CN117076816B - Response prediction method, response prediction apparatus, computer device, storage medium, and program product - Google Patents

Abstract

The present application relates to a response prediction method, device, computer equipment, storage medium and program product. The method includes: for a target dangerous traffic scene, starting from a preset stage of the target dangerous traffic scene, calculating the cumulative amount of target evidence information at preset time intervals, the target evidence information is the evidence information based on which the target decision is made; after each calculation, after the cumulative amount is obtained, determining whether the decision response condition is met according to the cumulative amount corresponding to the current calculation moment; if the decision response condition is met, determining the response time required for the driver to make the target decision according to the current calculation moment, in this process, there is no need for the big data-driven deep learning model in traditional technology, so the present application reduces the difficulty of predicting the driver's response time.

Description

Response prediction method, device, computer equipment, storage medium and program product

Technical Field

The present application relates to the field of prediction technology, and in particular to a response prediction method, apparatus, computer equipment, storage medium and program product.

Background technique

With the development of autonomous driving technology, driving assistance systems are becoming more and more intelligent, and users' requirements for driving assistance systems are also getting higher and higher. In dangerous traffic scenarios, predicting the driver's decision-making process and the response time of various decisions can provide more variable inputs for the driving assistance system to predict vehicle motion planning and paths.

In traditional technology, deep learning models are used to predict the driver's response time in normal traffic scenarios, but there is a lack of predictions in dangerous traffic scenarios. This is because deep learning models need to be driven by big data, and data for dangerous traffic scenarios is difficult to obtain. Therefore, in dangerous traffic scenarios, traditional technology has the problem of difficulty in predicting the driver's decision-making response.

Summary of the invention

Based on this, it is necessary to provide a response prediction method, device, computer equipment, storage medium and program product that can reduce the difficulty of predicting the driver's decision response in response to the above technical problems.

In the first aspect, the present application provides a response prediction method. The method includes: for a target dangerous traffic scene, starting from a preset stage of the target dangerous traffic scene, calculating the cumulative amount of target evidence information at preset time intervals, where the target evidence information is evidence information based on which a target decision is made; after each calculation, determining whether a decision response condition is met based on the cumulative amount corresponding to the current calculation moment; if the decision response condition is met, determining the response time required for the driver to make the target decision based on the current calculation moment.

In one of the embodiments, the cumulative amount of target evidence information is calculated at preset time intervals, including: at preset time intervals, obtaining the cumulative amount of target evidence information calculated at the previous calculation moment, and determining the first amount of evidence information that currently supports the target decision and the second amount of evidence information that opposes the target decision based on the road information at the current moment in the target dangerous traffic scene, and obtaining the cumulative amount of the target evidence information at the current calculation moment based on the cumulative amount corresponding to the previous calculation moment, the first amount of evidence information, and the second amount of evidence information.

In one embodiment, the cumulative amount of target evidence information at the current calculation moment is obtained based on the cumulative amount corresponding to the previous calculation moment, the first evidence information amount, and the second evidence information amount, including: obtaining the noise amount at the current calculation moment based on the Wiener process; obtaining the cumulative amount of target evidence information at the current calculation moment based on the cumulative amount corresponding to the previous calculation moment, the first evidence information amount, the second evidence information amount, and the noise amount.

In one of the embodiments, whether the decision response condition is met is determined based on the cumulative amount corresponding to the current calculation moment, including: determining whether the decision response condition is met based on the cumulative amount corresponding to the current calculation moment, driver parameters and vehicle parameters; wherein the driver parameters are related to the information of the driver in the target dangerous traffic scene, and the vehicle parameters are related to the state of the vehicle in the target dangerous traffic scene.

In one of the embodiments, whether the decision response conditions are met is determined based on the cumulative amount corresponding to the current calculation moment, the driver parameters and the vehicle parameters, including: determining whether the sum of the cumulative amount corresponding to the current calculation moment and the driver parameters is greater than or equal to the vehicle parameters; if it is greater than or equal to the vehicle parameters, it is determined that the decision response conditions are met.

In one of the embodiments, the vehicle parameters are calculated based on the state of the vehicle in the target dangerous traffic scene and the objective function, and the objective function is a function whose function value gradually decreases with time.

In one of the embodiments, the target decision is a decision in a decision pool, and the method further includes: determining the response time corresponding to each target decision in the decision pool; sorting the multiple target decisions included in the decision pool according to the response time corresponding to each target decision, and determining the driver's response process in the target dangerous traffic scenario.

In the second aspect, the present application also provides a response prediction device. The device includes: a calculation module, for a target dangerous traffic scene, starting from a preset stage of the target dangerous traffic scene, calculating the cumulative amount of target evidence information at preset time intervals, the target evidence information being the evidence information based on which the target decision is made; a first determination module, after each calculation, after obtaining the cumulative amount, determining whether the decision response condition is met according to the cumulative amount corresponding to the current calculation moment; a second determination module, if the decision response condition is met, determining the response time required for the driver to make the target decision according to the current calculation moment.

In one of the embodiments, the calculation module is specifically used to obtain the cumulative amount of target evidence information calculated at the previous calculation moment every preset time period, and determine the first amount of evidence information that currently supports the target decision and the second amount of evidence information that opposes the target decision based on the road information at the current moment in the target dangerous traffic scene, and obtain the cumulative amount of the target evidence information at the current calculation moment based on the cumulative amount corresponding to the previous calculation moment, the first amount of evidence information, and the second amount of evidence information.

In one of the embodiments, the calculation module is specifically used to obtain the noise amount at the current calculation moment based on the Wiener process; and obtain the cumulative amount of the target evidence information at the current calculation moment according to the cumulative amount corresponding to the previous calculation moment, the first evidence information amount, the second evidence information amount and the noise amount.

In one of the embodiments, the first determination module is specifically used to determine whether the decision response conditions are met based on the cumulative amount, driver parameters and vehicle parameters corresponding to the current calculation moment; wherein the driver parameters are related to the information of the driver in the target dangerous traffic scene, and the vehicle parameters are related to the state of the vehicle in the target dangerous traffic scene.

In one of the embodiments, the first determination module is specifically used to determine whether the sum of the cumulative amount corresponding to the current calculation moment and the driver parameter is greater than or equal to the vehicle parameter. If it is greater than or equal to the vehicle parameter, it is determined that the decision response condition is met.

In one of the embodiments, the vehicle parameters are calculated based on the state of the vehicle in the target dangerous traffic scene and the objective function, and the objective function is a function whose function value gradually decreases with time.

In one embodiment, the target decision is a decision in a decision pool, and the device also includes a third determination module for determining the response time corresponding to each target decision in the decision pool; according to the response time corresponding to each target decision, the multiple target decisions included in the decision pool are sorted to determine the driver's response process in the target dangerous traffic scenario.

In a third aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of any one of the methods described in the first aspect are implemented.

In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any one of the methods described in the first aspect are implemented.

In a fifth aspect, the present application further provides a computer program product, wherein the computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of any one of the methods described in the first aspect are implemented.

The above-mentioned response prediction method, device, computer equipment, storage medium and program product, for the target dangerous traffic scenario, starts from the preset stage of the target dangerous traffic scenario, calculates the cumulative amount of target evidence information at preset time intervals, wherein the target evidence information is the evidence information based on which the target decision is made, and then after each calculation, determines whether the decision response condition is met according to the cumulative amount corresponding to the current calculation moment. If the decision response condition is met, the response time required for the driver to make the target decision is determined according to the current calculation moment. This process does not require the big data-driven deep learning model in traditional technology, so the present application reduces the difficulty of predicting the driver's decision response.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic flow chart of a response prediction method in one embodiment;

FIG2 is a schematic diagram of an evidence accumulation model in one embodiment;

FIG3 is another response prediction method according to an embodiment;

FIG4 is a schematic diagram of a Bayesian framework in one embodiment;

FIG5 is a structural block diagram of a response prediction device in one embodiment;

FIG. 6 is a diagram showing the internal structure of a computer device in one embodiment.

Detailed ways

In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

Dangerous traffic scenarios refer to scenarios where a sudden danger source appears in the road traffic environment or a normal traffic participant turns into a danger source, affecting the normal driving of the vehicle. Dangerous traffic scenarios usually require the driver to actively respond (such as releasing the accelerator, braking, and steering) to avoid risks and avoid collisions.

Vehicles in a road traffic environment can often be divided into four stages from normal driving to collision. In the first stage, the vehicle is driving normally. Facing a complex traffic environment, the vehicle continuously exchanges information with the environment through sensors or drivers to achieve environmental monitoring and ensure the normal driving of the vehicle. When a dangerous condition occurs, the vehicle enters the second stage, the vehicle sensor or driver perceives the danger source, and the vehicle's active safety system intervenes. This stage often lasts from 6 seconds before the collision to 3 seconds before the collision. The probability of the active safety system successfully intervening and avoiding the collision is relatively high. Once the collision is not avoided, the vehicle enters the third stage of critical danger, that is, entering a dangerous traffic scene. The driver takes active response behavior (also known as active behavior) and intervenes in the vehicle to the maximum extent possible. At the same time, the intelligent restraint system is activated. The vehicle intervention may successfully avoid the collision, or it may not be successful but reduce the collision contact intensity. The driver's decision-making behavior and reaction time at this stage directly affect the driver's state at the zero moment of the collision, and thus affect the risk of injury. Finally, it enters the fourth stage of the collision stage, and the vehicle's active and passive safety integrated protection equipment provides targeted protection for the occupants in the vehicle.

In dangerous traffic scenarios, if the driver does not take active control of the vehicle, the dangerous traffic scenario will evolve into a serious traffic accident, threatening the safety of traffic participants. Therefore, the driver needs to actively respond to avoid collision or reduce the intensity of collision when the vehicle enters the third stage of critical dangerous state. The active response mainly focuses on the decision type and the response time of different types (i.e. reaction time). The different decision-making processes in the time course constitute the driver's active response process. Among them, the driver's reaction time in dangerous traffic scenarios is usually defined as the time from the appearance of the danger source or danger information (such as the leading vehicle in the following scenario starts braking) to the driver making an active response decision and taking collision avoidance actions (such as releasing the accelerator, braking, steering, etc.).

In traditional technologies, the research on driver decision type and reaction time prediction algorithms focuses on vehicle routine cruising tasks (such as lane changing, steering, following, fatigue detection, and identification of the driver's second task on autonomous vehicles), but lacks research on driver decision type inference and reaction time prediction in dangerous traffic scenarios. In addition, the traditional driver decision type inference and reaction time prediction models are divided into two technical solutions: decomposition and end-to-end. Among them, the decomposition solution is further divided into three modules: perception, control, and decision-making. It has clear division of labor and strong interpretability, but the system complexity is high and the amount of calculation is large; the end-to-end solution is based on machine learning models or deep learning models to simulate anthropomorphic driving behavior decisions. The amount of calculation is small and the system complexity is low, but the algorithm requirements are high, the safety is low, and the interpretability and reliability are poor. In addition, machine learning models and deep learning models require a lot of data to drive, which is a major limitation of dangerous traffic scene research.

In addition, the inference of driver decision type and prediction of reaction time through machine learning models or deep learning models such as decision trees and fuzzy logic control are mostly data-driven black box processes. Although they have good fitting accuracy, they cannot explain the logical relationship in the decision-making process and it is difficult to explain the distribution of driver reaction time and its changing forms.

In addition, physical quantities such as TTC (Time to Collision) and THW (Time Headway) that describe the urgency of the scene are used as references to infer the driver's decision type and predict the reaction time. However, people, vehicles, and roads constitute an extremely complex road traffic system, and people are the core elements of the system. In dangerous traffic scenes, drivers will perceive key elements, understand potential dangers in traffic scenes, predict the movement form of dangerous sources based on prior knowledge and scene characteristics, make collision avoidance decisions, and finally take collision avoidance actions. Therefore, in real dangerous traffic scenes, the driver's perception, decision-making, and active response are affected by multiple factors such as the driver (age, style, fatigue status, etc.), the vehicle (dynamic status, control mode, etc.), and the traffic scene (accident type, urgency, type of dangerous source, etc.). Therefore, the decision type and reaction time predicted by TTC and THW have the problem of low prediction accuracy.

Based on this, it is necessary to comprehensively consider multiple factors to predict the driver's decision type and reaction time in dangerous traffic scenarios.

Among them, reaction time is an important random variable that affects the development of dangerous situations. In the field of psychology or cognitive science, SDT (Signal Detection Theory) believes that when humans receive stimulus information and make decisions, the receptors in the process of perceiving stimulus signals and the random signal processing process of human brain neurons will generate noise, leading to uncertainty in human decision-making behavior. Moreover, when dealing with most road conflicts, drivers do not need to mobilize the high-level cognitive function areas of the brain for long-term deliberation, but only need to respond quickly based on the perceived information and prior knowledge.

In the field of theoretical neuroscience, decision-making models at different scales have been established for the decision-making process of the human brain under external stimuli. The NSP (Neuron Spiking Model) at the microscopic level can simulate the physiological process of neurons firing when the accumulated potential reaches the threshold after receiving the stimulation signal; the RRM (Reduced Rate Model) at the mesoscopic level is a simplified multi-neuron pulse firing model based on the dynamic stability characteristics of the functional neuron group; the DDM (Drift Diffusion Model) and LATER (Linear Approach to Threshold with Ergodic Rate) models at the macroscopic behavioral level can characterize the behavioral characteristics of people at the behavioral cognitive level. The DDM and LATER models at the behavioral level can characterize the relationship between human decision results, reaction time and changes in stimulation signals. However, compared with the single static decision-making task in traditional cognitive behavioral research, the driver's decision in dynamic traffic scenes is affected by many potential factors.

Most SDT-based models focus on constructing threshold model detection and reaction thresholds, that is, the driver reacts when the signal of interest exceeds the threshold at a certain moment, ignoring the time course of the development of dangerous traffic scenes. The evidence accumulation process emphasizes the evolution and accumulation of different signals accompanied by noise over time. Evidence accumulation models, such as DDM or LATER models, have been shown to fit cognitive decision-making tasks of different tasks and complexity, and studies have confirmed the corresponding neurophysiological mechanisms. However, there is no application in real traffic scenarios.

Therefore, with the help of the evidence accumulation process, the three influencing factors of human-vehicle-road can be integrated into a model to accurately predict the driver's decision type and reaction time. The specific process is described as follows.

In one embodiment, as shown in FIG1 , a response prediction method is provided, which is described by taking the method applied to a terminal as an example, and includes the following steps:

Step 101, for a target dangerous traffic scene, starting from a preset stage of the target dangerous traffic scene, the cumulative amount of target evidence information is calculated at preset time intervals, where the target evidence information is evidence information based on which a target decision is made.

The target dangerous traffic scenario refers to the third stage mentioned above, the driver's active response to avoid collision. Active response refers to the process in which the driver makes a decision and executes the decision based on the dangerous working conditions. For example, the driver sees the leading vehicle start to brake, and after a short thought, the driver decides to turn right and turns the steering wheel to the right.

The preset stage may be the start time of a pre-input target dangerous traffic scenario.

The target decision includes at least one of releasing the accelerator, braking and turning, etc. The evidence information refers to visual change stimuli (viewing angle θ, viewing angle change rate or a combination thereof), visual prompt information such as brake lights, and three-dimensional spatial information such as distance and speed.

Optionally, at preset time intervals, the cumulative amount of target evidence information at the current moment is calculated based on the cumulative amount of target evidence information calculated at the previous calculation moment, the amount of first evidence information supporting the target decision, and the amount of second evidence information opposing the target decision.

Step 102, after each calculation to obtain the cumulative amount, determine whether the decision response condition is met according to the cumulative amount corresponding to the current calculation moment.

Optionally, it is determined whether the cumulative amount corresponding to the current calculation moment is greater than or equal to a decision threshold. If it is greater than or equal to the decision threshold, it is determined that the decision response condition is met.

In another optional embodiment, it is determined whether the sum of the cumulative amount corresponding to the current calculation moment and the driver parameters is greater than or equal to the vehicle parameters. If so, it is determined that the decision response conditions are met, wherein the driver parameters are related to the information of the driver in the target dangerous traffic scene, and the vehicle parameters are related to the state of the vehicle in the target dangerous traffic scene.

Step 103: If the decision response condition is met, the response time required for the driver to make the target decision is determined according to the current calculation time.

Optionally, the cumulative amount of target evidence information for making a target decision will be calculated at preset time intervals, and a determination will be made as to whether the cumulative amount meets the decision response condition. If not, the cumulative amount of the target evidence information will continue to be calculated at the next calculation moment until the cumulative amount calculated at a certain calculation moment meets the decision response condition. This moment will then be taken as the response time required for the driver to make the target decision, that is, the driver's reaction time.

For example, the cumulative amount of target evidence information is calculated every 0.01 seconds. When the cumulative amount calculated at 1.15 seconds meets the decision condition, the response time required for the driver to make the target decision is 1.15 seconds.

To summarize, for the target dangerous traffic scenario, starting from the preset stage of the target dangerous traffic scenario, the cumulative amount of target evidence information is calculated at preset time intervals, wherein the target evidence information is the evidence information based on which the target decision is made. After each calculation, the cumulative amount is obtained, and then it is determined whether the decision response conditions are met according to the cumulative amount corresponding to the current calculation moment. If the decision response conditions are met, the response time required for the driver to make the target decision is determined according to the current calculation moment. In this process, there is no need for big data-driven deep learning models in traditional technologies, and thus the present application reduces the difficulty of predicting the driver's decision response.

In one of the embodiments, the cumulative amount of target evidence information is calculated at preset time intervals, including: at preset time intervals, obtaining the cumulative amount of target evidence information calculated at the previous calculation moment, and determining the first amount of evidence information that currently supports the target decision and the second amount of evidence information that opposes the target decision based on the road information at the current moment in the target dangerous traffic scene, and obtaining the cumulative amount of the target evidence information at the current calculation moment based on the cumulative amount corresponding to the previous calculation moment, the first amount of evidence information, and the second amount of evidence information.

In one embodiment, the cumulative amount of target evidence information at the current calculation moment is obtained based on the cumulative amount corresponding to the previous calculation moment, the first evidence information amount, and the second evidence information amount, including: obtaining the noise amount at the current calculation moment based on the Wiener process; obtaining the cumulative amount of target evidence information at the current calculation moment based on the cumulative amount corresponding to the previous calculation moment, the first evidence information amount, the second evidence information amount, and the noise amount.

Among them, road information is also road environment information, including dangerous scene types (such as rear-end collisions, conflicts at traffic intersections, oncoming vehicles, etc.), hazard source types (such as cars, trucks, pedestrians, cyclists, etc.), collision angles, and collision speeds.

Optionally, the terminal can query the first evidence information supporting the target decision and the second evidence information opposing the target decision from the decision database according to the road information, and respectively count the number of the first evidence information and the second evidence information to obtain the first evidence information amount and the second evidence information amount. The decision database stores multiple groups of corresponding relationships between the supporting and opposing evidence information corresponding to the road information and the target decision information.

The above embodiment can be expressed by the following mathematical formula:

E(t)＝E(t-Δt)+β(t)Δt+ρW(t)Δt (1)

Among them, E(t) represents the cumulative amount at time t, E(t-Δt) represents the cumulative amount corresponding to the previous calculation time, and Δt represents the preset duration. S(t) represents the amount of first evidence information supporting the target decision at time t, O(t) represents the amount of second evidence information opposing the target decision at time t, R represents the road information, g(y _i ) represents the information corresponding to the i-th item included in the road information, such as g(y ₁ ) represents the dangerous scene type, g(y ₂ ) represents the dangerous source type, g(y ₃ ) represents the collision angle, g(y ₄ ) represents the collision speed, and the difference β(t) between S(t) and O(t) is related to the road information and can also be related to the dangerous traffic scene interaction process (stimulus intensity and evidence quality). W(t) represents the Wiener process or Brownian motion, that is, the amount of noise at time t, ρ represents the scaling factor of the Wiener process, and the distribution of W(t)-W(t-Δt) in the Wiener process does not depend on time t, and is expected to be 0, which can be expressed by formula (3).

in addition, It represents the amount of noise evidence accumulated within the preset time period of Δt, Indicates the amount of evidence information accumulated within the preset time length Δt.

In one of the embodiments, whether the decision response condition is met is determined based on the cumulative amount corresponding to the current calculation moment, including: determining whether the decision response condition is met based on the cumulative amount corresponding to the current calculation moment, driver parameters and vehicle parameters; wherein the driver parameters are related to the information of the driver in the target dangerous traffic scene, and the vehicle parameters are related to the state of the vehicle in the target dangerous traffic scene.

In one of the embodiments, whether the decision response conditions are met is determined based on the cumulative amount corresponding to the current calculation moment, the driver parameters and the vehicle parameters, including: determining whether the sum of the cumulative amount corresponding to the current calculation moment and the driver parameters is greater than or equal to the vehicle parameters; if it is greater than or equal to the vehicle parameters, it is determined that the decision response conditions are met.

In one of the embodiments, the vehicle parameters are calculated based on the state of the vehicle in the target dangerous traffic scene and the objective function, and the objective function is a function whose function value gradually decreases with time.

Among them, the driver's information includes age, gender, driving style, driving experience, fatigue status, and distraction status; the vehicle's status includes position, heading angle, speed, acceleration, control status (automatic driving, manual driving, takeover critical state, etc.).

Optionally, the above embodiment can be expressed by the following mathematical formula:

S ₀ +E(t)≥α(t) (4)

E ₀ (t) = α (t) - S ₀ (5)

Among them, formula (4) is the decision response condition, _S0 represents the driver parameter, that is, the initial value of evidence accumulation, which is related to the characteristics of different individual/group drivers, that is, it is related to the driver's information D, f( _xi ) represents the information corresponding to the i-th item included in the driver's information, such as f( _x1 ) represents age, f( _x2 ) represents gender, f( _x3 ) represents driving style, f( _x4 ) represents driving experience, f( _x5 ) represents fatigue state, f( _x6 ) represents distraction state, and _S0 is assumed to obey the normal distribution N(α,σ). α(t) represents the vehicle parameters at time t, that is, the decision threshold at time t, which is related to the vehicle state V and the time pressure t. G(V,t) represents the objective function, and h(z _i ) represents the information corresponding to the i-th item included in the vehicle state, such as h(z ₁ ) represents the position, h(z ₂ ) represents the heading angle, h(z ₃ ) represents the speed, h(z ₄ ) represents the acceleration, and h(z ₅ ) represents the control state. E ₀ (t) represents the evidence accumulation threshold required to reach the decision threshold at time t, which is determined by the difference between the decision threshold at time t and the initial value.

The above formulas (1) to (7) can be considered as the constructed evidence accumulation model. As shown in Figure 2, a schematic diagram of the evidence accumulation model is provided. It can be seen that the initial value S ₀ and the decision threshold (response boundary) α in the evidence accumulation model jointly determine the evidence accumulation threshold E ₀ (t) required for making a decision response. Increasing the initial value or reducing the response boundary is conducive to responding with a small amount of evidence, thereby shortening the response time.

The initial value S ₀ is related to the driver's information. Drivers with rich driving experience can predict dangerous scenes and respond in advance, and drivers with conservative driving style will make decisions and respond as soon as possible to maintain a safer distance. Therefore, the initial values of different individuals or groups can be calculated from the six variables in the driver's information. Drivers with larger initial values can respond faster to dangerous traffic scenes with the same stimulus. The normal distribution represents the distribution forms of different individuals in the group and the random uncertainty of individual responses.

The evidence accumulation rate v corresponds to the strength of the stimulus signal in the dangerous traffic scene. The more urgent the dangerous traffic scene is, the stronger the stimulus signal (such as visual change stimulus, distance, speed, etc.), the greater the evidence accumulation rate, and the faster the evidence accumulation. In addition, the evidence accumulation rate v is also related to road information. The greater the amount of first evidence information supporting the target decision obtained based on the reasoning information, and the smaller the amount of second evidence information opposing the target decision, the greater the evidence accumulation rate.

The response boundary α is related to the state of the vehicle and time pressure. The initial response boundary of the dangerous traffic scene is a constant boundary. If the driver does not take an active response in the early stage, the dangerous traffic scene will deteriorate and face a more urgent situation, thus being forced to make a response decision. In the evidence accumulation model, it is manifested as a decrease in the response boundary over time (collapse boundary α′), that is, the amount of evidence required to trigger the decision becomes smaller as the decision time increases. On a macro level, it is manifested as an insufficient decision with a long reaction time. The state of the vehicle, such as the self-control mode (automatic driving, manual driving, critical state of takeover, etc.) or whether there are brake lights, in-car warning sounds, etc., can also change the response boundary, thereby helping the driver to make a quick decision response. E represents the cumulative amount of evidence information accumulated to make the target decision, that is, the cumulative posterior probability of the decision evidence.

In one of the embodiments, the target decision is a decision in a decision pool, and the method further includes: determining the response time corresponding to each target decision in the decision pool; sorting the multiple target decisions included in the decision pool according to the response time corresponding to each target decision, and determining the driver's response process in the target dangerous traffic scenario.

Optionally, the decision pool includes target decisions such as releasing the accelerator, braking and steering. For each target decision, the response time corresponding to each target decision can be obtained according to the above method. For example, the response time corresponding to releasing the accelerator is 0.70 seconds, the response time corresponding to braking is 0.92 seconds, and the response time corresponding to steering is 1.13 seconds. The driver's response process in the target dangerous traffic scenario is that the driver releases the accelerator at 0.70 seconds, brakes at 0.90 seconds, and turns at 1.13 seconds.

In summary, as shown in FIG3 , another response prediction method is provided, and the response prediction method includes the following steps:

Step 301, for the target dangerous traffic scenario, starting from the preset stage of the target dangerous traffic scenario, at every preset time interval, the cumulative amount of target evidence information calculated at the previous calculation moment is obtained, and the first amount of evidence information supporting the target decision and the second amount of evidence information opposing the target decision are determined based on the road information at the current moment in the target dangerous traffic scenario.

Step 302, obtaining the noise amount at the current calculation moment based on the Wiener process; obtaining the cumulative amount of the target evidence information at the current calculation moment according to the cumulative amount corresponding to the previous calculation moment, the first evidence information amount, the second evidence information amount and the noise amount.

Step 303, determine whether the sum of the cumulative amount corresponding to the current calculation moment and the driver parameters is greater than or equal to the vehicle parameters; if it is greater than or equal to the vehicle parameters, determine that the decision response conditions are met, wherein the driver parameters are related to the information of the driver in the target dangerous traffic scene, and the vehicle parameters are related to the state of the vehicle in the target dangerous traffic scene. The vehicle parameters are calculated based on the state of the vehicle in the target dangerous traffic scene and the objective function, and the objective function is a function whose function value gradually decreases with time.

Step 304: If the sum of the cumulative amount corresponding to the current calculation time and the driver parameter meets the decision response condition, the response time required for the driver to make the target decision is determined according to the current calculation time.

Step 305, the target decision is the decision in the decision pool, and the response time corresponding to each target decision in the decision pool is determined; according to the response time corresponding to each target decision, the multiple target decisions included in the decision pool are sorted to determine the driver's response process in the target dangerous traffic scenario.

This application also has the following advantages:

(1) With the help of cutting-edge research in cognitive psychology, cognitive neuroscience and other fields, an evidence accumulation model is established based on the evidence accumulation process. It is a driver's active response prediction model that can reflect the driver's perception decision-making mechanism. The model can comprehensively consider driver factors (corresponding to the above driver information), road environment factors (corresponding to the above road information), and vehicle state factors (corresponding to the above vehicle state) to infer the driver's decision type and reaction time. It can explain the impact of different factors and parameter changes on the driver's perception decision, improve the interpretability and reliability of the algorithm, and realize the white-box prediction of the driver's active response.

(2) The model can obtain relevant model parameters with a moderate amount of data, avoiding the large amount of data required by data-driven methods. This is more applicable to dangerous traffic scenarios where it is difficult to obtain data.

(3) The model can construct parameterized models that represent different groups and individuals through the representation of driver factors, thereby reflecting the individual differences of drivers. At the same time, it can infer and predict the driver's response process in dangerous traffic scenarios (such as braking time, turning time, etc.), thereby providing personalized driving assistance for different drivers and improving the comprehensibility of the human-computer interaction process.

In addition, the constructed evidence accumulation model is verified as follows: In order to verify the accuracy of the model, two verification methods are provided. The first method uses volunteers to obtain driving simulator test data and test data using real traffic video images as stimuli, respectively, to obtain data information on the driver's active response and decision-making behavior under dangerous conditions. Using simulator tests and video stimulation tests as true values, the accuracy and effectiveness of the established model are measured by comparing the reaction time error between the true value and the model prediction and the accuracy of decision type inference. At this time, attention should be paid to distinguishing between the data training set and the validation set. The reason for using real traffic video images as stimuli is to strive to restore the degree of stimulation of real traffic scenes in a laboratory environment. The second method uses published test data to infer the driver's decision type and predict the reaction time, and compares the results with the existing model established by the corresponding test data to verify the advanced nature of the invented model.

The construction of the above-mentioned evidence accumulation model is based on the Bayesian framework. As shown in Figure 4, a Bayesian framework principle diagram is provided. The idea of the Bayesian framework is as follows:

The Bayesian framework is an appropriate mathematical framework for making decisions based on uncertain data. Bayesian decision-making means that the results of experiments before the decision will affect the decision-making process, that is, it has the combined influence of prior knowledge and posterior knowledge. To make a decision, you must sample sequentially and accumulate evidence step by step until the evidence is convincing enough. The process of gradually sampling and accumulating evidence is the process of continuously using posterior knowledge to update the results in the Bayesian framework.

If there is a hypothesis H and an event E is observed, according to Bayes' theorem:

P(E)P(H|E)＝P(H)P(E|H) (8)

Rearranging this way, we can see that after observing event E, the estimated probability is updated from the old value P(H) (prior) to the new value P(H|E) (posterior):

The left side of the equal sign is the posterior odds, and the right side of the equal sign is the prior odds and likelihood ratio. The likelihood ratio continues to appear in a multiplicative manner during repeated sampling, so after converting the above formula into logarithmic form:

log(posteriorodds)＝log(priorodds)+log(likelihoodratio) (10)

The form of multiplication is transformed into addition, so the likelihood ratio of the evidence supporting a certain option increases after each sampling, and the logarithmic probability of supporting the correct hypothesis becomes larger and larger, representing the rate of information obtained each time. Therefore, through the Bayesian framework, it can be seen that what is needed is a decision signal S representing the logarithmic probability, the initial value of which is S ₀ , representing the logarithmic prior probability; when the number of times occurs, S begins to gradually increase at a rate v, which is determined by the logarithmic likelihood ratio, representing that the sampled information provides supporting evidence for a certain option; the final logarithmic probability will reach the level of decision making S _T. It can be seen that the decision process derived by the Bayesian framework is consistent with the evidence accumulation framework of the reaction time response process.

The modeling process of the above driver factor D, road environment factor R and vehicle state factor V is as follows:

(1) Data acquisition

The driver's perception decision affects the interaction process under dangerous conditions, directly affecting whether a collision can be avoided, the collision time and collision angle, and thus affecting the risk of injury to traffic participants during the collision. Some information about driver factors (such as age, gender, driving style, and driving experience) can be obtained through driver behavior questionnaires or inquiries, but the driver's fatigue and distraction status, as well as road environment factors and vehicle status factors are closely related to dangerous scenarios, and need to be specifically analyzed through dangerous scenario process information and collected physiological signals. The main way to obtain dangerous scenario data is through volunteer experiments. Through the driver behavior questionnaire survey before the test, the test task design and state evaluation to record the relevant information of volunteers for the analysis of driver factors affecting perception decision-making; real-time marking and recording of road environment parameter data during the test, as well as the vehicle status data of the vehicle itself, the factors affecting the driver's perception decision-making can be quantitatively analyzed from different levels.

The above volunteer tests are usually conducted on driving simulators, which can design different dangerous scenarios and perception decision-making tasks. Although real-car tests can provide the subjects with the most realistic driving experience and are most in line with the subjects' driving habits. However, it is impossible to artificially design dangerous traffic scenarios when driving on real roads, which is inefficient and not repeatable. Considering safety, serious dangerous scenario tests cannot be carried out. Relevant dangerous scenario data can also be obtained by analyzing natural driving data sets or accident analysis data sets, combining accident investigation reports, reproducing accident scene information, and completing the statistics of factors affecting driver perception decisions. However, due to the limited information records and the protection of the privacy of investigators, it is difficult to obtain all the required levels of information through this method, and it is impossible to provide sufficient data support for the construction of the model.

(2) Model building

The premise of modeling is to have sufficient data support, and the basic theoretical method is the control variable method. Next, we will take the driver factor as an example. The driver factor can be subdivided into six aspects: age x ₁ , gender x ₂ , driving style x ₃ , driving experience x ₄ , fatigue state x ₅ , and distraction state x _6. Among them, age can be approximated as a continuous variable, while gender, driving style, driving experience, fatigue state, and distraction state are discrete variables.

The classification of each variable is shown in Table 1. Further classification can be carried out according to the research situation and degree of sophistication, such as dividing the distraction state into different levels according to different secondary tasks. The classification of each factor can be specifically determined by the driver behavior questionnaire score and status assessment, and there is already a research basis in this regard. Using the control variable method, the specific impact of each parameter on the driver factor is studied and analyzed separately. It can be divided into two steps. First, determine the positive and negative impact of each factor on the driver's reaction time and determine the function trend line, and then determine the function coefficient value.

Table 1

Variable Name Variable Symbols Variable classification gender x ₂ Female: 0, Male: 1 Driving style x ₃ Conservative: 0, Normal: 1, Radical 2 Driving Experience x ₄ Not rich: 0, Rich: 1 Fatigue status x ₅ Awake: 0, Fatigue: 1 Distracted state x ₆ Concentration: 0, Distraction: 1

First, the effect of age x ₁ on driver factor D is studied. The statistical results and accident data investigation report show that the amount of accident data increases first and then decreases with the driver's age. Therefore, it can be approximately considered that there is a quadratic function relationship between D and x _1. Using the results of the known data set, the parameters a, b, and c of the quadratic function are calculated, and a clear relationship between D and x ₁ is obtained.

Take the fatigue state _x5 as an example to illustrate the influence of discrete variables on driver factor D. Existing research and statistical results show that drivers in a sober state can perceive dangers in traffic scenes more quickly and make decisions. The influence of discrete variables on the factor is expressed by coefficients. The size of the coefficient indicates the degree of influence on the factor. The positive or negative coefficient indicates the effect of the factor (extending or shortening the reaction time). The intercept represents the overall influence drift of the variable on the factor. Using the results of the known data set, the influence factors d and f of the variables are calculated, and a clear relationship between D and _x5 is obtained.

D＝dx ₅ +f

Similarly, a similar method can be used to find the functional relationship between the driver factor D and gender x ₂ , driving style x ₃ , driving experience x ₄ , fatigue state x ₅ , and distraction state x ₆ . Finally, the functional relationship between D and age, gender, driving style, driving experience, fatigue state, and distraction state can be quantified as follows:

Through the above process and steps, the quantitative influence relationship between the road environment factor R and the vehicle state factor V and their related variables can be derived, so as to finally complete the modeling of the factors affecting the driver's perception decision, and characterize its influencing factors and effects from three levels:

(3) Model verification

The established theoretical model needs to be verified experimentally, and two methods for verifying the established theoretical model are proposed. The first method is to verify the data set collected by conducting volunteer behavior experiments. For example, to study the impact of driver age or driving style on driver perception and decision-making characteristics, typical dangerous traffic scenarios can be designed in a targeted manner, and representative subjects can be selected to conduct dangerous perception experiments of drivers in dangerous scenarios. The model effect can be verified and the model parameters can be revised based on the results of the test records. In order to verify the application effect of the model in real traffic scenarios, the revised model can be applied to the natural driving data set for effectiveness verification. At this time, the evaluation model can be verified in a targeted manner based on a small amount of natural driving data type obtained. It should be noted that whether it is a volunteer behavior test data set or a natural driving data set, the training set and validation set of the model need to be effectively divided, and the accuracy of the established driver behavior mechanism model is measured according to the model accuracy verified on the validation set.

It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

Based on the same inventive concept, the embodiment of the present application also provides a response prediction device for implementing the above-mentioned response prediction method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more response prediction device embodiments provided below can refer to the limitations of the response prediction method above, and will not be repeated here.

In one embodiment, as shown in FIG5 , a response prediction device is provided. The response prediction device 500 includes: a calculation module 501, a first determination module 502, and a second determination module 503, wherein:

The calculation module 501 calculates the cumulative amount of target evidence information for the target dangerous traffic scene at a preset time interval starting from the preset stage of the target dangerous traffic scene, where the target evidence information is evidence information based on which the target decision is made;

A first determination module 502 determines whether a decision response condition is satisfied according to the accumulation amount corresponding to the current calculation moment after each calculation.

The second determination module 503 determines the response time required for the driver to make the target decision according to the current calculation time if the decision response condition is met.

In one embodiment, the calculation module 501 is specifically used to obtain the cumulative amount of target evidence information calculated at the previous calculation moment every preset time period, and determine the first amount of evidence information that currently supports the target decision and the second amount of evidence information that opposes the target decision based on the road information at the current moment in the target dangerous traffic scene, and obtain the cumulative amount of the target evidence information at the current calculation moment based on the cumulative amount corresponding to the previous calculation moment, the first amount of evidence information, and the second amount of evidence information.

In one embodiment, the calculation module 501 is specifically used to obtain the noise amount at the current calculation moment based on the Wiener process; and obtain the cumulative amount of the target evidence information at the current calculation moment according to the cumulative amount corresponding to the previous calculation moment, the first evidence information amount, the second evidence information amount and the noise amount.

In one embodiment, the first determination module 502 is specifically used to determine whether the decision response conditions are met based on the cumulative amount, driver parameters and vehicle parameters corresponding to the current calculation moment; wherein the driver parameters are related to the information of the driver in the target dangerous traffic scene, and the vehicle parameters are related to the state of the vehicle in the target dangerous traffic scene.

In one embodiment, the first determination module 502 is specifically used to determine whether the sum of the cumulative amount corresponding to the current calculation moment and the driver parameter is greater than or equal to the vehicle parameter. If it is greater than or equal to the vehicle parameter, it is determined that the decision response condition is met.

In one of the embodiments, the vehicle parameters are calculated based on the state of the vehicle in the target dangerous traffic scene and the objective function, and the objective function is a function whose function value gradually decreases with time.

In one embodiment, the target decision is a decision in a decision pool, and the device also includes a third determination module for determining the response time corresponding to each target decision in the decision pool; according to the response time corresponding to each target decision, the multiple target decisions included in the decision pool are sorted to determine the driver's response process in the target dangerous traffic scenario.

Each module in the above-mentioned response prediction device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be shown in FIG6. The computer device includes a processor, a memory, an input/output interface, a communication interface, a display unit, and an input device. The processor, the memory, and the input/output interface are connected via a system bus, and the communication interface, the display unit, and the input device are connected to the system bus via the input/output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input/output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a response prediction method is implemented. The display unit of the computer device is used to form a visually visible picture, which may be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

Those skilled in the art will understand that the structure shown in FIG. 6 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps described in any of the above method embodiments when executing the computer program.

In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps described in any of the above method embodiments are implemented.

In one embodiment, a computer program product is provided, including a computer program, which implements the steps described in any of the above method embodiments when executed by a processor.

It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims (10)

1. A method of response prediction, the method comprising:

for a target dangerous state traffic scene, starting from a preset stage of the target dangerous state traffic scene, calculating the accumulation amount of target evidence information at intervals of preset time, wherein the target evidence information is evidence information on which a target decision is made;

After the accumulated quantity is obtained through each calculation, determining whether a decision response condition is met or not according to the accumulated quantity corresponding to the current calculation moment;

if the decision response condition is met, determining the response time required by the driver to make the target decision according to the current calculation time;

The calculating the accumulated amount of the target evidence information every preset time length comprises the following steps: and acquiring the accumulated quantity of the target evidence information obtained by calculation at the last calculation time every preset time, determining the first evidence information quantity supporting the target decision currently and the second evidence information quantity opposite to the target decision currently according to the road information at the current time in the target dangerous state traffic scene, and acquiring the accumulated quantity of the target evidence information at the current calculation time according to the accumulated quantity corresponding to the last calculation time, the first evidence information quantity and the second evidence information quantity.

2. The method according to claim 1, wherein the obtaining the accumulated amount of the target evidence information at the current calculation time from the accumulated amount corresponding to the previous calculation time, the first evidence information amount, and the second evidence information amount includes:

Acquiring the noise amount at the current calculation time based on the wiener process;

And acquiring the accumulated quantity of the target evidence information at the current calculation time according to the accumulated quantity corresponding to the last calculation time, the first evidence information quantity, the second evidence information quantity and the noise quantity.

3. The method according to claim 1, wherein determining whether the decision response condition is satisfied according to the accumulated amount corresponding to the current calculation time comprises:

Determining whether the decision response condition is met according to the accumulated quantity, the driver parameter and the vehicle parameter corresponding to the current calculation moment;

Wherein the driver parameter is related to information of a driver in the target dangerous state traffic scene, and the vehicle parameter is related to a state of a vehicle in the target dangerous state traffic scene.

4. A method according to claim 3, wherein said determining whether the decision response condition is satisfied based on the accumulated amount corresponding to the current calculation time, the driver parameter, and the vehicle parameter comprises:

determining whether the sum of the accumulated amount corresponding to the current calculation time and the driver parameter is greater than or equal to the vehicle parameter;

And if the vehicle parameter is greater than or equal to the vehicle parameter, determining that the decision response condition is met.

5. A method according to claim 3, wherein the vehicle parameters are calculated from the state of the vehicle in the target dangerous state traffic scenario and an objective function, the objective function being a function of which the function value gradually decreases with time.

6. The method according to any one of claims 1 to 5, wherein the target decision is a decision in a decision pool, the method further comprising:

determining the response time length corresponding to each target decision in the decision pool;

And sequencing a plurality of target decisions included in the decision pool according to the response time length corresponding to each target decision, and determining the response process of the driver in the target dangerous state traffic scene.

7. A response predicting apparatus, the apparatus comprising:

the calculation module is used for calculating the accumulation of target evidence information for the target dangerous state traffic scene from the preset stage of the target dangerous state traffic scene at intervals of preset time, wherein the target evidence information is evidence information on which a target decision is made;

The first determining module is used for determining whether the decision response condition is met or not according to the accumulated quantity corresponding to the current calculation moment after the accumulated quantity is obtained through each calculation;

The second determining module is used for determining the response time required by the driver to make the target decision according to the current calculation moment if the decision response condition is met;

The calculation module is specifically configured to obtain, at intervals of the preset duration, an accumulated amount of the target evidence information obtained by calculation at a previous calculation time, determine, according to road information at a current time in the target dangerous state traffic scene, a first evidence information amount supporting the target decision currently and a second evidence information amount opposing the target decision, and obtain, according to the accumulated amount corresponding to the previous calculation time, the first evidence information amount and the second evidence information amount, an accumulated amount of the target evidence information at the current calculation time.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 6 when the computer program is executed.

9. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 6.