Vehicle safety domain modeling and conflict risk assessment method based on visual constraint

Abstract

The invention discloses a visual constraint-based vehicle safety domain modeling and conflict risk assessment method, which comprises the following steps of: analyzing and selecting a non-signalized intersection with frequent traffic accidents by using accident black spots, and acquiring multi-dimensional traffic information of the intersection through an unmanned aerial vehicle; building three vehicle safety domain models of straight running, right turning and left turning; by predicting a vehicle trajectory equation, determining an intersection conflict position, dividing intersection conflicts into two categories of cross conflicts and confluence conflicts according to a trajectory interaction mode, and in different conflict scenes, according to a constructed vehicle safety domain, performing quantitative calculation on a conflict risk of vehicles in a non-signalized intersection to obtain traffic conflict exposure time; and based on the traffic conflict exposure time data, determining an optimal clustering number K of a K-means algorithm through an elbow rule, clustering results of the traffic conflict exposure time by using the K-means algorithm, and grading conflict risks. According to the invention, the accuracy of vehicle conflict detection at the non-signalized intersection is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical fields of intelligent transportation, conflict risk, etc., in particular to a vehicle safety domain modeling and conflict risk assessment method based on visual constraints. BACKGROUND

[0002] With the continuous growth of urban traffic flow, the safety problem of intersection as the core node of traffic convergence is increasingly serious. Current traffic conflict detection is mainly based on time and space proximity indicators, but the actual application is mainly based on time indicators. In the conflict judgment process, on the one hand, the inherent geometric size of the vehicle needs to be considered, and the vehicle cannot be simply regarded as a point; on the other hand, the calculation complexity problem of risk detection based on vehicle geometric model also needs to be considered. At present, the common vehicle geometric models are circular model, rectangular model and double circle model.

[0003] Although the above research has made certain progress in modeling method and perception means, most of them still do not fully consider the driver's visual constraints. The traditional geometric symmetry model (such as circular / rectangular / double circle model) has significant defects in the kinematic description of vehicle, and the simplified framework based on axisymmetric assumption does not effectively integrate the forward and backward perception characteristics of driver's vision, resulting in the problem of missing visual domain constraints in dynamic steering process, and the mapping relationship between driver's line of sight and vehicle motion parameters is not established, which is difficult to represent the influence of forward-looking distance and rearward blind area on control decision. In view of the complex risk propagation mechanism in the scene of unsignalized intersection, it is of great theoretical significance and application value to study the conflict detection and risk assessment method which is closer to the driver's perception behavior. SUMMARY

[0004] The purpose of the present application is to provide a vehicle safety domain modeling and conflict risk assessment method based on visual constraints, which makes up for the deficiency of traditional geometric model in describing the kinematics of vehicle, more accurately captures the dynamic characteristics in the process of vehicle interaction at intersection, and effectively identifies the potential collision risk.

[0005] The technical solution to achieve the purpose of the present application is: a vehicle safety domain modeling and conflict risk assessment method based on visual constraints, comprising the following steps:

[0006] Step 1, selecting an unsignalized intersection with frequent traffic accidents by analyzing accident black spots, collecting multi-dimensional traffic information of the intersection by using unmanned aerial vehicle, including traffic volume, vehicle type, number of lanes and intersection channelization;

[0007] Step 2, determining the traffic capacity according to the traffic volume, determining the vehicle dynamics parameters and mixed traffic flow model according to the vehicle type, determining the road network structure according to the number of lanes and intersection channelization, and drawing the road network in simulation software;

[0008] Step 3, considering the driver's visual constraints, three vehicle safety domain models for straight, right turn and left turn are constructed. The vehicle safety domain refers to the domain where there are no obstacles or no overlap with other vehicle safety domains. The vehicle in this area is in a safe state;

[0009] Step 4, by predicting the vehicle trajectory equation, the intersection conflict position is determined. According to the trajectory interaction mode, the intersection conflict is divided into "cross conflict" and "merge conflict". In different conflict scenarios, the conflict risk of vehicles in the unsignalized intersection is quantitatively calculated according to the constructed vehicle safety domain, and the traffic conflict exposure time is obtained;

[0010] Step 5, based on the traffic conflict exposure time data, the elbow rule is used to determine the optimal clustering number K of K-means algorithm, and K-means algorithm is used to cluster the results of traffic conflict exposure time, and the conflict risk is classified.

[0011] Further, step 1, the accident black spot analysis is used to select the unsignalized intersection with frequent traffic accidents. The multi-dimensional traffic information of the intersection is collected by the unmanned aerial vehicle, including traffic flow, vehicle type, lane number and intersection channelization, wherein:

[0012] (1) The accident black spot analysis is used to select the unsignalized intersection with frequent traffic accidents;

[0013] The accident black spot analysis uses kernel density estimation and text similarity algorithm to select the unsignalized intersection with frequent traffic accidents. The specific process is as follows:

[0014] The positions of different vehicles in the unsignalized intersection are input for kernel density estimation to obtain the spatial density distribution of vehicles. The intersection with the top 20% density peak value is the hidden unsignalized intersection. The kernel density estimation formula is as follows:

[0015]

[0016] is the probability density estimation value at r, r is the current vehicle position of the unsignalized intersection, n is the number of all vehicles in the unsignalized intersection, h is the bandwidth, K is the kernel function, and Ri is the position of the i th vehicle in the unsignalized intersection;

[0017] The text similarity calculation is performed on the accident report or hidden trouble investigation report of the hidden unsignalized intersection to obtain the intersection with the highest accident frequency, which is listed as the unsignalized intersection for priority treatment. The text similarity calculation formula is as follows:

[0018]

[0019] w is the traffic accident text, v is the accident report, TF(w, v) is the frequency of w in the document v, DF(w) is the number of reports containing traffic accident text word w, and N is the total number of reports;

[0020] (2) Collect multi-dimensional traffic information of the intersection by the unmanned aerial vehicle, including traffic volume, vehicle type, lane number, intersection channelization;

[0021] Select the morning peak or evening peak as the unmanned aerial vehicle shooting time, collect multi-dimensional traffic video of the intersection by the unmanned aerial vehicle, and use UNet algorithm to extract data from the video shot by the unmanned aerial vehicle, including traffic volume, vehicle type, lane number, and intersection channelization.

[0022] Further, step 2, determine the traffic capacity according to the traffic volume, determine the vehicle dynamics parameters and the mixed traffic flow model according to the vehicle type, and determine the road network structure according to the lane number and intersection channelization, and draw the road network in the simulation software, the specific method is:

[0023] Determine the traffic capacity according to the traffic volume collected by the unmanned aerial vehicle, wherein the traffic capacity calculation formula is:

[0024]

[0025] T0 is the design traffic capacity, N is the number of vehicles in detention, and Nmax is the maximum number of vehicles that can be accommodated;

[0026] Set the vehicle dynamics parameters according to the vehicle type, including vehicle size, maximum speed, maximum acceleration and deceleration, determine the proportion of vehicle types in the mixed traffic flow model, determine the road network structure according to the lane number and intersection channelization, and draw the road network in the micro-traffic simulation software SUMO.

[0027] Further, step 3, consider the driver's visual constraints, and construct three vehicle safety domain models for straight, right turn and left turn, the vehicle safety domain is a domain in which there are no obstacles or the safety domain of other vehicles does not overlap, and the vehicle in the domain is in a safe state, wherein:

[0028] (1) Straight vehicle safety domain model;

[0029]

[0030] In the formula, (x, y) is the coordinate of the center of the vehicle, L is the width of the road, D / 2 is the radius of the rear semicircular domain, and a is the major axis of the ellipse;

[0031] (2) Right turn vehicle safety domain model

[0032]

[0033] (3) Left turn vehicle safety domain model

[0034]

[0035] Further, step 4, the intersection conflict position is determined by predicting the vehicle trajectory equation, the intersection conflict is divided into "cross conflict" and "merge conflict" according to the trajectory interaction mode, the conflict risk of vehicles in the unsignalized intersection is quantitatively calculated according to the constructed vehicle safety domain, and the traffic conflict exposure time is obtained, wherein:

[0036] The intersection conflict area is determined by predicting the vehicle trajectory equation, and the vehicle trajectory equation is:

[0037] s(t)-a0+a1t+a2t 2 +a3t 3 +a4t 4 +a5t 5

[0038] t is time, a0, a1, a2, a3, a4 and a5 are parameters of the trajectory equation fitting, and the intersection conflict position is solved by combining the vehicle trajectory equation;

[0039] According to the trajectory interaction mode, the intersection conflict area is divided into "cross conflict" and "merge conflict" two conflict scenes, when the included angle of different vehicles entering the intersection is greater than 90 degrees, it is cross conflict, and when the included angle is less than 90°, it is merge conflict; in different conflict scenes, the conflict risk of vehicles in the unsignalized intersection is quantitatively calculated according to the constructed vehicle safety domain:

[0040] (1) Cross conflict

[0041] I. Cross conflict between straight vehicles

[0042] The included angle a between vehicles is calculated:

[0043]

[0044] In the formula, (x1, y1) and (x2, y2) are the coordinates of the two vehicles, a is the included angle between the vehicles, is the vehicle heading angle;

[0045] The boundary distance l between the center position of straight vehicle 1 and the vehicle safety domain is calculated α1 :

[0046]

[0047] In the formula, a is the included angle between the vehicles, Rbf is the longitudinal radius of the front domain of straight vehicle 1, Rba is the longitudinal radius of the rear domain of straight vehicle 1, and l is the boundary distance between the center position of straight vehicle 1 and the vehicle safety domain.

[0048] Domain longitudinal radius;

[0049] Calculate the boundary distance l between the center position of the straight-going vehicle 2 and its safety domain. α2 :

[0050]

[0051] In the formula, Sb is the lateral radius of the vehicle traveling in a straight line. Let Sb be the heading angle of a straight-going vehicle. Since the rear region of the vehicle's safety domain is a semicircle, Sb = Rba, further simplifying:

[0052]

[0053] The safe distance d between vehicles is calculated as follows:

[0054] d = d0 - l α1 -l α2

[0055] In the formula, d0 is the distance between the centers of the two vehicles;

[0056] II. Crossing conflicts between left-turning and straight-going vehicles

[0057] Calculate the boundary distance l between the center position of the straight-going vehicle 1 and the vehicle's safety domain. α1 :

[0058]

[0059] Since the safety zone of left-turning vehicle 2 has been adjusted, calculate the boundary distance l between the center position of left-turning vehicle 2 and its safety zone. α2 :

[0060]

[0061] When vehicle 1 going straight approaches vehicle 2 turning left, which is closer to the driver's side, it uses a V-shaped movement. straight1 The calculation method is as follows: when the straight-going vehicle 1 is on the other side of the driver's side when the left-turning vehicle 2 was originally away from the driver, V is used. straight2 Calculation method;

[0062] The safe distance d between vehicles is calculated as follows:

[0063] d = d0 - l α1 -l α2

[0064] (2) Merging conflict

[0065] The characteristic of merging conflicts between right-turning and straight-going vehicles is that there is a clear following relationship between the vehicles. The boundary distance l between the center position of straight-going vehicle 1 and its safety domain is calculated. α1 :

[0066]

[0067] The boundary distance l between the center position of the right-turn vehicle 2 and the safety domain of the vehicle is calculated α2 :

[0068] l α2 = R ba

[0069] The inter-vehicle safety distance d is calculated as follows:

[0070] d = d0 - l α1 -l α2

[0071] In the intersection conflict and the merging conflict, if the safety domains of the vehicle 1 and the vehicle 2 overlap, that is, d < 0, the vehicle overlap duration is defined as a key indicator for measuring the degree of risk, that is, the traffic conflict exposure time Tvcr. The greater the Tvcr value, the longer the vehicle stays in the possible collision area, and the higher the conflict risk degree.

[0072] Further, step 5 determines the optimal cluster number K of the K-means algorithm based on the traffic conflict exposure time data, and classifies the conflict risk by using the K-means algorithm to cluster the results of the traffic conflict exposure time.

[0073] The elbow rule is an unsupervised learning method for determining the optimal cluster number K value in the K-means algorithm, setting the range of K, and calculating the error sum of squares SSE of the traffic conflict exposure time Tvcr:

[0074]

[0075] Wherein, Ci is the ith cluster, μi is the cluster center, K is the number of cluster clusters, and μ represents the traffic conflict exposure time Tvcr.

[0076] The K-means calculation formula is:

[0077]

[0078] K is the optimal cluster number, Ci is the data point set of the ith cluster, and μi is the centroid of the ith cluster.

[0079] According to the clustering results, the range of conflict risk classification is obtained.

[0080] A kind of vehicle safety domain modeling and conflict risk assessment system based on visual constraint, implement the vehicle safety domain modeling and conflict risk assessment method based on visual constraint described, realize the vehicle safety domain modeling and conflict risk assessment based on visual constraint, five modules are executed step 1~5 respectively.

[0081] A kind of computer equipment, comprising memory, processor and computer program stored on memory and can be run on processor, when the processor executes the computer program, implement the vehicle safety domain modeling and conflict risk assessment method based on visual constraint described, realize the vehicle safety domain modeling and conflict risk assessment based on visual constraint.

[0082] A kind of computer readable storage medium, computer program is stored on it, when the processor executes the computer program, implement the vehicle safety domain modeling and conflict risk assessment method based on visual constraint described, realize the vehicle safety domain modeling and conflict risk assessment based on visual constraint.

[0083] Compared with prior art, the present application has the following advantages: 1) by constructing front and rear vehicle safety domain considering the characteristics of driver's visual field, the shortcomings of traditional geometric model in describing vehicle kinematics are made up;2) the model is divided into straight, right turn and left turn, which makes up for the defect of single type of traditional model;3) the model can more accurately capture the dynamic characteristics of vehicle interaction at intersection, and effectively identify potential collision risk.

[0084] The application will be described in further detail below with reference to the drawings. BRIEF DESCRIPTION OF DRAWINGS

[0085] Figure 1 It is the flow chart of the vehicle safety domain modeling and conflict risk assessment method based on visual constraint of the present application.

[0086] Figure 2 It is the algorithm block diagram of the specific implementation method.

[0087] Figure 3 It is the traditional vehicle model diagram.

[0088] Figure 4 It is the straight vehicle safety domain diagram.

[0089] Figure 5 It is the right turn vehicle safety domain diagram.

[0090] Figure 6 It is the left turn vehicle safety domain diagram.

[0091] Figure 7 It is the potential conflict point diagram of unsignalized intersection.

[0092] Figure 8A straight vehicle conflict graph for one embodiment.

[0093] Figure 9 A vehicle risk score graph for one embodiment.

[0094] Figure 10 A straight vehicle intersection conflict calculation graph for one embodiment.

[0095] Figure 11 A straight vehicle rear-end conflict calculation graph for one embodiment.

[0096] Figure 12 A left-turn vehicle and other vehicle conflict graph for one embodiment.

[0097] Figure 13 A left-turn vehicle and other vehicle conflict calculation graph for one embodiment.

[0098] Figure 14 A right-turn vehicle and other vehicle conflict calculation graph for one embodiment.

[0099] Figure 15 A elbow rule K value determination graph for one embodiment.

[0100] Figure 16 A K-means clustering result graph for one embodiment.

[0101] Figure 17 A Tvcr cumulative risk value spatial distribution graph for one embodiment. DETAILED DESCRIPTION

[0102] The application will be further described below in conjunction with the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related application, and not to limit the application. After reading the application, those skilled in the art can modify various equivalent forms of the application, which fall within the scope defined by the claims attached hereto.

[0103] A vehicle safety domain modeling and conflict risk assessment method based on visual constraints, the method comprising the following steps:

[0104] Step 1, select a frequent traffic accident intersection by analyzing the accident black spots, collect multi-dimensional traffic information of the intersection by a drone, including traffic volume, vehicle type, lane number and intersection channelization.

[0105] (1) select a frequent traffic accident intersection by analyzing the accident black spots;

[0106] The accident black spot analysis selects a frequent traffic accident intersection by using kernel density estimation and text similarity algorithm, the specific process is as follows:

[0107] The position of different vehicles at the unsignalized intersection is input for kernel density estimation to obtain a vehicle spatial density distribution, and the intersection with a peak value of density in the first 20% is a hidden hazard unsignalized intersection, wherein the kernel density estimation calculation formula is:

[0108]

[0109] is a probability density estimation value at point r, r is the current vehicle position at the unsignalized intersection, n is the number of all vehicles at the unsignalized intersection, h is a bandwidth, K is a kernel function, and Ri is the position of the i-th vehicle at the unsignalized intersection;

[0110] Text similarity calculation is performed on the accident report and hidden hazard investigation report of the hidden hazard unsignalized intersection to obtain the intersection with the highest accident frequency, which is selected as the subsequent processing unsignalized intersection, wherein the text similarity calculation formula is:

[0111]

[0112] w is a traffic accident text, v is an accident report, TF(w, v) is the frequency of w appearing in the document v, DF(w) is the number of reports containing the traffic accident text word w, and N is the total number of reports;

[0113] (2) Collect multi-dimensional traffic information of the intersection by using a drone, including vehicle flow, vehicle type, lane number, and intersection channelization;

[0114] Select an early morning peak or late evening peak as the drone shooting time, collect multi-dimensional traffic video of the intersection by using a drone, and use a UNet algorithm to extract data from the video shot by the drone, including vehicle flow, vehicle type (sedan, bus, truck), lane number, and intersection channelization.

[0115] Step 2, determine the traffic capacity according to the vehicle flow, determine the vehicle dynamics parameters and mixed traffic flow model according to the vehicle type, and determine the road network structure according to the lane number and intersection channelization, and draw the road network in the simulation software, and the specific method is:

[0116] Determine the traffic capacity according to the vehicle flow collected by the drone, wherein the traffic capacity calculation formula is:

[0117]

[0118] T0 is the design traffic capacity, N is the number of vehicles in detention, and Nmax is the maximum number of vehicles that can be accommodated;

[0119] The vehicle dynamics parameters are set by vehicle type, including vehicle size, maximum speed, maximum acceleration and deceleration, the proportion of vehicle types in the mixed traffic flow model is determined, the road network structure is determined according to the number of lanes and intersection channelization, and the road network is drawn in the micro-traffic simulation software SUMO.

[0120] Step 3, considering the driver's visual constraints, three vehicle safety domain models for straight, right turn and left turn are constructed, and the vehicle safety domain refers to the domain in which the vehicle is in a safe state without obstacles or overlapping with other vehicle safety domains.

[0121] Specifically,

[0122] (1) Straight vehicle safety domain model

[0123] The forward safety domain of the straight vehicle safety domain model adopts a semi-elliptical body configuration, and the rearward semi-spherical domain compensates for the blind area, as shown in Figure 4 The calculation method is as follows:

[0124]

[0125] (x, y) is the coordinate of the vehicle center, D / 2 is the radius of the rear semicircular domain, a is the major axis of the ellipse, and D / 2-L / 2 is the safety distance.

[0126] (2) Right turn vehicle safety domain model

[0127] In the right turn scenario, the significant increase in the right safety domain is mainly due to the coupling effect of multiple factors: first, the dynamic superposition effect of the A-pillar blind area, the vehicle head blind area and the rearview mirror blind area forces the driver to frequently switch the visual focus; second, due to the inherent influence of the left driving position, the driver's visual threshold coverage rate on the right side is more than 30% lower than on the left side. The safety domain in front of the right side of the vehicle will be higher than that in front of the left side. Therefore, the vehicle safety domain for right turn is adjusted as shown in Figure 5 The calculation method is as follows:

[0128]

[0129] Where L is the width of the road.

[0130] (3) Left turn vehicle safety domain model

[0131] In the left turn scenario, the left A-pillar blind area has an angle of 6° (only 2° on the right side), forming a larger sector occlusion area. Since the driving position is on the left side, the A-pillar is closer to the driver's visual axis, and its geometric projection blind area can completely cover pedestrians or electric vehicles below 1.5 meters in height. Therefore, the vehicle safety domain for left turn is adjusted as shown in Figure 6 The calculation method is as follows:

[0132]

[0133] Step 4, determine the intersection conflict position by predicting the vehicle trajectory equation, divide the intersection conflict into two categories of "cross conflict" and "merge conflict" according to the trajectory interaction mode, and quantitatively calculate the conflict risk of vehicles in the unsignalized intersection according to the constructed vehicle safety domain to obtain the traffic conflict exposure time;

[0134] Determine the intersection conflict position by predicting the vehicle trajectory equation, and the vehicle trajectory equation is:

[0135] s(t) = a0 + a1t + a2t 2 +a3t 3 +a4t 4 +a5t 5

[0136] t is time, a0, a1, a2, a3, a4, a5 are parameters of the trajectory equation fitting, and the intersection conflict position is solved by simultaneously solving the vehicle trajectory equation; According to the trajectory interaction mode, the intersection conflict area is divided into two categories of "cross conflict" and "merge conflict" as shown in Figure 7 When the included angle of different vehicles entering the intersection is greater than 90 degrees, it is a cross conflict, and when the included angle is less than 90 degrees, it is a merge conflict;

[0137] In different conflict scenarios, the conflict risk of vehicles in the unsignalized intersection is quantitatively calculated according to the constructed vehicle safety domain:

[0138] (1) Cross conflict

[0139] I. Cross conflict between straight vehicles, as shown in Figure 10 , 11

[0140] Calculate the included angle a between vehicles:

[0141]

[0142] In the formula, (x1, y1) and (x2, y2) are the coordinates of the two vehicles, a is the included angle between the vehicles, is the vehicle heading angle;

[0143] Calculate the boundary distance l between the center position of straight vehicle 1 and the vehicle safety domain α1 :

[0144]

[0145] In the formula, a is the included angle between the vehicles, Rbf is the current vehicle front domain longitudinal radius, and Rba is the current vehicle rear domain longitudinal radius;

[0146] ​The boundary distance l between the center position of the straight vehicle 2 and the vehicle safety domain is calculated α2 :

[0147]

[0148] In the formula, Sb is the lateral radius of the straight vehicle 2, is the heading angle of the straight vehicle 2, and because the rear domain of the vehicle safety domain is a semicircle, Sb = Rba, which can be further simplified:

[0149]

[0150] The inter-vehicle safety distance d is calculated as follows:

[0151] d = d0 - l α1 -l α2

[0152] In the formula, d0 is the distance between the centers of the two vehicles;

[0153] In intersection conflicts and merging conflicts, if the safety domains of vehicle 1 and vehicle 2 overlap (d < 0), the vehicle overlap duration is defined as a key indicator for measuring the degree of risk, that is, the Time-in-Vehicle Collision Risk (Tvcr). The greater the Tvcr value, the longer the vehicle stays in the possible collision area, and the higher the risk of conflict.

[0154] II. Intersection conflict between left-turn and straight vehicles, as shown in Figure 13

[0155] Because the vehicle safety domain of the left-turn vehicle 2 is adjusted, the boundary distance l between the center position of the left-turn vehicle 2 and the vehicle safety domain is calculated α2 :

[0156]

[0157] When the straight vehicle 1 is on the driver's side of the left-turn vehicle, V straight1 is calculated, and when the straight vehicle 1 is on the other side of the driver of the left-turn vehicle, V straight2 is calculated.

[0158] The boundary distance l between the center position of the straight vehicle 1 and the vehicle safety domain is calculated α1 :

[0159]

[0160] The inter-vehicle safety distance d is calculated as follows:

[0161] d = d0 - l α1 -l​α2

[0162] (2) Merging conflict

[0163] The characteristic of merging conflicts between right-turning and straight-going vehicles is that there is a clear following relationship between the vehicles, such as... Figure 14 As shown, calculate the boundary distance l between the center position of the straight-going vehicle 1 and the vehicle's safety domain. α1 :

[0164]

[0165] Calculate the boundary distance l between the center position of the right-turning vehicle 2 and the vehicle's safety zone. α2 :

[0166] l α2 =R ba

[0167] The safe distance d between vehicles is calculated as follows:

[0168] d = d0 - l α1 -l α2

[0169] Step 5: Based on the traffic conflict exposure time data, determine the optimal number of clusters K for the K-means algorithm using the elbow rule, and use the K-means algorithm to cluster the results of traffic conflict exposure time to classify the conflict risk.

[0170] The elbow rule is an unsupervised learning method used to determine the optimal number of clusters K in the K-means algorithm. It sets the range of K (e.g., K = 1 to 10) and calculates the sum of squared errors (SSE) of the total number of clusters (Tvcr).

[0171]

[0172] Where Ci is the i-th cluster, μi is the cluster center, K is the number of clusters, and μ represents Tvcr in the dataset;

[0173] The formula for calculating K-means is:

[0174]

[0175] K is the optimal number of clusters, Ci is the set of data points in the i-th cluster, and μi is the centroid of the i-th cluster; based on the clustering results, the range of conflict risk level classification is obtained.

[0176] The application further provides a vehicle safety domain modeling and conflict risk assessment system based on visual constraints.

[0177] A computer device comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the vehicle safety domain modeling and conflict risk assessment method based on visual constraints.

[0178] A computer readable storage medium has a computer program stored thereon, wherein the computer program is executable on a processor to implement the vehicle safety domain modeling and conflict risk assessment method based on visual constraints.

[0179] Embodiments

[0180] Step 1: Select a traffic accident-prone unsignalized intersection by analyzing accident black spots, collect multi-dimensional traffic information of the intersection by a UAV, including vehicle flow, vehicle type, lane number and intersection channelization.

[0181] In one embodiment, a bus station in a city of Jiangsu Province is taken as a typical unsignalized intersection scene. According to the traffic data statistics from June to November 2024 for a total of five months, the daily vehicle flow of this intersection is 1100 vehicles, of which about 100 are large vehicles, accounting for 9.1%. The peak period is concentrated from 7:00 to 9:00, and the average vehicle running speed is about 70 km / h. Since this intersection lacks red and green light signal control, traffic conflicts occur frequently, and it is highly representative and challenging.

[0182] Step 2: Determine the traffic capacity according to the vehicle flow, determine the vehicle dynamics parameters and mixed traffic flow model according to the vehicle type, determine the road network structure according to the lane number and intersection channelization, and draw the road network in the simulation software.

[0183] An unsignalized intersection is built in the simulation area, which has two lanes in the east-west direction and one lane in the south-north direction. The maximum speed of vehicles passing through the intersection is set to 70 km / h. The signal timing is set to no signal control. The simulation time is set to 3600s, and the peak traffic period is simulated, with a vehicle flow of 90 vehicles. The simulation parameter input is shown in Table 1.

[0184] Table 1 Simulation input data

[0185]

[0186] Step 3, considering the driver's visual constraints, three vehicle safety domain models for straight, right turn, and left turn are constructed. The vehicle safety domain refers to the domain in which the vehicle is in a safe state if there are no obstacles or overlap with other vehicle safety domains.

[0187] Step 4, by predicting the vehicle trajectory equation, the intersection conflict position is determined. According to the trajectory interaction mode, the intersection conflict is divided into "cross conflict" and "merge conflict". In different conflict scenarios, the conflict risk of vehicles in the unsignalized intersection is quantitatively calculated based on the constructed vehicle safety domain, and the traffic conflict exposure time is obtained. The conflict risk quantization results of different vehicle types in the unsignalized intersection are shown in Table 2.

[0188] Table 2 Conflict risk quantization results

[0189]

[0190] According to the vehicle risk quantization data analysis in Table 2, the following conclusions can be drawn: small cars dominate in the highest risk interval (Tvcr>12), accounting for 75% of the total number of events; while large cars are significantly concentrated in the low risk interval (88.9% in Tvcr≤10, of which 72.2% are in the 0-2 interval). Based on this, a differentiated control strategy is proposed.

[0191] Step 5, based on the traffic conflict exposure time data, the elbow rule is used to determine the optimal clustering number K of the K-means algorithm, and the K-means algorithm is used to cluster the traffic conflict exposure time results, and the conflict risk is classified. The elbow rule is used to determine the coefficient K of the K-means algorithm, as shown in Figure 15 The clustering analysis of the Tvcr index is shown in Figure 16 , which reveals the hierarchical distribution characteristics of road conflict risk. The clustering results show that the risk probability presents a significant hierarchical decreasing law: the proportion of slight conflict (threshold range 0-4) is the highest, reaching 64.4%; the proportion of moderate conflict (threshold range 4-12) is 22.1%; while the proportion of severe conflict (threshold >12) is only 13.5%. This distribution characteristic intuitively reflects the risk probability gradient difference in different conflict intensity intervals, of which more than 60% of vehicles are in a low risk state, while high risk conflicts only account for a small number, which is in good agreement with the actual traffic flow characteristics. Figure 17 The spatial distribution of the cumulative risk value of Tvcr is analyzed by three-dimensional visualization, and the closer to the intersection center, the more severe the conflict.

[0192] Table 3 Experimental results

[0193]

Claims

1. A method for visual constraint based vehicle safety domain modeling and collision risk assessment, characterized in that, The method comprises the following steps: Step 1, selecting a frequent-traffic-accident intersection by accident black spot analysis, collecting multi-dimensional traffic information of the intersection by a UAV, including vehicle flow, vehicle type, lane number and intersection channelization; Step 2, determining the traffic capacity according to the vehicle flow, determining the vehicle dynamics parameters and the mixed traffic flow model according to the vehicle type, determining the road network structure according to the lane number and the intersection channelization, and drawing the road network in a simulation software; Step 3, considering the driver's visual constraints, constructing three vehicle safety domain models of straight, right turn and left turn, the vehicle safety domain being a domain in which there is no obstacle or the safety domain of other vehicles does not overlap, and the vehicle in the area is in a safe state; Step 4, determining the intersection conflict position by a vehicle trajectory prediction equation, dividing the intersection conflict into two categories of "crossing conflict" and "merging conflict" according to the trajectory interaction mode, and quantitatively calculating the conflict risk of vehicles in the intersection according to the constructed vehicle safety domain in different conflict scenarios to obtain traffic conflict exposure time; Step 5, determining the optimal clustering number K of the K-means algorithm by the elbow rule based on the traffic conflict exposure time data, and clustering the results of the traffic conflict exposure time by the K-means algorithm to classify the conflict risk.

2. The vision constraint based vehicle safety domain modeling and collision risk assessment method according to claim 1, wherein, Step 1, selecting a frequent-traffic-accident intersection by accident black spot analysis, collecting multi-dimensional traffic information of the intersection by a UAV, including vehicle flow, vehicle type, lane number and intersection channelization, wherein: (1) selecting a frequent-traffic-accident intersection by accident black spot analysis; The accident black spot analysis selects a frequent-traffic-accident intersection by using kernel density estimation and text similarity algorithm, and the specific process is as follows: Inputting the positions of different vehicles at the intersection to perform kernel density estimation to obtain vehicle spatial density distribution, and the intersection with the top 20% density peak value is a hidden intersection, wherein the kernel density estimation calculation formula is as follows: is the probability density estimation value at r, r is the current unsignalized intersection vehicle position, n is the number of all unsignalized intersection vehicles, h is the bandwidth, K is the kernel function, and Ri is the position of the ith unsignalized intersection vehicle. Performing text similarity calculation on the accident report or hidden intersection hazard investigation report to obtain the intersection with the highest accident frequency, and listing the intersection as a priority intersection for rectification, wherein the text similarity calculation formula is as follows: w is a traffic accident text, v is an accident report, TF(w, v) is the frequency of w in the document v, DF(w) is the number of reports containing the traffic accident text word w, and N is the total number of reports; (2) collecting multi-dimensional traffic information of the intersection by a UAV, including vehicle flow, vehicle type, lane number and intersection channelization; Selecting an early morning peak or late evening peak as the UAV shooting time, collecting multi-dimensional traffic video of the intersection by a UAV, and using UNet algorithm to extract data from the video shot by the UAV, including vehicle flow, vehicle type, lane number and intersection channelization.

3. The vision constraint based vehicle safety domain modeling and collision risk assessment method of claim 1, wherein, Step 2, determining the traffic capacity according to the vehicle flow, determining the vehicle dynamics parameters and the mixed traffic flow model according to the vehicle type, determining the road network structure according to the lane number and the intersection channelization, and drawing the road network in a simulation software, and the specific method is as follows: Determining the traffic capacity according to the vehicle flow collected by the UAV, wherein the traffic capacity calculation formula is as follows: T0 is the design capacity, N is the number of vehicles, and Nmax is the maximum number of vehicles that can be accommodated; The vehicle dynamics parameters are set by vehicle type, including vehicle size, maximum speed, maximum acceleration and deceleration, the proportion of vehicle types in the mixed traffic flow model is determined, the road network structure is determined according to the number of lanes and intersection channelization, and the road network is drawn in the micro-traffic simulation software SUMO.

4. The vision constraint based vehicle safety domain modeling and collision risk assessment method of claim 1, wherein, Step 3: Considering the driver's visual constraints, three vehicle safety domain models for straight, right turn and left turn are constructed. The vehicle safety domain refers to the domain where there are no obstacles or the safety domains of other vehicles do not overlap, and the vehicle in this domain is in a safe state. Among them: (1) Straight vehicle safety domain model; In the formula, (x, y) is the coordinate of the vehicle center, L is the width of the road, D / 2 is the radius of the rear semicircular domain, and a is the major axis of the ellipse; (2) Right turn vehicle safety domain model (3) Left turn vehicle safety domain model 5. The vision constraint based vehicle safety domain modeling and collision risk assessment method of claim 1, wherein, Step 4: By predicting the vehicle trajectory equation, the intersection conflict position is determined. According to the trajectory interaction mode, the intersection conflict is divided into two categories: "cross conflict" and "merge conflict". In different conflict scenarios, the conflict risk of vehicles in the unsignalized intersection is quantitatively calculated according to the constructed vehicle safety domain, and the traffic conflict exposure time is obtained. Among them: By predicting the vehicle trajectory equation, the intersection conflict area is determined. The vehicle trajectory equation is: s(t) - a0+ a1t + a2t 2 + a3t 3 + a4t 4 + a5t 5 t is the time, a0, a1, a2, a3, a4, a5 are the parameters of the trajectory equation fitting, and the intersection conflict position is solved by simultaneously solving the vehicle trajectory equation; According to the trajectory interaction mode, the intersection conflict area is divided into two categories: "cross conflict" and "merge conflict". When the angle between different vehicles entering the intersection is greater than 90 degrees, it is a cross conflict, and when the angle is less than 90 degrees, it is a merge conflict. In different conflict scenarios, the conflict risk of vehicles in the unsignalized intersection is quantitatively calculated according to the constructed vehicle safety domain: (1) Cross conflict I. Cross conflict between straight vehicles Calculate the angle α between vehicles: where (xl,yl) and (x2,y2) are the coordinates of the two vehicles, a is the included angle between the vehicles, is the vehicle heading angle; calculating a boundary distance l between the center position of the straight-ahead vehicle 1 and the safety domain of the vehicle α1 : In the formula, α is the angle between vehicles, Rbf is the longitudinal radius of the front domain of straight vehicle 1, and Rba is the longitudinal radius of the rear domain of straight vehicle 1. calculating a boundary distance l between the center position of the straight-ahead vehicle 2 and the safety domain of the vehicle α2 : where Sb is the lateral radius of the straight vehicle 2, is the heading angle of the straight vehicle 2, and since the rear region of the vehicle safety domain is a semi-circle, Sb = Rba, which further simplifies to: The safety distance d between vehicles is calculated as follows: d = do - l α1 -l α2 In the formula, d0 is the distance between the centers of the two vehicles. II. Cross conflict between left turn and straight vehicles calculating a boundary distance l between the center position of the straight-ahead vehicle 1 and the safety domain of the vehicle α1 : Since the vehicle safety domain of the left-turn vehicle 2 is adjusted, the boundary distance I between the center position of the left-turn vehicle 2 and the vehicle safety domain is calculated α2 : The straight vehicle 1 uses V straight1 The straight vehicle 1 uses V straight2 The straight vehicle 1 uses V The safety distance d between vehicles is calculated as follows: d = do - l α1 -l α2 (2) Merge conflict The characteristics of the merging conflict between the right-turn vehicle and the straight-going vehicle are that there is a clear following relationship between the vehicles, and the boundary distance l between the center position of the straight-going vehicle 1 and the safety domain of the vehicle is calculated α1 : calculating a boundary distance I between the center position of the right-turning vehicle 2 and the vehicle safety domain α2 : l α2 ＝R ba The safety distance d between vehicles is calculated as follows: d = do - l α1 -l α2 In the cross conflict and merge conflict, if the safety domain of vehicle 1 and vehicle 2 overlaps, i.e. d < 0, the vehicle overlap duration is defined as a key indicator to measure the risk level, i.e. traffic conflict exposure time Tvcr. The larger the Tvcr value, the longer the vehicle stays in the possible collision area, and the higher the conflict risk.

6. The vision constraint based vehicle safety domain modeling and collision risk assessment method of claim 1, wherein, Step 5: Based on the traffic conflict exposure time data, the elbow rule is used to determine the optimal clustering number K of the K-means algorithm, and the K-means algorithm is used to cluster the results of traffic conflict exposure time, and the conflict risk is classified. Elbow rule is an unsupervised learning method to determine the optimal number of clusters K in K-means algorithm, set the range of K, and calculate the sum of squared errors SSE of traffic conflict exposure time Tvcr: Where Ci is the ith cluster, μi is the cluster center, K is the number of clusters, and μ represents the traffic conflict exposure time Tvcr. The K-means calculation formula is: K is the optimal number of clusters, Ci is the data point set of the ith cluster, and μi is the centroid of the ith cluster. According to the clustering results, the range of conflict risk classification is obtained.

7. A vision constraint based vehicle safety domain modeling and collision risk assessment system, characterized in that, The method for modeling vehicle safety domain and evaluating conflict risk based on visual constraints according to any one of claims 1-6 is implemented to realize modeling vehicle safety domain and evaluating conflict risk based on visual constraints, and five modules are used to perform steps 1-5 respectively.

8. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the method for modeling vehicle safety domain and evaluating conflict risk based on visual constraints according to any one of claims 1-6 is implemented to realize modeling vehicle safety domain and evaluating conflict risk based on visual constraints.

9. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for modeling vehicle safety domain and evaluating conflict risk based on visual constraints according to any one of claims 1-6 is implemented to realize modeling vehicle safety domain and evaluating conflict risk based on visual constraints.

Images