A method for judging and evaluating lane changing and cutting in at an intersection

Abstract

The application discloses a kind of intersection motor vehicle lane changing and safety evaluation method, it is related to the field of intelligent transportation, by collecting the historical and real-time driving trajectory data of intersection internal motor vehicle, the typical driving path of different import road to corresponding export road is analyzed and extracted, the virtual lane center line reference system conforming to actual traffic characteristics is constructed;Combined with the lateral displacement variation characteristics of vehicle in the preset observation area, whether the vehicle has lane changing behavior is judged;Traffic safety risk assessment model is constructed, according to the identified type of rush, the weight of each index is adjusted, the comprehensive risk value is calculated, and the evaluation of intersection traffic operation safety level is realized.Solve the problem of lacking effective trajectory reference benchmark in intersection without lane line, the problem of complex rush behavior form in intersection, the problem that driving intention is difficult to distinguish, the problem that traditional safety evaluation method uses fixed weight and is difficult to reflect the risk difference of different rush modes.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation, specifically relating to a method for judging and assessing the safety of vehicles changing lanes and overtaking at intersections. Background Technology

[0002] With the gradual development of intelligent connected vehicle technology and vehicle-road cooperative systems, intersection traffic organization is shifting from traditional static management to a refined control method based on perception and computation. At the same time, the continuous increase in the number of motor vehicles and the increasing complexity of road structures and traffic participation factors have made traffic conflicts within intersections, especially under conditions without clear lane line constraints, increasingly prominent, becoming one of the important factors affecting urban road traffic safety. Existing roadside perception and traffic safety assessment methods struggle to stably and accurately identify and quantify abnormal behaviors such as lane changes and overtaking within complex intersection scenarios, especially when vehicle trajectories are intertwined and traffic directions are unclear, making it difficult to effectively characterize related risk features.

[0003] Currently, reference lines generated based on pure geometric planning have low fitting accuracy and are difficult to accurately represent vehicle travel paths. Existing technologies typically employ geometric topology-based planning methods (such as third-order Bézier curves) to construct virtual guide lines in intersection scenarios lacking lane lines. However, this method has significant drawbacks: (1) The trajectory is out of sync with actual driving habits (low fitting accuracy): Geometric programming relies solely on the physical positions (pstart, pend) and tangential vectors of the entrance and exit lanes. , Mathematical interpolation is used to generate smooth curves. However, these idealized mathematical curves cannot reflect the micro-behavioral characteristics of drivers in actual traffic (such as "natural trajectories" like turning and changing lanes), resulting in a large lateral spatial deviation between the constructed virtual reference line and the actual driving path of most vehicles.

[0004] (2) Large error in risk feature extraction: Due to the inaccuracy of the reference benchmark (virtual lane lines), the lateral offset of the vehicle (d value in the Frenet coordinate system) calculated based on this benchmark has a systematic error. This deviation will directly interfere with the accurate identification of lane-changing and overtaking behavior, causing the system to make false alarms or misses when assessing safety risks.

[0005] In summary, identifying lane-changing and overtaking behaviors of motor vehicles within intersections based on vehicle trajectory monitoring, and quantitatively assessing the risk characteristics of different types of overtaking behaviors, to provide reliable safety situation information for traffic control systems and intelligent vehicles, has become an urgent technical problem to be solved in the field of intersection traffic safety technology. Summary of the Invention

[0006] To address the technical contradictions mentioned in the background art, such as the low fitting degree of reference lines based on geometric programming in intersections lacking clear lane line constraints, the inability to reflect real driving behavior characteristics, and the resulting high false alarm and false alarm rates in safety assessments, the purpose of this invention is to provide a method for identifying and assessing the safety of lane-changing and overtaking behavior at intersections. This method solves the problem of lacking effective trajectory reference benchmarks in intersections without lane lines; it also solves the problem of complex overtaking behaviors and difficulty in distinguishing driving intentions within intersections; and it also addresses the problem that traditional safety assessment methods, which use fixed weights, cannot reflect the risk differences of different overtaking modes.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A method for judging and assessing the safety of vehicles changing lanes and overtaking at intersections includes the following steps: S1: Data acquisition: Real-time trajectory data of target vehicles within the intersection is acquired through roadside sensors, and the trajectory data is smoothed and denoised using the extended Kalman filter algorithm. The vehicle's state vector is then output as historical data storage. S2: The generation of virtual lane lines and virtual lane center lines is based on cluster analysis of historical trajectory data to extract typical driving paths for different flow directions and construct a virtual reference system composed of left virtual lane lines, right virtual lane lines and virtual lane center lines. S3: Determine the observation area and transform the coordinates. Set the observation area bounded by the stop line of the entrance lane and the start line of the exit lane. Map the position coordinates of the target vehicle in the Cartesian coordinate system to the Frenet coordinate system based on the center line of the virtual lane. Decouple the longitudinal displacement and lateral distance. S4: Lane change behavior determination. Based on the characteristics of the lateral distance change of the target vehicle within the observation area, determine whether a lane change behavior has occurred, and classify the lane change behavior into S-type or J-type lane change behavior based on the straightening situation when leaving the observation area. S5: Risk Quantitative Assessment, extracting the time risk index, spatial risk index and energy risk index of the target vehicle at time t, dynamically adjusting the weight of each risk index according to the behavior type determined in S4, and calculating the comprehensive risk value; S6: Traffic scenario safety classification determination, compare the comprehensive risk value with the preset safety classification threshold, output the intersection traffic operation safety level assessment result, and realize the safety level classification.

[0008] Furthermore, the historical data mentioned in step S1 refers to the target vehicle. The state vector at time t , , Here is the vehicle number, and t is the real-time observation time. The coordinates of the center position, For velocity components, The direction angle of the vehicle's travel. and These refer to the vehicle length and width, respectively.

[0009] Furthermore, the specific process of step S2 is as follows: 2.1) Based on the vehicle's entrance lane location, direction of travel, and corresponding exit lane, the historical trajectory data is spatially clustered and grouped to form multiple sets of traffic flow trajectories; 2.2) For each set of traffic flow trajectories, the position distribution of vehicle trajectories in the lateral direction is statistically analyzed along the driving direction. The 5% and 95% quantile points are selected as the left and right boundary points. The left and right virtual lane lines are generated by curve fitting to describe the actual passage range of vehicles within the intersection. 2.3) Based on the left and right virtual lane lines, calculate the geometric center point of the left and right virtual lane lines at the same longitudinal position, and perform smoothing to generate continuous virtual lane center lines as reference benchmarks for vehicle lateral offset and longitudinal position.

[0010] Furthermore, in step S3, the Frenet coordinate system uses the virtual lane centerline as a reference, the longitudinal displacement s is the path length along the centerline, the lateral distance d is the vertical distance of the vehicle's geometric center point relative to the centerline, and d is positive when the vehicle is on the left side of the centerline and negative when it is on the right side.

[0011] Furthermore, the logic for determining lane-changing behavior in step S4 is as follows: The lateral offset d of the vehicle at time t is continuously monitored within the observation area. (t) Record the maximum lateral offset value |d peak |;If|d peak |Exceeds the preset offset threshold D trigger If D is a lane change, then it is determined that the target vehicle has changed lanes; where D is a lane change. trigger Set according to the lane width of the intersection.

[0012] Furthermore, after determining that a lane change has occurred, the logic for distinguishing the type of lane change is as follows: Collect the lateral offset value |d| of the target vehicle at time t when it leaves the observation area. peak | and with the positive threshold δ return Compare; When |d peak |≤δ return If this occurs, it is judged as an S-shaped lane-changing maneuver, that is, overtaking and changing lanes and then returning to the original lane; When |d peak |>δ return If this occurs, it is considered a J-type lane-changing violation, meaning that the driver does not return to the original lane after overtaking or changing lanes.

[0013] Furthermore, it also includes a subdivision step based on the lateral displacement direction: According to d during the lane change process (t) The symbol, if If positive, it is determined to be a lane change on the left; if If the value is negative, it is determined as a lane change to the right; Based on the S-shaped or J-shaped determination results, the act of rushing can be further subdivided into: left-side S-shaped, right-side S-shaped, left-side J-shaped, and right-side J-shaped.

[0014] Furthermore, the time risk index, spatial risk index, and energy risk index mentioned in step S5 are calculated as follows: 1) Time risk indicators Calculated based on the ratio of the longitudinal distance difference between the two vehicles along the centerline of the virtual lane to their speed difference; 2) Spatial risk indicators : Calculated based on the overlap width of the vehicle body projections of adjacent vehicles in the lateral dimension; 3) Energy risk indicators Calculated based on the square of the ratio of the speed difference between the following vehicle and the preceding vehicle to the preset upper limit of relative speed.

[0015] Furthermore, the comprehensive risk value in step S5 Through the formula: ; The weighting adjustment rules for calculating the overall risk value are as follows: When a lane change is identified as a J-shaped maneuver, the time risk weight is increased. Reduce spatial risk weight When the relative velocity difference exceeds a preset threshold, the energy risk weight is locked. The value is high; when the lateral overlap exceeds the collision threshold, the collision is released. Inhibition; When an S-shaped lane change is identified, the spatial risk weight is locked. The value is high, and abnormal lateral interference behavior is judged based on the trend reversal frequency of lateral overlap. When a behavior is unclassified or lacks sufficient confidence, a balanced weighting, i.e., a time risk weighting, is used. Spatial risk weights Energy risk weights Distribute equally.

[0016] Furthermore, the security level classification and determination logic in step S6 is as follows: set a threshold of concern. and critical threshold ; like It is rated as Level 1: Safe; like It is rated as Level 2: Potential Conflict; like It is rated as Level 3: Dangerous overtaking.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention addresses the problem of the lack of effective trajectory reference benchmarks inside intersections without lane lines. By constructing a stable and continuous reference coordinate system based on virtual lanes and virtual lane centerlines generated by historical trajectory clustering, a foundation is provided for the unified description and quantitative analysis of vehicle driving behavior in the absence of physical lane lines.

[0018] 2. This invention addresses the problem of complex forms of lane-changing and lane-changing behaviors at intersections and the difficulty in distinguishing driving intentions. Under the aforementioned reference coordinate system, by analyzing the changes in the lateral displacement of vehicles and using whether the vehicle returns to its original driving path when leaving the intersection as a criterion, lane-changing and lane-changing behaviors are classified into S-shaped behaviors (returning to the original lane) and J-shaped behaviors (not returning to the original lane), thereby achieving effective identification and classification of different lane-changing behaviors.

[0019] 3. In view of the problem that traditional safety assessment methods use fixed weights and are difficult to reflect the risk differences of different rushing modes, this invention introduces a multi-dimensional risk dynamic weighted assessment mechanism based on the type of rushing behavior. The weights of time, space and energy risk indicators are adjusted according to the identified rushing type, and the priority of energy risk is increased when the relative speed difference is large, so that the risk assessment results match the actual rushing behavior characteristics.

[0020] 4. In the context of intersections without lane markings, this invention achieves effective identification of motor vehicle lane-changing and overtaking behavior by constructing a stable reference coordinate system, and completes risk assessment based on this.

[0021] 5. This invention addresses the problem of lack of trajectory reference at intersections without lane markings by dynamically constructing virtual reference lines, enabling vehicle trajectories to be described in a unified coordinate system, thereby ensuring the stability of behavior analysis.

[0022] 6. This invention introduces a discrimination method based on lateral displacement return characteristics to distinguish lane-changing and overtaking behaviors at intersections into S-type and J-type, thereby clarifying the risk characteristics of different overtaking behaviors.

[0023] 7. Differentiate the weighting of time, space, and energy risk indicators according to the type of predatory behavior, so that the risk assessment results can reflect the main sources of risk under different behavior patterns. Attached Figure Description

[0024] Figure 1 This is an overall flowchart of the method of the present invention; Figure 2The trajectory diagrams are for the four preemptive modes (S left / S right / J left / J right) of this invention; Figure 3 This is a comparison chart of the trajectories and Frenet curves for the four preemptive modes (S left / S right / J left / J right) of this invention; Figure 4 This is an original trajectory data diagram illustrating the construction principle of the virtual lane lines and center lines of this invention; Figure 5 This is a trajectory clustering and boundary extraction diagram illustrating the construction principle of virtual lane lines and center lines in this invention; Figure 6 This is a diagram illustrating the construction principle of virtual lane lines and center lines in this invention, showing the generation of virtual lane lines and center lines. Figure 7 This is a schematic diagram of the observation area definition and Frenet coordinate projection of the present invention; Figure 8 This is a flowchart illustrating the logic of dynamically adjusting risk weights based on behavior type in this invention. Figure 9 This is a safety classification judgment curve based on risk field threshold of the present invention.

[0025] The attached figures are labeled as follows: s, longitudinal displacement; d, lateral offset distance; R(t), comprehensive safety risk value; Weighting of time conflict indicators; Weighting of space encroachment indicators; Weighting of energy destruction indicators; T peak The value represents the moment when the comprehensive safety risk value R(t) between the target vehicle and other vehicles in the intersection reaches its highest peak within the intersection observation area. Detailed Implementation

[0026] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0027] A method for judging and assessing the safety of vehicles changing lanes and overtaking at intersections includes the following steps: S1: Data acquisition: Real-time trajectory data of target vehicles within the intersection is acquired through roadside sensors, and the trajectory data is smoothed and denoised using the extended Kalman filter algorithm. The vehicle's state vector is then output as historical data storage. S2: The generation of virtual lane lines and virtual lane center lines is based on cluster analysis of historical trajectory data to extract typical driving paths for different flow directions and construct a virtual reference system composed of left virtual lane lines, right virtual lane lines and virtual lane center lines. S3: Determine the observation area and transform the coordinates. Set the observation area bounded by the stop line of the entrance lane and the start line of the exit lane. Map the position coordinates of the target vehicle in the Cartesian coordinate system to the Frenet coordinate system based on the center line of the virtual lane. Decouple the longitudinal displacement and lateral distance. S4: Lane change behavior determination. Based on the characteristics of the lateral distance change of the target vehicle within the observation area, determine whether a lane change behavior has occurred, and classify the lane change behavior into S-type or J-type lane change behavior based on the straightening situation when leaving the observation area. S5: Risk Quantitative Assessment, extracting the time risk indicators, spatial risk indicators and energy risk indicators of the target vehicle, dynamically adjusting the weight of each risk indicator according to the behavior type determined in S4, and calculating the comprehensive risk value; S6: Traffic scenario safety classification determination, compare the comprehensive risk value with the preset safety classification threshold, output the intersection traffic operation safety level assessment result, and realize the safety level classification.

[0028] The method relies on a roadside integrated sensing system deployed at the intersection. This system includes a roadside sensing system (sensors installed on the roadside acquire real-time vehicle motion state data within the observation area), a roadside computing unit (smoothing the collected raw observation data using an extended Kalman filter algorithm, serving as historical data), and a cloud-based management platform (stores historical data). The various steps of this application are described in detail below: S1 Data Acquisition: Real-time trajectory data of vehicles passing through the intersection is recorded using sensors (e.g., roadside sensors). Then, clustering methods are used to identify the most frequently traveled routes, constructing natural reference lines for vehicle travel within the intersection, and using these as the standard for subsequent safety assessments.

[0029] 1.1 Data Collection and Preliminary Processing ① At each time point t, observe each vehicle and collect data, and establish a state vector for each vehicle i. : ; Where t is the current observation time; i is the unique number of each vehicle; x and y are the center positions of the vehicles in the planar coordinate system; and The speed of the vehicle in different directions; The direction angle of the vehicle's movement; L and W are the specific length and width dimensions of the vehicle, respectively.

[0030] ② Use the Extended Kalman Filter (EKF) algorithm to process the data. This method can filter out interference and noise generated by the sensor.

[0031] S2: Generation of virtual lane lines and virtual lane center lines. This step constructs a virtual reference system for intersection analysis based on historical trajectory data, and mainly includes the following three processes: 2.1 Clustering of trajectory data: Historical vehicle trajectories collected within the intersection are processed. Combining road structure and lane function, the trajectories are spatially clustered and grouped according to the vehicle's entrance lane location, travel direction, and corresponding exit lane, forming several sets of traffic flow trajectories with clear travel directions, providing basic data for virtual lane construction.

[0032] 2.2 Generation of virtual lane lines: For each set of traffic flow trajectories, the position distribution of vehicle trajectories in the lateral direction is statistically analyzed along the driving direction. The 5% and 95% quantiles of the lateral position distribution are selected as the left and right boundaries, and the corresponding left and right virtual lane lines are generated by curve fitting to describe the actual passage range of vehicles within the intersection.

[0033] 2.3 Calculation of virtual lane centerline: Based on the left and right virtual lane lines, the geometric center of the corresponding point at the same longitudinal position is calculated, and the obtained center points are subjected to curve fitting and smoothing to generate a continuous virtual lane centerline, which serves as a reference benchmark for subsequent analysis of vehicle lateral offset and longitudinal position.

[0034] S3: Determine the observation area and transform the coordinates. This step mainly involves defining the global physical boundary of the intersection and mapping the vehicle trajectory to a unified reference frame. The specific implementation is as follows: 3.1 Setting the observation area: To fully capture the traffic and interaction behavior of vehicles within the intersection, the observation range is set as a closed physical area enclosed by the stop lines of the approach lanes and the start lines of the exit lanes in each direction. The boundary of this area is defined as follows (shown in blue): ① Area Scope: A two-dimensional planar area covering the entire traffic space of the intersection is constructed, with the stop line of each approach lane as the starting point of the outer edge and the starting line of each exit lane as the ending point of the outer edge.

[0035] ② Coverage Dimensions: This area fully covers the physical space from the stop line of one side entrance lane to the start line of the opposite exit lane in both the longitudinal and lateral dimensions.

[0036] 3.2 Convert the position coordinates by converting the collected absolute coordinates (x,y) of the vehicle into Frenet coordinates (s,d) to quantify the vehicle's driving status inside the intersection.

[0037] Symbol meaning: s: longitudinal displacement, which refers to the length of the path the vehicle travels along the center line of the virtual lane.

[0038] d: Lateral distance, which refers to the vertical distance between the vehicle's geometric center point and the virtual lane centerline.

[0039] Numerical specification: When the vehicle is to the left of the center line of the virtual lane, d takes a positive value; When the vehicle is to the right of the center line of the virtual lane, d takes a negative value.

[0040] S4: Lane Change Behavior Judgment 4.1 To determine whether a vehicle has changed lanes within the observation area, the vehicle's driving status is continuously monitored after it crosses the stop line of the approach lane. The lateral distance of the vehicle in the Frenet coordinate system is used as the criterion. At each time t, the corresponding lateral offset d(t) of the vehicle is obtained, and its changes are recorded throughout the observation process. This yields the maximum lateral offset value |d| of the vehicle within the observation area. peak |

[0041] ①When |d peak |Exceeds the preset offset threshold D trigger This indicates that the vehicle has deviated from its original driving path and determines that the vehicle has changed lanes within the intersection. ②When |d peak If the threshold is not reached, it means that the vehicle only makes minor lateral adjustments during normal driving and does not change lanes. Wherein, the offset threshold D trigger The setting is based on the geometric width of the intersection lanes (usually half the lane width) to distinguish between real lane changes and trajectory noise or slight correction behavior.

[0042] 4.2 Lane Change Type Determination: After confirming that a vehicle has changed lanes, the type of lane change is further determined. The lane change type is divided into J-type and S-type. J-type: after overtaking and changing lanes, the vehicle does not return to the original lane; S-type: after overtaking, the vehicle returns to the original lane.

[0043] The lateral position of the vehicle when it leaves the observation area (starting line at the intersection exit) is used as the criterion for judgment. The lateral offset value |d| of the vehicle leaving the observation area is obtained through data collection. peak | t And compare this value with the positive threshold δ return Compare them.

[0044] ①When |d peak |≤δ return At that time, it is assumed that the vehicle has completed overtaking and lane changing and then returned to the original lane (S-shaped).

[0045] ②When |d peak |>δ return At that time, it was believed that the vehicle had not returned to its original lane after overtaking and changing lanes (J-type).

[0046] Among them, the positive threshold δ returnA pre-set distance threshold is used to limit the maximum lateral deviation that a vehicle is allowed to make while returning to its original lane.

[0047] 4.3 Subdivision of Lane-Changing Based on Lateral Displacement Direction: After determining the driving process type, the identified lane-changing behaviors are further classified to clarify the specific overtaking direction of the vehicle; the lateral displacement direction of the vehicle within the observation area is used as the classification basis; the overtaking direction of the vehicle is determined based on the sign change of the lateral offset d(t) during the lane-changing process. ① The maximum lateral offset value when a vehicle is changing lanes. If the value is positive, the vehicle will change lanes to the left. ② The maximum lateral offset value when the vehicle is changing lanes. If the value is negative, then the vehicle changed lanes on the right. Based on this, and combined with the driving process type determination results in 4.2, the lane-changing behavior is further subdivided, as shown in the figure: 1) When a vehicle is judged to have engaged in an S-shaped lane-changing maneuver, and the lane change is on the left, the behavior shall be judged as a left-side S-shaped lane-changing maneuver. 2) When a vehicle is judged to have engaged in an S-shaped lane-changing maneuver, and the lane change is on the right, the maneuver will be judged as a right-side S-shaped lane-changing maneuver. 3) When a vehicle is judged to have engaged in a J-shaped lane-changing maneuver, which is a lane change to the left, the behavior shall be judged as a left-side J-shaped lane-changing maneuver. 4) When a vehicle is judged to have engaged in a J-shaped lane-changing behavior and is changing lanes on the right, the behavior shall be judged as a right-side J-shaped lane-changing violation.

[0048] S5: A risk quantification assessment method based on dynamic coupling of spatiotemporal multidimensional indicators and adaptive weighting of behavior. This step is executed by the onboard processor or roadside computing unit, and the processor obtains the following respectively: The virtual lane centerline reference system established by S2, the Frenet coordinate projection data transformed by S3, and the lane change behavior classification results output by S4.

[0049] 5.1 The risk assessment component calculation processor is based on the Frenet coordinate state vector (s, d, v). s v d ), calculate the following normalized risk indicators: Time risk indicators Calculated based on the longitudinal distance and speed difference of the vehicle along the center line of the virtual lane; ; Symbol meanings: t is the current time; s front and s rear It refers to the longitudinal position of the vehicle in front and the vehicle behind on the center line; v front and vrear It is the speed of the two vehicles along the tangential direction.

[0050] Space risk indicators Calculate the lateral projection overlap based on the vehicle's lateral position and body width; ; Symbol explanation: It is the calculated overlap width (meters); d L and d R These are the lateral coordinates of the left and right boundaries of the vehicle body. and These represent the lateral coordinates of the left boundary of the front and rear vehicle bodies, respectively. This indicates the lateral coordinate position of the right boundary (R) of the front vehicle body in the Frenet coordinate system; This represents the lateral coordinate position of the right boundary (R) of the rear vehicle in the Frenet coordinate system. The formula works by finding the intersection of intervals. If the result is 0, it means the two vehicles do not overlap.

[0051] Energy risk indicators Calculate relative kinetic energy risk based on the relative speed difference between vehicles.

[0052] ; Symbol explanation: This indicates the longitudinal speed of the vehicle in front along the center line (tangent) of the virtual lane; This indicates the longitudinal speed of the following vehicle along the center line (tangent) of the virtual lane; It is the speed difference between the following vehicle and the vehicle in front; V limit This is the upper limit of relative speed, preferably 50 km / h (approximately 13.8 m / s). This value represents the physical limit at which a collision at an intersection would result in severe damage.

[0053] 5.2 Dynamic weighted fusion modeling: The processor weights and fuses the aforementioned risk indicators to construct a comprehensive risk model. ; Symbol explanation: R(t) is the final risk score (0 to 1), i.e., the comprehensive risk value; w1, w2, w3 are the weights of time, space, and energy.

[0054] Here, w1, w2, and w3 are the weights of the corresponding risk indicators, and they satisfy the weight normalization constraint. The processor dynamically adjusts the weights based on the lane change behavior type output by S4. When a J-shaped lane change is identified, the spatial risk weight w2 is reduced and the temporal risk weight w1 is increased; when the relative speed difference exceeds a preset threshold, the energy risk weight w3 is locked at a high value; when the lateral overlap exceeds the collision threshold, the suppression of w2 is lifted.

[0055] When an S-shaped lane change is identified, the spatial risk weight w2 is set to a high value, and abnormal lateral interference behavior is judged based on the trend reversal frequency of lateral overlap.

[0056] When a behavior is unclassified or lacks sufficient confidence, a balanced weighting, i.e., a time risk weighting, is used. Spatial risk weights Energy risk weights Distribute equally.

[0057] 5.3 Risk Result Output: The processor outputs the comprehensive risk value R(t), compares it with the preset threshold, generates the risk assessment result, and records it.

[0058] S6: A traffic scenario safety classification method based on extreme value characteristics of the risk field, using a roadside computing unit (RSU) or edge node. It receives the comprehensive risk value R(t) output by S5 and uses a locally preset threshold for comparison and judgment.

[0059] 6.1 Define the grading threshold parameters, load the key grading threshold processor, and pre-set two key thresholds, each corresponding to different driving psychological and physical boundaries: The processor preloads two key thresholds: the attention threshold. and critical threshold .

[0060] Parameter 1: Attention threshold

[0061] Physical meaning: The boundary between "free driving" and "defensive driving". Below this value, the vehicle is in the comfort zone; above this value, it indicates interference and the vehicle needs to enter a state of readiness.

[0062] Value selection method: Based on the statistical distribution of a large amount of normal driving data (usually the 95th percentile is optimal), specifically focusing on the threshold. This analysis is based on a large amount of historical, conflict-free traffic trajectory data from normal intersections, typically using the 90%–98% percentile of the overall risk distribution during normal driving as the standard for assessment. In practical applications, it is recommended to set a higher threshold. The value range is set between [0.3, 0.4].

[0063] Numerical example: In one embodiment, The value is 0.35.

[0064] Parameter 2: Critical threshold

[0065] Physical meaning: The critical point between "controlled collision" and "uncontrolled collision". A value higher than this indicates an extremely high probability of physical collision, requiring immediate irreversible braking or steering.

[0066] Value selection method: ROC curve analysis based on historical accident reconstruction data (selecting the corresponding value when the false negative rate is < 1%), specifically the critical threshold. Based on analysis of extensive historical accident reconstruction data from intersections and trajectory data from extremely high-risk interaction scenarios, the optimal trigger threshold with a false negative rate of less than 1% is typically selected as the standard calibration. In practical applications, it is recommended to use a critical threshold... The value range is set between [0.70, 0.85].

[0067] Numerical Example: In one embodiment, The value is 0.80.

[0068] 6.2 Rating Determination Level 1 (Safe): When .

[0069] Level 2 (Potential Conflict): When .

[0070] Level 3 (Dangerous Crossing): When .

[0071] In summary, this invention collects historical and real-time driving trajectory data of motor vehicles inside intersections using roadside sensors, performs cluster analysis on the trajectories to extract typical driving paths from different approach lanes to corresponding exit lanes, and then constructs a virtual lane centerline reference system that conforms to actual traffic characteristics. The spatial position of vehicles is mapped to the Frenet coordinate system to characterize their lateral and longitudinal motion states. Combining the lateral displacement change characteristics of vehicles within a preset observation area, the invention determines whether a vehicle has engaged in lane-changing behavior based on the maximum lateral offset and the vehicle's straightening behavior upon leaving the intersection. Furthermore, based on differences in straightening characteristics, the invention classifies lane-changing behavior into S-type or J-type lane-changing behaviors. On this basis, a multi-dimensional index characterizing time, space, and kinetic energy characteristics is introduced to construct a traffic safety risk assessment model. The weights of each index are adjusted according to the identified lane-changing type to calculate a comprehensive risk value, thereby achieving an assessment of the traffic safety level of the intersection.

[0072] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A method for judging and assessing the safety of vehicles changing lanes and overtaking at intersections, characterized in that, Includes the following steps: S1: Data acquisition: Real-time trajectory data of target vehicles within the intersection is acquired through roadside sensors, and the trajectory data is smoothed and denoised using the extended Kalman filter algorithm. The vehicle's state vector is then output as historical data storage. S2: The generation of virtual lane lines and virtual lane center lines is based on cluster analysis of historical trajectory data to extract typical driving paths for different flow directions and construct a virtual reference system composed of left virtual lane lines, right virtual lane lines and virtual lane center lines. S3: Determine the observation area and transform the coordinates. Set the observation area bounded by the stop line of the entrance lane and the start line of the exit lane. Map the position coordinates of the target vehicle in the Cartesian coordinate system to the Frenet coordinate system based on the center line of the virtual lane. Decouple the longitudinal displacement and lateral distance. S4: Lane change behavior determination. Based on the characteristics of the lateral distance change of the target vehicle within the observation area, determine whether a lane change behavior has occurred, and classify the lane change behavior into S-type or J-type lane change behavior based on the straightening situation when leaving the observation area. S5: Risk Quantitative Assessment, extracting the time risk index, spatial risk index and energy risk index of the target vehicle at time t, dynamically adjusting the weight of each risk index according to the behavior type determined in S4, and calculating the comprehensive risk value; S6: Traffic scenario safety classification determination, compare the comprehensive risk value with the preset safety classification threshold, output the intersection traffic operation safety level assessment result, and realize the safety level classification.

2. The method for judging and assessing the safety of vehicles changing lanes and overtaking at intersections according to claim 1, characterized in that, The historical data mentioned in step S1 refers to the target vehicle. The state vector at time t , , Here is the vehicle number, and t is the real-time observation time. The coordinates of the center position, For velocity components, The direction angle of the vehicle's travel. and These refer to the vehicle length and width, respectively.

3. The method for judging and assessing the safety of vehicles changing lanes and overtaking at intersections according to claim 1, characterized in that, The specific process of step S2 is as follows: 2.1) Based on the vehicle's entrance lane location, direction of travel, and corresponding exit lane, the historical trajectory data is spatially clustered and grouped to form multiple sets of traffic flow trajectories; 2.2) For each set of traffic flow trajectories, the position distribution of vehicle trajectories in the lateral direction is statistically analyzed along the driving direction. The 5% and 95% quantile points are selected as the left and right boundary points. The left and right virtual lane lines are generated by curve fitting to describe the actual passage range of vehicles within the intersection. 2.3) Based on the left and right virtual lane lines, calculate the geometric center point of the left and right virtual lane lines at the same longitudinal position, and perform smoothing to generate continuous virtual lane center lines as reference benchmarks for vehicle lateral offset and longitudinal position.

4. The method for judging and assessing the safety of vehicles changing lanes and overtaking at intersections according to claim 1, characterized in that, In step S3, the Frenet coordinate system uses the virtual lane centerline as a reference, the longitudinal displacement s is the path length along the centerline, and the lateral distance d is the vertical distance of the vehicle's geometric center point relative to the centerline. When the vehicle is on the left side of the centerline, d is positive, and when it is on the right side, d is negative.

5. The method for judging and assessing the safety of vehicles changing lanes and overtaking at intersections according to claim 1, characterized in that, The logic for determining lane-changing behavior in step S4 is as follows: The lateral offset d of the vehicle at time t is continuously monitored within the observation area. (t) Record the maximum lateral offset value |d peak |;If|d peak |Exceeds the preset offset threshold D trigger If D is a lane change, then it is determined that the target vehicle has changed lanes; where D is a lane change. trigger Set according to the lane width of the intersection.

6. The method for judging and assessing the safety of vehicles changing lanes and overtaking at intersections according to claim 5, characterized in that, After determining that a lane change has occurred, the logic for distinguishing the type of lane change is as follows: Collect the lateral offset value |d| of the target vehicle at time t when it leaves the observation area. peak | and with the positive threshold δ return Compare; When |d peak |≤δ return If this occurs, it is judged as an S-shaped lane-changing maneuver, that is, overtaking and changing lanes and then returning to the original lane; When |d peak |>δ return If this occurs, it is considered a J-type lane-changing violation, meaning that the driver does not return to the original lane after overtaking or changing lanes.

7. The method for judging and assessing the safety of vehicles changing lanes and overtaking at intersections according to claim 6, characterized in that, It also includes a subdivision step based on the lateral displacement direction: According to d during the lane change process (t) The symbol, if If positive, it is determined to be a lane change on the left; if If the value is negative, it is determined as a lane change to the right; Based on the S-shaped or J-shaped determination results, the act of rushing can be further subdivided into: left-side S-shaped, right-side S-shaped, left-side J-shaped, and right-side J-shaped.

8. The method for judging and assessing the safety of vehicles changing lanes and overtaking at intersections according to claim 1, characterized in that, The steps described in step S5 The calculation methods for the temporal risk index, spatial risk index, and energy risk index at any given moment are as follows: 1) Time risk indicators Calculated based on the ratio of the longitudinal distance difference between the two vehicles along the centerline of the virtual lane to their speed difference; 2) Spatial risk indicators : Calculated based on the overlap width of the vehicle body projections of adjacent vehicles in the lateral dimension; 3) Energy risk indicators Calculated based on the square of the ratio of the speed difference between the following vehicle and the preceding vehicle to the preset upper limit of relative speed.

9. A method for judging and assessing the safety of vehicles changing lanes and overtaking at intersections according to claim 8, characterized in that, Comprehensive risk value in step S5 Through the formula: ; The weighting adjustment rules for calculating the overall risk value are as follows: When a lane change is identified as a J-shaped maneuver, the time risk weight is increased. Reduce spatial risk weight ; When the relative velocity difference exceeds a preset threshold, the energy risk weight is locked. The value is high; when the lateral overlap exceeds the collision threshold, the collision is released. Inhibition; When an S-shaped lane change is identified, the spatial risk weight is locked. The value is high, and abnormal lateral interference behavior is judged based on the trend reversal frequency of lateral overlap. When a behavior is unclassified or lacks sufficient confidence, a balanced weighting, i.e., a time risk weighting, is used. Spatial risk weights Energy risk weights Distribute equally.

10. The method for judging and assessing the safety of vehicles changing lanes and overtaking at intersections according to claim 1, characterized in that, The security level classification and determination logic in step S6 is as follows: Set the attention threshold. and critical threshold ; like It is rated as Level 1: Safe; like It is rated as Level 2: Potential Conflict; like It is rated as Level 3: Dangerous overtaking.

Images