A risk assessment method for human-vehicle interactive game based on risk perception

Abstract

The present invention discloses a risk assessment method for human-vehicle interaction game based on risk perception, extracting the interaction trajectory of pedestrians and vehicles from the recorded video of the non-signalized intersection area; determining the risk perception influencing factors of pedestrians and vehicles in the process of human-vehicle interaction; constructing a human-vehicle risk perception model through a long short-term memory network to determine the risk perception value of pedestrians and vehicles; according to the human-vehicle risk perception model, respectively determining the risk perception value range of pedestrians and vehicles under the conditions of going first and giving way as the acceptable risk perception level of pedestrians and vehicles; using the game model, analyzing the relationship between the risk perception value of pedestrians and vehicles, the risk perception level of pedestrians and vehicles, and the motion state of pedestrians and vehicles during the game time period, and constructing a human-vehicle interaction behavior model. The human-vehicle interaction simulation framework thus established shows high accuracy and is of great value for characterizing the interaction behavior of pedestrians and vehicles.

Description

Technical Field

[0001] The present invention relates to the technical field of road traffic safety, and in particular to a risk assessment method for human-vehicle interactive game based on risk perception. Background Art

[0002] In recent years, urban walking has attracted much attention as a sustainable mode of transportation, which not only helps to alleviate traffic congestion but also improves physical health. However, compared with other modes of transportation, pedestrians are more likely to suffer serious injuries or deaths on the road, especially at unsignalized intersections. About 70% of traffic accidents involving pedestrians and vehicles occur at crosswalks. Therefore, analyzing the interaction between pedestrians and vehicles at unsignalized intersections is crucial to improving long-term road safety.

[0003] The occurrence and severity of pedestrian-vehicle conflicts depend on the behavior of pedestrians and drivers. Analyzing the information of pedestrians and drivers, especially risk perception, is a key issue in understanding pedestrian crossing behavior and driver driving behavior. In the dynamic interaction of the pedestrian-vehicle environment system, the risk perception level of pedestrians and drivers changes dynamically. Once the perceived risk exceeds the acceptable range (i.e., the acceptable risk perception level), the driver will accelerate or decelerate accordingly to adjust his perceived risk. At the same time, pedestrians may switch between sudden stops and fast walks, which is difficult for surrounding vehicles to predict, resulting in collisions. Therefore, analyzing the interaction behavior between pedestrians and vehicles at unsignalized intersections, establishing risk perception models for pedestrians and vehicles, and determining the acceptable risk perception level are crucial for analyzing the human-vehicle game. Summary of the invention

[0004] The purpose of the present invention is to provide a risk assessment method based on risk-perceived human-vehicle interactive game. The present invention introduces a human-vehicle interactive framework and incorporates the human-vehicle game considering risk perception to predict the decisions made by pedestrians and vehicles when crossing the road. By drawing on cognitive psychology theory and applying the long short-term memory (LSTM) network, a quantitative model of risk perception of pedestrians and vehicles during the interaction process is established. Based on game theory, the benefits of pedestrians and vehicles are determined by their risk perception level and motion state. The human-vehicle interactive simulation framework thus established shows high accuracy and is of great value for characterizing the interactive behavior of pedestrians and vehicles.

[0005] To achieve the above objectives, this application adopts the following scheme:

[0006] On the one hand, the present invention provides a risk assessment method for human-vehicle interactive game based on risk perception, which specifically includes the following steps:

[0007] S1. Obtain recorded video of an unsignalized intersection area, and extract interaction trajectories of pedestrians and vehicles from the recorded video;

[0008] S2. Determine the influencing factors of pedestrian risk perception and vehicle risk perception in the process of human-vehicle interaction according to the interaction trajectories of pedestrians and vehicles;

[0009] S3. Based on the factors affecting pedestrian risk perception and vehicle risk perception, a human-vehicle risk perception model is constructed through a long short-term memory network to determine the risk perception values ​​of pedestrians and vehicles;

[0010] S4. Determine the acceptable risk perception level of pedestrians and vehicles based on cognitive psychology risk compensation theory and risk homeostasis theory;

[0011] S5. Construct a human-vehicle interaction game model, analyze the risk perception values ​​of pedestrians and vehicles, the risk perception levels of pedestrians and vehicles, and the relationship between the movement states of pedestrians and vehicles during the game period, and finally obtain a human-vehicle interaction behavior model for evaluating the risks of human-vehicle interaction games.

[0012] In some specific implementations, factors affecting pedestrian risk perception include the speed of the vehicle. , the relative distance between pedestrians and vehicles , the time when the vehicle arrives at the conflict point , the ratio of the distance from vehicles to the conflict point to the distance from pedestrians to the conflict point ;

[0013] Factors influencing vehicle risk perception include pedestrian speed , vehicle speed , the distance from the pedestrian to the conflict point The time required for pedestrians to reach the conflict point .

[0014] In some specific implementations, the human-vehicle interaction game model includes:

[0015] The set of participants in constructing the human-vehicle interaction game model is P =( p , v ), p represents pedestrians, v represents vehicles, and the game time period is defined as: [ t a , t e ], divide the time period into n time steps, and define the strategy set of pedestrians at each time step as , the vehicle’s strategy set is , and They represent the acceleration strategies of pedestrians and vehicles at each time step respectively.

[0016] In some specific implementation schemes, the acceleration strategy for pedestrians and vehicles at each time step includes setting a corresponding number of acceleration values ​​according to preset increments within a preset acceleration range.

[0017] In some specific implementation schemes, the gaming process specifically performs the following process:

[0018] S51, obtaining the acceleration strategy of pedestrians and the acceleration strategy of vehicles at the current time step, combining them to obtain several strategy combinations, obtaining the corresponding pedestrian and vehicle risk perception influencing factors and motion state information at the current time step, and using the established pedestrian and vehicle risk perception model to calculate the speed change and corresponding risk perception value of pedestrians and vehicles at each strategy combination;

[0019] S52, according to the speed change of pedestrians and vehicles under each strategy combination and the corresponding risk perception value at the current time step, using the benefit function of the constructed human-vehicle interaction game model, calculate the benefit value of pedestrians and vehicles under each strategy combination at the current time step;

[0020] S53, according to the benefit values ​​of pedestrians and vehicles in each strategy combination at the current time step, using Nash equilibrium solution to determine the Nash equilibrium strategy of pedestrians and vehicles at the current time step; determine the strategy combination of pedestrians and vehicles according to the Nash equilibrium strategy, and calculate the motion state information of pedestrians and vehicles at the next time step according to the determined strategy combination;

[0021] S54. Repeat steps S51-S53 until all time steps are calculated.

[0022] In some specific implementation schemes, the calculation method for calculating the motion state information of the next time step in step S53 is:

[0023] The calculation method for calculating the motion state information of the next time step in step S53 is:

[0024]

[0025] in, j = { p , v} ; i =1, 2, ..., n -1, is at the time step i length of time, represents the speed of participant j at the current time step i, represents the acceleration value corresponding to participant j in the strategy combination determined at time step i, represents the distance from participant j to the conflict point at the current time step i, represents the speed of participant j at the next time step i+1, represents the distance from participant j to the conflict point at the next time step i+1, Represents the speed change of participant j at the next time step i+1.

[0026] In some specific implementation schemes, the method for determining the risk perception influencing factors of the downline at time step i+1 is:

[0027]

[0028] The method for calculating the risk perception influencing factors of the vehicle at time step i+1 is:

[0029]

[0030] in, , , They represent the speed, acceleration and risk perception value of the pedestrian at time step i, respectively. , , They represent the speed, acceleration and risk perception value of the vehicle at time step i, respectively. and Respectively represent the speed of pedestrians and vehicles at the beginning of time step i+1, and Respectively represent the distances from pedestrians and vehicles to the conflict point at time step i+1, and They represent the time from pedestrian to vehicle to the conflict point at time step i+1, represents the relative distance between pedestrians and vehicles at time step i+1, and Represent the speed changes of pedestrians and vehicles at time step i+1, respectively. , ), ( , )and( , ) are pedestrians, vehicles, and conflict points. i+1 The coordinates of time.

[0031] In some specific implementations, the benefit function for calculating the benefit value of pedestrians and vehicles is:

[0032]

[0033] in, j = { p , v} ; represents the normalized change in velocity of participant j at time step i, represents the risk perception value of participant j at time step i Normalization of represents the risk perception value of participant j at time step i, represents the expected coefficient of the speed change level of participant j, represents the expected coefficient of risk perception of participant j, , , .

[0034] In some embodiments, in the payoff function, the participant's expected coefficient for the level of speed change is It is determined by the risk perception value of the participant at the current time step and the acceptable risk perception level of the participant. The method is:

[0035] ,

[0036] in, represents the risk perception value of participant j at time step i, j = { p , v} , [ α j ，β j ] represents the acceptable risk perception level of participant j.

[0037] In some specific implementation schemes, the specific process of step S4 is:

[0038] According to the risk perception model of pedestrians and vehicles, the risk perception value range C of pedestrians under the leading condition and the risk perception value range D of pedestrians under the yielding condition are determined respectively, and the intersection of C and D is solved, and the intersection is taken as the acceptable risk perception level of pedestrians;

[0039] According to the human-vehicle risk perception model, the risk perception value range C1 of the vehicle under the leading condition and the risk perception value range D1 under the yielding condition are determined respectively, and the intersection of C1 and D1 is solved, and the intersection is taken as the acceptable risk perception level of the vehicle.

[0040] The present invention has the beneficial effects:

[0041] The main purpose of the present invention is to establish a human-vehicle interaction game behavior model that takes into account risk perception, which is of great value for characterizing the interactive behavior of pedestrians and vehicles and provides reference and reference for traffic safety assessment. This application introduces a human-vehicle interaction framework to incorporate the human-vehicle game that takes into account risk perception to predict the decisions made by pedestrians and vehicles when crossing the road. By drawing on cognitive psychology theory and applying the long short-term memory (LSTM) network, a quantitative model of risk perception of pedestrians and vehicles during the interaction process was established. Game theory was used to analyze the quantitative value of risk perception, the acceptable risk perception level, and the relationship between the motion state of pedestrians and vehicles, and a human-vehicle interaction behavior model was established. The human-vehicle interaction simulation framework thus established shows high accuracy and is of great value for characterizing the interactive behavior of pedestrians and vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flow chart of a method for constructing a human-vehicle interactive game model based on risk perception provided by an embodiment of the present invention;

[0043] Figure 2 A schematic diagram of the road geometry structure of the data collection point provided by an embodiment of the present invention;

[0044] Figure 3 A schematic diagram of trajectory information of pedestrians and vehicles provided in an embodiment of the present invention;

[0045] Figure 4 A flowchart of a human-vehicle game considering risk perception provided by an embodiment of the present invention;

[0046] Figure 5 A schematic diagram of actual and predicted output risk perception values ​​of pedestrians in a specific implementation scheme of the present invention;

[0047] Figure 6 A schematic diagram of actual and predicted output risk perception values ​​of a vehicle in a specific embodiment of the present invention;

[0048] Figure 7 A distribution diagram of pedestrian risk perception values ​​in a specific implementation scheme of the present invention;

[0049] Figure 8 A distribution diagram of vehicle risk perception values ​​in a specific implementation scheme of the present invention;

[0050] Fig. 9 It is a schematic diagram of the probability density function of pedestrian acceleration and the probability density function of vehicle acceleration in a specific implementation scheme of the present invention;

[0051] Fig.10 A schematic diagram of the speed of pedestrians and vehicles at each time step in a specific implementation scheme of the present invention;

[0052] Fig.11This is a schematic diagram of the distances from pedestrians and vehicles to the conflict point at each time in a specific implementation manner of the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] The relative arrangement of components and steps, the numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention unless specifically stated otherwise.

[0055] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0056] Additionally, descriptions of well-known structures, functions, and configurations may be omitted for clarity and conciseness.One of ordinary skill in the art will recognize that various changes and modifications may be made to the examples described herein without departing from the spirit and scope of the present disclosure.

[0057] Technologies, methods, and apparatus known to ordinary technicians in the relevant field may not be discussed in detail, but where appropriate, such technologies, methods, and apparatus should be considered part of the authorization specification.

[0058] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0059] Example 1

[0060] like Figure 1 As shown, this embodiment provides a risk assessment method for human-vehicle interactive game based on risk perception, which specifically includes the following steps:

[0061] S1. Data collection and processing: Obtain recorded videos of the intersection area without signal control, and extract the interaction trajectories of pedestrians and vehicles from the recorded videos;

[0062] Unsignalized intersections are frequent sites of human-vehicle interaction, posing a high risk to pedestrians. The dynamic interaction between pedestrians and vehicles results in a constantly changing risk perception on both sides. Therefore, analyzing the association between risk-aware pedestrian and vehicle behavior at unsignalized intersections is critical to improving road safety. Unsignalized intersection areas are selected for video recording by drones, and then the trajectories of pedestrians and vehicles can be extracted using the semi-automated software PeTrack.

[0063] S2. Determine the influencing factors of pedestrian risk perception and vehicle risk perception in the process of human-vehicle interaction according to the interaction trajectories of pedestrians and vehicles;

[0064] By analyzing the trajectories of pedestrians and vehicles, the risk perception of pedestrians and vehicles is quantified, and the important factors affecting risk perception in the process of human-vehicle interaction are identified. Specifically, for pedestrians, the factors affecting pedestrian risk perception include the speed of the vehicle. , the relative distance between pedestrians and vehicles , the time when the vehicle arrives at the conflict point , The ratio of the distance from the vehicle to the conflict point to the distance from the pedestrian to the conflict point ;

[0065] Factors influencing vehicle risk perception include pedestrian speed , vehicle speed , the distance from the pedestrian to the conflict point The time required for pedestrians to reach the conflict point .

[0066] S3. Based on the factors affecting pedestrian risk perception and vehicle risk perception, a human-vehicle risk perception model is constructed through a long short-term memory network to determine the risk perception values ​​of pedestrians and vehicles;

[0067] S4. Determine the acceptable risk perception level of pedestrians and vehicles based on cognitive psychology risk compensation theory and risk homeostasis theory;

[0068] According to the risk perception model of pedestrians and vehicles, the risk perception value range C of pedestrians under the leading condition and the risk perception value range D of pedestrians under the yielding condition are determined respectively, and the intersection of C and D is solved, and the intersection is taken as the acceptable risk perception level of pedestrians;

[0069] According to the human-vehicle risk perception model, the risk perception value range C1 of the vehicle under the leading condition and the risk perception value range D1 under the yielding condition are determined respectively, and the intersection of C1 and D1 is solved, and the intersection is taken as the acceptable risk perception level of the vehicle.

[0070] By applying cognitive psychology theory and long short-term memory (LSTM) network, a quantitative model of risk perception of pedestrians and vehicles in the interaction process is established to determine the acceptable risk perception level of pedestrians and vehicles. When the acceptable risk perception level of pedestrians or vehicles is conceptualized as [α, β] range, when the perceived risk value of pedestrians or vehicles is lower than α When the risk perception value exceeds β , pedestrians or vehicles will slow down. When the risk perception value is [ α , β ], pedestrians or vehicles will accelerate, decelerate or keep a constant speed. After determining the acceptable risk perception level, the game theory model is used to solve the acceleration strategy at each moment. By continuously iterating and updating the acceleration strategy at each moment, it can be determined whether pedestrians or vehicles go first or give way in the end. The specific process is as follows:

[0071] S5. Construct a human-vehicle interaction game model, analyze the risk perception values ​​of pedestrians and vehicles, the risk perception levels of pedestrians and vehicles, and the relationship between the movement states of pedestrians and vehicles during the game period, and finally obtain a human-vehicle interaction behavior model for evaluating the risks of human-vehicle interaction games.

[0072] Specifically, in the game model, the definitions of factors such as participants, strategies, profit functions, and game solutions include:

[0073] (1) Participants and Strategies

[0074] According to game theory, the set of participants in constructing the game model is P =( p , v ), p represents pedestrians, v represents vehicles, pedestrians and vehicles can adopt different strategies, such as acceleration, constant speed or deceleration. Assume that pedestrians and vehicles are at time t a Start the game decision of both parties and at time t e Completed, the game time period is defined as: [ t a , t e ], the time period is divided into n time steps, and the length of each time step is ∆ t， Define the strategy set of pedestrians at each time step as , the vehicle’s strategy set is , and They represent the acceleration strategies of pedestrians and vehicles at each time step respectively.

[0075] Here, the acceleration strategy for each time step is preset, and the acceleration strategy for pedestrians and vehicles in each time step includes setting corresponding acceleration values ​​according to preset increments within the preset acceleration range. For example, assuming that the acceleration range of pedestrians is -3 to 3m / s 2 , assuming the increment is 0.2 m / s², the acceleration range of the vehicle is -1.5 m / s² to 1.5 m / s², and the increment is 0.1 m / s², then pedestrians and vehicles can get 31 acceleration values ​​at each time step, resulting in 31*31 strategy combinations.

[0076] (2) Profit function

[0077] Assume that at time step i At the beginning, the speeds of pedestrians and vehicles are and , their accelerations (or decelerations) are and , and the distances to the conflict point are and , the risk perception values ​​of pedestrians and vehicles are and . It is also assumed that all pedestrians and vehicles will be i Continuous movement with different accelerations eventually reaches the collision point.

[0078] Here, and represent the risk perception values ​​of pedestrians and vehicles at time step i, respectively, which can be obtained according to the risk perception model of pedestrians and vehicles established in step S3. i +1, the speed of pedestrians and vehicles at the start ( and ), the distance from pedestrians and vehicles to the conflict point ( and ), and the speed changes of pedestrians and vehicles ( and ) can be calculated by the following formula:

[0079] (1)

[0080] in j = p , v ; i =1, 2, ..., n -1. is at the time step i length of time.

[0081] At time step i +1, pedestrian's risk perception value ( ) can use the parameter and Determine and calculate through the established pedestrian risk perception model, the formula is as follows:

[0082] (2)

[0083] in, They are pedestrians, vehicles, and conflict points. i+ 1 The time step corresponds to the coordinates of the instant.

[0084] Likewise, at time step i +1, the vehicle's risk perception value ( ) can use the parameter and , and is determined by the vehicle risk perception model, and the calculation formula is as follows:

[0085] (3)

[0086] in, and Respectively represent the speed of pedestrians and vehicles at the beginning of time step i+1, and Respectively represent the distances from pedestrians and vehicles to the conflict point at time step i+1, and They represent the time from pedestrian to vehicle to the conflict point at time step i+1, represents the relative distance between pedestrians and vehicles at time step i+1; under each strategy combination, , Substitute the corresponding acceleration values ​​for calculation one by one.

[0087] Based on the above factors, the pedestrian and vehicle benefit functions are as follows:

[0088] (4)

[0089] in j = p , v ; i =1, 2, ..., n . Respectively Normalization of represents the risk perception value of participant j at time step i, represents the expected coefficient of the speed change level of participant j, represents the expected coefficient of risk perception of participant j, , , .

[0033] In the benefit function, the participant's expected coefficient of the speed change level is determined by the participant's risk perception value at the current time step and the participant's acceptable risk perception level. If the risk perception value of pedestrians and vehicles is lower than the acceptable level, priority is given to improving their efficiency. On the contrary, if the perceived risk value exceeds its acceptable level, priority is given to improving safety. However, if the risk perception value of pedestrians and vehicles is within an acceptable range, it indicates that safety and efficiency are equally important to the benefits of pedestrians and vehicles. Therefore, the coefficient of the speed change level can be derived based on the acceptable risk perception level of pedestrians and vehicles:

[0090]

[0091] in, represents the risk perception value of participant j at time step i, j = { p , v} , [ α j ，β j ] represents the acceptable risk perception level of participant j, α j represents the lower limit of the acceptable risk level for participant j, β j It represents the upper limit of the acceptable risk level for participant j.

[0092] (3) Game solution

[0093] The model is solved based on Nash equilibrium. In human-vehicle interaction, there are multiple strategies to choose from. Here, it is assumed that the equilibrium point is selected probabilistically.

[0094] Specifically, the specific process of the game is:

[0095] S51, obtaining the acceleration strategy of pedestrians and the acceleration strategy of vehicles at the current time step, combining them to obtain several strategy combinations, obtaining the corresponding pedestrian and vehicle risk perception influencing factors and motion state information at the current time step, and using the established pedestrian and vehicle risk perception model to calculate the speed change and corresponding risk perception value of pedestrians and vehicles at each strategy combination;

[0096] S52, according to the speed change of pedestrians and vehicles under each strategy combination and the corresponding risk perception value at the current time step, using the benefit function of the constructed human-vehicle interaction game model, calculate the benefit value of pedestrians and vehicles under each strategy combination at the current time step;

[0097] S53, according to the benefit values ​​of pedestrians and vehicles in each strategy combination at the current time step, using Nash equilibrium solution to determine the Nash equilibrium strategy of pedestrians and vehicles at the current time step; determine the strategy combination of pedestrians and vehicles according to the Nash equilibrium strategy, and calculate the motion state information of pedestrians and vehicles at the next time step according to the determined strategy combination;

[0098] S54. Repeat steps S51-S53 until all time steps are calculated.

[0099] S6. Visualization and analysis of experimental results,display the simulation results. By comparing the speed and trajectory information,calculated by the model with the actual values, the model effect is evaluated,using the mean absolute error (MAE) and root mean square error (RMSE) indicators.

[0100] It is understandable that the application scenario of this embodiment is an intersection without signal guidance. Since intersections without signal guidance and control are frequent places where human-vehicle interactions occur, they pose a higher risk to pedestrians. The dynamic interaction between pedestrians and vehicles causes the risk perception of both parties to change constantly. Therefore, it is crucial to analyze the relationship between pedestrian and vehicle behaviors that consider risk perception at intersections without signal control to improve road safety. This embodiment establishes a quantitative model of risk perception of pedestrians and vehicles during the interaction process by applying cognitive psychology theory and long short-term memory (LSTM) networks, and then determines the acceptable risk perception level of pedestrians and vehicles. Secondly, game theory is used to analyze the quantitative value of risk perception, the acceptable risk perception level, and the relationship between the motion state of pedestrians and vehicles, and a human-vehicle interaction behavior model is established. Finally, the model effect is verified through the measured trajectory data of human-vehicle interaction.

[0101] In order to better demonstrate the above method, this embodiment provides a specific implementation method. In order to ensure the accuracy of the model and the authenticity of the data, a DJI Mini 3 Pro drone is used to Figure 2 Video recording and acquisition are performed at the intersection shown in the figure, wherein the road geometry of the intersection has symmetrical structural dimensions in the north-south and east-west directions. The unsignalized intersection is located near schools, hotels, commercial centers, and residential areas, and a large number of pedestrians and vehicles often cross the intersection, which facilitates the data collection work of this embodiment. The DJI Mini 3 Pro drone used in this embodiment supports 4K (3840×2160 pixels) resolution and a frame rate of 30 frames per second. In order to fully cover the relevant intersection area, the aircraft hovers at a predetermined position and maintains an altitude of 100 meters during each recording. In the end, a total of 214 interactions between pedestrians in the north-south direction and vehicles in the east-west direction were extracted. The trajectories of pedestrians and vehicles were further extracted using the semi-automated software PeTrack, as shown below. Figure 3 The track information is shown.

[0102] Secondly, the risk perception values ​​of pedestrians and vehicles are quantified. Based on the selected model parameters, the LSTM model is used to determine the risk perception values ​​of pedestrians and vehicles. In the LSTM network model, 171 sets of human-vehicle interaction data (accounting for 80% of all data) are designated as the training set, while 43 sets of human-vehicle interaction data (accounting for 20% of all data) are used for the test set. Figure 5-Figure 6 The actual and predicted risk perception values ​​of pedestrians and vehicles are displayed. It can be seen that the model results can effectively reflect the changes in risk perception of pedestrians and vehicles, and the model error is small (pedestrians: MAE= 0.09, RMSE= 0.18; vehicles: MAE=0.11, RMSE= 0.21). This shows that the model meets actual needs.

[0103] Furthermore, the determined acceptable risk perception level is closely related to the risk perception ability of pedestrians and vehicles. Through the established quantitative method, the risk perception range of pedestrians and vehicles under the go-first and go-by conditions can be derived. Subsequently, the acceptable risk level can be determined based on these two ranges. Assuming that the overlapping area of ​​the two ranges is [A, B], when the perceived risk value falls within the range of [A, B], pedestrians and vehicles can choose to give way or go first. Therefore, in the present invention, [A, B] is regarded as the acceptable risk level [ α , β ]. Through quantitative methods, it is determined that the range of pedestrian risk perception under the prior condition is [0.47,1.23] (e.g. Figure 7 ). In contrast, the risk perception range under the yield condition is [-0.18, 0.58] (e.g. Figure 7 Red column). Therefore, the overlap range of the two is [0.47, 0.58], which is determined to be the acceptable risk perception level for pedestrians. Similarly, for vehicles, the risk perception ranges under the first-pass and yield conditions are [-0.08, 0.78] and [0.33, 1.13], respectively (e.g. Figure 8 ). Therefore, the overlapping range of [0.33, 0.78] is determined as the acceptable risk perception level of the vehicle.

[0104] Next, based on game theory, a human-vehicle interaction game behavior model considering risk perception is established. The overall game process is as follows: Figure 4 The specific implementation process of each time step is as follows:

[0105] Step 1: Initialization provides the initial positions and velocities of pedestrians and vehicles in the first time step. Using the established motion state and risk perception model of pedestrians and vehicles, the speed changes and risk perception values ​​of the two participants under different strategy combinations are calculated.

[0106] Step 2: Apply the profit function in formula (4) to calculate the profit values ​​of pedestrians and vehicles under different strategy combinations based on the current participant status.

[0107] Step 3: Use Nash equilibrium solution to determine the Nash equilibrium strategy of pedestrians and vehicles in the first time step. Then, establish the behavior strategy (strategy combination) of both participants. Calculate the motion state information of both participants in the second time step based on the determined behavior strategy.

[0108] Step 4: According to the motion state and the established risk perception model, calculate the speed change and risk perception value of the two participants under different strategy combinations at the second time step. Repeat steps 2 and 3 to obtain the acceleration strategy of the two participants at the second time step and the motion state information at the third time step.

[0109] Step 5: Continue by calculating time step 3 to time step n The behavioral strategies (strategy combinations) and movement state information of both players until the end of the simulation (when one of the parties crosses the conflict point).

[0110] Finally, the experimental results are visualized and analyzed. For a deeper understanding, the pedestrians and vehicles in a specific time period are analyzed in detail. t a , t e ] game behavior. Here, [ t a , t e ] represents the 3-second duration before one of the parties (pedestrian or vehicle) reaches the conflict point. This time period is divided into 15 time steps, each lasting 0.2 seconds. The acceleration of pedestrians and vehicles is analyzed based on the data collected on site, and the acceleration probability density function is as follows: Fig. 9 As shown in Figure 2. Among the acceleration strategies for pedestrians and vehicles, it is assumed that the acceleration range of pedestrians is -3 m / s² to 3 m / s² with an increment of 0.2 m / s², and the acceleration range of vehicles is -1.5 m / s² to 1.5 m / s² with an increment of 0.1 m / s². This results in 31 acceleration strategies for pedestrians and vehicles respectively.

[0111] Taking a specific case as an example, the initial positions and motion states of pedestrians and vehicles at the beginning of the first time step are as follows: =57.03m, =7.00m, = 1.20m / s, =-1.03m, = 2.23m, = 11.39 m / s, =57.26m, = 3.42m. As described above, the acceptable risk perception level of pedestrians is [0.47, 0.58]. The acceptable risk perception level of vehicles is [0.33, 0.78].

[0112] By calculating the Nash equilibrium strategy, the present invention determines the optimal strategy of pedestrians and vehicles at each time step, and calculates the motion state information of the participants at the beginning of the next time step based on the estimated strategy. This process is repeated iteratively to obtain the Nash equilibrium strategy for each time step. The specific Nash equilibrium strategy is shown in Table 1.

[0113] Table 1

[0114]

[0115] Based on the estimated policy, the speed of pedestrians and vehicles and the distance to the conflict point are calculated at each time step. Figure 10-11 The comparison between the simulation results and the actual results is shown. Fig.10 In the actual interaction scenario, the vehicle is decelerating constantly while the pedestrian is keeping a constant speed. The results show that the simulated speed of the vehicle is almost consistent with the actual speed. However, after the sixth time step, the error between the simulated speed and the actual speed of the pedestrian gradually increases. In addition, the distance between the pedestrian and the vehicle to the conflict point at each time step is calculated. Fig.11 As shown in Figure 2, the estimated distances of pedestrians and vehicles to the conflict point accurately reflect the actual distances. In addition, the distances of pedestrians and vehicles to the conflict point can be used to determine which participant arrives at the conflict point first at the last time step. In the last time step, the participant closer to the conflict point is expected to arrive first. Fig.11 In , the pedestrian is estimated to be closer to the conflict point than the vehicle in the last time step. Therefore, it is expected that the pedestrian will make a proactive decision and pass the conflict point first. This is consistent with the actual observed data.

[0116] At the same time, the present invention simulated an additional 43 cases (test set, i.e., 20% of all vehicle-vehicle interaction data). Table 2 shows the simulation errors of the speed and the distances of both parties to the conflict point for all selected test set cases. As shown in Table 2, the error between the simulation results and the actual results is acceptable, indicating that the established model is effective in describing the human-vehicle interaction game behavior. When comparing the estimated and actual go-ahead / give-way decisions in all 43 cases, 40 of them were accurately estimated, and the model prediction accuracy reached 93.02%.

[0117] Table 2

[0118]

[0119] It can be seen that the advantages of this application are:

[0120] (1) Improved model accuracy: By introducing risk perception factors, the present invention takes into account more dynamic factors in the modeling of pedestrian-vehicle interaction behavior, effectively improving the accuracy of modeling the interaction between pedestrians and vehicles at unsignalized intersections. The use of a quantitative model based on long short-term memory (LSTM) can more effectively capture the dynamic changes in risk perception of pedestrians and vehicles during the interaction process.

[0121] (2) Game strategy disclosure: Based on game theory, this paper defines the payoff function of pedestrians and vehicles and reveals the game strategies of both parties under the risk perception level and speed change level. This helps to have a deeper understanding of the decision-making mechanism of both parties in the process of human-vehicle interaction and provides a theoretical basis for traffic management at intersections.

[0122] (3) Verification of consistency with actual scenarios: By verifying the consistency between simulation results and actual scenarios, the present invention not only provides a method for accurately describing human-vehicle interaction behaviors, but also provides a reliable reference for traffic management and road safety at unsignalized intersections. This means that the present invention has the feasibility and reliability of practical application.

[0123] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. According to the technical essence of the present invention, within the spirit and principles of the present invention, any simple modification, equivalent replacement and improvement made to the above embodiment still falls within the protection scope of the technical solution of the present invention.

Claims

1. A risk assessment method for human-vehicle interactive game based on risk perception, characterized in that: The specific steps include: S1. Obtain recorded video of an unsignalized intersection area, and extract interaction trajectories of pedestrians and vehicles from the recorded video; S2. Determine the influencing factors of pedestrian risk perception and vehicle risk perception in the process of human-vehicle interaction according to the interaction trajectories of pedestrians and vehicles; S3. Based on the factors affecting pedestrian risk perception and vehicle risk perception, a human-vehicle risk perception model is constructed through a long short-term memory network to determine the risk perception values ​​of pedestrians and vehicles; S4. Determine the acceptable risk perception level of pedestrians and vehicles based on cognitive psychology risk compensation theory and risk homeostasis theory; S5. Construct a human-vehicle interaction game model, analyze the risk perception values ​​of pedestrians and vehicles, the risk perception levels of pedestrians and vehicles, and the relationship between the movement states of pedestrians and vehicles during the game time period, and finally obtain a human-vehicle interaction behavior model for evaluating the risk of human-vehicle interaction games; The human-vehicle interaction game model includes: The set of participants in constructing the human-vehicle interaction game model is P =( p , v ), p represents pedestrians, v represents vehicles, and the game time period is defined as: [ t a , t e ], divide the time period into n time steps, and define the strategy set of pedestrians at each time step as , the vehicle’s strategy set is , and Respectively represent the acceleration strategies of pedestrians and vehicles at each time step; The game process specifically implements the following process: S51, obtaining the acceleration strategy of pedestrians and the acceleration strategy of vehicles at the current time step, combining them to obtain several strategy combinations, obtaining the corresponding pedestrian and vehicle risk perception influencing factors and motion state information at the current time step, and using the established pedestrian and vehicle risk perception model to calculate the speed change and corresponding risk perception value of pedestrians and vehicles at each strategy combination; S52, according to the speed change of pedestrians and vehicles under each strategy combination and the corresponding risk perception value at the current time step, using the benefit function of the constructed human-vehicle interaction game model, calculate the benefit value of pedestrians and vehicles under each strategy combination at the current time step; S53, according to the benefit values ​​of pedestrians and vehicles in each strategy combination at the current time step, using Nash equilibrium solution to determine the Nash equilibrium strategy of pedestrians and vehicles at the current time step; determine the strategy combination of pedestrians and vehicles according to the Nash equilibrium strategy, and calculate the motion state information of pedestrians and vehicles at the next time step according to the determined strategy combination; S54. Repeat steps S51-S53 until all time steps are calculated.

2. The risk assessment method for human-vehicle interactive game based on risk perception according to claim 1 is characterized in that: Factors affecting pedestrian risk perception include vehicle speed. , the relative distance between pedestrians and vehicles , the time when the vehicle arrives at the conflict point , the ratio of the distance from vehicles to the conflict point to the distance from pedestrians to the conflict point ; Factors influencing vehicle risk perception include pedestrian speed , vehicle speed , the distance from the pedestrian to the conflict point The time required for pedestrians to reach the conflict point .

3. The risk assessment method for human-vehicle interactive game based on risk perception according to claim 1 is characterized in that: The acceleration strategy for pedestrians and vehicles at each time step includes setting corresponding acceleration values ​​according to preset increments within a preset acceleration range.

4. The risk assessment method for human-vehicle interactive game based on risk perception according to claim 1 is characterized in that: The calculation method for calculating the motion state information of the next time step in step S53 is: in, j = { p , v} ; i =1, 2, ..., n -1, is at the time step i length of time, represents the speed of participant j at the current time step i, represents the acceleration value corresponding to participant j in the strategy combination determined at time step i, represents the distance from participant j to the conflict point at the current time step i, represents the speed of participant j at the next time step i+1, represents the distance from participant j to the conflict point at the next time step i+1, Represents the speed change of participant j at the next time step i+1.

5. The risk assessment method for human-vehicle interactive game based on risk perception according to claim 4 is characterized in that: The method to determine the factors affecting the risk perception of the downline at time step i+1 is: The method for calculating the risk perception influencing factors of the vehicle at time step i+1 is: in, , , They represent the speed, acceleration and risk perception value of the pedestrian at time step i, respectively. , , They represent the speed, acceleration and risk perception value of the vehicle at time step i, respectively. and Respectively represent the speed of pedestrians and vehicles at the beginning of time step i+1, and Respectively represent the distances from pedestrians and vehicles to the conflict point at time step i+1, and They represent the time from pedestrian to vehicle to the conflict point at time step i+1, represents the relative distance between pedestrians and vehicles at time step i+1, and Represent the speed changes of pedestrians and vehicles at time step i+1, respectively. , ), ( , )and( , ) are pedestrians, vehicles, and conflict points. i+1 The coordinates of time.

6. The risk assessment method for human-vehicle interactive game based on risk perception according to claim 5 is characterized in that: The profit function for calculating the profit value of pedestrians and vehicles is: in, j = { p , v} ; represents the change in speed of participant j at time step i Normalization of represents the risk perception value of participant j at time step i Normalization of represents the risk perception value of participant j at time step i, represents the expected coefficient of the speed change level of participant j, represents the expected coefficient of risk perception of participant j, , , .

7. The risk assessment method for human-vehicle interactive game based on risk perception according to claim 6 is characterized in that: In the payoff function, the participant's expected coefficient of the speed change level It is determined by the risk perception value of the participant at the current time step and the acceptable risk perception level of the participant. The method is: in, represents the risk perception value of participant j at time step i, j = { p , v} , [ α j ，β j ] represents the acceptable risk perception level of participant j.

8. The risk assessment method for human-vehicle interactive game based on risk perception according to claim 1 is characterized in that: The specific process of step S4 is: According to the risk perception model of pedestrians and vehicles, the risk perception value range C of pedestrians under the leading condition and the risk perception value range D of pedestrians under the yielding condition are determined respectively, and the intersection of C and D is solved, and the intersection is taken as the acceptable risk perception level of pedestrians; According to the human-vehicle risk perception model, the risk perception value range C1 of the vehicle under the leading condition and the risk perception value range D1 under the yielding condition are determined respectively, and the intersection of C1 and D1 is solved, and the intersection is taken as the acceptable risk perception level of the vehicle.

Images