A traffic conflict identification method based on time series data of vehicle motion status

Abstract

The present invention discloses a traffic conflict discrimination method based on automobile motion state time series data, including: 1 extracting and subtracting the time domain data of the lateral and longitudinal speeds and lateral and longitudinal accelerations of vehicles in the conflict candidate events within the scope of the road intersection; 2 using discrete Fourier transform to transform the time domain data into frequency domain data and obtain the area of ​​the spectrum diagram; 3 using zero-mean normalization to normalize the data matrix constructed by the area data, and then using the 2DPCA algorithm to reduce the dimension of the matrix data; 4 finally using the fuzzy C-means algorithm for clustering, and combining the indicators TTC / T2 / PET to divide the results into two categories: traffic conflict and non-traffic conflict. The present invention helps to improve the problems of strong subjectivity and low efficiency in traditional traffic conflict discrimination, thereby improving the accuracy and portability of traffic conflict discrimination between motor vehicles.

Description

Technical Field

[0001] The invention relates to a traffic conflict determination method based on automobile motion state time series data, belonging to the field of traffic safety analysis based on traffic conflicts. Background Art

[0002] With the rapid development of the road traffic industry and the increasing investment in transportation construction, road traffic safety issues have attracted widespread attention from all walks of life. In recent years, with the extensive development of relevant theoretical research and engineering practice, road traffic accident black spot identification technology has become an effective means for countries around the world to prevent road traffic accidents and improve traffic safety management.

[0003] The rapid development of machine learning, data mining and other technologies and theories, as well as the gradual popularization of machine vision technology and multi-source vehicle-mounted sensor equipment, have created new conditions for road traffic data collection. Although multi-source data collection technology provides rich information for road accident black spot identification, traditional black spot identification is mainly based on accident frequency prediction models. Its disadvantage is that it requires the collection of a large amount of historical traffic accident data, which is not suitable for newly built roads or existing roads where accident data collection is difficult; and the prediction model constructed based on historical accident and risk factor data is difficult to have good timeliness. The process and mechanism of traffic conflict are closely related to traffic accidents, and are characterized by real-time and frequent occurrence. Therefore, using multi-source data collection technology to obtain traffic conflict data, and by building a relationship between traffic conflicts and traffic accidents, road safety analysis and evaluation can be carried out, which can effectively solve the problem that traditional black spot identification theory relies on historical accident data and has weak timeliness.

[0004] However, the existing traffic conflict identification methods for motor vehicles at road intersections based on traffic conflict theory have the following problems: First, most of the existing studies use a single traffic conflict index or a combination of two conflict indicators to divide traffic conflicts / non-conflicts, but due to the complexity of model calculation within the existing theoretical framework, the traffic conflict index is difficult to expand to multiple factors; Second, when solving the traffic conflict / non-conflict threshold, the high-dimensional data consistency and complementary information composed of multiple indicators are not fully utilized, and the identification accuracy of traffic conflict events needs to be improved; Third, traditional methods often determine the threshold of traffic conflict indicators based on experience or according to the correlation between conflict identification results and accident data, and use fixed values. In fact, the threshold of the indicator is affected by factors such as traffic facility characteristics, traffic conditions, and weather environment. Therefore, the portability and scalability of existing traffic conflict identification methods are poor. Summary of the invention

[0005] The present invention is to avoid the problems existing in the above-mentioned prior art and provide a traffic conflict identification method based on automobile motion state time series data, so as to achieve efficient and objective identification of traffic conflicts between motor vehicles at road intersections, and improve the accuracy and scientificity of traffic conflict identification, thereby improving the traffic safety level of motor vehicles at road intersections.

[0006] The present invention adopts the following technical solutions to solve the technical problems:

[0007] The traffic conflict determination method based on automobile motion state time series data of the present invention is characterized in that it is performed according to the following steps:

[0008] Step 1: selection of road intersections and collection and preprocessing of vehicle kinematic state data;

[0009] Step 1.1, collecting morning rush hour video data with a frame rate of θ at z intersections from the road traffic travel information system, and extracting the kinematic state data of any motor vehicle A and any motor vehicle B from the i-th road intersection, including: motor vehicle position, speed and acceleration, lateral and longitudinal speeds and lateral and longitudinal accelerations; i = 1, 2, ..., z;

[0010] Step 1.2: If motor vehicle A and motor vehicle B approach each other on the same horizontal straight line at the i-th road intersection, then execute steps 1.2.1 to 1.2.2; if motor vehicle A and motor vehicle B approach each other in a staggered manner at the i-th road intersection, then execute steps 1.2.3 to 1.2.4;

[0011] Step 1.2.1: If motor vehicle A and motor vehicle B are traveling in the same direction on the same horizontal straight line at the i-th road intersection, and motor vehicle A is the front vehicle and motor vehicle B is the rear vehicle, then use formula (1) to calculate the collision time TTC of motor vehicle A and motor vehicle B at the i-th road intersection at time t: A,B (t), thus obtaining the same-direction collision time of the ith road intersection at all n moments and selecting the minimum value, recorded as min TTC A,B :

[0012]

[0013] In formula (1), L A,B (t) represents the distance between the front of the rear vehicle B and the rear of the front vehicle A at the i-th road intersection at time t, v B (t) represents the speed of the rear vehicle B at the i-th road intersection at time t, v A (t) represents the speed of the preceding vehicle A at the i-th road intersection at time t;

[0014] If motor vehicle A and motor vehicle B are currently traveling in opposite directions on the same horizontal straight line at the i-th road intersection, the collision time TTC′ between motor vehicle A and motor vehicle B at the i-th road intersection at time t is calculated using formula (1): A,B (t), thus obtaining the same-direction collision time of the ith road intersection at all n moments and selecting the minimum value, recorded as min TTC A,B ;

[0015]

[0016] In formula (1), L′ A,B (t) represents the distance between the front of motor vehicle A and motor vehicle B at the i-th road intersection at time t, v A (t), v B (t) respectively represent the speeds of motor vehicle A and motor vehicle B at the i-th road intersection at time t;

[0017] Step 1.2.2: Use formula (3) to calculate the traffic conflict risk metric index of motor vehicle A and motor vehicle B at the i-th road intersection at time t: Thus, the traffic conflict risk measurement index of the ith road intersection at all n moments is obtained and the minimum value is selected, which is recorded as

[0018]

[0019] In formula (3), D A (t) represents the distance between the intersection point of the expected trajectories of motor vehicle A and motor vehicle B at the i-th road intersection at time t when the two vehicles keep their current driving directions unchanged, D B (t) represents the distance between the intersection point of the expected trajectories of motor vehicle A and motor vehicle B and motor vehicle B at the i-th road intersection at time t when the two vehicles maintain their current driving directions;

[0020] Step 1.3: Determine k candidate events of motor vehicle conflict at the i-th road intersection, where the determination rule for each candidate event of motor vehicle conflict is:

[0021] If the absolute value of the longitudinal acceleration of motor vehicle A or motor vehicle B at the i-th road intersection is ≥ g1 and minTTC A，B ≤κ, it means that motor vehicle A and motor vehicle B at the i-th road intersection are candidate events of rear-end collision or oncoming conflict;

[0022] If the absolute value of the lateral acceleration of motor vehicle A or motor vehicle B at the i-th road intersection is ≥ g2 and This means that motor vehicle A and motor vehicle B at the i-th road intersection are candidate events with angle conflict;

[0023] Where g1 represents the absolute value threshold of the longitudinal acceleration of motor vehicle A or motor vehicle B in the candidate event of rear-end collision or oncoming collision, and κ represents the min TTC of motor vehicle A and motor vehicle B in the candidate event of rear-end collision or oncoming collision. A,B The threshold value, g2 represents the absolute value threshold value of the longitudinal acceleration of motor vehicle A or motor vehicle B in the candidate event of angle conflict, and τ represents the absolute value threshold value of the longitudinal acceleration of motor vehicle A and motor vehicle B in the candidate event of angle conflict. Threshold value;

[0024] Step 1.4: Use formula (4) to calculate the rear intrusion time index of motor vehicle A and motor vehicle B in any motor vehicle conflict candidate event at the i-th road intersection at all n moments:

[0025] PET A,B =t A -t B (4)

[0026] In formula (4), t A represents the time when the front of motor vehicle A, which passes the intersection point of the expected trajectories of the two vehicles, reaches the intersection point at the i-th road intersection, t B It represents the time when the rear end of the motor vehicle B that first passes the intersection point of the expected trajectories of the two vehicles leaves the intersection point at the i-th road intersection;

[0027] Step 1.5: After subtracting the kinematic state data of the two motor vehicles in the candidate event of motor vehicle conflict at the i-th road intersection, the lateral differential speed v of the motor vehicle in the j-th candidate event of conflict at the i-th road intersection is obtained. xij , longitudinal differential speed v yij , lateral differential acceleration a xij , longitudinal differential acceleration a yij ; i=1,2,3,...,z, j=1,2,...,k;

[0028] Step 2: According to the kinematic state, a discrete Fourier transform model is established for each road intersection, and the area data is obtained by using the spectrum obtained by the transformation to construct a data matrix;

[0029] Step 2.1, define the step size as n;

[0030] Step 2.2: Set v xij 、v yij 、a xij 、a yij Any difference data in the time domain is recorded as time domain data x t , and use formula (5) to calculate x t After discrete Fourier transform, the frequency domain data X is obtained when the frequency band is f f :

[0031]

[0032] In formula (5), f = 0, 1, 2, ..., n-1; t = 0, 1, 2, ..., n-1; represents an imaginary unit;

[0033] Step 2.3, set the cutoff frequency to Draw the X f The spectrum of the obtained spectrum and the area enclosed by the coordinate axis are divided into Ψ parts in the frequency direction, and the area of ​​each part {S μ |μ=2,3,...,Ψ} and the total area S, where S μ represents the area of ​​the μth part;

[0034] Step 2.4: Use formula (6) to construct the data matrix X of the jth conflict candidate event at the i-th road intersection: ij :

[0035]

[0036] In formula (6), Respectively represent v yij The area and total area of ​​the μth part of the spectrum after discrete Fourier transformation, Respectively represent v xij The area and total area of ​​the μth part of the spectrum after discrete Fourier transformation, Respectively represent a yij The area and total area of ​​the μth part of the spectrum after discrete Fourier transformation, Respectively represent a xij The area and total area of ​​the μth part of the spectrum after discrete Fourier transformation;

[0037] Step 3: Use formula (7) to establish the normalized matrix data X of the jth motor vehicle conflict candidate event at the i-th road intersection: i,j,norm :

[0038]

[0039] In formula (7), Represents X i,j The mean of σ i,j Represents X i,j The standard deviation of

[0040] Step 4: Establish a dimensionality reduction model of the 2DPCA algorithm for the i-th road intersection;

[0041] Step 4.1: Use formula (8) to obtain the covariance matrix G of the i-th road intersection: i :

[0042]

[0043] In formula (8), represents the average matrix of the i-th road intersection, and T stands for transpose;

[0044] Step 4.2: Find the covariance matrix G i After sorting the four eigenvalues ​​in descending order, the sorted eigenvalues ​​{λ i,l |l=1,2,3,4} and its corresponding orthogonal eigenvectors {π i,l |l=1,2,3,4}; where λ i,l Represents G i The lth eigenvalue of i,l Represents G i The eigenvector corresponding to the lth eigenvalue of ;

[0045] Step 4.3, define the number of selected eigenvalues ​​as d, and initialize d = 1;

[0046] Step 4.4: Calculate the cumulative contribution rate Among them, λ i,l′ represents the l′th eigenvalue;

[0047] Step 4.5, if Q ≥ Q0; then set the d eigenvalues ​​{λ i,l′ |l′=1,2,..d.}, the corresponding orthogonal eigenvectors {π i,l′ |l′=1,2,...,d} form the optimal projection matrix V opt Otherwise, assign d+1 to d and return to step 4.4 to execute sequentially until d>4, where Q0 represents the cumulative contribution rate threshold, π i,l′ represents the l′th eigenvalue λ i,l′ The corresponding orthogonal eigenvectors;

[0048] Step 4.5: Use formula (9) to transform the matrix data Perform projection dimensionality reduction to obtain the dimensionality reduction matrix Y of the j-th candidate motor vehicle conflict event at the i-th road intersection i,j :

[0049]

[0050] Step 5: Establish a fuzzy C-means clustering model for each road intersection to identify traffic conflicts;

[0051] Step 5.1: Y i,j Expand into vector data by column, denoted as Y′ i,j ={y i,j,1 ,y i,j,2 ,...,y i,j,r ,...y i,j,Ψ×d}, where y i,j,r Represents the rth value of the jth vector data at the i-th intersection, 1≤r≤Ψ×d;

[0052] Step 5.2: Initialize the number of clusters to q and set the maximum number of iterations to b max And the termination threshold is ε, and the centroid of the sth cluster is defined as c s , Y′ ij The cluster centroid c s The membership degree is denoted as u i,j,s , define the current number of iterations as b; 1≤s≤q, 1≤j≤k;

[0053] Initialize b = 1; randomly initialize the cluster centroid at the b-th iteration to

[0054] Step 5.3: Use formula (10) to calculate the membership matrix under the b-th iteration:

[0055]

[0056] In formula (10), Y′ ij To cluster centroid The Lagrangian distance of Y′ ij To the s′th cluster centroid Lagrangian distance; m is the fuzzy weighting parameter;

[0057] Step 5.4: Use formula (11) to calculate the sth cluster center in the b+1th iteration

[0058]

[0059] Step 5.5: Use formula (12) to construct the objective function for the bth iteration:

[0060]

[0061] Step 5.6: After assigning b+1 to b, use formula (12) to calculate the objective function at the b+1th iteration:

[0062] Step 5.7: If b>bmax or with This means that the q best clusters of the i-th road intersection in the b+1-th iteration are obtained, otherwise, return to step 5.3 and execute sequentially;

[0063] Step 5.8, using formula (1), formula (2), formula (3), formula (4), calculate the minimum same-direction collision time, the minimum traffic conflict risk measurement index, and the post-intrusion time index of each conflict candidate event corresponding to the s-th best clustering cluster of the i-th road intersection data; thereby obtaining the minimum same-direction collision time, the minimum traffic conflict risk measurement index, and the post-intrusion time index of all conflict candidate events corresponding to the q-th best clustering cluster of the i-th road intersection;

[0064] Step 5.9: All the best clusters whose minimum same-direction collision time, minimum traffic conflict risk metric index, and post-intrusion time index of all conflict candidate events corresponding to the q best clusters are less than the threshold Δ for more than 3 / 4 are classified as traffic conflict classes, and the remaining best clusters are classified as non-traffic conflict classes.

[0065] An electronic device of the present invention includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the traffic conflict determination method, and the processor is configured to execute the program stored in the memory.

[0066] The present invention provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and the computer program executes the steps of the traffic conflict determination method when the computer program is executed by a processor.

[0067] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0068] 1. Compared with the traditional method of using a single traffic conflict index or a combination of two conflict indexes to distinguish traffic conflicts / non-conflicts, the present invention fully considers multiple time domain data indicators such as the lateral and longitudinal speeds and lateral and longitudinal accelerations of motor vehicles, and comprehensively considers the correlation information of multiple indicators, thereby improving the precision and accuracy of traffic conflict judgment between motor vehicles.

[0069] 2. The present invention takes into account the consistency and complementary information of high-dimensional data composed of multivariate indicators, and adopts the 2DPCA algorithm to project and reconstruct the matrix data, which greatly reduces the dimension and spatial redundancy of the high-dimensional matrix data and retains the local spatial relationship information and structural characteristics of the matrix data, thereby improving the efficiency and accuracy of traffic conflict judgment.

[0070] 3. The present invention takes into account the problem that the threshold division of traffic conflict indicators at different road intersections is highly subjective and has poor portability, and proposes to execute the fuzzy C-means clustering algorithm on the data of different road intersections respectively, thereby enhancing the theoretical basis of traffic conflict discrimination and the portability of the discrimination method. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 A basic flow chart of a traffic conflict determination method based on automobile motion state time series data of the present invention;

[0072] Figure 2 This is a spectrum diagram of the longitudinal differential velocity after discrete Fourier transformation of the present invention;

[0073] Figure 3 This is a spectrum diagram of the lateral differential velocity after discrete Fourier transformation of the present invention;

[0074] Figure 4 It is a spectrum diagram of the longitudinal differential acceleration after discrete Fourier transformation of the present invention;

[0075] Figure 5 It is a spectrum diagram of the lateral differential acceleration after discrete Fourier transformation of the present invention;

[0076] Figure 6 This is the traffic conflict judgment result diagram. DETAILED DESCRIPTION

[0077] In this embodiment, see Figure 1 ,A traffic conflict identification method based on the time series data of the vehicle motion state is carried out in the following steps:

[0078] Step 1: selection of road intersections and collection and preprocessing of vehicle kinematic state data;

[0079] Step 1.1, collect morning rush hour video data with a frame rate of θ=29 frames / second at z=12 intersections from the Florida Department of Transportation's FL511 Advanced Traveler Information System (ATIS), and extract the kinematic state data of any motor vehicle A and any motor vehicle B from the i-th road intersection, including: motor vehicle position, speed and acceleration, lateral and longitudinal speeds, and lateral and longitudinal accelerations; i=1,2,...,12;

[0080] Step 1.2: If motor vehicle A and motor vehicle B approach each other on the same horizontal straight line at the i-th road intersection, then execute steps 1.2.1 to 1.2.2; if motor vehicle A and motor vehicle B approach each other in a staggered manner at the i-th road intersection, then execute steps 1.2.3 to 1.2.4;

[0081] Step 1.2.1: If motor vehicle A and motor vehicle B are traveling in the same direction on the same horizontal straight line at the i-th road intersection, and motor vehicle A is the front vehicle and motor vehicle B is the rear vehicle, then use formula (1) to calculate the collision time TTC of motor vehicle A and motor vehicle B at the i-th road intersection at time t: A,B (t), thus obtaining the same-direction collision time of the ith road intersection at all n moments and selecting the minimum value, recorded as min TTC A,B :

[0082]

[0083] In formula (1), L A,B (t) represents the distance between the front of the rear vehicle B and the rear of the front vehicle A at the i-th road intersection at time t, v B (t) represents the speed of the rear vehicle B at the i-th road intersection at time t, v A (t) represents the speed of the preceding vehicle A at the i-th road intersection at time t;

[0084] If motor vehicle A and motor vehicle B are currently traveling in opposite directions on the same horizontal straight line at the i-th road intersection, the collision time TTC′ between motor vehicle A and motor vehicle B at the i-th road intersection at time t is calculated using formula (1): A,B (t), thus obtaining the same-direction collision time of the ith road intersection at all n moments and selecting the minimum value, recorded as min TTC A,B ;

[0085]

[0086] In formula (1), L′ A,B (t) represents the distance between the front of motor vehicle A and motor vehicle B at the i-th road intersection at time t, v A (t), v B (t) respectively represent the speeds of motor vehicle A and motor vehicle B at the i-th road intersection at time t;

[0087] Step 1.2.2: Use formula (3) to calculate the traffic conflict risk metric index of motor vehicle A and motor vehicle B at the i-th road intersection at time t: Thus, the traffic conflict risk measurement index of the ith road intersection at all n moments is obtained and the minimum value is selected, which is recorded as

[0088]

[0089] In formula (3), D A(t) represents the distance between the intersection point of the expected trajectories of motor vehicle A and motor vehicle B at the i-th road intersection at time t when the two vehicles keep their current driving directions unchanged, D B (t) represents the distance between the intersection point of the expected trajectories of motor vehicle A and motor vehicle B and motor vehicle B at the i-th road intersection at time t when the two vehicles maintain their current driving directions;

[0090] Step 1.3: Determine k candidate events of motor vehicle conflict at the i-th road intersection, where the determination rule for each candidate event of motor vehicle conflict is:

[0091] If the absolute value of the longitudinal acceleration of motor vehicle A or motor vehicle B at the i-th road intersection is ≥ g1 = 0.45g and min TTC A,B ≤κ=4 seconds, it means that motor vehicle A and motor vehicle B at the i-th road intersection are candidate events of rear-end collision or oncoming collision; g represents the acceleration of gravity, which is g=9.8m / s 2 ;

[0092] If the absolute value of the lateral acceleration of motor vehicle A or motor vehicle B at the i-th road intersection is ≥ g2 = 0.5g and seconds; it means that motor vehicle A and motor vehicle B at the i-th road intersection are candidate events with angle conflict;

[0093] Where g1 represents the absolute value threshold of the longitudinal acceleration of motor vehicle A or motor vehicle B in the candidate event of rear-end collision or oncoming collision, and κ represents the min TTC of motor vehicle A and motor vehicle B in the candidate event of rear-end collision or oncoming collision. A,B The threshold value, g2 represents the absolute value threshold value of the longitudinal acceleration of motor vehicle A or motor vehicle B in the candidate event of angle conflict, and τ represents the absolute value threshold value of the longitudinal acceleration of motor vehicle A and motor vehicle B in the candidate event of angle conflict. Threshold value;

[0094] Step 1.4: Use formula (4) to calculate the rear intrusion time index of motor vehicle A and motor vehicle B in any motor vehicle conflict candidate event at the i-th road intersection at all n moments:

[0095] PET A,B =t A -t B (4)

[0096] In formula (4), t A represents the time when the front of motor vehicle A, which passes the intersection point of the expected trajectories of the two vehicles, reaches the intersection point at the i-th road intersection, t B It represents the time when the rear end of the motor vehicle B that first passes the intersection point of the expected trajectories of the two vehicles leaves the intersection point at the i-th road intersection;

[0097] Step 1.5: After subtracting the kinematic state data of the two motor vehicles in the candidate event of motor vehicle conflict at the i-th road intersection, the lateral differential speed v of the motor vehicle in the j-th candidate event of conflict at the i-th road intersection is obtained. xij , longitudinal differential speed v yij , lateral differential acceleration a xij , longitudinal differential acceleration a yij ; j = 1, 2, ..., k, the value of k varies according to the road intersection, see Table 2;

[0098] Step 2: According to the kinematic state, a discrete Fourier transform model is established for each road intersection, and the area data is obtained by using the spectrum obtained by the transformation to construct a data matrix;

[0099] Step 2.1, define the step size as n;

[0100] Step 2.2: Set v xij 、v yij 、a xij 、a yij Any difference data in the time domain is recorded as time domain data x t , and use formula (5) to calculate x t After discrete Fourier transform, the frequency domain data X is obtained when the frequency band is f f :

[0101]

[0102] In formula (5), f = 0, 1, 2, ..., n-1; t = 0, 1, 2, ..., n-1; represents an imaginary unit;

[0103] Step 2.3, set the cutoff frequency to Draw the X f The spectrum diagram of Figure 2 , Figure 3 , Figure 4 and Figure 5 As shown, the area enclosed by the obtained spectrum and the coordinate axis is divided into Ψ = 8 parts in the frequency direction, and the area of ​​each part {S μ |μ=2,3,...,8} and the total area S, where S μ represents the area of ​​the μth part;

[0104] Step 2.4: Use formula (6) to construct the data matrix X of the jth conflict candidate event at the i-th road intersection: ij :

[0105]

[0106] In formula (6), Respectively represent v yij The area and total area of ​​the μth part of the spectrum after discrete Fourier transformation, Respectively represent v xij The area and total area of ​​the μth part of the spectrum after discrete Fourier transformation, Respectively represent a yij The area and total area of ​​the μth part of the spectrum after discrete Fourier transformation, Respectively represent a xij The area and total area of ​​the μth part of the spectrum after discrete Fourier transformation;

[0107] Step 3: Use formula (7) to establish the normalized matrix data X of the jth motor vehicle conflict candidate event at the i-th road intersection: i,j,norm :

[0108]

[0109] In formula (7), Represents X i,j The mean of σ i,j Represents X i,j The standard deviation of

[0110] Step 4: Establish a dimensionality reduction model of the 2DPCA algorithm for the i-th road intersection;

[0111] Step 4.1: Use formula (8) to obtain the covariance matrix G of the i-th road intersection: i :

[0112]

[0113] In formula (8), represents the average matrix of the i-th road intersection, and T stands for transpose;

[0114] Step 4.2: Find the covariance matrix G i After sorting the four eigenvalues ​​in descending order, the sorted eigenvalues ​​{λ i,l |l=1,2,3,4} and its corresponding orthogonal eigenvectors {π i,l |l=1,2,3,4}; where λ i,l Represents G i The lth eigenvalue of i,l Represents G i The eigenvector corresponding to the lth eigenvalue of ;

[0115] Step 4.3, define the number of selected eigenvalues ​​as d, and initialize d = 1;

[0116] Step 4.4: Calculate the cumulative contribution rate Among them, λ i,l′ represents the l′th eigenvalue;

[0117] Step 4.5, if Q≥Q0=0.85; then set the d eigenvalues ​​{λ i,l′ |l′=1,2,...,d} corresponding to the orthogonal eigenvector {π i,l′ |l′=1,2,...,d} form the optimal projection matrix V opt Otherwise, assign d+1 to d and return to step 4.4 to execute sequentially until d>4, where Q0 represents the cumulative contribution rate threshold, π i,l′ represents the l′th eigenvalue λ i,l′ The corresponding orthogonal eigenvectors, when l′=d=1, the calculation results of the cumulative contribution rate Q of the 12 road intersections are shown in Table 1:

[0118] Table 1 Cumulative contribution rate Q

[0119]

[0120] Q≥Q0=0.85 is satisfied, so d takes the value 1.

[0121] Step 4.5: Use formula (9) to transform the matrix data Perform projection dimensionality reduction to obtain the dimensionality reduction matrix Y of the j-th candidate motor vehicle conflict event at the i-th road intersection i,j :

[0122]

[0123] In formula (9), V opt =[π i,1 ];

[0124] Step 5: Establish a fuzzy C-means clustering model for each road intersection to identify traffic conflicts;

[0125] Step 5.1: Y i,j Expand into vector data by column, denoted as Y′ i,j ={y i,j,1 ,y i,j,2 ,...,y i,j,r ,...y i,j,8}, where y i,j,r Represents the rth value of the jth vector data at the i-th intersection, 1≤r≤8;

[0126] Step 5.2: Initialize the number of clusters to q = 2 and set the maximum number of iterations to b max = 1500 and the termination threshold is ε = 0.005, and the centroid of the sth cluster is defined as c s , Y′ ij The cluster centroid c s The membership degree is denoted as u i,j,s , define the current number of iterations as b; 1≤s≤2, 1≤j≤k;

[0127] Initialize b = 1; randomly initialize the cluster centroid at the b-th iteration to

[0128] Step 5.3: Use formula (10) to calculate the membership matrix under the b-th iteration:

[0129]

[0130] In formula (10), Y′ ij To cluster centroid The Lagrangian distance of Y′ ij To the s′th cluster centroid Lagrange distance; m is the fuzzy weighting parameter, which takes the value of 2;

[0131] Step 5.4: Use formula (11) to calculate the sth cluster center in the b+1th iteration

[0132]

[0133] Step 5.5: Use formula (12) to construct the objective function for the bth iteration:

[0134]

[0135] Step 5.6: After assigning b+1 to b, use formula (12) to calculate the objective function at the b+1th iteration:

[0136] Step 5.7: If b>b max =1500 or This means that q = 2 optimal clusters are obtained for the i-th road intersection at the b+1-th iteration. Otherwise, return to step 5.3 and execute sequentially.

[0137] Step 5.8: Calculate the conflict candidate event corresponding to the sth best cluster of the i-th road intersection data using equations (1), (2), (3), and (4). value; thus, the values ​​of all candidate conflict events corresponding to the q best clusters at the i-th road intersection are obtained. value;

[0138] Step 5.9: All the best clusters with values ​​generally less than the threshold Δ = 3 seconds are classified as traffic conflict classes, named cluster 1, and the remaining best clusters are classified as non-traffic conflict classes, named cluster 2. Figure 6 As shown;

[0139] In this embodiment, an electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0140] In this embodiment, a computer-readable storage medium stores a computer program on the computer-readable storage medium, and the computer program executes the steps of the above method when executed by a processor.

[0141] The results of motor vehicle traffic conflict identification at 12 road intersections are shown in Table 2:

[0142] Table 2 Traffic conflict identification results of 2DPCA-FCM

[0143]

[0144]

[0145] The total traffic accident data of 12 road intersections from 2012 to 2021 were collected, and the correlation analysis between the traffic conflict identification results and the historical number of traffic accidents was conducted.

[0146] As shown in Table 3, the traffic conflict judgment result set of the road intersection traffic conflict judgment method based on the time series data of vehicle motion state, namely 2DPCA-FCM, is H 2DPCA ={h′1,h′2,h′3,...,h′ 12}, the threshold is Δ = 3 seconds The discriminant result set is The historical accident data set of each road intersection is H wreck ={h″′1,h″′2,h″′3,...,h″′ 12}; where h′ i represents the traffic conflict judgment number of the ith intersection of the 2DPCA-FCM method, where h″ i The threshold is Δ = 3 seconds. The traffic conflict discrimination number of the ith intersection of the method, where h″′ i Represents the number of traffic accidents at the i-th intersection.

[0147] Table 3 Traffic conflict identification results and historical traffic accident data

[0148]

[0149]

[0150] Calculate the traffic conflict judgment results of road intersections based on 2DPCA-FCM and the threshold value of 3 seconds The Pearson correlation coefficient and Spearman correlation coefficient of the discrimination results and the number of historical traffic accidents are shown in Table 4. The differences in the accuracy of traffic conflict discrimination between the two methods are compared and the reasons are analyzed.

[0151] Table 4 Correlation test between traffic conflict identification and accident data

[0152]

[0153] It can be seen from Table 4 that at 95% confidence level (p value less than 0.05), based on 2DPCA-FCM, There is a significant positive correlation between traffic conflict events and traffic accidents identified by the two methods. The Pearson and Spearman correlation coefficients are 0.6948, 0.7694, 0.660 and 0.6449 respectively.

[0154] It can be seen that the discrimination results based on 2DPCA-FCM have a higher correlation with traffic accidents. This may be because the method of the present invention adopts a method of conducting in-depth analysis on each intersection separately, which can better capture the traffic conflicts at different road intersections, thereby improving the discrimination accuracy of traffic conflict events. The correlation of the method is lower, which may be because the same fixed threshold is used for different intersections, which cannot fully reflect the characteristics of different intersections, and some traffic conflict events are misjudged, reducing the accuracy of the judgment results.

Claims

1. A traffic conflict identification method based on automobile motion state time series data, characterized in that: The steps are as follows: Step 1: selection of road intersections and collection and preprocessing of vehicle kinematic state data; Step 1.1, collecting morning rush hour video data with a frame rate of θ at z intersections from the road traffic travel information system, and extracting the kinematic state data of any motor vehicle A and any motor vehicle B from the i-th road intersection, including: motor vehicle position, speed and acceleration, lateral and longitudinal speeds and lateral and longitudinal accelerations; i = 1, 2, ..., z; Step 1.2: If motor vehicle A and motor vehicle B approach each other on the same horizontal straight line at the i-th road intersection, then execute steps 1.2.1 to 1.2.2; if motor vehicle A and motor vehicle B approach each other in a staggered manner at the i-th road intersection, then execute steps 1.2.3 to 1.2.4; Step 1.2.1: If motor vehicle A and motor vehicle B are traveling in the same direction on the same horizontal straight line at the i-th road intersection, and motor vehicle A is the front vehicle and motor vehicle B is the rear vehicle, then use formula (1) to calculate the collision time TTC of motor vehicle A and motor vehicle B at the i-th road intersection at time t: A,B (t), thus obtaining the same-direction collision time of the ith road intersection at all n moments and selecting the minimum value, recorded as minTTC A,B : In formula (1), L A,B (t) represents the distance between the front of the rear vehicle B and the rear of the front vehicle A at the i-th road intersection at time t, v B (t) represents the speed of the rear vehicle B at the i-th road intersection at time t, v A (t) represents the speed of the preceding vehicle A at the i-th road intersection at time t; If motor vehicle A and motor vehicle B are currently traveling in opposite directions on the same horizontal straight line at the i-th road intersection, the collision time TTC′ between motor vehicle A and motor vehicle B at the i-th road intersection at time t is calculated using formula (1): A,B (t), thus obtaining the same-direction collision time of the ith road intersection at all n moments and selecting the minimum value, recorded as minTTC A,B ; In formula (1), L′ A,B (t) represents the distance between the front of motor vehicle A and motor vehicle B at the i-th road intersection at time t, v A (t), v B (t) respectively represent the speeds of motor vehicle A and motor vehicle B at the i-th road intersection at time t; Step 1.2.2: Use formula (3) to calculate the traffic conflict risk metric index of motor vehicle A and motor vehicle B at the i-th road intersection at time t: Thus, the traffic conflict risk measurement index of the ith road intersection at all n moments is obtained and the minimum value is selected, which is recorded as In formula (3), D A (t) represents the distance between the intersection point of the expected trajectories of motor vehicle A and motor vehicle B at the i-th road intersection at time t when the two vehicles keep their current driving directions unchanged, D B (t) represents the distance between the intersection point of the expected trajectories of motor vehicle A and motor vehicle B and motor vehicle B at the i-th road intersection at time t when the two vehicles maintain their current driving directions; Step 1.3: Determine k candidate events of motor vehicle conflict at the i-th road intersection, where the determination rule for each candidate event of motor vehicle conflict is: If the absolute value of the longitudinal acceleration of motor vehicle A or motor vehicle B at the i-th road intersection is ≥ g1 and minTTC A,B ≤κ, it means that motor vehicle A and motor vehicle B at the i-th road intersection are candidate events of rear-end collision or oncoming conflict; If the absolute value of the lateral acceleration of motor vehicle A or motor vehicle B at the i-th road intersection is ≥ g2 and This means that motor vehicle A and motor vehicle B at the i-th road intersection are candidate events with angle conflict; Where g1 represents the absolute value threshold of the longitudinal acceleration of motor vehicle A or motor vehicle B in the candidate event of rear-end collision or oncoming conflict, and κ represents the minTTC of motor vehicle A and motor vehicle B in the candidate event of rear-end collision or oncoming conflict. A,B The threshold value, g2 represents the absolute value threshold value of the longitudinal acceleration of motor vehicle A or motor vehicle B in the candidate event of angle conflict, and τ represents the absolute value threshold value of the longitudinal acceleration of motor vehicle A and motor vehicle B in the candidate event of angle conflict. Threshold value; Step 1.4: Use formula (4) to calculate the rear intrusion time index of motor vehicle A and motor vehicle B in any motor vehicle conflict candidate event at the i-th road intersection at all n moments: PET A,B =t A -t B (4) In formula (4), t A represents the time when the front of motor vehicle A, which passes the intersection point of the expected trajectories of the two vehicles, reaches the intersection point at the i-th road intersection, t B It represents the time when the rear end of the motor vehicle B that first passes the intersection point of the expected trajectories of the two vehicles leaves the intersection point at the i-th road intersection; Step 1.5: After subtracting the kinematic state data of the two motor vehicles in the candidate event of motor vehicle conflict at the i-th road intersection, the lateral differential speed v of the motor vehicle in the j-th candidate event of conflict at the i-th road intersection is obtained. xij , longitudinal differential speed v yij , lateral differential acceleration a xij , longitudinal differential acceleration a yij ;i=1,2,3,...,z, j=1,2,...,k; Step 2: According to the kinematic state, a discrete Fourier transform model is established for each road intersection, and the area data is obtained by using the spectrum obtained by the transformation to construct a data matrix; Step 2.1, define the step size as n; Step 2.2: Set v xij 、v yij 、a xij 、a yij Any difference data in the time domain is recorded as time domain data x t , and use formula (5) to calculate x t After discrete Fourier transform, the frequency domain data X is obtained when the frequency band is f f : In formula (5), f = 0, 1, 2, ..., n-1; t = 0, 1, 2, ..., n-1; θ represents an imaginary unit; Step 2.3, set the cutoff frequency to Draw the X f The spectrum of the obtained spectrum and the area enclosed by the coordinate axis are divided into Ψ parts in the frequency direction, and the area of ​​each part {S μ |μ=2,3,...,Ψ} and the total area S, where S μ represents the area of ​​the μth part; Step 2.4: Use formula (6) to construct the data matrix X of the jth conflict candidate event at the i-th road intersection: ij : In formula (6), Respectively represent v yij The area and total area of ​​the μth part of the spectrum after discrete Fourier transformation, Respectively represent v xij The area and total area of ​​the μth part of the spectrum after discrete Fourier transformation, Respectively represent a yij The area and total area of ​​the μth part of the spectrum after discrete Fourier transformation, Respectively represent a xij The area and total area of ​​the μth part of the spectrum after discrete Fourier transformation; Step 3: Use formula (7) to establish the normalized matrix data X of the jth motor vehicle conflict candidate event at the i-th road intersection: i,j,norm : In formula (7), Represents X i,j The mean of σ i,j Represents X i,j The standard deviation of Step 4: Establish a dimensionality reduction model of the 2DPCA algorithm for the i-th road intersection; Step 4.1: Use formula (8) to obtain the covariance matrix G of the i-th road intersection: i : In formula (8), represents the average matrix of the i-th road intersection, and T stands for transpose; Step 4.2: Find the covariance matrix G i After sorting the four eigenvalues ​​in descending order, the sorted eigenvalues ​​{λ i,l |l=1,2,3,4} and its corresponding orthogonal eigenvectors {π i,l |l=1,2,3,4}; where λ i,l Represents G i The lth eigenvalue of i,l Represents G i The eigenvector corresponding to the lth eigenvalue of ; Step 4.3, define the number of selected eigenvalues ​​as d, and initialize d = 1; Step 4.4: Calculate the cumulative contribution rate Among them, λ i,l′ represents the l′th eigenvalue; Step 4.5, if Q ≥ Q0; then set the d eigenvalues ​​{λ i,l′ |l′=1,2,..d.}, the corresponding orthogonal eigenvectors {π i,l′ |l′=1,2,...,d} form the optimal projection matrix V opt Otherwise, assign d+1 to d and return to step 4.4 to execute sequentially until d>4, where Q0 represents the cumulative contribution rate threshold, π i,l′ represents the l′th eigenvalue λ i,l′ The corresponding orthogonal eigenvectors; Step 4.5: Use formula (9) to transform the matrix data Perform projection dimensionality reduction to obtain the dimensionality reduction matrix Y of the j-th candidate motor vehicle conflict event at the i-th road intersection i,j : Step 5: Establish a fuzzy C-means clustering model for each road intersection to identify traffic conflicts; Step 5.1: Y i,j Expand into vector data by column, denoted as Y′ i,j ={y i,j,1 ,y i,j,2 ,...,y i,j,r ,...y i,j,Ψ×d }, where y i,j,r Represents the rth value of the jth vector data at the i-th intersection, 1≤r≤Ψ×d; Step 5.2: Initialize the number of clusters to q and set the maximum number of iterations to b max And the termination threshold is ε, and the centroid of the sth cluster is defined as c s , Y′ ij The cluster centroid c s The membership degree is denoted as u i,j,s , define the current number of iterations as b; 1≤s≤q, 1≤j≤k; Initialize b = 1; randomly initialize the cluster centroid at the b-th iteration to Step 5.3: Use formula (10) to calculate the membership matrix under the b-th iteration: In formula (10), Y′ ij To cluster centroid The Lagrangian distance of Y′ ij To the s′th cluster centroid Lagrangian distance; m is the fuzzy weighting parameter; Step 5.4: Use formula (11) to calculate the sth cluster center in the b+1th iteration Step 5.5: Use formula (12) to construct the objective function for the bth iteration: Step 5.6: After assigning b+1 to b, use formula (12) to calculate the objective function at the b+1th iteration: Step 5.7: If b>b max or with This means that the q best clusters of the i-th road intersection in the b+1-th iteration are obtained, otherwise, return to step 5.3 and execute sequentially; Step 5.8, using formula (1), formula (2), formula (3), formula (4), calculate the minimum same-direction collision time, the minimum traffic conflict risk measurement index, and the post-intrusion time index of each conflict candidate event corresponding to the s-th best clustering cluster of the i-th road intersection data; thereby obtaining the minimum same-direction collision time, the minimum traffic conflict risk measurement index, and the post-intrusion time index of all conflict candidate events corresponding to the q-th best clustering cluster of the i-th road intersection; Step 5.9: All the best clusters whose minimum same-direction collision time, minimum traffic conflict risk metric index, and post-intrusion time index of all conflict candidate events corresponding to the q best clusters are less than the threshold Δ for more than 3 / 4 are classified as traffic conflict classes, and the remaining best clusters are classified as non-traffic conflict classes.

2. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the traffic conflict determination method according to claim 1, and the processor is configured to execute the program stored in the memory.

3. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the traffic conflict determination method according to claim 1 are executed.

Images