A vehicle forward collision warning method and system based on driver's real-time reaction time

Abstract

The present invention discloses a vehicle forward collision warning method based on the real-time reaction time of the driver, which belongs to the field of vehicle collision warning. By taking the driver's reaction time, a factor with distinct driver individuality, as an independent variable for real-time calculation and incorporating it into the prediction of the warning system, the actual driving risk of the vehicle is reduced and the driver's acceptance of driving information is improved. By combining driving information with the driver's real-time reaction time, a personalized vehicle forward collision warning method and system are developed, which can greatly improve the driver's acceptance of warning information, improve actual driving safety, and relieve the increased road traffic pressure.

Description

Technical Field

[0001] The present invention belongs to the field of vehicle collision warning, and in particular to a vehicle forward collision warning method and system based on the real-time reaction time of a driver. Background Art

[0002] With the rapid development of vehicle technology and the continuous improvement of social economy, the number of road traffic accidents and huge economic losses are also rising. From a large number of traffic accident statistics, it can be concluded that human errors such as drivers' inaccurate understanding of surrounding road traffic conditions and drivers' lack of concentration are the main causes of traffic accidents. The driver group is facing the current situation of a surge in the number of drivers, an increase in the age span, and uneven driving skills. The standard forward collision warning information cannot adapt to the personalized driving population and urgently needs to be optimized from a personalized perspective.

[0003] In traditional personalized forward collision warning systems, most of them use perception-based models (TTC model, etc.) or kinematics-based models (SDA model, etc.) to warn of potential dangers. Some machine learning-based methods use non-parametric forms to warn in order to enhance the system's adaptability to complex and changeable driver behaviors and improve the driver's acceptance of driving information. However, in the optimization process, a fixed driver reaction time is used for warning. Driver reaction time has individual differences and behavioral fluctuations, and using a fixed time will greatly reduce the degree of personalization of the system. Summary of the invention

[0004] In order to solve the problems existing in the prior art, the present invention provides a vehicle forward collision warning method and system based on the real-time reaction time of the driver, which is used to solve the problem that the optimization process in the prior art uses a fixed driver reaction time for warning.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A vehicle forward collision warning method based on the driver's real-time reaction time comprises the following steps:

[0007] Step 1, obtain the vehicle information and vehicle motion parameters of all vehicles on the current road section, and calculate the distance between two vehicles and the headway time;

[0008] Step 2, based on the data obtained in step 1, according to the similarity and delay of the spatiotemporal images of the host vehicle and the leading vehicle in a stable following state, a curve matching function is designed to calculate the driver's real-time reaction time; the curve matching function uses the Euclidean distance as a measurement indicator of the curve matching degree;

[0009] Step 3, output the real-time driver reaction time with a timestamp, and calculate the real-time driver comfort warning area;

[0010] Step 4: Based on the data obtained in step 1, the vehicle acceleration information of the current road section within the preset time is collected, and the cumulative frequency distribution diagram of all accelerations within the preset time of the current road section is integrated to obtain the acceleration value at 1% as an indicator for determining whether hard braking occurs, and transmit it to the host vehicle;

[0011] Step 5: The host vehicle distinguishes between positive events and negative events based on whether an alarm signal is generated and whether the driver confirms the alarm information within the comfort warning range after the system triggers the alarm, and determines whether to trigger the braking action to determine a real event or a false event;

[0012] Step 6, expressing the decision problem of forward collision warning as a Markov decision process model based on the driver's real-time reaction time;

[0013] Step 7: Construct a DQN model based on the Markov decision model in step 6 to train the data set in step 1;

[0014] Step 8: Based on the training results in step 7, a warning operation is performed on the driver.

[0015] Preferably, in step 1, vehicle information of all vehicles on the current road section is obtained through roadside equipment, and the vehicle information includes time, speed, acceleration, number of vehicles, and vehicle number; at the same time, vehicle motion parameters are obtained through on-board equipment, and the vehicle motion parameters include the speed, acceleration, position information, and vehicle number of the main vehicle and the vehicle immediately preceding it in the same lane.

[0016] Preferably, the stable following state is a situation where the vehicle information of the leading vehicle remains unchanged, the host vehicle has no intention of changing lanes or overtaking, and the distance between the two vehicles is relatively stable.

[0017] Preferably, step 2 specifically includes the following steps:

[0018] Step 2.1, mark each frame of data obtained in step 1 as a data point in the time-space diagram; translate each data point in the trajectory of the main vehicle and overlap it with the data point of the front vehicle to obtain a motion vector V;

[0019] Step 2.2, after translating the entire trajectory of the main vehicle along the motion vector V, the Euclidean distance between the data points of the overlapping parts is calculated; the Euclidean distance (D E ) is calculated as follows:

[0020]

[0021] in, is the data point of the host vehicle, is the data point of the preceding vehicle, i and j represent the data points currently matched between the two vehicles, and n is the number of matched data points;

[0022] Step 2.3, matching each data point of the main vehicle according to step 2.1 and step 2.2 to obtain the corresponding Euclidean distance; after completing a round of traversal, selecting the motion vector (V) corresponding to the minimum Euclidean distance as the final matching scheme to obtain the component of V in the time direction, that is, the driver's real-time reaction time;

[0023] Step 2.4, when executing curve matching, set a fixed-size window, and each window only matches the trajectory in the window; the result obtained by the curve matching algorithm is the driver's real-time reaction time of the last frame in the current window. A trajectory curve match is performed once for each step, and the parameters are updated once.

[0024] Preferably, in step 3, the driver's comfort warning zone is calculated as a confidence interval of ±5% of the driver's real-time reaction time.

[0025] Preferably, step 6 specifically includes the following steps:

[0026] Step 6.1, driving information and driver personalized information are combined into a state set in the MDP model to form a vector set as follows:

[0027]

[0028] Where t is the current time, and the driving information data includes the speed of the main vehicle Acceleration Headway and the gap between the main vehicle and the front vehicle (gap t );Driver personalized information is the driver's real-time reaction time

[0029] Step 6.2, the action set in the MDP model is the activation state of the warning information;

[0030] Step 6.3, classify the current state according to the five different categories (TP, FP-E, FP-L, TN, FN) in the data obtained by the vehicle terminal, and design the reward function;

[0031] The Markov decision process model makes a decision on whether to trigger an alarm based on the reward function value, and then monitors the driver's behavior to see whether the driver will confirm the alarm information.

[0032] Preferably, step 7 specifically includes the following steps:

[0033] Step 7.1, establish the data set required for neural network training in the DQN model, and the data set is the personal data set of each driver;

[0034] Step 7.2, the driving data of the frequently used drivers is saved separately. When the actual driver of the main vehicle changes, the corresponding driver data set can be selected to meet the driving needs of different drivers;

[0035] Step 7.3, calculate the Q value in the DQN model. The DQN model takes the state and action set as the input of the neural network and outputs the corresponding Q value. At the same time, build a target neural network to generate a target Q value. After the target neural network is introduced, the target Q value will remain unchanged for a period of time.

[0036] The model parameters of the policy neural network are updated according to the loss function; after N iterations, the parameters of the policy neural network are copied to the target neural network.

[0037] Preferably, in step 8, when the warning operation is performed, human-computer interaction is performed on the HMI interface; the warning modes of the human-computer interaction include a normal warning mode and an increased attention warning mode.

[0038] A vehicle forward collision warning system based on driver's real-time reaction time, comprising a standard value calculation module, a driver's real-time reaction time calculation module, a data division module, a DQN-based warning decision module and an HMI display module;

[0039] The driver's real-time reaction time calculation module is used to calculate the driver's real-time reaction time, and the driver's real-time reaction time calculation module and the standard value calculation module are both connected to the input end of the data division module;

[0040] The output end of the data division module is connected to the input end of the DQN-based early warning decision module, and the output end of the DQN-based early warning decision module is connected to the HMI display module.

[0041] Compared with the prior art, the present invention has the following beneficial technical effects:

[0042] The present invention provides a vehicle forward collision warning method based on the driver's real-time reaction time. By taking the driver's reaction time, a factor with distinct driver individuality, as an independent variable for real-time calculation and incorporating it into the prediction of the warning system, the actual driving risk of the vehicle is reduced and the driver's acceptance of driving information is improved. By combining driving information with the driver's real-time reaction time, a personalized vehicle forward collision warning method and system are developed, which can greatly improve the driver's acceptance of warning information, improve actual driving safety, and relieve the increased road traffic pressure. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a system structure diagram of an embodiment of the present invention;

[0044] Figure 2 A schematic diagram of a curve matching function according to an embodiment of the present invention;

[0045] Figure 3 Schematic diagram of data classification according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The present invention is further described in detail below in conjunction with specific embodiments, which are intended to explain the present invention rather than to limit it.

[0047] The present invention provides a vehicle forward collision warning method based on the driver's real-time reaction time, thereby reducing the actual driving risk of the vehicle and improving the driver's acceptance of driving information. The main innovation is to use the driver's real-time reaction time, a factor with distinct driver personalization, as an independent variable for real-time calculation and incorporate it into the prediction of the warning system. A non-parametric forward collision warning model based on the DQN (Deep Q-Learning) model that combines driving information with personalized factors is proposed.

[0048] A vehicle forward collision warning method based on the driver's real-time reaction time of the present invention specifically includes the following steps:

[0049] Step 1: Obtain vehicle information (time, speed, acceleration, number of vehicles, vehicle number) of all vehicles on the current road section through roadside equipment; at the same time, obtain vehicle motion parameters (including the speed, acceleration, position information, and vehicle number of the main vehicle and its adjacent preceding vehicle in the same lane) through on-board equipment, and calculate the vehicle spacing and headway between the two vehicles through the collected data.

[0050] Step 2, using the data obtained by the vehicle-mounted device in step 1 above, design a curve matching function to calculate the driver's real-time reaction time based on the similarity and delay between the time-space images of the main vehicle and the front vehicle in a stable following state. The stable following state is when the vehicle information of the front vehicle remains unchanged, the main vehicle has no intention of changing lanes or overtaking, and the distance between the two vehicles is relatively stable. The specific function design is as follows: Taking the Euclidean distance (D E ) is used as a measure of the degree of curve matching.

[0051] Step 2.1, mark each frame of acquired data as a data point in the time-space diagram. Each data point in the trajectory of the main vehicle is translated and overlapped with the data point of the front vehicle to obtain a motion vector (V).

[0052] Step 2.2, after translating the entire trajectory of the main vehicle along the motion vector V, the Euclidean distance between the data points of the overlapping parts is calculated. E ) is calculated as shown in Formula I below.

[0053]

[0054] in is the data point of the host vehicle, is the data point of the preceding vehicle, i and j represent the data points currently matched between the two vehicles, and n is the number of matched data points.

[0055] Step 2.3, match each data point of the host vehicle according to the above loop method to obtain the corresponding Euclidean distance. After a round of traversal, select the motion vector (V) corresponding to the minimum Euclidean distance as the final matching solution to obtain the component of V in the time direction, that is, the driver's real-time reaction time.

[0056] Step 2.4, when performing curve matching, set a fixed-size window (15 seconds, 150 frames), and each match only matches the trajectory in the window. The result obtained by the curve matching algorithm is the real-time reaction time of the driver in the last frame of the current window. Each step performs a trajectory curve match and updates the parameters. Step 2.4 is used to determine the range of the curves that need to be matched (because of the fluctuations in the driver's behavior, all trajectory curves cannot be used for matching), and specifies which frame the current matching result is used as the reaction time. For example: using the trajectory from frame 1 to frame 150 for curve matching, the actual reaction time of frame 150 will be obtained, and then the trajectory from frame 2 to frame 151 will be used for curve matching to obtain the actual reaction time of frame 151.

[0057] Step 3: Output and record the calculated real-time driver reaction time with a timestamp. Calculate the real-time driver's comfort warning area (the warning information issued in the current area is most likely to be accepted by the driver), calculated as a confidence interval of ±5% of the driver's actual reaction time.

[0058] Step 4, based on the data obtained by the roadside equipment in step 1, the vehicle acceleration information in the past hour of the current section is collected, and the cumulative frequency distribution diagram of all accelerations in the past hour of the current section is integrated, and the acceleration value at 1% is obtained as an indicator for judging whether hard braking occurs (whether there is a dangerous situation), and transmitted to the host vehicle. The hard braking indicator refers to whether the braking action is triggered in step 5, which is used to judge whether the driver confirms the alarm information generated by the system.

[0059] In step 5, the host vehicle uses whether the system of the present invention generates an alarm signal to distinguish positive events from negative events, and whether the driver confirms the alarm information within a certain time (driver comfort warning interval) after the system triggers the alarm, that is, whether the braking action is triggered (that is, the index value calculated in step 4) to judge real events or false events. According to the above standards, the data obtained by the vehicle-mounted terminal in step 1 is divided into the following four types: true positive (TP), false positive (FP), false negative (FN) and true negative (TN). True (TP) data includes the situation where the CWS is correctly alarmed within an appropriate time interval. False alarm (FP) data includes false alarm events, which can be divided into two situations: early alarm (FP-E) and late alarm (FP-L), according to whether the warning is sent earlier or later than a given time interval. FP will reduce the driver's trust in CWS and may also interfere with the driver's normal driving. False negative (FN) data includes lost alarm events, which are dangerous but not detected. True negative (TN) data is safe driving data and there is no safety hazard. Step 5 determines the data type for the subsequent generation of reward functions for reinforcement learning.

[0060] Step 6: Formulate the decision problem of forward collision warning as a Markov decision process (MDP) model. The specific steps are as follows.

[0061] Step 6.1, the state set in the MDP model is a combination of driving information and driver personalized information, which is described as a collection of vectors shown in the following formula II.

[0062]

[0063] Where t is the current time, and the driving information data includes the speed of the main vehicle Acceleration Headway and the gap between the main vehicle and the front vehicle (gap t );Driver personalized information is the driver's real-time reaction time

[0064] Step 6.2, the action set in the MDP model is the activation state of the warning information, which is defined as follows: Equation III.

[0065] actions = {0,1} Formula III

[0066] Step 6.3, classify the current state according to the five different categories (TP, FP-E, FP-L, TN, FN) in step 5 of the above data description part, and design the reward function. The specific reward function design is shown in Formula IV below.

[0067] R(s,t)=I(t)*R′(s,t) Formula IV

[0068] The time-varying component I(t) is a binary indication signal, and its calculation formula is shown in the following formula V.

[0069] I(t)=I1(t)*I2(t) Formula V

[0070] Among them, I1(t)∈{-1,1} reflects whether the system generates an alarm signal, and I2(t)∈{-1,1} reflects whether the driver confirms the alarm information, that is, whether an emergency braking operation occurs.

[0071] The model will first make a judgment based on the previous reward function value and take corresponding actions (whether to trigger the alarm), and then monitor the driver's behavior to see whether the driver will confirm the alarm information (i.e., whether to perform hard braking). The R′(s,t) function at time t is designed as shown in Formula VI, Formula VII, and Formula VIII.

[0072]

[0073]

[0074]

[0075] where t h It is the time point when the danger actually occurs (the driver performs hard braking), the parameters u and p are 1000 and 1, and the parameters v, w, and o are: 600, 600, and 850.

[0076] Step 7: Build a DQN model based on the Markov decision model proposed in step 6. The specific steps are as follows.

[0077] Step 7.1, the data set required for neural network training in the DQN model is a personal data set for each driver. In the initial stage of use, you can choose to import the driver's previous data or select the cautious, aggressive, and medium driver data sets as the initial data set. In the early stage, the data set is updated every hour, the existing data is written into the data set and the data in the initial data set is randomly cleaned until all the initial data is rewritten, and new data is recorded into the data set after every 5 hard brakes.

[0078] Step 7.2, the driving data of the frequently used drivers are saved separately. When the actual driver of the main vehicle changes, the corresponding driver data set can be selected to meet the driving needs of different drivers.

[0079] Step 7.3, for the Q value in the DQN model, the calculation formula is shown in the following formula IX.

[0080] Q(s,a)* ←Q(s,a)+α(R+γmax a′ (s ′ ,a ′ )-Q(s,a))Formula IX

[0081] Where α is the learning efficiency that represents the degree of learning of the deviation, R represents the value of the reward function, and γ represents the decay value of future rewards.

[0082] The DQN model takes the state and action set as the input of the neural network and outputs the corresponding Q value. At the same time, a target neural network is constructed to generate a target Q value. After the target neural network is introduced, the target Q value will remain unchanged for a period of time. T ) is calculated as shown in Formula X.

[0083] Q T =R+γmax a′ (s ′ ,a ′ ,θ) formula X

[0084] Where θ represents the network parameters. Then, the model parameters of the policy neural network are updated according to the loss function. After N iterations, the parameters of the policy neural network are copied to the target neural network. The formula of the loss function is shown in Equation XI below.

[0085] L(θ)=E[(Q T -Q(s,a,θ)) 2 Formula XI

[0086] Step 8, according to the operation provided by the model in step 7, the driver is warned and human-computer interaction is performed in the HMI interface. You can choose between two warning modes. The first one is the "normal warning" mode, which only performs visual warnings and is suitable for people with rich driving experience, that is, only provides red light alarms in the display panel. The second one is the "increased attention warning" mode, which performs visual alarms and auditory alarms. It is suitable for people with relatively poor driving experience or the elderly or those with poor attention. That is, a red light alarm is given in the display panel and a "beep" sound is used to warn the driver at the same time to increase the driver's attention.

[0087] The present invention also proposes a vehicle forward collision warning system based on the driver's real-time reaction time according to the forward collision warning method described above. The system includes two parts: roadside equipment and vehicle-mounted equipment.

[0088] The roadside equipment can be divided into three modules: standard value calculation module and V2I information transmission module. The vehicle-mounted equipment can be specifically divided into: V2I information receiving module, driver real-time reaction time calculation module, data division module, DQN-based warning decision module and HMI display interface. The specific description is as follows.

[0089] Example

[0090] like Figure 1 As shown, a system structure diagram of a vehicle forward collision warning system based on the real-time reaction time of the driver according to an embodiment of the present invention can be divided into two parts: roadside equipment and vehicle-mounted equipment. The real-time driver reaction time calculation module in the vehicle-mounted equipment obtains the driver's personalized indicators, and obtains the real-time road information indicators obtained by the standard value calculation module in the roadside equipment through the V2I communication module. The two together serve as the distinction criteria of the data division module to distinguish the corresponding data. The divided data is applied to the DQN-based warning decision module to obtain personalized warning information and present it on the human-computer interaction HMI interface to achieve the goal of providing drivers with real-time, reliable and adaptive warning information.

[0091] like Figure 2 As shown in FIG. 1 , an example diagram of the curve matching function in the driver real-time reaction time module of the present invention is shown, wherein the blue dots represent the data points of the preceding vehicle, and the red dots represent the data points of the host vehicle. The green dots represent the position of each point after the red dots are translated along the matching vector, and the points in the box represent the points for which the Euclidean distance needs to be calculated.

[0092] The data classification diagram of the present invention is shown in Figure 3, where t represents the current time, The DRT represents the time t, the red box represents the confidence interval (driver comfort warning interval), and the red explosion point represents the time when the hard braking operation occurs. When the system triggers the warning and hard braking operation occurs in the comfort warning area, the data at this time is classified as true positive data (TP); if the alarm is outside the comfort area, it is classified as false positive data (FP); if it causes interference alarm before the area, it is classified as premature alarm (FP-E); if there is a safety hazard afterwards, it is classified as late alarm (FP-L). When the system does not trigger the alarm and there is no hard braking operation in the comfort area, it is classified as true negative data (TN); if hard braking operation occurs, it is false negative data (FN).

[0093] Specific implementation use case 1:

[0094] When the driver is an elderly person and goes out in rainy or snowy weather, the elderly person's ability to react to dangerous situations during actual driving is relatively slow, and their control over the vehicle speed will be relatively weak. Rainy and snowy weather also presents potential dangers such as low visibility and slippery roads, making collision accidents more likely to occur. Therefore, a personalized warning system is needed to provide drivers with appropriate and safe collision warning prompts to improve driving safety.

[0095] When conducting forward collision warning risk assessment for actual road traffic conditions, the present invention sets an independent driver data set for common drivers and independently performs parameter training for the machine learning network; at the same time, two levels of warning modes are provided, so that the system gives adaptive warning information when facing different drivers, thereby improving driving safety and comfort.

[0096] When elderly drivers are preparing to start a journey, they can select their previous driving data set in advance and select the "increased attention warning" mode to make the driving process more comfortable and safe.

[0097] When the elderly driver starts to drive normally, in order to better grasp the driver's real-time reaction during the actual driving process, the present invention takes into account the driver's real-time reaction time when making warning information decisions, and monitors the driver's personalized data in real time; at the same time, the driving data is updated in real time, and the system parameters are continuously iterated to be closest to the current state, so that the present invention can generate warning information in real time, so as to achieve the adaptability of the system and meet daily driving needs.

[0098] At the same time, the vehicle conditions in rainy and snowy weather are more special than those in normal weather, and the judgment conditions for potential dangerous situations are also different. When making early warning decisions, the present invention obtains the real-time potential danger judgment index data of the current road section in the past hour transmitted by the roadside equipment through V2I communication, and divides the data in real time to apply it to the early warning decision of the main vehicle, making the driving process safer.

Claims

1. A vehicle forward collision warning method based on the driver's real-time reaction time, characterized in that: The following steps are included: Step 1, obtain the vehicle information and vehicle motion parameters of all vehicles on the current road section, and calculate the distance between two vehicles and the headway time; Step 2, based on the data obtained in step 1, according to the similarity and delay of the spatiotemporal images of the host vehicle and the leading vehicle in a stable following state, a curve matching function is designed to calculate the driver's real-time reaction time; the curve matching function uses the Euclidean distance as a measurement indicator of the curve matching degree; Step 3, output the real-time driver reaction time with a timestamp, and calculate the real-time driver comfort warning area; Step 4: Based on the data obtained in step 1, the vehicle acceleration information of the current road section within the preset time is collected, and the cumulative frequency distribution diagram of all accelerations within the preset time of the current road section is integrated to obtain the acceleration value at 1% as an indicator for determining whether hard braking occurs, and transmit it to the host vehicle; Step 5: The host vehicle distinguishes between positive events and negative events based on whether an alarm signal is generated and whether the driver confirms the alarm information within the comfort warning range after the system triggers the alarm, and determines whether to trigger the braking action to determine a real event or a false event; Step 6, expressing the decision problem of forward collision warning as a Markov decision process model based on the driver's real-time reaction time; Step 7: Construct a DQN model based on the Markov decision model in step 6 to train the data set in step 1; Step 8: Based on the training results in step 7, a warning operation is performed on the driver.

2. The vehicle forward collision warning method based on the driver's real-time reaction time according to claim 1, characterized in that: In step 1, the vehicle information of all vehicles on the current road section is obtained through the roadside equipment, and the vehicle information includes time, speed, acceleration, number of vehicles, and vehicle number; at the same time, the vehicle motion parameters are obtained through the on-board equipment, and the vehicle motion parameters include the speed, acceleration, position information, and vehicle number of the main vehicle and the vehicle immediately preceding it in the same lane.

3. The vehicle forward collision warning method based on the driver's real-time reaction time according to claim 1, characterized in that: The stable following state is a situation where the vehicle information of the leading vehicle remains unchanged, the host vehicle has no intention of changing lanes or overtaking, and the distance between the two vehicles is relatively stable.

4. The vehicle forward collision warning method based on the driver's real-time reaction time according to claim 1, characterized in that: Step 2 specifically includes the following steps: Step 2.1, mark each frame of data obtained in step 1 as a data point in the time-space diagram; translate each data point in the trajectory of the main vehicle and overlap it with the data point of the front vehicle to obtain a motion vector V; Step 2.2, after translating the entire trajectory of the main vehicle along the motion vector V, the Euclidean distance between the data points of the overlapping parts is calculated; the Euclidean distance (D E ) is calculated as follows: in, is the data point of the host vehicle, is the data point of the preceding vehicle, i and j represent the data points currently matched between the two vehicles, and n is the number of matched data points; Step 2.3, matching each data point of the main vehicle according to step 2.1 and step 2.2 to obtain the corresponding Euclidean distance; after completing a round of traversal, selecting the motion vector (V) corresponding to the minimum Euclidean distance as the final matching scheme to obtain the component of V in the time direction, that is, the driver's real-time reaction time; Step 2.4, when executing curve matching, set a fixed-size window, and each window only matches the trajectory in the window; the result obtained by the curve matching algorithm is the driver's real-time reaction time of the last frame in the current window. A trajectory curve match is performed once for each step, and the parameters are updated once.

5. The vehicle forward collision warning method based on the driver's real-time reaction time according to claim 1, characterized in that: In step 3, the driver's comfort warning zone is calculated as a ±5% confidence interval of the driver's real-time reaction time.

6. The vehicle forward collision warning method based on the driver's real-time reaction time according to claim 1, characterized in that: Step 6 specifically includes the following steps: Step 6.1, driving information and driver personalized information are combined into a state set in the MDP model to form a vector set as follows: Where t is the current time, and the driving information data includes the speed of the main vehicle Acceleration Headway and the gap between the main vehicle and the front vehicle (gap t );Driver personalized information is the driver's real-time reaction time Step 6.2, the action set in the MDP model is the activation state of the warning information; Step 6.3, classify the current state according to the five different categories (TP, FP-E, FP-L, TN, FN) in the data obtained by the vehicle terminal, and design the reward function; The Markov decision process model makes a decision on whether to trigger an alarm based on the reward function value, and then monitors the driver's behavior to see whether the driver will confirm the alarm information.

7. The vehicle forward collision warning method based on the driver's real-time reaction time according to claim 1, characterized in that: Step 7 specifically includes the following steps: Step 7.1, establish the data set required for neural network training in the DQN model, and the data set is the personal data set of each driver; Step 7.2, the driving data of the frequently used drivers is saved separately. When the actual driver of the main vehicle changes, the corresponding driver data set can be selected to meet the driving needs of different drivers; Step 7.3, calculate the Q value in the DQN model. The DQN model takes the state and action set as the input of the neural network and outputs the corresponding Q value. At the same time, build a target neural network to generate a target Q value. After the target neural network is introduced, the target Q value will remain unchanged for a period of time. The model parameters of the policy neural network are updated according to the loss function; after N iterations, the parameters of the policy neural network are copied to the target neural network.

8. The vehicle forward collision warning method based on the driver's real-time reaction time according to claim 1, characterized in that: In step 8, when performing the warning operation, human-computer interaction is performed on the HMI interface; the warning modes of human-computer interaction include a normal warning mode and an increased attention warning mode.

9. A vehicle forward collision warning system based on the driver's real-time reaction time, characterized in that: It includes standard value calculation module, driver real-time reaction time calculation module, data division module, DQN-based warning decision module and HMI display module; The driver's real-time reaction time calculation module is used to calculate the driver's real-time reaction time, and the driver's real-time reaction time calculation module and the standard value calculation module are both connected to the input end of the data division module; The output end of the data division module is connected to the input end of the DQN-based early warning decision module, and the output end of the DQN-based early warning decision module is connected to the HMI display module.

Images