A driving style dynamic identification method in a vehicle networking environment

Abstract

This invention relates to a dynamic driving style recognition method in a vehicle-to-everything (V2X) environment, belonging to the field of traffic safety. Drivers' decisions are subjective, and driving assistance systems that consider these differences in decision-making habits can improve the efficiency and accuracy of active safety. Traditional driving style recognition methods primarily rely on a series of traffic psychology questionnaires to investigate and score drivers' driving habits, thereby obtaining their driving style. However, the results often differ from actual road performance, and the scoring methods are not suitable for real-time driving style detection in a V2X environment. This invention utilizes multi-scenario trajectory data and unsupervised machine learning methods to learn and train driving style recognition thresholds. It then leverages the communication and computing capabilities of the central cloud control platform and the vehicle-mounted platform in a connected environment to detect aggressive drivers in real time, ultimately formulating personalized warning strategies. This invention can be applied to active safety warnings in the context of intelligent connected vehicle technology.

Description

Technical Field

[0001] This invention belongs to the field of traffic safety evaluation, and specifically refers to a new method for dynamic recognition of driving style in a vehicle-to-everything (V2X) environment. Background Technology

[0002] With the advancement of the national transportation power strategy and the new infrastructure construction strategy, and the rapid development of intelligent connected and vehicle-road cooperative technologies, configuring advanced advanced driver assistance systems (ADAS) has become an important means to ensure traffic safety and alleviate traffic congestion. Traditional ADAS mainly considers a general control model and algorithm to achieve safe vehicle operation. However, driver decision-making is subjective, and different drivers react and drive differently to the same traffic conditions or events. Driver assistance systems that consider differences in driver decision-making habits can provide personalized safety warnings and proactive control for drivers with different driving styles, improving the efficiency and accuracy of proactive safety. Traditional driving style recognition methods mainly obtain driving styles by investigating and scoring drivers' driving habits through a series of traffic psychology questionnaires. However, the results obtained often differ from the driver's performance on actual roads, and the scoring method is not suitable for real-time driving style detection in a vehicle-to-everything (V2X) environment. With the application and development of vehicle micro-trajectory data, some scholars have begun to explore methods for obtaining driving styles through trajectory data. Trajectory data provides a large amount of driver motion characteristics such as speed and acceleration. By capturing the intensity of changes in driver behavior over a certain time series, it may be possible to distinguish driving styles, and then provide real-time information prompts to drivers with different styles to ensure traffic safety.

[0003] Therefore, how to express driving style, especially by using the perception capabilities of the connected environment to assess driving style in real time, will help improve the provision of personalized active safety warning information. Summary of the Invention

[0004] To address the aforementioned problems, this invention proposes a novel method for dynamic recognition of driving styles in a connected environment. This method utilizes the intensity of changes in the driving behavior characteristics of the vehicle and surrounding vehicles over time to represent driving styles, and determines the recognition thresholds for different driving styles based on an unsupervised learning method. Finally, the proposed online sliding window recognition method is used to identify more aggressive drivers in real time and provide them with appropriate information prompts to ensure safe driving. This invention provides theoretical support for personalized safety warnings and control of vehicles in a connected environment.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for dynamic recognition of driving style in a vehicle-to-everything (V2X) environment, characterized by the following steps:

[0007] Step 1: Construct a driving style database

[0008] 1. Driving Style Database

[0009] Driving style refers to a driver's driving habits and reactions in a specific environment. Therefore, drivers' driving characteristics generally differ in different driving environments, and the threshold for classifying driving styles also varies. This study aims to construct a driving style database for different driving environments to achieve driving style recognition in these environments.

[0010] This invention primarily classifies driving styles by identifying the intensity of driver variation in a specified trajectory parameter time series within vehicle micro-trajectory data. Therefore, the driving style database should contain driver micro-trajectory data for different environments. Let D represent the driver database, then the database can be defined as follows: Where a and b represent the environment sets, respectively. Represents a region i under environment a. This represents a region i under environment b.

[0011] 2. Introduction to Trajectory Data Examples

[0012] For a dataset of a certain region under a certain environment, the vehicle micro-trajectory parameters that should be included at least include: vehicle number, frame number, coordinates, speed, acceleration and other information, as shown in Table 1.

[0013] Table 1. Trajectory data fields used in this invention

[0014] Field Name meaning Frames The interval between each two data points is 25 frames per second (one frame every 0.04 seconds). Vehicle number Vehicle number X Vehicle X coordinate Y Vehicle Y-coordinate speed vehicle speed acceleration Vehicle acceleration front vehicle number The vehicle number of the car in front of this car Rear vehicle number The vehicle number of the car following this vehicle Left-forward vehicle number The vehicle number in front of the car in the left lane is... Left-hand rear vehicle number The vehicle number of the car behind this car in the left lane Right-forward vehicle number The vehicle number in front of the car in the right lane is... Right-hand rear vehicle number The vehicle number of the car behind this vehicle in the right lane

[0015] Step 2: Extraction of driving behavior features based on vehicle micro-trajectory

[0016] To identify driving styles, the first step is to extract vehicle kinematic parameters from the vehicle's micro-trajectory data. This includes extracting parameters from the target vehicle (the vehicle whose driving style is being identified) and adjacent vehicles, totaling 27 parameters. Adjacent vehicles are: the vehicle behind the target vehicle, the vehicle in front of the target vehicle, the vehicle behind in the left lane, the vehicle in front in the left lane, the vehicle behind in the right lane, and the vehicle in front in the right lane.

[0017] 1. Extraction of target vehicle parameters

[0018] Extract the speed of the target vehicle acceleration Position coordinates These are the three basic parameters of the target vehicle. The superscript t... i Representing time i, here are two consecutive t. i The time interval is 1 frame; the subscript 's' represents the target vehicle.

[0019] 2. Extraction of parameters of adjacent vehicles

[0020] A vehicle's driving style depends not only on its motion characteristics but also on the surrounding traffic flow environment. The surrounding vehicles influence the target vehicle's decisions, and the target vehicle behaves differently under different traffic conditions. This behavior reflects the driver's habits and driving style under a given traffic situation. Therefore, the interaction parameters between the target vehicle and adjacent vehicles are extracted as fundamental data characterizing driving style. The extracted fields mainly include the speed difference, acceleration difference, distance difference in the x-direction, and distance difference in the y-direction between the target vehicle and adjacent vehicles, totaling 24 interaction parameters, as detailed below:

[0021] 1) Calculate the target vehicle at time t i The speed difference between the current time and the speed of the six adjacent vehicles;

[0022] 2) Calculate the target vehicle at time t i The difference in acceleration between the vehicle and its six adjacent vehicles;

[0023] 3) Calculate the target vehicle at time t i The distance difference in the x-direction between the vehicle and its six adjacent vehicles at any given time;

[0024] 4) Calculate the target vehicle at time t i Distance difference in the y-direction between the current time and the 6 adjacent vehicles

[0025] Step 3: Construction of Driving Style Characteristics Based on Time-Based Driving Volatility

[0026] This invention uses the concept of driving volatility to characterize driving behavior. Driving volatility represents the degree to which a driver's driving indicator deviates from its normal value within a specified time range. The greater the degree of change, the more intense the driver's operational fluctuations within that time range, representing a more aggressive driving style. In step two, 27 features were extracted, including the driver's speed, acceleration, and speed, acceleration, and distance differences with adjacent vehicles. Driving style is characterized by calculating the driving volatility of these 27 features within a specified time period. The calculation functions for the five driving volatility indicators are as follows:

[0027] 1. Time series standard deviation of characteristic parameters

[0028]

[0029] In the formula: S dev The standard deviation is represented by x, T is the number of feature parameters (in this invention, T is the length of a certain feature time series), and x is the number of feature parameters. i For feature parameter values, The mean of the feature parameters;

[0030] 2. Time series coefficient of variation of characteristic parameters

[0031]

[0032] In the formula: C v S represents the coefficient of variation. dev The time series standard deviation of the characteristic parameter, It represents the absolute value of the mean of the characteristic parameters.

[0033] 3. Time series mean absolute deviation of characteristic parameters

[0034]

[0035] In the formula: D mean Represents the mean absolute deviation, T is the number of characteristic parameters, and x i For feature parameter values, The mean of the feature parameters;

[0036] 4. Time series quartile variation coefficient of characteristic parameters

[0037]

[0038] In the formula: Q cv The coefficients of variation represent the quartiles, with Q1 being the lower quartile (the value corresponding to the 25th percentile of the data) and Q3 being the upper quartile (the value corresponding to the 75th percentile of the data).

[0039] 5. Time-varying stochastic volatility

[0040]

[0041] In the formula: v f Represents time-varying stochastic volatility, r i The characteristic change per unit time, x t and x t-1 The values ​​of the characteristic parameters at times t and t-1 are... is the average value of the characteristic change per unit time, and n is the number of characteristic parameters.

[0042] 6. For each specified time series T feature, calculate the above 5 driving volatility indicators. Finally, for each group of samples, 135 volatility indicators can be calculated to characterize driving style.

[0043] Step 4: Dynamic driving style recognition and personalized information prompts

[0044] 1. Standardize data units

[0045] The time-driven volatility data obtained in step three is represented by X. Z-Score normalization is then applied to X to obtain the normalized matrix X0. * :

[0046] 2. Eliminate collinearity characteristics

[0047] The characteristic matrix of order n×p that may have collinearity is compressed into an n×k non-collinear matrix. The specific calculation method is as follows:

[0048] 1) For matrix X * Decentralization yields a new matrix That is, each column is zero-mean normalized, which means subtracting the mean of that column.

[0049] 2) Calculation The covariance matrix C.

[0050] 3) Perform eigenvalue decomposition on the covariance matrix to find the eigenvalues ​​λ of the covariance matrix. k and the corresponding feature vector v k .

[0051] 4) Arrange the eigenvectors in descending column order from left to right according to their corresponding eigenvalues ​​to form a matrix, and take the first k columns to form matrix W, which is an n×k matrix.

[0052] 5) Through Y=X * W calculates Y as an n×k order non-collinear matrix, i.e., the k features after compression.

[0053] 6) To ensure that the compressed features represent at least 95% of the contribution variance of the original features, the value of k is defined as the minimum k value that makes the following equation hold:

[0054]

[0055] 3. Driving style marking

[0056] The 135(p) driving volatility indicators were ultimately compressed into k features Y. Although Y represents driving style, the threshold for classifying driving styles has not yet been determined. Assuming Y has a certain shape distribution, unsupervised learning methods can divide the distributed data according to the degree of clustering. This study adopts the k-means clustering algorithm to classify driving styles based on the threshold, clustering the feature data group Y (an n×k matrix with n groups of k features) into z classes of driving styles. The specific process is as follows:

[0057] 1) The algorithm randomly selects the initial z0 class centers:

[0058] 2) For each sample Y i (Y = [Y1, Y2, ..., Y) i ,…Y n ]), i∈n, and label it as the category closest to the category center, that is:

[0059]

[0060] 3) Update each category center to the mean of all samples belonging to that category;

[0061]

[0062] c j x represents the number of samples in each category. i For each category, there are samples.

[0063] 4) Repeat the last two steps until the category centers converge. At this point, the driving style threshold is:

[0064] 5) Driving styles are categorized into three types: cautious, normal, and aggressive. Therefore, the final driving style thresholds are defined as: (0-u1], (u1-u2], (u2-u3]. Driving volatility reflects the degree to which a driver deviates from their normal state within a certain timeframe; therefore, a higher value indicates a more aggressive driving style. Thus, 0-u1 represents a cautious driver, u1-u2 a normal driver, and u2-u3 an aggressive driver. (For the driving style database...) In any environment and any region, the driving style recognition threshold (0-u1], (u1-u2], (u2-u3) should be calculated according to the method in the above steps.

[0065] 4. Dynamic detection of driving style and personalized information prompts

[0066] To ensure traffic safety, personalized alerts should primarily target aggressive drivers. Drivers may exhibit increased driving volatility for a short period under certain unforeseen circumstances, leading to their detection as aggressive drivers. To eliminate false alarms caused by unforeseen circumstances, a rolling time window detection method is used to identify aggressive drivers. The specific identification process is as follows:

[0067] 1) First, the vehicle network cloud control platform should locate the vehicle's position and send the driving style discrimination threshold of the target vehicle to the vehicle terminal according to the driver's driving environment area: (0-u1], (u1-u2], (u2-u3).

[0068] 2) Secondly, the target vehicle acquires the speed, acceleration, and relative distance of the target vehicle and adjacent vehicles at certain time intervals Δt based on the vehicle's on-board perception and computing capabilities in the connected environment.

[0069] 3) Calculate the vehicle's time-driven volatility function in any time series T and obtain the target vehicle's driving style parameters according to the method in step four.

[0070] 4) If If the driver is an aggressive type, record the driver's vehicle number (note that (u2-u3] is the driving style threshold in a specific environmental area);

[0071] 5) After the time interval step ΔT, re-label the driving style for the data of time series length T after T+ΔT;

[0072] 6) Reassess whether the driver is an aggressive driver at the beginning of each new testing cycle;

[0073] 7) If the license plate number is recorded in three consecutive rolling detection cycles, the driver is defined as an aggressive type.

[0074] 8) Provide warnings to drivers marked as having an aggressive driving style on the in-vehicle terminal, reminding them to adjust their driving style. Attached Figure Description

[0075] Figure 1 Invention Flowchart

[0076] Figure 2 Vehicle Relationship Diagram

[0077] Figure 3 Vehicle parameter calculation diagram

[0078] Figure 4 Schematic diagram of driving style detection and personalized warning in a connected environment Detailed Implementation

[0079] The specific embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this disclosure.

[0080] The flowchart of this invention is as follows Figure 1 As shown, this method mainly consists of four steps: 1. Introduction of data source; 2. Extraction of driving behavior features based on vehicle micro-trajectory; 3. Construction of driving style features based on time-dependent driving volatility; and 4. Dynamic driving style recognition and personalized information prompts. Step 1 introduces the data format used for driving style recognition; step 2 extracts the basic parameters for identifying driving style from the trajectory data; step 3 constructs the feature parameters for identifying driving style; and step 4 determines the threshold for driving style and the identification logic and personalized information prompt strategy for aggressive drivers.

[0081] The specific calculation process of this invention is described below.

[0082] The specific execution steps are as follows:

[0083] This invention uses vehicle micro-trajectory data for driving style recognition. Specific data extraction examples and methods are as follows.

[0084] Step 1: Construct a driving style database

[0085] 1. Driving Style Database

[0086] Driving style refers to a driver's driving habits and reactions in a specific environment. Therefore, drivers' driving characteristics generally differ in different driving environments, and the threshold for classifying driving styles also varies. This study aims to construct a driving style database for different driving environments to achieve driving style recognition in these environments.

[0087] This invention primarily classifies driving styles by identifying the intensity of driver variation in a specified trajectory parameter time series within vehicle micro-trajectory data. Therefore, the driving style database should contain driver micro-trajectory data for different environments. Let D represent the driver database, then the database can be defined as follows: Where a and b represent the environment sets, respectively. Represents a region i under environment a. This represents a region i under environment b.

[0088] Example: If the driving environment in a city is divided into two environments: highways and urban roads, then the database D should contain or be defined as D = {Highway: [basic road segments, weaving areas, tunnels, diverging areas, merging areas, ...], Urban roads: [intersections, expressways, commercial areas, industrial areas, residential areas, ...]}

[0089] 2. Introduction to Trajectory Data Examples

[0090] For a dataset of a certain region under a certain environment, the vehicle micro-trajectory parameters that should be included at least include: vehicle number, frame number, coordinates, speed, acceleration and other information, as shown in Table 1.

[0091] Table 1. Trajectory data fields used in this invention

[0092] Field Name meaning Frames The interval between each two data points is 25 frames per second (one frame every 0.04 seconds). Vehicle number Vehicle number X Vehicle X coordinate Y Vehicle Y-coordinate speed vehicle speed acceleration Vehicle acceleration front vehicle number The vehicle number of the car in front of this car Rear vehicle number The vehicle number of the car following this vehicle Left-forward vehicle number The vehicle number in front of the car in the left lane is... Left-hand rear vehicle number The vehicle number of the car behind this car in the left lane Right-forward vehicle number The vehicle number in front of the car in the right lane is... Right-hand rear vehicle number The vehicle number of the car behind this vehicle in the right lane

[0093] Example: Taking a basic road segment area in a highway environment as an example, the dataset format is shown in Table 2. Each vehicle number should have vehicle micro-motion parameters for consecutive frames. This invention references open-source vehicle micro-trajectories from basic highway segments in Germany. For more detailed information, please refer to the link (https: / / www.highd-dataset.com / ).

[0094] Table 2 Examples of trajectory data used in this invention

[0095]

[0096]

[0097] Note: In this article, the vehicle whose driving style is to be identified is defined as the target vehicle, and the vehicles in front of and behind the target vehicle, the vehicles in front of and behind the target vehicle in the left lane, and the vehicles in front of and behind the target vehicle in the right lane are defined as adjacent vehicles.

[0098] Step 2: Extraction of driving behavior features based on vehicle micro-trajectory

[0099] To identify driving styles, the first step is to extract vehicle kinematic parameters from the vehicle's micro-trajectory data. This includes extracting parameters from the target vehicle (the vehicle whose driving style is being identified) and adjacent vehicles, totaling 27 parameters. Adjacent vehicles are: the vehicle behind the target vehicle, the vehicle in front of the target vehicle, the vehicle behind in the left lane, the vehicle in front in the left lane, the vehicle behind in the right lane, and the vehicle in front in the right lane.

[0100] 1. Extraction of target vehicle parameters

[0101] Extract the speed of the target vehicle acceleration Coordinates serve as the basic parameters of the target vehicle. The superscript t... i Representing time i, here are two consecutive t. i The time interval is 1 frame; the subscript 's' represents the target vehicle.

[0102] Example: This represents the speed of the target vehicle at time i.

[0103] 2. Extraction of parameters of adjacent vehicles

[0104] A vehicle's driving style depends not only on its motion characteristics but also on the surrounding traffic flow. The surrounding vehicles influence the target vehicle's decisions, and the target vehicle behaves differently under different traffic conditions. This behavior reflects the driver's habits and driving style under a given traffic situation. Therefore, the interaction parameters between the target vehicle and adjacent vehicles are extracted as fundamental data characterizing driving style. The extracted fields mainly include the speed difference, acceleration difference, distance difference in the x-direction, and distance difference in the y-direction between the target vehicle and adjacent vehicles, totaling 24 interaction parameters, such as... Figure 3 As shown. The calculation method is as follows:

[0105] 1) Calculate the target vehicle at time t i The speed difference between the current time and the speed of the six adjacent vehicles;

[0106] 2) Calculate the target vehicle at time t i The difference in acceleration between the vehicle and its six adjacent vehicles;

[0107] 3) Calculate the target vehicle at time t iThe distance difference in the x-direction between the vehicle and its six adjacent vehicles at any given time;

[0108] 4) Calculate the target vehicle at time t i Distance difference in the y-direction between the current time and the 6 adjacent vehicles

[0109] 1) Calculate the target vehicle at time t i Speed ​​difference between the current time and the speed of adjacent vehicles j∈[1,6];

[0110]

[0111]

[0112]

[0113]

[0114]

[0115]

[0116] In the formula: For t i The speed difference between the target vehicle and the following vehicle at any given moment;

[0117] For t i The speed difference between the vehicle in front and the target vehicle at any given moment;

[0118] For t i The speed difference between the target vehicle and the vehicle following in the left lane at any given moment;

[0119] For t i The speed difference between the vehicle in front and the target vehicle in the left lane at any given moment;

[0120] For t i The speed difference between the target vehicle and the vehicle following in the right lane at any given moment;

[0121] For t i The speed difference between the vehicle in front in the right lane and the target vehicle at any given moment;

[0122] 2) Calculate the target vehicle at time t i Acceleration difference between the vehicle and adjacent vehicles j∈[1,6];

[0123]

[0124]

[0125]

[0126]

[0127]

[0128]

[0129] In the formula: For t i The difference in acceleration between the target vehicle and the following vehicle at any given moment;

[0130] For t i The difference in acceleration between the vehicle in front and the target vehicle at any given moment;

[0131] For t i The difference in acceleration between the target vehicle and the vehicle following in the left lane at any given moment;

[0132] For t i The difference in acceleration between the vehicle in front and the target vehicle in the left lane at any given moment;

[0133] For t i The difference in acceleration between the target vehicle and the vehicle following in the right lane at any given moment;

[0134] For t i The difference in acceleration between the vehicle in front and the target vehicle in the right lane at any given moment;

[0135] 3) Calculate the target vehicle at time t i Distance difference in the x direction between the vehicle and the adjacent vehicle at any time j∈[1,6];

[0136]

[0137]

[0138]

[0139]

[0140]

[0141]

[0142] In the formula: For t i The distance difference between the target vehicle and the vehicle behind it at any given time;

[0143] For t iThe distance difference between the vehicle in front and the target vehicle at any given time;

[0144] For t i The distance difference between the target vehicle and the vehicle behind it in the left lane at any given time;

[0145] For t i The distance difference between the vehicle in front and the target vehicle in the left lane at any given time;

[0146] For t i The distance difference between the target vehicle and the vehicle behind it in the right lane at any given time;

[0147] For t i The distance difference between the vehicle in front in the right lane and the target vehicle at any given time;

[0148] L: Vehicle length, assuming all vehicles are the same length.

[0149] 4) Calculate the target vehicle at time t i Distance difference in the y-direction between the vehicle and its neighboring vehicle at any given time j∈[1,6];

[0150]

[0151]

[0152]

[0153]

[0154]

[0155]

[0156] In the formula: For t i The distance difference between the target vehicle and the vehicle behind it at any given time;

[0157] For t i The distance difference between the vehicle in front and the target vehicle at any given time;

[0158] For t i The distance difference between the target vehicle and the vehicle behind it in the left lane at any given time;

[0159] For t i The distance difference between the vehicle in front and the target vehicle in the left lane at any given time;

[0160] For t iThe distance difference between the target vehicle and the vehicle behind it in the right lane at any given time;

[0161] For t i The distance difference between the vehicle in front in the right lane and the target vehicle at any given time;

[0162] 5) A total of 27 kinematic parameters of the target vehicle and adjacent vehicles can be extracted in each frame.

[0163] Step 3: Construction of Driving Style Characteristics Based on Time-Based Driving Volatility

[0164] This invention uses the concept of driving volatility to characterize driving behavior. Driving volatility represents the degree to which a driver's driving indicator deviates from its normal value within a specified time range. The greater the degree of change, the more intense the driver's operational fluctuations within that time range, representing a more aggressive driving style. In step two, 27 features were extracted, including the driver's speed, acceleration, and speed, acceleration, and distance differences with adjacent vehicles. Driving style is characterized by calculating the driving volatility of these 27 features over a specified time period. The driving volatility function is calculated as follows:

[0165] 1. Time series standard deviation of characteristic parameters

[0166]

[0167] In the formula: S dev The standard deviation is represented by x, T is the number of feature parameters (in this invention, T is the length of a certain feature time series), and x is the number of feature parameters. i For feature parameter values, The mean of the feature parameters;

[0168] 2. Time series coefficient of variation of characteristic parameters

[0169]

[0170] In the formula: C v S represents the coefficient of variation. dev The time series standard deviation of the characteristic parameter, It represents the absolute value of the mean of the characteristic parameters.

[0171] 3. Time series mean absolute deviation of characteristic parameters

[0172]

[0173] In the formula: D mean Represents the mean absolute deviation, T is the number of characteristic parameters, and x i For feature parameter values, The mean of the feature parameters;

[0174] 4. Time series quartile variation coefficient of characteristic parameters

[0175]

[0176] In the formula: Q cv The coefficients of variation represent the quartiles, with Q1 being the lower quartile (the value corresponding to the 25th percentile of the data) and Q3 being the upper quartile (the value corresponding to the 75th percentile of the data).

[0177] 5. Time-varying stochastic volatility

[0178] r i =In(x t / x t-1 )

[0179]

[0180] In the formula: v f Represents time-varying stochastic volatility, r i The characteristic change per unit time, x t and x t-1 The values ​​of the characteristic parameters at times t and t-1 are... is the average value of the characteristic change per unit time, and n is the number of characteristic parameters.

[0181] 6. For each specified time series T feature, calculate the above 5 driving volatility indicators. Finally, for each group of samples, 135 volatility indicators can be calculated to characterize driving style.

[0182] Example: If a total of 300 sets of data on target vehicles and adjacent vehicles are extracted from the basic road section of a highway, and the time series T is 3 seconds, with 1 frame of trajectory data representing 0.04 seconds, then the length of the time series is: This means that for each set of target vehicle and adjacent vehicle data, 75 frames of data need to be extracted to calculate the driving volatility. The dimensions of the 300 sets of data are (300×75×27, where 75 is the time series length and 27 is the 27 extracted parameters). The driving volatility calculated from each set of 75 frames of data is ultimately an S... dev A C v A D mean A Q cv A v f Therefore, the data from 300 sets of target vehicles and adjacent vehicles were ultimately calculated to obtain 300×135 (27×5) dimensional data.

[0183] Step 4: Dynamic driving style recognition and personalized information prompts

[0184] 1. Standardize data units

[0185] Let X represent the time-driven volatility data obtained in step three. X can be represented by a matrix as follows:

[0186]

[0187] Where n is the number of groups of data for the target vehicle and adjacent vehicles, i.e., the number of samples; p is the feature dimension, which in this invention represents 135 features calculated from 27 parameters using time-based driving volatility. To eliminate unit differences between features, it is necessary to unify the dimensions of the p features by performing Z-Score standardization on the original data to obtain the standardization matrix X. * :

[0188]

[0189] In the formula

[0190]

[0191] 2. Eliminate collinearity characteristics

[0192] The extracted features may contain redundancy. The n×p-order feature matrix, which may exhibit collinearity, is compressed into an n×k-order non-collinear matrix. The specific calculation method is as follows:

[0193] 1) For matrix X * Decentralization yields a new matrix That is, each column is zero-mean normalized, which means subtracting the mean of that column.

[0194]

[0195]

[0196] 2) Calculation The covariance matrix C.

[0197] Where C is a p×p matrix

[0198] 3) Perform eigenvalue decomposition on the covariance matrix to find the eigenvalues ​​λ of the covariance matrix. k and the corresponding feature vector v k .

[0199] Cv k =λ k v k

[0200] 4) Arrange the eigenvectors in descending column order from left to right according to their corresponding eigenvalues ​​to form a matrix, and take the first k columns to form matrix W, which is an n×k matrix.

[0201] 5) Through Y=X * W calculates Y as an n×k order non-collinear matrix, i.e., the k features after compression.

[0202]

[0203] 6) To ensure that the compressed features represent at least 95% of the contribution variance of the original features, the value of k is defined as the minimum k value that makes the following equation hold:

[0204]

[0205] 3. Driving style marking

[0206] The 135(p) driving volatility indicators were ultimately compressed into k features Y. Although Y represents driving style, the threshold for classifying driving styles has not yet been determined. Assuming Y has a certain shape distribution, unsupervised learning methods can divide the distributed data according to the degree of clustering. This study uses the k-means clustering algorithm to classify driving styles based on the threshold. The feature data group Y (an n×k matrix, n groups of k features) is clustered into z classes of driving styles. The specific process is as follows:

[0207] 1) Randomly select the initial z0 class centers:

[0208] 2) For each sample Y i (Y = [Y1, Y2, ..., Y) i ,…Y n ]), i∈n, and label it as the category closest to the category center, that is:

[0209]

[0210] 3) Update each category center to the mean of all samples belonging to that category;

[0211]

[0212] c j x represents the number of samples in each category. i For each category, there are samples.

[0213] 4) Repeat the last two steps until the category centers converge. At this point, the driving style threshold is:

[0214] 5) Driving styles are categorized into three types: cautious, normal, and aggressive. Therefore, the final driving style thresholds are defined as: (0-u1], (u1-u2], (u2-u3]. Driving volatility reflects the degree to which a driver deviates from their normal state within a certain timeframe; therefore, a higher value indicates a more aggressive driving style. Thus, 0-u1 represents a cautious driver, u1-u2 a normal driver, and u2-u3 an aggressive driver. (For the driving style database...) In any environment and any region, the driving style recognition threshold (0-u1], (u1-u2], (u2-u3) should be calculated according to the method in the above steps.

[0215] 4. Dynamic detection of driving style and personalized information prompts

[0216] To ensure traffic safety, personalized warnings should primarily target aggressive drivers. Drivers may exhibit increased driving volatility for a short period under certain unforeseen circumstances, leading to their detection as aggressive drivers. To eliminate false alarms caused by unforeseen circumstances, a rolling time window detection method is used to identify aggressive drivers. The flowchart for driving style identification and warning for a specific target vehicle is as follows: Figure 4 As shown, the specific recognition process is as follows:

[0217] 1) First, the vehicle network cloud control platform should locate the vehicle's position and send the driving style discrimination threshold of the target vehicle to the vehicle terminal according to the driver's driving environment area: (0-u1], (u1-u2], (u2-u3).

[0218] 2) Secondly, the target vehicle acquires the speed, acceleration, and relative distance of the target vehicle and adjacent vehicles at certain time intervals Δt based on the vehicle's on-board perception and computing capabilities in the connected environment.

[0219] 3) Calculate the vehicle's time-driven volatility function in any sequence T, and obtain the target vehicle's driving style parameters according to the method in step four.

[0220] 4) If If the driver is an aggressive type, record the driver's vehicle number (note that (u2-u3] is the driving style threshold in a specific environmental area);

[0221] 5) After the time interval step size ΔT, re-label the driving style for the data of time series length T after T+ΔT;

[0222] 6) Reassess whether the driver is an aggressive driver at the beginning of each new testing cycle;

[0223] 7) If the license plate number is recorded in three consecutive rolling detection cycles, the driver is defined as an aggressive type.

[0224] 8) Provide warnings to drivers marked as having an aggressive driving style on the in-vehicle terminal, reminding them to adjust their driving style.

Claims

1. A method for dynamic recognition of driving style in a vehicle-to-everything (V2X) environment, characterized by comprising the following steps: Step 1: Construct a driving style database 1.1 Driving Style Database Driving styles are categorized by identifying the intensity of driver changes in specified trajectory parameters over time within vehicle micro-trajectory data. The driving style database should include driver micro-trajectory data for different environments. Let D represent the driver database, then the database can be defined as follows: in, a and b represent the environment sets, respectively. Represents a region i under environment a. Represents a region i under environment b; 1.2 Introduction to Trajectory Data Examples For a dataset of a specific region under a given environment, the minimum vehicle micro-trajectory parameters that should be included are: vehicle number, frame number, coordinates, speed, and acceleration, as shown below: Step 2: Extraction of driving behavior features based on vehicle micro-trajectory To identify driving styles, the first step is to extract vehicle kinematic parameters from the vehicle's micro-trajectory data. Specifically, this includes extracting parameters from the target vehicle (the vehicle whose driving style is being identified) and adjacent vehicles, totaling 27 parameters. Among these, adjacent vehicles are: the vehicle behind the target vehicle, the vehicle in front of the target vehicle, the vehicle behind the vehicle in the left lane, the vehicle in front of the vehicle in the left lane, the vehicle behind the vehicle in the right lane, and the vehicle in front of the vehicle in the right lane. 2.1 Target Vehicle Parameter Extraction Extract the speed of the target vehicle v s ti acceleration a s ti Location coordinates These are the three basic parameters of the target vehicle; where the superscript t i Representing time i, here are two consecutive t. i The time interval is 1 frame; the subscript 's' represents the target vehicle; 2.2 Extraction of parameters of adjacent vehicles Interaction parameters between the target vehicle and adjacent vehicles were also extracted as basic data characterizing driving style; the extracted fields mainly include the speed difference, acceleration difference, distance difference in the x-direction, and distance difference in the y-direction between the target vehicle and adjacent vehicles, totaling 24 interaction parameters, as detailed below: 1) Calculate the speed difference of the target vehicle with the adjacent 6 vehicles at t i time; 2) Calculate the acceleration difference of the target vehicle with the 6 adjacent vehicles at t i ; 3) Calculate the x-direction distance difference between the target vehicle and the adjacent 6 vehicles at t i ; 4) Calculate the y-direction distance difference of the target vehicle with the adjacent 6 vehicles at t i ; Step 3: Construction of Driving Style Characteristics Based on Time-Based Driving Volatility In step two, 27 features were extracted, including the driver's speed, acceleration, and speed, acceleration, and distance differences with adjacent vehicles. Driving style was characterized by calculating the driving volatility of these 27 features over a specified time period. The calculation functions for the five driving volatility indicators are as follows: 3.1 Time series standard deviation of characteristic parameters In the formula: S dev The standard deviation of the time series representing the feature parameter, where T is the number of feature parameters and T is the length of a certain feature time series, x i For feature parameter values, The mean of the feature parameters; 3.2 Time series coefficient of variation of characteristic parameters In the formula: C v S represents the coefficient of variation. dev The time series standard deviation of the characteristic parameter, It is the absolute value of the mean of the feature parameters; 3.3 Time series mean absolute deviation of characteristic parameters In the formula: D mean Represents the mean absolute deviation, T is the number of characteristic parameters, and x i For feature parameter values, The mean of the feature parameters; 3.4 Time series quartile variation coefficient of characteristic parameters wherein: Q cv Q represents the coefficient of quartile variation, Q1 is the lower quartile, i.e. the value corresponding to the 25% quantile of the data, and Q3 is the upper quartile, i.e. the value corresponding to the 75% quantile of the data; 3.5 Time-varying stochastic volatility r i = In(x t / x t-1 ) In the formula: v f Represents time-varying stochastic volatility, r i For the characteristic change per unit time, x t and x t-1 The values ​​of the characteristic parameters at times t and t-1 are... is the average value of the characteristic change per unit time, and n is the number of characteristic parameters; 3.

6. For each specified time series T feature, calculate the above 5 driving volatility indicators. Finally, for each group of samples, 135 volatility indicators can be calculated to characterize driving style. Step 4: Dynamic driving style recognition and personalized information prompts 4.1 Standardize data units Let X denote the time driving volatility data obtained in Step Three. Apply a Z-Score standardization transformation to X to obtain a standardized matrix X : 4.2 Eliminating Collinearity Features The characteristic matrix of order n×p that may have collinearity is compressed into an n×k non-collinear matrix. The specific calculation method is as follows: 1) For matrix X Decentralization yields a new matrix This means that each column is zero-mean normalized, which is achieved by subtracting the mean of that column. ; 2) compute the covariance matrix C; 3) eigen-decomposition of the covariance matrix to obtain the eigenvalues λ k of the covariance matrix, and the corresponding eigenvectors v k ; 4) Arrange the eigenvectors in descending column order from left to right according to their corresponding eigenvalues ​​to form a matrix, and take the first k columns to form matrix W, which is an n×k matrix; 5) by Y = X W computes Y as an n x k order non-collinear matrix, i.e. k features after compression; 6) To ensure that the compressed features represent at least 95% of the contribution variance of the original features, the value of k is defined as the minimum k value that makes the following equation hold: 4.3 Driving Style Marking The p driving volatility indicators are ultimately compressed into k features Y. Although Y is used to represent driving style, the threshold for classifying driving styles has not yet been determined. The k-means clustering algorithm is used to classify driving styles based on this threshold. The feature data set Y, i.e., an n×k matrix, with n groups of k features, is clustered into z categories of driving style data. The specific process is as follows: 1) The algorithm randomly selects the initial z0 class centers: ; 2) For each sample Y i (Y = [Y1, Y2, ..., Y) i ,…Y n ]), i∈n, mark it as the category closest to the category center, that is: 3) Update each category center to the mean of all samples belonging to that category; c j the number of samples contained in each class, x i samples in each class; 4) Repeat the last two steps until the category centers converge. At this point, the driving style thresholds are: (0-u1], (u1-u2], (u2-u3], ... ; 5) Driving styles are categorized into three types: cautious, normal, and aggressive. Therefore, the final driving style thresholds are defined as: (0-u1], (u1-u2], (u2-u3], where 0-u1 represents a cautious driver, u1-u2 represents a normal driver, and u2-u3 represents an aggressive driver. For the driving style database... In any environment and any region, the driving style recognition thresholds (0-u1], (u1-u2], (u2-u3]) should be calculated according to the method in the above steps; 4.4 Driving style dynamic detection and personalized information prompts Aggressive drivers are identified using a rolling time window detection method; the specific identification process is as follows: 1) First, the vehicle network cloud control platform should locate the vehicle's position and send the driving style discrimination threshold for the target vehicle's on-board terminal according to the driver's driving environment area: (0-u1], (u1-u2], (u2-u3]; 2) Secondly, the target vehicle acquires the speed, acceleration, and relative distance of the target vehicle and adjacent vehicles at certain time intervals Δt based on the vehicle's on-board perception and computing capabilities in the connected environment. 3) Calculate the vehicle's time-driven volatility function and obtain the target vehicle's driving style parameters within any time series T. ; 4) If If the driver is an aggressive type, record the driver's vehicle number; 5) After the time interval step ΔT, re-label the driving style for the data of time series length T after T+ΔT; 6) Reassess whether the driver is an aggressive driver at the beginning of each new testing cycle; 7) If the license plate number is recorded in three consecutive rolling detection cycles, the driver is defined as having an aggressive style; 8) Provide warnings to drivers marked as having an aggressive driving style on the in-vehicle terminal, reminding them to adjust their driving style.

Images