Personalized lane changing early warning system and method considering different speed conditions for curve tunnel

Abstract

The invention discloses a curve tunnel personalized lane change early warning system and method considering different speed conditions. The method comprises the following steps: collecting multi-source data; driving behavior characteristics are extracted based on the obtained data, and drivers are divided into a rash type, a stable type and a conservative type; establishing a nonlinear basic safety distance model under different speed intervals and different relative speed intervals of the two vehicles, and dynamically correcting a formula of the basic safety distance model in combination with a slope angle and a curve radius to obtain a three-dimensional safety model; generating an early warning rule in combination with the TTC threshold value and the safe distance model; and graded real-time early warning is conducted, specifically, graded early warning is triggered according to the safety indexes and early warning rules matched with the current situation, and early warning comprises visual prompt, sound warning and active braking intervention adopted according to the driving style. According to the method, driving style classification and a dynamic curvature-slope compensation algorithm are fused, and accurate lane changing early warning under different speed conditions in the curve tunnel is achieved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent driving assistance technology, and relates to a personalized lane change warning system and method for curved tunnels that takes into account different speed conditions. Background Technology

[0002] The lane-changing problems encountered in existing curved tunnels mainly stem from the combined effects of three-dimensional dynamic distortion, driver behavior differences, and sensor noise interference within the tunnel. Therefore, personalized lane-changing warnings for curved tunnels can achieve adaptability to the curved environment and integrate driving style classification. This is reflected in differentiated lane-changing threshold matching. Sensor noise interference is mitigated through redundancy-based anti-interference measures, enabling multi-source sensor fusion within the curved tunnel. This highlights the necessity of personalized lane-changing warnings for curved tunnels. Considering lane-changing warning systems under different speed conditions, this system can cover the full speed threshold conditions of curved tunnels. Furthermore, given the varying curve radii and gradients of curved tunnels, speed can be used to calculate lane-changing indices, achieving dynamic coupling of multiple parameters. Considering different speed conditions, this system can be integrated with driving style classification, providing earlier warnings for aggressive drivers on high-speed curves. By implementing the aforementioned personalized lane-changing warning system and method for curved tunnels under different speed conditions, full-speed-domain driving style-based lane-changing warnings can be achieved in this unique scenario, significantly improving lane-changing safety within curved tunnels.

[0003] Patent CN119872579A discloses a slope-compensated acceleration estimation method based on trajectory data. It explains that curved tunnels typically have slopes of varying angles, and the method compensates for driving acceleration based on the slope. Slope compensation is incorporated into the lane-changing model to correct the vehicle's lane-changing distance. Patent CN114633766A discloses a vehicle curve control compensation method, device, equipment, and medium. It mentions that when a vehicle travels on roads with discontinuous curvature or S-shaped bends, lane-changing behavior may cause deviations in the vehicle's trajectory. Therefore, it is necessary to correct the lane-changing model according to different radii. This eliminates the influence of the curved tunnel radius and slope on lane-changing behavior, enabling accurate assessment of safe lane-changing distances in curved tunnel environments.

[0004] Existing traditional lane change warning systems are mostly based on fixed-threshold safety distance models (such as TTC and TIV algorithms) or static kinematic equations. Static models do not fully consider the impact of vehicle speed segments, leading to deviations in safety distance calculations from actual risks. Furthermore, existing technologies are mostly designed for straight, flat roads, lacking dynamic compensation mechanisms for three-dimensional road features such as curve radius changes and gradient angles in curved tunnels, resulting in significant lateral offset errors. In addition, lane change models incorporating driving style can reduce false triggering rates and effectively prevent the warning threshold from becoming fixed. However, recent implementations of dynamic model-based solutions still have limitations. For example, speed range divisions are relatively coarse, failing to cover high-speed sections, and the simplistic environmental compensation mechanism makes real-time fitting of curved tunnels difficult. Simultaneously, with the popularization of intelligent driving, lane change models also need to overcome corresponding technical barriers, such as adaptive models across the entire speed range, multi-physics coupling designs, and personalized human-machine interaction with dynamic adjustments to warning parameters based on driver behavior characteristics. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a personalized lane-change warning system for curved tunnels that considers different speed conditions. It integrates driving style classification and dynamic curvature-slope compensation algorithms to achieve accurate lane-change warnings under different speed conditions within curved tunnels.

[0006] Another objective of this invention is to provide a personalized lane-changing early warning method for curved tunnels that takes into account different speed conditions.

[0007] The technical solution adopted in this invention is a personalized lane-changing early warning method for curved tunnels considering different speed conditions, comprising the following steps: S1. Multi-source data acquisition: Real-time acquisition of main vehicle speed, adjacent lane following vehicle speed, lateral offset, relative distance, acceleration and steering wheel angle data through on-board multiple sensors; S2. Based on the data obtained in S1, driving behavior features are extracted. These features include the average headway of the vehicle during multiple lane change events, the average minimum collision time, the rate of change of steering wheel angle, and the standard deviation of lateral acceleration. Using the average headway and the average minimum collision time as clustering features, drivers are classified into three types: aggressive, steady, and conservative. S3. Establish a nonlinear basic safety distance model under different speed ranges and different relative speed ranges between the two vehicles. Combine the formula for dynamically correcting the basic safety distance model with the slope angle and the curve radius to obtain a three-dimensional safety model. Generate early warning rules by combining the TTC threshold and the basic safety distance model, or by combining the TTC threshold and the three-dimensional safety model. S4. Tiered Real-Time Warning: Triggers tiered warnings based on warning rules adapted to the current situation. Warnings include visual cues, audible alerts, and active braking interventions based on driving style.

[0008] Furthermore, in S2, the driver segmentation uses the Mini-Batch k-means clustering algorithm: Using average headway and average minimum collision time as data input, infeasible and low-risk samples were excluded, and the data was standardized using Z-score. Select k data points as initial cluster centers, where k=3, corresponding to aggressive, stable, and conservative clusters; to eliminate the correlation of the covariance matrix, calculate the Mahalanobis distance from the sample to each cluster center; Samples are assigned to the nearest cluster based on Mahalanobis distance, and the cluster centers are updated. The maximum number of iterations is set, and the iteration frequency is synchronized with the vehicle CAN bus refresh rate. If the continuous iteration loss decrease rate is less than 1%, the batch size is increased. The final cluster labels and centroid coordinates are output. The results of each cluster label are used as the standard for driving style classification, thereby classifying drivers into aggressive, steady, and conservative types.

[0009] Furthermore, the Z-score normalization process is as follows: ; in, z i Indicates the first i Standardized values ​​for each sample For the first in the original data i Each sample value; The mean of the dataset; σ The standard deviation of the dataset; Mahalanobis distance The calculation formula is: ; in, Indicates sample With cluster center The transpose of the difference vector; This represents the inverse of the covariance matrix.

[0010] Furthermore, in S2, the criteria for classifying driving styles are as follows: For aggressive drivers: MTH ≤ 1.36s, TTC_min ≤ 4.21s; Stable driving: 1.36s <MTH≤1.83s,4.21s<TTC_min≤7.62s; Conservative driver: MTH > 1.83s, TTC_min > 7.62s; Where MTH represents the average headway of multiple lane change events by the driver, and TTC_min represents the average minimum collision time of multiple lane change events by the driver.

[0011] Furthermore, in S3, the method for constructing the basic safety distance model is as follows: Assuming the main vehicle and the following vehicle in the adjacent lane maintain a constant speed during lane change, and the speed of the main vehicle during lane change is divided into multiple intervals, the safety distance DSS calculation formula is simplified based on the average lane change time in each speed interval. The simplified safety distance DSS calculation formula is then combined with TTC. When the absolute value of the relative speed between the main vehicle and the following vehicle in the adjacent lane is >15 km / h, the time interval T is set to TTC_min; when the absolute value of the relative speed between the main vehicle and the following vehicle in the adjacent lane is ≤15 km / h, the time interval T is set to MTH. Here, MTH represents the average headway of multiple lane change events, and TTC_min represents the average minimum collision time of multiple lane change events. This leads to the establishment of a nonlinear basic safety distance model under different speed intervals and different relative speed intervals between the two vehicles.

[0012] Furthermore, in S3, the basic safety distance model is as follows: ; DWS stands for Basic Safety Distance Threshold.

[0013] Furthermore, in S3, the method for determining the three-dimensional security model is as follows: ; in, This represents the three-dimensional safety distance threshold, while DWS represents the basic safety distance threshold obtained from the basic safety distance model. This represents the safety distance correction based on the curve radius. This represents the safety distance correction based on the slope angle; k is an empirical coefficient. φ This is the ratio of lateral acceleration to gravitational acceleration. This reflects the dynamic impact of lateral forces on a vehicle in a curve; R represents the real-time curve radius. μ The coefficient of friction of the road surface. The slope angle is denoted by .

[0014] Furthermore, in S3, the warning rule is as follows: The triggering conditions for a Level 1 warning are: actual distance ∈ [80%·DWS, 100%·DWS] or TTC ∈ [5s, 8s); The triggering conditions for a Level 2 warning are: actual distance ∈ [60%·DWS, 80%·DWS) or TTC ∈ [3s, 5s); The trigger condition for a Level 3 warning is: actual distance < 60% DWS or TTC < 3s; When two safety indicators conflict, the one with the higher risk level will be selected for early warning.

[0015] Furthermore, in S4, the tiered early warning is as follows: Level 1 Warning: The HUD displays a yellow warning box, and the dashboard flashes a lane change risk icon; Level 2 warning: An audible warning is given with a frequency that increases linearly from 1.5kHz to 3kHz, with the frequency gradually increasing with the level of danger; Level 3 Warning: Intervene with active braking based on driving style. Active braking includes braking force control and steering wheel angle limitation. Braking force control applies 20%-30% of the basic braking force to aggressive drivers; 30%-40% of the basic braking force to moderate drivers; and 40%-50% of the basic braking force to conservative drivers. Steering wheel angle limitation restricts the maximum angle to -20° to 20° for aggressive drivers, -15° to 15° for moderate drivers, and -10° to 10° for conservative drivers. The dynamic delay compensation mechanism is as follows: when TTC < 2s, pre-filled hydraulic braking is activated after a fixed base oil pressure has been established in advance; if the first-level warning continues for a fixed time without being lifted, it will automatically be upgraded to the second-level warning.

[0016] A personalized lane-change early warning system for curved tunnels considering different speed conditions, employing the aforementioned personalized lane-change early warning method for curved tunnels considering different speed conditions, includes: The multi-source data acquisition module is used to collect data on the speed of the main vehicle, the speed of vehicles following in adjacent lanes of the target lane, lateral offset, curve radius, slope angle, relative distance, acceleration, and steering wheel angle through millimeter-wave radar, LiDAR, camera, IMU, and GPS; and synchronizes the time through the V2X module. The driving style classification module uses average headway and average minimum collision time as clustering features to classify drivers into aggressive, steady, and conservative types. The lane change warning model building module is used to establish a nonlinear basic safety distance model based on different speed ranges and different relative speeds between two vehicles. It generates warning rules by combining the TTC threshold and the basic safety distance model, or by combining the TTC threshold and the three-dimensional safety model, and sends the warning rules to the real-time warning module. The three-dimensional safety model calculation module is used to combine the formula of slope angle and curve radius to dynamically correct the basic safety distance model and obtain the three-dimensional safety model; The real-time warning module is used to trigger tiered warnings based on warning rules adapted to the current situation. Warnings include visual cues, audible alerts, and active braking interventions based on driving style.

[0017] The beneficial effects of this invention are: 1. This invention constructs a nonlinear safety distance formula for speed intervals, solving the problem that traditional static models cannot adapt to different vehicle speeds; considering differences in driving styles, it performs cluster analysis on different driving styles, constructs a safety model for speed intervals and personalized active braking intervention, and uses the Mini-Batch K-means clustering method to achieve real-time warning acceleration processing.

[0018] 2. This invention combines a safe lane-change early warning model with the curve radius and slope factors of curved tunnels for dynamic correction, and applies it to lane-change early warning in curved tunnel scenarios; it uses LiDAR point cloud to fit the curve radius in real time, integrates IMU / GPS data to dynamically correct the slope influence, and establishes a three-dimensional trajectory compensation model (three-dimensional environmental compensation) to improve the accuracy of lane-change early warning in curved tunnels.

[0019] 3. This invention utilizes the absolute value of relative speed to select the shortest headway (TTCi). Here, the average headway and the average shortest collision time are used for early warning, denoted as MTH and TTC_min, respectively, or the shortest collision time (TTC) and the average shortest collision time (TTC_min) to construct the early warning model, optimizing the early warning accuracy in high relative speed scenarios (with particular attention to dangerous lane-changing scenarios where the speed of the vehicle in front is greater than the speed of the following vehicle in the adjacent lane).

[0020] 4. This invention also provides a new approach to lane change warning systems in ordinary tunnel scenarios. In curved tunnel scenarios, it leverages a three-dimensional trajectory compensation model to improve lane change accuracy. In ordinary tunnels, i.e., straight and gentle slope scenarios, it switches between disabling curve radius and slope compensation items. This is achieved by loading a scenario configuration file into the vehicle system for identification and disabling, enabling lane change warning systems in different speed ranges within ordinary tunnels. It is applicable to ordinary tunnel scenarios and functions fully in curved tunnel scenarios. The sensors and execution layer devices meet the perception and execution requirements of ordinary tunnels. The driving style classification algorithm is scenario-independent and also applicable to ordinary tunnels. The dynamic safety distance model for different speed ranges can be directly applied to ordinary tunnels. Lateral offset detection in ordinary tunnels can utilize cameras for lane line recognition. The warning rules in the graded warning system are also applicable to ordinary tunnels. Attached Figure Description

[0021] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a system architecture diagram according to an embodiment of the present invention.

[0023] Figure 2 This is a flowchart illustrating the calculation process of the dynamic safety distance model in an embodiment of the present invention.

[0024] Figure 3 This is a schematic diagram of the three-dimensional environmental compensation mechanism in an embodiment of the present invention.

[0025] Figure 4 This is a schematic diagram of the hardware deployment and interface of an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0027] The basic concept of this invention is as follows: By acquiring data from multiple sensors and fusing external data, the system extracts slope angle, lateral offset, vehicle speed, speed of adjacent vehicles in the target lane, relative speed, acceleration, steering wheel angle and rate of change, relative distance, curve radius, and road surface friction coefficient. Driving behavior characteristics are then extracted. Using the Mini-Batch K-means algorithm, drivers are categorized into three driving styles: aggressive, stable, and conservative. For different speed ranges and driving styles, a dynamic safe distance model and personalized active braking intervention are constructed to achieve personalized lane change warnings. A relative speed segmentation strategy is introduced to distinguish the relative speed conditions between the vehicle in the main lane (SV) and the adjacent vehicle in the front lane (FV), implementing graded warnings based on driving behavior. This effectively reduces the risk of lane change accidents and increases user acceptance of the automated driving assistance system. Furthermore, by fitting the curve radius using high-precision maps and real-time LiDAR scanning, and combining IMU and GPS data to calculate the slope angle and curve compensation, a three-dimensional safe distance model is constructed. This model is then applied to curved tunnel scenarios, and a tiered warning system is implemented to alert and proactively intervene in dangerous driving behaviors. Lane-changing warnings are issued for dangerous lane-changing behaviors within the tunnel to reduce traffic accidents caused by illegal lane-changing. The scene recognition module automatically switches between ordinary tunnel and curved tunnel warning modes, with the curve radius and slope compensation items disabled in the ordinary tunnel mode.

[0028] A personalized lane-change early warning system for curved tunnels that considers different speed conditions, such as Figure 1 As shown, it includes: The multi-source data acquisition module is used to collect data on the speed of the main vehicle, the speed of vehicles following in adjacent lanes of the target lane, lateral offset, relative distance, acceleration, and steering wheel angle through millimeter-wave radar, LiDAR, camera, IMU, and GPS; time synchronization is achieved through the V2X module (PTP protocol, error <1ms); sensor data cleaning is performed; and features such as curve radius fitting, slope angle calculation, relative speed, and distance are extracted.

[0029] The driving style classification module takes MTH (mean headway) and TTC_min (mean minimum collision time) as input, matches different driving styles through cluster analysis, and outputs driving style labels and dynamic parameters of safe distance (referring to lane change duration characteristics, i.e., aggressive drivers have a peak lane change duration of 3-4 seconds, steady drivers 4-5 seconds, and conservative drivers 5-6 seconds). The Mini-Batch K-means algorithm is used to classify drivers into aggressive, steady, and conservative types.

[0030] The lane change warning model construction module is used to establish a nonlinear basic safety distance model based on different speed ranges and different relative speeds of two vehicles. It generates warning rules by combining the TTC threshold and the basic safety distance model, or by combining the TTC threshold and the three-dimensional safety model, and sends the warning rules to the real-time warning module.

[0031] The 3D safety model calculation module integrates the curve radius from LiDAR and high-precision maps with the slope angle from IMU and GPS to calculate the safety distance correction based on the curve radius and the safety distance correction based on the slope angle. It then applies a 3D correction to the formula of the basic safety distance model to obtain the 3D safety model, which is then sent to the lane change warning model construction module.

[0032] The real-time early warning module is used to trigger tiered early warnings based on early warning rules adapted to the current situation. Early warnings include visual cues, audible alerts, and active braking interventions.

[0033] Among them, the multi-source data acquisition module, driving style classification module and three-dimensional safety model calculation module constitute the prediction module 1, which is used for environmental perception (data acquisition and fusion) and decision calculation, and is deployed in the vehicle computing unit (such as TI TDA4VM).

[0034] The safety indicator dynamic selection module, lane change warning model construction module, and real-time warning module constitute warning module 2, which is used for command execution and dynamic control. It is integrated into the vehicle execution system (ESP / EPS) and its core functions include graded warning triggering, personalized control strategy execution, and vehicle dynamic intervention.

[0035] Multi-sensor redundancy anti-interference design, eliminating metal reflection noise by fusing millimeter-wave radar and LiDAR point cloud; The 3D trajectory prediction module dynamically adjusts the control points for lane-changing trajectories based on Bézier curves, with the number of control points adaptively selected according to the curvature of the curve.

[0036] A personalized lane-change early warning method for curved tunnels considering different speed conditions, employing the aforementioned personalized lane-change early warning system for curved tunnels considering different speed conditions, includes the following steps: S1. Multi-source data acquisition and processing: Vehicle sensor data: Real-time acquisition of the main vehicle's speed, the speed of vehicles following in adjacent lanes in the target lane, relative distance, acceleration, and steering wheel angle data through multiple onboard sensors.

[0037] Specifically, millimeter-wave radar is used to measure the relative speed and relative distance of the target lane in real time; LiDAR is used to fit the contours of adjacent lane following vehicles through point cloud data to calculate the lateral offset (accuracy ±0.3 meters) and relative distance (accuracy ±0.3 meters); and the RANSAC algorithm is combined to optimize noise interference.

[0038] Cameras are used to assist in perceiving the surrounding environment; an IMU (Inertial Measurement Unit) is used to collect dynamic parameters such as vehicle acceleration and yaw rate; GPS is used to obtain vehicle position and elevation data; IMU and GPS are fused by Kalman filtering to combine IMU acceleration and GPS elevation data and calculate the slope angle in real time.

[0039] The CAN bus is used to connect sensing units such as millimeter-wave radar, LiDAR, IMU, and GPS with onboard computing units (such as NVIDIA Xavier AGX) to achieve synchronous transmission of multi-source data (such as vehicle speed, yaw rate, and distance to the target vehicle). It transmits warning commands in real time to execution units such as electronic power steering (EPS) and electronic stability program (ESP) to ensure low-latency execution of active control actions (such as increased steering resistance and gradient braking), with a latency of <50ms.

[0040] External data: Static road information such as curve radius and slope of curved tunnels is pre-stored through high-precision maps; V2X communication module is used to receive real-time traffic information from roadside units (RSU) to enhance environmental perception capabilities.

[0041] The main models of the aforementioned equipment include: Sensing layer devices: millimeter-wave radar (Bosch MRR5), LiDAR (Hesai AT128), camera (Sony IMX490), IMU (ADI ADIS16470), V2X communication module (Huawei MH5000 (5G+V2X)).

[0042] Computing and control layer devices: Data fusion processor (NVIDIA Xavier AGX), warning command redundancy controller (Infineon AURIX TC397).

[0043] Actuation layer equipment: HUD (Continental AR-HUD Gen3), brake actuator (Bosch ESP® Hev II), steering actuator (ZF EPS-MD2), such as Figure 4 As shown.

[0044] S2. Data processing: Based on the data obtained in step S1, calculate in real time the speed of the main vehicle, the relative speed (Δv) of the vehicles following in the adjacent lanes of the target lane, the lateral offset, and the slope angle; further extract driving behavior features: average headway, minimum collision time, steering wheel angle change rate, and lateral acceleration standard deviation.

[0045] Among them, the speed of the main vehicle serves as the basic input to the safety model, and the calculation of the safe distance needs to be based on the speed of the main vehicle and divide the speed range; the relative speed of the following vehicles in the adjacent lanes of the target lane is used to distinguish the relative speed scenario to calculate TTC_min (mean shortest collision time) and Time Gap (mean headway); the lateral offset is used to monitor whether the vehicle deviates from the lane to determine the lane-changing intention, and can be used to calculate the compensation amount in combination with the curve radius; the slope angle is used to correct the safe distance model, and a shorter safe distance is required when going uphill and a longer safe distance is required when going downhill.

[0046] Average headway and average minimum collision time are key features for classifying driving styles; steering wheel angle change rate and lateral acceleration standard deviation are used to assist in judging driving style.

[0047] Using average headway (MTH) and average minimum collision time (TTC_min) as clustering features, the risk preferences of driving styles are reflected (adventurous drivers have short MTH and low TTC_min; conservative drivers have the opposite). Through cluster analysis, the results, together with the driving style labels, are output as lane change duration features, that is, adventurous drivers have short lane change durations (peak 3-4 seconds), steady drivers have 4-5 seconds, and conservative drivers have 5-6 seconds.

[0048] S3. Driving style classification: Construct a standard for driving style classification, and use the Mini-Batch k-means clustering algorithm to classify drivers into aggressive, steady, and conservative types.

[0049] The Mini-Batch K-means algorithm uses the average headway (MTH) and average minimum collision time (TTC_min) as input data. It removes samples where MTH < 0.5 seconds or TTC_min > 15 seconds to exclude infeasible and low-risk samples. The data (average headway and average minimum collision time) are then Z-score standardized.

[0050] in, z i Indicates the first i Standardized values ​​for each sample x i For the first in the original data i Individual sample values ​​(i.e., a single data point vector); The mean of the dataset; σ denoted as the standard deviation of the dataset.

[0051] Select k data points as the initial clustering centers (k = 3, corresponding to aggressive, stable, and conservative types), and take 256 as the batch size for real-time processing; to eliminate the correlation of the covariance matrix, calculate the Mahalanobis distance from the samples to each clustering center :

[0052] : The mean vector of the mean distribution of the data set; Σ: The covariance matrix of the distribution (reflecting the correlation between variables); Σ −1 : The inverse matrix of the covariance matrix; Among them, among them, represents a single data point vector and the difference vector transpose between the mean vector of the mean distribution of the data set (clustering center), indicating the deviation between the two in each feature dimension; represents a single data point vector and the mean vector of the mean distribution of the data set, which is the inverse matrix of the covariance matrix of the difference vector, used to eliminate the correlation between features and adjust the weights of each dimension.

[0053] Allocate samples to the nearest cluster according to the Mahalanobis distance and update the clustering centers. The maximum number of iterations is 100, and the iteration frequency is synchronized with the vehicle CAN bus refresh rate. If the continuous iteration loss decrease rate < 1%, increase the batch size to 512, and output the final clustering labels and centroid coordinates. Obtain the results of each clustering label as the standard for driving style classification.

[0054] Obtain the results of each clustering label: Aggressive drivers: MTH ≤ 1.36 seconds, TTC_min ≤ 4.21 seconds; Stable drivers: 1.36 seconds < MTH ≤ 1.83 seconds, 4.21 seconds < TTC_min ≤ 7.62 seconds; Conservative drivers: MTH > 1.83 seconds, TTC_min > 7.62 seconds.

[0055] S4. As Figure 2 shown, divide the main vehicle speed (v s ) into four intervals: low-speed interval (v s ≤ 70 km / h); medium-speed interval (70 km / h < v s ≤ 90 km / h); high-speed interval (90 km / h < v s ≤ 110 km / h); ultra-high-speed interval (v s>110 km / h).

[0056] Will Divided into three sections: . The relative speed between the two vehicles (SV and FV) is (i.e., the speed of the main vehicle minus the speed of the following vehicle in the adjacent lane), in km / h.

[0057] Speed ​​range and relative speed between the two vehicles (SV and FV) Constructing a safe distance model:

[0058] DWS stands for Basic Safety Distance Threshold.

[0059] Considering that driving safety cannot be guaranteed when the speed of the lane SV is faster than that of the adjacent lane FV, and given that the deceleration of the FV is relatively large when the SV changes lanes frequently, this situation is taken into account when constructing the lane change analysis and warning model in this embodiment of the invention. When the SV is faster than the FV, the relative speed is positive; when the FV is faster than the SV, the relative speed is negative.

[0060] When the absolute value of the relative speed is greater than 15 km / h: the value of TTC_min is relatively stable, and TTC_min is used as the time indicator for calculating the safe distance threshold. According to relevant references, T=TTC_min=5s is selected here, as the recognition accuracy is higher than that when T=3s.

[0061] When the absolute value of the relative speed is ≤15km / h: the headway is relatively stable, and MTH is used as the indicator for calculating the safe distance threshold. Based on the reference of the headway warning value of relevant following warning systems, T=MTH=0.6s is selected here.

[0062] In actual warnings, the actual TTC and MTH are calculated by the system and compared with the threshold corresponding to the driving style to trigger graded warnings; the calculation data source is: real-time acquisition by sensors (frequency 50-100Hz) and transmission via CAN bus.

[0063] Warning rules are generated based on a safe distance model combined with the average headway (MTH) and the TTC threshold (TTC_min): The triggering conditions for Level 1 warning (low risk) are: actual distance ∈ [80%·DWS, 100%·DWS] or TTC ∈ [5s, 8s); The triggering conditions for a Level 2 warning (medium risk) are: actual distance ∈ [60%·DWS, 80%·DWS) or TTC ∈ [3s, 5s); The trigger condition for a Level 3 warning (high risk) is: actual distance <60%·DWS or TTC <3s.

[0064] In this embodiment of the invention, spatial DWS and temporal TTC_min are selected as safety indicators for judgment. When the two safety indicators conflict, the one with the higher risk level is selected for early warning.

[0065] In curved tunnel scenarios, the lateral offset is calculated and the early warning response is optimized by combining dynamic correction of slope and curve radius.

[0066] The method for calculating lateral offset is as follows: Lane lines and vehicle contours are fitted using LiDAR point cloud data, combined with the lateral displacement formula in the main vehicle coordinate system.

[0067] in, Indicates the lateral offset. The coordinates are for the lane centerline, with an accuracy of ±0.3m. This is the vehicle's current lateral coordinate, representing the position of the vehicle's centerline in the lane coordinate system, with an accuracy of ±0.3m. This reflects the error range of the LiDAR sensor in measuring the vehicle.

[0068] Lateral offset is crucial for calculating curve radius compensation. It significantly impacts safety distances, enabling early prediction of curves. Calculating lateral offset reduces misjudgments caused by lateral displacement in curves, improving warning accuracy. Lateral offset is greater in road sections with smaller curve radii. Incorporating empirical coefficients reflects the sensitivity of lateral offset to curve radius compensation.

[0069] S5, Tiered Real-time Early Warning; When the real-time distance between the main vehicle and the following vehicle in the adjacent lane of the target lane is less than the safe distance model threshold (DWS), a graded warning is triggered, including visual cues, audible warnings, and active braking intervention.

[0070] Level 1 Warning (Low Risk): The HUD displays a yellow warning box, and the dashboard flashes a lane change risk icon; Level 2 Warning (Medium Risk): An audible warning is given with a frequency that increases linearly from 1.5kHz to 3kHz, with the frequency gradually increasing with the level of danger. Level 3 Warning (High Risk): Intervention is implemented using active braking based on driving style. Active braking methods include brake force control and steering wheel angle limitation. Brake force control applies 20%-30% of the base braking force to aggressive drivers (the baseline value is 2% increase for every 10 km / h increase in speed); 30%-40% of the base braking force to moderate drivers; and 40%-50% of the base braking force to conservative drivers. Steering wheel angle limitation restricts the maximum angle to -20° to 20° for aggressive drivers, -15° to 15° for moderate drivers, and -10° to 10° for conservative drivers.

[0071] The dynamic delay compensation mechanism is as follows: when TTC < 2s, pre-filled hydraulic braking is activated after 10% base oil pressure is established in advance; if the first-level warning is not lifted for 1.5s, it is automatically upgraded to the second-level warning; the braking system is in a "ready to go" state, and the algorithm predicts the hardware status in advance, achieving millisecond-level response in extreme scenarios and breaking through the physical delay bottleneck of the hydraulic system.

[0072] This invention's multi-sensory, progressive warning system effectively balances alertness and distraction. Traditional warning systems, with their singular nature, often lead to driver neglect. This invention employs a linearly increasing sound frequency, which attracts more of the brain's attention (the "strange sphere effect" in psychology), resulting in a significantly better warning effect than fixed-frequency sounds. Combined with an automatic escalation mechanism that upgrades the warning if it persists for 1.5 seconds, a smooth, time-based warning progression is created, providing drivers with reasonable reaction time. This multi-coupling design of "intensity-time-sensory dimensions" accurately simulates the gradual psychological tension experienced by humans when facing potential threats. It not only physically calculates risk but also optimizes human-computer interaction efficiency at a neuropsychological level. This deep, interdisciplinary optimization leads to an experience improvement that traditional single-threshold systems struggle to achieve.

[0073] Example 1

[0074] Lane change event data collected from real-world road tests (including 2752 lane change events involving 50 drivers) was used. Lane change scenarios with following vehicles (FVs) in adjacent lanes were selected, totaling 1496 valid events. Events with mean headway (MTH) <0.5 seconds or time to turn (TTC) were excluded. min Samples with an average minimum collision time > 15 seconds were excluded from infeasible and low-risk samples. Through the above clustering steps (data preprocessing, Mahalanobis distance calculation, and dynamic iterative optimization, accurate classification of driving styles was achieved), the specific classification criteria for each cluster label were obtained: Aggressive drivers: MTH ≤ 1.36 seconds, TTC_min ≤ 4.21 seconds; Steady drivers: 1.36 s < MTH ≤ 1.83 s, 4.21 s < TTC_min ≤ 7.62 s; Conservative drivers: MTH > 1.83 s, TTC_min > 7.62 s.

[0075] The above classification criteria are statistically obtained based on large-scale historical data (50 drivers, 2,752 lane-changing events), which can be used as the basic classification benchmark for the system. It is determined once during system deployment and is valid for a long time. The classification criteria themselves do not need to be re-determined every time a lane change occurs, but the classification labels of specific drivers need to be updated in real time.

[0076] By speed interval and relative speed Construct a non-linear safe distance model, dynamically correct it in combination with slope and curve radius, and calculate the lateral offset to optimize the warning response.

[0077] During the entire lane-changing process, considering that the SV (subject vehicle) must always be in front of the FV (following vehicle in the adjacent lane) and there is a certain distance between the two vehicles, the final safe distance DSS is as follows:

[0078] In the formula, D R is the longitudinal distance between the SV and the FV at the start of the lane change, Max{D R , 0} is the distance between the two vehicles at the start of the lane change (taking the maximum value of D R and 0), and B0 is the relative distance between the SV and the FV after entering the adjacent lane.

[0079] Specifically, the minimum distance between the two vehicles at the start of the lane change is set to 0 when the FV cannot keep up with the SV.

[0080] The longitudinal distance D R between the SV and the FV at the start of the lane change should satisfy:

[0081] Among them, D S is the longitudinal distance when the SV completely enters the adjacent lane (the distance between the rear of the SV that has completely entered the adjacent lane and the rear of the SV that has not entered the adjacent lane), D F is the longitudinal distance of the FV (the distance between the front of the FV when the SV completes the lane-changing process), D W is the width of the SV, and θ is the angle between the front of the SV and the boundary line of the adjacent lane.

[0082] To ensure that the FV is always behind the SV, it is necessary to satisfy:

[0083] To avoid a potential collision when two vehicles enter adjacent lanes, the following conditions should be met:

[0084] Among them, B S B is the driving distance of SV. F B0 represents the distance traveled by FV, and B0 represents the relative distance between SV and FV after they enter the adjacent lane.

[0085] Based on the safe distance model including headway (specifically, the headway model, i.e., the headway distribution for different drivers), B0 should meet the following conditions:

[0086] The safe distance DSS is thus obtained as follows: (1) Where: t represents the duration of the lane change; SV is the speed of the main vehicle at the starting point of the lane change. The speed of FV when SV begins to change lanes; The relative speed between the two vehicles (SV and FV); The acceleration of SV; Let FV be the acceleration; D W SV width (1.8m); θ SV is the angle between the front of the vehicle and the boundary line of the adjacent lane; T is the time interval value used to ensure a low false alarm rate, and is the value used when the absolute value of relative speed is ≤15km / h (0.6s); when the absolute value of relative speed is >15km / h, the value of TTC_min is selected.

[0087] Considering that most drivers do not have obvious acceleration or deceleration operations during lane changes, it is assumed that SV and FV maintain a constant speed during lane changes, and the acceleration value is set to 0; the speed of SV during lane changes is divided into four intervals: ≤70 km / h, 70 km / h - 90 km / h, 90 km / h - 110 km / h, and ≥110 km / h; under relatively stable acceleration conditions, based on the average lane change time of each speed interval, formula (1) is simplified to obtain:

[0088] In the selected specific embodiments, through experimental data analysis and calculation, the average duration of lane changes within each speed range was 5.3s, 5.1s, 4.9s, and 4.7s, respectively. The average speeds during lane changes within each speed range were 60 km / h, 79 km / h, 99 km / h, and 116 km / h, respectively. θ was taken as 1° from the reference, and a simplified safety distance was used, combined with TTC (Traffic Time Control). When the absolute relative speed was >15 km / h, TTC_min was used as the safety indicator; when the absolute relative speed was ≤15 km / h, the headway was used as the safety indicator. Finally, a lane change warning model was established for different relative speed ranges.

[0089] DWS stands for Basic Safety Distance Threshold.

[0090] The above safety distance model is applied to curved tunnel scenarios, with the addition of correction formulas for radius and slope angle.

[0091] The curve radius compensation includes fitting the real-time curve radius using a high-precision map or LiDAR point cloud; the curve radius compensation formula for dynamically adjusting the safety distance is:

[0092] in, This represents the safety distance correction based on the curve radius; k is an empirical coefficient (taken as 0.15) to ensure that the safety distance compensation matches the actual lateral offset. φ This is the ratio of lateral acceleration to gravitational acceleration. R represents the dynamic influence of lateral forces on vehicles in a curve; R represents the real-time curve radius, which is the current dynamic curve radius fitted by point cloud. R is a dynamic value that is sensed and calculated in real time by LiDAR sensors, and can reflect the actual curvature of the curve where the vehicle is currently located, describing the geometric parameters of the degree of curvature of the tunnel.

[0093]

[0094] Indicates the three-dimensional safety distance threshold. This represents the basic safety distance threshold.

[0095] Specifically, in S4 above, the slope angle correction formula for dynamically adjusting the safety distance by integrating IMU and GPS data is as follows:

[0096] in, This represents the safety distance correction based on the slope angle; μThe coefficient of friction of the road surface is taken as 0.7; the slope θ > 0 when going uphill and θ < 0 when going downhill.

[0097] By overlaying the curve radius compensation and slope correction results, a three-dimensional safety distance model is generated, achieving dynamic parameter fusion:

[0098] The data source for constructing the three-dimensional safety distance model mainly consists of the static curve radius R of curved tunnels pre-stored in high-precision maps. j ; and using LiDAR for real-time scanning, fitting the current dynamic curve radius through point cloud, and combining Kalman filtering to optimize noise interference, such as Figure 3 As shown. Represents the three-dimensional safety distance threshold, in Replace the basic safe distance threshold (DWS) in the S4 warning rules, predict the warning level, and apply it to special scenarios such as curved tunnels to optimize warnings. This enables the application of lane-changing models based on different speeds in curved tunnels.

[0099] The steps for optimizing noise interference using Kalman filtering are as follows: Define the state variables:

[0100] R k This indicates the radius of the curve at the current moment. This represents the rate of change of the curve radius (assuming uniform change), and k represents the time step number. It is used to discretize the change of the curve radius over a continuous time period. For example, k=0 represents the initial time, and k=1 represents the next sampling time. Given the true state vector at time k, determine the state transition model:

[0101] Where: F is the state transition matrix:

[0102] The time step (refers to the filtering period, which is synchronized with the sensor data refresh rate. That is, if the sensor collects data once per second, then the time step is 1 second). For process noise, express k Prediction of prior states at time t. This represents the posterior state estimate at time k-1.

[0103] Observation model:

[0104] HThis represents the observation matrix, whose purpose is to establish state variables (such as the curve radius R). k ) and sensor measurements (such as the radius of a LiDAR point cloud fit) The linear relationship between the two. The direct observation radius represents the sensor measurement value. Directly corresponds to R in the state variable k H defines how to extract information from the state vector. Extract the physical quantities that the sensors can directly measure, and establish the mapping relationship between the internal state of the system and the measurements of the external sensors; For measuring noise.

[0105] Prediction phase: Predicted status:

[0106] Represents the state vector, used for matrix calculations of multidimensional state vectors. This represents the optimal state estimate of multiple state components at the previous time step (k-1). F This represents the state transition matrix (describing the uniform change of the curve radius).

[0107] Prediction error covariance:

[0108] This represents the updated prediction error covariance; This represents the post-test error covariance of the previous time step; This represents the transpose of the state transition matrix; This represents the process noise covariance, such as a sudden change in a curve; its overall physical meaning is the covariance of the error from the previous moment. Through the state transition matrix F Mapped to the current time step. The prediction covariance error is calculated by overlaying process noise.

[0109] The state transition matrix F is derived based on the physical and dynamic characteristics of the curve radius change. The state transition matrix F is the key bridge connecting the system's physical characteristics and the estimation algorithm; it is obtained through the following steps: Step 1: Continuous-time dynamics modeling: Establish the continuous-time differential equations of the system state:

[0110] Where x is the state vector; A is the continuous-time state matrix; and t is the continuous-time variable.

[0111] Step 2: Discretization: Using the zero-order preserved discretization method, the continuous-time system is converted into a discrete-time system, and the final state transition matrix is ​​obtained:

[0112] Update phase: Calculate the Kalman gain:

[0113] This represents the weighting coefficient, which balances the degree of confidence between predicted and observed values. This represents the transpose of the observation matrix; This represents the observation noise covariance matrix.

[0114] Updated state estimate:

[0115] This represents the updated state estimate; This represents the weighting coefficient, which balances the degree of confidence between the predicted and observed values. State vector A single component in the system (a single variable is each independent component or physical quantity in the system state vector, representing radius, velocity, acceleration, etc.).

[0116] Update error covariance:

[0117] This represents the updated error covariance; Represents the identity matrix.

[0118] Determine the noise covariance matrix: Process noise covariance: set based on the physical characteristics of curve radius variation.

[0119] The variance of the random disturbance of the curve radius (such as fluctuations in radius measurement due to uneven road surface, unit: m²) is determined by calculating the error variance based on the known true curve radius and the radius fitted by the sensor, after calibration based on sensor performance.

[0120] The random disturbance variance of the curve radius refers to the random fluctuation variance of the curve radius measurement value caused by factors such as uneven road surface and sensor noise. It is used to quantify the measurement uncertainty.

[0121] Calculation method: Step 1: Calculate the error sequence between the obtained true radius and the curve radius obtained from N independent tests, i.e. , where e i e represents the error of the i-th measurement. 测 and e 真These represent the radius measured by the sensor and the true radius, respectively.

[0122] Step 2: Calculate the mean of the error sequence, i.e. In ideal circumstances, u e If u approaches zero e A value that is significantly non-zero indicates the presence of systematic measurement bias.

[0123] Step 3: Calculate the error variance, i.e. .

[0124] Measurement covariance matrix: set according to LiDAR accuracy:

[0125] This represents the variance of the radius measured by LiDAR compared to the known true radius. The method for determining this variance is to calibrate the radius using LiDAR and then calculate the variance between the LiDAR and the known true radius. This represents the variance of the radius estimate.

[0126] In this process, it is necessary to limit the Kalman filter update frequency to improve the calculation speed.

[0127] Example 2

[0128] Assuming the vehicle enters a curved tunnel, the initial LiDAR fitting radius R... init =200 m, Kalman filter history estimate is R k−1 =195 m, r k−1 =0.1 m / s, time step Δt=0.1 s.

[0129] R k−1 and r k−1 These represent the curve radius in the previous time step and the rate of change of the radius in the previous time step, respectively. predict: R k∣k−1 =195 + 0.1 × 0.1 = 195.01 m; R k∣k−1 This represents the curve radius value predicted based on the previous time step (k-1).

[0130] Observation Update: If the LiDAR measurement value z k =198 m. After Kalman gain calculation, the final estimated value converged to 196.5 m (the corrected value was obtained by calculating the observation residuals and rounding them through radius correction and rate of change correction), which is significantly better than the original fitting result (initial fitting radius Rinit=200 m).

[0131] Tiered real-time early warning: When the actual distance between the main vehicle and the following vehicle in the adjacent lane (FV) is less than the threshold of the three-dimensional safe distance model, a graded warning signal is triggered, including visual cues, audible warnings, and active braking intervention. The interactive unit used is an AR-HUD (Augmented Reality Head-Up Display): 10° field of view, dynamically adjustable brightness (300-1000 cd / m²); and an in-vehicle speaker: frequency response range 500Hz-5kHz, sound pressure level 70dB@1m.

[0132] The hardware architecture includes a redundant MCU (such as InfineonAurix TC397) deployed in the vehicle's ESP / EPS controller, a main control unit supporting ASIL-D level functional safety, and actuator interfaces, including a braking system that interacts with the ESP via the CAN-FD protocol to control the braking force gradient (0-100% adjustable, 0.5% resolution); an EPS interface using the SENT protocol for a steering system with a steering wheel angle limit range of ±30° and a resolution of 0.1°; LVDS video transmission (60Hz refresh rate), audio PWM modulation (48kHz sampling rate), and a HUD / speaker.

[0133] The real-time communication protocol includes time synchronization based on the PTP protocol (IEEE 1588), with an end-to-end latency of <50ms in the sensor-controller-actuator link.

[0134] Experimental Verification: The effects of existing technologies and experiments are compared and verified in terms of warning accuracy, warning time, error in three-dimensional scenes, and the accuracy of driving style classification using the above clustering algorithm, as well as the effect of existing lane change warning systems. By using existing survey data and simulation data, the false alarm rate in the same curved tunnel scenario is compared; the warning lead time in the high-speed range (>90km / h); and the illegal lane change suppression rate in the scenario of reckless drivers.

[0135] The objects to be tested were divided into three groups: Control group 1: Traditional fixed threshold model (TTC≤3s triggers warning, no speed / curve compensation); Control group 2: Dynamic model (only speed range is divided, without driving style classification and 3D compensation); Experimental group: The system of this invention embodiment (speed range division + three-dimensional compensation + driving style adaptation); The test results are shown in Table 1.

[0136] Table 1. Test results of the present invention compared with the control group.

[0137] The above test results demonstrate that the lane-changing model based on speed ranges of the present invention improves the accuracy of early warnings and the advance time of high-speed warnings. Furthermore, for lane-changing conditions with different driving styles, the speed range reduces the false alarm rate for reckless driving. Considering the three-dimensional compensation in curved tunnel scenarios, the warning error is reduced.

[0138] Based on the above experimental results, the comparison data between ordinary tunnel scenarios considering speed ranges and driving style classifications, but removing the three-dimensional compensation model, and curved tunnels is shown in Table 2.

[0139] Table 2 Inspection Results of Curved Tunnels and Ordinary Tunnels

[0140] The results show that the accuracy error of the warning is less than 1.5%, which meets the safety requirements. The false alarm rate is almost the same. After disabling the three-dimensional compensation, the computational load is reduced, which shortens the system latency. At the same time, the camera's accuracy in recognizing lane lines is also high. This indicates that the system can be used in ordinary tunnel scenarios, but it can play a more complete role in complex scenarios such as curved tunnels.

[0141] In summary, the technical solution of this invention proposes a personalized lane-change warning system and method for curved tunnels considering different speed conditions. On one hand, it dynamically models safe distances for different driving speed ranges based on ordinary roads and tunnels, and combines the Mini-Batch K-means algorithm to classify driving styles in real time, adjusting the lane-change safe distance model for different driving styles in real time, thereby achieving real-time warnings for lane-change behavior. On the other hand, by incorporating a curved tunnel scenario, it introduces curve radius compensation and slope correction into the safety distance model constructed for each speed range, building a three-dimensional safe distance model. Furthermore, it dynamically divides the parameters for graded warnings based on real-time predicted driving styles, achieving real-time warnings for lane-change behavior in the complex scenario of curved tunnels. This significantly improves the accuracy of lane-change warnings under traditional fixed thresholds and reduces warning errors for lane-change models in speed ranges within curved tunnel scenarios, thus achieving the goal of real-time lane-change warnings. Seamless switching between warning modes for ordinary tunnels and curved tunnels is required, overcoming the reliance on manual configuration in traditional methods.

[0142] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A personalized lane-changing early warning method for curved tunnels considering different speed conditions, characterized in that, Includes the following steps: S1. Multi-source data acquisition: Real-time acquisition of main vehicle speed, adjacent lane following vehicle speed, lateral offset, relative distance, acceleration and steering wheel angle data through on-board multiple sensors; S2. Based on the data obtained in S1, driving behavior features are extracted. The average headway and average minimum collision time are used as clustering features to classify drivers into three types: aggressive, steady, and conservative. S3. Establish a nonlinear basic safety distance model for different speed ranges of the main vehicle and different relative speed ranges of the two vehicles. Combine the formula for dynamically correcting the basic safety distance model with the slope angle and the curve radius to obtain a three-dimensional safety model. Generate early warning rules by combining the TTC threshold and the basic safety distance model, or by combining the TTC threshold and the three-dimensional safety model. S4. Tiered Real-Time Warning: Triggers tiered warnings based on warning rules adapted to the current situation. Warnings include visual cues, audible alerts, and active braking interventions based on driving style.

2. The personalized lane-changing early warning method for curved tunnels considering different speed conditions according to claim 1, characterized in that, In S2, the driver segmentation uses the Mini-Batch k-means clustering algorithm: Using average headway and average minimum collision time as data input, infeasible and low-risk samples were excluded, and the data was standardized using Z-score. Select k data points as initial cluster centers, where k=3, corresponding to aggressive, stable, and conservative clusters; to eliminate the correlation of the covariance matrix, calculate the Mahalanobis distance from the sample to each cluster center; Samples are assigned to the nearest cluster based on Mahalanobis distance, and the cluster centers are updated. The maximum number of iterations is set, and the iteration frequency is synchronized with the vehicle CAN bus refresh rate. If the continuous iteration loss decrease rate is less than 1%, the batch size is increased. The final cluster labels and centroid coordinates are output. The results of each cluster label are used as the standard for driving style classification, thereby classifying drivers into aggressive, steady, and conservative types.

3. The personalized lane-changing early warning method for curved tunnels considering different speed conditions according to claim 2, characterized in that, The Z-score normalization process is as follows: ； in, z i Indicates the first i Standardized values ​​for each sample For the first in the original data i Each sample value; The mean of the dataset; σ The standard deviation of the dataset; Mahalanobis distance The calculation formula is: ； in, Indicates sample With cluster center The transpose of the difference vector; This represents the inverse of the covariance matrix.

4. The personalized lane-changing early warning method for curved tunnels considering different speed conditions according to claim 2, characterized in that, The criteria for classifying driving styles in S2 are as follows: For aggressive drivers: MTH ≤ 1.36s, TTC_min ≤ 4.21s; For stable drivers: 1.36s < MTH ≤ 1.83s, 4.21s < TTC_min ≤ 7.62s; Conservative driver: MTH > 1.83s, TTC_min > 7.62s; Where MTH represents the average headway of multiple lane change events by the driver, and TTC_min represents the average minimum collision time of multiple lane change events by the driver.

5. The personalized lane-changing early warning method for curved tunnels considering different speed conditions according to claim 1, characterized in that, The method for constructing the basic safety distance model in S3 is as follows: Assuming the main vehicle and the following vehicle in the adjacent lane maintain a constant speed during lane change, and the speed of the main vehicle during lane change is divided into multiple intervals, the safety distance DSS calculation formula is simplified based on the average lane change time in each speed interval. The simplified safety distance DSS calculation formula is then combined with TTC. When the absolute value of the relative speed between the main vehicle and the following vehicle in the adjacent lane is >15 km / h, the time interval T is set to TTC_min; when the absolute value of the relative speed between the main vehicle and the following vehicle in the adjacent lane is ≤15 km / h, the time interval T is set to MTH. Here, MTH represents the average headway of multiple lane change events, and TTC_min represents the average minimum collision time of multiple lane change events. This leads to the establishment of a nonlinear basic safety distance model under different speed intervals and different relative speed intervals between the two vehicles.

6. The personalized lane-changing early warning method for curved tunnels considering different speed conditions according to claim 5, characterized in that, In S3, the basic safety distance model is: ； DWS stands for Basic Safety Distance Threshold.

7. The personalized lane-changing early warning method for curved tunnels considering different speed conditions according to claim 1, characterized in that, In S3, the method for determining the three-dimensional security model is as follows: ； in, This represents the three-dimensional safety distance threshold, while DWS represents the basic safety distance threshold obtained from the basic safety distance model. This represents the safety distance correction based on the curve radius. This represents the safety distance correction based on the slope angle; k is an empirical coefficient. φ This is the ratio of lateral acceleration to gravitational acceleration. This reflects the dynamic impact of lateral forces on a vehicle in a curve; R represents the real-time curve radius. μ The coefficient of friction of the road surface. The slope angle is denoted by .

8. The personalized lane-changing early warning method for curved tunnels considering different speed conditions according to claim 1, characterized in that, In S3, the warning rule is as follows: The triggering conditions for a Level 1 warning are: actual distance ∈ [80%·DWS, 100%·DWS] or TTC ∈ [5s, 8s); The triggering conditions for a Level 2 warning are: actual distance ∈ [60%·DWS, 80%·DWS) or TTC ∈ [3s, 5s); The triggering condition for a Level 3 warning is: actual distance < 60% DWS or TTC < 3s; When two safety indicators conflict, the one with the higher risk level will be selected for early warning.

9. The personalized lane-changing early warning method for curved tunnels considering different speed conditions according to claim 1, characterized in that, In S4, the tiered early warning system is as follows: Level 1 Warning: The HUD displays a yellow warning box, and the dashboard flashes a lane change risk icon; Level 2 warning: An audible warning is given with a frequency that increases linearly from 1.5kHz to 3kHz, with the frequency gradually increasing with the level of danger; Level 3 Warning: Intervene with active braking based on driving style. Active braking includes braking force control and steering wheel angle limitation. Braking force control applies 20%-30% of the basic braking force to aggressive drivers; 30%-40% of the basic braking force to moderate drivers; and 40%-50% of the basic braking force to conservative drivers. Steering wheel angle limitation restricts the maximum angle to -20° to 20° for aggressive drivers, -15° to 15° for moderate drivers, and -10° to 10° for conservative drivers. The dynamic delay compensation mechanism is as follows: when TTC < 2s, pre-filled hydraulic braking is activated after a fixed base oil pressure is established in advance; if the first-level warning is not lifted for a fixed time, it is automatically upgraded to the second-level warning.

10. A personalized lane-changing early warning system for curved tunnels considering different speed conditions, characterized in that, The method for personalized lane-changing early warning in curved tunnels considering different speed conditions, as described in claim 1, includes: The multi-source data acquisition module is used to collect data on the speed of the main vehicle, the speed of vehicles following in adjacent lanes of the target lane, lateral offset, curve radius, slope angle, relative distance, acceleration, and steering wheel angle through millimeter-wave radar, LiDAR, camera, IMU, and GPS; and synchronizes the time through the V2X module. The driving style classification module uses average headway and average minimum collision time as clustering features to classify drivers into aggressive, steady, and conservative types. The lane change warning model building module is used to establish a nonlinear basic safety distance model based on different speed ranges and different relative speeds between two vehicles. It generates warning rules by combining the TTC threshold and the basic safety distance model, or by combining the TTC threshold and the three-dimensional safety model, and sends the warning rules to the real-time warning module. The three-dimensional safety model calculation module is used to combine the formula of slope angle and curve radius to dynamically correct the basic safety distance model and obtain the three-dimensional safety model; The real-time warning module is used to trigger tiered warnings based on warning rules adapted to the current situation. Warnings include visual cues, audible alerts, and active braking interventions based on driving style.

Images