A method and system for actively identifying traffic accidents based on vehicle networking

Abstract

The application discloses a kind of method and system of active identification traffic accident based on Internet of Vehicles, it is related to accident identification technical field.By confirming and testing current vehicle operating state gradually, the accuracy and reliability of traffic accident detection can be significantly improved.Real-time acquisition vehicle operating state data and preliminary abnormality judgment based on preset basic threshold, by combining historical operating state data to calculate trend difference, and set standard threshold to further verify the authenticity of preliminary abnormality event, significant test is carried out on trend difference, its statistical significance is evaluated by error probability, ensure that only truly meaningful abnormal event is marked as potential accident, to avoid misjudgment caused by random noise or short-term fluctuations.Integrated analysis trend difference and operating state data using machine learning model, output specific traffic accident type, so that the system can not only confirm abnormal state, but also accurately classify different types of traffic accident.

Description

Technical Field

[0001] This invention relates to the field of accident recognition technology, and in particular to a method and system for actively recognizing traffic accidents based on the Internet of Vehicles. Background Technology

[0002] With the rapid development of Intelligent Transportation Systems (ITS), Internet of Vehicles (IoV) technology is gradually becoming an important means to improve road traffic safety. Through real-time communication between vehicles (V2V) and between vehicles and infrastructure (V2I), IoV can collect and analyze vehicle operating status data (such as speed, acceleration, and heading angle), thereby enabling the identification and early warning of potential traffic accidents. However, existing accident identification methods still have significant technical shortcomings, especially in the mechanism for determining abnormal events.

[0003] Currently, most vehicle-to-everything (V2X) accident detection systems rely primarily on a single threshold-based method. Specifically, these systems preset a fixed threshold for vehicle operating status data (such as speed and acceleration). When detected data exceeds this threshold, the system identifies it as an abnormal event and triggers a corresponding warning mechanism. While this method is simple and easy to implement, it has revealed numerous problems in practical applications:

[0004] Vehicles are inevitably affected by various external factors during operation, such as changes in road conditions, weather conditions, and driver operating habits. These factors may cause drastic fluctuations in vehicle operating data within a short period of time, but such fluctuations do not necessarily indicate the occurrence of an abnormal event. For example:

[0005] When braking suddenly on a wet or slippery surface, the acceleration may increase dramatically in a short period of time.

[0006] When overtaking or swerving on a highway, a vehicle's speed and heading angle may change drastically.

[0007] However, because current technology relies solely on a single threshold determination method, these normal driving behaviors may be misjudged as abnormal events, leading to frequent false alarms from the system. Such false alarms not only reduce the system's reliability but may also interfere with the driver's normal operation and even trigger unnecessary panic. Summary of the Invention

[0008] The purpose of this invention is to provide a method and system for proactively identifying traffic accidents based on the Internet of Vehicles, in order to solve the problem that relying solely on a single threshold determination method may lead to normal driving behaviors being misjudged as abnormal events, resulting in the system frequently issuing false alarms.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a method for actively identifying traffic accidents based on the Internet of Vehicles, comprising:

[0010] Step 100: Acquire the current vehicle's operating status data in real time and preset a basic threshold for the operating status data. When the operating status data is greater than the basic threshold, the current vehicle's operating status is determined to be a preliminary abnormal event and step 200 is executed; when the operating status data is less than or equal to the basic threshold, the current vehicle's operating status is determined to be a normal event.

[0011] Step 200: Obtain historical operating status data corresponding to the operating status data; obtain trend differences based on historical operating status data and operating status data; and preset a standard threshold. When the trend difference is greater than the standard threshold, mark the preliminary abnormal event as an event to be analyzed and execute step 300; when the trend difference is less than or equal to the standard threshold, mark the preliminary abnormal event as a normal event.

[0012] Step 300: Perform a significance test on the trend difference to obtain the error probability; and preset a test threshold. When the error probability is less than or equal to the test threshold, the preliminary abnormal event is defined as an abnormal event and step 400 is executed; when the error probability is greater than the test threshold, the preliminary abnormal event is defined as a normal event.

[0013] Step 400: Build a machine learning model by inputting trend differences and operational status data into the machine learning model and outputting the traffic accident type.

[0014] In another implementation, real-time acquisition of the vehicle's operating status data includes: acquiring the vehicle's current time and data prior to the current time. The operational status data at each historical moment are integrated into real-time time series data in chronological order.

[0015] In another implementation, both historical operating status data and operating status data include: speed, acceleration, heading angle, environmental data, and tire pressure.

[0016] In another implementation, historical operating status data corresponding to the operating status data is obtained, and trend differences are derived based on the historical operating status data and the operating status data, including the following steps:

[0017] Step 210: Obtain historical operating status data corresponding to the same time period as the operating status data, and integrate them into historical time series data in chronological order;

[0018] Step 220: Define weighting coefficients for each time point in the real-time time series data and the historical time series data based on the running status data and historical running status data;

[0019] Step 230: Define environmental impact factors based on operational status data and historical operational status data;

[0020] Step 240: Based on the operating status data and historical operating status data, calculate the absolute deviation and mean square error between the historical time series data and the real-time time series data;

[0021] Step 250: Calculate the heading angle deviation between historical time series data and real-time time series data based on the heading angles in the operational status data and historical operational status data; and obtain the maximum heading angle from the historical time series data and real-time time series data.

[0022] Step 260: Calculate the trend difference based on the weighting coefficient, environmental impact factor, absolute deviation, mean square error, heading angle deviation, and maximum heading angle.

[0023] In another implementation, environmental impact factors are defined based on operational status data and historical operational status data, including the following steps:

[0024] Step 231: Determine environmental factors based on environmental data;

[0025] Step 232: Define a standard tire pressure, and calculate the braking distance change ratio based on the tire pressure and the standard tire pressure.

[0026] Step 233: Calculate the acceleration deviation between historical time series data and real-time time series data based on the acceleration in the running status data and historical running status data; and obtain the maximum acceleration from the historical time series data and real-time time series data.

[0027] Step 234: Calculate the environmental impact factor based on the environmental factors, the braking distance change ratio, the acceleration deviation, and the maximum acceleration.

[0028] In another implementation, constructing a machine learning model includes: acquiring sample data using oversampling and undersampling methods, each sample data including a set of historical operating data; determining historical trend deviation, historical acceleration, historical heading angle, historical cumulative rate of change of speed, and historical environmental data based on the sample data; using the historical trend deviation, historical acceleration, historical heading angle, historical cumulative rate of change of speed, and historical environmental data as the training set for the machine learning model, and labeling each historical operating data with a traffic accident type; training the machine learning model using the training set to obtain a trained machine learning model.

[0029] In another implementation, operational status data and trend differences are input into a trained machine learning model, which outputs the type of traffic accident.

[0030] This invention also discloses a system for actively identifying traffic accidents based on the Internet of Vehicles, comprising the following modules:

[0031] Preliminary Abnormal Event Detection Module: This module is used to acquire the current vehicle's operating status data in real time and preset a basic threshold for the operating status data. When the operating status data is greater than the basic threshold, the current vehicle's operating status is determined to be a preliminary abnormal event; when the operating status data is less than or equal to the basic threshold, the current vehicle's operating status is determined to be a normal event.

[0032] Continue analyzing the event labeling module: connected to the preliminary abnormal event judgment module, used to obtain historical operating status data corresponding to the operating status data, obtain trend differences based on historical operating status data and operating status data; and preset a standard threshold. When the trend difference is greater than the standard threshold, the preliminary abnormal event is marked as an event to continue analysis; when the trend difference is less than or equal to the standard threshold, the preliminary abnormal event is marked as a normal event.

[0033] Significance testing module: Connected to the continued analysis event labeling module, it is used to perform significance testing on trend differences to obtain error probabilities; and preset a testing threshold. When the error probability is less than or equal to the testing threshold, the preliminary abnormal event is defined as an abnormal event; when the error probability is greater than the testing threshold, the preliminary abnormal event is defined as a normal event.

[0034] Prediction module: Connected to the significance test module, it is used to build a machine learning model. Trend differences and operational status data are input into the machine learning model, and the output is the type of traffic accident.

[0035] The technical effects and advantages of this invention are as follows:

[0036] This invention significantly improves the accuracy and reliability of traffic accident detection by progressively confirming and verifying the current vehicle operating status. First, real-time vehicle operating status data is acquired and preliminary anomaly judgments are made based on preset thresholds. This process aims to quickly identify potential risk states. Next, trend differences are calculated by combining historical operating status data, and standard thresholds are set to further verify the authenticity of preliminary anomalies. This method not only utilizes real-time data but also incorporates historical context information to enhance the accuracy of the judgment. Subsequently, the trend differences are tested for significance, and their statistical significance is assessed using error probability to ensure that only truly meaningful anomalies are marked as potential accidents, thus avoiding misjudgments due to random noise or short-term fluctuations. Finally, a machine learning model is used to comprehensively analyze trend differences and operating status data, outputting specific traffic accident types. This allows the system to not only confirm anomalies but also accurately classify different types of traffic accidents. Overall, this method, through multi-stage confirmation and verification of the current vehicle operating status, combined with real-time monitoring, historical data analysis, and advanced statistical and machine learning techniques, achieves efficient and accurate traffic accident identification, greatly improving the system's reliability and practicality, and helping to take timely measures to prevent or mitigate the impact of traffic accidents. Attached Figure Description

[0037] Figure 1 A flowchart illustrating a method for proactively identifying traffic accidents based on the Internet of Vehicles (IoV) provided in this embodiment;

[0038] Figure 2 This is a structural diagram of a system for actively identifying traffic accidents based on the Internet of Vehicles (IoV) provided in this embodiment. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] This embodiment significantly improves the accuracy and reliability of traffic accident detection by progressively confirming and verifying the current vehicle operating status. First, real-time vehicle operating status data is acquired and preliminary anomaly judgments are made based on preset thresholds. This process aims to quickly identify potential risk states. Next, trend differences are calculated by combining historical operating status data, and standard thresholds are set to further verify the authenticity of preliminary anomalies. This method not only utilizes real-time data but also incorporates historical background information to enhance the accuracy of the judgment. Subsequently, the trend differences are tested for significance, and their statistical significance is assessed using error probability to ensure that only truly meaningful anomalies are marked as potential accidents, thereby avoiding misjudgments due to random noise or short-term fluctuations. Finally, a machine learning model is used to comprehensively analyze trend differences and operating status data, outputting specific traffic accident types. This allows the system to not only confirm anomalies but also accurately classify different types of traffic accidents. Overall, this method, through multi-stage confirmation and verification of the current vehicle operating status, combined with real-time monitoring, historical data analysis, and advanced statistical and machine learning techniques, achieves efficient and accurate traffic accident identification, greatly improving the system's reliability and practicality, and helping to take timely measures to prevent or mitigate the impact of traffic accidents.

[0041] like Figure 1 As shown, this embodiment discloses a method for actively identifying traffic accidents based on the Internet of Vehicles, including:

[0042] Step 100: Acquire the current vehicle's operating status data in real time and preset a basic threshold for the operating status data. When the operating status data is greater than the basic threshold, the current vehicle's operating status is determined to be a preliminary abnormal event and step 200 is executed; when the operating status data is less than or equal to the basic threshold, the current vehicle's operating status is determined to be a normal event.

[0043] Operational status data includes: speed, acceleration, heading angle, environmental data, and tire pressure.

[0044] Speed: Obtained via the vehicle's built-in speed sensor.

[0045] Acceleration: The acceleration or deceleration of a vehicle is measured using an accelerometer.

[0046] Environmental data, such as temperature and humidity, can be collected by environmental sensors installed on the outside of the vehicle.

[0047] Tire pressure: Monitored by pressure sensors inside each tire.

[0048] The heading angle is an important parameter describing a vehicle's current direction of travel, typically used to characterize the angle of deviation of the vehicle relative to a reference direction (such as geographic north). More specifically, the heading angle is the angle between the vehicle's current direction of travel and geographic north, usually positive clockwise, and ranges from 0° to 360°. For example: a heading angle of 0° indicates the vehicle is traveling due north; a heading angle of 90° indicates the vehicle is traveling east; a heading angle of 180° indicates the vehicle is traveling due south; and a heading angle of 270° indicates the vehicle is traveling west.

[0049] The calculation of the heading angle usually relies on the GPS system, and the specific method is as follows:

[0050] GPS-based heading angle calculation

[0051] GPS systems calculate a vehicle's coordinates (latitude and longitude) on the Earth's surface by receiving signals from multiple satellites. When a vehicle is moving, its direction of travel can be estimated by acquiring its coordinates at two consecutive points in time.

[0052] Let the coordinates of the first position be... , The longitude of the first location. Here are the latitudes of the first location. The coordinates of the second location are... , The longitude of the second location. The latitude is the coordinate of the second location. Based on the coordinate difference between the two locations, calculate the vehicle's azimuth angle (i.e., heading angle):

[0053] ;

[0054] ;

[0055] ; Due to longitude difference, Due to latitude difference, It is a four-quadrant arctangent function, ensuring that the calculation result is within the range of 0° to 360°.

[0056] The basic threshold was set based on the analysis of a large amount of historical data:

[0057] Speed: Taking into account road speed limits and safe driving standards, a basic speed threshold is established by adding a certain buffer value (10%) to the maximum permissible speed for a specific road section. .

[0058] Acceleration: For acceleration, a threshold can be set by analyzing the maximum acceleration range under normal driving behavior. For example, on ordinary urban roads, sudden braking or sudden acceleration may exceed the range of -3m / s² to 3m / s², so anything outside this range can be considered abnormal. Therefore, the basic threshold for acceleration is [-3, 3].

[0059] Heading angle change rate: If a vehicle undergoes a significant change in direction within a short period of time (e.g., exceeding 30 degrees / second), it may indicate an emergency avoidance or other dangerous situation. This type of change rate can be used as the basic threshold for heading angle change rate: [0, 30].

[0060] Environmental data and tire pressure: For environmental data and tire pressure, refer to the manufacturer's recommended safe operating range. For example, if the tire pressure is 20% below the recommended value, it increases braking distance and affects driving safety; therefore, [20%, 100%] is used as the basic threshold for tire pressure.

[0061] If the operating status data is greater than the basic threshold, it means that any one of the above operating status data is not within the corresponding basic threshold, and the current vehicle operating status is judged as a preliminary abnormal event; if the operating status data is less than or equal to the basic threshold, it means that all the above operating status data are within the corresponding basic threshold, and the current vehicle operating status is judged as a preliminary abnormal event.

[0062] When the operating status data exceeds a preset threshold, the current vehicle operating status is judged as a preliminary abnormal event. This is because data exceeding the threshold usually indicates that certain key vehicle parameters (such as speed and acceleration) have deviated from the normal operating range, potentially posing a safety hazard. However, simply relying on data exceeding the threshold at a single moment is insufficient to definitively conclude that a true anomaly has occurred. In real-world driving environments, numerous factors can cause temporary data fluctuations, such as emergency avoidance, rapid lane changes, or short-term acceleration and deceleration. In these situations, although the data momentarily exceeds the set threshold, it does not necessarily mean that a traffic accident will occur or that the vehicle is in an unsafe state. Therefore, further processing of trend differences is needed (step 200). By analyzing the data's changing trend over a period of time, short-term data fluctuations can be distinguished from persistent abnormal states, thereby reducing false alarm rates and improving the accuracy of the warning system.

[0063] Conversely, when the operational status data is less than or equal to the baseline threshold, the system classifies it as a normal event. This is based on the following considerations: Under normal driving conditions, the vehicle's various operating parameters should remain within a relatively stable and safe range, without frequently touching or exceeding the set safety limits. This means that if all monitored data points do not exceed the threshold, it can be reasonably inferred that the vehicle is currently in a stable and controllable driving state, with no immediate risk of causing an accident. Therefore, in this case, there is no need to initiate a more complex analysis process; it can be directly marked as a normal event. This not only simplifies the system's computational burden but also avoids a large number of false alarms caused by oversensitivity, allowing the driver and the monitoring system to focus more on the situations that truly require attention.

[0064] Step 200: Obtain historical operating status data corresponding to the operating status data, and obtain the trend difference based on the historical operating status data and the operating status data; and preset a standard threshold. When the trend difference is greater than the standard threshold, mark the preliminary abnormal event as an event to be analyzed and execute step 300; when the trend difference is less than or equal to the standard threshold, mark the preliminary abnormal event as a normal event.

[0065] The process of obtaining historical operational status data corresponding to operational status data, and determining trend differences based on historical operational status data and operational status data, includes the following steps:

[0066] Step 210: Obtain historical operating status data corresponding to the same time period as the operating status data, and integrate them into historical time series data in chronological order.

[0067] Real-time acquisition of the vehicle's current operating status data, including: the current vehicle's current moment and the time elapsed before the current moment. The operational status data at each historical moment is integrated into real-time time series data in chronological order. It's worth noting that the operational status data and historical operational status data are collected at a certain frequency (e.g., once per second / hour / day, etc.), using a sliding window technique, meaning that not only current data but also past data are recorded. Data from each time point is used to form a real-time time series data stream.

[0068] Historical operational status data refers to data collected during the same period as the operational status data. For example, operational status data is data collected on April 12, 2025, with an hourly collection frequency; operational status data is data collected every hour on April 12, 2025; historical operational data refers to data collected every hour on April 11, 2025.

[0069] Historical operating status data and operating status data both include: speed, acceleration, heading angle, environmental data, and tire pressure.

[0070] Step 220: Define weighting coefficients for each time point in the real-time time series data and the historical time series data based on the running status data and historical running status data;

[0071] The weighting coefficient for each time point is calculated based on the velocity at each time point in the real-time time series data and the maximum velocity in both the real-time and historical time series data.

[0072] ; For the first Weighting coefficients at each time point For the first in real-time time series data The speed at each point in time This represents the maximum speed in both real-time and historical time-series data.

[0073] Weighting coefficients are used to quantify the importance of data at different time points. When the speed at a certain time point is close to or equal to the maximum speed, its weighting coefficient is close to 1, indicating that the data at that time point has a significant impact on the overall trend and may require further attention. Conversely, if the speed is much lower than the maximum speed, its weighting coefficient is smaller, indicating that the data at that time point is relatively stable and has a smaller impact on the overall trend. This design helps the system more accurately identify abnormal events, reduces false alarms caused by data fluctuations, and ensures timely response when real anomalies occur. By comparing the speed at each time point with the maximum speed, those time points that are close to or exceed the maximum speed can be highlighted, which helps identify potential abnormal events and has a strong theoretical basis.

[0074] Step 230: Define environmental impact factors based on operational status data and historical operational status data.

[0075] Step 231: Determine environmental factors based on environmental data.

[0076] In rainy conditions, slippery roads reduce the friction between tires and the ground, significantly increasing braking distance. Rain increases braking distance by 50%, hence the environmental impact factor for rainy weather is 1.5, meaning that at the same speed, braking distance in the rain is 1.5 times that in sunny weather. This reflects the need for a longer distance to safely stop in the rain, increasing driving risk.

[0077] In contrast, under sunny conditions, the road surface is dry and the friction is high, resulting in better vehicle braking performance. The environmental impact factor for sunny conditions is 1.0, indicating that under normal circumstances, the braking distance and driving safety in sunny conditions are at the baseline level.

[0078] Besides rainy and sunny days, other environmental conditions can also affect the operating status of vehicles, such as:

[0079] Snowy weather: The road surface is more slippery in snowy weather, the friction is further reduced, and a longer braking distance is required. The environmental impact factor is 1.8.

[0080] Foggy weather: Foggy weather reduces visibility and prolongs the driver's reaction time. Although it does not directly affect braking distance, it increases driving difficulty and risk. The environmental impact factor is 1.3.

[0081] Nighttime: Insufficient light at night limits the driver's visibility, which also increases the difficulty of driving. The environmental impact factor is 1.3.

[0082] Step 232: Define a standard tire pressure and calculate the braking distance change ratio based on the tire pressure and the standard tire pressure.

[0083] The standard tire pressure is determined by the vehicle manufacturer based on factors such as the vehicle's weight, tire size, and design performance, and is indicated in the vehicle owner's manual or on the tire sidewall.

[0084] ; For the current moment The percentage change in braking distance The current tire pressure. Standard tire pressure This is the tire pressure adjustment factor.

[0085] The process of obtaining the tire pressure adjustment coefficient is as follows:

[0086] First, we collected a large amount of historical data, including actual braking distances under different tire pressures, as shown in the table below:

[0087]

[0088] Then, calculate the ratio of change in tire pressure to braking distance:

[0089] Using the data from serial number 1 as a baseline (standard pressure 2.5 bar, braking distance 40 m), calculate the pressure change ratio and braking distance change ratio for other data points.

[0090] Serial number 2:

[0091] The calculated pressure change percentage is 0.1, which means a decrease of 10%.

[0092] The calculated change in braking distance is 0.02, which is an increase of 2%.

[0093] Serial number 3:

[0094] The calculated pressure change percentage is 0.2, which means a decrease of 20%.

[0095] The calculated change in braking distance is 0.04, which is an increase of 4%.

[0096] Serial number 4:

[0097] The calculated pressure change percentage is 0.3, which means a decrease of 30%.

[0098] The calculated change in braking distance is 0.06, which is an increase of 6%.

[0099] Serial number 5:

[0100] The calculated pressure change percentage is 0.4, which is a decrease of 40%.

[0101] The calculated change in braking distance is 0.08, which is an increase of 8%.

[0102] During the analysis, a linear relationship was found between tire pressure and braking distance. That is, when tire pressure decreases by a certain percentage, the braking distance increases by a fixed percentage accordingly. When tire pressure is gradually reduced from the standard value of 2.5 bar to 1.5 bar, a regular increase in braking distance can be observed as the pressure decreases. This regular increase indicates an approximately linear relationship between the two; therefore, in this embodiment, the 2% increase in braking distance (number 2) is used as the tire pressure adjustment coefficient.

[0103] Step 233: Calculate the acceleration deviation between historical time series data and real-time time series data based on the acceleration in the running status data and historical running status data; and obtain the maximum acceleration from the historical time series data and real-time time series data.

[0104] Step 234: Calculate the environmental impact factor based on the environmental factors, the braking distance change ratio, the acceleration deviation, and the maximum acceleration.

[0105] ;

[0106] In the formula, For the current moment Environmental factors, For the current moment The percentage change in braking distance For the current moment acceleration, For a historic moment Historical acceleration For acceleration deviation, For maximum acceleration, This is the weighted index.

[0107] By combining environmental factors, the percentage change in braking distance, and the ratio of acceleration deviation to maximum acceleration, a foundation is provided for the accurate assessment of trend differences. Specifically, environmental factors and the percentage change in braking distance reflect the impact of external conditions on vehicle driving safety, while acceleration deviation captures the difference between the current operating state and historical patterns. This design ensures that trend differences consider not only direct changes in speed or acceleration but also indirect but crucial factors such as weather conditions and tire pressure. In this way, the formula enhances the accuracy and comprehensiveness of trend difference calculations, helping to more finely distinguish between short-term fluctuations in normal operation and actual potential risks, thereby improving the overall system's ability and reliability in identifying abnormal events.

[0108] This formula is designed based on the theory of multi-factor comprehensive assessment, aiming to accurately identify potential anomalies in vehicle operating conditions by quantifying different environmental influencing factors. Its core principle lies in the impact of environmental conditions (such as weather conditions) on driving safety. The percentage increase in braking distance due to changes in tire pressure This, along with the deviation between real-time and historical acceleration, is integrated to form a comprehensive indicator reflecting the vehicle's current risk status. Specifically, and The system captures changes in both the external environment and the vehicle's mechanical state, which together affect vehicle safety. The introduction of the ratio of acceleration deviation to maximum acceleration measures the degree to which the vehicle's current behavior deviates from the normal pattern, helping to distinguish between routine driving operations and potentially dangerous behavioral patterns. The entire formula follows the principles of multi-level data analysis in systems engineering, namely, achieving refined management of complex systems through weighted combination of data from multiple dimensions, thus providing solid theoretical support and technical assurance for the proactive identification of traffic accidents in the Internet of Vehicles (IoV).

[0109] Step 240: Based on the operating status data and historical operating status data, calculate the absolute velocity deviation between historical time series data and real-time time series data, and the mean square error of velocity between historical time series data and real-time time series data.

[0110] Step 250: Calculate the heading angle deviation between historical time series data and real-time time series data based on the heading angles in the operating status data and historical operating status data; and obtain the maximum heading angle from the historical time series data and real-time time series data.

[0111] Step 260: Calculate the trend difference based on the weighting coefficient, environmental impact factor, absolute speed deviation, mean square speed error, heading angle deviation, and maximum heading angle.

[0112] The formula for calculating trend difference is:

[0113] ;

[0114] In the formula, Due to trend differences, For the first Weighting coefficients at each time point The heading angle for real-time time series data. For the first in real-time time series data The heading angle at each point in time, The heading angle for historical time series data. For the first in historical time series data The heading angle at each point in time, For heading angle deviation, This is the heading angle deviation adjustment factor. For the speed of real-time time series data, For the first in real-time time series data The speed at each point in time For the speed of historical time series data, For the first in historical time series data The speed at each point in time The mean squared error of velocity in real-time time series data compared to historical time series data. The absolute velocity error between real-time time series data and historical time series data. This is the adjustment factor for the mean square error. As an environmental impact factor, This represents the total number of time points in historical time series data and / or real-time time series data. This represents the maximum heading angle of the real-time time series data.

[0115] Compared to relying solely on mean squared error and absolute error for trend analysis, this embodiment introduces weighting coefficients, environmental impact factors, heading angle deviation, and maximum heading angle to calculate trend deviation, achieving a more detailed and accurate analysis of initial anomalies. Traditional methods typically only reflect numerical differences between data points, failing to comprehensively consider the impact of external environmental changes and vehicle dynamics. This embodiment, however, assigns a speed-based proportional weight to each time point, ensuring sensitivity in anomaly detection under different driving conditions. The inclusion of environmental impact factors allows the model to adjust its judgment criteria for different weather or road conditions, improving system adaptability. Furthermore, considering changes in heading angle deviation and maximum heading angle not only focuses on speed fluctuations but also deeply analyzes directional instability factors, thus capturing potential risks more comprehensively. These additional dimensions of data processing and integration make this method superior in distinguishing between transient fluctuations in normal operation and actual safety hazards, effectively reducing false alarm rates and improving the reliability of the early warning system, providing a more solid foundation for proactive traffic accident identification in the vehicle-to-everything (V2X) environment.

[0116] By filtering multiple sets of historical time series data from the same period and calculating the historical trend differences between them, the distribution characteristics of these trend differences (such as mean, standard deviation, quantiles, etc.) can be obtained. If the trend differences are mainly concentrated within a certain range (e.g., 95% of the data are distributed between [0, 0.1]), the standard threshold can be set to the upper limit of this range, 0.1, to ensure that most normal events are not misjudged as abnormal. When the trend difference is greater than the standard threshold, the initial abnormal event is marked as an event to be analyzed further; when the trend difference is less than or equal to the standard threshold, the initial abnormal event is marked as a normal event.

[0117] When the trend difference exceeds a standard threshold, the initial anomaly is marked as an event requiring further analysis. This is primarily because a large trend difference may indicate a significant change in the system state or environment, exceeding the expected range of normal operation. These changes could be due to potential safety hazards, system errors, or data fluctuations. Continued analysis helps to further confirm whether a risk actually exists. Conversely, when the trend difference is less than or equal to the standard threshold, it means that the currently observed change is within an acceptable range, consistent with historical behavior and preset standards, indicating that the vehicle's operating status is stable and as expected. In this case, it can be reasonably considered normal because the trend difference is merely a routine fluctuation in daily operation and does not show any abnormal signs requiring immediate attention. This distinction helps improve the efficiency and accuracy of the monitoring system, ensuring that only events that truly require attention are marked for in-depth analysis, while normal operation is not unnecessarily disturbed.

[0118] Step 300: Perform a significance test on the trend difference to obtain the error probability; and preset a test threshold. When the error probability is less than or equal to the test threshold, the preliminary abnormal event is defined as an abnormal event and step 400 is executed; when the error probability is greater than the test threshold, the preliminary abnormal event is defined as a normal event.

[0119] The significance test for trend differences is used to scientifically assess whether the event actually exists or whether it is a false positive result caused by random fluctuations or data noise. The significance test is a standard method in statistics used to determine whether the observed data differences are statistically significant, that is, whether the differences are large enough to be unlikely to be caused by random factors.

[0120] First, we set the null hypothesis and the alternative hypothesis. The null hypothesis means that there is no significant difference between the current data and the historical data, and the alternative hypothesis means that there is a significant difference between the current data and the historical data.

[0121] This embodiment adopts Test, chi-square test or Significance is assessed using methods such as tests to obtain test values, and the corresponding error probabilities are obtained by looking up the test values ​​in a table.

[0122] This embodiment uses For example, in the test, the test value is: , The mean of the trend differences in the running status data. This represents the mean of historical trend differences. The variance of the trend difference in the running status data. For historical trend differences in variance, The number of samples representing trend differences in the operational status data. This represents the number of historical trend samples.

[0123] The mean trend difference of the operational status data is calculated based on the trend difference at each time point in the operational status data. The trend difference at each time point is:

[0124] The same principle applies to variance calculation. The sample size is determined based on the time point.

[0125] The threshold value is usually called the significance level. In statistics, significance levels are typically set to 0.05, 0.01, or 0.1, corresponding to error rates of 5%, 1%, and 10%, respectively. A lower significance level indicates a lower false positive rate (the probability of misclassifying a normal event as an abnormal event), but a higher false negative rate (the probability of misclassifying an abnormal event as a normal event). In traffic safety scenarios, false alarms can cause frequent driver interference (e.g., frequent warnings), so a lower significance level should be set. In this embodiment, a significance level of 0.01 is preferred.

[0126] If the error probability is less than the preset significance level, the null hypothesis is rejected, and the trend difference is considered to be significantly different from the historical trend difference; the preliminary abnormal event should be marked as an abnormal event. If the error probability is greater than or equal to the preset significance level, the null hypothesis is accepted, and the trend difference is considered to be not significantly different from the historical trend difference; the preliminary abnormal event should be marked as a normal event.

[0127] Step 400: Build a machine learning model by inputting trend differences and operational status data into the machine learning model and outputting the traffic accident type.

[0128] Even if the error probability is less than the preset significance level after a significance test, thus rejecting the null hypothesis and indicating a significant difference between the current trend and historical trends, and the initial anomaly is marked as such, further model identification is still needed to confirm the cause and nature of the event. This is because a significance test only tells us whether the observed data deviates from expectations sufficiently to be unlikely to be the result of random variation, but it does not provide detailed information about the specific causes of this difference or its actual impact. By using machine learning models, these initial anomalies can be analyzed more deeply, such as distinguishing different types of traffic accidents, assessing severity, or predicting possible trends. This not only helps improve the accuracy of anomaly detection and reduce false alarms, but also provides more valuable information support for subsequent emergency response and decision-making. Therefore, this multi-level verification mechanism ensures that only events confirmed through comprehensive analysis are ultimately identified as true anomalies, thereby improving the reliability and effectiveness of the entire monitoring system.

[0129] The process of constructing a machine learning model includes: acquiring sample data using oversampling and undersampling methods, with each sample data set comprising a set of historical operational data; determining historical trend deviation, historical acceleration, historical heading angle, historical cumulative rate of change of speed, and historical environmental data based on the sample data; using the historical trend deviation, historical acceleration, historical heading angle, historical cumulative rate of change of speed, and historical environmental data as the training set for the machine learning model, and labeling each historical operational data set with a traffic accident type; and training the machine learning model using the training set to obtain a trained machine learning model.

[0130] The training set selects one set of historical operational data, specifically two sets of historical operational data from the same time period. Based on each set of historical operational data, the historical trend deviation, historical acceleration, historical heading angle, and historical environmental data are obtained for each historical time point. The historical cumulative rate of change of velocity is calculated by summing the rate of change of velocity between adjacent historical time points of the historical operational data, over the entire historical time range.

[0131] The specific training process follows existing technology, primarily involving calculating the error during training until the error reaches its minimum or the maximum number of iterations is reached, indicating that the model training is complete. This embodiment will not elaborate further.

[0132] The machine learning model is LSTM. Since this embodiment is applied to vehicle traffic accident recognition, a lightweight model is preferred.

[0133] LSTM (Long Short-Term Memory) is a special type of recurrent neural network, particularly suitable for processing and predicting time-series data because it can learn long-term dependencies. In vehicle accident recognition applications, LSTM models can help identify potential accident patterns from continuous sensor data. To meet the demands of real-time processing and considering computational resource constraints, this embodiment selects a lightweight version of the LSTM model.

[0134] Traditional LSTMs consist of multiple stacked LSTM layers to capture more complex features. However, for lightweight applications, computational complexity can be reduced by decreasing the number of LSTM layers. For example, a deep network consisting of 3-4 layers can be simplified to a shallow network with only 1-2 LSTM layers. This simplification can provide sufficient performance and significantly improve operational efficiency.

[0135] Within each LSTM layer, reducing the number of LSTM units (also known as memory units or hidden units) can effectively reduce model complexity. Fewer units mean less information to process in each layer, thus reducing computational burden. The appropriate number of units can be chosen based on specific task requirements and available computing resources. For example, in some cases, it can be reduced from hundreds to dozens of units.

[0136] The design of the complexity of the lightweight LSTM model described above is something that those skilled in the art can obtain without any technical effort, and the structural design will not be described in detail in this embodiment.

[0137] This embodiment is applied to multiple fields such as intelligent driving and accident recognition, and performs real-time identification and decision-making based on real-time data. Besides reducing the number of model layers and lowering complexity, it also introduces edge computing methods, distributing some computational tasks to devices closer to the data source (such as in-vehicle computers) instead of relying entirely on remote servers. This reduces data transmission latency because most processing is done locally, improving system response speed and enabling immediate action upon detecting anomalies. Currently, many intelligent driving technologies already achieve real-time decision-making based on real-time data, such as Tesla's Autopilot system, SuperVision platform, and Drive autonomous driving system. Therefore, existing autonomous driving technologies and research results have demonstrated the feasibility and effectiveness of real-time decision-making methods.

[0138] The operational status data and trend differences are input into the trained machine learning model, which outputs the traffic accident type. The traffic accident types in this embodiment include, but are not limited to, the following categories: rear-end collisions, side collisions, rollovers, single-vehicle loss of control, and emergencies caused by brake failure or tire blowouts.

[0139] like Figure 2 As shown, this embodiment also discloses a system for actively recognizing traffic accidents based on the Internet of Vehicles, including the following modules:

[0140] Preliminary Abnormal Event Detection Module: This module is used to acquire the current vehicle's operating status data in real time and preset a basic threshold for the operating status data. When the operating status data is greater than the basic threshold, the current vehicle's operating status is determined to be a preliminary abnormal event; when the operating status data is less than or equal to the basic threshold, the current vehicle's operating status is determined to be a normal event.

[0141] Continue analyzing the event labeling module: connected to the preliminary abnormal event judgment module, used to obtain historical operating status data corresponding to the operating status data, obtain trend differences based on historical operating status data and operating status data; and preset a standard threshold. When the trend difference is greater than the standard threshold, the preliminary abnormal event is marked as an event to continue analysis; when the trend difference is less than or equal to the standard threshold, the preliminary abnormal event is marked as a normal event.

[0142] Significance testing module: Connected to the continued analysis event labeling module, it is used to perform significance testing on trend differences to obtain error probabilities; and preset a testing threshold. When the error probability is less than or equal to the testing threshold, the preliminary abnormal event is defined as an abnormal event; when the error probability is greater than the testing threshold, the preliminary abnormal event is defined as a normal event.

[0143] Prediction module: Connected to the significance test module, it is used to build a machine learning model. It inputs trend differences and operational status data into the machine learning model and outputs the type of traffic accident.

Claims

1. A method for actively identifying traffic accidents based on Internet of Vehicles, characterized in that, Comprise: Step 100, real-time acquisition of the current vehicle running state data, and preset the basic threshold value for the running state data, when the running state data is greater than the basic threshold value, then the current vehicle running state is determined as a preliminary abnormal event and step 200 is executed; When the operating status data is less than or equal to the basic threshold, the current vehicle operating status is determined to be a normal event; wherein, real-time acquisition of the current vehicle's operating status data includes: acquiring the current vehicle's current time and the data prior to the current time. The operational status data at each historical moment are integrated into real-time time series data in chronological order; Step 200, obtaining the historical running state data corresponding to the running state data, the historical running state data and the running state data both include: speed, acceleration, heading angle, environmental data, tire pressure; According to the trend difference between the historical running state data and the running state data, comprising the following steps: Step 210, obtaining the historical running state data corresponding to the same time period of the running state data, and integrating into historical time series data in time sequence; Step 220, defining the weight coefficient for each time point in real-time time series data and historical time series data according to the running state data and historical running state data; Step 230, defining the environmental impact factor according to the running state data and historical running state data; Step 240, calculating the absolute deviation and mean square error between the historical time series data and the real-time time series data according to the running state data and historical running state data; Step 250, calculating the heading angle deviation between the historical time series data and the real-time time series data according to the heading angle in the running state data and historical running state data; and obtaining the maximum heading angle from the historical time series data and the real-time time series data; Step 260, calculating the trend difference according to the weight coefficient, environmental impact factor, absolute deviation, mean square error, heading angle deviation and maximum heading angle: ； In the formula, is the trend difference, is the weight coefficient of the time point, is the heading angle of real-time time series data, is the heading angle of the time point in real-time time series data, is the heading angle of historical time series data, is the heading angle of the time point in historical time series data, is the heading angle deviation, is the heading angle deviation adjustment coefficient, is the speed of real-time time series data, is the speed of the time point in real-time time series data, is the speed of historical time series data, is the speed of the time point in historical time series data, is the mean square error of speed in real-time time series data and historical time series data, is the absolute error of speed in real-time time series data and historical time series data, is the adjustment coefficient of the mean square error, is the environmental influence factor, is the total number of time points of historical time series data and / or real-time time series data, is the maximum heading angle of real-time time series data and real-time time series data; And preset a standard threshold value, when the trend difference is greater than the standard threshold value, the preliminary abnormal event is marked as a continue analysis event and step 300 is executed; when the trend difference is less than or equal to the standard threshold value, the preliminary abnormal event is marked as a normal event; Step 300, performing significance test on the trend difference to obtain error probability; and presetting a test threshold value, when the error probability is less than or equal to the test threshold value, then the preliminary abnormal event is defined as an abnormal event and step 400 is executed; when the error probability is greater than the test threshold value, then the preliminary abnormal event is defined as a normal event; Step 400, constructing a machine learning model, inputting the trend difference and running state data into the machine learning model, and outputting the traffic accident type. 2.The method of actively identifying traffic accidents based on the Internet of Vehicles according to claim 1, characterized in that: According to the speed of each time point in the real-time time series data and the maximum speed in the real-time time series data and the historical time series data, the weight coefficient of each time point is calculated. 3.The method of claim 2, wherein: According to the running state data and historical running state data, the environmental impact factor is defined, comprising the following steps: Step 231, determining the environmental factor according to the environmental data; Step 232, defining a tire standard pressure, and calculating the braking distance change ratio according to the tire pressure and the tire standard pressure; Step 233, calculating the acceleration deviation between the historical time series data and the real-time time series data according to the acceleration in the running state data and historical running state data; and obtaining the maximum acceleration from the historical time series data and the real-time time series data; At step 234, the environmental influence factor is calculated according to the environmental factor, the braking distance change ratio, the acceleration deviation and the maximum acceleration. 4.The method of claim 3, wherein: The machine learning model is constructed, including: obtaining sample data in a manner of oversampling and undersampling, each sample data including a set of historical running data, determining historical trend deviation, historical acceleration, historical heading angle, historical speed cumulative change rate and historical environment data according to the sample data; taking the historical trend deviation, the historical acceleration, the historical heading angle, the historical speed cumulative change rate and the historical environment data as a training set of the machine learning model, and labeling the traffic accident type for each historical running data; training the machine learning model through the training set to obtain the trained machine learning model. 5.The method of claim 3, wherein the method further comprises: The running state data and the trend difference are input into the trained machine learning model, and the traffic accident type is output.

6. A system for actively identifying traffic accidents based on vehicle networking, configured to perform the method for actively identifying traffic accidents based on vehicle networking according to any one of claims 1-5. The method comprises the following modules: A preliminary abnormal event determination module is configured to acquire running state data of a current vehicle in real time, and preset a basic threshold for the running state data, when the running state data is greater than the basic threshold, the running state of the current vehicle is determined as a preliminary abnormal event; when the running state data is less than or equal to the basic threshold, the running state of the current vehicle is determined as a normal event; A continued analysis event determination module is connected with the preliminary abnormal event determination module, configured to acquire historical running state data corresponding to the running state data, and obtain a trend difference according to the historical running state data and the running state data; and preset a standard threshold, when the trend difference is greater than the standard threshold, the preliminary abnormal event is marked as a continued analysis event; when the trend difference is less than or equal to the standard threshold, the preliminary abnormal event is marked as a normal event; A significance test module is connected with the continued analysis event determination module, configured to perform significance test on the trend difference to obtain an error probability; and preset a test threshold, when the error probability is less than or equal to the test threshold, the preliminary abnormal event is defined as an abnormal event; when the error probability is greater than the test threshold, the preliminary abnormal event is defined as a normal event; A prediction module is connected with the significance test module, configured to construct a machine learning model, input the trend difference and the running state data into the machine learning model, and output a traffic accident type.

Images