Irregular overtaking behavior identification method and device, storage medium and program product

Abstract

The invention discloses a non-standard overtaking behavior identification method and system, and belongs to the field of intelligent traffic. According to the method, trajectory data streams of a plurality of vehicles are obtained, and an interaction analysis group is constructed in real time based on space-time proximity. In the group, a spherical trigonometric function is utilized to accurately calculate the relative orientation between the vehicles, and the occurrence of a non-standard overtaking event is judged by monitoring the change of the position relation of the vehicles from the rear to the front and combining a preset speed characteristic rule. According to the method, kinematics parameters, front potential conflicts and high-risk geofence information in the overtaking process are further fused, and multi-dimensional quantitative risk assessment is carried out on the overtaking event. The limitation of a single vehicle visual angle is overcome, and the accuracy of non-standard overtaking identification is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent transportation, and in particular, to a method and device for identifying non-standard overtaking behavior, a storage medium, and a program product. BACKGROUND

[0002] In the road traffic safety prevention and control work in China, the "key vehicles" composed of heavy trucks, long-distance buses and dangerous goods transport vehicles have become the top priority of supervision because they account for the highest proportion in serious traffic accidents. Data analysis shows that overtaking behavior is extremely common in the huge flow of highway networks, and one type of dangerous driving behavior known as "non-standard overtaking" is a serious hidden danger that induces serious traffic accidents. According to statistics, among the overtaking behaviors that occur every day, tens of thousands of them are non-standard overtaking, and directly lead to dozens of traffic accidents. Therefore, effective prevention and control of this behavior is a top priority.

[0003] In some related technologies, such as advanced driver assistance systems (ADAS) based on vehicle-mounted perception units or roadside perception solutions, there are fundamental technical limitations in identifying overtaking behavior. The core problem of these solutions lies in their inherent "single vehicle perspective" and "physical space limitations". Specifically, the solutions rely on the sensors (such as cameras, millimeter wave radars, etc.) of the vehicle itself, whose perception range is not only limited by physical distance, but also has physical obstruction problems. For example, when a large truck is parallel to the side of the vehicle, its sensor field of view will be completely blocked, making it difficult to predict the dynamics of other road users outside the blocked area in advance, especially dangerous overtaking vehicles approaching at high speed from the perception blind area. The risk is often only perceived at the last moment when the dangerous behavior occurs, and by then the best warning and avoidance opportunity has been missed. This kind of calculation based on single observation naturally has delay and error, and it is difficult to form an accurate "fact consensus". Therefore, such methods have the problem of too single evaluation dimension, leading to inaccurate risk identification. SUMMARY

[0004] The present application provides a method and device for identifying non-standard overtaking behavior, a storage medium, and a program product, to solve the problem of too single evaluation dimension in related technologies, leading to not comprehensive risk identification.

[0005] In a first aspect, the application provides a method for identifying non-standard overtaking behavior, which is applied to a non-standard overtaking behavior identification system. The method comprises: obtaining trajectory data streams of a plurality of vehicles running in parallel, the trajectory data stream comprising at least real-time position coordinates, running heading angle, speed data and time stamp, and the trajectory data stream corresponding to a vehicle; based on the time stamp and the real-time position coordinates, performing real-time horizontal correlation on the current trajectory data stream to determine an interaction analysis group, the current trajectory data stream being the trajectory data stream of at least two vehicles located in a preset spatial adjacent area within the same time window; in the interaction analysis group, converting the real-time position coordinates into radian coordinates; in the interaction analysis group, selecting one real-time position coordinate as a reference object; calculating the absolute bearing angle between the reference object and a target object according to the radian coordinates and spherical trigonometric functions; the target object being another real-time position coordinate in the interaction analysis group; performing difference calculation on the absolute bearing angle and the running heading angle of the reference object to obtain a relative bearing angle, the relative bearing angle being the bearing angle with the running direction of the reference object as the reference; mapping the relative bearing angle to a preset standard interval to obtain front-rear relationship data of the target object and the reference object; extracting the speed data when the front-rear relationship of the target object and the reference object changes from rear to front in the front-rear relationship data; and determining and generating a target overtaking event when the speed data meets a preset target overtaking characteristic rule.

[0006] The above embodiment collects and correlates the trajectory data streams of a plurality of vehicles to construct an interaction analysis group beyond the physical perception range of a single vehicle, and identifies non-standard overtaking behavior based on this. The method first obtains and fuses the trajectory data of a plurality of vehicles in the same spatio-temporal adjacent area, breaking through the limitations of "single vehicle perspective" and sensor physical obstruction in related technologies. Instead of relying on the limited and easily obstructed sensor field of view of a single vehicle, it forms a shared, blind area-free God's perspective, thereby enabling continuous and complete observation of the mutual relationship between vehicles running in parallel, and improving the problem that dangerous behavior cannot be predicted in advance due to the perception blind area or physical obstruction. Further, the application converts high-precision position coordinates into radian coordinates, and calculates the relative bearing angle by using spherical trigonometric functions in combination with the running heading angle, accurately capturing the position relationship change from "rear" to "front" between vehicles. This calculation method provides an objective and stable geometric basis for the determination of overtaking behavior, and improves the accuracy and timeliness of event identification compared to the estimation based on single observation.

[0007] In some embodiments of the first aspect, in some embodiments, the determining of the target overtaking feature rule comprises: calculating a minimum lateral distance, a longitudinal cutback spacing, and a relative speed difference based on the trajectory data streams of the target object and the reference object; the minimum lateral distance is a minimum value of the lateral distance between the target object and the reference object during parallel driving; the longitudinal cutback spacing is a distance between the front of the target object and the reference object after overtaking is completed; the relative speed difference is a speed difference between the target object and the reference object during overtaking; and generating a kinematic danger score representing a danger level of the target overtaking event by a preset quantization model based on the minimum lateral distance, the longitudinal cutback spacing, and the relative speed difference.

[0008] The above embodiment further extracts three key kinematic parameters, i.e., the minimum lateral distance, the longitudinal cutback spacing, and the relative speed difference, at the time of the overtaking event, and inputs them into a preset quantization model to generate a kinematic danger score. This score deepens the macroscopic identification of non-standard overtaking behavior into a microscopic quantitative evaluation of its inherent danger. Instead of simply determining whether the event occurs or not, it provides an objective measure of the risk level through specific dynamic interaction data, thereby improving the problem in the related art that only a rough qualitative judgment can be made and different overtaking behaviors cannot be effectively distinguished in terms of danger level.

[0009] In some embodiments of the first aspect, in some embodiments, the method further comprises: identifying a target lane used by the target object during overtaking, and searching for a potential conflict object in front of the target lane in the driving direction of the target object; after the potential conflict object is searched in front of the target lane in the driving direction of the target object, calculating a predicted collision time between the target object and the potential conflict object; and generating a scene danger score representing a comprehensive danger level of the target overtaking event by a preset risk fusion model based on the predicted collision time and the kinematic danger score.

[0010] The above embodiment extends the analysis dimension from the interaction between vehicles to the external traffic environment in which the vehicles are located. By identifying the target lane occupied by the overtaking action and actively searching for a potential conflict object in front of the lane, the system can calculate a predicted collision time. Then, the time dimension information with foresight is fused with the previously generated kinematic danger score to generate a comprehensive scene danger score. This process changes the risk assessment from a isolated and retrospective kinematic analysis to a dynamic and predictive scene deduction, thereby improving the limitation in the related art that the risk assessment is passive in response to the danger that has occurred due to the lack of prediction ability for the front environment, and making the risk assessment closer to the complex interaction scene of the real road.

[0011] In some embodiments of the first aspect, the method further comprises: obtaining map data of high-risk geofences; the high-risk geofences are pre-labeled and indicate geographic locations including a curve, a slope, a tunnel, or a ramp merge entrance; obtaining a geographic location where the target overtaking event occurs, and determining whether the geographic location falls within the high-risk geofence, to obtain a determination result; and based on the scene danger score, combining the determination result of whether the geographic location falls within the high-risk geofence, and generating a final comprehensive danger level through a preset spatiotemporal fusion rule.

[0012] The above embodiments introduce static geographic space information into the risk assessment model. By obtaining geofence data of high-risk road segments such as curves, slopes, tunnels, or ramp merge entrances, and determining whether the location where the overtaking event occurs falls within these specific areas, the system combines the determination result of the geographic location with the aforementioned scene danger score according to a preset spatiotemporal fusion rule to generate a final comprehensive danger level. This makes the danger level assessment go beyond pure dynamic traffic flow analysis, organically unifying the instantaneous danger of the event (dynamic interaction) and the inherent risk of the location (static environment). This improves the related art, which ignores static background risks such as fixed road facilities and geographic environment when performing risk assessment, resulting in a deviation between the assessment result and the real-world risk. The final danger level is more comprehensive and accurate.

[0013] In some embodiments of the first aspect, the target overtaking feature rule further comprises: when any of the following determinations is yes, the target overtaking feature rule is satisfied: determining whether the speed of the target object is greater than the speed of the reference object, and whether the speed difference between the two is less than a preset first speed threshold; determining whether the speed of the target object exceeds the legal speed limit of the road on which it is located; determining whether the speed difference between the target object and the reference object is greater than a preset second speed threshold.

[0014] The above embodiment, when identifying the target overtaking event, does not make a general behavior judgment, but sets more detailed speed characteristic rules to construct a preposed risk characteristic screening layer. First, the scheme can identify the "suspended" overtaking behavior that the target vehicle is parallel to the reference vehicle for a long time and cannot complete the overtaking by judging whether the speed difference between the target object and the reference object is less than the preset first speed threshold. The judgment blind spot caused by the related art due to the inability to identify this specific dangerous mode with long duration and large potential conflict window is improved. Further, by introducing the judgment dimensions of whether the vehicle itself is speeding and whether the speed difference between the two vehicles is too large, the scheme also includes the behaviors of speeding overtaking and high-speed forced overtaking, which have clear high-risk characteristics, into the rule category, making up for the deficiency of the previous technology that only focuses on whether the overtaking action is completed, but ignores the key risk element of speed. Through the combination of multi-dimensional speed rules, the overtaking behaviors with different risk characteristics can be preliminarily classified in the identification stage, providing more accurate and focused event input for subsequent quantitative risk assessment, and improving the pertinence and efficiency of the whole dangerous event identification system.

[0015] In combination with some embodiments of the first aspect, in some embodiments, the method further comprises: generating corresponding risk warning information based on the scene danger score, the size of the scene danger score determining at least one output attribute of the risk warning information; the output attribute including the type of the warning information, the intensity of the warning information, and the presentation mode of the warning information.

[0016] The above embodiment directly associates the scene danger score with the output attribute (such as type and intensity) of the warning information, and establishes a hierarchical warning strategy. By matching the warning intensity with the actual severity of the risk, the "alarm fatigue" problem caused by non-critical alarms can be improved. At the same time, the hierarchical warning intuitively conveys the urgency of the risk to the driver, assisting him / her to make more timely decisions and improving the effectiveness of risk intervention.

[0017] In combination with some embodiments of the first aspect, in some embodiments, the method further comprises: collecting a plurality of final comprehensive danger levels associated with a specific vehicle within a preset time period to form a danger level time sequence; and performing statistical analysis on the danger level time sequence to extract at least one driving behavior feature, and generating a driving behavior portrait of the specific vehicle based on the driving behavior feature.

[0018] The above embodiments expand the dimension of risk assessment from isolated instantaneous events to a continuous time dimension. First, by aggregating multiple final comprehensive hazard levels of a specific vehicle within a preset period, a hazard level time series is formed. This transforms discrete, single-event assessments into a continuous observational data stream of the vehicle's long-term risk performance, overcoming the limitation of related technologies that can only judge independent events and cannot form longitudinal tracking and comparisons. Furthermore, by statistically analyzing this time series and extracting driving behavior characteristics, this solution can quantify the vehicle's dynamic performance over a period of time into stable data indicators such as "frequency of high-risk events" and "average risk score." Finally, the driving behavior profile generated based on these quantified characteristics achieves a hierarchical leap from "event risk" assessment to qualitative "driver style," providing a basis for evaluating a driver's habitual risk level and compensating for the shortcomings of previous technologies that lacked long-term data accumulation and analysis methods, thus failing to make fundamental judgments about a driver's risk tendencies.

[0019] Secondly, embodiments of this application provide an irregular overtaking behavior recognition device, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, which includes computer instructions, and the one or more processors call the computer instructions to cause the irregular overtaking behavior recognition device to perform the method described in the first aspect and any possible implementation thereof.

[0020] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on an irregular overtaking behavior recognition device, cause the irregular overtaking behavior recognition device to execute the method described in the first aspect and any possible implementation thereof.

[0021] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on an irregular overtaking behavior recognition device, cause the irregular overtaking behavior recognition device to perform the method described in the first aspect and any possible implementation thereof.

[0022] Understandably, the non-standard overtaking behavior recognition device provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the method provided in the embodiments of this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0023] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. This application constructs an interactive analysis group by aggregating multi-vehicle trajectory data streams, breaking through the limitations of traditional single-vehicle sensors in physical space and field of view, and forming a blind-spot-free collaborative perception perspective. This fundamentally improves the problem in the background technology that dangerous behaviors (especially irregular overtaking) cannot be observed in advance and completely due to physical obstruction and limited perception range, providing an objective and comprehensive data foundation for subsequent accurate identification and risk assessment.

[0024] 2. This application establishes a multi-level risk assessment system from macro-level screening to micro-level quantification. First, high-risk overtaking behaviors are pre-screened using speed characteristic rules, improving identification efficiency. Then, by integrating the kinematic parameters of the overtaking process, potential conflict objects on the road ahead, and the static geographical environment risks of the incident site, the judgment of irregular overtaking behavior is deepened from a simple qualitative identification of "whether it has occurred" to a precise quantitative classification of the degree of danger, improving the shortcomings of related technologies that can only make rough judgments and cannot effectively distinguish the level of danger.

[0025] 3. This application expands the dimension of risk assessment from isolated instantaneous events to a continuous time dimension. By statistically analyzing the comprehensive hazard level sequence of a specific vehicle over a long period, a driving behavior profile is ultimately generated. This achieves a key leap from "single-event risk assessment" to "qualitative characterization of driver habitual styles," filling the technological gap where previous technologies, lacking long-term data accumulation and analysis methods, could not effectively judge the fundamental risk tendencies of the driving subject. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating a method for identifying irregular overtaking behavior in an embodiment of this application. Figure 2 This is a flowchart illustrating a hazard refinement scoring method in an embodiment of this application; Figure 3 This is a flowchart illustrating the scenario hazard scoring method in an embodiment of this application; Figure 4 This is a flowchart illustrating a geofencing-based risk assessment method in an embodiment of this application. Figure 5 This is a schematic diagram of the physical device structure of the non-standard overtaking behavior recognition device in the embodiments of this application. Detailed Implementation

[0027] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0028] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0029] To facilitate understanding, the application scenarios of the embodiments of this application are described below.

[0030] In recent years, key vehicles (passenger vehicles, dangerous goods transport vehicles, and freight trucks weighing over 12 tons) have accounted for the highest proportion of serious traffic accidents, making them a top priority in traffic safety prevention and control. Data analysis shows that out of a daily average traffic flow of 2.31 million vehicles, 1.386 million overtaking incidents occurred, accounting for 60%. Among these, 67,000 were improper overtaking maneuvers, resulting in 67 accidents. Therefore, prevention and control are of paramount importance.

[0031] Currently, overtaking risk assessment technologies primarily rely on real-time dynamic traffic parameters acquired by onboard sensors. For example, they analyze the relative position and speed of the vehicle and surrounding targets to determine the potential for an instantaneous collision. However, a core problem with this approach is its overly simplistic assessment dimensions, leading to incomplete risk identification. Specifically, its assessment is almost entirely limited to the vehicle's interpretation of the dynamic traffic scenario at that particular moment, neglecting other equally crucial risk dimensions.

[0032] This application proposes a novel overtaking risk assessment method and system. Its innovation lies in the deep integration of dynamic traffic game theory, static geographical environment risk, and target vehicle driving behavior profiles to construct a comprehensive spatiotemporal multi-dimensional assessment model, thereby achieving more accurate and predictive safety decisions. The specific implementation methods of this application will be described in detail below.

[0033] To facilitate understanding, the method provided in this implementation will be described in detail below, using the above scenario as an example. Please refer to [link / reference]. Figure 1 This is a flowchart illustrating a method for identifying irregular overtaking behavior in an embodiment of this application.

[0034] S101. Obtain trajectory data streams of multiple vehicles traveling in parallel. The trajectory data streams include at least real-time position coordinates, heading angles, speed data, and timestamps. Each trajectory data stream corresponds to a vehicle.

[0035] Trajectory data stream refers to a set of data continuously generated and reported by in-vehicle intelligent terminals, especially devices equipped with BeiDou high-precision positioning modules, at a preset frequency (e.g., 5-10 times per second), used to describe the spatiotemporal motion state of a single vehicle; its essence is a time series. Real-time location coordinates refer to the high-precision (e.g., centimeter-level) vehicle geographic location information calculated in real time by the BeiDou satellite navigation system or other Global Navigation Satellite Systems (GNSS), typically expressed as latitude and longitude in the WGS-84 or CGCS2000 coordinate system, used to accurately depict the vehicle's absolute physical location.

[0036] The trigger condition for this step is that the irregular overtaking behavior recognition system described in this application starts and begins monitoring a designated area or a designated type of vehicle (such as "key vehicles"). This step is the starting point of the entire recognition method, and its function is to serve as a continuously running data acquisition layer, providing basic data input for all subsequent analysis steps. Specifically, this step transmits the collected trajectory data stream to the cloud data center in real time or near real time through the communication module of the vehicle terminal. This aggregates the scattered and isolated dynamic data of individual vehicles in the road network into a centralized data pool that can be analyzed uniformly. It provides the necessary data foundation for this application to subsequently build an "interactive analysis group" that goes beyond the perspective of a single vehicle, and on this basis, identify complex vehicle interaction behaviors (such as irregular overtaking), which is a prerequisite for realizing the core technical concept of this application.

[0037] S102. Based on timestamps and real-time location coordinates, the current trajectory data stream is horizontally correlated in real time and identified as an interactive analysis group. The current trajectory data stream consists of the trajectory data streams of at least two vehicles located in the same spatial adjacent area within the same time window.

[0038] Real-time lateral correlation refers to a data processing action. Its core is to dynamically and in real-time identify vehicles traveling in parallel or with potential lateral interactions (such as overtaking or lane changing) within a massive, independent stream of vehicle trajectory data, based on the principle of spatiotemporal proximity, and logically bind their data. An interaction analysis group refers to a temporary, dynamic data set constructed through real-time lateral correlation. This set contains the trajectory data streams of at least two vehicles that are physically adjacent at the same time, constituting a closed "micro-scene" suitable for subsequent relative motion analysis.

[0039] This embodiment operates in a general-purpose server environment integrating a message queue (such as Kafka) and an in-memory database (such as Redis). Specifically, upon receiving trajectory data from vehicle A, the system immediately uses the Redis GEORADIUS command to efficiently query neighboring vehicles (such as vehicle B) within a 150-meter radius. Subsequently, by comparing timestamps, vehicles with data latency exceeding 500ms are filtered out. Finally, the spatiotemporally synchronized data from vehicles A and B are packaged into an interactive analysis group and pushed to the downstream processing queue. This step, through centralized real-time computation, virtually constructs a "cooperative perception domain" for spatiotemporally proximate vehicles that surpasses the physical perception capabilities of any single vehicle, providing a complete data foundation previously unavailable for subsequent hazardous behavior analysis.

[0040] It should be noted that the implementation method of this application is not limited to this. For example, optionally, a stream processing engine such as Apache Flink can be used to directly generate interaction groups in the data stream through its geographic partitioning and time window operators to achieve high throughput. Alternatively, a database that supports spatiotemporal queries (such as PostgreSQL / PostGIS) can be used to directly extract vehicle groups that meet the conditions from the database in batches using an optimized SQL compound query. It is understood that any technology that can efficiently achieve spatiotemporal correlation to construct multi-vehicle interaction scenarios can be adopted, and no limitation is made here.

[0041] S103. Within the interactive analysis group, convert the real-time position coordinates into radian coordinates.

[0042] This step involves standardizing and preprocessing the data within the "interactive analysis group" generated in the previous step. Real-time location coordinates refer to the original latitude and longitude coordinates of each vehicle in the data group (e.g., [longitude 116.404°, latitude 39.915°]), while radian coordinates are the numerical representation of latitude and longitude converted using the mathematical formula (radians = angle × π / 180).

[0043] This step is performed immediately after the interactive analysis group is generated and before any relative relationship calculations are performed. Specifically, the data processing program iterates through the vehicle data within the interactive analysis group. For example, an interactive analysis group contains vehicle A and vehicle B, with their original WGS-84 coordinates as follows: Vehicle A: lon=116.404°, lat=39.915°; Vehicle B: lon=116.405°, lat=39.916°. The program will apply the following formulas to calculate the coordinates of vehicle A: Longitude in radians_A=116.404*3.1415926 / 180≈2.03149; Latitude in radians_A=39.915*3.1415926 / 180≈0.69678. Similarly, the coordinates of vehicle B are transformed: longitude in radians_B = 116.405 * 3.1415926 / 180 ≈ 2.03151; latitude in radians_B = 39.916 * 3.1415926 / 180 ≈ 0.69680. After the transformation, the program stores these radian values ​​as new fields in the corresponding vehicle's data structure. This step unifies all geographic coordinates into a standard input format suitable for subsequent spherical geometric calculations such as the Haversine formula, laying the foundation for efficiently calculating the precise distance and azimuth between vehicles, avoiding repeated transformations in each calculation, thereby improving the computational efficiency and real-time performance of the entire interactive analysis process.

[0044] S104. Within the interactive analysis group, select a real-time location coordinate as the reference object.

[0045] Specifically, once the system constructs an interactive analysis group (e.g., containing vehicles A, B, and C), it needs an efficient and reproducible method to select a reference. This embodiment preferably employs a "unique identifier sorting" rule. For example, the platform extracts the unique IDs (such as license plate numbers or device IDs) of each vehicle within the group, which are ID_A="...3A5C", ID_B="...1B8F", and ID_C="...8E22". By sorting these ID strings lexicographically, ID_B is the smallest. Therefore, the platform selects vehicle B as the reference object for this analysis. This step provides an automated and deterministic method for macro-level traffic analysis systems to dynamically anchor the analysis focus. Faced with tens of thousands of concurrent interactive events that may occur every second on a cloud platform, this scalable and standardized rule can efficiently and unambiguously transform the chaotic multi-vehicle absolute motion data stream into structured relative motion analysis problems centered on specific vehicles, improving the system's scalability and computational consistency in handling complex traffic scenarios.

[0046] It should be noted that this application can also achieve the selection of reference objects in various ways. For example, optionally, an "event-triggered" rule can be used. Specifically, the upper-level platform can continuously monitor key events in the vehicle data stream, such as emergency braking, abnormal lane changes, or triggering of the warning system. Once a vehicle (e.g., vehicle C) is detected to have triggered such an event, the system will immediately and dynamically designate the "event-triggered vehicle" as the reference object and build an interaction analysis group around it. It is understood that any method that can determine a unique and stable analysis benchmark for the interaction analysis group can be used, and no limitation is made here.

[0047] S105. Calculate the absolute azimuth angle between the reference object and the target object based on radian coordinates and spherical trigonometric functions.

[0048] This step is performed for each target object within the interaction analysis group after the reference object is determined. It is the first step in transforming two independent absolute coordinate points into points with a directional relative relationship. Specifically, the host platform extracts the radian coordinates of the reference object and the target object from the database. To illustrate this more clearly, a calculation example is given here: Initial coordinates: Reference object A: lon_ref≈2.031439rad, lat_ref≈0.696342rad.

[0049] Target object B: lon_tgt≈2.031457rad, lat_tgt≈0.696351rad.

[0050] Azimuth calculation: Apply the spherical azimuth calculation formula Bearing = atan2(Y, X), where: Y=sin(lon_tgt-lon_ref)*cos(lat_tgt), X=cos(lat_ref)*sin(lat_tgt)-sin(lat_ref)*cos(lat_tgt)*cos(lon_tgt-lon_ref), Substituting the values, we get Y≈0.0000134, X≈0.0000873.

[0051] Interpretation of the results: Executing atan2(0.0000134, 0.0000873) yields a result of approximately 0.1523 rad. Converting this to degrees (degree = radians × 180 / π), the absolute azimuth is approximately 8.72°. This result clearly indicates that, from the position of vehicle A, vehicle B is located approximately 8.72 degrees east of due north.

[0052] This step uses a spherical model for calculations, rather than a simplified planar approximation. This solves the technical problem of significant errors in azimuth calculations due to the Earth's curvature in large-span or high-latitude regions, ensuring that the calculated direction remains lane-level accurate even when vehicles are hundreds of meters or more apart. This provides highly reliable source data for subsequent advanced functions such as relative position calculations and collision risk assessments.

[0053] S106. The relative azimuth angle is obtained by calculating the difference between the absolute azimuth angle and the driving heading angle of the reference object. The relative azimuth angle is the azimuth angle with the driving direction of the reference object as the reference.

[0054] This step, performed after obtaining the absolute azimuth of the target object, is the core of constructing a perception view centered on the reference vehicle. Specifically, the platform performs a simple subtraction: relative azimuth = absolute azimuth - reference object's heading angle. Due to the periodicity of angle calculations, the result needs to be normalized, typically adjusted to the range of [-180°, 180°] to clearly distinguish left from right. Continuing with the example from S105: the absolute azimuth of target object B relative to reference object A is known to be 8.72°. At this time, the platform obtains the real-time heading angle of reference object A as 350° (meaning the vehicle's front is facing 10 degrees west of north). The initial relative azimuth is then calculated as 8.72° - 350° = -341.28°. After normalization (adding 360°), the final relative azimuth is 18.72°. The physical meaning of this result is: target vehicle B is located approximately 18.72 degrees to the right and front of reference vehicle A in its direction of travel. The innovation of this step lies in its large-scale, real-time transformation from a "God's-eye view" to a "driver's-eye view" on a cloud platform. This perspective shift is the logical foundation for realizing collaborative intelligent driving applications (such as lane change assistance and forward collision warning), enabling the platform to understand and assess the situation and risks of multi-vehicle interactions in a way that aligns with human driving intuition.

[0055] S107. Map the relative azimuth angle to a preset standard range to obtain the front-back relationship data between the target object and the reference object.

[0056] The preset standard intervals refer to one or more angle ranges predefined by the system developers based on business needs. The most basic setting is to divide the 360-degree area around the reference vehicle into two regions: "front" and "rear." For example, the interval [-90°, 90°] is defined as "front," and the union of (-180°, -90°) and (90°, 180°) is defined as "rear." The front-rear relationship data is the final output classification result; it is a simplified data label, such as a string ("FRONT" / "REAR") or an enumeration value, used to identify whether the target object is in the front or rear view of the reference object.

[0057] This step, following the relative azimuth angle calculation, is used to coarsely categorize a massive number of interactive objects, simplifying the complexity of subsequent analysis. Specifically, after receiving the relative azimuth angle output by S106, the platform performs a simple logical judgment. Continuing with the previous example: the relative azimuth angle between reference object A and target object B is 18.72°. The platform loads a preset interval definition, i.e., forward = [-90°, 90°]. Since 18.72° falls within this interval, the platform marks the relationship data between target object B and reference object A as "forward". By performing real-time, large-scale forward / backward relationship classification on all traffic participants, the allocation of computing resources is optimized. For example, when performing collaborative analysis for Forward Collision Warning (FCW) or Automatic Emergency Braking (AEB), the system can instantly filter out all vehicle pairs marked as "rear", thus concentrating valuable computing power on targets that truly pose a forward collision risk.

[0058] S108. Extract the velocity data when the front-back relationship between the target object and the reference object changes from back to front in the front-back relationship data.

[0059] This step acts as an event trigger in the system logic. It continuously monitors the relational stream output by S107, and immediately executes data extraction upon detecting a specified state change. Specifically, the analysis engine of the upper platform maintains a context in memory for each "reference-target" vehicle pair, containing at least the state of the previous time step. When the relational data Rel(t) of the current time step t is received, the system compares it with the stored relational data Rel(t-1) of the previous time step. The event is triggered only when both Rel(t-1) and Rel(t) are "behind" and "forward" are met simultaneously. At this time, the system immediately extracts the speed V_ref(t) of the reference object and the speed V_tgt(t) of the target object from the raw data packet of the current time step t. For example, at time t1, car B is behind car A; at the next data point t2, car B is determined to be in front of car A. The system detects this change and immediately captures the data at time t2: car A's speed is 100 km / h, and car B's speed is 120 km / h. The data pair (100, 120) is then extracted and appended with a timestamp and vehicle ID for subsequent risk assessment or driving behavior analysis.

[0060] S109. When the speed data meets the preset target overtaking feature rules, determine and generate a target overtaking event.

[0061] In this case, the "target overtaking characteristic rule" specifically refers to the set of logical conditions used to define "irregular" or "dangerous" overtaking. It no longer judges whether overtaking has occurred, but rather whether the overtaking behavior is compliant and safe. These rules aim to quantify overtaking patterns that violate traffic regulations or exhibit high-risk characteristics in terms of dynamics. Correspondingly, a target overtaking event specifically refers to an "irregular overtaking event" or a "dangerous overtaking event."

[0062] In some embodiments, the target overtaking feature rule includes: the target overtaking feature rule is satisfied when any of the following is true: determining whether the speed of the target object is greater than the speed of the reference object, and whether the speed difference between the two is less than a preset first speed threshold; determining whether the speed of the target object exceeds the legal speed limit of the road where it is located; determining whether the speed difference between the target object and the reference object is greater than a preset second speed threshold.

[0063] This step acts as a risk arbiter in the system logic, performing a mandatory safety review on each overtaking candidate event. Specifically, when S108 inputs a set of speed data, such as the target object's speed V_tgt = 135 km / h, the reference object's speed V_ref = 90 km / h, and the system knows that the current road segment's speed limit is 100 km / h, the judgment logic immediately starts. The system will check three preset rules one by one: First, it checks whether the "slow overtaking" rule is met. The speed difference is calculated as 135 - 90 = 45 km / h, which is not less than the first speed threshold of 5 km / h, so this rule is not met. Next, it checks whether the "speeding overtaking" rule is met. V_tgt is 135 km / h, which exceeds the legal speed limit of 100 km / h, so this rule is met. Since the rule is "any one must be met," the system could actually make a final judgment at this point, but it will continue to check in order to record the complete reason. Finally, the system determines whether the overtaking is considered "excessive overtaking." A speed difference of 45 km / h exceeds the second speed threshold of 40 km / h, which also meets this criterion. In summary, since at least one rule is met, the system ultimately classifies the overtaking as irregular and generates a "target overtaking event." The event details can include specific violation codes, such as "speeding" and "excessive speed difference." This step engineered the previously vague concept of "dangerous driving" through a set of clear, configurable, and quantitative rules, achieving automated, standardized, and real-time detection of irregular overtaking behavior. This provides a reliable and objective basis for subsequent applications such as risk warning, liability clarification, and insurance pricing.

[0064] The above embodiments solve the qualitative identification problem of whether or not irregular overtaking behavior occurs. To further achieve a more refined measurement of the degree of danger of an event, the following embodiments will, on this basis, explain how to extract key kinematic parameters and establish a quantitative model to generate an objective kinematic hazard score, thereby elevating the analysis from a judgment of "whether it has occurred" to an assessment of "how dangerous it is".

[0065] The following provides a more detailed description of the process of the method provided in this implementation. Please refer to [link / reference]. Figure 2 This is another flowchart illustrating the hazard refinement scoring method in this application embodiment.

[0066] S201. Based on the trajectory data streams of the target object and the reference object, calculate the minimum lateral distance, the longitudinal cut-off distance, and the relative speed difference; the minimum lateral distance is the minimum lateral distance between the target object and the reference object during parallel driving; the longitudinal cut-off distance is the distance between the front of the target object and the reference object after overtaking; the relative speed difference is the speed difference between the corresponding vehicles of the target object and the reference object during the overtaking process. This step is the "feature engineering" stage of the risk quantification model. It is triggered after the system determines that a target overtaking event (as described in Example 1) has occurred. Its purpose is to prepare standardized and quantifiable input for the subsequent hazard score calculation (S202). Specifically, the system first locks the complete trajectory data stream segment corresponding to this overtaking event. To calculate the minimum lateral distance, the system iterates through each frame of data during the "parallel driving period," calculating the absolute value of the difference between the two vehicles' centers of mass in the lateral coordinates in each frame: |y_target(t) - y_reference(t)|. Finally, it takes the minimum value within the entire period as the result. To calculate the longitudinal cut-back distance, the system needs to accurately identify the "cut-back completion" time t_cut_in. At this time, it reads the longitudinal coordinates of the two vehicles' front ends and calculates x_target_head(t_cut_in) - x_reference_head(t_cut_in) to obtain the value. To calculate the relative speed difference, the system calculates the speed difference V_target(t) - V_reference(t) for each frame within the time window of the overtaking process. Then, it takes the arithmetic mean of all these speed differences to obtain a stable value that represents the average dynamics of the entire process. This step abstracts and reduces the complexity and dimensionality of a complex, time-varying overtaking process into three core feature parameters with clear dimensions, distinct physical meaning, and high correlation to risk through kinematic modeling. This simplifies the complexity of the subsequent quantification model and makes it interpretable.

[0067] It should be noted that, in order to construct a more comprehensive and robust hazard feature space, this application is not limited to the three core parameters mentioned above. Other kinematic features capable of characterizing specific risk dimensions of overtaking behavior can be further introduced and used as input to the pre-defined quantitative model to achieve a more comprehensive auxiliary quantification of the degree of danger. For example, maximum lateral acceleration can be introduced to quantify the aggressiveness of the maneuver and the risk of loss of control; total overtaking time can be introduced to assess the duration of risk exposure of the vehicle on abnormal paths; or the cut-back instantaneous collision time (TTCi) can be introduced to more accurately measure the urgency of the direct collision caused to the following vehicle by the cut-in action. It is understood that any kinematic or dynamic parameters that can be extracted from the trajectory data stream to characterize the potential danger of overtaking events can be included in the calculation scope of this application, without exhaustive limitations.

[0068] S202. Based on the minimum lateral distance, longitudinal turning distance, and relative speed difference, a kinematic hazard score is generated through a preset quantization model to characterize the degree of danger of the target overtaking event.

[0069] The pre-defined quantitative model refers to a mathematical or logical structure built into the system that maps multiple input parameters to a single output score. Its core function is to simulate the expert's perception and judgment standards for dangerous driving behavior, making the abstract "degree of danger" concrete and numerical. The kinematic hazard score represents the final result output by the model. It is a standardized numerical value (e.g., 0-100 points) used to directly and quantitatively represent the level of danger of the target overtaking event at a purely kinematic level. The higher the score, the greater the potential collision risk, loss of control risk, or interference with other road users.

[0070] This step is a crucial execution step in achieving "quantitative" rather than "qualitative" risk assessment. Specifically, the system inputs the three values ​​output by S201—minimum lateral distance, longitudinal cut-off distance, and relative speed difference—as independent variables into the preset quantitative model. The internal logic of the model embodies the core technical idea of ​​this application: it not only independently evaluates the risk contribution of each parameter (e.g., the smaller the lateral distance, the higher the risk; the smaller the cut-off distance, the higher the risk), but more importantly, it comprehensively considers the coupling effect between these three parameters. For example, a high relative speed difference is not necessarily extremely dangerous in itself, but if it is accompanied by a very small longitudinal cut-off distance, it means a violent and highly aggressive cut-in, and its danger level will increase dramatically and non-linearly. This step transforms a complex, multi-dimensional overtaking risk assessment problem into a simple and clear numerical scoring process, enabling overtaking events of different times, locations, and vehicles to be compared and ranked in terms of danger under the same standard, providing accurate and reliable data support for subsequent driving behavior analysis and risk warning strategies.

[0071] Furthermore, to enable the quantification model to accurately adapt to complex interactive scenarios and reflect a deep understanding of different traffic environments, a preferred embodiment of this application introduces an adaptive adjustment mechanism of "dynamic weights." This means that when calculating the final hazard score, the risk weights of core parameters such as minimum lateral distance, longitudinal turning distance, and relative speed difference are not fixed, but are dynamically adjusted according to the real-time attributes or status of the reference object (i.e., the overtaken vehicle).

[0072] For example, when the vehicle being overtaken is a large truck, considering the strong crosswinds (crosswind effect) generated by the truck and the physical pressure and psychological stress on the driver caused by its large size, the model will automatically increase the risk weight of the "minimum lateral distance" parameter. Similarly, if the model determines that the reference object has poor driving stability (such as frequent lateral swaying within the lane) by analyzing the trajectory data of the reference object, it will simultaneously increase the weight of the lateral and longitudinal distance related parameters, because being close to a traffic participant with unstable behavior means higher unpredictable risks.

[0073] In this way, the quantitative model of this application is no longer a rigid static evaluator, but has evolved into an intelligent risk assessment system capable of sensing the environment and understanding interactions. The kinematic hazard scores it generates can thus more realistically and accurately reflect the comprehensive hazard level under specific dynamic scenarios, enhancing the accuracy and robustness of risk assessment.

[0074] The aforementioned method quantifies the inherent danger of overtaking maneuvers by analyzing the kinematic parameters of both vehicles, yielding a kinematic hazard score. However, this score primarily reflects the internal interaction risk between vehicles, while the overall danger of an overtaking maneuver also depends on the external traffic environment. Therefore, this application further extends the assessment dimension to the scenario ahead. The following embodiments illustrate how to identify potential conflict objects ahead, calculate the predicted time of collision (TTC), and fuse this with the kinematic score to generate a more comprehensive scenario hazard score.

[0075] The following provides a more detailed description of the process of the method provided in this implementation. Please refer to [link / reference]. Figure 3 This is a flowchart illustrating a scenario hazard scoring method in an embodiment of this application.

[0076] S301. Identify the target lane used by the target object during the overtaking process, and search for potential conflicting objects in front of the target object in the target lane in the direction of travel. Potential collision objects refer to any dynamic or static object within the target lane and located in front of the target object's path of travel, which constitutes a source of forward collision risk, such as a slow-moving vehicle or a stationary obstacle ahead.

[0077] This step is triggered when the system detects a vehicle beginning to overtake or change lanes, and is used to assess the forward collision risk posed by occupying an adjacent lane. Specifically, the system first identifies the "target lane" the vehicle is about to enter based on the vehicle's pose and lane line recognition results. Then, the system immediately focuses its perception resources on the area ahead of that lane, continuously searching and marking any objects within this area as "potential collision targets," and recording their position, speed, and other information. This step filters out the main collision threats in overtaking scenarios, providing clear and accurate target input for subsequent risk calculations (such as TTC).

[0078] S302. After a potential conflict object is found ahead of the target object's driving direction, calculate the estimated collision time between the target object and the potential conflict object.

[0079] Specifically, after acquiring the ID of a potential collision object, the system immediately retrieves the complete kinematic state information of that object and the target object (the vehicle itself) from the trajectory data stream. This includes at least their precise positions and longitudinal velocities. The system first calculates the longitudinal distance (Δd) between the two objects along their travel paths, as well as their relative longitudinal velocities (Δv). A crucial step is determining whether the two objects are approaching or moving away from each other. The calculation of the Time to Collision (TTC) is only meaningful when the target object's velocity is faster than the potential collision object (Δv > 0), i.e., when they are approaching. At this point, the system calculates the estimated collision time using the formula TTC = Δd / Δv. If the two objects are moving away from each other or their relative velocity is zero, the TTC can be considered infinite, representing no collision risk in the current state. This step transforms the abstract "possibility" of danger into a concrete, quantifiable time value, providing a crucial quantitative input representing the "scenario risk" for the subsequent risk fusion model (S303).

[0080] S303. Based on the estimated collision time and kinematic hazard score, a scenario hazard score representing the comprehensive hazard level of the target overtaking event is generated through a preset risk fusion model.

[0081] This step involves a comprehensive, multi-dimensional risk assessment of overtaking events, avoiding the bias inherent in single-dimensional assessments. Specifically, the system has two key inputs: a kinematic hazard score representing "self-operational risk" and a predicted time of collision (TTC) representing "external environmental risk." Since TTC is measured in seconds, a smaller value indicates higher risk. Therefore, it needs to be transformed into a positive "scenario risk index" with the same dimensions as the kinematic hazard score using a mapping function (such as an inverse proportional function or a piecewise function). Subsequently, the risk fusion model uses these two risk indices as input and calculates them using pre-defined logic (such as weighted summation or fuzzy inference) to ultimately generate a "scenario hazard score" that comprehensively reflects the overall risk level of the current overtaking event. This step represents a leap from "single risk source" assessment to "multi-risk source comprehensive assessment," organically combining driver behavior risk with external environmental risk. This makes the risk assessment results more closely resemble the complexities of real-world situations, significantly improving the accuracy and reliability of the early warning system.

[0082] S304. Based on the scene hazard score, generate corresponding risk warning information. The magnitude of the scene hazard score determines at least one output attribute of the risk warning information. The output attributes include the type of warning information, the intensity of the warning information, and the presentation method of the warning information.

[0083] This step is triggered after the "scenario hazard score" is calculated. Its core function is to transform the abstract risk score into a concrete warning that the driver can intuitively perceive. Specifically, the system has a pre-set mapping table between risk levels and warning strategies. Upon receiving a scenario hazard score, the system first matches it to the corresponding hazard level (e.g., low, medium, high), and then retrieves the corresponding output attribute combination from the strategy library. For example, a high hazard score will trigger a combined warning consisting of high-intensity visual, auditory, and tactile signals. The technical effect of this step is to implement a graded warning mechanism that precisely matches the level of risk. By dynamically adjusting warning attributes, it avoids excessive interference with the driver while ensuring safety, thus improving the effectiveness of human-computer interaction.

[0084] The above method analyzes the potential conflict ahead to assess the risk of overtaking in the dynamic traffic flow, generating a scenario hazard score. This score primarily reflects the danger of dynamic interactions and does not yet consider the static risks of the road itself. To achieve a unified risk assessment across the spatiotemporal dimensions, the following implementation introduces high-risk geofencing (such as curves and tunnels). By determining whether the overtaking event occurred within these areas and combining this with the scenario hazard score, a final comprehensive hazard level that takes into account geographical environmental factors is generated.

[0085] The following provides a more detailed description of the process of the method provided in this implementation. Please refer to [link / reference]. Figure 4This is another flowchart illustrating the geofencing comprehensive risk assessment method in this application embodiment.

[0086] S401. Obtain map data for high-risk geofences; high-risk geofences are pre-marked geographical locations that indicate bends, ramps, tunnels, or ramp entrances. High-risk geofences refer to virtual geographic boundaries pre-defined on digital maps to identify specific road sections with inherent overtaking risks, such as curves and tunnels. In this application, map data specifically refers to a dedicated data layer containing the location, extent, and type attributes of these high-risk geofences.

[0087] This step, serving as data preparation before risk assessment, is typically executed during system startup or route planning. It preloads static risk information related to the road. Specifically, the onboard computing unit requests and retrieves map data from a local high-precision map or cloud-based map service, identifying and caching high-risk geofence layers within the data. This step loads the inherent static risk information of the road into the system, providing a data foundation for the subsequent spatial location determination in step S402.

[0088] S402. Obtain the geographical location of the target overtaking event and determine whether the geographical location falls within the high-risk geofence, and obtain the judgment result.

[0089] Specifically, once the perception fusion module confirms the overtaking intention or action, the system immediately obtains the vehicle's current precise geographic coordinates from the positioning module. Subsequently, the computing unit matches these coordinates with the high-risk geofence data loaded into memory in step S401. This calculation is essentially a geometric judgment of "point within polygon," used to determine whether the vehicle's current position has entered any predefined danger zone. The direct technical effect of this step is to output a Boolean (yes / no) result, which clarifies whether the overtaking behavior occurred in a statically high-risk geographic environment, providing crucial spatial scene input for the subsequent risk fusion decision in S403.

[0090] S403. Based on the scene hazard score and the judgment result of whether the geographical location falls within the high-risk geofence, the final comprehensive hazard level is generated through preset spatiotemporal fusion rules.

[0091] The pre-defined spatiotemporal fusion rules represent a decision-making logic or mathematical model, whose core function is to integrate dynamic "scenario" risks with static "geographical" risks. The final comprehensive risk level is the conclusion output after fusion calculation, usually expressed as a classification such as "low," "medium," "high," or "dangerous," used to intuitively represent the overall risk level of this overtaking behavior.

[0092] This step is the final decision-making stage in the entire overtaking risk assessment process, executed after S402 completes the spatial judgment. Specifically, the calculation unit sends two dimensions of input—the "scene hazard score" representing dynamic risk and the "geographical location judgment result" (yes / no) representing static risk—to the spatiotemporal fusion rule module. The core of this rule is that the geographical location judgment result plays a crucial role in correcting or weighting the scene hazard score. For example, even if the scene score is low (e.g., few surrounding vehicles), if the judgment indicates the vehicle is within a high-risk geographical fence such as a curve or tunnel, the fusion rule will force an increase in the final hazard level. The direct technical effect of this step is to output a more accurate and comprehensive hazard level than a single-dimensional assessment, effectively avoiding misjudgments caused by neglecting static geographical risks.

[0093] In some embodiments, the method further includes creating a driving behavior profile for the target vehicle: within a preset time period, multiple final comprehensive hazard levels associated with a specific vehicle are aggregated to form a hazard level time series; statistical analysis is performed on the hazard level time series to extract at least one driving behavior feature, and a driving behavior profile for the specific vehicle is generated based on the driving behavior feature.

[0094] Driving behavior characteristics are indicators extracted from the time series using statistical methods, quantifying driving habits such as the frequency of high-risk events and average hazard level. After multiple interactions with a specific vehicle (continuously trackable via unique identifiers like vehicle ID), driving patterns are extracted from historical behavior. Specifically, the system establishes a temporary database of continuously tracked vehicles in the vicinity, aggregating their "final comprehensive hazard level" assessed by the vehicle each time. When the data accumulates to a certain amount or reaches a preset time period, the system initiates an analysis program to statistically process the vehicle's hazard level time series, calculating driving behavior characteristics such as the proportion of high-risk levels, the mean, and variance of hazard levels. Finally, based on these characteristics, the vehicle is categorized into a preset driving behavior profile. This step upgrades instantaneous, isolated risk assessments to long-term, trend-based behavioral cognition of other road users, providing deeper predictive input for the vehicle's decision-making and planning (such as adjusting following distance and choosing overtaking opportunities).

[0095] The following describes the non-standard overtaking behavior recognition device in the embodiments of this invention from the perspective of hardware processing. Please refer to [link / reference]. Figure 5 This is a schematic diagram of the physical device structure of the non-standard overtaking behavior recognition device in the embodiments of this application.

[0096] It should be noted that, Figure 5 The structure of the irregular overtaking behavior recognition device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0097] like Figure 5 As shown, the irregular overtaking behavior recognition device includes a Central Processing Unit (CPU) 501, which can perform various appropriate actions and processes based on a program stored in Read-Only Memory (ROM) 502 or a program loaded from storage section 508 into Random Access Memory (RAM) 503, such as performing the methods described in the above embodiments. The RAM 503 also stores various programs and data required for system operation. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0098] The following components are connected to I / O interface 505: input section 506 including audio input devices, push-button switches, etc.; output section 507 including liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 508 including hard disks, etc.; and communication section 509 including network interface cards such as LAN (Local Area Network) cards, modems, etc. Communication section 509 performs communication processing via a network such as the Internet. Drive 510 is also connected to I / O interface 505 as needed. Removable media 511, such as disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on drive 510 as needed so that computer programs read from them can be installed into storage section 508 as needed.

[0099] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is executed by central processing unit (CPU) 501, it performs the various functions defined in the present invention.

[0100] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0101] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0102] Specifically, the non-standard overtaking behavior recognition device in this embodiment includes a processor and a memory. The memory stores a computer program, and when the computer program is executed by the processor, it implements the non-standard overtaking behavior recognition method provided in the above embodiment.

[0103] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the irregular overtaking behavior recognition device described in the above embodiments; or it may exist independently and not assembled into the irregular overtaking behavior recognition device. The storage medium carries one or more computer programs, which, when executed by a processor of the irregular overtaking behavior recognition device, cause the irregular overtaking behavior recognition device to implement the irregular overtaking behavior recognition method provided in the above embodiments.

[0104] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0105] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0106] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for identifying irregular overtaking behavior, characterized in that, The method, applied to a system for recognizing irregular overtaking behavior, includes: Acquire trajectory data streams of multiple vehicles traveling in parallel, wherein the trajectory data streams include at least real-time position coordinates, driving heading angles, speed data, and timestamps, and the trajectory data streams correspond one-to-one with the vehicles; Based on the timestamp and the real-time location coordinates, the current trajectory data stream is horizontally correlated in real time and determined as an interactive analysis group. The current trajectory data stream is the trajectory data stream of at least two vehicles located in a preset spatial adjacent area within the same time window. Within the interactive analysis group, the real-time location coordinates are converted into radian coordinates; Within the interactive analysis group, select a real-time location coordinate as a reference object; The absolute azimuth angles between the reference object and the target object are calculated based on the radian coordinates and spherical trigonometric functions; the target object is another real-time position coordinate in the interactive analysis group. The relative azimuth angle is obtained by calculating the difference between the absolute azimuth angle and the driving heading angle of the reference object. The relative azimuth angle is the azimuth angle with the driving direction of the reference object as the reference. By mapping the relative azimuth angle to a preset standard range, the front-back relationship data between the target object and the reference object is obtained; Extract the velocity data from the front-back relationship data when the front-back relationship between the target object and the reference object changes from back to front. When the speed data meets the preset target overtaking feature rules, a target overtaking event is determined and generated.

2. The method according to claim 1, characterized in that, The determination steps for the target overtaking feature rule include: Based on the trajectory data streams of the target object and the reference object, the minimum lateral distance, the longitudinal cut-off distance, and the relative speed difference are calculated; the minimum lateral distance is the minimum value of the lateral distance between the target object and the reference object during parallel driving; the longitudinal cut-off distance is the distance between the front ends of the target object and the reference object after overtaking; the relative speed difference is the speed difference between the corresponding vehicles of the target object and the reference object during the overtaking process. Based on the minimum lateral distance, the longitudinal turning distance, and the relative speed difference, a kinematic hazard score is generated using a preset quantization model to characterize the degree of danger of the target overtaking event.

3. The method according to claim 2, characterized in that, The method further includes: Identify the target lane used by the target object during the overtaking process, and search for potential conflicting objects ahead of the target object in the target lane in the direction of travel of the target object; After a potential conflict object is detected ahead of the target object's direction of travel, the estimated collision time between the target object and the potential conflict object is calculated. Based on the predicted collision time and the kinematic hazard score, a scenario hazard score representing the overall hazard level of the target overtaking event is generated through a preset risk fusion model.

4. The method according to claim 3, characterized in that, The method further includes: Obtain map data for high-risk geofences; the high-risk geofences are pre-marked geographical locations indicating locations including curves, ramps, tunnels, or ramp entrances; Obtain the geographical location of the target overtaking event, determine whether the geographical location falls within the high-risk geofence, and obtain the determination result; Based on the scenario hazard score and the judgment result of whether the geographical location falls within the high-risk geofence, a final comprehensive hazard level is generated through preset spatiotemporal fusion rules.

5. The method according to claim 1, characterized in that, The target overtaking feature rules also include: The target overtaking feature rule is satisfied when any of the following conditions are met: Determine whether the speed of the target object is greater than the speed of the reference object, and whether the speed difference between the two is less than a preset first speed threshold; Determine whether the speed of the target object exceeds the legal speed limit of the road in which it is located; Determine whether the speed difference between the target object and the reference object is greater than a preset second speed threshold.

6. The method according to claim 3, characterized in that, The method further includes: Based on the scenario hazard score, corresponding risk warning information is generated. The magnitude of the scenario hazard score determines at least one output attribute of the risk warning information. The output attribute includes the type of warning information, the intensity of the warning information, and the presentation method of the warning information.

7. The method according to claim 4, characterized in that, The method further includes: Within a preset time period, multiple final comprehensive hazard levels associated with a specific vehicle are aggregated to form a hazard level time series; Statistical analysis is performed on the time series of the hazard level to extract at least one driving behavior feature, and a driving behavior profile of the specific vehicle is generated based on the driving behavior feature.

8. A device for recognizing irregular overtaking behavior, characterized in that, The irregular overtaking behavior recognition device includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the irregular overtaking behavior recognition device to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the non-standard overtaking behavior recognition device, the non-standard overtaking behavior recognition device performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the non-standard overtaking behavior recognition device, the non-standard overtaking behavior recognition device performs the method as described in any one of claims 1-7.

Images