A road alignment safety evaluation analysis optimization method

Abstract

The present application belongs to the technical field of road alignment optimization evaluation, and specifically discloses a road alignment safety evaluation analysis optimization method, which captures the vehicle trajectories of traffic participants passing through the geometric alignment section in road operation, and identifies trajectory abnormal events and risk conflict events by using the deviation of the vehicle trajectories from the standard trajectories and the spatiotemporal proximity of the trajectories between the traffic participants. Meanwhile, the cause-effect correlation between the trajectory abnormal events and the risk conflict events is established, the risk inducing position is located, the local alignment parameters are extracted in combination with the high-precision map, the risk inducing defect type is inferred according to the diagnosis rule base, and the real traffic behavior data driven road alignment optimization is realized. The present application breaks through the hysteresis of the traditional design simulation or accident statistics, has the advantages of dynamic response, accurate attribution, closed-loop verification, etc., and improves the scientificity and effectiveness of the optimization.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of road alignment optimization evaluation, and specifically discloses a road alignment safety evaluation analysis optimization method. BACKGROUND

[0002] With the continuous improvement of the transportation network, the public's demand for travel safety, comfort and efficiency is constantly increasing, and the traditional road design mode of reaching the standard as soon as it is built has been difficult to meet the demand. According to statistics, a large number of traffic accidents are related to line design defects, such as small radius curves, steep slope curves, insufficient sight distance, etc., which are easy to cause vehicle deviation, rear-end accidents, etc., so that the driving behavior is easily induced by the line and improper. Therefore, carrying out scientific and precise road alignment optimization and safety evaluation has become a key path to improve the level of road essential safety.

[0003] There are technical solutions for road alignment design and evaluation in the prior art. For example, a kind of road alignment design scheme comprehensive evaluation and optimization method based on BIM is proposed in Chinese patent CN116186858A, which establishes a parameterized BIM model and combines it with a vehicle driving simulation module to simulate, evaluate and iteratively optimize the safety, comfort and economy of the road alignment in the design stage. This kind of method realizes the visualization, parameterization and quantitative evaluation of the design process to a certain extent, and improves the quality and efficiency of the design stage.

[0004] However, when traffic accidents occur frequently due to line defects in the built and operated road, specific line defect diagnosis and analysis need to be carried out on the existing road, and the road at the specific line defect position needs to be reconstructed or re-constructed and handled. In the optimization of road alignment, this kind of method based on BIM simulation still belongs to the category of pre-evaluation in essence, and its evaluation results depend on the pre-set, ideal simulation model and parameters such as standard vehicle model, constant design speed, etc., which leads to the following limitations: (1) The simulation environment is difficult to completely reproduce the complex traffic flow state and diversified driver behavior in the real world, so the simulation evaluation result may be significantly different from the actual operation performance of the road.

[0005] (2) Focus on optimization in the design stage, once the road is built and opened to traffic, its model is relatively fixed and cannot be continuously evaluated and optimized based on the real operation state of the line using the massive real-time data generated during the operation period.

[0006] (3) Although the simulation can output the dynamic response of the vehicle, these responses are idealized results under the preset conditions of the model, and it is difficult to accurately establish the quantitative causal chain between the specific road alignment defects and the specific dangerous driving behaviors caused thereby, such as abnormal trajectories and subsequent actual risks. Due to the breakage of this causal chain, the optimization process can only rely on the experience cycle of hypothesis-simulation-adjustment, rather than data-driven decision-making of observation-attribution-targeted intervention, resulting in ambiguous or even biased optimization direction, which is difficult to target to solve the root problem. SUMMARY

[0007] Therefore, one purpose of the embodiments of the present application is to provide a road alignment safety evaluation analysis optimization method capable of fusing real operation data, revealing behavior-risk causal mechanisms, and realizing accurate diagnosis of alignment defects, effectively solving the problems mentioned in the background art.

[0008] The purpose of the present application can be achieved by the following technical solutions: a road alignment safety evaluation analysis optimization method, comprising the following steps: step 1: collecting real-time space-time coordinate data of traffic participants from a multi-source sensor network deployed on the entire road to generate high-precision vehicle trajectories, and simultaneously simulating and generating a theoretical driving trajectory of a vehicle at a specific road location based on road design alignment parameters and vehicle dynamics characteristics as a standard trajectory.

[0009] Step 2: Calculate the lateral offset and heading angle deviation of the vehicle trajectory and the standard trajectory at each sampling point to generate a trajectory deviation index, based on which to identify trajectory anomaly events and record the occurrence time, location, and involved vehicles of the trajectory anomaly events.

[0010] Step 3: Calculate the space-time proximity index between traffic participants based on the vehicle trajectory, based on which to identify risk conflict events and record the occurrence time, location, and involved vehicles of the risk conflict events.

[0011] Step 4: Establish a space-time causal association model of trajectory anomaly events and risk conflict events, and identify risk inducing locations based on the causal association results.

[0012] Step 5: Spatially superimpose the risk inducing locations and the road alignment digital map, extract the road alignment parameters around the risk inducing locations, infer the road alignment defect type causing the risk based on a pre-defined diagnosis rule library, and optimize the road according to the road alignment defect type.

[0013] In combination with all the technical solutions described above, the application has the following positive effects: 1. Based on the vehicle trajectory data during the road operation period, the application identifies the causal relationship between the trajectory abnormal events and the risk conflict events, locates the risk inducing position, extracts the local linear parameters in combination with the high-precision map, and infers the risk-causing defect type according to the diagnostic rule base, thereby realizing the road linear optimization driven by real traffic behavior data, breaking through the lag of traditional design simulation or accident statistics, and having the advantages of dynamic response, accurate attribution, and closed-loop verification, thereby improving the scientificity and effectiveness of optimization.

[0014] 2. The application identifies the risk inducing position and infers the cause of the linear defect type, realizes accurate positioning and defect classification of the section to be optimized, and generates targeted optimization suggestions for the root risk mechanism, compared with the traditional experience-based optimization, the method can clearly point to specific defects, significantly improve the effectiveness of the treatment measures, avoid resource mismatch, and realize the precise road safety management from passive response to active prevention. BRIEF DESCRIPTION OF DRAWINGS

[0015] The application will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of the following drawings.

[0016] Figure 1 The method embodiment of the application is shown in the following steps.

[0017] Figure 2 The determination diagram of the occurrence time and position of the trajectory abnormal event in the application is shown.

[0018] Figure 3 The identification and linear defect inference implementation diagram of the risk inducing position in the application is shown. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0020] Referring to Figure 1 The application proposes a road linear safety evaluation analysis and optimization method, including the following steps: step 1: real-time collection of the space-time coordinate data of traffic participants from the multi-source sensor network deployed on the whole road to generate high-precision vehicle trajectories, and generation of the theoretical driving trajectory of the vehicle at a specific road position as a standard trajectory based on the road design linear parameters and vehicle dynamics characteristics simulation.

[0021] As an optional implementation of the above scheme, the high-precision vehicle trajectory is generated as follows: millimeter wave radar and video monitoring equipment are arranged to form a sensor network along the geometric linear section of the road.

[0022] As an application of the above scheme, the geometric linear section refers to the key section of the road where the driving behavior is prone to instability due to complex combination of horizontal and vertical lines or sudden changes in operating environment, including small-radius horizontal curves, large vertical slopes, limited-sight vertical curve sections, linear mutation points, interchange ramp merging / outgoing sections, and tunnel entrances and exits.

[0023] The selection of the above section as the focus of sensor deployment is based on the consensus of road safety design theories such as "Technical Standards for Highway Engineering" and "Specifications for Highway Route Design": these sections are often the cause of trajectory deviation and conflict events due to high driving complexity and high probability of human error. By preferentially deploying sensing devices in these sections, targeted monitoring of potential safety risks can be achieved, improving system diagnosis efficiency and optimization relevance.

[0024] The motion point cloud data of the traffic participants passing through the geometric linear section is obtained by millimeter wave radar, which includes the spatial position and motion speed vector of the traffic participants, reflecting their dynamic behavior in the road environment, and is the basic observation data for vehicle trajectory formation.

[0025] The vehicle visual feature data of the traffic participants passing through the geometric linear section is obtained by video monitoring equipment, which includes the appearance contour, color texture, and vehicle model contour, reflecting the static appearance attributes and category features of the target, and can be used for precise vehicle identity association to enhance the tracking continuity of target recognition.

[0026] The motion point cloud data and visual feature data are time-synchronized and spatially registered.

[0027] As can be understood, the observation data of different sensors is in their respective local coordinate systems and needs to be geometrically aligned under a unified geographical reference through spatial calibration and coordinate transformation; at the same time, there is clock drift and sampling asynchrony in data collection, which must be time-synchronized to ensure strict alignment and consistent association of point cloud and visual data in the space-time dimension, ensuring the accuracy and reliability of multi-source information fusion.

[0028] The registered data is applied to generate a continuous traffic participant motion trajectory sequence by associating the continuous observation values of the same target's motion point cloud data through target tracking.

[0029] The generated traffic participant motion trajectory sequence is applied to reconstruct a high-precision vehicle trajectory that is continuous in space and time by using trajectory smoothing and interpolation algorithms.

[0030] It should be noted that the original motion trajectory is often affected by sensor noise, multipath effect or temporary target occlusion, resulting in data jitter or intermittent sampling, which affects the continuity and accuracy of the trajectory. Trajectory smoothing and interpolation techniques can be used to restore the true motion path, aiming to improve the geometric rationality and spatio-temporal continuity of the trajectory, and provide a reliable data basis for subsequent trajectory analysis and risk identification.

[0031] As a further optional implementation of the above scheme, the specific generation process of the standard trajectory is as follows: extracting planar linear feature point coordinates, longitudinal section slope change point coordinates and transverse section lane width data according to road design drawings.

[0032] In the example of the above operation, the planar linear feature points are used to construct the geometric shape of the road, including the start and end points of the straight line segment, the start and end points of the easement curve, and the center and end points of the circular curve. These points define the lateral path of the vehicle.

[0033] The longitudinal section slope change points include the slope start, slope end and slope change points, which determine the slope change of the vehicle in the longitudinal direction.

[0034] The transverse section lane width data describes the specific width and layout of the lane, which affects the operation space of the vehicle and the behavior decision of the driver.

[0035] It should be noted that accurate road geometric information is the basis for establishing high-fidelity vehicle driving simulation. Only by accurately describing the planar, longitudinal and transverse characteristics of the road can the dynamic response of the vehicle in complex road conditions be truly reproduced, and the generated standard trajectory can have high authenticity and representativeness.

[0036] Simulate the theoretical driving path of the vehicle through the geometric linear section based on the vehicle dynamics and design speed as the standard trajectory.

[0037] It should be understood that the vehicle dynamics refers to the physical model that characterizes the motion response and handling performance of the vehicle under different driving conditions, covering longitudinal, lateral and vertical dynamics. By integrating road geometric information and vehicle dynamics, the theoretical driving path of the vehicle can be simulated under the given design speed. This method belongs to the mature technology in the fields of traffic engineering and intelligent driving, and its technical details have been well developed, which will not be described here.

[0038] The standard trajectory obtained by the above simulation is not simply following the road center line or lane line, but is a comprehensive product of road geometric design specifications and vehicle dynamics response characteristics, reflecting the optimal driving path expectation under the human-vehicle-road cooperation. Its generation process fully considers the continuity of the linear shape, driving comfort and safety boundary, ensuring its scientificity and rationality as an evaluation of the deviation degree of the actual trajectory.

[0039] Step 2: Calculate the lateral offset and heading angle deviation of the vehicle trajectory and the standard trajectory at each sampling point to generate the trajectory deviation index, by which the trajectory anomaly event is identified, and the time and location of the trajectory anomaly event are recorded.

[0040] As a way to achieve the above scheme, the trajectory anomaly event identification is specifically operated as follows: the vehicle trajectory of the traffic participant is sampled at equal intervals at a fixed interval to obtain a discrete trajectory sampling point sequence, and for each sampling point, the corresponding reference position is matched on the standard trajectory by nearest neighbor search, thereby calculating the lateral offset and heading angle deviation of the vehicle trajectory and the standard trajectory.

[0041] Understandably, actual trajectory data is usually collected by asynchronous sensors such as radar and video, with a non-fixed sampling frequency, clock drift or data packet loss, resulting in uneven time intervals. Equal-interval resampling converts the original non-uniform trajectory into a time-aligned discrete sequence, meeting the basic requirements of subsequent sliding window analysis algorithms for equal-time input. When matching the standard trajectory, it is necessary to ensure that the actual trajectory point and the standard trajectory reference point are aligned on the same time axis. Equal-interval sampling provides a unified time reference, facilitating nearest neighbor matching, time domain alignment, and cross-sequence association.

[0042] Further, based on the geometric projection principle and path tracking theory, the actual trajectory is regarded as a deviation process from the standard trajectory, spatial alignment is achieved by nearest neighbor search, the corresponding reference point of each point on the standard trajectory is obtained, and basic data is provided for subsequent construction of trajectory deviation indicators, so that the deviation is transformed from qualitative description to calculable and comparable numerical indicators.

[0043] The lateral offset in the above is the projection distance of the actual position relative to the normal direction of the standard trajectory, and the specific expression is wherein represents the lateral offset of the i-th sampling point, represents the unit normal vector of the standard trajectory at the i-th sampling point, , respectively represent the actual position of the i-th sampling point and its projection reference position on the standard trajectory, represents the sampling point number.

[0044] The heading angle deviation is the angle difference between the actual heading angle and the tangent direction of the standard trajectory.

[0045] It should be noted that under ideal road conditions, the vehicle should run smoothly along the designed expected path, and its spatial position and motion direction should be highly consistent with the standard trajectory. Significant deviation indicates that there is a driving behavior disturbance or insufficient linear guidance. Among them, the lateral offset reflects the degree to which the vehicle deviates from the designed path in the lateral direction, and embodies the path keeping ability; the heading angle deviation reflects the matching of the vehicle driving direction and the road direction, and embodies the direction compliance. Both of them jointly quantify the deviation degree of the actual driving state from the design intention, and constitute the basis input of micro driving behavior anomaly recognition.

[0046] The continuous lateral offset sequence and the heading angle deviation sequence are filtered and normalized, and a trajectory deviation index is generated by weighted fusion.

[0047] The above filtering of the lateral offset sequence and the heading angle deviation sequence is to suppress noise and transient jitter, and the normalization of the two sequences is to map them to a unified dimension interval to eliminate the difference in magnitude.

[0048] Further, the normalized lateral offset sequence and the heading angle deviation sequence are weighted and fused to generate a trajectory deviation index, aiming to overcome the limitations of a single-dimensional index in describing complex driving behavior. The cooperative fusion of lateral offset and heading angle deviation can more comprehensively capture the driving behavior instability characteristics.

[0049] Specifically, the weights of the lateral offset and the heading angle deviation can be calculated by principal component analysis to obtain the contribution rate of the two types of deviations in the total variance, and the contribution rate is used as the fusion weight, reflecting the relative importance of the two types of deviations in representing the driving behavior variation characteristics.

[0050] Synchronously collect cross-section traffic flow data in road geometric linear sections, extract cross-section average speed, draw traffic flow state curve of cross-section average speed evolution over time, and extract traffic flow state change rate sequence by first-order differential operation on the curve.

[0051] It can be explained that the cross-section average speed reflects the overall traffic efficiency and traffic flow running state of a specific road section per unit time, and is an important macroscopic index for measuring traffic flow state stability. The traffic flow state curve drawn based on this data describes the dynamic process of traffic flow evolution over time, and further uses the first derivative of the cross-section average speed to extract the traffic flow state change rate as a dynamic disturbance representation at the group level.

[0052] The trajectory deviation sequence and the traffic flow state change rate sequence are time-domain aligned, and the two sequences are segmented by a sliding time window. The local covariance of the two sequences at each center time is calculated as the trajectory-flow line coupling index.

[0053] Continuing to be understandable, the local covariance of the trajectory deviation sequence and the traffic flow state change rate sequence in the sliding time window is used as the trajectory-flow coupling index, aiming to quantify the phased dynamic synergy between individual trajectory anomalies and group traffic flow disturbances. The index reflects the degree of synchronization and the consistency of the direction of the change trend of the two in the local period. The greater the absolute value of the coupling index, the higher the synchronization of the individual trajectory deviation and the group speed fluctuation in time, the change direction is coordinated, and there is a significant statistical correlation. Such strong coupling characteristics suggest that the abnormal event is not an isolated random driving behavior, but has an inherent relationship with the evolution of the macro traffic flow state, which may be induced by common external factors such as road alignment defects, deteriorating visibility conditions, and then trigger the coordinated response of individuals and groups.

[0054] The trajectory-flow coupling index corresponding to the center time of each sliding time window is compared with the preset reference range, and the trajectory deviation sequence is compared with the allowable deviation threshold. If any of the following conditions is met, it is determined that a trajectory anomaly event occurs: Condition A: The trajectory deviation index continuously exceeds the allowable deviation threshold for a safe time length.

[0055] The allowable deviation threshold in the above reflects the maximum trajectory deviation level that a vehicle can accept in a normal driving state. It can be referred to the provisions of the "Highway Alignment Design Specification" on lane keeping, visibility requirements, and lateral acceleration limits to deduce the allowable lateral offset and heading angle deviation threshold; The safe time length refers to the minimum time that the abnormal state needs to exist, which is used to exclude false positives caused by transient disturbances or measurement noise, and is usually set to 2-3 seconds.

[0056] Condition B: The trajectory-flow coupling index deviates from the reference range.

[0057] The reference range of the coupling index in the above represents the natural fluctuation level between individual trajectories and group flow states under normal traffic operation conditions. The time series distribution of the trajectory-flow coupling index can be statistically analyzed based on historical operation data under free flow or stable flow conditions, and the reference range can be constructed using the mean value ± 2 times the standard deviation.

[0058] The trajectory anomaly event discrimination mechanism constructed in the present application serves as the abnormal perception center of the road alignment optimization and safety evaluation system. The core function is to convert high-precision vehicle trajectory data into structured event information with physical meaning and safety semantics, providing high-confidence input basis for subsequent road alignment defect identification.

[0059] In the judgment process, the micro-behavior deviation and the macro-flow coupling are fused, breaking through the limitations of traditional methods that rely only on individual behavior threshold discrimination, and can effectively identify those implicit risk sources that do not directly lead to traffic conflicts but have caused group flow disturbances, significantly improving the robustness and fault tolerance of anomaly detection.

[0060] Referring to Figure 2 As shown, as a further implementation mode capable of being realized by the above scheme, after identifying the trajectory abnormal event, the following operation is performed: when the trajectory abnormal event is determined to occur by condition A, the first sampling point satisfying the trajectory deviation degree higher than the allowed deviation threshold is searched in the trajectory sampling sequence of the traffic participant, and the timestamp corresponding to the sampling point is taken as the occurrence time of the trajectory abnormal event.

[0061] This time marks the critical transition of individual behavior from normal to deviated state, and has clear behavior starting semantics.

[0062] The spatial position of the sampling point in the vehicle trajectory is taken as the occurrence position of the trajectory abnormal event.

[0063] When the trajectory abnormal event is determined to occur by condition B, the time corresponding to the maximum trajectory-flow coupling index in the time set in which all trajectory-flow coupling indexes deviate from the reference range is taken as the occurrence time of the trajectory abnormal event.

[0064] This time represents the strongest transient coupling of individual behavior and group flow state, which is the peak point of system disturbance.

[0065] The spatial position of the traffic participant in the vehicle trajectory at this time is taken as the occurrence position of the trajectory abnormal event.

[0066] When the trajectory abnormal event is determined to occur by both condition A and condition B, the time and position determined by condition A are taken as the occurrence time and occurrence position of the trajectory abnormal event.

[0067] It should be noted that when the trajectory abnormal event satisfies both condition A and condition B, it indicates that the event has the dual characteristics of individual sustained deviation and strong coupling of group flow state, and the time and position determined by condition A are taken as the occurrence time and occurrence position of the trajectory abnormal event, because condition A captures the starting point of individual behavior instability, representing the cause of abnormal behavior, and condition B reflects the response coupling of the abnormality and the group flow state, which belongs to the effect or the collaborative evolution stage. From the cause-effect logic, the initial triggering time of the abnormal behavior should be taken as the event occurrence time, which conforms to the time sequence logic that behavior precedes influence.

[0068] The different occurrence time and occurrence location determination strategies for different determination conditions are due to the essential differences in the cause mechanism and dynamic characteristics of the two types of trajectory anomaly events. Condition A reflects the starting point of individual continuous deviation behavior, and the first time the limit is exceeded is used as the trigger benchmark for spatial behavior instability. Condition B reflects the strong coupling disturbance between the individual and the group flow state, and the peak coupling strength time is used as the representative time point of systematic anomaly. This conditional positioning mechanism ensures the clear physical meaning and causal traceability of the event space-time anchor point, and its purpose is to provide reliable and interpretable input basis for the subsequent spatio-temporal causal correlation analysis between trajectory anomaly events and risk conflict events.

[0069] Step 3: Calculate the spatio-temporal proximity index between traffic participants based on vehicle trajectories, identify risk conflict events, and record the occurrence time, occurrence location, and involved vehicles of the risk conflict events.

[0070] As a preferred implementation of the above scheme, the risk conflict event is specifically identified as follows: For any pair of traffic participants, the trajectory sampling points obtained by equally sampling the actual vehicle trajectories of the two participants are time-aligned for each sampling point, ensuring that the state information of both participants at each time comes from the same time reference, forming a series of synchronized sampling point groups as the basic unit for subsequent interaction analysis.

[0071] For each sampling point group, the spatial position and instantaneous velocity vector at the corresponding time are extracted from their respective trajectories to predict their motion paths in the future time domain. The minimum Euclidean distance between the two predicted trajectories in the corresponding time domain is calculated as the spatio-temporal proximity index.

[0072] It should be understood that vehicle motion has inertial characteristics, and in a short-term time domain, usually ≤3s, the future trajectory can be extrapolated based on the current position and velocity vector with constant speed, which conforms to the short-term dynamic evolution law. Compared with the method of only calculating the instantaneous Euclidean distance between sampling points, this method can identify potential conflicts by searching for the minimum distance between predicted trajectories in the time domain, overcoming the limitation that instantaneous distance cannot reflect dynamic approaching relationship, significantly improving the predictability and reliability of risk identification.

[0073] The spatio-temporal proximity index of each sampling point group is compared with the preset safety threshold. When the spatio-temporal proximity index is lower than the safety threshold, it is determined that a risk conflict event occurs.

[0074] The safety threshold in the above embodiment represents the minimum acceptable safety distance between traffic participants, which can be specifically referred to the lateral and longitudinal safety spacing recommended in the "Guidelines for Road Safety Evaluation".

[0075] As a further preferred implementation of the above scheme, after identifying the risk conflict event, the trajectory abnormal event occurrence time and occurrence location are recorded as follows: the timestamp of the first spatiotemporal proximity index corresponding to the sampling point group below the safety threshold is taken as the occurrence time of the risk conflict event, and the center point between the spatial positions of the sampling point group in the respective trajectories is taken as the occurrence location of the risk conflict event.

[0076] Understandably, the risk conflict is essentially a spatial intrusion behavior between two bodies, and its occurrence location is not a single vehicle location, but the core of the interaction region formed by the two, and the center point can effectively represent the geographical center of the potential risk region.

[0077] Step 4: Establish a spatiotemporal causal association model between the trajectory abnormal event and the risk conflict event, and then identify the risk inducing location based on the causal association result.

[0078] In the specific implementation of the above scheme, the spatiotemporal causal association model is established as follows: extract all involved vehicles of the recorded trajectory abnormal events and risk conflict events, and match and classify the two types of events with the common involved vehicles as the association key: for any trajectory abnormal event and risk conflict event, if they contain at least one same involved vehicle, the pair of events is classified into the same control event group.

[0079] As applied to the above implementation, each control event group represents whether a vehicle participates in a subsequent risk conflict after a trajectory abnormality occurs, and is a basic unit for causal analysis.

[0080] For each control event group, the occurrence location of the trajectory abnormal event and the occurrence location of the risk conflict event are extracted, and the Euclidean space distance between them is calculated, which reflects the spatial continuity between the abnormal behavior occurrence point and the conflict occurrence point.

[0081] The occurrence time of the trajectory abnormal event and the occurrence time of the risk conflict event are extracted, and the time lag interval between them is calculated, which reflects the time continuity between the abnormal behavior occurrence and the conflict occurrence.

[0082] The Euclidean space distance and the time lag interval of each control event group are introduced into a causal confidence score model , and the causal confidence of each control event group is calculated, wherein represents the causal confidence, , represents the Euclidean space distance and the time lag interval of the control event group, respectively, , represents the maximum space threshold and the maximum allowed time delay, respectively.

[0083] In the optimized implementation of the above scheme, a spatio-temporal rationality constraint is applied to all control event groups before causal confidence calculation: only event pairs with a Euclidean space distance less than or equal to the maximum space threshold and a time lag interval less than or equal to the maximum allowed time delay are retained. This pre-screening step aims to exclude spatially distant or temporally disjoint event combinations, avoid incorporating unrelated trajectory anomalies and conflict events into causal analysis, and thus prevent false associations from interfering with model output, improving the accuracy and computational efficiency of causal inference.

[0084] As an explanation of the causal confidence scoring model described above, introduce and As a physically interpretable spatio-temporal proximity boundary, its value should be set based on traffic operation characteristics, where the maximum space threshold needs to cover the influence range of typical road functional sections such as curves and ramps, and the maximum allowed time delay is generally 5-10 seconds, which conforms to the time scale of driving behavior response and risk evolution.

[0085] By introducing spatio-temporal constraints, , are dimensionless and take values between representing spatial deviation and temporal lag, , The smaller the value, the , greater, indicating that the trajectory anomaly event and the risk conflict event are closer in space and time, and the causal probability is higher. The use of and as the causal confidence is because when either dimension is small, the overall causal confidence is small, reflecting the joint probability characteristics under strong constraints, ensuring that only under the condition of spatio-temporal proximity is high confidence given, which conforms to the causal logic of trajectory anomalies inducing subsequent conflicts.

[0086] As an explanation of the above scheme, consider trajectory anomalies as unsafe driving behavior representations and risk conflicts as near-accident states. By building a causal relationship between the two, it is revealed that risk conflicts do not occur randomly, but can be traced back to the preceding trajectory anomaly event. Trajectory anomalies themselves have a strong correlation with road geometric linear features, so this mechanism provides a logical chain and data support for subsequent causal inference of road linear defects. Compared to the correlation assumption in traditional methods that directly infers linear defects from conflict positions, this method can maximize the avoidance of misjudging occasional conflicts as caused by linear defects by introducing trajectory anomalies as an intermediate variable.

[0087] In further specific implementation of the above scheme, the risk inducing position is identified based on the result of causal association, see the following: comparing the causal confidence of each control event group with the configured causal significance threshold, screening the control event group greater than the causal significance threshold, and marking the occurrence position of the trajectory abnormal event in the control event group as the risk inducing position.

[0088] In the above, the causal significance threshold can be calculated using historical data or control event group samples collected on typical safe road segments, to obtain an empirical distribution of causal confidence, and select a high quantile value such as P90 or P95 as the threshold.

[0089] Step 5: Spatially superimpose the risk inducing position with the high-precision road alignment digital map, extract the road alignment parameters around the risk inducing position, and infer the road alignment defect type causing the risk based on a predefined diagnostic rule base.

[0090] Referring to Figure 3 The above step is implemented as follows: map the risk inducing position to the corresponding coordinates of the high-precision road alignment digital map to achieve spatial alignment of the event and the road infrastructure.

[0091] With each risk inducing position as the center, a circular analysis area of a preset radius is constructed, focusing on the most likely cause of the linear parameter, avoiding the introduction of irrelevant remote information, and improving the diagnostic relevance.

[0092] In the above, the preset radius is generally taken as 50-100m, considering that the driver's response to road alignment usually occurs within a certain distance before and after the defect point.

[0093] In the circular analysis area, detailed design parameters of the road alignment are extracted from the digital map, including the curvature radius of the planar alignment, the slope and vertical curve radius of the longitudinal section alignment, and the lane width and shoulder width of the transverse section alignment.

[0094] It can be understood that in the above, the curvature radius of the planar alignment reflects the geometric characteristics of the curve, a small radius is easy to cause lateral deviation, the slope and vertical curve radius of the longitudinal section alignment affect the vehicle acceleration and deceleration behavior and the sight distance condition, steep slope or convex vertical curve top is easy to cause speed out of control, and the lane width and shoulder width of the transverse section alignment determine the vehicle operating space, narrow lane or no hard shoulder increases the driving trajectory fluctuation.

[0095] The extracted road alignment design parameters are associated with the recorded trajectory abnormal event characteristics at the risk inducing position, and the linear defect type is inferred according to the predefined diagnostic rule base.

[0096] Applying to the above operation, the diagnostic rule base includes the following: when the risk inducing position is located in the area of the plane curve, and the recorded trajectory anomaly event characteristics are that the curvature radius of the vehicle trajectory is greater than or less than the design curvature radius of the road, it is diagnosed as a plane linear curvature design unreasonable defect.

[0097] When the risk inducing position is located at the vertical slope change point or the top of the convex vertical curve area, and the recorded trajectory anomaly event characteristics are that the vehicle speed abnormally fluctuates or the trajectory frequently corrects laterally, it is diagnosed as a vertical section linear sight distance deficiency or slope mutation defect.

[0098] When the risk inducing position is located within the lane range, and the recorded trajectory anomaly event characteristics are that the vehicle trajectory deviates laterally relative to the lane center line, it is diagnosed as a transverse section linear lane width deficiency.

[0099] It can be understood that the diagnostic rule base is based on the technical limits and safety design principles in the Highway Route Design Specification, and constructs a structured knowledge graph by integrating knowledge in the field of traffic engineering, which semantically associates road linear parameters, driving behavior responses, and safety risk patterns. By establishing an abnormal symptom-cause mechanism causal pattern matching mechanism, the explainable attribution analysis from observed phenomena to design defects is realized, providing engineering-based, targeted technical support for road linear optimization.

[0100] It is further pointed out that the above-mentioned linear defect diagnosis based on the rule base is a knowledge-driven static inference, which still needs to be combined with the reproduction of the driving process on this section through real vehicle tests or driving simulators to observe the driver's behavior response and vehicle dynamic performance, and to verify whether the inferred defect truly causes abnormalities, in order to improve the accuracy and reliability of the diagnosis.

[0101] Further, after inferring the type of road linear defect, the road linear optimization is performed according to the type of road linear defect.

[0102] In the specific implementation of the above scheme, when the type of road linear defect of the risk inducing position is a plane linear curvature design unreasonable defect, the specific optimization measures are to redesign the curve when the space conditions permit, to increase the curvature radius to above the minimum limit radius or the recommended radius corresponding to the design speed, to reduce the lateral acceleration; or to insert a sufficient length of transition curve between the straight line and the circular curve to realize the smooth transition of the curvature from zero to the target value, and to avoid the sudden change of direction.

[0103] When the road line defect type of the risk inducing position is the longitudinal section line shape sight distance deficiency or slope mutation defect, the optimization measure is to enlarge the convex or concave vertical curve, recalculate the required minimum radius according to the design speed, and ensure that the parking sight distance requirement is met;Adjust the horizontal and vertical combination, follow the principle of horizontal package and vertical original, ensure that the horizontal curve covers the longitudinal slope point, and prevent visual and operation double conflict.

[0104] When the road line defect type of the risk inducing position is the horizontal section line shape lane width deficiency defect, the optimization measure is to increase the lane width to the specification requirement value;Add hard shoulder or adjust the transverse slope.

[0105] The present application analyzes and diagnoses the vehicle trajectory of the traffic participants on the road in real time, outputs the road line defect type of the risk inducing position, optimizes the road line, and the essence is active safety decision based on real-time data, although the immediate form of the physical road is not changed, but the real-time generation and release of risk identification and optimization suggestions are realized, the immediate, accurate and operable line improvement basis is provided for the road operation unit, the response cycle of traditional accident statistics-manual investigation-design reconstruction is significantly shortened, and the foresight and initiative of the intelligent transportation system in safety governance is embodied.

[0106] The parameters involved in the above formula are de-dimensioned to calculate their numerical values, the formula is obtained by collecting a large amount of data to simulate the nearest real situation, and the preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0107] The above embodiments can be realized wholly or partially by software, hardware, firmware or any combination thereof. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product.

[0108] Those skilled in the art can realize that the modules and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0109] In addition, the functional modules in each embodiment of the present application can be integrated in one processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0110] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0111] Finally, the above merely provides the preferred embodiments of the present application, but is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for optimizing safety evaluation analysis of road alignment, characterized in that, The application relates to a method for identifying road line defects based on trajectory anomaly events and risk conflict events. The method comprises the following steps: Step 1: collecting real-time space-time coordinate data of traffic participants from a multi-source sensor network deployed along the whole road, and generating high-precision vehicle trajectories; Based on the road design line parameters and the vehicle dynamics characteristics, the theoretical driving trajectory of the vehicle at a specific road location is simulated to generate a standard trajectory; Step 2: calculating the lateral offset and heading angle deviation of the vehicle trajectory and the standard trajectory at each sampling point to generate a trajectory deviation index, which is used to identify trajectory anomaly events and record the occurrence time, location and involved vehicle of the trajectory anomaly events; Step 3: calculating the space-time proximity index between the vehicle trajectories based on the vehicle trajectories to identify risk conflict events and record the occurrence time, location and involved vehicle of the risk conflict events; Step 4: establishing a space-time causal correlation model of the trajectory anomaly events and the risk conflict events, and identifying the risk inducing locations based on the causal correlation results; 2. A method for road alignment safety evaluation analysis optimization as claimed in claim 1 wherein: Step 5: superimposing the risk inducing locations and the road line digital map, extracting the road line parameters around the risk inducing locations, inferring the road line defect types causing the risk based on a pre-defined diagnosis rule base, and optimizing the road according to the road line defect types. The high-precision vehicle trajectory is generated in the following manner: Millimeter wave radars and video monitoring devices are arranged along the geometric line segments of the road to form a sensor network; The motion point cloud data of the traffic participants passing through the geometric line segments are obtained by the millimeter wave radars, and the vehicle visual feature data of the traffic participants passing through the geometric line segments are obtained by the video monitoring devices; The motion point cloud data and the visual feature data are subjected to time synchronization and space registration processing; The registered data are subjected to target tracking to associate the continuous observation values of the motion point cloud data of the same target to generate continuous traffic participant motion trajectory sequences; 3. A method for road alignment safety evaluation analysis optimization as claimed in claim 1 wherein: The generated traffic participant motion trajectory sequences are subjected to trajectory smoothing and interpolation algorithm to reconstruct the high-precision vehicle trajectories. The standard trajectory is generated in the following manner: The planar line feature point coordinates, the longitudinal section slope change point coordinates and the transverse section lane width data are extracted from the road design drawings; 4. The method of claim 1, wherein: The theoretical driving path of the vehicle passing through the high-risk geometric line segment is simulated based on the vehicle dynamics characteristics and the design speed as the standard trajectory. The step 2 comprises the following contents: The vehicle trajectories of the traffic participants are sampled at fixed intervals to obtain a discrete trajectory sampling point sequence; The corresponding reference position on the standard trajectory is matched by the nearest neighbor search for each sampling point, so that the lateral offset and the heading angle deviation of the vehicle trajectory and the standard trajectory are calculated; The continuous lateral offset sequence and the heading angle deviation sequence are subjected to filtering and normalization processing, and the trajectory deviation index is generated by weighted fusion; The cross-section traffic flow data are synchronously collected at the road geometric line segments, the cross-section average speed is extracted, the traffic flow state curve of the cross-section average speed with time evolution is drawn, and the traffic flow state change rate sequence is extracted by performing first-order differential operation on the curve. The trajectory deviation degree sequence and the traffic flow state change rate sequence are time-domain aligned, and the two sequences are segmented by a sliding time window. The local covariance of the two sequences at each central time is calculated as the trajectory-flow coupling index; The trajectory-flow coupling index at the central time of each sliding time window is compared with the preset reference range, and the trajectory deviation degree value sequence is compared with the allowed deviation threshold. If any one of the following conditions is met or both conditions are met, it is determined that a trajectory anomaly event occurs: Condition A: The trajectory deviation degree index continuously exceeds the allowed deviation threshold for a safe time length; Condition B: The trajectory-flow coupling index deviates from the reference range.

5. A method for road alignment safety evaluation analysis optimization as claimed in claim 4 wherein: The occurrence time and location of the trajectory anomaly event are as follows: When the trajectory anomaly event is determined to occur by condition A, the first sampling point in the trajectory sampling sequence of the traffic participant that satisfies the trajectory deviation degree higher than the allowed deviation threshold is found. The timestamp of the sampling point is taken as the occurrence time of the trajectory anomaly event, and the spatial position of the sampling point in the vehicle trajectory is taken as the occurrence location of the trajectory anomaly event; When the trajectory anomaly event is determined to occur by condition B, the maximum trajectory-flow coupling index corresponding to the time in the set of times when the trajectory-flow coupling index deviates from the reference range is taken as the occurrence time of the trajectory anomaly event. The spatial position of the traffic participant in the vehicle trajectory at the time is taken as the occurrence location of the trajectory anomaly event; When the trajectory anomaly event is determined to occur by both condition A and condition B, the time and location determined by condition A are taken as the occurrence time and occurrence location of the trajectory anomaly event.

6. A method for road alignment safety evaluation analysis optimization as claimed in claim 4 wherein: The step 3 is as follows: For any pair of traffic participants, the trajectory sampling point sequences obtained by equally sampling the trajectories are time-aligned, and each sampling point in the sequences is time-aligned to form a sampling point group; For each sampling point group, the spatial position and instantaneous speed vector at the corresponding time are extracted from the respective trajectories to predict the motion path in the future time domain. The minimum Euclidean distance between the two predicted trajectories in the corresponding time domain is calculated as the spatiotemporal proximity index; The spatiotemporal proximity index of each sampling point group is compared with the preset safety threshold. When the spatiotemporal proximity index is lower than the safety threshold, it is determined that a risk conflict event occurs.

7. A method for road alignment safety evaluation analysis optimization as claimed in claim 6 wherein: The occurrence time and location of the risk conflict event are as follows: The timestamp of the sampling point group corresponding to the first spatiotemporal proximity index lower than the safety threshold is taken as the occurrence time of the risk conflict event, and the center point between the spatial positions in the respective trajectories of the sampling point group is taken as the occurrence location of the risk conflict event.

8. The method of road alignment safety evaluation analysis optimization as claimed in claim 1 wherein: The spatiotemporal causal association model is established as follows: All involved vehicles of the recorded trajectory anomaly events and risk conflict events are extracted. The two types of events are matched and classified using the common involved vehicles as the association key: for any trajectory anomaly event and risk conflict event, if they contain at least one same involved vehicle, the pair of events is classified into the same control event group; For each control event group, the occurrence location of the trajectory anomaly event and the occurrence location of the risk conflict event are extracted, and the Euclidean space distance between them is calculated. Extract the occurrence time of the trajectory anomaly event and the risk conflict event, and calculate the time lag interval of the two; introducing the Euclidean space distance and the time lag interval of each control event group into a causal confidence score model , and calculating the causal confidence of each control event group, wherein represents the causal confidence, 、 respectively represent the Euclidean space distance and the time lag interval of the control event group, 、 respectively represent the maximum space threshold and the maximum allowed time delay.

9. A method for road alignment safety evaluation analysis optimization as claimed in claim 8 wherein: The risk inducing position is identified based on the causal correlation result, see the following content: Compare the causal confidence of each control event group with the configured causal significance threshold, and screen out the control event group greater than the causal significance threshold, and take the occurrence position of the trajectory anomaly event in the control event group as the risk inducing position.

10. The method of road alignment safety evaluation analysis optimization as claimed in claim 1 wherein: The step 5 includes the following contents: Map the risk inducing position to the corresponding coordinate of the high-precision road line digital map; Take each risk inducing position as the center, and construct a circular analysis area with a preset radius; Extract the detailed design parameters of the road line from the digital map in the circular analysis area, including the curvature radius of the plane line, the slope and vertical curve radius of the longitudinal section line, and the lane width and shoulder width of the transverse section line; Correlate the extracted road line design parameters with the recorded trajectory anomaly event characteristics at the risk inducing position, and infer the line defect type according to the pre-defined diagnosis rule library.

Images