Collision avoidance control method for mixed intersection based on laser radar perception

Abstract

The application provides a mixed intersection vehicle-road cooperation anti-collision control method based on laser radar perception, and relates to the field of traffic control. Relying on radar point cloud data, point cloud attenuation, driver attention, automatic driving level and penetration rate parameters are introduced to build a power average nonlinear mixed visibility model; relying on the disturbance-resilience response surface, a three-dimensional evaluation system is established to calculate the visibility deficiency probability and the intersection resilience index; through SHAP analysis to screen key factors, a resilience and engineering cost double objective optimization model is built, and the NSGA-III algorithm is improved to solve the static optimal visibility; combined with short-term traffic disturbance prediction, the visibility threshold is dynamically adjusted, the graded anti-collision instruction is generated, and the vehicle is issued by the vehicle-road cooperation platform. The problems of model linearization, step fragmentation and ambiguity in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of traffic control, and in particular to a vehicle-road cooperative collision avoidance control method for mixed-traffic intersections based on lidar perception. Background Technology

[0002] Existing intersection line-of-sight design methods are mainly based on the characteristics of human driving behavior, without considering the point cloud attenuation characteristics of sensors such as LiDAR in rainy and foggy weather, nor distinguishing the dynamic line-of-sight requirements under different levels of autonomous driving and penetration rates. At the same time, traditional methods mostly treat line-of-sight as a static geometric design parameter, lacking the means to use it as a real-time control variable for collision avoidance decisions, making it difficult to cope with sudden disturbances such as adverse weather or temporary obstacles that reduce perception capabilities.

[0003] Therefore, a vehicle-road cooperative collision avoidance control method based on lidar perception is needed at mixed-traffic intersections to solve the above problems. Summary of the Invention

[0004] The purpose of this invention is to provide a vehicle-road cooperative collision avoidance control method for mixed-traffic intersections based on lidar perception, which solves the problems of model linearization, step fragmentation, and ambiguity in the prior art.

[0005] To achieve the above objectives, this invention provides a vehicle-road cooperative collision avoidance control method for mixed-traffic intersections based on lidar perception, comprising the following steps: S1: Obtain LiDAR point cloud data within the intersection area, and construct a nonlinear human-vehicle hybrid line-of-sight model based on power average by combining the point cloud attenuation coefficient, driver visual attention coefficient, autonomous driving level, and autonomous driving penetration rate. S2: Establish a three-dimensional quantitative evaluation system based on the disturbance-toughness response surface, use the hybrid sight distance model constructed in S1 to calculate the probability of insufficient sight distance, quantify the comprehensive toughness index of the intersection area under different disturbance intensities, and output toughness index data under different design parameters. S3: SHAP interpretability analysis was used to screen key influencing factors, a dual-objective optimization model of resilience and engineering cost was constructed, and the Pareto optimal static sight distance configuration was solved by improving the NSGA-III algorithm; S4: Based on short-term traffic disturbance prediction, and using the optimal static sight distance configuration obtained in S3 as a benchmark, the control sight distance threshold is dynamically adjusted to generate hierarchical collision avoidance control commands and send them to vehicles through the vehicle-road cooperative system. The vehicles adjust the target speed, safe following distance and recommended driving trajectory according to the hierarchical collision avoidance control commands to achieve closed-loop collision avoidance control.

[0006] Preferably, in S1, roadside lidar fixedly installed on intersection poles collects lidar point cloud data within the intersection area. After denoising, ground segmentation, and clustering operations, the vehicle's position and speed are extracted from the lidar point cloud data; the autonomous driving level L and autonomous driving penetration rate p are obtained from the roadside unit (RSU); and the standard lidar detection distance is determined. Attenuation constant k; Point cloud attenuation coefficient α corresponding to the current weather; Camera detection distance Vehicle-road cooperative communication distance Standard sight distance for drivers and driver visual attention coefficient β ; The nonlinear human-vehicle hybrid line-of-sight model based on power-means is as follows: ; In the formula, The power-mean order; For manually driven vehicles, the line of sight is required. Line of sight for autonomous vehicles; This indicates the overall perceptual reduction caused by the combined effects of point cloud decay and decreased attention; the output shows a mixed line of sight. .

[0007] Preferably, the comprehensive resilience index of the three-dimensional quantitative evaluation system in S2 for: ; In the formula, To enhance resistance to disturbances; To restore ability; This provides the ability to prevent cascading failures. , , These are the weighting coefficients; satisfying... + + =1; By changing the disturbance intensity and design parameters, the following can be obtained: Response surface as a function of disturbance intensity and design parameters; output comprehensive toughness index And data that varies with different design parameters.

[0008] Preferably, the design parameters include intersection geometric design parameters, traffic parameters, and environmental parameters; Intersection geometric design parameters include turning radius and obstacle distance; traffic parameters include autonomous driving penetration rate; environmental parameters include point cloud attenuation coefficient α and driver visual attention coefficient. β .

[0009] Preferably, in S3, the SHAP interpretability analysis method is used to calculate the interpretability of each feature pair. The marginal contribution; for each sample, the SHAP value satisfies: ; in As the baseline value, Let be the SHAP value of the m-th feature; for all samples The average value is taken to obtain the feature importance ranking, and key impact factors with a cumulative contribution of over 90% are selected as optimization variables. ; Construct a dual-objective optimization model for resilience and engineering cost: ; In the formula, Geometric and traffic constraints; engineering costs Including the cost of upgrading roadside lidar Cost of intersection sight distance modification Cost of vehicle-road cooperative communication facilities The specific expression is: ; In the formula, Positively correlated with lidar detection range Positively correlated with obstacle clearance distance Positively correlated with autonomous driving penetration rate p; An adaptive adjustment strategy is introduced, and an improved NSGA-III algorithm is used to solve the toughness-engineering cost dual-objective optimization model: ; ; In the formula, Indicates the adaptive crossover probability; Indicates the probability of mutation; This represents the current iteration number. This represents the maximum number of iterations. , These are the maximum and minimum crossover probabilities, respectively; , These are the maximum and minimum values ​​of the mutation probability, respectively; After obtaining the Pareto optimal solution set, the TOPSIS method is used to select the optimal compromise solution. The hybrid sight distance corresponding to the optimal compromise solution is the optimal static sight distance configuration, denoted as . .

[0010] Preferably, S4 is obtained from S3. Based on the mixed line-of-sight data updated in real time via S1 Calculate the dynamic control line of sight: ; In the formula, The base scaling factor is set according to the disturbance level; This is the sensitivity coefficient. Indicates taking a positive value; according to It generates target speed, safe following distance, and recommended driving trajectory, and sends control commands to autonomous vehicles through the vehicle-road cooperative system.

[0011] Preferably, the disturbance prediction level includes three levels: slight, moderate, and severe; The formula for calculating the target velocity is: ; In the formula, Speed ​​limits at intersections; For comfortable deceleration, a value of 2.5 m / s² is used. 2 ; To maintain a safe following distance; The recommended driving trajectory uses a fifth-order polynomial curve. The polynomial coefficients are solved based on the vehicle's current position, speed, heading angle, as well as the target position, target speed, and target heading angle. The time interval between trajectory points is set to 0.1 seconds.

[0012] Therefore, the present invention adopts the above-mentioned vehicle-road cooperative anti-collision control method for mixed intersections based on lidar perception, which directly uses lidar perception data for intersection anti-collision decision-making, significantly improving the traffic resilience and safety of intersections under mixed traffic conditions, and is applicable to intersection controllers in intelligent connected environments. Attached Figure Description

[0013] Figure 1 This is a flowchart of the vehicle-road cooperative collision avoidance control method for mixed-traffic intersections based on lidar perception, as described in this invention. Figure 2 This is a flowchart of the human-vehicle mixed line-of-sight calculation model in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the calculation of the three-dimensional toughness comprehensive index for different design parameters in an embodiment of the present invention. Figure 4 This is a flowchart of the SHAP feature selection and bi-objective optimization solution for finding the optimal static line of sight in an embodiment of the present invention; Figure 5 This is a flowchart of the predictive hierarchical dynamic line-of-sight control and vehicle-road cooperative instruction generation in an embodiment of the present invention. Detailed Implementation

[0014] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0015] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0016] Example 1 like Figures 1-5 As shown, this invention provides a vehicle-road cooperative collision avoidance control method for mixed-traffic intersections based on lidar perception, comprising the following steps: S1: Obtain LiDAR point cloud data within the intersection area, and construct a nonlinear human-vehicle hybrid line-of-sight model based on power average by combining the point cloud attenuation coefficient, driver visual attention coefficient, autonomous driving level, and autonomous driving penetration rate. In S1, roadside lidar fixedly installed on intersection poles collects lidar point cloud data within the intersection area. After denoising, ground segmentation, and clustering operations, the vehicle's position and speed are extracted from the lidar point cloud data. The autonomous driving level L and autonomous driving penetration rate p are obtained from the roadside unit (RSU). The standard lidar detection range... Attenuation constant k; Point cloud attenuation coefficient α corresponding to the current weather; Camera detection distance Vehicle-road cooperative communication distance Standard sight distance for drivers and driver visual attention coefficient β ; The attenuation constant k is related to the model of the lidar; the point cloud attenuation coefficient α corresponding to the current weather is specifically set as follows: α>0 in rainy / foggy weather, α=0 in sunny weather. Point cloud intensity data of roadside lidar under different weather conditions are collected in a standard test field, the ratio of actual detection distance to nominal distance is calculated, and the exponential attenuation curve is fitted to obtain the α value; in this embodiment, the driver's standard line of sight is... The value is 100m, and the driver's visual attention coefficient is... β Related to the driver's age and fatigue level; The actual effective detection range of roadside lidar is calculated using the following formula: ; The nonlinear human-vehicle hybrid line-of-sight model based on power-means is as follows: ; In the formula, The power mean order; when It degenerates into a linear weighted average when... When it approaches the geometric mean; When the power average assigns a higher weight to the smaller value, it reflects the weakest link effect—the weaker of the two, autonomous driving line of sight and human line of sight, has a greater impact on the hybrid line of sight. For manually driven vehicles, the line of sight is required. Line of sight for autonomous vehicles; This indicates the overall perceptual reduction caused by the combined effects of point cloud decay and decreased attention; the larger the product of the two, the more severe the reduction. The output is a mixed line-of-sight distance. ; The line-of-sight distance for autonomous vehicles is calculated using the following formula, which takes the minimum value across all perception methods: ; The following formula is used to calculate the sight distance of a manually driven vehicle: ; S2: Establish a three-dimensional quantitative evaluation system based on the perturbation-resilience response surface, and calculate the probability of insufficient line of sight using the hybrid line-of-sight model constructed in S1: ; in, To ensure the safe line of sight required based on the design speed, This is the sensitivity parameter; the Sigmoid function smoothly maps the line-of-sight gap to a probability between 0 and 1: when hour ;when hour .

[0017] Quantify the comprehensive toughness index of the intersection area under different disturbance intensities, and output the toughness index data under different design parameters; The comprehensive resilience index of the three-dimensional quantitative evaluation system in S2 for: ; In the formula, To enhance resistance to disturbances; To restore ability; This provides the ability to prevent cascading failures. , , These are the weighting coefficients; satisfying... + + =1; By changing the disturbance intensity and design parameters, the following can be obtained: Response surface as a function of disturbance intensity and design parameters; output comprehensive toughness index And data that varies with different design parameters.

[0018] Using the probability of insufficient line of sight as a reduction factor, the corrected disturbance resistance capability is obtained. : ; recovery ability Defined as: ; In the formula, This refers to the actual recovery time. In this embodiment, the maximum allowable recovery time is set to 60 minutes. Cascade failure prevention capability Defined as: ; in, This refers to the number of road segments associated with the intersection. For road section i Traffic capacity; This refers to the loss of traffic capacity caused by cascading failure; For road section i Initial passage capacity; Design parameters include intersection geometric design parameters, traffic parameters, and environmental parameters; Intersection geometric design parameters include turning radius and obstacle distance; traffic parameters include autonomous driving penetration rate; environmental parameters include point cloud attenuation coefficient α and driver visual attention coefficient. β By iterating through the range of these parameters, the comprehensive toughness index for each parameter combination is calculated. This allows us to obtain a three-dimensional response surface that represents the combined index of disturbance strength, design parameters, and toughness.

[0019] Specifically, in this embodiment, the design parameters vary as follows: turning radius: 10m~30m (interval of 2m); obstacle distance: 5m~20m (interval of 1m); autonomous driving penetration rate p: 0~0.5 (interval of 0.05); point cloud attenuation coefficient α: 0~1.0 (interval of 0.1, corresponding to sunny to rainy days); driver visual attention coefficient. β : 0.5~1.0 (interval 0.05 corresponds to a range of attention from severely distracted to highly focused).

[0020] Comprehensive toughness index under different disturbance intensities and the above five types of design parameters Dataset, for each set of design parameters The corresponding comprehensive toughness index is calculated. This forms a sample set. ,in For the first j Group design parameter vector, This corresponds to the resilience output.

[0021] S3: SHAP interpretability analysis was used to screen key influencing factors, a dual-objective optimization model of resilience and engineering cost was constructed, and the Pareto optimal static sight distance configuration was solved by improving the NSGA-III algorithm; S3 employs the SHAP interpretability analysis method to calculate the interpretability of each feature pair. The marginal contribution; for each sample, the SHAP value satisfies: ; in As the baseline value, Let be the SHAP value of the m-th feature; for all samples The average value is taken to obtain the feature importance ranking, and key impact factors with a cumulative contribution of over 90% are selected as optimization variables. ; Construct a dual-objective optimization model for resilience and engineering cost: ; In the formula, With design parameters The mapping relationship between them is based on multiple sets of data already calculated in S2. The data is fitted using a neural network surrogate model. The structure of the neural network surrogate model is as follows: the number of input nodes equals the number of key factors after screening, 1 output node, 2 hidden layers with 32 neurons each, ReLU activation function, and 200 training rounds. Geometric and traffic constraints; engineering costs Including the cost of upgrading roadside lidar Cost of intersection sight distance modification Cost of vehicle-road cooperative communication facilities The specific expression is: ; In the formula, Positively correlated with lidar detection range Positively correlated with obstacle clearance distance Positively correlated with autonomous driving penetration rate p; An adaptive adjustment strategy is introduced, and an improved NSGA-III algorithm is used to solve the toughness-engineering cost dual-objective optimization model: ; ; In the formula, Indicates the adaptive crossover probability; Indicates the probability of mutation; This represents the current iteration number. This represents the maximum number of iterations. , These are the maximum and minimum crossover probabilities, respectively; , These are the maximum and minimum values ​​of the mutation probability, respectively; After running the NSGA-III algorithm, a set of Pareto optimal solutions is obtained. ,in This represents the number of solutions on the Pareto front. Each solution corresponds to a pair of objective values: resilience index. and engineering costs After obtaining the Pareto optimal solution set, the TOPSIS method is used to select the optimal compromise solution. The hybrid sight distance corresponding to the optimal compromise solution is the optimal static sight distance configuration, denoted as . .

[0022] S4: Based on short-term traffic disturbance prediction, and using the optimal static sight distance configuration obtained in S3 as a benchmark, the control sight distance threshold is dynamically adjusted to generate tiered collision avoidance control commands. These commands are then sent to vehicles via the vehicle-to-infrastructure (V2I) system. Vehicles adjust their target speed, safe following distance, and recommended driving trajectory according to these commands, achieving closed-loop collision avoidance control. S4 uses the optimal static sight distance configuration obtained in S3 as a benchmark. Based on the mixed line-of-sight data updated in real time via S1 Calculate the dynamic control line of sight: In the formula, The base scaling factor is set according to the disturbance level; This is the sensitivity coefficient. Indicates taking a positive value; according to It generates target speed, safe following distance, and recommended driving trajectory, and sends control commands to autonomous vehicles through the vehicle-road cooperative system.

[0023] This study uses an LSTM neural network to predict short-term traffic disturbances within the next 5-15 minutes, such as increased rainfall, sudden congestion, and accidents. The input to the LSTM is historical time series data, including traffic flow, speed, and weather data from the past 30 minutes. The output is a disturbance prediction level for the next 10 minutes, categorized as minor, moderate, and severe. Slight perturbation: ; Typical disturbances: ; Severe disturbance: ; The formula for calculating the target velocity is: ; In the formula, Speed ​​limits at intersections; For comfortable deceleration, a value of 2.5 m / s² is used. 2 ; To maintain a safe following distance, commands can be issued directly. It is recommended to use a fifth-order polynomial curve for the driving trajectory. The polynomial coefficients are calculated based on the vehicle's current position, speed, heading angle, as well as the target position, target speed, and target heading angle. The time interval between trajectory points is set to 0.1 seconds.

[0024] The aforementioned instructions are broadcast to the Onboard Unit (OBU) of the autonomous vehicle via the Roadside Unit (RSU). The vehicle adjusts its speed, following distance, and trajectory according to the instructions to achieve closed-loop collision avoidance control. Simultaneously, the Roadside Unit continuously updates the real-time data of S1, resulting in rolling optimization.

[0025] Example 2 This embodiment uses a cross-shaped intersection as an example, with a design speed of 50 km / h and the corresponding required sight distance. The weather was moderate rain, the autonomous driving level was L=3, and the penetration rate was p=0.3. All sensing devices were roadside lidar, fixedly installed on poles in the four cardinal directions of the intersection, at a height of 5 meters and tilted downwards at 15 degrees.

[0026] Step S1 Example: Acquire roadside lidar point cloud data and calculate mixed line-of-sight distance: Roadside LiDAR Point Cloud Acquisition and Processing: The roadside LiDAR (model RS-LiDAR-32) scans the intersection area at a frequency of 10Hz to generate a 3D point cloud. Outliers are removed using a filtering algorithm, and the Random Sample Consensus (RANSAC) algorithm is used for ground segmentation. The remaining point cloud is then clustered using the DBSCAN algorithm to identify vehicle targets, extracting the position, speed, and heading angle of each vehicle. Simultaneously, the current weather is recorded as moderate rain, with a point cloud attenuation coefficient α = 0.6, obtained through rainy day calibration experiments. Parameter acquisition: Standard detection range of roadside lidar The attenuation constant k=0.5, according to the lidar datasheet. Calculate the actual effective detection range: ; Roadside camera detection range Vehicle-road cooperative communication distance Then the equivalent perception line of sight for autonomous driving takes the minimum value: ; Standard sight distance for drivers The attention coefficient β is monitored in real time by the driver's eye tracker. In this embodiment... β =0.8, indicating that the driver is older and has a less focused field of vision; Then manual driving visibility ; Nonlinear hybrid line-of-sight calculation: taking the power-mean order .

[0027] First calculate : ; calculate : ; The weighted sum of powers is: ; open Power (i.e., take 1 / 0.7 ≈ 1.4286): ; Multiply by the exponential decay term: ; Final mixed line of sight: ; S1 output: , , α=0.6, β=0.8. These data are saved to the roadside unit's memory database for use in step S2.

[0028] Step S2: Calculate the probability of insufficient sight distance; Take sensitivity parameters Required sight distance .

[0029] ; because Extremely small, approximately 0, therefore: ; This indicates that the current visibility is severely insufficient, and a visibility-deficient event is almost inevitable.

[0030] Traffic capacity data: An intersection model was built using the traffic simulation software SUMO, and traffic flow under moderate rain conditions was input. The simulation yielded the intersection's traffic capacity under normal sunny conditions. Minimum traffic capacity during moderate rain disturbance .but: ; This indicates that the resistance has dropped to zero because the probability of insufficient visibility is 100%.

[0031] Recovery ability: After moderate rain lasted for 30 minutes, it turned into light rain; actual recovery time The maximum allowable recovery time from when the rain stops until traffic capacity is restored to 90%. .

[0032] Cascading failure blocking capability: Number of associated road segments at the intersection n=4, traffic capacity of each road segment Initial passage capacity Cascaded loss .

[0033] ; First, calculate the molecules: ; The total number of numerators is 100,000 + 180,000 + 40,000 + 300,000 = 620,000.

[0034] Calculate the denominator again: ; The sum of the denominators = 1 + 1.44 + 0.64 + 2.25 = 5.33 × 10^6.

[0035] ; Resilience Comprehensive Index: Weights are determined using the Analytic Hierarchy Process (AHP), and a judgment matrix is ​​constructed. ; Calculate the eigenvector corresponding to the largest eigenvalue and normalize it to obtain... This embodiment (Empirical values ​​can also be used).

[0036] ; S2 Output: Overall Resilience Index under Current Moderate Rain Conditions The value is significantly lower than the approximately 0.85 under clear weather conditions, indicating a need to optimize the line-of-sight design. Simultaneously, step S2 outputs multiple sets of resilience index data under different design parameters (turning radius, obstacle distance, permeability, point cloud attenuation coefficient, and attention coefficient), forming a sample set. This is for use in step S3.

[0037] Step S3: Complete SHAP filtering and bi-objective optimization; Using the sample data calculated in step S2, and by changing the turning radius, obstacle distance, permeability, point cloud attenuation coefficient, and attention coefficient in step S2, the comprehensive toughness index under different working conditions has been calculated accordingly in step S2. This embodiment collected 200 different combinations of design parameters and their corresponding toughness index results. Each sample includes: turning radius. (10~30m) Obstacle Distance (5~20m), Autonomous Driving Penetration Rate (0~0.5), point cloud attenuation coefficient (0~1.0) Visual attention coefficient (0.5~1.0).

[0038] SHAP Analysis: Calculating each feature pair using Python's SHAP library. The contribution. Specific steps: Train an XGBoost regression model ( , =100) is used as a surrogate model, and then shap.Explainer is called to calculate the SHAP value for each sample. The average of |SHAP| for all samples is taken to obtain the feature importance ranking: Turning radius: Average SHAP value 0.12; Obstacle distance: 0.09; Autonomous driving penetration rate p: 0.07; Point cloud attenuation coefficient α: 0.05; Visual attention coefficient β : 0.01; The features that contribute more than 90% cumulatively are: turning radius (contribution percentage 0.12 / 0.36 = 33%), obstacle distance (25%), penetration rate (19%), and point cloud attenuation coefficient (14%), with the first four contributing 91% cumulatively. For simplicity, this embodiment uses the first three as optimization variables. .

[0039] Establish an optimization model: Objective 1: Maximize The mapping relationship of S2 was fitted using a neural network surrogate model (3 inputs, 1 output, 2 hidden layers with 32 neurons each, ReLU activation, trained for 200 rounds).

[0040] Objective 2: Minimize project costs .

[0041] The cost function is defined as: ; The larger the turning radius, the lower the cost (reduced demolition); the greater the distance to obstacles, the higher the cost (requiring the relocation of obstacles or widening of the field of vision); and the higher the visual attention coefficient penetration rate p, the higher the cost (investment in vehicle-road cooperative facilities).

[0042] Add constraints: The turning radius and lane width must meet minimum safety requirements. (The design speed of 50km / h corresponds to a minimum turning radius of 18m), which is met in this embodiment.

[0043] Improved NSGA-III solution: Set population size to 100, maximum number of iterations Crossover probability range Mutation probability range Adaptive crossover / mutation probability: ; The algorithm yielded the Pareto front (30 non-dominated solutions). The TOPSIS method was then employed to... The bigger the better. Choose the optimal solution based on the criterion that smaller is better: , , ; Corresponding mixed line of sight Resilience index ,cost Ten thousand yuan.

[0044] S3 Output: Optimal Static View Distance Write it into the static configuration table of the roadside unit.

[0045] Step S4 Example: Dynamic control and command issuance; Perturbation prediction: An LSTM model with 30 time steps × 5 features in the input layer, two LSTM layers each with 64 neurons, and Dropout=0.2 is used. The output layer uses softmax to output probabilities for three perturbation levels. Inputs include traffic flow, average speed, rainfall intensity, visibility, and point cloud attenuation coefficient α over the past 30 minutes. The prediction for the next 10 minutes is: rainfall intensity increasing from moderate to heavy rain. The prediction model outputs a "moderate perturbation" probability of 0.85, classifying it as a "moderate perturbation," with a base scaling factor. .

[0046] Real-time hybrid sight distance update: S1 calculates the hybrid sight distance in real time. Under heavy rain conditions, α=0.8, and the attention coefficient β=0.8 (driver attention decreases due to windshield wiper interference). Other parameters are the same as in step S1. ;; ; ; Power average: ; ; Weighted sum = 0.3 × 25.2 + 0.7 × 21.48 = 7.56 + 15.036 = 22.596; Root of 1.4286: ; Multiply by the exponential decay term: ; ; visibility gap ; Dynamic control line-of-sight calculation: taking the sensitivity coefficient .

[0047] ; Generate control commands including target speed, safe following distance, and recommended driving trajectory: Target speed: take deceleration Safe following distance , ; Since 14.36 m / s = 51.7 km / h, which exceeds the intersection speed limit of 50 km / h, therefore we take... However, for safety redundancy, a yellow alert is used, and a speed limit of 40 km / h is reduced by 20%.

[0048] Safe following distance: based on the two-second rule. Rounded down to 25m.

[0049] Recommended driving trajectory: Assume the vehicle's current position is (0,0), speed is 10m / s, and heading is 0 degrees; the target exit position is (80m,0), target speed is 11.1m / s (corresponding to 40km / h), and target heading is 0 degrees. A fifth-order polynomial is used. ; ; Boundary conditions: at t=0, x=0, v=10, a=0; at t=T (4 seconds), x=80, v=11.1, a=0. After obtaining the coefficients, a trajectory point is generated every 0.1 seconds.

[0050] Vehicle-to-everything (V2X) communication: The Roadside Unit (RSU) encapsulates the instructions into MAP and SPAT formats in the SAE J2735 standard message and broadcasts them via DSRC (Dedicated Short Range Communication) in the 5.9 GHz band, covering a radius of 300 meters. The autonomous vehicle receives the instructions through its Onboard Unit (OBU), parses out the speed limit of 40 km / h, the safe following distance of 25 m, and the list of trajectory points, and executes the instructions through the longitudinal controller (PID) and the lateral controller (pure tracking), successfully avoiding potential side collisions due to insufficient line of sight.

[0051] Therefore, the present invention adopts the above-mentioned vehicle-road cooperative collision avoidance control method for mixed traffic intersections based on lidar perception, and directly uses lidar perception data for intersection collision avoidance decision-making, which significantly improves the traffic resilience and safety of intersections under mixed traffic.

[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A vehicle-road cooperative collision avoidance control method for mixed-traffic intersections based on lidar perception, characterized in that, Includes the following steps: S1: Obtain LiDAR point cloud data within the intersection area, and construct a nonlinear human-vehicle hybrid line-of-sight model based on power average by combining the point cloud attenuation coefficient, driver visual attention coefficient, autonomous driving level, and autonomous driving penetration rate. S2: Establish a three-dimensional quantitative evaluation system based on the disturbance-toughness response surface, use the hybrid sight distance model constructed in S1 to calculate the probability of insufficient sight distance, quantify the comprehensive toughness index of the intersection area under different disturbance intensities, and output toughness index data under different design parameters. S3: SHAP interpretability analysis was used to screen key influencing factors, a dual-objective optimization model of resilience and engineering cost was constructed, and the Pareto optimal static sight distance configuration was solved by improving the NSGA-III algorithm; S4: Based on short-term traffic disturbance prediction, and using the optimal static sight distance configuration obtained in S3 as a benchmark, the control sight distance threshold is dynamically adjusted to generate hierarchical collision avoidance control commands and send them to vehicles through the vehicle-road cooperative system. The vehicles adjust the target speed, safe following distance and recommended driving trajectory according to the hierarchical collision avoidance control commands to achieve closed-loop collision avoidance control.

2. The vehicle-road cooperative collision avoidance control method for mixed-traffic intersections based on lidar perception according to claim 1, characterized in that, In S1, roadside lidar fixedly installed on intersection poles collects lidar point cloud data within the intersection area. After denoising, ground segmentation, and clustering operations, the vehicle's position and speed are extracted from the lidar point cloud data. The autonomous driving level L and autonomous driving penetration rate p are obtained from the roadside unit (RSU). The standard lidar detection range... Attenuation constant k; Point cloud attenuation coefficient α corresponding to the current weather; Camera detection distance Vehicle-road cooperative communication distance Standard sight distance for drivers and driver visual attention coefficient β ; The nonlinear human-vehicle hybrid line-of-sight model based on power-means is as follows: ； In the formula, The power-mean order; For manually driven vehicles, the line of sight is required. Line of sight for autonomous vehicles; This indicates the overall perceptual reduction caused by the combined effects of point cloud decay and decreased attention; the output shows a mixed line of sight. .

3. The vehicle-road cooperative collision avoidance control method for mixed-traffic intersections based on lidar perception according to claim 1, characterized in that, The comprehensive resilience index of the three-dimensional quantitative evaluation system in S2 for: ； In the formula, To enhance resistance to disturbances; To restore ability; This provides the ability to prevent cascading failures. , , These are the weighting coefficients; satisfying... + + =1; By changing the disturbance intensity and design parameters, the following can be obtained: Response surface as a function of disturbance intensity and design parameters; output comprehensive toughness index And data that varies with different design parameters.

4. The vehicle-road cooperative collision avoidance control method for mixed-traffic intersections based on lidar perception according to claim 3, characterized in that, Design parameters include intersection geometric design parameters, traffic parameters, and environmental parameters; Intersection geometric design parameters include turning radius and obstacle distance; traffic parameters include autonomous driving penetration rate; environmental parameters include point cloud attenuation coefficient α and driver visual attention coefficient. β .

5. The vehicle-road cooperative collision avoidance control method for mixed-traffic intersections based on lidar perception according to claim 1, characterized in that, S3 employs the SHAP interpretability analysis method to calculate the interpretability of each feature pair. The marginal contribution; for each sample, the SHAP value satisfies: ； in As the baseline value, Let be the SHAP value of the m-th feature; for all samples The average value is taken to obtain the feature importance ranking, and key impact factors with a cumulative contribution of over 90% are selected as optimization variables. ; Construct a dual-objective optimization model for resilience and engineering cost: ； In the formula, Geometric and traffic constraints; engineering costs Including the cost of upgrading roadside lidar Cost of intersection sight distance modification Cost of vehicle-road cooperative communication facilities The specific expression is: ； In the formula, Positively correlated with lidar detection range Positively correlated with obstacle clearance distance Positively correlated with autonomous driving penetration rate p; An adaptive adjustment strategy is introduced, and an improved NSGA-III algorithm is used to solve the toughness-engineering cost dual-objective optimization model: ； ； In the formula, Indicates the adaptive crossover probability; Indicates the probability of mutation; This represents the current iteration number. This represents the maximum number of iterations. , These are the maximum and minimum crossover probabilities, respectively; , These are the maximum and minimum values ​​of the mutation probability, respectively; After obtaining the Pareto optimal solution set, the TOPSIS method is used to select the optimal compromise solution. The hybrid sight distance corresponding to the optimal compromise solution is the optimal static sight distance configuration, denoted as . .

6. The vehicle-road cooperative collision avoidance control method for mixed-traffic intersections based on lidar perception according to claim 5, characterized in that, S4 is obtained from S3 Based on the mixed line-of-sight data updated in real time via S1 Calculate the dynamic control line of sight: ； In the formula, The base scaling factor is set according to the disturbance level; This is the sensitivity coefficient. Indicates taking a positive value; according to It generates target speed, safe following distance, and recommended driving trajectory, and sends control commands to autonomous vehicles through the vehicle-road cooperative system.

7. The vehicle-road cooperative collision avoidance control method for mixed-traffic intersections based on lidar perception according to claim 6, characterized in that, The disturbance prediction levels are divided into three levels: slight, moderate, and severe. The formula for calculating the target velocity is: ; In the formula, Speed ​​limits at intersections; For comfortable deceleration, a value of 2.5 m / s² is used. 2 ; To maintain a safe following distance; The recommended driving trajectory uses a fifth-order polynomial curve. The polynomial coefficients are solved based on the vehicle's current position, speed, heading angle, as well as the target position, target speed, and target heading angle. The time interval between trajectory points is set to 0.1 seconds.

Images