Human-vehicle interactive simulation method and system based on improved social force model and application

Abstract

The invention relates to the field of urban traffic management and supervision and the field of traffic simulation, and provides a human-vehicle interactive simulation method and system based on an improved social force model and application. The method comprises the following steps: S1, constructing an improved social force model; s2, constructing a decision-making model, wherein the decision-making model is used for determining a behavior mode when the pedestrian and the vehicle meet; and S3, giving initial basic data of pedestrians and vehicles, and performing interactive simulation on human-vehicle movement conditions by using the improved social force model and the decision model. The method can be applied to scenes where vehicles and pedestrians are crossed and mixed, such as intersections without traffic lights, including crossroads and multi-directional intersections. According to the method, the trueness of the model in a complex traffic environment is remarkably improved, and behavior decisions of pedestrians in an actual scene can be more accurately simulated, especially under the condition that potential conflicts occur between the pedestrians and vehicles; through the introduction of modular design and transfer learning, the model can quickly adapt to a dynamically changing environment.

Description

Technical Field

[0001] The present invention relates to the fields of urban traffic management, supervision and traffic simulation, and in particular to a human-vehicle interaction simulation method, system and application based on an improved social force model, which can be used in traffic management and supervision, and is particularly suitable for information processing systems for traffic prediction purposes. Background Art

[0002] In urban mixed traffic systems, pedestrians will have frequent and complex interactions with different traffic participants and the surrounding environment. These interactions involve individual pedestrians' path selection, speed adjustment, and avoidance of potential risks. In order to better understand and simulate pedestrian behavior, social force models are widely used in pedestrian simulation. This model can effectively simulate how pedestrians adjust their movement trajectories in the traffic environment to avoid collisions with others. It can also combine environmental factors such as buildings and obstacles to further optimize the simulation of pedestrian movement.

[0003] In the social force model, pedestrians are regarded as moving particles subject to force. The movement trajectory is determined by the motion simulation equation in the model. The behavior of each pedestrian is described by the psychological force and physical force in the equation. The psychological force reflects the pedestrian's inner intention, and the physical force reflects the collision avoidance behavior generated by the pedestrian to avoid collision. At each simulation moment, pedestrians dynamically adjust their motion state according to the resultant force obtained from the model equation, thereby truly reproducing the motion behavior of pedestrians in real scenes.

[0004] Authorized Chinese patent CN115092166B is an interactive autonomous driving speed planning method based on a social force model, which solves the problem that the current speed planning method ignores the interaction between people, resulting in poor rationality of the planned speed. Authorized Chinese patent CN110414365B is a pedestrian trajectory prediction method, system and medium based on a social force model, which can improve the safety of autonomous driving vehicles in zebra crossing areas with mixed pedestrians and vehicles and reduce vehicle delays. Authorized Chinese patent CN102682303B is a crowd abnormal event detection method based on the LBP weighted social force model, which detects abnormal crowd behavior by combining optical flow and LBP spectrum to perform innovative calculations on social forces.

[0005] The social force model has shown good results in most pedestrian simulations, but it has obvious limitations when dealing with complex human-vehicle interaction scenarios. First, it ignores speed differences. The formula of the social force model does not fully consider the psychological force differences that vehicles of different speeds exert on pedestrians. In actual scenarios, pedestrians' perception and reaction to vehicles often vary significantly with changes in vehicle speed. Vehicles traveling at high speeds usually make pedestrians feel greater pressure and urgency, which leads to pedestrians being more cautious or quick when making avoidance decisions. However, when calculating the interaction force between pedestrians and vehicles, the social force model usually fails to adjust the weight of the vehicle's speed as an influencing factor, resulting in an inaccurate simulation of pedestrian reactions and an inability to truly reflect pedestrians' psychological pressure and behavioral choices. The second is the problem of decision-making limitations. The model relies on simple mechanical calculations and mainly describes the movement behavior of pedestrians through physical and psychological forces. However, in the process of human-vehicle interaction, pedestrians' decisions are often affected by multiple factors, such as the perception and judgment of vehicle speed, intention, and surrounding environmental risks, and these complex cognitive processes are difficult to accurately simulate with a simple mechanical model.

[0006] Existing traffic management requires real-time monitoring of traffic flow and even early prediction. There is an urgent need to develop effective traffic data processing methods and systems with predictive purposes. Summary of the invention

[0007] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a human-vehicle interaction simulation method, system and application based on an improved social force model, which solves the problem of pedestrian behavior simulation in complex human-vehicle interaction scenarios by introducing an improved social force model and a decision-making model.

[0008] The present invention adopts the following technical solution:

[0009] In one aspect, the present invention provides a method for simulating human-vehicle interaction based on an improved social force model, comprising:

[0010] S1. Construct an improved social force model, which is used to simulate the influence of driving force, human-vehicle interaction force, pedestrian-to-pedestrian interaction force, and obstacle repulsion on pedestrian travel speed; the improved social force model optimizes the influence of speed, distance, and interaction angle between people and vehicles on pedestrian travel;

[0011] S2. Constructing a decision model, wherein the decision model is used to determine the behavior pattern of pedestrians and vehicles when they meet; the decision model optimizes the influence of speed, distance and interaction angle between pedestrians and vehicles on pedestrian decision-making;

[0012] S3. Given the initial basic data of pedestrians and vehicles, the improved social force model of step S1 and the decision model of step S2 are used to interactively simulate the movement of pedestrians and vehicles.

[0013] Any possible implementation described above further provides an implementation, in step S1, the improved social force model is expressed as:

[0014]

[0015] (2)

[0016]

[0017] f ij psy = A ped exp. [ ( r ij - d ij ) / B ped ] n ij ⃗ （ 4 ）

[0018]

[0019]

[0020] f io psy = A obs exp. [ ( r i - d io ) / B obs ] n io ⃗ （ 7 ）

[0021]

[0022]

[0023]

[0024] (11)

[0025] (12)

[0026] Where: The destination is the driving force for pedestrians, driving them to the destination. Pedestrians adjust their current speed and direction according to the desired speed and direction within a certain interval; For the interaction between pedestrians; is the repulsive force of obstacles on pedestrians; is the interaction force between pedestrians and vehicles; is a random factor; N1, N2, and N are the total number of vehicles, obstacles, and pedestrians, respectively;

[0027] , is the mass and speed of pedestrians, , They represent the pedestrian’s expected speed and expected direction of movement, respectively. is the interval time;

[0028] They are the psychological and physical forces of interaction between pedestrians; is the distance between the centroids of pedestrians, , are two pedestrian radii When two pedestrians approach each other, a psychological force is generated to avoid each other; the physical force is determined by Decide, is a piecewise function. When the distance between pedestrians is less than a certain distance, > , physical force is generated; when ≤ hour, , no physical force is generated; , is a parameter used to adjust the intensity of interaction between pedestrians. , are the elastic coefficient and sliding friction coefficient of the human body, which respectively reflect the coefficient of the human body's reaction to collision or approach and the friction coefficient generated when approaching. is the direction of the force exerted by pedestrian j on target pedestrian i, is the speed difference between pedestrians in the tangential direction;

[0029] They are the psychological force and physical force of the obstacles’ repulsive force on pedestrians, Pedestrian radius, is the distance between the center of mass of the pedestrian and the obstacle. When the pedestrian and the obstacle are close, a psychological force will be generated to avoid getting close to each other. The physical force is determined by Determine that when the distance between pedestrians and obstacles is less than a certain distance, > , physical force is generated; when ≤ hour, ; , It is a parameter used to adjust the intensity of the obstacle’s effect on pedestrians. , are the elastic coefficient and sliding friction coefficient of the human body, which respectively reflect the coefficient of the human body's reaction to collision or approach and the friction coefficient generated when approaching. is the direction of the force acting on the obstacle. is the unit vector tangential to the pedestrian and the obstacle; is the velocity vector of pedestrian i;

[0030] , Parameters used to adjust the intensity of the vehicle's impact on pedestrians, is the distance between the pedestrian and the vehicle; the anisotropy coefficient is used to describe the relationship between the interaction force and the angle between the vehicle driving direction and the pedestrian moving direction. is the parameter to adjust the anisotropic characteristics, Indicates the angle between the pedestrian's movement direction and the vehicle's travel direction, ranging from arrive between; is the unit vector of pedestrian i pointing to the vehicle;

[0031] Used to assess whether the speed difference between vehicles and pedestrians exceeds a set threshold Parameters, is the speed difference between pedestrians and vehicles, u is the parameter for adjusting the speed difference affecting the force, is the vehicle speed.

[0032] Any possible implementation described above further provides an implementation in which the parameters in the improved social force model are , , , , , , , u is optimized by Bayesian optimization method; the Bayesian optimization method is as follows:

[0033] Z1, the Bayesian model is:

[0034] Parameter Space ,in ={ , , , , , , , }, each Represents a parameter group used to perform an objective function evaluation;

[0035] Objective function F( ) = Y: Y is the input parameter group The output target value after the simulation is obtained, and the output target value Y is the average displacement error of the pedestrians; the predicted trajectories of all pedestrians are obtained after a certain simulation process, and the average displacement error of the pedestrians is calculated based on the predicted trajectory and the actual trajectory; the average displacement error The formula is:

[0036] ;

[0037] is the x-coordinate of the predicted position of pedestrian i at time t, is the y coordinate of the predicted position of pedestrian i at time t, is the x-coordinate of the real position of pedestrian i at time t, is the y coordinate of the real position of pedestrian i at time t, and N is the total number of pedestrians;

[0038] Acquisition function S: used to select the optimal parameter set for the next evaluation ;

[0039] ;

[0040] Dataset D: contains all parameter sets that have been evaluated and the corresponding target value ;

[0041] , n is the total number of evaluations;

[0042] Z2, initialization:

[0043] Randomly select K parameter groups for initial evaluation of the objective function to obtain the data set D, where K is the number of initialization samplings;

[0044]

[0045] Z3, iterative optimization:

[0046] At each iteration, Bayesian optimization performs the following steps:

[0047] Z31, Gaussian process modeling:

[0048] Assume that the objective function is continuous in the parameter space and satisfies the Gaussian process; this means that the value of the objective function at any position in the parameter space can be described by a mean function and a variance function; specifically, the distribution form of the value of the objective function at a given position is expressed as:

[0049] ;

[0050] is the mean of all objective function values ​​that have been iterated so far, is the variance of all objective function values ​​that have been iterated; P() is the probability distribution, D t is the current data set, x t+1 is the parameter group for the next iteration;

[0051] Z32, select the next parameter group: At each iteration, Bayesian optimization selects the next set of parameters to evaluate; through the acquisition function S, the parameters with the greatest potential improvement are selected :

[0052] ;

[0053] ;

[0054] To iterate to the optimal target value of the current step, is the objective function value of the parameter group;

[0055] Z33, Update the data set: The selected parameter points are passed through the objective function to calculate the objective function value, and the result is added to the existing training data; then, the updated data set is used to refit the Gaussian process model to make more accurate predictions in the next iteration:

[0056]

[0057] Z34, termination condition: reaching the maximum number of iterations or satisfying the convergence condition, and obtaining the optimal parameter set ;

[0058] ={ , , , , , , , }.

[0059] According to any possible implementation described above, a further implementation is provided, in step S2, the decision method adopted by the decision model is specifically:

[0060] X1. Calculate the future conflict point between pedestrians and vehicles

[0061] X2. If there is a conflict point, calculate the time it takes for pedestrians and vehicles to reach the conflict point according to their current speeds and directions, and determine the order of arrival;

[0062] X3. If the vehicle reaches the conflict point first, then:

[0063] X31. Calculate the position of the pedestrian when the vehicle reaches the conflict point, and determine the risk area and the danger area of ​​the vehicle at the conflict point according to the driving direction of the vehicle; the risk area is an area where there is a potential conflict between the pedestrian and the vehicle, and the danger area is an area where the conflict between the pedestrian and the vehicle cannot be avoided;

[0064] X32, determine whether the pedestrian's future position is within the risk area;

[0065] If the pedestrian's future position is not within the risk area of ​​the vehicle, the pedestrian's motion state is calculated according to the improved social force model;

[0066] If the pedestrian is in the risk area, it is further determined whether it is in the dangerous area;

[0067] If the pedestrian is in the risk area and not in the danger area, the time difference between the pedestrian and the vehicle reaching the conflict point is calculated; if the time difference does not exceed the set threshold, the decision output is to slow down in the current direction; if the time difference exceeds the threshold, the decision output is to enter hesitation;

[0068] If the pedestrian is in the danger zone, the interaction angle between the pedestrian and the vehicle is further determined; when the interaction angle is lateral interaction, the decision output is an emergency stop in the current direction; when the interaction angle is front-to-back interaction, the decision output is a sharp turn;

[0069] X4. If the pedestrian reaches the conflict point first, then:

[0070] X41. Calculate the position of the vehicle when the person arrives at the conflict point, and determine the risk area and danger area of ​​the vehicle at the conflict point based on the vehicle's driving direction;

[0071] X42. Determine whether the vehicle is located in the risk area;

[0072] If the pedestrian is not in the risk area, the motion state calculation continues based on the social force model;

[0073] If the pedestrian is in the risk area, it is further determined whether the pedestrian is in the danger area or the risk area;

[0074] If the pedestrian is in the risk area and not in the dangerous area, the time difference between the pedestrian and the vehicle reaching the conflict point is calculated; if the time difference does not exceed the set threshold, the output decision is to slowly accelerate in the current direction; if the time difference exceeds the threshold, the output enters a hesitant state;

[0075] If the pedestrian is in a dangerous area, a judgment is made based on the interaction angle; when the interaction angle is a lateral interaction, the output decision is to run and accelerate in the current direction; when the interaction angle is a forward and backward interaction, the output decision is a sharp turn.

[0076] Any possible implementation described above further provides an implementation, in which the specific method of step S3 includes:

[0077] S31, inputting initial basic data of pedestrians and vehicles, wherein the basic data includes initial positions, speeds, time steps, and destinations of pedestrians and vehicles;

[0078] S32, calculating the target position of the pedestrian at the current time step, wherein the target position is determined according to the planned path between the actual position at the current time step and the destination;

[0079] According to the Astar algorithm, the final destination location is input, and the Astar algorithm will plan an optimal path (shortest or lowest cost) for pedestrians to reach the final destination from the current location. This path is composed of a series of discrete points, each of which can be regarded as a node on the path. The algorithm will output the target location (the next node to move towards at the current moment);

[0080] S33, inputting the pedestrian's current velocity vector, current position and target position into the improved social force model to calculate the resultant force on the pedestrian;

[0081] S34, entering the decision model, and generating a decision output for the current time step based on the relative distance and speed between the pedestrian and the vehicle;

[0082] S35, adjusting the resultant force on the pedestrian according to the decision output obtained at the current time step;

[0083] S36, adjusting the speed of the pedestrian according to the resultant force adjusted in step S35, and calculating the actual position of the pedestrian in the next time step;

[0084] S37. Repeat steps S32-S36 until all time steps are completed.

[0085] According to any of the possible implementations described above, a further implementation is provided. In step S35, the specific method for adjusting the resultant force on the pedestrian according to the decision output is:

[0086] When pedestrians choose to make a sharp turn, the improved social force model adjusts the social force to a force perpendicular to the direction of the vehicle according to the relative position of the pedestrian and the vehicle. The direction of the force depends on the position of the pedestrian relative to the vehicle and is determined according to the direction to be avoided.

[0087] When the pedestrian chooses to accelerate slowly or run, the social force is adjusted to the acceleration force along the current speed direction, and the pedestrian's walking speed and acceleration upper limit are adjusted to a higher running speed and acceleration;

[0088] When the pedestrian chooses to slow down or stop suddenly, the social force is adjusted to the acceleration force along the current reverse speed, and the pedestrian's acceleration is replaced by a higher emergency stop acceleration;

[0089] When the pedestrian leaves the danger zone but is still in the risk area, maintain the previous decision and continue running, slowing down or turning until leaving the risk area.

[0090] In step S32, the pedestrian target position is calculated using the Astar algorithm.

[0091] On the other hand, the present invention also provides a human-vehicle interaction simulation system based on an improved social force model, the system is used to implement the above method, the system comprises:

[0092] The social force calculation module calculates the social forces on pedestrians based on the improved social force model;

[0093] A decision module is used to determine the behavior patterns of pedestrians and vehicles when they meet;

[0094] The interactive simulation module is configured to interactively simulate the movement of pedestrians and vehicles by using the social force calculation module and the decision-making module given the initial basic data of pedestrians and vehicles.

[0095] On the other hand, the present invention also provides an application of a human-vehicle interaction simulation method based on an improved social force model, wherein the method is applied to a scene where vehicles and pedestrians cross and mix.

[0096] As any possible implementation described above, an implementation is further provided, wherein the method is applied to an intersection without a traffic light; the intersection includes a cross intersection and a multi-directional intersection.

[0097] The beneficial effects of the present invention are:

[0098] 1. Improve the realism of simulation trajectories: The present invention significantly improves the realism in complex traffic environments by combining the Astar algorithm, an improved social force model, and a decision-making model based on human-vehicle interaction. Traditional models often ignore subtle interaction details when dealing with the dynamic interaction between pedestrians and vehicles, such as the immediate impact of vehicle speed changes on pedestrian behavior. By introducing dynamic factors such as relative speed and distance, the present invention can more accurately simulate the behavioral decisions of pedestrians in actual scenarios, especially in situations where there is a potential conflict between pedestrians and vehicles. Compared with the traditional social force model, the improved social force model reduces the average displacement error by 11%.

[0099] 2. Improve the safety of simulation results. By combining with the decision-making module, pedestrians can more accurately predict and respond to potential collision risks during their interaction with vehicles. Compared with the traditional social force model, the improved social force model reduces the relative error of the post-encroachment time by an average of 8%.

[0100] Post-encroachment time is an important indicator used in the traffic field to assess potential conflict risks. It refers to the difference in passing time between two traffic participants (such as pedestrians and vehicles) at a common conflict point.

[0101]

[0102] is the time it takes for the vehicle to reach the conflict point at its current speed and direction. is the time it takes for the pedestrian to reach the conflict point at the current speed and direction.

[0103] The calculation formula of the relative error of post-occupancy time is:

[0104]

[0105] is the PET value of the simulation trajectory, is the PET value of the true trajectory.

[0106] 3. The simulation system of the present invention can be applied to traffic flow management and supervision in real scenarios to warn of possible traffic accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0107] Figure 1 Shown is a schematic diagram of a decision model in an embodiment of the present invention.

[0108] Figure 2 Shown is a flow chart of interactive simulation in an embodiment.

[0109] Figure 3 Shown is a schematic diagram of a vehicle danger zone in an embodiment. DETAILED DESCRIPTION

[0110] The specific embodiments of the present invention will be described in detail below in conjunction with specific drawings. It should be noted that the technical features or combinations of technical features described in the following embodiments should not be considered isolated, and they can be combined with each other to achieve better technical effects.

[0111] An embodiment of the present invention provides a human-vehicle interaction simulation method based on an improved social force model, comprising:

[0112] S1. Construct an improved social force model, which is used to simulate the influence of driving force, human-vehicle interaction force, pedestrian-to-pedestrian interaction force, and obstacle repulsion on pedestrian travel speed; the improved social force model optimizes the influence of speed, distance, and interaction angle between people and vehicles on pedestrian travel;

[0113] S2. Constructing a decision model, wherein the decision model is used to determine the behavior pattern of pedestrians and vehicles when they meet; the decision model optimizes the influence of speed, distance and interaction angle between pedestrians and vehicles on pedestrian decision-making;

[0114] S3. Given the initial basic data of pedestrians and vehicles, the improved social force model of step S1 and the decision model of step S2 are used to interactively simulate the movement of pedestrians and vehicles.

[0115] The specific process is as follows Figure 2 shown.

[0116] 1. Traditional social force model:

[0117] In the social force model, it can be divided into driving force, human-vehicle interaction force, pedestrian interaction force, obstacle repulsion and a random force, as shown in formula (1).

[0118]

[0119] is the driving force of the destination on the pedestrian, driving the pedestrian to the destination. As shown in formula (2), the pedestrian will adjust the current speed and direction according to the expected speed and direction within a certain interval. , is the mass and speed of pedestrians, , represents the pedestrian’s expected speed and expected direction of movement, For the interval time.

[0120]

[0121] The interaction forces between pedestrians can be divided into psychological forces and physical forces, as shown in formulas (3), (4), and (5); is the distance between the centroids of pedestrians, , are two pedestrian radii When two pedestrians are not in physical contact ( < ), a psychological force will be generated to avoid getting close to each other, and the physical force is determined by Determine that when the distance between pedestrians is less than a certain distance ( > ) generates physical force; when ≤ hour, , no physical force is generated; , is a parameter used to adjust the intensity of interaction between pedestrians. , are the elastic coefficient and sliding friction coefficient of the human body, which respectively reflect the coefficient of the human body's reaction to collision or approach and the friction coefficient generated when approaching. is the direction of the force exerted by pedestrian j on target pedestrian i, is the speed difference between pedestrians in the tangential direction.

[0122]

[0123] f ij psy = A i exp. [ ( r ij - d ij ) / B i ] n ij ⃗ （ 4 ）

[0124]

[0125] (11-1)

[0126] The repulsive force of obstacles on pedestrians can also be divided into psychological force and physical force, as shown in formulas (6), (7), and (8): The radius of the pedestrian, is the distance between the pedestrian and the obstacle’s center of mass. When there is no physical contact between the pedestrian and the obstacle ( < ), a psychological force will be generated to avoid getting close to each other, and the physical force is determined by Determine that when the distance between the pedestrian and the obstacle is less than a certain distance ( > ) generates physical force, A and B are parameters used to adjust the strength of the wall's effect on pedestrians. , are the elastic coefficient and sliding friction coefficient, is the direction of the force acting on the obstacle. is the unit vector tangential to the pedestrian and the obstacle.

[0127]

[0128] f io psy = A obs exp. [ ( r i - d io ) / B obs ] n io ⃗ （ 7 ）

[0129]

[0130] (11-2)

[0131] The interaction force between pedestrians and vehicles is mainly composed of the attenuation function and the anisotropy function, as shown in formulas (a), (b) and (c). A and B are used to adjust the parameters of the intensity of the vehicle's effect on pedestrians. is the distance between the pedestrian and the vehicle. The anisotropy coefficient is used to describe the relationship between the interaction force and the angle between the vehicle driving direction and the pedestrian moving direction. is the parameter to adjust the anisotropic characteristics, Indicates the angle between the pedestrian's movement direction and the vehicle's travel direction, ranging from arrive between.

[0132] ( ,A,B)

[0133]

[0134]

[0135] 2. Improved social force model:

[0136] Current pedestrian-vehicle social force models mainly consider the impact of the distance and interaction angle between pedestrians and vehicles on the forces acting on pedestrians. However, in actual traffic scenarios, the speed difference between pedestrians and vehicles is often very significant, especially during vehicle acceleration or emergency braking, when the time and distance required are quite different from the motion characteristics of pedestrians. Therefore, considering only the distance and angle is not enough to fully describe the forces exerted by vehicles on pedestrians. For example, under the same interaction distance and angle conditions, a vehicle traveling at a low speed and approaching a pedestrian will have a completely different impact on pedestrians than a vehicle traveling at a high speed.

[0137] Therefore, the speed difference effect is added to the vehicle force calculation. A parameter used to evaluate whether the speed of vehicles and pedestrians exceeds a set threshold. When the vehicle speed is within the acceptable range for pedestrians, the force exerted by the vehicle on pedestrians can ignore the impact of speed differences. However, when the vehicle speed exceeds this threshold, it means that pedestrians may perceive that there is a potential safety risk in interacting with the vehicle, which in turn has a significant impact on their behavior. The speed difference between pedestrians and vehicles is used to adjust the parameters of the force affected by the speed difference.

[0138]

[0139]

[0140] (12)

[0141] Parameter optimization:

[0142] The parameters in the improved social force model , , , , , , , u is optimized by Bayesian optimization method; the Bayesian optimization method is as follows:

[0143] Z1, the Bayesian model is:

[0144] Parameter Space ,in ={ , , , , , , , }, each Represents a parameter group used to perform an objective function evaluation;

[0145] Objective function F( ) = Y: Y is the input parameter group The output target value after the simulation is obtained, and the output target value Y is the average displacement error of the pedestrians; the predicted trajectories of all pedestrians are obtained after a certain simulation process, and the average displacement error of the pedestrians is calculated based on the predicted trajectory and the actual trajectory; the average displacement error The formula is:

[0146] ;

[0147] is the x-coordinate of the predicted position of pedestrian i at time t, is the y coordinate of the predicted position of pedestrian i at time t, is the x-coordinate of the real position of pedestrian i at time t, is the y coordinate of the real position of pedestrian i at time t, and N is the total number of pedestrians;

[0148] Acquisition function S: used to select the optimal parameter set for the next evaluation ;

[0149] ;

[0150] Dataset D: contains all parameter sets that have been evaluated and the corresponding target value ;

[0151] , n is the total number of evaluations;

[0152] Z2, initialization:

[0153] Randomly select K parameter groups for initial evaluation of the objective function to obtain the data set D, where K is the number of initialization samplings;

[0154]

[0155] Z3, iterative optimization:

[0156] At each iteration, Bayesian optimization performs the following steps:

[0157] Z31, Gaussian process modeling:

[0158] Assume that the objective function is continuous in the parameter space and satisfies the Gaussian process; this means that the value of the objective function at any position in the parameter space can be described by a mean function and a variance function; specifically, the distribution form of the value of the objective function at a given position is expressed as:

[0159] ;

[0160] is the mean of all objective function values ​​that have been iterated so far, is the variance of all objective function values ​​that have been iterated; P() is the probability distribution, D t is the current data set, x t+1 is the parameter group for the next iteration;

[0161] Z32, select the next parameter group: At each iteration, Bayesian optimization selects the next set of parameters to evaluate; through the acquisition function S, the parameters with the greatest potential improvement are selected :

[0162] ;

[0163] ;

[0164] To iterate to the optimal target value of the current step, is the objective function value of the parameter group;

[0165] Z33, Update the data set: The selected parameter points are passed through the objective function to calculate the objective function value, and the result is added to the existing training data; then, the updated data set is used to refit the Gaussian process model to make more accurate predictions in the next iteration:

[0166]

[0167] Z34, termination condition: reaching the maximum number of iterations or satisfying the convergence condition, and obtaining the optimal parameter set ;

[0168] ={ , , , , , , , }.

[0169] 3. Decision model

[0170] like Figure 1 As shown, the decision-making method adopted by the decision model is specifically:

[0171] X1. Calculate the future conflict point between pedestrians and vehicles

[0172] X2. If there is a conflict point, calculate the time it takes for pedestrians and vehicles to reach the conflict point according to their current speeds and directions, and determine the order of arrival;

[0173] X3. If the vehicle reaches the conflict point first, then:

[0174] X31. Calculate the position of the pedestrian when the vehicle reaches the conflict point, and determine the risk area and the danger area of ​​the vehicle at the conflict point according to the driving direction of the vehicle; the risk area is an area where there is a potential conflict between the pedestrian and the vehicle, and the danger area is an area where the conflict between the pedestrian and the vehicle cannot be avoided;

[0175] In a specific embodiment, there is a certain buffer distance between the risk area and the non-hazardous area, and the conflict has not reached an emergency level. The hazardous area directly causes a collision or a serious accident.

[0176] like Figure 3 As shown, assuming the width of the vehicle is w and the length is l, the danger zone is the current width and length of the vehicle extending outwards. The distance is set to a length of The buffer part, ,k is a hyperparameter, is the speed of the vehicle.

[0177] X32, determine whether the pedestrian's future position is within the risk area;

[0178] If the pedestrian's future position is not within the risk area of ​​the vehicle, the pedestrian's motion state is calculated according to the improved social force model;

[0179] If the pedestrian is in the risk area, it is further determined whether it is in the dangerous area;

[0180] If the pedestrian is in the risk area and not in the danger area, the time difference between the pedestrian and the vehicle reaching the conflict point is calculated; if the time difference does not exceed the set threshold, the decision output is to slow down in the current direction; if the time difference exceeds the threshold, the decision output is to enter hesitation;

[0181] If the pedestrian is in the danger zone, the interaction angle between the pedestrian and the vehicle is further determined; when the interaction angle is lateral interaction, the decision output is an emergency stop in the current direction; when the interaction angle is front-to-back interaction, the decision output is a sharp turn;

[0182] X4. If the pedestrian reaches the conflict point first, then:

[0183] X41. Calculate the position of the vehicle when the person arrives at the conflict point, and determine the risk area and danger area of ​​the vehicle at the conflict point based on the vehicle's driving direction;

[0184] X42. Determine whether the vehicle is located in the risk area;

[0185] If the pedestrian is not in the risk area, the motion state calculation continues based on the social force model;

[0186] If the pedestrian is in the risk area, it is further determined whether the pedestrian is in the danger area or the risk area;

[0187] If the pedestrian is in the risk area and not in the dangerous area, the time difference between the pedestrian and the vehicle reaching the conflict point is calculated; if the time difference does not exceed the set threshold, the output decision is to slowly accelerate in the current direction; if the time difference exceeds the threshold, the output enters a hesitant state;

[0188] If the pedestrian is in a dangerous area, a judgment is made based on the interaction angle; when the interaction angle is a lateral interaction, the output decision is to run and accelerate in the current direction; when the interaction angle is a forward and backward interaction, the output decision is a sharp turn.

[0189] The dynamic speed adjustment mechanism in the hesitant state is used when the pedestrian cannot clearly judge whether he can safely pass the conflict point. When the pedestrian clearly judges whether he can safely pass the conflict point, he will temporarily maintain the current speed and wait until the distance with the vehicle is closer before making a judgment.

[0190] In the above-mentioned slow acceleration, the acceleration is less than the set upper limit value; in the running acceleration, the acceleration is greater than the set upper limit value.

[0191] Any possible implementation described above further provides an implementation, in which the specific method of step S3 includes:

[0192] S31, inputting initial basic data of pedestrians and vehicles, wherein the basic data includes initial positions, speeds, time steps, and destinations of pedestrians and vehicles;

[0193] S32, calculating the target position of the pedestrian at the current time step, wherein the target position is determined according to the planned path between the actual position at the current time step and the destination;

[0194] S33, inputting the pedestrian's current velocity vector, current position and target position into the improved social force model to calculate the resultant force on the pedestrian;

[0195] S34, entering the decision model, and generating a decision output for the current time step based on the relative distance and speed between the pedestrian and the vehicle;

[0196] S35, adjusting the resultant force on the pedestrian according to the decision output obtained at the current time step;

[0197] S36, adjusting the speed of the pedestrian according to the resultant force adjusted in step S35, and calculating the actual position of the pedestrian in the next time step;

[0198] S37. Repeat steps S32-S36 until all time steps are completed.

[0199] The interactive simulation process is as follows Figure 2 shown.

[0200] In a specific embodiment, in step S32, the Astar algorithm is used to calculate the target position of the pedestrian at the next time step. According to the Astar algorithm, the final destination position is input, and the Astar algorithm plans an optimal path (shortest or lowest cost) for the pedestrian to reach the final destination from the current position. This path is composed of a series of discrete points, each point can be regarded as a node on the path, and the algorithm outputs the target position (the next node to move towards at the current moment).

[0201] In a specific embodiment, in step S35, the specific method for adjusting the resultant force on the pedestrian according to the decision output is:

[0202] When pedestrians choose to make a sharp turn, the improved social force model adjusts the social force to a force perpendicular to the direction of the vehicle according to the relative position of the pedestrian and the vehicle. The direction of the force depends on the position of the pedestrian relative to the vehicle and is determined according to the direction to be avoided.

[0203] When the pedestrian chooses to accelerate slowly or run, the social force is adjusted to the acceleration force along the current speed direction, and the pedestrian's walking speed and acceleration upper limit are adjusted to a higher running speed and acceleration;

[0204] When the pedestrian chooses to slow down or stop suddenly, the social force is adjusted to the acceleration force along the current reverse speed, and the pedestrian's acceleration is replaced by a higher emergency stop acceleration;

[0205] When the pedestrian leaves the danger zone but is still in the risk area, maintain the previous decision and continue running, slowing down or turning until leaving the risk area.

[0206] The higher running speeds and accelerations mentioned above are all set with minimum limits based on experiments or statistical data, and the higher emergency stop accelerations are also set with minimum limits based on experiments or statistical data.

[0207] An embodiment of the present invention provides a human-vehicle interaction simulation system based on an improved social force model, the system being used to implement the above method, and the system comprising:

[0208] The social force calculation module calculates the social forces on pedestrians based on the improved social force model;

[0209] A decision module is used to determine the behavior patterns of pedestrians and vehicles when they meet;

[0210] The interactive simulation module is configured to interactively simulate the movement of pedestrians and vehicles by using the social force calculation module and the decision-making module given the initial basic data of pedestrians and vehicles.

[0211] An embodiment of the present invention discloses an application of a human-vehicle interaction simulation method based on an improved social force model, wherein the method is applied to a scene where vehicles and pedestrians cross and mix.

[0212] In a specific embodiment, the method is applied to an intersection without a traffic light; the intersection includes a cross intersection and a multi-directional intersection.

[0213] The traditional social force model does not fully consider the impact of vehicle speed when dealing with pedestrian-vehicle interactions, especially when vehicles approach or pass by at high speed. The present invention improves the social force model and combines the dynamic changes of the resultant force on pedestrians with the environment to enhance the model's ability to describe the impact of vehicle speed. It can more accurately simulate the force state of pedestrians, especially in complex traffic scenarios with large speed differences between pedestrians and vehicles.

[0214] Optimization of decision-making model based on human-vehicle interaction: Traditional decision-making models usually ignore the complex interaction process between pedestrians and vehicles and the diversity of pedestrian decision-making behaviors during the interaction between pedestrians and vehicles, resulting in insufficient accuracy of the model when dealing with human-vehicle interaction scenarios. This invention generates reasonable risk avoidance decisions based on the relative speed, position and distance between pedestrians and vehicles by designing a special decision-making module, which significantly improves the safety and rationality of decision-making of pedestrians in complex traffic environments.

[0215] Although several embodiments of the present invention have been given herein, those skilled in the art should understand that the embodiments of the present invention may be modified without departing from the spirit of the present invention. The above embodiments are merely exemplary and should not be used as a limitation on the scope of the present invention.

Claims

1. A human-vehicle interaction simulation method based on an improved social force model, characterized in that: The method comprises: S1. Construct an improved social force model, which is used to simulate the influence of driving force, human-vehicle interaction force, pedestrian-to-pedestrian interaction force, and obstacle repulsion on pedestrian travel speed; the improved social force model optimizes the influence of speed, distance, and interaction angle between people and vehicles on pedestrian travel; S2. Constructing a decision model, wherein the decision model is used to determine the behavior pattern of pedestrians and vehicles when they meet; the decision model optimizes the influence of speed, distance and interaction angle between pedestrians and vehicles on pedestrian decision-making; S3. Given the initial basic data of pedestrians and vehicles, the improved social force model of step S1 and the decision model of step S2 are used to interactively simulate the movement of pedestrians and vehicles.

2. The human-vehicle interaction simulation method based on the improved social force model as claimed in claim 1, characterized in that: In step S1, the improved social force model is expressed as: ； （2）； ； ； ； ； ； ； ； ； （11）； （12）； Where: The driving force of the destination on pedestrians, For the interaction between pedestrians, is the repulsive force of obstacles on pedestrians, is the interaction force between pedestrians and vehicles, is a random factor, N1, N2, and N are the total number of vehicles, obstacles, and pedestrians, respectively; , is the mass and speed of pedestrians, , They represent the pedestrian’s expected speed and expected direction of movement, respectively. is the interval time; They are the psychological and physical forces of interaction between pedestrians; is the distance between the centroids of pedestrians, , are two pedestrian radii When two pedestrians approach each other, a psychological force is generated to avoid each other; the physical force is determined by Decide, is a piecewise function. When the distance between pedestrians is less than a certain distance, > , physical force is generated; when ≤ hour, , no physical force is generated; , is a parameter used to adjust the intensity of interaction between pedestrians. , are the elastic coefficient and sliding friction coefficient of the human body, is the direction of the force exerted by pedestrian j on target pedestrian i, is the speed difference between pedestrians in the tangential direction; They are the psychological force and physical force of the obstacles’ repulsive force on pedestrians, is the pedestrian radius, is the distance between the center of mass of the pedestrian and the obstacle. When the pedestrian and the obstacle are close to each other, a psychological force is generated to avoid getting close to each other. The physical force is determined by Determine that when the distance between pedestrians and obstacles is less than a certain distance, > , physical force is generated; when ≤ hour, , no physical force is generated; , It is a parameter used to adjust the intensity of the obstacle’s effect on pedestrians. is the direction of the force acting on the obstacle. is the unit vector tangential to the pedestrian and the obstacle; is the velocity vector of pedestrian i; , Parameters used to adjust the intensity of the vehicle's impact on pedestrians, is the distance between the pedestrian and the vehicle; the anisotropic characteristics describe the relationship between the interaction force and the angle between the vehicle driving direction and the pedestrian moving direction. is the parameter to adjust the anisotropic characteristics, Indicates the angle between the pedestrian's movement direction and the vehicle's travel direction, ranging from arrive between; is the unit vector of pedestrian i pointing to the vehicle; Used to assess whether the speed difference between vehicles and pedestrians exceeds a set threshold Parameters, is the speed difference between pedestrians and vehicles, u is the parameter for adjusting the speed difference affecting the force, is the vehicle speed.

3. The human-vehicle interaction simulation method based on the improved social force model as claimed in claim 2, characterized in that: The parameters in the improved social force model , , , , , , , u is optimized by Bayesian optimization method; the Bayesian optimization method is as follows: Z1, the Bayesian model is: Parameter Space ,in ={ , , , , , , , }, each Represents a parameter group used to perform an objective function evaluation; Objective function F( ) = Y: Y is the input parameter group The output target value after the simulation is obtained, and the output target value Y is the average displacement error of the pedestrians; the predicted trajectories of all pedestrians are obtained after a certain simulation process, and the average displacement error of the pedestrians is calculated based on the predicted trajectory and the actual trajectory; the average displacement error The formula is: ； is the x-coordinate of the predicted position of pedestrian i at time t, is the y coordinate of the predicted position of pedestrian i at time t, is the x-coordinate of the real position of pedestrian i at time t, is the y coordinate of the real position of pedestrian i at time t, and N is the total number of pedestrians; Acquisition function S: used to select the optimal parameter set for the next evaluation ; ； Dataset D: contains all parameter sets that have been evaluated and the corresponding target value ; , n is the total number of evaluations; Z2, initialization: Randomly select K parameter groups for initial evaluation of the objective function to obtain the data set D, where K is the number of initialization samplings; ； Z3, iterative optimization: At each iteration, Bayesian optimization performs the following steps: Z31, Gaussian process modeling: Assume that the objective function is continuous in the parameter space and satisfies the Gaussian process; this means that the value of the objective function at any position in the parameter space can be described by a mean function and a variance function; specifically, the distribution form of the value of the objective function at a given position is expressed as: ； is the mean of all objective function values ​​that have been iterated so far, is the variance of all objective function values ​​that have been iterated; P() is the probability distribution, D t is the current data set, x t+1 is the parameter group for the next iteration; Z32, select the next parameter group: At each iteration, Bayesian optimization selects the next set of parameters to evaluate; through the acquisition function S, the parameters with the greatest potential improvement are selected : ； ； To iterate to the optimal target value of the current step, is the objective function value of the parameter group; Z33, Update the data set: The selected parameter points are passed through the objective function to calculate the objective function value, and the result is added to the existing training data; then, the updated data set is used to refit the Gaussian process model to make more accurate predictions in the next iteration: ； Z34, termination condition: reaching the maximum number of iterations or satisfying the convergence condition, and obtaining the optimal parameter set ; ={ 、 、 、 、 、 、 、 }。 4. The human-vehicle interaction simulation method based on the improved social force model as claimed in claim 1, characterized in that: In step S2, the decision-making method adopted by the decision model is specifically: X1. Calculate the future conflict point between pedestrians and vehicles X2. If there is a conflict point, calculate the time it takes for pedestrians and vehicles to reach the conflict point according to their current speeds and directions, and determine the order of arrival; X3. If the vehicle reaches the conflict point first, then: X31. Calculate the position of the pedestrian when the vehicle reaches the conflict point, and determine the risk area and the danger area of ​​the vehicle at the conflict point according to the driving direction of the vehicle; the risk area is an area where there is a potential conflict between the pedestrian and the vehicle, and the danger area is an area where the conflict between the pedestrian and the vehicle cannot be avoided; X32, determine whether the pedestrian's future position is within the risk area; If the pedestrian's future position is not within the risk area of ​​the vehicle, the pedestrian's motion state is calculated according to the improved social force model; If the pedestrian is in the risk area, it is further determined whether it is in the dangerous area; If the pedestrian is in the risk area and not in the danger area, the time difference between the pedestrian and the vehicle arriving at the conflict point is calculated; If the time difference does not exceed the set threshold, the decision output is to slow down in the current direction; If the time difference exceeds the threshold, the decision output is to enter hesitation; If the pedestrian is in the danger zone, the interaction angle between the pedestrian and the vehicle is further determined; when the interaction angle is lateral interaction, the decision output is an emergency stop in the current direction; when the interaction angle is front-to-back interaction, the decision output is a sharp turn; X4. If the pedestrian reaches the conflict point first, then: X41. Calculate the position of the vehicle when the person arrives at the conflict point, and determine the risk area and danger area of ​​the vehicle at the conflict point based on the vehicle's driving direction; X42. Determine whether the vehicle is located in the risk area; If the pedestrian is not in the risk area, the motion state calculation continues based on the social force model; If the pedestrian is in the risk area, it is further determined whether the pedestrian is in the danger area or the risk area; If the pedestrian is in the risk area and not in the dangerous area, the time difference between the pedestrian and the vehicle reaching the conflict point is calculated; If the time difference does not exceed the set threshold, the output decision is to slowly accelerate in the current direction; If the time difference exceeds the threshold, the output enters a hesitant state; If the pedestrian is in a dangerous area, a judgment is made based on the interaction angle; when the interaction angle is a lateral interaction, the output decision is to run and accelerate in the current direction; when the interaction angle is a forward and backward interaction, the output decision is a sharp turn.

5. The human-vehicle interaction simulation method based on the improved social force model as claimed in claim 1, characterized in that: The specific method of step S3 includes: S31, inputting initial basic data of pedestrians and vehicles, wherein the basic data includes initial positions, speeds, time steps, and destinations of pedestrians and vehicles; S32, calculating the target position of the pedestrian at the current time step, wherein the target position is determined according to the planned path between the actual position at the current time step and the destination; S33, inputting the pedestrian's current velocity vector, current position and target position into the improved social force model to calculate the resultant force on the pedestrian; S34, entering the decision model, and generating a decision output for the current time step based on the relative distance and speed between the pedestrian and the vehicle; S35, adjusting the resultant force on the pedestrian according to the decision output obtained at the current time step; S36, adjusting the speed of the pedestrian according to the resultant force adjusted in step S35, and calculating the actual position of the pedestrian in the next time step; S37. Repeat steps S32-S36 until all time steps are completed.

6. The human-vehicle interaction simulation method based on the improved social force model as claimed in claim 5, characterized in that: In step S35, the specific method for adjusting the resultant force on the pedestrian according to the decision output is: When pedestrians choose to make a sharp turn, the improved social force model adjusts the social force to a force perpendicular to the direction of the vehicle according to the relative position of the pedestrian and the vehicle. The direction of the force depends on the position of the pedestrian relative to the vehicle and is determined according to the direction to be avoided. When the pedestrian chooses to accelerate slowly or run, the social force is adjusted to the acceleration force along the current speed direction, and the pedestrian's walking speed and acceleration upper limit are adjusted to a higher running speed and acceleration; When the pedestrian chooses to slow down or stop suddenly, the social force is adjusted to the acceleration force along the current reverse speed, and the pedestrian's acceleration is replaced by a higher emergency stop acceleration; When the pedestrian leaves the danger zone but is still in the risk area, maintain the previous decision and continue running, slowing down or turning until leaving the risk area.

7. The human-vehicle interaction simulation method based on the improved social force model as claimed in claim 6, characterized in that: In step S32, the target position of the pedestrian is calculated using the Astar algorithm.

8. A human-vehicle interaction simulation system based on an improved social force model, characterized in that: The system is used to implement the method according to any one of claims 1 to 7, and the system includes: The social force calculation module calculates the social forces on pedestrians based on the improved social force model; A decision module is used to determine the behavior patterns of pedestrians and vehicles when they meet; The interactive simulation module is configured to interactively simulate the movement of pedestrians and vehicles by using the social force calculation module and the decision-making module given the initial basic data of pedestrians and vehicles.

9. An application of a human-vehicle interaction simulation method based on an improved social force model, characterized in that: The method as described in any one of claims 1 to 7 is applied to a scene where vehicles and pedestrians are crossing and mixing.

10. The application of the human-vehicle interaction simulation method based on the improved social force model as claimed in claim 9, characterized in that: The method is applied to an intersection without a traffic light; the intersection includes a cross intersection and a multi-directional intersection.

Images