An intersection traffic conflict event identification method based on 4D millimeter wave radar

Abstract

The application relates to a kind of intersection traffic conflict event identification methods based on 4D millimeter wave radar, and is related to the technical field of road traffic safety, in particular to an intersection traffic conflict event identification method based on 4D millimeter wave radar.The purpose of the present application is to solve the problem that the existing method is not suitable for radar sparse point cloud, low-speed detection and trajectory fracture, and it is difficult to express the risk of multiple types of traffic conflicts by using a single index.The process is as follows:4D millimeter wave radar point cloud data is collected at a certain road intersection;continuous trajectory information of vehicles in the measured road intersection is obtained;candidate conflict vehicle pairs are screened;the conflict type of the candidate conflict vehicle pairs is determined;multiple risk indicators are selected according to the conflict type, and the comprehensive risk indicator and risk identification confidence are calculated by fusing multiple risk indicators;according to the comprehensive risk indicator value of each vehicle pair, the conflict severity is classified by combining the preset threshold range, and the conflict severity evaluation result is obtained.

Description

Technical Field

[0001] This invention relates to the field of road traffic safety technology, specifically to a method for identifying traffic conflict events at intersections based on 4D millimeter-wave radar, which belongs to the field of traffic engineering. Background Technology

[0002] Traffic accidents are characterized by low frequency, randomness, and severe consequences. Relying solely on historical accident data for safety assessments often suffers from long sample accumulation periods, incomplete accident records, and an inability to promptly reflect road operational risks. Traffic conflicts occur when road users are unusually close in time and space, and maintaining their current state of motion may result in a collision, or force one party to brake, stop, or swerve to avoid a collision. Because traffic conflicts occur more frequently than accidents, using traffic conflict events for proactive safety assessments has become an important means of road traffic risk monitoring.

[0003] Current traffic conflict identification methods typically utilize video, drone trajectories, or lidar. Video perception is susceptible to lighting conditions, rain, snow, occlusion, and perspective errors; lidar requires large amounts of data and is expensive; 4D millimeter-wave radar can simultaneously provide distance, azimuth, elevation, and radial velocity, and features all-weather, long-range, and direct speed measurement capabilities, making it suitable for continuously acquiring vehicle motion information at intersections and road segments. However, roadside radar point clouds suffer from problems such as long-range sparseness, static background reflection, adhesion of adjacent vehicle point clouds, and continuous missed detections when vehicles are slow or stationary. If traffic conflict indicators are calculated directly on the original detection points or broken trajectories, trajectory identity switching, interpolation repair, and speed fluctuations can be amplified into false sudden decelerations or false spatial proximity.

[0004] Therefore, a traffic conflict event identification method is needed that takes into account the characteristics of 4D millimeter-wave radar data. Based on the formation of stable continuous vehicle trajectories, the method should consider the quality of trajectory observation, integrate multiple risk indicators according to the conflict type, and identify and classify the complete traffic conflict process. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing methods in adapting to sparse radar point clouds, low-speed missed detections, and trajectory breaks, as well as the difficulty in fully expressing the risks of multiple types of traffic conflicts using a single index. This invention provides a method for identifying traffic conflict events at intersections based on 4D millimeter-wave radar.

[0006] The specific process of a method for identifying traffic conflict events at intersections based on 4D millimeter-wave radar is as follows:

[0007] Step 1: Collect 4D millimeter-wave radar point cloud data at the roadside of a certain intersection; the specific process is as follows:

[0008] A 4D millimeter-wave radar was installed on the side of a road at an intersection, and point cloud data of the 4D millimeter-wave radar was acquired based on the 4D millimeter-wave radar.

[0009] Based on 4D millimeter-wave radar point cloud data, the timestamp, radial distance, azimuth angle, elevation angle, radial velocity, and echo quality of the point cloud data are obtained.

[0010] The echo quality refers to echo power, signal-to-noise ratio, and noise.

[0011] Step 2: Process the point cloud data to obtain continuous trajectory information of vehicles at the intersection of the measured road segment;

[0012] Step 3: Filter candidate conflicting vehicle pairs based on continuous vehicle trajectory information;

[0013] Step 4: Determine the conflict type for candidate conflict vehicle pairs. Conflict types are divided into same-direction following conflict, cross-path conflict, merging or lateral intrusion conflict.

[0014] Step 5: Select multiple risk indicators based on the type of conflict, and combine the multiple risk indicators to calculate the comprehensive risk indicator and the risk identification confidence level;

[0015] Step 6: Based on the comprehensive risk index value of each vehicle and the preset threshold range, classify the severity of the conflict to obtain the conflict severity assessment result.

[0016] The beneficial effects of this invention are as follows:

[0017] This invention utilizes direct measurement data from 4D millimeter-wave radar to improve the ability to identify traffic conflict risks under adverse weather and complex lighting conditions;

[0018] This invention integrates vehicle movement direction, lane topology, and predicted trajectory spatial relationships to classify conflicts into same-direction following conflicts, intersecting path conflicts, and merging or lateral intrusion conflicts. It selects corresponding risk indicators for different conflict types and performs fusion calculations, which can reduce misjudgments caused by using the same discrimination standard for different conflict mechanisms and improve the accuracy and applicability of traffic conflict identification at complex intersections.

[0019] This invention distinguishes and marks the original trajectory points, repaired trajectory points, and low-quality associated points, and calculates the risk identification confidence level by combining radar observation quality, trajectory association quality, repair point ratio, and prediction uncertainty.

[0020] Therefore, this invention can not only output the degree of conflict risk, but also characterize the credibility of the risk result, reducing the impact of missing trajectories, incorrect associations, or trajectory repair on the reliability of risk identification results. Attached Figure Description

[0021] Figure 1This is a flowchart illustrating the overall process of the traffic conflict event identification method of the present invention.

[0022] Figure 2 This is a data type diagram of the 4D millimeter-wave radar used in this invention;

[0023] Figure 3 This is a schematic diagram of the effect of clustering to generate vehicle video frames according to the present invention;

[0024] Figure 4 In response to Figure 3 Clustering generates vehicle target image;

[0025] Figure 5 This is a schematic diagram illustrating the same-direction car-following conflict type discrimination of the present invention;

[0026] Figure 6 This is a schematic diagram illustrating the cross-path conflict type discrimination of the present invention;

[0027] Figure 7 This is a schematic diagram illustrating the discrimination between merging and lateral intrusion conflict types according to the present invention. Detailed Implementation

[0028] Specific Implementation Method 1: The specific process of this implementation method for identifying intersection traffic conflict events based on 4D millimeter-wave radar is as follows:

[0029] Step 1: Collect 4D millimeter-wave radar point cloud data at the roadside of a certain intersection; the specific process is as follows:

[0030] A 4D millimeter-wave radar was installed on the side of a road at an intersection, and point cloud data of the 4D millimeter-wave radar was acquired based on the 4D millimeter-wave radar.

[0031] Based on 4D millimeter-wave radar point cloud data, the timestamp, radial distance, azimuth angle, elevation angle, radial velocity, and echo quality of the point cloud data are obtained.

[0032] The echo quality refers to echo power, signal-to-noise ratio, and noise.

[0033] Step 2: Process the point cloud data to obtain continuous trajectory information of vehicles at the intersection of the measured road segment;

[0034] Step 3: Filter candidate conflicting vehicle pairs based on continuous vehicle trajectory information;

[0035] Initial screening was conducted based on spatial distance, temporal synchronization relationship, and lane topology relationship, with a focus on retaining vehicle pairs with potential interaction relationships;

[0036] Step 4: Determine the conflict type for candidate conflict vehicle pairs. Conflict types are divided into same-direction following conflict, cross-path conflict, merging or lateral intrusion conflict.

[0037] Step 5: Select multiple risk indicators based on the type of conflict, and combine the multiple risk indicators to calculate the comprehensive risk indicator and the risk identification confidence level;

[0038] Step 6: Based on the comprehensive risk index value of each vehicle and the preset threshold range, classify the severity of the conflict to obtain the conflict severity assessment result.

[0039] Specific Implementation Method Two: This implementation method differs from Specific Implementation Method One in that: in step two, the point cloud data is processed to obtain continuous trajectory information of vehicles at the intersection of the measured road segment;

[0040] The specific process is as follows:

[0041] Step 2: 1. Convert the coordinates of the measured point cloud data to the road coordinate system; 2. Remove invalid points from the converted point cloud data to obtain the point cloud data after removing invalid points.

[0042] Step 22: Cluster the radar points based on their spatial location, radial velocity, echo quality, and altitude;

[0043] Steps 2 and 3: Associate the trajectories of vehicles belonging to the same type to generate the vehicle target trajectory;

[0044] Step 24: Match the measured target trajectory of each vehicle with the actual lane to obtain the lane matching result.

[0045] The other steps and parameters are the same as in Specific Implementation Method 1.

[0046] Specific Implementation Method 3: This implementation method differs from Specific Implementation Method 1 or 2 in that: Step 21 involves converting the measured point cloud data coordinates to the road coordinate system; invalid points are removed from the converted coordinate point cloud to obtain point cloud data after invalid point removal.

[0047] The specific process is as follows:

[0048] Step 211: When using radar to perform roadside measurements, the measurement direction will have a certain angle with the road travel direction. In order to align the radar coordinates with the road direction, it is necessary to perform coordinate transformation on the original point cloud coordinates.

[0049] Transform the radial distance of the point cloud data measured in step one into the road coordinate system to obtain the radial distance of the point cloud data in the road coordinate system.

[0050] Transform the azimuth angle of the point cloud data measured in step one into the road coordinate system to obtain the azimuth angle of the point cloud data in the road coordinate system.

[0051] Transform the pitch angle of the point cloud data measured in step one into the road coordinate system to obtain the pitch angle of the point cloud data in the road coordinate system.

[0052] Transform the radial velocity of the point cloud data in the coordinate system measured in step one to the road coordinate system to obtain the radial velocity of the point cloud data in the road coordinate system.

[0053] The echo power of the point cloud data measured in step one is converted to the road coordinate system to obtain the echo power of the point cloud data in the road coordinate system.

[0054] Transform the signal-to-noise ratio of the point cloud data measured in step one into the road coordinate system to obtain the signal-to-noise ratio of the point cloud data in the road coordinate system.

[0055] The noise in the point cloud data coordinate system measured in step one is transformed to the road coordinate system to obtain the noise in the point cloud data in the road coordinate system.

[0056] The X-axis of the point cloud data coordinate system (the data measured by 4D millimeter-wave radar is itself in the point cloud data coordinate system) is the measurement direction, the Y-axis is perpendicular to the X-axis, and the Z-axis is perpendicular to the XY plane.

[0057] The road coordinate system is defined with the radar projection point on the road surface as the origin, the road travel direction as the X-axis, the perpendicular direction of the road travel direction as the Y-axis, and the vertically upward direction as the Z-axis.

[0058] Step 212: Remove invalid points from the point cloud data in the road coordinate system to obtain the point cloud data after removing invalid points;

[0059] The specific process is as follows:

[0060] Step 2121: Crop the region of interest for each frame of point cloud in the road coordinate system;

[0061] Step 2: Filter outliers in the radar point cloud data within the region of interest to obtain filtered radar point cloud data.

[0062] The specific process is as follows:

[0063] If the signal-to-noise ratio of the radar point cloud data in the region of interest is less than 8 dB, the radar point cloud is determined to be a low-quality point and is removed.

[0064] If the signal-to-noise ratio of the radar point cloud data in the region of interest is greater than or equal to 8 dB and less than 10 dB, the radar point cloud will be retained but marked as a low-quality point.

[0065] If the signal-to-noise ratio of the radar point cloud data in the region of interest is greater than or equal to 10 dB, the radar point cloud is marked as a normal point.

[0066] Step 2123: Perform background removal on the filtered radar point cloud data to obtain radar point cloud data with background removal.

[0067] The specific process is as follows:

[0068] 1) In this invention, the region of interest is divided into grids with a vertical dimension of 1m and a horizontal dimension of 0.5m;

[0069] 2) Statistical analysis of the occurrence of each grid cell throughout the entire time period, including the total number of radar points, the number of low-speed static points, the number of obviously moving points, and the time span of occurrence;

[0070] Among them, low-speed static points are defined as radar points with an absolute radial velocity of no more than 0.25 m / s;

[0071] A point of obvious motion is defined as a radar point with an absolute radial velocity of not less than 0.8 m / s;

[0072] 3) If a grid cell meets all of the following conditions, then the corresponding grid cell will be marked as a candidate background grid cell:

[0073] First, the time span of the grid appearance must be no less than 600 seconds;

[0074] Second, the number of low-speed static points within the grid shall not be less than 100;

[0075] Third, the proportion of low-speed static points within a grid to the total number of radar points within the corresponding grid shall not be less than 90%;

[0076] Fourth, the ratio of the number of obviously moving points to the number of low-speed static points within the grid should not exceed 3%;

[0077] Grids that meet the above conditions are usually characterized by long-term existence, fixed position, near-stationary speed, and lack of moving targets passing by, and are therefore more likely to correspond to static background reflections of roads.

[0078] 4) To avoid accidentally deleting queued vehicles and stopped vehicles, this paper further introduces lane protection zones;

[0079] Lane protection zones are automatically determined based on the distribution of obvious movement points in the lateral direction throughout the day, and are used to indicate the main lanes where vehicles frequently travel.

[0080] A lane protection zone is set up within 2.0 m on both sides of the center line of each lane;

[0081] Low-speed static points located within lane protection zones will not be directly deleted as background targets, even if the absolute value of their radar radial velocity is less than or equal to 0.25 m / s.

[0082] 5) During the background removal stage, a radar point is only identified as a background point and removed if it simultaneously meets the following conditions:

[0083] The radar point is located in a background candidate grid; the absolute value of the radial velocity of the radar point is no greater than 0.25 m / s; the radar point is not located within the lane protection zone.

[0084] In this way, this paper can effectively eliminate long-term fixed static background reflections, while preserving queuing vehicles, stopped vehicles, and low-speed vehicle targets as much as possible.

[0085] Steps 21-24: Perform ground clutter suppression on the radar point cloud data obtained in Steps 21-23 after background removal; the specific process is as follows:

[0086] Radar points with a height of less than 0.10 m relative to the road surface, an absolute radial velocity of less than 0.30 m / s, and fewer than two adjacent points within a radius of 0.50 m are identified as ground clutter and are removed.

[0087] Other steps and parameters are the same as in specific implementation method one or two.

[0088] Specific Implementation Method Four: This implementation method differs from Specific Implementation Methods One to Three in that: Step Two, clustering is performed based on the spatial location, radial velocity, echo quality, and height of the radar points;

[0089] The specific process is as follows:

[0090] Step 221: Calculate the joint feature distance between point clouds; the process is as follows:

[0091] Any two radar points and The joint feature distance is defined as:

[0092]

[0093] In the formula, This represents the joint characteristic distance between two radar points; , Representing radar points and The three-dimensional spatial position; , Representing radar points and radial velocity; , Representing radar points and Height; , , These represent the weights of spatial distance, radial velocity difference, and height difference, respectively.

[0094] Step 2: Based on the joint feature distance between point clouds, the DBSCAN neighborhood distance threshold, and the minimum number of cluster points, the DBSCAN clustering method is used to form vehicle target point clusters (one point cluster represents one vehicle).

[0095] For each point cluster, the centroid position, target size, average radial velocity, number of radar points, mean echo power, mean signal-to-noise ratio, mean noise, and dispersion are calculated to obtain the vehicle's observed target.

[0096] The other steps and parameters are the same as those in one of the specific implementation methods one to three.

[0097] Specific Implementation Method Five: This implementation method differs from Specific Implementation Methods One to Four in that: in steps two and three, the trajectories of vehicles belonging to the same type are associated to generate vehicle target trajectories;

[0098] The specific steps are as follows:

[0099] Steps 2 and 31: For the vehicle Establish motion state vector:

[0100]

[0101] In the formula, Indicates vehicle At any moment The state of motion; These represent vehicles in the X and Y directions of the road coordinate system, respectively. At any moment Positional components; These represent vehicles in the X and Y directions of the road coordinate system, respectively. At any moment velocity components; These represent vehicles in the X and Y directions of the road coordinate system, respectively. At any moment The acceleration components; Indicates vehicle At any moment The heading angle; Indicates vehicle At any moment angular velocity;

[0102] The superscript T indicates transpose;

[0103] Step 232: Using the Kalman filter model, predict the motion state of each vehicle in the next moment based on the current motion state of each vehicle, and initially generate the short trajectory of each vehicle.

[0104] Steps 2 and 3: Calculate the shortest trajectory for each vehicle. With trajectory The association cost between them is defined as:

[0105]

[0106] In the formula, This represents the cost associated with the trajectory and the measurement. This represents the Mahalanobis distance between the predicted and measured locations. Indicates the measured speed; Indicates the predicted speed; Indicates the measured heading angle; Indicates the predicted heading angle; This represents the intersection-union ratio (IU) between the predicted vehicle region and the measured vehicle region. , , , Represents the weight of each associated feature;

[0107] Steps two, three, and four: exclude trajectory pairs whose association cost exceeds the threshold, and then use the Hungarian algorithm to complete trajectory matching for the filtered trajectory pairs;

[0108] Steps 2, 3, and 5: Perform uniform interpolation on the trajectory pairs that have completed trajectory matching to fill in the missing trajectory points and generate a continuous long trajectory for each vehicle (the vehicle's target trajectory).

[0109] During this process, the interpolated and repaired trajectory points and the trajectory pairs with associated costs less than or equal to the threshold are marked.

[0110] The other steps and parameters are the same as in any of the specific implementation methods one to four.

[0111] Specific Implementation Method Six: This implementation method differs from Specific Implementation Methods One through Five in that: in step two four, the measured target trajectory of each vehicle is matched with the actual lane to obtain the lane matching result; the specific steps are as follows:

[0112] The vehicle's position is projected onto the intersection lane topology map. The lane in which the vehicle is located is determined based on the vehicle's lateral distance, vehicle heading difference, and historical lane continuity. The specific formula is as follows:

[0113]

[0114] In the formula, Indicates vehicle At any moment Lane matching results; Indicates the lane to be matched; Indicates that the vehicle has entered the lane. Lateral distance of the centerline; Indicates the vehicle's heading angle; Indicates the direction of the lane centerline; This is an indicator function; it takes the value 1 when the lane changes and 0 otherwise. , , Indicates the corresponding weight; Indicates vehicle At any moment The lane matching result; the argmin function selects the lane with the lowest overall cost.

[0115] The other steps and parameters are the same as those in any of the specific implementation methods one to five.

[0116] Specific Implementation Method Seven: This implementation method differs from one of Specific Implementation Methods One to Six in that: Step Three involves filtering candidate conflicting vehicle pairs based on the continuous trajectory information of the vehicles;

[0117] Initial screening was conducted based on spatial distance, temporal synchronization relationship, and lane topology relationship, with a focus on retaining vehicle pairs with potential interaction relationships;

[0118] The specific process is as follows:

[0119] Step 31: Filter candidate conflict vehicle pairs based on spatial distance; the specific process is as follows:

[0120] Future prediction time window Predict the positions of the two vehicles and calculate the minimum prediction distance; expressed as:

[0121]

[0122] In the formula, Indicates the prediction time window Inner vehicle and vehicles The minimum spatial distance;

[0123] Indicates the length of the prediction time window; Indicates the prediction time offset; and Representing vehicles and vehicles Predicted location at a future time; Indicates time;

[0124] For intersecting or merging lanes, calculate vehicles. and vehicles The predicted time difference of arrival at the area of ​​public conflict; expressed as:

[0125]

[0126] In the formula, Indicates vehicle and vehicles Time difference in arrival at the area of ​​public conflict; , Representing vehicles and vehicles Expected time of entry into the conflict zone (vehicle position, speed, and acceleration calculated) and vehicles (Expected time of entry into the conflict zone);

[0127] A vehicle pair is retained as a candidate conflict vehicle pair when any of the following conditions are met; denoted as:

[0128] or

[0129] In the formula, Indicates the spatial distance filtering threshold; This indicates the time difference threshold for reaching the conflict zone;

[0130] Step 32: Based on the candidate conflict vehicle pairs retained in Step 31, filter the candidate conflict vehicle pairs according to the time synchronization relationship. The specific process is as follows:

[0131] For any vehicle and vehicles Calculate the time overlap length of the trajectory:

[0132]

[0133] In the formula, Indicates vehicle and vehicles The length of time overlap of the trajectory; , Indicates vehicle and vehicles The starting time of the trajectory; , Indicates vehicle and vehicles The moment the trajectory terminates;

[0134] Only vehicle pairs that satisfy the following formula are retained:

[0135]

[0136] In the formula, Indicates the preset minimum effective time overlap length;

[0137] Time interpolation is performed on the trajectories of the selected vehicles to give them a unified timestamp.

[0138] Step 33: Based on the candidate conflict vehicle pairs retained in Step 32, filter the candidate conflict vehicle pairs according to the lane topology relationship. The specific process is as follows:

[0139] Based on the lane matching results from step two and four and the lane topology map, determine each candidate conflict vehicle pair retained in step three and two.

[0140] Whether the lane in question belongs to the same lane, upstream / downstream connecting lane, intersecting lane, merging lane, or adjacent lane where lane changing is permitted;

[0141] If it does not belong to the category, the corresponding vehicle pair will be excluded;

[0142] If it belongs to the category, the corresponding vehicle pair will be retained.

[0143] The other steps and parameters are the same as those in one of the specific implementation methods one to six.

[0144] Specific Implementation Method Eight: This implementation method differs from one of the specific implementation methods one to seven in that: in step four, the candidate conflict vehicle pairs are judged for conflict type, and the conflict type is divided into same-direction following conflict, cross-path conflict, merging or lateral intrusion conflict.

[0145] The specific process is as follows:

[0146] Traffic conflict types are categorized into three types (see example). Figure 5 , Figure 6 , Figure 7 As shown, the specific explanation is as follows:

[0147] Following collision: This type of collision occurs when vehicles in the same lane or on the same approach direction experience a rapid decrease in longitudinal distance due to the vehicle in front slowing down, stopping, or the vehicle behind traveling at excessive speed. The following vehicle then needs to brake to avoid a rear-end collision. This type of collision mainly occurs when vehicles are queuing at red lights, starting at green lights, crawling at low speeds, and during the dissipation of queues.

[0148] Cross-path conflict: Vehicles traveling in different directions have intersecting trajectories within the intersection, such as left-turning vehicles approaching oncoming straight-ahead vehicles, or lateral vehicles approaching longitudinal vehicles within the conflict area. A cross-path conflict is considered to exist when the time interval between the arrival of two vehicles at the conflict area is small, or when the predicted trajectories overlap within a short period of time.

[0149] Merging and lateral intrusion conflicts: When vehicles are parallel, merging, changing lanes, or approaching laterally near an approach lane, turning area, or lane boundary, if the lateral distance between the vehicles continues to decrease while the longitudinal clearance is insufficient, a lateral collision may occur or one party may be forced to slow down and avoid a collision. This type of conflict is more common in multi-lane approach lanes, right-turn merging, and irregular driving scenarios.

[0150] The candidate vehicle pairs are classified into conflict types based on vehicle movement direction, lane topology, and trajectory spatial relationships. The specific classification rules are as follows:

[0151] Vehicles are considered as candidates for a same-direction following conflict when they simultaneously meet the following conditions: (1) the two vehicles are in the same lane or consecutive lanes upstream and downstream; (2) the difference in heading angle between the two vehicles is not greater than 30°; (3) the net longitudinal distance between the two vehicles is not greater than 30 m; (4) the difference in speed between the following vehicle and the preceding vehicle is not less than 0.20 m / s; (5) the minimum net distance between the two vehicles within 5.0 s is predicted to be not greater than 8.0 m.

[0152] Vehicles are considered candidates for cross-path conflict when they simultaneously meet the following conditions: (1) there is a spatial intersection relationship between the lanes of the two vehicles; (2) the angle between the headings of the two vehicles is greater than 30° and less than 150°; (3) the intersection of the predicted trajectories is located in the region of interest or the preset conflict zone; (4) the time difference between the arrival of the two vehicles at the intersection of the trajectories or the common conflict zone is not greater than 4.0 s; (5) the minimum predicted distance is not greater than 6.0 m.

[0153] Vehicles are considered as candidates for merging or lateral intrusion when they simultaneously meet the following conditions: (1) The two vehicles are in adjacent lanes where lane changing is permitted or in lanes where merging is possible; (2) The current lateral distance is not greater than 4.0 m; (3) The absolute value of the longitudinal distance is not greater than 15 m; (4) The lateral distance has been decreasing for at least 4 adjacent moments within the last 0.5 s; (5) The lateral approach speed is not less than 0.20 m / s; (6) The predicted minimum distance is not greater than 5.0 m.

[0154] The other steps and parameters are the same as those in any of the specific implementation methods one to seven.

[0155] Specific Implementation Method Nine: This implementation method differs from Specific Implementation Methods One to Eight in that: in step five, multiple risk indicators are selected according to the conflict type, and multiple risk indicators are integrated to calculate a comprehensive conflict risk indicator and risk identification confidence level;

[0156] The specific process is as follows:

[0157] Step 51: Select risk conflict indicators, including headway (THW), time to collision (TTC), time to rear intrusion (PET), deceleration required to avoid collision (DRAC), predicted minimum distance (DCPA), and lateral velocity change rate.

[0158] The specific process is as follows:

[0159] Step 5.11: Calculate the headway (THW); expressed as:

[0160]

[0161] In the formula, express The car behind and the car in front Headway of the train;

[0162] , They represent The car behind and the car in front speed, and They represent The car behind and the car in front The vertical position, Indicates the vehicle in front Train Length;

[0163] Step 512: Calculate the collision time TTC; expressed as:

[0164]

[0165] In the formula, express The car behind and the car in front The collision time;

[0166] when At this time, there is no closed-loop speed relationship between the two vehicles, and TTC is taken as infinity; the smaller the TTC, the shorter the time available for the vehicle to react and avoid the collision, and the higher the risk of conflict.

[0167] Step 513: Calculate the post-invasive time PET; expressed as:

[0168]

[0169] In the formula, Indicates the following vehicle and the car in front The time after intrusion;

[0170] Indicates vehicle Time of entry into the conflict zone; Indicates vehicle Time of departure from the conflict zone;

[0171] A smaller PET indicates that the two vehicles pass through the conflict zone at similar times, and the risk of intersection conflicts is higher. This indicator is particularly applicable to scenarios involving left turns and straight-ahead travel, lateral crossings, and weaving paths within intersections.

[0172] Step 514: Calculate the deceleration DRAC required to avoid a collision; expressed as:

[0173]

[0174] In the formula, Indicates the following vehicle and the car in front The deceleration required to avoid a collision;

[0175] When the speed of the following vehicle is no higher than that of the preceding vehicle, or when the distance between the two vehicles is large, the DRAC value is smaller. The larger the DRAC value, the more powerful the braking is required to avoid a collision, and the higher the risk level.

[0176] Step 515: Calculate the following vehicle and the car in front exist The predicted minimum distance DCPA at time t is expressed as:

[0177]

[0178] In the formula, Indicates the following vehicle and the car in front exist The minimum predicted distance at any given time; express The car behind and the car in front The distance between, , For prediction in the time domain;

[0179] Step 516: The lateral velocity change rate is used to characterize the degree of change in the lateral motion state of a vehicle; its essence is lateral acceleration.

[0180] Calculate vehicles exist The rate of change of lateral velocity at time t is defined as:

[0181]

[0182] In the formula, Indicates vehicle exist Lateral velocity at any given moment; Indicates vehicle exist The rate of change of lateral velocity at time t;

[0183] It should be noted that absolute values ​​are generally used when making risk assessments, and the sign of the lateral velocity rate of change only indicates the direction of lateral acceleration;

[0184] Step 52: Normalize the headway THW, collision time TTC, rear intrusion time PET, deceleration required to avoid collision DRAC, predicted minimum distance DCPA, and lateral velocity change rate respectively to obtain the normalized headway THW, collision time TTC, rear intrusion time PET, deceleration required to avoid collision DRAC, predicted minimum distance DCPA, and lateral velocity change rate.

[0185] The specific process is as follows:

[0186] Because different risk indicators differ in their dimensions, value ranges, and risk directions, they cannot be directly linearly superimposed. Therefore, this paper first normalizes various risk indicators, transforming them into a unified risk score. The risk score range is as follows: The higher the value, the higher the risk of conflict for the vehicle at the current moment;

[0187] For indicators such as Time to Collision (TTC), Time to Rear Intrusion (PET), Predicted Minimum Distance (DCPA), and Headway (THW), which indicate higher risk with smaller values, a decreasing risk mapping function is used for normalization; expressed as:

[0188] At that time, the risk score is 1;

[0189] At that time, the risk score was ;

[0190] At that time, the risk score is 0;

[0191] in, This indicates the time to collision (TTC), time to intrusion (PET), predicted minimum distance (DCPA), or time to frontage (THW).

[0192] Indicates a low-risk boundary; Indicates a high-risk boundary;

[0193] For indicators such as the deceleration rate (DRAC) required to avoid a collision and the rate of change of lateral velocity, which indicate higher risk with larger values, an incremental risk mapping function is used for normalization; expressed as:

[0194] At that time, the risk score is 0;

[0195] At that time, the risk score was ;

[0196] At that time, the risk score is 1;

[0197] in, This represents the deceleration rate (DRAC) or lateral velocity change rate required to avoid a collision.

[0198] Indicates a low-risk boundary; Indicates a high-risk boundary;

[0199] Step 53: Based on the normalized headway (THW), collision time (TTC), rear intrusion time (PET), deceleration required to avoid collision (DRAC), predicted minimum distance (DCPA), and lateral velocity change rate, calculate the vehicle-to-vehicle (V2V) collision distance. At any moment The comprehensive conflict risk indicators are shown below:

[0200]

[0201] In the formula, Indicates the type of conflict. Represents the normalized i-th Risk scores for each risk conflict indicator Indicates the type of conflict The weights of the corresponding indicators are as follows: This represents the lane topology correction term. Weights for topology correction terms;

[0202] Among them, the lane topology correction term is used to reflect whether there is a potential actual conflict between vehicle pairs;

[0203] The conflict types are same-direction following conflict, cross-path conflict, or merging or lateral intrusion conflict;

[0204] Risk conflict indicators include headway (THW), time to collision (TTC), time to rear intrusion (PET), deceleration required to avoid collision (DRAC), predicted minimum distance (DCPA), or rate of change of lateral velocity.

[0205] Different conflict types are assigned different weights; for same-direction car-following conflicts, the weights of TTC, DRAC, and THW are increased; for cross-path conflicts, the weights of PET and DCPA are increased; for merging conflicts, the weights of DCPA and lateral distance change rate are increased; through categorized weighting, the comprehensive risk score can more accurately reflect the degree of danger in different traffic conflict scenarios.

[0206] Step 54: Considering the possibility of repair points and low-quality correlation points in the radar trajectory, calculate the risk identification confidence level; as shown below:

[0207]

[0208] In the formula, Indicates the following vehicle and the car in front exist Risk identification confidence level at any given moment;

[0209] and The following car and the car in front exist The quality of the trajectory points at any given moment;

[0210] Indicates the following vehicle and the car in front The proportion of repair points within the risk assessment time window (1.0s);

[0211] It does not directly change whether a conflict exists, but rather is used to mark the credibility of risk identification results. When the comprehensive risk index value is high but the confidence level is low, the result can be marked as "suspected risk" to avoid misjudgment caused by trajectory repair errors.

[0212] Trajectory point quality or Determined according to the following rules:

[0213] 1) If The car behind The corresponding point clusters have a signal-to-noise ratio (SNR) of at least 15 dB and a number of points of at least 5. hour, Set the value to 1.0;

[0214] like The car in front The corresponding point clusters have a signal-to-noise ratio (SNR) of at least 15 dB and a number of points of at least 5. hour, Set the value to 1.0;

[0215] 2) If The car behind The corresponding point clusters have a mean signal-to-noise ratio of not less than 10dB and a number of points of not less than 3. hour, Take 0.8;

[0216] like The car in front The corresponding point clusters have a mean signal-to-noise ratio of not less than 10dB and a number of points of not less than 3. hour, Take 0.8;

[0217] 3) If The car behind The mean signal-to-noise ratio of the corresponding point clusters is greater than 8dB and less than 10dB or hour, Take 0.6;

[0218] like The car behind The mean signal-to-noise ratio of the corresponding point clusters is greater than 8dB and less than 10dB or hour, Take 0.6;

[0219] 4) If the following vehicle If there are interpolated trajectory points that can be repaired, then... Set to 0;

[0220] If the car in front If there are interpolated trajectory points that can be repaired, then... Take 0.

[0221] The other steps and parameters are the same as those in one of the specific implementation methods one to eight.

[0222] Specific Implementation Method 10: This implementation method differs from Specific Implementation Methods 1 to 9 in that: in step 6, based on the comprehensive risk index value of each vehicle and combined with a preset threshold range, the severity of the conflict is classified to obtain the conflict severity assessment result.

[0223] The specific process is as follows:

[0224] The risk level is divided into four levels, and three thresholds are set: the risk of concern threshold, the potential conflict threshold, and the serious conflict threshold.

[0225] When a vehicle's comprehensive risk index value is less than three thresholds at the same time, the risk level is 0, the vehicle can pass normally, the vehicle spacing is sufficient, the TTC, PET and other indicators are relatively large, and there is no obvious sudden deceleration or trajectory weaving.

[0226] When a vehicle's comprehensive risk index value is greater than the risk threshold of concern but less than the potential conflict threshold and the serious conflict threshold, the risk level is Level 1. The vehicles are showing a certain tendency to approach each other, but no obvious conflict has yet occurred.

[0227] When the comprehensive risk index value of a vehicle is greater than the potential conflict threshold but less than the severe conflict threshold, the risk level is level 2. Indicators such as TTC, PET, and DRAC indicate that the vehicles have a strong spatial and temporal proximity relationship, requiring the driver to slow down or avoid them.

[0228] When the comprehensive risk index value of a vehicle is greater than the severe conflict risk threshold, the risk level is level 3. The predicted trajectories of the two vehicles overlap briefly, the TTC or PET is extremely small, or the required braking intensity is large, indicating a high risk of collision.

[0229] The thresholds for risk of concern are set at 0.25, for potential conflict at 0.50, and for serious conflict at 0.75.

[0230] The other steps and parameters are the same as those in any of the specific implementation methods one to nine.

[0231] The beneficial effects of the present invention are verified using the following embodiments:

[0232] Example 1:

[0233] This embodiment takes a two-way four-lane urban intersection as an example to illustrate the traffic conflict event identification method based on 4D millimeter-wave radar described in this invention.

[0234] S1. Collect and preprocess 4D millimeter-wave radar point cloud data at a roadside intersection. The specific implementation plan is as follows: Install a 77 GHz 4D millimeter-wave radar at the roadside of the intersection. The radar installation height is... The angle between the radar's horizontal measurement direction and the road centerline direction is The top-down view is The point cloud output frequency is The format of the measured data is shown in the appendix. Figure 2 As shown.

[0235] S2. Process the point cloud data to obtain continuous trajectory information of vehicles at intersections in the measured road segment. The specific implementation plan is as follows:

[0236] S21. Perform coordinate transformation and invalid point removal on the data. Take the radar projection point on the road surface as the origin of the road coordinate system, the road travel direction as the X-axis, the direction perpendicular to the road travel direction as the Y-axis, and the vertical upward direction as the Z-axis.

[0237] Set the effective detection range of the radar to The region of interest is along the X-axis. Y-axis direction: ; Relative to the Z-axis direction of the road surface: .

[0238] echo signal-to-noise ratio is lower than Radar points were identified as low-quality points and removed. For a signal-to-noise ratio of... The radar points are retained, but marked as low-quality points; the signal-to-noise ratio is not lower than [missing value]. The radar points are marked as normal points.

[0239] Before the formal inspection, choose a location where no vehicles are passing by. A fixed background template is established for each time period. The region of interest is divided into a grid with a side length of 0.25 m. When the occupancy rate of a grid is not less than 90% during the calibration period, and the standard deviation of the radar point positions within the grid is not greater than 0.15 m, the grid is marked as a fixed background grid. During the formal detection process, points that fall into the fixed background grid and are no more than 0.30 m away from the background points are removed.

[0240] To prevent the omission of low-speed or stationary vehicles, low radial velocity is not used as a sole background rejection criterion. For stationary or low-speed radar points located within the lane area that form a cluster of vehicle-sized points in three or more consecutive frames, even if their absolute radial velocity value is less than [a certain value], they are rejected. , and will also be retained.

[0241] For a distance of less than 0.10 m relative to the road surface height and a radial velocity absolute value of less than Furthermore, radar points with fewer than two neighboring points within a radius of 0.50m are identified as ground clutter and are removed.

[0242] S21. Cluster the preprocessed radar points using the DBSCAN algorithm. (The last part, "Any two radar points," appears to be a separate, unrelated sentence fragment and is left untranslated.) and The joint feature distance between them is:

[0243]

[0244] In the formula, This represents the joint characteristic distance between two radar points; , Representing radar points and The three-dimensional spatial position; , Representing radar points and radial velocity; , Representing radar points and Height; , , These represent the weights of spatial distance, radial velocity difference, and height difference, respectively.

[0245] In this embodiment, spatial distance weight is used. Radial velocity difference weighting Height difference weight DBSCAN neighborhood distance threshold Minimum number of cluster points .

[0246] When a point cluster contains the number of radar points , length is Width is Height is At that time, the cluster of points is identified as a vehicle observation target. For clusters of 3 points or with an average signal-to-noise ratio below [value missing], [the following applies]. Clusters of points are marked as low-quality vehicle observation targets, but not directly deleted, in order to improve the detection capability of vehicles in sparse point clouds. See the example for the clustering effect. Figure 3 , Figure 4 As shown.

[0247] S23, with Using a Kalman filter to predict vehicle states for a given filtering period, a vehicle trajectory is confirmed when a newly detected target is successfully associated with at least 3 out of 5 consecutive frames. The association cost is as follows:

[0248]

[0249] In the formula, This represents the cost associated with the trajectory and the measurement. This represents the Mahalanobis distance between the predicted and measured locations. Indicates the measured speed; Indicates the predicted speed; Indicates the measured heading angle; Indicates the predicted heading angle; This represents the intersection-union ratio (IU) between the predicted vehicle region and the measured vehicle region. , , , Represents the weight of each associated feature;

[0250] when A value not greater than 0.75 is allowed to participate in Hungarian matching; The associated combination is excluded when the value is greater than 0.75. When this happens, the corresponding trajectory point is marked as a low-quality correlation point.

[0251] The vehicle trajectory is lost for no more than 5 consecutive frames, that is, the loss time does not exceed [a certain value]. In this case, a uniform velocity model is used for interpolation repair, and the supplementary points are marked as repair points. (Continuous loss) In frame-by-frame analysis, only trajectory prediction is performed, and the predicted points are not used for high-confidence collision identification; if more than 8 frames are lost consecutively, i.e., exceeding [a certain threshold], [the situation is considered resolved]. The trajectory should be terminated at that time. The effective trajectory duration for traffic conflict identification should be no less than [number missing]. .

[0252] S24. Match the measured vehicles with the actual lanes. The specific steps are as follows: project the vehicle position onto the intersection lane topology map, and determine the vehicle's lane based on the comprehensive lateral distance, heading difference, and historical lane continuity. The specific formula is as follows:

[0253]

[0254] In the formula, Indicates vehicle At any moment Lane matching results; Indicates the lane to be matched; Indicates that the vehicle has entered the lane. Lateral distance of the centerline; Indicates the vehicle's heading angle; Indicates the direction of the lane centerline; This is an indicator function; it takes the value 1 when the lane changes and 0 otherwise. , , This indicates the corresponding weight; the argmin function selects the lane with the lowest overall cost.

[0255] In this case , , .

[0256] Lane switching is confirmed only when the vehicle matches the new lane for 3 consecutive frames and the vehicle's center of gravity crosses the original lane boundary by at least 0.30 m; otherwise, the original lane matching result is maintained to suppress frequent jumps in matching results near the lane boundary.

[0257] S3. Screening candidate conflicting vehicle pairs. Initial screening is performed based on spatial distance, temporal synchronization relationship, and lane topology, with a focus on retaining vehicle pairs with potential interaction relationships. The specific implementation plan is as follows:

[0258] S31. Candidate conflict vehicle pairs are selected based on spatial distance. The specific process is as follows: within the future prediction time window... Predict the positions of the two vehicles and calculate the minimum prediction distance:

[0259]

[0260] In the formula, Indicates the prediction time window Inner vehicle and vehicles The minimum spatial distance;

[0261] Indicates the length of the prediction time window; Indicates the prediction time offset; and Representing vehicles and vehicles Predicted location at a future time; Indicates time;

[0262] For intersecting or merging lanes, calculate the predicted time difference between the arrival times of the two vehicles at the common conflict zone: ;

[0263] In the formula, Indicates vehicle and vehicles Time difference in arrival at the area of ​​public conflict; , Representing vehicles and vehicles Expected time of entry into the conflict zone (vehicle position, speed, and acceleration calculated) and vehicles (Expected time of entry into the conflict zone);

[0264] The vehicle pair will be retained as a candidate conflict vehicle pair if any of the following conditions are met: or ;

[0265] In the formula, Indicates the spatial distance filtering threshold; This indicates the time difference threshold for reaching the conflict zone;

[0266] S32. Based on the time synchronization relationship, candidate conflicting vehicle pairs are selected. The specific implementation process is as follows:

[0267] For any vehicle and vehicles Calculate the time overlap length of its trajectory:

[0268]

[0269] In the formula, Indicates vehicle and vehicles The length of time overlap of the trajectory; , Indicates vehicle and vehicles The starting time of the trajectory; , Indicates vehicle and vehicles The moment the trajectory terminates;

[0270] Only vehicle pairs that satisfy the following formula are retained: ;

[0271] In the formula, Indicates the preset minimum effective time overlap length;

[0272] Time interpolation is performed on the trajectories of the selected vehicles to give them a unified timestamp.

[0273] S33. Filter candidate conflicting vehicle pairs based on lane topology relationships. In S24, this method has already matched vehicles with lanes. Based on the lane topology map, it is determined whether the lanes of the previously filtered vehicle pairs belong to the same lane, upstream and downstream connecting lanes, intersecting lanes, merging lanes, or adjacent lanes that allow lane changes. Vehicle pairs that do not have the above topology relationships are excluded.

[0274] This case study adopts the approach described in S31-S32. The prediction time window H, to The predicted vehicle position is calculated based on the prediction step size. The candidate vehicle selection parameters are set as follows: minimum prediction distance threshold. Time difference threshold for reaching public conflict zones Minimum trajectory time overlap length .

[0275] When the minimum predicted distance between the two vehicles within the prediction time window Not greater than Or the time difference between the two vehicles' expected entry into the conflict zone. Not greater than At that time, the vehicle pair will be retained.

[0276] The retained vehicles should also have no less than Vehicle pairs whose shared trajectory time is specified, and whose lanes belong to the same lane, upstream / downstream connecting lanes, intersecting lanes, merging lanes, or adjacent lanes where lane changes are permitted, are excluded.

[0277] S4. Identify Traffic Conflict Types. Based on vehicle movement direction, lane topology, and predicted trajectory spatial relationships, candidate vehicle pairs are classified into three conflict categories: same-direction following conflict, intersecting path conflict, and merging or lateral intrusion conflict. The specific implementation plan is as follows:

[0278] S41. Traffic conflict types are classified into three categories (see example). Figure 5 , Figure 6 , Figure 7 As shown, the specific explanation is as follows:

[0279] Following-in-the-line conflict: This occurs when vehicles in the same lane or on the same approach direction experience a rapid decrease in longitudinal distance due to the vehicle in front slowing down, stopping, or the vehicle behind traveling at excessive speed, requiring the following vehicle to brake to avoid a rear-end collision. This type of conflict mainly occurs during red light queuing, green light starting, slow-moving vehicles, and queue dissipation phases.

[0280] Cross-path conflict: Vehicles traveling in different directions have intersecting trajectories within the intersection, such as left-turning vehicles approaching oncoming straight-ahead vehicles, or lateral vehicles approaching longitudinal vehicles within the conflict area. A cross-path conflict is considered to exist when the time interval between the arrival of two vehicles at the conflict area is small, or when their predicted trajectories overlap within a short period of time.

[0281] Merging and lateral intrusion conflicts: When vehicles are parallel, merging, changing lanes, or approaching laterally near an approach lane, turning area, or lane boundary, if the lateral distance between the vehicles continues to decrease while the longitudinal clearance is insufficient, a lateral collision may occur or one party may be forced to slow down and avoid a collision. This type of conflict is more common in multi-lane approach lanes, right-turn merging, and irregular driving scenarios.

[0282] S42. Based on the vehicle's direction of motion, lane topology, and trajectory spatial relationships, the candidate vehicle pairs are determined to have conflict types. The specific determination rules are as follows:

[0283] Vehicles are considered candidates for a same-direction following conflict when both of the following conditions are met: (1) the two vehicles are in the same lane or consecutive lanes upstream and downstream; (2) the difference in heading angle between the two vehicles is not greater than 1. (3) The net longitudinal distance between the two vehicles shall not exceed 30 m; (4) The difference between the speed of the rear vehicle and the speed of the front vehicle shall not be less than (5) Prediction The minimum clearance between the two inner vehicles shall not be greater than .

[0284] Vehicles are considered candidates for cross-path conflict when they simultaneously meet the following conditions: (1) there is a spatial intersection relationship between the lanes of the two vehicles; (2) the angle between the headings of the two vehicles is... (3) The predicted trajectory intersection point is located within the region of interest or the preset conflict zone; (4) The time difference between the arrival of the two vehicles at the trajectory intersection point or the common conflict zone is no greater than 100%. (4) The predicted minimum distance is not greater than .

[0285] Vehicles are considered merging or lateral intrusion candidates when both of the following conditions are met: (1) the two vehicles are in adjacent lanes where lane changing is permitted or in lanes where merging is possible; (2) the current lateral distance is not greater than [missing information]. (3) The absolute value of the longitudinal distance is not greater than ; (4) Recently (5) The lateral distance between two adjacent points in time decreases at least four times; (6) The predicted minimum distance is not greater than .

[0286] S5. Integrate multiple indicators to calculate a comprehensive risk index and risk identification confidence level. Multiple risk indicators are selected based on risk categories, and these indicators are calculated into a single comprehensive risk index value using a formula. Considering the potential presence of repair points and low-quality correlation points in the vehicle trajectory, the risk identification confidence level is further calculated. The specific implementation plan is as follows:

[0287] S51. Select classic collision indicators, including time to collision (TTC), time to intrusion (PET), deceleration required to avoid collision (DRAC), predicted minimum distance (DCPA), headway (THW), and rate of change of lateral distance, and calculate the specific values ​​using formulas.

[0288]

[0289] In the formula, express The car behind and the car in front Headway of the train;

[0290] , They represent The car behind and the car in front speed, and They represent The car behind and the car in front The vertical position, Indicates the vehicle in front Train Length;

[0291] Calculate the collision time TTC; expressed as:

[0292]

[0293] In the formula, express The car behind and the car in front The collision time;

[0294] when At this time, there is no closed-loop speed relationship between the two vehicles, and TTC is taken as infinite. The smaller the TTC, the shorter the time available for the vehicle to react and avoid a collision, and the higher the risk of conflict.

[0295] The post-invasive time PET is calculated and expressed as:

[0296]

[0297] In the formula, Indicates the following vehicle and the car in front The time after intrusion;

[0298] Indicates vehicle Time of entry into the conflict zone; Indicates vehicle Time of departure from the conflict zone;

[0299] The smaller the PET value, the closer the two vehicles pass through the conflict zone, and the higher the risk of intersection conflicts. This indicator is particularly applicable to scenarios involving left turns and straight-ahead travel, lateral crossings, and weaving paths within intersections.

[0300] Calculate the deceleration DRAC required to avoid a collision; expressed as:

[0301]

[0302] In the formula, Indicates the following vehicle and the car in front The deceleration required to avoid a collision;

[0303] When the speed of the following vehicle is no higher than that of the preceding vehicle, or when the distance between the two vehicles is large, the DRAC value is smaller. The larger the DRAC value, the more powerful the braking is required to avoid a collision, and the higher the risk level.

[0304] The rate of change of lateral velocity is used to characterize the degree of change in the lateral motion state of a vehicle; its essence is lateral acceleration. (Calculation of vehicle...) exist The rate of change of lateral velocity at time t is defined as:

[0305]

[0306] In the formula, Indicates vehicle exist Lateral velocity at any given moment; Indicates vehicle exist The rate of change of lateral velocity at time t;

[0307] It should be noted that absolute values ​​are generally used when making risk assessments, and the sign of the lateral velocity change rate only indicates the direction of lateral acceleration.

[0308] Calculate the following vehicle and the car in front exist The predicted minimum distance DCPA at time t is expressed as:

[0309]

[0310] In the formula, Indicates the following vehicle and the car in front exist The minimum predicted distance at any given time; express The car behind and the car in front The distance between, , For prediction in the time domain;

[0311] S52. For indicators such as Time to Collision (TTC), Time to Rear Intrusion (PET), Predicted Minimum Distance (DCPA), and Headway (THW), which are considered to have higher risk with smaller values, a decreasing risk mapping function is used for normalization; expressed as:

[0312] At that time, the risk score is 1;

[0313] At that time, the risk score was ;

[0314] At that time, the risk score is 0;

[0315] in, This indicates the time to collision (TTC), time to intrusion (PET), predicted minimum distance (DCPA), or time to frontage (THW).

[0316] Indicates a low-risk boundary; Indicates a high-risk boundary;

[0317] For indicators such as the deceleration rate (DRAC) required to avoid a collision and the rate of change of lateral velocity, which indicate higher risk with larger values, an incremental risk mapping function is used for normalization; expressed as:

[0318] At that time, the risk score is 0;

[0319] At that time, the risk score was ;

[0320] At that time, the risk score is 1;

[0321] in, This indicates the time to collision (TTC), time to intrusion (PET), predicted minimum distance (DCPA), or time to frontage (THW).

[0322] Indicates a low-risk boundary; Indicates a high-risk boundary;

[0323]

[0324] S53, Calculate vehicle pairs At any moment The comprehensive conflict risk indicators are as follows:

[0325]

[0326] In the formula, Indicates the type of conflict. Represents the normalized i-th Risk scores for each risk conflict indicator Indicates the type of conflict The weights of the corresponding indicators are as follows: This represents the lane topology correction term. Weights for topology correction terms;

[0327] Among them, the lane topology correction term is used to reflect whether there is a possibility of actual conflict between vehicle pairs.

[0328]

[0329] When the lanes of the two vehicles are directly intersecting, are in adjacent lanes, or are directly merging, Take 1; when two vehicles are in adjacent lanes where lane changing is permitted, Set the value to 0.6; to reduce fluctuations in the point cloud of a single frame, use the median of the risk values ​​obtained at the current time and the two previous times as the current output risk value.

[0330] S54. Considering the possible existence of repair points and low-quality correlation points in the radar trajectory, the risk identification confidence level is calculated as follows:

[0331]

[0332] In the formula, Indicates the following vehicle and the car in front exist Risk identification confidence level at any given moment;

[0333] and The following car and the car in front exist The quality of the trajectory points at any given moment;

[0334] Indicates the following vehicle and the car in front The proportion of repair points within the risk assessment time window;

[0335] It does not directly change whether a conflict exists, but rather is used to mark the credibility of risk identification results. When the comprehensive risk index value is high but the confidence level is low, the result can be marked as "suspected risk" to avoid misjudgment caused by trajectory repair errors.

[0336] Trajectory point quality The following rules shall be applied: (1) The signal-to-noise ratio of the point cluster shall not be less than 15 dB, and the number of points shall not be less than 5. hour, Take 1.0; (2) The signal-to-noise ratio of the point cluster is not less than 10 dB and the number of points is not less than 3 and hour, Take 0.8; (3) Satisfies the minimum detection and correlation threshold, but the signal-to-noise ratio is or hour, Take 0.6; (4) Interpolation repair point Take 0. See S23.

[0337] and Take the time before the risk assessment respectively Average quality of non-repaired trajectory points within the time window. Proportion of repaired points. Take the two cars at this The ratio of the total number of repair points to the total number of trajectory points within the time window.

[0338] The confidence thresholds are as follows: High confidence level, output is confirmed risk;

[0339] Medium confidence level, output as suspected risk;

[0340] Low confidence level: Only retain risk records, not as confirmation of conflict events;

[0341] At that time, regardless of the magnitude of the overall risk value, it is marked as low confidence.

[0342] S6. Conflict Risk Level Classification. Based on the comprehensive risk index value of each vehicle and a preset threshold range, the severity of the conflict is classified to obtain the conflict severity assessment result. The specific implementation plan is as follows:

[0343] The risk level is divided into four levels, with three thresholds set: a concern risk threshold of 0.25, a potential conflict threshold of 0.50, and a severe conflict threshold of 0.75. When the comprehensive risk index value of a vehicle is less than all three thresholds, the risk level is 0, indicating normal traffic flow, sufficient distance between vehicles, high TTC and PET values, and no obvious sudden deceleration or trajectory interweaving. When the comprehensive risk index value of a vehicle is greater than the concern risk threshold but less than the other two, the risk level is 1, indicating a certain tendency for vehicles to approach each other, but no obvious conflict has yet occurred. When the comprehensive risk index value of a vehicle is greater than the potential conflict threshold but less than the severe conflict threshold, the risk level is 2, indicating a strong spatiotemporal proximity between vehicles, requiring the driver to slow down or avoid them. When the comprehensive risk index value of a vehicle is greater than the severe conflict risk threshold, the risk level is 3, indicating a short-term overlap of the predicted trajectories of the two vehicles, extremely low TTC or PET values, or a large braking intensity required, posing a high risk of collision.

[0344] A risk event is formed when five consecutive frames of Level 1 risk, three consecutive frames of Level 2 risk, or two consecutive frames of Level 3 risk occur. If this occurs... , , or If any of the conditions are met, a Level 3 risk event will be immediately output.

[0345] The specific vehicle pair calculation example is as follows:

[0346] At a certain moment, two vehicles are detected in the same lane, one in front of the other. The speed of the vehicle behind is... The speed of the car in front is The net longitudinal distance between the two vehicles is For a conflict arising from car-following in the same direction, the calculation yields:

[0347] ;

[0348] ;

[0349] ;

[0350] Prediction within 5.0 seconds ;

[0351] The lane topology correction term is 1.

[0352] After normalization, the TTC risk score is approximately 0.71, the THW risk score is approximately 0.98, the DRAC risk score is 0, and the DCPA risk score is 1. The overall risk value is:

[0353] .

[0354] Therefore, the vehicle was classified as a Level 2 potential conflict.

[0355] If the masses of the trajectory points of the two vehicles are 0.90 and 0.88 respectively, and the proportion of repair points within the most recent 1.0 s is 0.10, then: .

[0356] because If the value is greater than 0.65, the final output result is a high-confidence level 2 co-directional racing conflict.

[0357] This invention may have other embodiments. Without departing from the spirit and essence of this invention, those skilled in the art can make various corresponding changes and modifications according to this invention, but these corresponding changes and modifications should all fall within the protection scope of the appended claims.

Claims

1. A method for identifying traffic conflict events at intersections based on 4D millimeter-wave radar, characterized in that: The specific process of the method is as follows: Step 1: Collect 4D millimeter-wave radar point cloud data at the roadside of a certain intersection; the specific process is as follows: A 4D millimeter-wave radar was installed on the side of a road at an intersection, and point cloud data of the 4D millimeter-wave radar was acquired based on the 4D millimeter-wave radar. Based on 4D millimeter-wave radar point cloud data, the timestamp, radial distance, azimuth angle, elevation angle, radial velocity, and echo quality of the point cloud data are obtained. The echo quality refers to echo power, signal-to-noise ratio, and noise. Step 2: Process the point cloud data to obtain continuous trajectory information of vehicles at the intersection of the measured road segment; Step 3: Filter candidate conflicting vehicle pairs based on continuous vehicle trajectory information; Step 4: Determine the conflict type for candidate conflict vehicle pairs. Conflict types are divided into same-direction following conflict, cross-path conflict, merging or lateral intrusion conflict. Step 5: Select multiple risk indicators based on the type of conflict, and combine the multiple risk indicators to calculate the comprehensive risk indicator and the risk identification confidence level; Step 6: Based on the comprehensive risk index value of each vehicle and the preset threshold range, classify the severity of the conflict to obtain the conflict severity assessment result.

2. The method for identifying traffic conflict events at intersections based on 4D millimeter-wave radar according to claim 1, characterized in that: Step two involves processing the point cloud data to obtain continuous trajectory information of vehicles at the intersection of the measured road segment. The specific process is as follows: Step 2:

1. Convert the coordinates of the measured point cloud data to the road coordinate system; 2. Remove invalid points from the converted point cloud data to obtain the point cloud data after removing invalid points. Step 22: Cluster the radar points based on their spatial location, radial velocity, echo quality, and altitude; Steps 2 and 3: Associate the trajectories of vehicles belonging to the same type to generate the vehicle target trajectory; Step 24: Match the measured target trajectory of each vehicle with the actual lane to obtain the lane matching result.

3. The method for identifying traffic conflict events at intersections based on 4D millimeter-wave radar according to claim 2, characterized in that: Step 21: Convert the measured point cloud data coordinates to the road coordinate system; remove invalid points from the converted coordinate point cloud to obtain point cloud data after removing invalid points; The specific process is as follows: Step 2: Convert the radial distance measured in Step 1 to the road coordinate system to obtain the radial distance of the point cloud data in the road coordinate system. Transform the azimuth angle measured in step one to the road coordinate system to obtain the azimuth angle of the point cloud data in the road coordinate system; Transform the pitch angle measured in step one to the road coordinate system to obtain the pitch angle of the point cloud data in the road coordinate system; The radial velocity measured in step one is converted to the road coordinate system to obtain the radial velocity of the point cloud data in the road coordinate system. The echo power measured in step one is converted to the road coordinate system to obtain the echo power of the point cloud data in the road coordinate system; The signal-to-noise ratio measured in step one is converted to the road coordinate system to obtain the signal-to-noise ratio of the point cloud data in the road coordinate system; The noise measured in step one is converted to the road coordinate system to obtain the noise of the point cloud data in the road coordinate system; The road coordinate system is defined with the radar projection point on the road surface as the origin, the road travel direction as the X-axis, the perpendicular direction of the road travel direction as the Y-axis, and the vertically upward direction as the Z-axis. Step 212: Remove invalid points from the point cloud data in the road coordinate system to obtain the point cloud data after removing invalid points; The specific process is as follows: Step 2121: Crop the region of interest for each frame of point cloud in the road coordinate system; Step 2122: Filter outliers in the radar point cloud data within the region of interest to obtain filtered radar point cloud data; the specific process is as follows: If the signal-to-noise ratio of the radar point cloud data in the region of interest is less than 8 dB, the radar point cloud is determined to be a low-quality point and is removed. If the signal-to-noise ratio of the radar point cloud data in the region of interest is greater than or equal to 8 dB and less than 10 dB, the radar point cloud will be retained but marked as a low-quality point. If the signal-to-noise ratio of the radar point cloud data in the region of interest is greater than or equal to 10 dB, the radar point cloud is marked as a normal point. Steps 2-3: The filtered radar point cloud data undergoes background removal to obtain radar point cloud data with background removed; the specific process is as follows: 1) Divide the region of interest into grids with a vertical dimension of 1m and a horizontal dimension of 0.5m; 2) Statistical analysis of the occurrence of each grid cell throughout the entire time period, including the total number of radar points, the number of low-speed static points, the number of obviously moving points, and the time span of occurrence; Among them, low-speed static points are defined as radar points with an absolute radial velocity of no more than 0.25 m / s; A point of obvious motion is defined as a radar point with an absolute radial velocity of not less than 0.8 m / s; 3) If a grid cell meets all of the following conditions, then the corresponding grid cell will be marked as a candidate background grid cell: First, the time span of the grid appearance must be no less than 600 seconds; Second, the number of low-speed static points within the grid shall not be less than 100; Third, the proportion of low-speed static points within a grid to the total number of radar points within the corresponding grid shall not be less than 90%; Fourth, the ratio of the number of obviously moving points to the number of low-speed static points within the grid should not exceed 3%; 4) A 2.0 m radius on either side of the center line of each lane is designated as a lane protection zone; Low-speed static points located within lane protection zones will not be directly deleted as background targets, even if the absolute value of their radar radial velocity is less than or equal to 0.25 m / s. 5) During the background removal stage, a radar point is only identified as a background point and removed if it simultaneously meets the following conditions: The radar point is located in a background candidate grid; the absolute value of the radial velocity of the radar point is no greater than 0.25 m / s; the radar point is not located within the lane protection zone. Steps 21-24: Perform ground clutter suppression on the radar point cloud data obtained in Steps 21-23 after background removal; the specific process is as follows: Radar points with a height of less than 0.10 m relative to the road surface, an absolute radial velocity of less than 0.30 m / s, and fewer than two adjacent points within a radius of 0.50 m are identified as ground clutter and are removed.

4. The method for identifying traffic conflict events at intersections based on 4D millimeter-wave radar according to claim 3, characterized in that: Step two involves clustering radar points based on their spatial location, radial velocity, echo quality, and altitude. The specific process is as follows: Step 221: Calculate the joint feature distance between point clouds; the process is as follows: Any two radar points and The joint feature distance is defined as: In the formula, This represents the joint characteristic distance between two radar points; , Representing radar points and The three-dimensional spatial position; , Representing radar points and radial velocity; , Representing radar points and Height; , , These represent the weights of spatial distance, radial velocity difference, and height difference, respectively. Step 222: Based on the joint feature distance between point clouds, the DBSCAN neighborhood distance threshold, and the minimum number of cluster points, the DBSCAN clustering method is used to form vehicle target point clusters; For each point cluster, the centroid position, target size, average radial velocity, number of radar points, mean echo power, mean signal-to-noise ratio, mean noise, and dispersion are calculated to obtain the vehicle's observed target.

5. The method for identifying traffic conflict events at intersections based on 4D millimeter-wave radar according to claim 4, characterized in that: Steps two and three involve associating the trajectories of vehicles belonging to the same type to generate the vehicle target trajectory; the specific steps are as follows: Steps 2 and 31: For the vehicle Establish motion state vector: In the formula, Indicates vehicle At any moment The state of motion; These represent vehicles in the X and Y directions of the road coordinate system, respectively. At any moment Positional components; These represent vehicles in the X and Y directions of the road coordinate system, respectively. At any moment velocity components; These represent vehicles in the X and Y directions of the road coordinate system, respectively. At any moment The acceleration components; Indicates vehicle At any moment The heading angle; Indicates vehicle At any moment angular velocity; The superscript T indicates transpose; Step 232: Using the Kalman filter model, predict the motion state of each vehicle in the next moment based on the current motion state of each vehicle, and initially generate the short trajectory of each vehicle. Steps 2 and 3: Calculate the shortest trajectory for each vehicle. With trajectory The association cost between them is defined as: In the formula, This represents the cost associated with the trajectory and the measurement. This represents the Mahalanobis distance between the predicted and measured locations. Indicates the measured speed; Indicates the predicted speed; Indicates the measured heading angle; Indicates the predicted heading angle; This represents the intersection-union ratio (IU) between the predicted vehicle region and the measured vehicle region. , , , Represents the weight of each associated feature; Steps two, three, and four: exclude trajectory pairs whose association cost exceeds the threshold, and then use the Hungarian algorithm to complete trajectory matching for the filtered trajectory pairs; Steps 2, 3, and 5: Perform uniform interpolation on the trajectory pairs that have completed trajectory matching to fill in the missing trajectory points and generate a continuous long trajectory for each vehicle. The interpolated repaired trajectory points and the trajectory pairs with associated costs less than or equal to the threshold are marked.

6. The method for identifying traffic conflict events at intersections based on 4D millimeter-wave radar according to claim 5, characterized in that: Step two and four involve matching the measured target trajectory of each vehicle with the actual lane to obtain the lane matching result. The specific steps are as follows: The vehicle's position is projected onto the intersection lane topology map. The lane in which the vehicle is located is determined based on the vehicle's lateral distance, vehicle heading difference, and historical lane continuity. The specific formula is as follows: In the formula, Indicates vehicle At any moment Lane matching results; Indicates the lane to be matched; Indicates that the vehicle has entered the lane. Lateral distance of the centerline; Indicates the vehicle's heading angle; Indicates the direction of the lane centerline; This is an indicator function; it takes the value 1 when the lane changes and 0 otherwise. , , Indicates the corresponding weight; Indicates vehicle At any moment The lane matching result; the argmin function selects the lane with the lowest overall cost.

7. The method for identifying traffic conflict events at intersections based on 4D millimeter-wave radar according to claim 6, characterized in that: Step three involves filtering candidate conflicting vehicle pairs based on the vehicle's continuous trajectory information; the specific process is as follows: Step 31: Filter candidate conflict vehicle pairs based on spatial distance; the specific process is as follows: Future prediction time window Predict the positions of the two vehicles and calculate the minimum prediction distance; expressed as: In the formula, Indicates the prediction time window Inner vehicle and vehicles The minimum spatial distance; Indicates the length of the prediction time window; Indicates the prediction time offset; and Representing vehicles and vehicles Predicted location at a future time; Indicates time; For intersecting or merging lanes, calculate vehicles. and vehicles The predicted time difference of arrival at the area of ​​public conflict; expressed as: In the formula, Indicates vehicle and vehicles Time difference in arrival at areas of public conflict; , Representing vehicles and vehicles Expected time of entry into the conflict zone; A vehicle pair is retained as a candidate conflict vehicle pair when any of the following conditions are met; denoted as: or In the formula, Indicates the spatial distance filtering threshold; This indicates the time difference threshold for reaching the conflict zone; Step 32: Based on the candidate conflict vehicle pairs retained in Step 31, filter the candidate conflict vehicle pairs according to the time synchronization relationship. The specific process is as follows: For any vehicle and vehicles Calculate the time overlap length of the trajectory: In the formula, Indicates vehicle and vehicles The length of time overlap of the trajectory; , Indicates vehicle and vehicles The starting time of the trajectory; , Indicates vehicle and vehicles The moment the trajectory terminates; Only vehicle pairs that satisfy the following formula are retained: In the formula, Indicates the preset minimum effective time overlap length; Time interpolation is performed on the trajectories of the selected vehicles to give them a unified timestamp. Step 33: Based on the candidate conflict vehicle pairs retained in Step 32, filter the candidate conflict vehicle pairs according to the lane topology relationship. The specific process is as follows: Based on the lane matching results from step two and four and the lane topology map, determine each candidate conflict vehicle pair retained in step three and two. Whether the lane in question belongs to the same lane, upstream / downstream connecting lane, intersecting lane, merging lane, or adjacent lane where lane changing is permitted; If it does not belong to the category, the corresponding vehicle pair will be excluded; If it belongs to the category, the corresponding vehicle pair will be retained.

8. The method for identifying traffic conflict events at intersections based on 4D millimeter-wave radar according to claim 7, characterized in that: Step four involves identifying the conflict type of candidate conflicting vehicle pairs. Conflict types are categorized as same-direction following conflict, cross-path conflict, merging or lateral intrusion conflict. The specific process is as follows: Vehicles are considered as candidates for a same-direction following conflict when they simultaneously meet the following conditions: (1) the two vehicles are in the same lane or consecutive lanes upstream and downstream; (2) the difference in heading angle between the two vehicles is not greater than 30°; (3) the net longitudinal distance between the two vehicles is not greater than 30 m; (4) the difference in speed between the following vehicle and the preceding vehicle is not less than 0.20 m / s; (5) the minimum net distance between the two vehicles within 5.0 s is predicted to be not greater than 8.0 m. Vehicles are considered candidates for cross-path conflict when they simultaneously meet the following conditions: (1) there is a spatial intersection relationship between the lanes of the two vehicles; (2) the angle between the headings of the two vehicles is greater than 30° and less than 150°; (3) the intersection of the predicted trajectories is located in the region of interest or the preset conflict zone; (4) the time difference between the arrival of the two vehicles at the intersection of the trajectories or the common conflict zone is not greater than 4.0 s; (5) the minimum predicted distance is not greater than 6.0 m. Vehicles are considered as candidates for merging or lateral intrusion when they simultaneously meet the following conditions: (1) The two vehicles are in adjacent lanes where lane changing is permitted or in lanes where merging is possible; (2) The current lateral distance is not greater than 4.0 m; (3) The absolute value of the longitudinal distance is not greater than 15 m; (4) The lateral distance has been decreasing for at least 4 adjacent moments within the last 0.5 s; (5) The lateral approach speed is not less than 0.20 m / s; (6) The predicted minimum distance is not greater than 5.0 m.

9. The method for identifying traffic conflict events at intersections based on 4D millimeter-wave radar according to claim 8, characterized in that: Step five involves selecting multiple risk indicators based on the conflict type, and then integrating these multiple risk indicators to calculate a comprehensive conflict risk indicator and a risk identification confidence level. The specific process is as follows: Step 51: Select risk conflict indicators, including headway (THW), time to collision (TTC), time to rear intrusion (PET), deceleration required to avoid collision (DRAC), predicted minimum distance (DCPA), and lateral velocity change rate. The specific process is as follows: Step 5.11: Calculate the headway (THW); expressed as: In the formula, express The car behind and the car in front Headway of the train; , They represent The car behind and the car in front speed, and They represent The car behind and the car in front The vertical position, Indicates the vehicle in front Train Length; Step 512: Calculate the collision time TTC; expressed as: In the formula, express The car behind and the car in front The collision time; Step 513: Calculate the post-invasive time PET; expressed as: In the formula, Indicates the following vehicle and the car in front The time after intrusion; Indicates vehicle Time of entry into the conflict zone; Indicates vehicle Time of departure from the conflict zone; Step 514: Calculate the deceleration DRAC required to avoid a collision; expressed as: In the formula, Indicates the following vehicle and the car in front The deceleration required to avoid a collision; Step 515: Calculate the following vehicle and the car in front exist The predicted minimum distance DCPA at time t is expressed as: In the formula, Indicates the following vehicle and the car in front exist The minimum predicted distance at any given time; express The car behind and the car in front The distance between, , For prediction in the time domain; Step 516: Calculate the vehicle exist The rate of change of lateral velocity at time t is defined as: In the formula, Indicates vehicle exist Lateral velocity at any given moment; Indicates vehicle exist The rate of change of lateral velocity at time t; Step 5.2: Normalize the headway (THW), collision time (TTC), rear intrusion time (PET), deceleration required to avoid collision (DRAC), predicted minimum distance (DCPA), and lateral velocity change rate, respectively, to obtain the normalized headway (THW), collision time (TTC), rear intrusion time (PET), deceleration required to avoid collision (DRAC), predicted minimum distance (DCPA), and lateral velocity change rate; the specific process is as follows: For collision time TTC, rear intrusion time PET, predicted minimum distance DCPA, and headway THW, a decreasing risk mapping function is used for normalization; expressed as: At that time, the risk score is 1; At that time, the risk score was ; At that time, the risk score is 0; in, This indicates the time to collision (TTC), time to intrusion (PET), predicted minimum distance (DCPA), or time to frontage (THW). Indicates a low-risk boundary; Indicates a high-risk boundary; The deceleration rate (DRAC) and lateral velocity change rate required to avoid a collision are normalized using an incremental risk mapping function, expressed as: At that time, the risk score is 0; At that time, the risk score was ; At that time, the risk score is 1; in, This represents the deceleration rate (DRAC) or lateral velocity change rate required to avoid a collision. Indicates a low-risk boundary; Indicates a high-risk boundary; Step 53: Based on the normalized headway (THW), collision time (TTC), rear intrusion time (PET), deceleration required to avoid collision (DRAC), predicted minimum distance (DCPA), and lateral velocity change rate, calculate the vehicle-to-vehicle (V2V) collision distance. At any moment The comprehensive conflict risk indicators are shown below: In the formula, Indicates the type of conflict. Represents the normalized i-th Risk scores for each risk conflict indicator Indicates the type of conflict The weights of the corresponding indicators are as follows: This represents the lane topology correction term. Weights for topology correction terms; The conflict types are same-direction following conflict, cross-path conflict, or merging or lateral intrusion conflict; Risk conflict indicators include headway (THW), time to collision (TTC), time to rear intrusion (PET), deceleration required to avoid collision (DRAC), predicted minimum distance (DCPA), or rate of change of lateral velocity. Step 54: Calculate the risk identification confidence level; as shown below: In the formula, Indicates the following vehicle and the car in front exist Risk identification confidence level at any given moment; and The following car and the car in front exist The quality of the trajectory points at any given moment; Indicates the following vehicle and the car in front The proportion of repair points within the risk assessment time window; Trajectory point quality or Determined according to the following rules: 1) If The car behind The corresponding point clusters have a signal-to-noise ratio (SNR) of at least 15 dB and a number of points of at least 5. hour, Set the value to 1.0; like The car in front The corresponding point clusters have a signal-to-noise ratio (SNR) of at least 15 dB and a number of points of at least 5. hour, Set the value to 1.0; 2) If The car behind The corresponding point clusters have a mean signal-to-noise ratio of not less than 10dB and a number of points of not less than 3. hour, Take 0.8; like The car in front The corresponding point clusters have a mean signal-to-noise ratio of not less than 10dB and a number of points of not less than 3. hour, Take 0.8; 3) If The car behind The mean signal-to-noise ratio of the corresponding point clusters is greater than 8dB and less than 10dB or hour, Take 0.6; like The car behind The mean signal-to-noise ratio of the corresponding point clusters is greater than 8dB and less than 10dB or hour, Take 0.6; 4) If the following vehicle If there are interpolated trajectory points that can be repaired, then... Set to 0; If the car in front If there are interpolated trajectory points that can be repaired, then... Take 0.

10. The method for identifying traffic conflict events at intersections based on 4D millimeter-wave radar according to claim 9, characterized in that: Step six involves classifying the severity of conflicts based on the comprehensive risk index value for each vehicle, combined with a preset threshold range, to obtain the conflict severity assessment result. The specific process is as follows: The risk level is divided into four levels, and three thresholds are set: the risk of concern threshold, the potential conflict threshold, and the serious conflict threshold. When a vehicle's comprehensive risk index value is less than three thresholds simultaneously, the risk level is 0. When a vehicle’s comprehensive risk index value is greater than the risk threshold but less than the potential conflict threshold and the serious conflict threshold, the risk level is Level 1. When the vehicle's comprehensive risk index value is greater than the potential conflict threshold but less than the severe conflict threshold, the risk level is level 2. When the vehicle's comprehensive risk index value is greater than the severe conflict risk threshold, the risk level is level 3; The thresholds for risk of concern are set at 0.25, for potential conflict at 0.50, and for serious conflict at 0.75.

Images