction problem of "longitudinal advancement + lateral offset".

[0036] First, the longitudinal progression. The acceleration expression for the basic FVD (Full Velocity Difference model) model is: ; in, , These are the acceleration and velocity of the following vehicle, respectively; v Let be the speed difference between the two vehicles at time t; This represents the driver's sensitivity coefficient to the difference in desired speed. This is the sensitivity coefficient to the difference in relative velocity.

[0037] To characterize the impact of DTI on car-following behavior, this section proposes an embedding structure for the distance compensation term, based on the behavioral assumption that "DTI affects expected speed." The FVD-DTI-2 formula is as follows: ; Where s represents the headway between the leading vehicle and the following vehicle; l represents the length of the leading vehicle; and V1, V2, C1, and C2 are the parameters to be calibrated for the optimal speed function. This represents the driver's sensitivity coefficient to DTI (Distance Compensation Intensity), where D is the distance from the intersection.

[0038] Next is the lateral offset, where the lane change starts at t=0 and the lane change duration is T. lc Let the lateral displacement be y(t). Let the lateral position of the lane change starting point be y0, and the lateral position of the target lane center be y0. f The total lateral offset is: .

[0039] To ensure a smooth lane change process and avoid abrupt changes in lateral velocity and acceleration, this paper uses a fifth-order polynomial to describe the lateral trajectory: .

[0040] Then, apply the following boundary conditions at the start and end of the lane change: .

[0041] The polynomial coefficients can be uniquely determined by the above six boundary conditions. This can be written in normalized form. Then, the lateral trajectory can be represented as: .

[0042] As can be seen from the above, the urgency of lane changing increases as the vehicle approaches the stop line, and the acceptable time-distance threshold dynamically shrinks as the distance to the stop line decreases. To ensure that the trajectory layer is consistent with this decision logic, this invention will further increase the lane changing duration T. lc The table is designed to be at a distance d from the stop line. e The function in the table states that the closer the vehicle is to the stop line, the shorter the lane-changing time and the more compact the trajectory; the farther the vehicle is from the stop line, the longer the lane-changing time and the smoother the trajectory. This can be expressed in the following piecewise linear form: ; Among them, T max and T min These represent the lane change durations corresponding to the far end and the near end, respectively, which are designed to be 4s and 2s in this invention.

[0043] Considering lane change duration T lc When the distance function from the stop line is used, it only reflects the urgency of the action and does not necessarily guarantee that the trajectory meets the vehicle's kinematic constraints. Therefore, this patent considers this time as the initial lane-changing time. Constraint verification is performed after the trajectory is generated.

[0044] If the generated trajectory does not meet the constraints of steering angle, steering angular velocity, or lateral acceleration, then the lane-changing time is corrected using an incremental method while keeping the trajectory boundary conditions unchanged. ; The trajectory is then regenerated and constraints are checked until the constraints are met or the maximum permissible lane change time is reached. T The design time is 0.2s. Therefore, the lane change time ultimately executed by this patent can be expressed as: ; in, This represents the minimum feasible lane change time that satisfies the vehicle's kinematic constraints.

[0045] Step S15: Based on the start and end times of the lane change and the feasible lane change trajectory of the vehicle, deduce the longitudinal interaction process between the lane-changing vehicle and the vehicle in front of the target lane and the vehicle behind the target lane to obtain lane change risk scenario samples, and determine the vehicle lane change risk prediction result based on the lane change risk scenario samples.

[0046] In this embodiment, the step of deriving the longitudinal interaction process between the lane-changing vehicle and the vehicle in front and the vehicle behind in the target lane based on the lane-changing start and end times and the feasible lane-changing trajectory of the vehicle to obtain lane-changing risk scenario samples includes: dividing the value range of risk tendency stratification parameters to obtain each risk tendency layer, and randomly sampling within each risk tendency layer to obtain the corresponding lane-changing scenario parameter combination; the risk tendency stratification parameters include the lane-changing duration and the target forward time distance and target backward time distance corresponding to the lane-changing end time in the forward and backward time distances; determining the longitudinal gap between the lane-changing vehicle and the vehicle in front and the vehicle behind in the target lane at the lane-changing end time based on the lane-changing scenario parameter combination and the speed of each vehicle, and recursively deriving the initial longitudinal position of each vehicle at the initial time based on the longitudinal gap; and iteratively deriving the longitudinal interaction process within a preset simulation time with a preset deviation step length based on the initial longitudinal position and the feasible lane-changing trajectory of the vehicle to obtain lane-changing risk scenario samples carrying the corresponding risk tendency layer identifiers.

[0047] During the approach to the stop line at an intersection, vehicles in adjacent lanes attempt to merge into the target lane from their original lanes. The vehicle in front in the target lane is traveling ahead, and the vehicle behind in the target lane is traveling behind in its own lane. After the lane change begins, a new longitudinal interaction is formed with the vehicles in front and behind in the target lane, thus evolving into a merging conflict. To improve sample coverage without screening for risk indicators, this application employs a risk-prone stratified sampling strategy to randomly generate initial conditions. Considering that the risk of a forced merge by an adjacent vehicle is mainly determined by the aggressiveness of the lane change and the available space before and after the merge, this invention selects the following three parameters as the basis for risk-prone stratification: lane change duration T... lc Table; Forward target distance THW upon completion of cut-in front target ; THW (Time Distance to Target) upon Completion of Cut-in rear target .

[0048] The scenario involves three vehicles: **Ego (Lane Changing Vehicle):** The ego is the vehicle approaching from the adjacent lane. Before the lane change begins, it maintains a constant speed within its original lane. After the lane change, its lateral trajectory is generated by a fifth-order polynomial, while its longitudinal motion is responded to using the FVD-DTI-2 car-following model based on the status of the vehicle in front in the target lane. **Lead Vehicle (Target Lane Lead):** The lead vehicle in the target lane primarily provides forward constraints in the simulation. Its longitudinal motion adopts an approximately constant speed driving mode, with a slight deceleration bias applied when approaching the stop line to simulate the natural deceleration behavior of vehicles approaching the intersection from the intersection entrance. **Rear Vehicle (Target Lane Following Vehicle):** The rear vehicle follows in the target lane. Before the ego enters, the rear travels at an approximately constant speed along the target lane. When the ego begins the lane change, the rear treats the ego as the new vehicle in front and responds longitudinally according to the FVD-DTI-2 car-following model, thus forming the process of rearward pressure and conflict evolution in the target lane.

[0049] To ensure that the lane-changing trajectory has clear phase divisions and reproducible event logic, ego behavior is divided into three phases, and its state transition process can be represented as: 1) Approach phase; in At that time, ego has not yet started changing lanes, maintaining its original lane and traveling straight at a constant speed, with no change in lateral position. 2) Lane changing phase: When At this point, the ego enters the lane-changing phase. The ego's lateral displacement is generated by a fifth-order polynomial trajectory to ensure that the lateral velocity and lateral acceleration are continuously zero at both the start and end points of the lane change, thus obtaining a smooth and controllable lateral trajectory. Simultaneously, the ego's longitudinal movement responds by following the vehicle in front in the target lane. 3) Lane-changing completion phase: When At this time, the ego completes the lane change and stably occupies the target lane, with its lateral position fixed in the center of the target lane. Subsequently, it continues to form a new longitudinal interaction relationship with the vehicles in front and behind in the target lane.

[0050] In this application, at time t=0, the following parameter is set: the initial distance of the lane-changing vehicle from the stop line. Initial velocity of the lane-changing vehicle Initial velocity of the vehicle in front in the target lane Initial velocity of the vehicle following in the target lane Initial longitudinal position of the lane-changing vehicle Initial longitudinal position of the vehicle ahead in the target lane Initial longitudinal position of the vehicle behind in the target lane The stop line is set at X=0, with a value taken upstream of the entrance lane. Therefore, the initial position of each vehicle is in front of the stop line.

[0051] The lane change event of ego is controlled by the following parameters: the start time of the lane change, t lc，start Lane change duration T lc; The target forward time distance between ego and the vehicle in front in the target lane when the cut is completed, THW front target ; Target rearward time distance between ego and the vehicle behind in the target lane when the cut is completed, THW rear target .

[0052] Among them, the following conditions must be met at the end of the lane change: .

[0053] It should be noted that this application does not directly and randomly sample the initial longitudinal positions of the three vehicles. Instead, it first determines the target forward and backward time distance at the time of lane change completion, and then uses the initial velocity of each vehicle to infer the longitudinal relationship at the time of lane change completion, thereby obtaining the initial positions of the three vehicles at t=0.

[0054] Furthermore, to improve sample coverage without screening for risk indicators, a risk-prone stratified sampling strategy is adopted to randomly generate initial conditions. Considering that the risk of a forced lane change by a neighboring vehicle is mainly determined by the aggressiveness of the lane change and the available space before and after the lane change, this paper selects the following three parameters as the basis for risk-prone stratification: lane change duration T. lc Table; Forward target distance THW upon completion of cut-in front target ; THW (Time Distance to Target) upon Completion of Cut-in rear target The samples are divided into three risk propensity layers: high, medium, and low. This is used to guide the distribution of samples in the parameter space, rather than serving as the final risk label. The parameter design table for the adjacent vehicle entry scenario is shown in Table 1 below: Table 1. Parameter Design Table for Adjacent Vehicle Interception Scenarios .

[0055] Specifically, the total simulation duration is set to 15s, and the discrete step size is set to 0.1s. There are three scene participants in total: the lane-changing vehicle (ego) travels straight at a constant speed in the original lane before lane changing, the lateral trajectory after lane changing is generated by a quintic polynomial, and the longitudinal motion adopts the FVD-DTI-2 car-following model to respond to the preceding vehicle in the target lane; the preceding vehicle in the target lane (lead) travels at an approximately constant speed, and a slight deceleration offset is applied when approaching the stop line to simulate the natural deceleration behavior; the following vehicle in the target lane (rear) regards the ego as a new preceding vehicle after the ego starts lane changing and performs a longitudinal response according to the FVD-DTI-2 model, forming a rearward compression and conflict evolution process. The behavior of ego is divided into three stages: in the approaching stage (when t<tlc,start), it keeps traveling straight at a constant speed in the original lane; in the lane changing stage (from tlc,start to tlc,end), a quintic polynomial trajectory is adopted in the lateral direction, and the car-following model response is adopted in the longitudinal direction; in the lane changing completion stage (t>tlc,end), it stably occupies the center of the target lane and continues to form a longitudinal interaction relationship with the preceding and following vehicles. The parameter ranges of the three types of risk tendency layers are mainly set based on the natural driving data analysis results: for the high-risk layer, Tlc is [2, 2.8]s, the forward time headway is [0.8, 1.4]s, and the rearward time headway is [0.8, 1.3]s; for the medium-risk layer, Tlc is [2.8, 3.8]s, the forward time headway is [1.4, 2]s, and the rearward time headway is [1.3, 2]s; for the low-risk layer, Tlc is [3.8, 5]s, the forward time headway is [2, 3]s, and the rearward time headway is [2, 3]s. The initial longitudinal positions of the three vehicles are not directly sampled randomly, but the target forward and rearward time headways at the lane changing completion moment are determined first, and then the longitudinal positions of each vehicle at t=0 are back-calculated according to the initial speed of each vehicle, so as to ensure the parameter logic consistency of the lane changing event, and batch generate lane changing risk scene samples with risk tendency layer identifiers.

[0056] As can be seen, in this embodiment, driving segment video frames containing lane-changing behavior are extracted from the natural driving dataset based on vehicle operating parameters to obtain a target driving image sequence; preset target prompt words and the target driving image sequence are input into a preset video large language model, and lane-changing risk causation parameters are determined based on the structured risk causation analysis results output by the preset video large language model; the lane-changing risk causation parameters include lane-changing start and end times, lane-changing duration, forward and backward time distances, and vehicle speeds; the vehicle speeds include the speeds of the lane-changing vehicle, the vehicle in front of the target lane, and the vehicle behind the target lane; based on the forward and backward time distances and the vehicle speeds, the lane-changing risk causation parameters are determined. Speed ​​determination is used to characterize the feasibility judgment signal for allowing the lane-changing vehicle to change lanes; based on the feasibility judgment signal and the lane-changing duration, displacement trajectory planning is performed on the lane-changing vehicle to obtain an initial lane-changing trajectory, and the initial lane-changing trajectory is verified based on preset vehicle kinematic constraints to obtain a feasible lane-changing trajectory; based on the lane-changing start and end times and the feasible lane-changing trajectory, the longitudinal interaction process between the lane-changing vehicle and the vehicle in front and behind in the target lane is deduced to obtain lane-changing risk scenario samples, and the vehicle lane-changing risk prediction result is determined based on the lane-changing risk scenario samples. Figure 5 As shown, the video large language model automatically extracts the parameters causing lane-changing risks, avoiding the subjective differences and high costs of manual frame-by-frame annotation. By designing the gap acceptance threshold as a dynamic function of the distance from the stop line, the system accurately depicts the forced lane-changing decision mechanism of drivers at intersection entrances who compress the acceptable gap due to limited remaining distance. Through five-order polynomial lateral trajectory and kinematic constraint verification, the smoothness and physical feasibility of the lane-changing trajectory are ensured. Through a risk-prone hierarchical sampling strategy, the system achieves batch generation of entry scenarios covering high, medium, and low risk levels, providing rich and reliable data support for intelligent driving system testing and risk assessment.

[0057] refer to Figure 6 The present application also discloses a vehicle lane-changing risk prediction device, comprising: The image sequence acquisition module 11 is used to extract driving segment video frames containing vehicle lane-changing behavior from the natural driving dataset based on vehicle operating parameters to obtain the target driving image sequence. The parameter determination module 12 is used to input the preset target prompt words and the target driving image sequence into the preset video large language model, and determine the lane change risk causation parameters based on the structured risk causation analysis results output by the preset video large language model; the lane change risk causation parameters include the start and end times of the lane change, the duration of the lane change, the forward and backward time distances, and the speeds of each vehicle; the speeds of each vehicle include the speeds of the lane-changing vehicle, the vehicle in front of the target lane, and the vehicle behind the target lane; Signal generation module 13 is used to determine a feasibility judgment signal for allowing the lane-changing vehicle to change lanes based on the forward and backward time distances and the speeds of each vehicle; The trajectory planning module 14 is used to plan the displacement trajectory of the lane-changing vehicle based on the feasibility judgment signal and the lane-changing duration to obtain the initial vehicle lane-changing trajectory, and to verify the initial vehicle lane-changing trajectory based on the preset vehicle kinematic constraints to obtain the vehicle feasible lane-changing trajectory. The risk prediction module 15 is used to deduce the longitudinal interaction process between the lane-changing vehicle and the vehicle in front of the target lane and the vehicle behind the target lane based on the start and end times of the lane change and the feasible lane-changing trajectory of the vehicle, so as to obtain lane-changing risk scenario samples and determine the vehicle lane-changing risk prediction result based on the lane-changing risk scenario samples.

[0058] As can be seen, in this embodiment, the above modules cooperate with each other to realize a complete processing flow from automatic extraction of natural driving data, risk cause analysis of video large language model, dynamic gap acceptance judgment, kinematic constraint trajectory generation to batch risk scenario output. It overcomes the shortcomings of existing technologies such as high cost of manual annotation, fixed thresholds that cannot reflect the dynamic urgency of intersections, simple interpolation that cannot guarantee the physical executability of the trajectory, and limited scenario diversity. It provides rich and reliable data support for intelligent driving system testing and risk assessment.

[0059] In some specific embodiments, the image sequence acquisition module 11 may specifically include: The data acquisition unit is used to collect natural driving data within a preset distance range before the stop line of the intersection approach lane to obtain a natural driving dataset; the natural driving data includes driving videos and vehicle operating parameters synchronized with the driving videos in time; The segment filtering unit is used to filter driving segments containing vehicle lane-changing behavior from the natural driving dataset based on the vehicle operating parameters. The video frame extraction unit is used to uniformly extract video frames from the driving video corresponding to the driving segment at preset time intervals based on the time axis, and arrange the extracted video frames according to the extraction time sequence to obtain a target driving image sequence that retains the temporal context.

[0060] In some specific embodiments, the parameter determination module 12 may specifically include: The prompt word construction unit is used to construct preset target prompt words, including role definition fields, task planning fields, and output constraint fields; The model analysis unit is used to input the preset target prompt words and the target driving image sequence into the preset video big language model to obtain the structured risk causation analysis results output by the preset video big language model; The parameter determination unit is used to locate key risk events based on the structured risk causation analysis results, and to determine the lane change risk causation parameters corresponding to the key risk events based on the distribution of lane change behavior parameters corresponding to the key risk events in the natural driving dataset.

[0061] In some specific embodiments, the signal generation module 13 may specifically include: A time-distance determination unit is used to determine the forward time distance and the backward time distance in the forward and backward time distances; the forward time distance is the ratio of the forward clearance between the lane-changing vehicle and the vehicle in front of the target lane to the speed of the lane-changing vehicle among the speeds of each vehicle, and the forward clearance is the value obtained by subtracting the vehicle length of the vehicle in front of the target lane from the difference between the longitudinal position of the vehicle in front of the target lane and the longitudinal position of the lane-changing vehicle; the backward time distance is the ratio of the backward clearance between the lane-changing vehicle and the vehicle behind the vehicle in the target lane to the speed of the vehicle behind the vehicle among the speeds of each vehicle, and the backward clearance is the value obtained by subtracting the vehicle length of the lane-changing vehicle from the difference between the longitudinal position of the lane-changing vehicle and the longitudinal position of the vehicle behind the vehicle in the target lane. The function construction unit is used to determine the target distance between the lane-changing vehicle and the intersection stop line, and constructs the forward time distance threshold and the backward time distance threshold as a three-segment linear function of the target distance based on the preset far-end boundary distance and the preset near-end critical distance; The signal determination unit is used to generate a feasibility determination signal indicating that lane changing is permitted when the forward time distance is not less than the forward time distance threshold and the backward time distance is not less than the backward time distance threshold.

[0062] In some specific embodiments, the trajectory planning module 14 may specifically include: The model building unit is used to build the vehicle kinematics model of the lane-changing vehicle. The vehicle kinematics model uses the center of mass position, heading angle, longitudinal velocity and equivalent front wheel steering angle of the lane-changing vehicle as state variables, and the rate of change of the heading angle is the ratio of the product of the longitudinal velocity and the tangent of the equivalent front wheel steering angle to the wheelbase of the lane-changing vehicle. The lateral trajectory construction unit is used to take the lane-changing duration as the initial lane-changing duration based on the feasibility determination signal, and construct the lateral displacement trajectory of the lane-changing vehicle from the center line of the current lane to the center line of the target lane using a fifth-order polynomial with the initial lane-changing duration as the time domain and according to the lateral boundary conditions; the lateral boundary conditions are that the lateral position at the start of the lane change is the lateral position of the lane-changing starting point, the lateral position at the end of the lane change is the lateral position of the center of the target lane, and the lateral velocity and lateral acceleration at the start and end of the lane change are both zero; The longitudinal trajectory construction unit is used to determine the longitudinal acceleration response of the lane-changing vehicle to the vehicle in front of the target lane based on a preset car-following model, and to integrate the longitudinal acceleration response in sequence to obtain the longitudinal displacement trajectory of the lane-changing vehicle. The vehicle trajectory construction unit is used to fuse the lateral displacement trajectory and the longitudinal displacement trajectory based on the vehicle kinematic model to obtain the initial vehicle lane-changing trajectory.

[0063] In some specific embodiments, the trajectory planning module 14 may specifically include: The constraint construction unit is used to construct preset vehicle kinematic constraints based on the vehicle kinematic model; the preset vehicle kinematic constraints include steering angle constraints, steering angular velocity constraints, and lateral acceleration constraints; The speed verification unit is used to determine the equivalent front wheel angle, equivalent front wheel angle change rate, and lateral acceleration at each moment based on the initial vehicle lane-changing trajectory, and to verify the equivalent front wheel angle, equivalent front wheel angle change rate, and lateral acceleration based on the preset vehicle kinematic constraints. The first trajectory determination unit is used to determine the initial vehicle lane-changing trajectory as the vehicle's feasible lane-changing trajectory if the verification is successful. The second trajectory determination unit is used to keep the lateral boundary conditions unchanged if the verification fails, increase the initial lane change duration by a preset time step and regenerate the initial vehicle lane change trajectory until the verification passes, so as to take the initial vehicle lane change trajectory when the verification passes as the vehicle's feasible lane change trajectory.

[0064] In some specific embodiments, the risk prediction module 15 may specifically include: A risk stratification unit is used to divide the value range of the risk tendency stratification parameters to obtain each risk tendency layer, and to randomly sample within each risk tendency layer to obtain the corresponding lane change scenario parameter combination; the risk tendency stratification parameters include the lane change duration, and the target forward time distance and target backward time distance corresponding to the lane change end time in the forward and backward time distances; The position determination unit is used to determine the longitudinal gap between the lane-changing vehicle and the vehicle in front of the target lane and the vehicle behind the target lane at the end of the lane-changing time based on the combination of lane-changing scenario parameters and the speed of each vehicle, and to reverse the longitudinal gap to obtain the initial longitudinal position of each vehicle at the initial time. The sample generation unit is used to iteratively simulate the longitudinal interaction process within a preset simulation time using a preset deviation step length based on the initial longitudinal position and the feasible lane-changing trajectory of the vehicle, so as to obtain lane-changing risk scenario samples carrying corresponding risk tendency layer identifiers.

[0065] Furthermore, embodiments of this application also disclose an electronic device, Figure 7 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0066] Figure 7 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the vehicle lane-changing risk prediction method disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0067] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0068] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0069] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the vehicle lane-changing risk prediction method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0070] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned vehicle lane-changing risk prediction method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0071] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0072] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0073] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0074] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0075] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting vehicle lane-changing risks, characterized in that, include: Based on vehicle operating parameters, driving video frames containing vehicle lane-changing behavior are extracted from the natural driving dataset to obtain the target driving image sequence. The preset target prompt words and the target driving image sequence are input into the preset video big language model, and the lane change risk causal parameters are determined based on the structured risk causal analysis results output by the preset video big language model. The lane change risk causation parameters include the lane change start and end times, lane change duration, forward and backward time distances, and the speeds of each vehicle; the speeds of each vehicle include the speeds of the lane-changing vehicle, the vehicle in front of the target lane, and the vehicle behind the target lane. Based on the forward and backward time distances and the speeds of each vehicle, a feasibility judgment signal is determined to characterize allowing the lane-changing vehicle to change lanes. Based on the feasibility determination signal and the lane change duration, the displacement trajectory of the lane-changing vehicle is planned to obtain the initial vehicle lane-changing trajectory, and the initial vehicle lane-changing trajectory is verified based on the preset vehicle kinematic constraints to obtain the vehicle feasible lane-changing trajectory. Based on the start and end times of the lane change and the feasible lane change trajectory of the vehicle, the longitudinal interaction process between the lane-changing vehicle and the vehicle in front of and behind the target lane is deduced to obtain lane change risk scenario samples, and the vehicle lane change risk prediction result is determined based on the lane change risk scenario samples.

2. The vehicle lane-changing risk prediction method according to claim 1, characterized in that, The step of extracting driving video frames containing vehicle lane-changing behavior from a natural driving dataset based on vehicle operating parameters to obtain a target driving image sequence includes: Natural driving data is collected within a preset distance range before the stop line of the intersection approach lane to obtain a natural driving dataset; the natural driving data includes driving videos and vehicle operating parameters synchronized with the driving videos in time; Based on the vehicle operating parameters, driving segments containing vehicle lane-changing behavior are filtered from the natural driving dataset. Based on the time axis, video frames are uniformly extracted from the driving video corresponding to the driving segment at preset time intervals, and the extracted video frames are arranged according to the extraction time sequence to obtain a target driving image sequence that retains the temporal context.

3. The vehicle lane-changing risk prediction method according to claim 1, characterized in that, The step of inputting the preset target prompt words and the target driving image sequence into a preset video large language model, and determining the lane change risk causation parameters based on the structured risk causation analysis results output by the preset video large language model, includes: Construct preset target prompts that include role definition fields, task planning fields, and output constraint fields; The preset target prompt words and the target driving image sequence are input into the preset video big language model to obtain the structured risk causation analysis results output by the preset video big language model; Based on the structured risk causation analysis results, key risk events are located, and based on the distribution of lane-changing behavior parameters corresponding to the key risk events in the natural driving dataset, the lane-changing risk causation parameters corresponding to the key risk events are determined.

4. The vehicle lane-changing risk prediction method according to claim 1, characterized in that, The determination of the feasibility judgment signal for allowing the lane-changing vehicle to change lanes, based on the forward and backward time distances and the speeds of each vehicle, includes: The forward and backward time distances are determined as follows: the forward time distance is the ratio of the forward clearance between the lane-changing vehicle and the vehicle in front of the target lane to the speed of the lane-changing vehicle among the vehicle speeds; the forward clearance is the value obtained by subtracting the vehicle length of the vehicle in front of the target lane from the difference between the longitudinal position of the vehicle in front of the target lane and the longitudinal position of the lane-changing vehicle; the backward time distance is the ratio of the backward clearance between the lane-changing vehicle and the vehicle behind the vehicle in the target lane to the speed of the vehicle behind the vehicle among the vehicle speeds; the backward clearance is the value obtained by subtracting the vehicle length of the lane-changing vehicle from the difference between the longitudinal position of the lane-changing vehicle and the longitudinal position of the vehicle behind the vehicle in the target lane. Determine the target distance between the lane-changing vehicle and the intersection stop line, and construct the forward time distance threshold and the backward time distance threshold as a three-segment linear function of the target distance based on the preset far-end boundary distance and the preset near-end critical distance; If the forward time distance is not less than the forward time distance threshold and the backward time distance is not less than the backward time distance threshold, a feasibility determination signal representing that lane changing is allowed is generated.

5. The vehicle lane-changing risk prediction method according to any one of claims 1 to 4, characterized in that, The step of planning the displacement trajectory of the lane-changing vehicle based on the feasibility determination signal and the lane-changing duration to obtain the initial vehicle lane-changing trajectory includes: Construct a vehicle kinematic model of the lane-changing vehicle; the vehicle kinematic model uses the center of mass position, heading angle, longitudinal velocity and equivalent front wheel steering angle of the lane-changing vehicle as state variables, and the rate of change of the heading angle is the ratio of the product of the longitudinal velocity and the tangent of the equivalent front wheel steering angle to the wheelbase of the lane-changing vehicle. Based on the feasibility determination signal, the lane-changing duration is taken as the initial lane-changing duration. Using the initial lane-changing duration as the time domain, a fifth-order polynomial is used to construct the lateral displacement trajectory of the lane-changing vehicle from the center line of the current lane to the center line of the target lane according to the lateral boundary conditions. The lateral boundary conditions are that the lateral position at the start of the lane change is the lateral position of the lane-changing starting point, and the lateral position at the end of the lane change is the lateral position of the center of the target lane. The lateral velocity and lateral acceleration at the start and end of the lane change are both zero. The longitudinal acceleration response of the lane-changing vehicle to the vehicle in front in the target lane is determined based on a preset car-following model, and the longitudinal acceleration response is integrated sequentially to obtain the longitudinal displacement trajectory of the lane-changing vehicle. The lateral displacement trajectory and the longitudinal displacement trajectory are fused based on the vehicle kinematic model to obtain the initial vehicle lane-changing trajectory.

6. The vehicle lane-changing risk prediction method according to claim 5, characterized in that, The step of verifying the initial vehicle lane-changing trajectory based on preset vehicle kinematic constraints to obtain a feasible lane-changing trajectory includes: Preset vehicle kinematic constraints are constructed based on the vehicle kinematic model; the preset vehicle kinematic constraints include steering angle constraints, steering angular velocity constraints, and lateral acceleration constraints; Based on the initial vehicle lane-changing trajectory, the equivalent front wheel angle, equivalent front wheel angle change rate, and lateral acceleration at each moment are determined, and the equivalent front wheel angle, equivalent front wheel angle change rate, and lateral acceleration are verified based on the preset vehicle kinematic constraints. If the verification is successful, the initial vehicle lane-changing trajectory will be taken as the feasible lane-changing trajectory of the vehicle. If the verification fails, the lateral boundary conditions remain unchanged, the initial lane-changing duration is increased by a preset time step, and the initial vehicle lane-changing trajectory is regenerated until the verification passes, so that the initial vehicle lane-changing trajectory when the verification passes is taken as the vehicle's feasible lane-changing trajectory.

7. The vehicle lane-changing risk prediction method according to claim 6, characterized in that, The process of deduceing the longitudinal interaction between the lane-changing vehicle and the vehicle in front and behind in the target lane based on the start and end times of the lane change and the feasible lane-changing trajectory of the vehicle, in order to obtain lane-changing risk scenario samples, includes: The range of values ​​for the risk propensity stratification parameters is divided to obtain each risk propensity layer, and random sampling is performed within each risk propensity layer to obtain the corresponding lane change scenario parameter combination; the risk propensity stratification parameters include the lane change duration, and the target forward time distance and target backward time distance corresponding to the lane change end time in the forward and backward time distances; Based on the combination of lane-changing scenario parameters and the speed of each vehicle, the longitudinal gap between the lane-changing vehicle and the vehicle in front of the target lane and the vehicle behind the target lane at the end of the lane change is determined, and the initial longitudinal position of each vehicle at the initial time is obtained by reverse calculation based on the longitudinal gap. Based on the initial longitudinal position and the vehicle's feasible lane-changing trajectory, the longitudinal interaction process is iteratively simulated within a preset simulation time using a preset distance step length, so as to obtain lane-changing risk scenario samples carrying corresponding risk tendency layer identifiers.

8. A vehicle lane-changing risk prediction device, characterized in that, include: The image sequence acquisition module is used to extract driving segment video frames containing vehicle lane-changing behavior from the natural driving dataset based on vehicle operating parameters to obtain the target driving image sequence. The parameter determination module is used to input the preset target prompt words and the target driving image sequence into the preset video big language model, and determine the lane change risk cause parameters based on the structured risk cause analysis results output by the preset video big language model. The lane change risk causation parameters include the lane change start and end times, lane change duration, forward and backward time distances, and the speeds of each vehicle; the speeds of each vehicle include the speeds of the lane-changing vehicle, the vehicle in front of the target lane, and the vehicle behind the target lane. The signal generation module is used to determine a feasibility judgment signal that characterizes allowing the lane-changing vehicle to change lanes, based on the forward and backward time distances and the speeds of each vehicle. The trajectory planning module is used to plan the displacement trajectory of the lane-changing vehicle based on the feasibility judgment signal and the lane-changing duration to obtain the initial vehicle lane-changing trajectory, and to verify the initial vehicle lane-changing trajectory based on the preset vehicle kinematic constraints to obtain the feasible vehicle lane-changing trajectory. The risk prediction module is used to deduce the longitudinal interaction process between the lane-changing vehicle and the vehicle in front of and behind the target lane based on the start and end times of the lane change and the feasible lane-changing trajectory of the vehicle, so as to obtain lane-changing risk scenario samples, and determine the vehicle lane-changing risk prediction result based on the lane-changing risk scenario samples.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the vehicle lane change risk prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store a computer program, which, when executed by a processor, implements the vehicle lane change risk prediction method as described in any one of claims 1 to 7.

Images