Method and system for quantifying safety degree of road section pedestrian crossing scene

Abstract

The invention relates to the technical field of automatic driving, and particularly discloses a method and system for quantifying the safety degree of a road section pedestrian crossing scene, and the method comprises the steps: determining a risk field range of pedestrians in the crossing scene; acquiring a traffic accident case and motor vehicle driving scene data, and determining main factors influencing pedestrian crossing safety in a risk field range; extracting related factor characteristics in the traffic accident case and the motor vehicle driving scene data based on the main factors, establishing a sample data set, training an XGBoost model used for dividing safety levels, and setting the safety levels; and identifying a target vehicle based on the trained XGBoost model to obtain a safety level. According to the road section pedestrian crossing safety degree quantification method, the TTC theory and the XGBoost model are integrated, the safety level of pedestrians crossing the street in the road section is efficiently and accurately found, it is guaranteed that the pedestrians can safely cross the street in the road section, and the method has important significance.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a method and system for quantifying the safety of pedestrian crossing scenes on a road section. Background Art

[0002] With the development of technology, autonomous driving has become a hot topic. It is becoming more and more normal for autonomous driving to be approved for use on the road. The forms of road traffic participants have also gradually changed from traditional drivers, vehicles and pedestrians to human-driven, human-machine co-driving, vehicles and pedestrians. Road safety will also face more severe challenges. Especially involving pedestrian traffic safety. Pedestrians are traffic participants without protective measures and are most vulnerable to injury in road traffic accidents. Among them, traffic accidents involving pedestrians crossing the road are an important part of pedestrian traffic accidents. This makes the interaction between autonomous driving, human-driven driving, and pedestrian crossing the street particularly important in the new form. In the traditional human-driven traffic mode, the driver will judge the situation of pedestrians crossing the street by himself. In the existing mixed traffic mode, when the driver ignores the communication with pedestrians when crossing the street due to focusing on secondary tasks, such as eye contact, gesture communication, honking, etc. This will also bring great hidden dangers to road traffic safety. How autonomous driving determines the safety of pedestrians crossing the street becomes crucial. Therefore, in view of how machines can determine the safety of pedestrians crossing the street under autonomous driving and avoid traffic accidents, it is of great significance to provide a method to quantify the safety of pedestrians crossing the street and ensure that pedestrians can cross the street safely on the road section. Summary of the invention

[0003] The purpose of the present invention is to provide a method and system for quantifying the safety of pedestrian crossing scenes on a road section, so as to solve the problems raised in the above background technology.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A method for quantifying the safety of a pedestrian crossing scene on a road section, the method comprising:

[0006] Determine the risk range of pedestrians in street crossing scenarios;

[0007] Obtain traffic accident cases and motor vehicle driving scenario data to determine the main factors affecting pedestrian crossing safety within the risk field; wherein the main factors include vehicle speed, operating margin, pedestrian initial speed, pedestrian acceleration, pedestrian type, pedestrian sight, weather conditions and other vehicles; the operating margin is time, indicating the length of time the driver is allowed to operate;

[0008] Based on the main factors, relevant factor features are extracted from traffic accident cases and motor vehicle driving scene data, a sample data set is established, an XGBoost model for dividing safety levels is trained, and safety levels are set;

[0009] Identify the target vehicle based on the trained XGBoost model to obtain the safety level.

[0010] As a further solution of the present invention: the step of determining the risk field range of pedestrians in the crosswalk scenario includes:

[0011] Obtain the vehicle speed;

[0012] Detect pedestrians based on on-vehicle radar to obtain the initial speed and acceleration of the pedestrians;

[0013] Create a plane rectangular coordinate system based on the vehicle, obtain the pedestrian coordinates based on on-vehicle radar, and fill them into the plane rectangular coordinate system; wherein, the plane rectangular coordinate system takes the road cross-section direction as the y-axis and the vehicle driving direction as the x-axis;

[0014] Determine the risk field range in the plane rectangular coordinate system.

[0015] As a further solution of the present invention: the step of determining the risk field range in the plane rectangular coordinate system includes:

[0016] Determine the time for the pedestrian to reach the vehicle driving path according to the initial speed, acceleration and coordinates of the pedestrian, as the first time;

[0017] Calculate the driving distance of the vehicle within the first time;

[0018] Calculate the remaining distance from the vehicle to the conflict point based on the driving distance;

[0019] Calculate the operation margin based on the remaining distance;

[0020] Obtain the braking acceleration of the vehicle, calculate the braking distance according to the braking acceleration, and calculate the risk field radius according to the braking distance, operation margin and preset reaction time;

[0021] Determine the risk field range according to the risk field radius.

[0022] As a further solution of the present invention: the step of obtaining traffic accident cases and motor vehicle driving scenario data and determining the main factors affecting pedestrian crosswalk safety within the risk field range includes:

[0023] According to the traffic accident cases of motor vehicles and pedestrians when driving on the road section and the motor vehicle driving scenario data of the road section, use the Bayesian algorithm to extract the relevant static factors leading to traffic accidents; the relevant static factors include: road flatness, lane markings, road width and whether there is a central median;

[0024] Within the established risk field, use in-vehicle radar to find dynamic factors related to traffic accidents; the dynamic factors include vehicle speed, operating margin, pedestrian initial speed, pedestrian acceleration, pedestrian type, pedestrian line of sight, weather conditions, and other vehicles.

[0025] As a further solution of the present invention: the steps of extracting relevant factor features from traffic accident cases and motor vehicle driving scenario data based on the main factors, establishing a sample data set, training an XGBoost model for dividing safety levels, and setting safety levels include:

[0026] Extract relevant factor features from traffic accident cases and motor vehicle driving scenario data based on the main factors, and establish a sample data set;

[0027] Determine and quantify the relative weights of various factors within the risk field in the sample data set;

[0028] Divide the data set, divide the established standardized sample data set, randomly select 30% of the sample data in the sample data set as the test set, and 70% of the sample data as the training set;

[0029] Train an XGBoost model for dividing safety levels based on the training set and the test set, and set safety levels;

[0030] As a further solution of the present invention: the steps of determining and quantifying the relative weights of various factors within the risk field in the sample data set include:

[0031] Perform standardized processing on the original data in the sample data set;

[0032] Calculate the mean and standard deviation of each index;

[0033] Calculate the coefficient of variation of each index, and calculate the weight of each index based on the coefficient of variation;

[0034] Calculate the overall safety index, and perform standardized processing on the overall safety index.

[0035] As a further solution of the present invention: the steps of training an XGBoost model for dividing safety levels based on the training set and the test set, and setting safety levels include:

[0036] Initialize the XGBoost model;

[0037] Perform H iterations on the initialized XGBoost model;

[0038] After H iterations, the model reaches the maximum number of iterations, stops iterating, and finally obtains an overall safety index model based on the XGBoost algorithm;

[0039] Set the security level;

[0040] The step of performing H iterations on the initialized XGBoost model includes:

[0041] Calculate the residual in the H-th iteration process;

[0042] Generate the weak training learner for the H-th iteration;

[0043] Calculate the best fit value;

[0044] Update the simulated prediction value.

[0045] As a further solution of the present invention: The step of setting the security level includes:

[0046] Read the overall security index under the trained XGBoost algorithm and sort it in ascending order;

[0047] Use the trained XGBoost model to set the security level and calculate the positions of 85% and 15% of the overall security index;

[0048] Read the corresponding values in the sorted overall security index according to the positions of 85% and 15% of the overall security index. The value corresponding to 85% of the overall security index is the first value, and the value corresponding to 15% of the overall security index is the second value;

[0049] When the overall security index is less than the second value, the security level is unsafe. When the overall security index is between the second value and the first value, the security level is generally safe. When the overall security index is greater than the first value, the security level is very safe.

[0050] As a further solution of the present invention: The step of identifying the target vehicle based on the trained XGBoost model to obtain the security level includes:

[0051] Taking the security level set by the trained model as the standard, during driving, the vehicle uses the on-board radar to obtain the risk field range, extracts the relevant factor features within the risk field range, and substitutes the extracted relevant factor features into the trained XGBoost model to obtain the real-time security level.

[0052] The technical solution of the present invention also provides a quantization system for the safety degree of the pedestrian crossing scenario on a road section. The system includes:

[0053] A range determination module for determining the risk field range of pedestrians in the crossing scenario;

[0054] A factor determination module is used to obtain traffic accident cases and motor vehicle driving scenario data, and determine the main factors affecting pedestrian crossing safety within the risk field. Among them, the main factors include vehicle speed, operation margin, pedestrian initial speed, pedestrian acceleration, pedestrian type, pedestrian line of sight, weather conditions, and other vehicles. The operation margin is the time representing the duration allowed for the driver to operate.

[0055] A model training module is used to extract relevant factor features from traffic accident cases and motor vehicle driving scenario data based on the main factors, establish a sample data set, train an XGBoost model for classifying safety levels, and set the safety levels.

[0056] A model application module is used to identify a target vehicle based on the trained XGBoost model to obtain the safety level.

[0057] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0058] 1. More comprehensive risk assessment factors:

[0059] When evaluating the safety of pedestrian crossing, the present invention takes into account a rich variety of factors. It not only includes traditional factors such as vehicle speed, pedestrian speed, and driver reaction time, but also deeply incorporates pedestrian type, pedestrian psychological factors (such as pedestrian line of sight), road flatness, lane markings, weather conditions, and other vehicles. By comprehensively considering various factors such as vehicles, roads, and the environment, the safety of pedestrian crossing can be evaluated more accurately.

[0060] 2. Provide decision support for the autonomous driving system:

[0061] Autonomous driving vehicles can obtain the risk field range and relevant factor data through on-vehicle radar, input them into the trained XGBoost model, and obtain the safety level in real time. Then, reasonable decisions can be made based on the safety level, such as decelerating, avoiding, or continuing to drive, etc. This helps to improve the safety and reliability of the autonomous driving system, reduce traffic accidents, and is of great significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0063] Figure 1 It is the first flow chart of the method for quantifying the safety of the pedestrian crossing scenario of the road section.

[0064] Figure 2 It is the second flow chart of the method for quantifying the safety of the pedestrian crossing scenario of the road section. Detailed implementation manners

[0065] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0066] Please refer to Figures 1 to 2 , in an embodiment of the present invention, a method for quantifying the safety degree of a pedestrian crossing scenario on a road section, the method includes:

[0067] Step 1, determine the risk field range of pedestrians in the crossing scenario;

[0068] Assume that an autonomous driving target vehicle is traveling on a certain road section at a constant speed V 车 When it is found that there are pedestrians passing through the road section ahead, use the on-vehicle radar on the autonomous driving target vehicle to determine the initial speed v0 and acceleration a of the pedestrians (assuming that the pedestrians are moving in a uniform straight line) 行人 , and obtain the function of time change as v p (t)=v0 + a 行人 ·t. Establish a plane rectangular coordinate system with the center point of the vehicle itself as the origin, where the y-axis is perpendicular to the road driving direction and the x-axis is the vehicle driving direction. The on-vehicle radar measures the initial distance between the vehicle and the pedestrians in the direction perpendicular to the road (i.e., the distance that the pedestrians need to cross horizontally) d y , and the initial distance d from the vehicle along the road driving direction to the conflict point x , specifically, it is carried out according to the following process:

[0069] The steps of determining the risk field range of pedestrians in the crossing scenario include:

[0070] Obtain the vehicle speed;

[0071] Detect pedestrians based on the on-vehicle radar, and obtain the initial speed and acceleration of the pedestrians;

[0072] Create a plane rectangular coordinate system based on the vehicle, and obtain the pedestrian coordinates based on the on-vehicle radar and fill them into the plane rectangular coordinate system; wherein, the plane rectangular coordinate system has the road cross-section direction as the y-axis and the vehicle driving direction as the x-axis;

[0073] Determine the risk field range in the plane rectangular coordinate system;

[0074] The steps of determining the risk field range in the plane rectangular coordinate system include:

[0075] Determine the time for the pedestrians to reach the vehicle driving path according to the initial speed, acceleration and pedestrian coordinates of the pedestrians, and use it as the first time;

[0076] Calculate the driving distance of the vehicle within the first period of time;

[0077] Calculate the remaining distance from the vehicle to the conflict point based on the driving distance;

[0078] Calculate the operation margin based on the remaining distance;

[0079] Obtain the braking acceleration of the vehicle, calculate the braking distance according to the braking acceleration, and calculate the risk field radius according to the braking distance, operation margin and preset reaction time;

[0080] Determine the risk field range according to the risk field radius.

[0081] The specific description of the above content is as follows:

[0082] Step 1.1: Calculate the time t for the pedestrian to reach the vehicle's driving path p (p is the point where the vehicle and the pedestrian conflict):

[0083] From That is It can be obtained that It can be solved that (where t p Take the positive value and discard the negative value);

[0084] Step 1.2: Calculate the distance traveled by the vehicle within the time period t p :

[0085] Since the vehicle travels at a constant speed V 车 Traveling, From Step 1 and Step 1.1, it can be known that V 车 , t p . It can be obtained that d 车 =V 车 ×t p ;

[0086] Step 1.3: Calculate the remaining distance from the vehicle to the conflict point: d 剩余 =d x -d 车 ;

[0087] Step 1.4: Calculate TTC: The TTC is also called the operation margin;

[0088] At this time, because the pedestrian only has a lateral speed, when calculating the TTC, only the remaining distance d 剩余 along the driving direction of the road of the vehicle and the speed V 车 of the vehicle need to be concerned, so

[0089] Step 1.5: Determine the scope of the risk field:

[0090] Step 1.5.1: Consider the driver's reaction time as t r , when the vehicle detects a pedestrian ahead and starts braking and decelerating, the magnitude of the acceleration is a 制动 , and the direction of the acceleration is opposite to the direction of motion of the autonomous driving target vehicle. Therefore Therefore, D = V 车 ·TTC - V·t r + d b ;

[0091] In the formula, d b is the driving distance of the autonomous driving target vehicle from the start of braking to a complete stop, in meters. a 制动 is the acceleration when the vehicle detects a pedestrian ahead and starts braking and decelerating, in m / s 2 , and D is the distance from the final stopping position of the vehicle to the conflict point, in meters;

[0092] Step 1.5.2: Risk field range. Establish a circle with the center point of the vehicle itself and a radius of D to obtain the risk field range.

[0093] Step 2: Obtain traffic accident cases and motor vehicle driving scenario data, and determine the main factors affecting pedestrian crossing safety within the risk field range; among them, the main factors include vehicle speed, operation margin, pedestrian initial speed, pedestrian acceleration, pedestrian type, pedestrian line of sight, weather conditions, and other vehicles; the operation margin is time, representing the duration allowed for the driver to operate;

[0094] The steps of obtaining traffic accident cases and motor vehicle driving scenario data and determining the main factors affecting pedestrian crossing safety within the risk field range include:

[0095] According to traffic accident cases where motor vehicles have collided with pedestrians during road section driving and road section motor vehicle driving scenario data, use the Bayesian algorithm to extract relevant static factors that lead to traffic accidents; the relevant static factors include: road roughness, lane markings, road width, and whether there is a central median;

[0096] Within the established risk field range, use on-vehicle radar to find dynamic factors related to traffic accidents; the dynamic factors include vehicle speed, operation margin, pedestrian initial speed, pedestrian acceleration, pedestrian type, pedestrian line of sight, weather conditions, and other vehicles.

[0097] The specific description of Step 2 is as follows:

[0098] Step 2.1: According to traffic accident cases where motor vehicles have collided with pedestrians during road section driving and road section motor vehicle driving scenario data, use the Bayesian algorithm to extract relevant static factors that lead to traffic accidents, specifically including: road roughness, lane markings, road width, whether there is a central median, and record them;

[0099] Step 2.2: Within the established risk field range, use on-vehicle radar to find dynamic factors related to traffic accidents, specifically including: vehicle speed, TTC, pedestrian initial speed, pedestrian acceleration, pedestrian type, pedestrian line of sight, weather conditions, and the number of other vehicles.

[0100] Step 3: Extract relevant factor features from traffic accident cases and motor vehicle driving scenario data based on the main factors, establish a sample data set, train an XGBoost model for classifying safety levels, and set safety levels.

[0101] The steps of extracting relevant factor features from traffic accident cases and motor vehicle driving scenario data based on the main factors, establishing a sample data set, training an XGBoost model for classifying safety levels, and setting safety levels include:

[0102] Extract relevant factor features from traffic accident cases and motor vehicle driving scenario data based on the main factors, and establish a sample data set;

[0103] Determine and quantify the relative weights of each factor within the risk field range in the sample data set;

[0104] Divide the data set, divide the established standardized sample data set, randomly select 30% of the sample data in the sample data set as the test set, and 70% of the sample data as the training set;

[0105] Train an XGBoost model for classifying safety levels based on the training set and the test set, and set safety levels;

[0106] The steps of determining and quantifying the relative weights of each factor within the risk field range in the sample data set include:

[0107] Perform standardization processing on the original data in the sample data set;

[0108] Calculate the mean and standard deviation of each index;

[0109] Calculate the coefficient of variation of each index, and calculate the weight of each index based on the coefficient of variation;

[0110] Calculate the overall safety index, and perform standardization processing on the overall safety index.

[0111] The steps of training an XGBoost model for classifying safety levels based on the training set and the test set, and setting safety levels include:

[0112] Initialize the XGBoost model;

[0113] Perform H iterations on the initialized XGBoost model;

[0114] After H iterations, the model reaches the maximum number of iterations and stops iterating, finally obtaining the overall safety index model based on the XGBoost algorithm;

[0115] Set the safety level;

[0116] The steps of performing H iterations on the initialized XGBoost model include:

[0117] Calculate the residuals in the H-th iteration process;

[0118] Generate the weak training learner for the H-th iteration;

[0119] Calculate the best fit value;

[0120] Update the simulated prediction value.

[0121] The steps of setting the safety level include:

[0122] Read the overall safety index under the trained XGBoost algorithm and sort it in ascending order;

[0123] Use the trained XGBoost model to set the safety level and calculate the positions of 85% and 15% of the overall safety index;

[0124] Read the corresponding values in the sorted overall safety index according to the positions of 85% and 15% of the overall safety index. The value corresponding to 85% of the overall safety index is the first value, and the value corresponding to 15% of the overall safety index is the second value;

[0125] When the overall safety index is less than the second value, the safety level is unsafe. When the overall safety index is between the second value and the first value, the safety level is generally safe. When the overall safety index is greater than the first value, the safety level is very safe.

[0126] Step 3 is the training process of the model, which is specifically described as follows:

[0127] Step 3.1: According to the traffic accident cases of motor vehicles colliding with pedestrians when driving on the road section, the motor vehicle driving scenario data on the road section, and each factor within the established risk field in step 2, extract the relevant factor characteristics leading to traffic accidents and establish a sample data set;

[0128] Step 3.2: Determine and quantify the relative weights of each factor within the risk field in the sample data set. Using the information entropy weight method, by establishing the original data matrix of relevant factors, that is, there are n evaluation indicators and the number of samples is l, so establish the original matrix Y=(y ij ) l×n , where y ijDenote the sample of the \(i\)-th row and the index of the \(j\)-th column, where \(j = 1, 2, \ldots, n\);

[0129] Step 3.2.1: Standardize the original data where is the maximum value of the \(j\)-th column index, is the minimum value of the \(j\)-th column index. When all the data in the \(j\)-th column are the same, this column of data can be uniformly assigned a value of 0, indicating no relative change. The same applies to the data in other columns, where \(j = 1, 2, \ldots, n\). Finally, the standardized matrix \(Q=(x ij ) l×n ;

[0130] Step 3.2.2: Calculate the mean of each index, where \(j = 1, 2, \ldots, n\);

[0131] Step 3.2.3: Calculate the standard deviation of each index, where \(j = 1, 2, \ldots, n\);

[0132] Step 3.2.4: Calculate the coefficient of variation of each index, where \(CV j is the coefficient of variation of the \(j\)-th type of index. When all the data in the \(j\)-th column are the same, it can be considered that there is no dispersion in this column of data. Therefore, the coefficient of variation is uniformly considered to be 0, where \(j = 1, 2, \ldots, n\);

[0133] Step 3.2.5: Calculate the weight of each index, where \(CV k is the coefficient of variation of the \(k\)-th type of index, \(j = 1, 2, \ldots, n\);

[0134] Step 3.2.6: Calculate the overall safety index, where \(Z i is the overall safety index of the \(i\)-th row sample, \(w j is the weight of the \(j\)-th column index, \(i = 1, 2, \ldots, l\), \(j = 1, 2, \ldots, n\);

[0135] Step 3.2.7: Standardize the overall safety index where \(i = 1, 2, \ldots, l\);

[0136] Step 3.3: Divide the dataset. Divide the established standardized sample dataset, randomly select 30% of the sample dataset as the test set, and 70% of the sample data as the training set;

[0137] Step 3.4: Initialize the XGBoost model, that is:

[0138] where is the initial predicted value of the standardized overall safety index for the q-th row of samples, S q is the initial value of the standardized overall safety index for the q-th row of samples, q = 1, 2, …, N, where N is the number of training set samples, is the mean of the sum of the standardized overall safety indices of the training set samples;

[0139] Step 3.5: Construct H iterations (h = 1, 2, …, H), and loop to execute Step 3.5.1, Step 3.5.2, Step 3.5.3, and Step 3.5.4;

[0140] Step 3.5.1: Calculate the residual γ of the h-th iteration qh ;

[0141] where γ qh is the residual of the h-th iteration of the overall safety index of the q-th sample in the training set samples, q = 1, 2, …, …, h = 1, 2, …, H;

[0142] Step 3.5.2: Obtain the weak training learner T of the h-th iteration h (x q ), where q = 1, 2, …, N, h = 1, 2, …, H, and x q is the feature vector of the q-th sample in the training set. The value of the weak training learner T h (x q ) is determined by the data characteristics and practical significance of different indicators; the weak training learner T h (x q ) is a specific application of the extreme gradient boosting algorithm. On the decision tree, finding the optimal tree structure and leaf nodes is to find the optimal solution most similar to the objective function. In this process, this step is equivalent to a model. Since the tree shape and leaf partitioning method are different, this model is different, so only this letter can be used to represent this model.

[0143] Step 3.5.3: Calculate the best fit value;

[0144] where γ h represents the goodness of fit of the residual in the h-th iteration, where i = 1, 2, …, l, j = 1, 2, …, n, h = 1, 2, …, H;

[0145] Step 3.5.4: Update the model predicted value;

[0146] where η is the learning rate, γ hrepresents the goodness of fit of the residual in the h-th iteration, where h = 1, 2, …, H and q = 1, 2, …, N;

[0147] Step 3.6: After H iterations, the model reaches the maximum number of iterations and stops iterating, finally obtaining the overall safety index model based on the XGBoost algorithm;

[0148] The overall safety index model under the XGBoost algorithm is where indicates that the final predicted overall safety index is the sum of the initial predicted value and the correction amount of the prediction by the weak learner in each round of iteration, represents the initial predicted value of the overall safety index of the q-th row sample, η is the learning rate, and γ h is the goodness of fit of the residual in the h-th iteration, and T h (x q ) is the weak learner in the h-th iteration, where h = 1, 2, …, H and q = 1, 2, …, N;

[0149] Step 3.7: Safety level setting:

[0150] Step 3.7.1: According to the overall safety index under the trained XGBoost algorithm sort them in ascending order (when the values of the overall safety index are the same, still sort according to the above requirements), where q = 1, 2, …, N; among them, q is equivalent to the i-th sample in the overall sample.

[0151] Step 3.7.2: Use the trained XGBoost model to set the safety level, calculate the positions of 85% and 15% of the overall safety index (round down each obtained position), the position of 85% is f1 = [0.85 × N], and the position of 15% is f2 = [0.15 × N];

[0152] Step 3.7.3: According to the calculated positions of 85% and 15%, find the values corresponding to the 85% and 15% positions when the overall safety index is sorted in ascending order;

[0153] Step 3.7.4: Establish the safety level. When the overall safety index is less than or equal to the value corresponding to f2, it is considered unsafe; when the overall safety index is greater than the value corresponding to f2 and less than or equal to the value corresponding to f1, it is considered generally safe; when the overall safety index is greater than the value corresponding to f1, it is considered very safe.

[0154] Step 4: Identify the target vehicle based on the trained XGBoost model to obtain the safety level.

[0155] The steps of identifying the target vehicle based on the trained XGBoost model to obtain the safety level include:

[0156] Taking the safety level set by the trained model as the standard, when the vehicle is driving, the on-vehicle radar is used to obtain the risk field range, relevant factor features are extracted within the risk field range, and the extracted relevant factor features are substituted into the trained XGBoost model to obtain the real-time safety level.

[0157] The description of step 4 is as follows:

[0158] For safety level identification, taking the safety level set by the trained model as the standard, when the target vehicle uses the on-vehicle radar to obtain the risk field range during driving and substitutes the data of each factor within the extracted risk field range into the trained XGBoost model, the real-time safety level can be obtained.

[0159] As a preferred embodiment of the technical solution of the present invention, a quantification system for the safety degree of the pedestrian crossing scenario of a road section is further provided. The system includes:

[0160] A range determination module for determining the risk field range of pedestrians in the crossing scenario;

[0161] A factor determination module for obtaining traffic accident cases and motor vehicle driving scenario data and determining the main factors affecting pedestrian crossing safety within the risk field range; wherein, the main factors include vehicle speed, operation margin, pedestrian initial speed, pedestrian acceleration, pedestrian type, pedestrian line of sight, weather conditions, and other vehicles; the operation margin is time, representing the duration allowed for the driver to operate;

[0162] A model training module for extracting relevant factor features from traffic accident cases and motor vehicle driving scenario data based on the main factors, establishing a sample data set, training an XGBoost model for dividing the safety level, and setting the safety level;

[0163] A model application module for identifying the target vehicle based on the trained XGBoost model to obtain the safety level.

[0164] In an example of the technical solution of the present invention, a specific example is provided as follows:

[0165] Step 1. Establishment of the risk field range: The autonomous driving target vehicle travels at a constant speed on a certain road section. It is found that there are pedestrians passing through the road section ahead. The on-vehicle radar on the autonomous driving target vehicle is used to determine the initial speed acceleration of the pedestrians (assuming the pedestrians are moving in a uniform linear motion), and the function of the time change is obtained as v py(t) = 2 + 0.2t. A rectangular coordinate system is established based on the dynamics of the vehicle itself, with the road cross-section direction as the y-axis and the vehicle driving direction as the x-axis. The on-vehicle radar measures the initial distance between the vehicle and the pedestrian in the direction perpendicular to the road (i.e., the distance the pedestrian needs to cross horizontally) d y = 15m, and the initial distance d x of the vehicle along the road driving direction is 180m. Specifically, it is carried out according to the following process:

[0166] Step 1.1: Calculate the time t for the pedestrian to reach the vehicle driving path p (p is the point where the vehicle and the pedestrian conflict): From That is: It can be obtained that It can be solved that That is, t p = 5.36s (where t p takes the positive value and the negative value is discarded).

[0167] Step 1.2: Calculate the distance traveled by the vehicle during the t p time period: Since the vehicle travels at a constant speed V 车 travels, From Step 1 and Step 1.1, it can be known that t p = 5.36s. It can be obtained that d 车 = 30×5.36 = 160.8m.

[0168] In the formula: V 车 is the constant speed of the autonomous driving target vehicle, unit: m / s, t p is the time for the pedestrian to reach the vehicle driving path (p is the point where the vehicle and the pedestrian conflict), unit: s, d 车 is the distance traveled by the vehicle during the t p time period, unit: m.

[0169] Step 1.3: Calculate the remaining distance from the vehicle to the conflict point: d 剩余 = d x - d 车 , that is, d 剩余 = 180 - 160.8 = 19.2m.

[0170] Step 1.4: Calculate TTC: At this time, since the pedestrian only has a horizontal speed, when calculating TTC, only the remaining distance d 剩余 along the road driving direction of the vehicle and the speed V 车 of the vehicle need to be concerned. Therefore:

[0171] Step 1.5. Determine the scope of the risk field:

[0172] Step 1.5.1. Consider the driver's reaction time as t r = 1 s. When the vehicle detects a pedestrian ahead and starts to brake and decelerate, the magnitude of the acceleration is The direction of the acceleration is opposite to the direction of motion of the autonomous driving target vehicle.

[0173]

[0174] Therefore, D = V 车 ·TTC - V 车 ·t r + d b = 30×0.64 - 30×1 + 90 = 79.2 m.

[0175] In the formula: d b is the driving distance of the autonomous driving target vehicle from the start of braking to a complete stop, unit: m, a 制动 is the acceleration of the vehicle when it detects a pedestrian ahead and starts to brake and decelerate, unit: m / s 2 , D is the distance from the final stopping position of the vehicle to the conflict point;

[0176] Step 1.5.2. For the scope of the risk field in Step 1.5.2, establish a circle with the center point of the vehicle itself and a radius of D = 79.2 m to obtain the scope of the risk field;

[0177] Step 2. Identify the main factors affecting the safety of pedestrians crossing the street within the risk field:

[0178] Step 2.1. According to the cases of traffic accidents between motor vehicles and pedestrians during road section driving and the road section motor vehicle driving scenario data, use the Bayesian algorithm to extract the relevant static factors leading to traffic accidents, specifically including: road flatness, lane markings, road width, whether there is a central median strip, and record them. Use road section crossing monitoring and radar to collect the road flatness of the road in the above Step 1, assume it is good, coded as 0, the lane markings are clear, coded as 0, assume the road width is 8.5 m, assume there is no central median strip, coded as 1;

[0179] Step 2.2. Within the established scope of the risk field, use on-vehicle radar to find the dynamic factors related to traffic accidents, specifically: vehicle speed V = 30 m / s, TTC = 0.64 s, pedestrian initial speed v0 = 2 m / s, pedestrian acceleration a 行人 = 0.2 m / s 2 , the pedestrian type is an ordinary adult, and assume it is coded as 1, the pedestrian's line of sight is normal and observed, and assume it is coded as 0, weather conditions, sunny coded as 0, the number of other vehicles is 0;

[0180] Step 3: Construct an XGBoost model for dividing safety levels and set safety levels:

[0181] Step 3.1: According to the traffic accident cases between motor vehicles and pedestrians during driving on sections, the pedestrian crossing scene data on sections, and various factors within the established risk field in Step 2, extract the relevant factor characteristics leading to traffic accidents and establish a sample data set; assume the specific coding of the extracted data is as follows: Pedestrian type: Assume that ordinary adults are coded as 1, children are coded as 2, and the elderly are coded as 3; Pedestrian's line of sight is observing the surroundings normally and coded as 0, and the distracted state can be coded as 1; Road flatness: Assume good is coded as 0, bumpy is coded as 1; Lane markings: Clearly visible is coded as 0, blurred is coded as 1; Weather conditions: Sunny is coded as 0, rainy is coded as 1, snowy is coded as 2, foggy is coded as 3, Median strip: With is coded as 0, without is coded as 1, and the remaining variables can be obtained according to the measurement, see the original matrix for details;

[0182] Step 3.2: Determine and quantify the relative weights of various factors within the risk field in the sample data set. Adopt the information entropy weight method. By establishing the original data matrix of relevant factors, that is, assume that the extracted data has n = 12 evaluation indicators and the number of samples is l = 10, and use the extracted data to establish the original matrix as: Where each column from left to right is vehicle speed, TTC, pedestrian initial speed, pedestrian acceleration, pedestrian type, pedestrian line of sight, road flatness, lane markings, weather conditions, whether there is a median strip, number of other vehicles, and road width in turn;

[0184] Step 3.2.1: Standardize the original data Where is the maximum value of the j-th column index, is the minimum value of the j-th column index. When all the data in the 7th column index (road flatness) are the same, this column of data can be uniformly assigned a value of 0, indicating no relative change. Finally, the standardized matrix is Where each column from left to right is vehicle speed, TTC, pedestrian initial speed, pedestrian acceleration, pedestrian type, pedestrian line of sight, road flatness, lane markings, weather conditions, whether there is a median strip, number of other vehicles, and road width in turn;

[0185] Step 3.2.2: Calculate the mean value of each index. Taking vehicle speed as an example, The same applies to the remaining indicators. Here, the mean values of other indicators are calculated as follows: TTC is The initial speed of the pedestrian is x3 = 0.449, the pedestrian acceleration is x4 = 0.394, the pedestrian type is x5 = 0.350, the pedestrian's line of sight is x6 = 0.500, the road roughness is x7 = 0, the lane markings are x8 = 0.300, the weather condition is x9 = 0.433, whether there is a median strip is x 10 = 0.500, the number of other vehicles is x 11 = 0.500, the road width is x 12 = 0.488;

[0186] Step 3.2.3, calculate the standard deviation of each index. Taking the vehicle speed as an example, The standard deviations of the remaining indexes can be obtained in the same way. Here, the standard deviations of the other indexes are: for TTC, σ2 = 0.329; for the initial speed of the pedestrian, σ3 = 0.344; for the pedestrian acceleration, σ4 = 0.355; for the pedestrian type, σ5 = 0.412; for the pedestrian's line of sight, σ6 = 0.527; for the road roughness, σ7 = 0; for the lane markings, σ8 = 0.483; for the weather condition, σ9 = 0.387; for whether there is a median strip, σ 10 = 0.527, for the number of other vehicles, σ 11 = 0.527, for the road width, σ 12 = 0.284;

[0188] Step 3.2.4, calculate the coefficient of variation of each index. Taking the vehicle speed as an example, then The coefficients of variation of the remaining indexes can be obtained in the same way. Here, the coefficients of variation of the other indexes are: for TTC, CV2 = 0.731; for the initial speed of the pedestrian, CV3 = 0.766; for the pedestrian acceleration, CV4 = 0.901; for the pedestrian type, CV5 = 1.177; for the pedestrian's line of sight, CV6 = 1.054. Since all the data in the 7th column index (road roughness) are the same, it can be considered that there is no dispersion degree in this column of data, so it is considered that the coefficient of variation CV7 = 0 at this time; for the lane markings, CV8 = 1.610; for the weather condition, CV9 = 0.894; for whether there is a median strip, CV 10 = 1.054, for the number of other vehicles, CV 11 = 1.054, for the road width, CV 12 = 0.582;

[0189] Step 3.2.5, calculate the weight of each index. Taking the vehicle speed as an example, it can be known that Then the weight of the vehicle speed index is The same applies to the remaining indicators. Here, the weights for calculating other indicators are as follows: for TTC, w2 = 0.070; for the initial speed of pedestrians, w3 = 0.073; for the acceleration of pedestrians, w4 = 0.086; for the type of pedestrians, w5 = 0.112; for the line of sight of pedestrians, w6 = 0.100; for the road roughness, w7 = 0; for the lane markings, w8 = 0.153; for the weather conditions, w9 = 0.085; for whether there is a median strip, w 10 = 0.100; for the number of other vehicles, w 11 = 0.100; for the road width, w 12 = 0.055;

[0191] Step 3.2.6, Calculate the overall safety index. Taking the first sample as an example,

[0192] The same applies to the remaining indicators. Here, the overall safety indices for other samples are calculated as follows: for the second sample, Z2 = 0.435; for the third sample, Z3 = 0.346; for the fourth sample, Z4 = 0.765; for the fifth sample, Z5 = 0.151; for the sixth sample, Z6 = 0.643; for the seventh sample, Z7 = 0.221; for the eighth sample, Z8 = 0.691; for the ninth sample, S9 = 0.593; for the tenth sample, Z 10 = 0.188;

[0193] Step 3.2.7, Standardize the overall safety index. Taking the overall safety index of the first sample as an example, the standardized overall safety index of the first sample is:

[0194]

[0195] The standardized data for the remaining samples are respectively S2 = 0.463, S3 = 0.318, S4 = 1.000, S5 = 0, S6 = 0.801, v7 = 0.114, S8 = 0.879, S9 = 0.720, S 10 = 0.060;

[0196] Step 3.3, Divide the dataset. Divide the established standardized sample dataset. Randomly select 30% of the sample dataset as the test set and 70% of the sample data as the training set. Since it is assumed that there are 10 data, it is assumed that the first, fourth, and seventh samples are randomly selected as the test set, and the remaining seven samples are used as the training set;

[0197] Step 3.4, Initialize the XGBoost model, that is: assume that the overall safety indices of the seven samples in the training set are 0.463, 0.318, 0, 0.801, 0.879, 0.720, 0.060 respectively, then Then the first sample in the training set The second sample The third sample The fourth sample The fifth sample The sixth sample The seventh sample

[0198] Step 3.5: Construct H iterations (h = 1, 2, …, H). Here, assume that 3 iterations are constructed, so h = 1, 2, 3. Each iteration goes through Step 3.5.1, Step 3.5.2, Step 3.5.3, and Step 3.5.4. Here, take the first iteration of the first sample in the training set as an example;

[0199] Step 3.5.1: Calculate the residual γ of the h = 1st iteration 11 , where γ 11 is the residual of the first iteration of the overall safety index of the first sample in the training set. Similarly, the residuals of the first iteration of the other samples in the training set are: for the second sample in the training set, γ 21 = -0.145; for the third sample in the training set, γ 31 = -0.463; for the fourth sample in the training set, γ 41 = 0.338; for the fifth sample in the training set, γ 51 = 0.416; for the sixth sample in the training set, γ 61 = 0.257; for the seventh sample in the training set, γ 71 = -0.403.

[0200] Step 3.5.2: Obtain the weak training learner T1(x1) of the h = 1st iteration. Assume that T1(x1) = 0.3 is the weak learner of the first iteration of the first sample in the training set, the weak learner of the first iteration of the second sample in the training set is T1(x2) = -0.3, the weak learner of the first iteration of the third sample in the training set is T1(x3) = 0.3, the weak learner of the first iteration of the fourth sample in the training set is T1(x4) = -0.3, the weak learner of the first iteration of the fifth sample in the training set is T1(x5) = -0.3, the weak learner of the first iteration of the sixth sample in the training set is T1(x6) = 0.3, and the weak learner of the first iteration of the seventh sample in the training set is T1(x7) = -0.3.

[0201] Step 3.5.3: Calculate the best fit value;

[0202] Assume that here, taking the first iteration as an example, from Step 3.5.1 and Step 3.5.2, the value of the first iteration is:

[0203] Step 3.5.4 Update the model prediction value;

[0204] Taking the first sample of the training set in the first iteration to update the model prediction value as an example, assuming the learning rate η = 0.1 and the best fit value of the residual γ1 = -0.197 in the first iteration, then Similarly, for the second sample of the training set in the first iteration, the updated model prediction value is For the third sample of the training set in the first iteration, the updated model prediction value is For the fourth sample of the training set in the first iteration, the updated model prediction value is For the fifth sample of the training set in the first iteration, the updated model prediction value is For the sixth sample of the training set in the first iteration, the updated model prediction value is For the seventh sample of the training set in the first iteration, the updated model prediction value is

[0205] Step 3.6. Assume that after 3 iterations, the model reaches the maximum number of iterations and stops iterating, and finally obtains the overall safety index model based on the XGBoost algorithm;

[0206] Taking the first sample of the training set as an example, assuming the initial prediction value of the overall safety index of the sample in the i = 1 row The adjustment values for each iteration of the first sample are: -0.006, -0.005, -0.005 respectively. Then the final predicted overall safety index is the sum of the initial prediction value and the correction amount of the prediction by the weak learner in each round of iteration, which is Similarly, for the second sample, the adjustment values for each iteration are 0.006, 0.005, 0.005 respectively. Then the final predicted overall safety index is the sum of the initial prediction value and the correction amount of the prediction by the weak learner in each round of iteration, which is For the third sample, the adjustment values for each iteration are -0.006, -0.005, -0.005 respectively. Then the final predicted overall safety index is the sum of the initial prediction value and the correction amount of the prediction by the weak learner in each round of iteration, which is For the fourth sample, the adjustment values for each iteration are 0.006, 0.005, 0.005 respectively. Then the final predicted overall safety index is the sum of the initial prediction value and the correction amount of the prediction by the weak learner in each round of iteration, which is For the fifth sample, the adjustment values for each iteration are 0.006, 0.005, 0.005 respectively. Then the final predicted overall safety index is the sum of the initial prediction value and the correction amount of the prediction by the weak learner in each round of iteration, which is The adjustment values for the sixth sample in each iteration are -0.006, -0.005, -0.005 respectively. Then the final predicted overall safety index is the sum of the initial predicted value and the correction amounts of the weak learners in each round of iteration, which is The adjustment values for the seventh sample in each iteration are 0.006, 0.005, 0.005 respectively. Then the final predicted overall safety index is the sum of the initial predicted value and the correction amounts of the weak learners in each round of iteration, which is

[0207] Step 3.7 Safety level setting:

[0208] Step 3.7.1. According to the overall safety index under the trained XGBoost algorithm Sort them in ascending order (when the values of the overall safety index are the same, still sort according to the above requirements), then we can get 0.447, 0.447, 0.447, 0.479, 0.479, 0.479, 0.479;

[0209] Step 3.7.2. Use the trained XGBoost model to set the safety level, and calculate the positions of 85% and 15% of the overall safety index (round down each obtained position). The position of 85% is f1 = [0.85×7] = 5, and the position of 15% is f2 = [0.15×7] = 1;

[0210] Step 3.7.3. According to the calculated positions of 85% and 15%, find the overall safety index Sorted in ascending order, the value corresponding to the 85% position is 0.479 and the value corresponding to 15% is 0.447;

[0211] Step 3.7.4. Safety level setting. When the overall safety index is less than or equal to the value corresponding to f2, that is at this time, it is considered very unsafe; when the overall safety index is greater than the value corresponding to f2 and less than or equal to the value corresponding to f1, that is at this time, it is considered generally safe; when the overall safety index is greater than the value corresponding to f1, that is at this time, it is considered very safe;

[0212] Step 4. Safety level identification. Assume that first, according to Step 1.5.2, the risk field range is obtained by using the on-vehicle radar during the driving of the target vehicle, and then the data of each factor within the risk field range extracted in Step 2 are substituted into the trained XGBoost model. The obtained safety index is 0.447 < 0.451 ≤ 0.479, that is, the overall safety index is greater than the value corresponding to f1 and less than or equal to the value corresponding to f2. Therefore, it is considered that the pedestrian crossing is in a generally safe state.

[0213] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for quantifying the safety of pedestrian crossing scenes on a road section, characterized in that: The method comprises: Determine the risk range of pedestrians in street crossing scenarios; Obtain traffic accident cases and motor vehicle driving scenario data to determine the main factors affecting pedestrian crossing safety within the risk field; wherein the main factors include vehicle speed, operating margin, pedestrian initial speed, pedestrian acceleration, pedestrian type, pedestrian sight, weather conditions and other vehicles; the operating margin is time, indicating the length of time the driver is allowed to operate; Based on the main factors, relevant factor features are extracted from traffic accident cases and motor vehicle driving scene data, a sample data set is established, an XGBoost model for dividing safety levels is trained, and safety levels are set; The target vehicle is identified based on the trained XGBoost model to obtain the safety level.

2. The method for quantifying the safety of pedestrian crossing scenes on a road section according to claim 1, characterized in that: The step of determining the risk field range of pedestrians in the street crossing scene includes: Get vehicle speed; Detect pedestrians based on vehicle-mounted radar and obtain the initial speed and acceleration of pedestrians; A plane rectangular coordinate system is created based on the vehicle, and the coordinates of the pedestrian are obtained based on the vehicle-mounted radar and filled into the plane rectangular coordinate system; wherein the plane rectangular coordinate system takes the road cross-section direction as the y-axis and the vehicle travel direction as the x-axis; Determine the risk field range in the plane rectangular coordinate system.

3. The method for quantifying the safety of pedestrian crossing scenes on a road section according to claim 2, characterized in that: The step of determining the risk field range in the plane rectangular coordinate system comprises: Determine the time when the pedestrian arrives at the vehicle's travel path according to the pedestrian's initial speed, acceleration, and pedestrian coordinates as the first time; Calculate the vehicle's travel distance in the first time; Calculate the remaining distance from the vehicle to the conflict point based on the travel distance; Calculate the operating margin based on the remaining distance; Obtaining the braking acceleration of the vehicle, calculating the braking distance according to the braking acceleration, and calculating the risk field radius according to the braking distance, the operating margin and the preset reaction time; The risk field range is determined according to the risk field radius.

4. The method for quantifying the safety of pedestrian crossing scenes on a road section according to claim 1, characterized in that: The steps of obtaining traffic accident cases and motor vehicle driving scene data and determining the main factors affecting pedestrian crossing safety within the risk field include: Based on the existing traffic accident cases between motor vehicles and pedestrians and the driving scene data of motor vehicles on the road section, the Bayesian algorithm is used to extract the relevant static factors that lead to traffic accidents; the relevant static factors include: road flatness, lane markings, road width and whether there is a central dividing strip; Within the established risk field, the vehicle-mounted radar is used to find dynamic factors related to traffic accidents; the dynamic factors include vehicle speed, operating margin, pedestrian initial speed, pedestrian acceleration, pedestrian type, pedestrian line of sight, weather conditions and other vehicles.

5. The method for quantifying the safety of pedestrian crossing scenes on a road section according to claim 1, characterized in that: The steps of extracting relevant factor features from traffic accident cases and motor vehicle driving scene data based on the main factors, establishing a sample data set, training an XGBoost model for dividing safety levels, and setting safety levels include: Based on the main factors, relevant factor features are extracted from traffic accident cases and motor vehicle driving scene data to establish a sample data set; Establish and quantify the relative weights of various factors within the risk field in the sample data set; Divide the data set, divide the established standardized sample data set, randomly select 30% of the sample data in the sample data set as the test set, and 70% of the sample data as the training set; Based on the training set and the test set, the XGBoost model for classifying security levels is trained and the security level is set.

6. The method for quantifying the safety of pedestrian crossing scenes on a road section according to claim 5, characterized in that: The step of establishing and quantifying the relative weights of various factors within the risk field range in the sample data set includes: Standardize the raw data in the sample data set; Calculate the mean and standard deviation of each indicator; Calculate the coefficient of variation of each indicator, and calculate the weight of each indicator based on the coefficient of variation; Calculate the overall safety index and standardize the overall safety index.

7. The method for quantifying the safety of pedestrian crossing scenes on a road section according to claim 5, characterized in that: The steps of training the XGBoost model for dividing the security level based on the training set and the test set and setting the security level include: Initialize the XGBoost model; Iterate the initialized XGBoost model H times; After H iterations, the model reaches the maximum number of iterations and stops iterating, and finally obtains the overall safety index model based on the XGBoost algorithm; Set security level; The step of iterating the initialized XGBoost model H times includes: Calculate the residual of the Hth iteration process; Generate the Hth iterative weak training learner; Calculate the best fit value; Update simulation predictions.

8. The method for quantifying the safety of pedestrian crossing scenes on a road section according to claim 7, characterized in that: The steps of setting the security level include: Read the overall safety index of the trained XGBoost algorithm and sort it in ascending order; Use the trained XGBoost model to set the safety level and calculate the 85% and 15% positions of the overall safety index; According to the positions of the overall safety index 85% and 15%, the corresponding values ​​are read from the sorted overall safety index, the value corresponding to the overall safety index 85% is the first value, and the value corresponding to the overall safety index 15% is the second value; When the overall safety index is less than the second value, the safety level is unsafe; when the overall safety index is between the second value and the first value, the safety level is generally safe; when the overall safety index is greater than the first value, the safety level is very safe.

9. The method for quantifying the safety of pedestrian crossing scenes on a road section according to claim 1, characterized in that: The step of identifying the target vehicle based on the trained XGBoost model and obtaining the safety level includes: Taking the safety level set by the trained model as the standard, the vehicle uses the on-board radar to obtain the risk field range while driving, extracts the relevant factor features within the risk field range, and substitutes the extracted relevant factor features into the trained XGBoost model to obtain the real-time safety level.

10. A quantification system for the safety of pedestrian crossing scenes on a road section, characterized in that: The system comprises: The range determination module is used to determine the risk range of pedestrians in the street crossing scene; The factor determination module is used to obtain traffic accident cases and motor vehicle driving scene data to determine the main factors affecting pedestrian crossing safety within the risk field; wherein the main factors include vehicle speed, operating margin, pedestrian initial speed, pedestrian acceleration, pedestrian type, pedestrian sight, weather conditions and other vehicles; the operating margin is time, indicating the length of time allowed for the driver to operate; The model training module is used to extract relevant factor features from traffic accident cases and motor vehicle driving scene data based on the main factors, establish a sample data set, train the XGBoost model for classifying safety levels, and set the safety level; The model application module is used to identify the target vehicle based on the trained XGBoost model and obtain the safety level.

Images