A driving risk monitoring method for emergent targets

Abstract

This invention proposes a driving risk monitoring method for sudden targets, belonging to the field of intelligent driving. The monitoring method includes: acquiring historical environmental data associated with the sudden target, establishing a model of impulsive crossing by the sudden target, and predicting the probability of impulsive crossing; constructing a driving risk field by combining the probability of impulsive crossing with other state information of the intelligent vehicle, and judging the driving risk based on the value of the driving risk field. This invention provides a risk threshold for safe driving of intelligent vehicles by constructing a driving risk field. Once the driving risk field value of the intelligent vehicle exceeds the risk field threshold, the intelligent vehicle will issue a warning, effectively avoiding collisions when a sudden target impulsively crosses in a blind spot environment.

Description

Technical Field

[0001] This invention belongs to the field of intelligent driving, specifically relating to a method for monitoring driving risks for suddenly appearing targets. Background Technology

[0002] With the rapid increase in car ownership, traffic safety issues have become increasingly serious. According to the latest report from the World Health Organization, more than 3,200 people die in traffic accidents globally every day, and more than two people die in traffic accidents every minute. Traffic accidents are a major cause of death for children and young people aged 5-29. Meanwhile, the fatality rate of traffic accidents involving pedestrians, non-motorized vehicles, and motorized vehicles is high, with "ghost pedestrians" (or "suddenly appearing from behind a vehicle") being a frequent example. When a vehicle is driving normally, obstructions to the driver's view caused by side parking, buildings, bus stops, overpass piers, and green belts create blind spots. When a pedestrian crosses the road ahead, the driver cannot see them in time and take timely and effective evasive action, resulting in a collision between the pedestrian and the vehicle, commonly known as a "ghost pedestrian." Due to the suddenness of these accidents and the vulnerability of pedestrians and non-motorized vehicles, they often cause significant property damage and serious injuries or fatalities. Currently, there is limited theoretical research on the risks of sudden obstacles appearing in blind spots, and research on road driving risk warning methods for sudden objects appearing in blind spots is even more lacking. Summary of the Invention

[0003] In view of this, the present invention proposes a driving risk monitoring method for sudden targets. This driving risk warning method aims to provide a driving risk field that can quantify the driving risk field caused by the visual blind spot environment during vehicle operation, and use this risk field to determine whether the vehicle's current driving speed is safe.

[0004] The objective of this invention is achieved by providing a method for monitoring driving risks against suddenly appearing targets. A suddenly appearing target is defined as a traffic participant who suddenly appears and poses a significant threat to vehicle safety, including but not limited to pedestrians, animals, and non-motorized vehicles. The driving risk monitoring method includes establishing a model of impulsive crossing by a suddenly appearing target, predicting the probability of a suddenly appearing target impulsively crossing ahead using this model, constructing a driving risk field based on the probability of the impulsive crossing, and determining the driving risk based on the field value of the driving risk field and taking corresponding early warning strategies. The specific steps are as follows:

[0005] Step 1: Obtain historical environmental data of the possible scenarios where the sudden target may appear, including visibility A1, time period A2, pedestrian flow A3, precipitation A4, traffic flow A5, and ambient temperature A6, and generate a sample set D;

[0006] Step 2: Build a decision tree model and use the sample set D as the decision tree training set. Perform classification training based on the C4.5 decision tree algorithm to obtain the impulsive traversal model of sudden target.

[0007] Step 3: Obtain real-time environmental data including visibility A1, time period A2, pedestrian flow A3, precipitation A4, traffic flow A5, and ambient temperature A6, and input them into the impulsive crossing model of sudden targets obtained in Step 2. The output is the probability state of impulsive crossing of sudden targets.

[0008] The probability states of the sudden target impulsive crossing include three states, which are respectively denoted as high probability state, medium probability state and low probability state.

[0009] Step 4: Given the probability δ of impulsive crossing of a sudden target, construct a driving risk field based on the probability δ of impulsive crossing of a sudden target, judge the driving risk based on the field value of the driving risk field, and take corresponding early warning strategies.

[0010] Step 4.1: The onboard sensors acquire vehicle status information and real-time environmental information;

[0011] The on-board sensor is installed at the center of the front bumper of the intelligent vehicle; the vehicle status information includes the speed and position of the intelligent vehicle.

[0012] Step 4.2, given the value of the probability δ: high probability state, δ = 0.8; medium probability state, δ = 0.5; low probability state, δ = 0.2;

[0013] Step 4.3: Denote the current time of the intelligent vehicle's driving as time t, and establish the driving risk field at time t;

[0014] Step 4.3.1: Record the time when the front of the intelligent vehicle reaches the conflict area at time t as t_t. c1 (t), whose expression is as follows:

[0015]

[0016] The conflict area refers to the line segment in the path of the emerging target that coincides with the path of the intelligent vehicle. l Let v be the distance along the vehicle's direction of travel between the front of the intelligent vehicle and the conflict area at time t. c (t) represents the speed of the intelligent vehicle at time t;

[0017] Step 4.3.2: Determine whether the vehicle's onboard sensors detect a sudden target at time t. If no sudden target is detected, proceed to step 4.3.3. If a sudden target is detected, proceed to step 4.3.4.

[0018] Step 4.3.3: Establish the first driving risk field of the intelligent vehicle in the visual blind spot environment at time t, and denote it as the first driving risk field E(t)1, the expression of which is as follows:

[0019]

[0020] Where θ(t) is the blind spot detection angle of the vehicle sensor, and d l ×tanθ(t) represents the width of the lateral field of view of the intelligent vehicle;

[0021] Proceed to step 5.1;

[0022] Step 4.3.4: Establish the second driving risk field of the intelligent vehicle in the visual blind spot environment at time t, and denote it as the second driving risk field E(t)2;

[0023] Let θ be the angle at which the onboard sensor of an intelligent vehicle detects a suddenly appearing target. r Choose from the following:

[0024] When the onboard sensors of the intelligent vehicle first detect a sudden target, take θ r =θ(t);

[0025] When the onboard sensors of an intelligent vehicle detect a sudden target for the first time, take θ. r =θ′(t), where θ′(t) is the detection angle when the on-board sensor of the intelligent vehicle detects a sudden target;

[0026] Calculate the time t it takes for the suddenly appearing target to reach the conflict zone at time t. r1 (θ r The calculation formula is as follows:

[0027]

[0028] In the formula, W c For the width of the intelligent vehicle, v p To highlight the speed of the target;

[0029] Let the time difference of the conflict be Δt, and its expression is:

[0030] Δt=t c1 (t)-t r1 (θ r )

[0031] The expression for the second driving risk field E(t)2 is as follows:

[0032]

[0033] Proceed to step 5.2;

[0034] Step 5, Risk Warning;

[0035] Step 5.1, let the probability δ of the sudden target impulsively crossing be 0.5, and the vehicle speed v c The risk value generated when (t) equals the preset vehicle speed and the lateral field of vision equals the preset width is the first risk threshold E. thre (t)1, the following methods shall be used for judgment and risk warning:

[0036] When E(t) ≥ E thre At time (t)1, the intelligent vehicle travels at a speed v c (t) Driving is considered high-risk; a high-risk warning is issued.

[0037] When E(t)1 <E thre At time (t)1, the intelligent vehicle travels at a speed v c (t) The driving is low-risk and requires no warning;

[0038] Step 5.2, let the vehicle speed v c The risk value generated when (t) equals the preset vehicle speed and the conflict time difference Δt equals the preset time difference is the second risk threshold E. thre (t)2, the following methods shall be used for judgment and risk warning:

[0039] When E(t)2≥E thre At time (t)2, the intelligent vehicle travels at a speed v c (t) Driving is considered high-risk; a high-risk warning is issued.

[0040] When E(t)2 <E thre At time (t)2, the intelligent vehicle travels at a speed v c (t) The driving is of low risk and no warning is required.

[0041] Preferably, the implementation process of step 2 is as follows:

[0042] Step 2.1: Define visibility A1, time period A2, pedestrian flow A3, precipitation A4, vehicle flow A5, and ambient temperature A6 as nodes, and denote them as node A. i Let i = 1, 2, 3, 4, 5, 6, and construct the node attribute set A of the decision tree, A = {A1, A2, A3, A4, A5, A6};

[0043] Based on node attributes, assign each node A i Each is divided into V probability fluctuation threshold intervals;

[0044] Step 2.2: Denote the probability of a sudden target appearing in samples within the same probability fluctuation threshold interval as ó, and divide the V probability fluctuation threshold intervals into high-probability intervals, medium-probability intervals, and low-probability intervals based on the value of ó, as follows:

[0045] When ó > 0.1, it is a high probability interval;

[0046] When 0.1 ≥ 0.01, the probability interval is medium.

[0047] When ó < 0.01, it is a low probability interval;

[0048] Step 2.3: Define the probability classification rules for sudden target appearances during intelligent vehicle operation, as follows:

[0049] If one node attribute is in a high probability range, and the other node attributes are in high, medium, or low probability ranges, then the probability of a target impulsively crossing the road during the intelligent vehicle's operation is high.

[0050] If one of the nodes has a probability fluctuation threshold range that is in the medium probability range, and the other nodes have probability fluctuation threshold ranges that are in the low probability or medium probability range, then the probability of a target impulsively crossing the road during the intelligent vehicle's operation is medium probability.

[0051] If the probability fluctuation threshold range of all nodes' node attributes is in the low probability range, then the probability of a target impulsively crossing during the intelligent vehicle's operation is low.

[0052] Step 2.4: Use the sample set D as the decision tree training set, perform classification training based on the C4.5 decision tree algorithm, and use the generated decision tree model as the impulsive traversal model of sudden target.

[0053] Preferably, step 2.1 involves classifying node A according to its node attributes. i The specific details of dividing the probability fluctuation threshold into V intervals are as follows:

[0054] Visibility A1 (km) is divided into 4 probability fluctuation threshold intervals, namely [0, 0.1], [0.1, 1], [1, 10], and [10, +∞].

[0055] The time period A2 is divided into 6 probability fluctuation threshold intervals, namely [0:00, 7:00], [7:00, 9:00], [9:00, 12:00], [12:00, 17:00], [17:00, 19:00], and [19:00, 24:00];

[0056] The pedestrian flow A3 (people / minute) is divided into 5 probability fluctuation threshold intervals, namely [0, 1], [1, 5], [5, 20], [20, 50], and [50, +∞].

[0057] The precipitation A4 (mm) is divided into 5 probability fluctuation threshold intervals, namely [0, 10], [10, 20], [20, 50], [50, 150], and [150, +∞].

[0058] Traffic flow A5 (vehicles / hour) is divided into 4 probability fluctuation threshold intervals, namely [0, 50], [50, 300], [300, 500], and [500, +∞].

[0059] The ambient temperature A6 (°C) is divided into 5 probability fluctuation threshold intervals, namely [-∞, 0], [0, 10], [10, 20], [20, 30], and [30, +∞].

[0060] Preferably, step 2.4, the classification training based on the C4.5 decision tree algorithm, includes:

[0061] Calculate the information gain ratio Gr(D, A) of the node attributes respectively. i The node feature with the highest information gain ratio is selected as the root node. In the branch nodes generated after dividing based on the root node, the information gain ratio is also used to divide the branch nodes. When the branch node and the parent node are of the same category, the division is stopped and the node is taken as a leaf node.

[0062] Preferably, the information gain ratio Gr(D,A) of the node attribute is... i The calculation is as follows:

[0063] Let any one of the high probability interval, medium probability interval and low probability interval defined in step 2.2 be denoted as the k-th probability interval, from high to low, k = 1, 2, 3;

[0064] The information entropy E(D) of a node attribute is calculated using the following expression:

[0065]

[0066] Where, p k This represents the proportion of samples in the sample set D that fall within different probability intervals;

[0067] Calculate the information gain G(D, A) of node attributes. i ) and the dynamic gain I(A) of node attributes i The expressions are as follows:

[0068]

[0069] Where v is any one of the V probability fluctuation threshold intervals, v = 1, 2, ..., V; |D v| represents the number of samples in the sample set D that fall within the v-th probability fluctuation threshold interval of the node attribute; E(D v ) is |D v |Information entropy;

[0070] Calculate the information gain ratio Gr(D, A) of node attributes. i The formula for its calculation is:

[0071]

[0072] Preferably, the scenarios in which the sudden target may appear include intelligent vehicles, the current lane, a zebra crossing, and the right side of the road. The intelligent vehicle is driving in the current lane, and there is a stationary obstacle parked on the right side of the road in front of the intelligent vehicle. There is a zebra crossing in front of the stationary obstacle, and the sudden target may appear on the zebra crossing at any time.

[0073] Compared with the prior art, the beneficial effects of the present invention are reflected in:

[0074] 1. This invention does not require blind spot information obtained from third-party devices such as roadside vehicle camera information and red light violation monitoring and processing devices in zebra crossing areas. Therefore, the application scenarios of the risk warning method of this invention are not limited by third-party devices.

[0075] 2. This invention activates a risk warning mode as soon as a blind spot is detected, prompting the driver to drive defensively. It does not only alert the driver when the camera detects a prominent target. This can effectively deal with collisions where the vehicle cannot brake in time when it detects an obstacle, especially at high speeds. Attached Figure Description

[0076] Figure 1 This is a top view of the target scene in an embodiment of the present invention;

[0077] Figure 2 This is a diagram showing the installation locations of the onboard sensors of the intelligent vehicle in this embodiment of the invention;

[0078] Figure 3 This is a flowchart of the driving risk monitoring method for sudden targets in an embodiment of the present invention;

[0079] Figure 4 This is a scene diagram in this embodiment of the invention where the vehicle-mounted sensor did not detect a sudden target at time t.

[0080] Figure 5 This is a scene diagram of a sudden target detected by the vehicle-mounted sensor at time t in an embodiment of the present invention. Detailed Implementation

[0081] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0082] Figure 1 This is a top view of the scenario where the sudden target may appear, as described in this embodiment of the invention. Figure 1 As can be seen, the driving risk monitoring method involves scenarios including the current lane, zebra crossing, and right side of the road. The intelligent vehicle is driving in the current lane, there is a stationary obstacle parked on the right side of the road in front of the intelligent vehicle, there is a zebra crossing in front of the stationary obstacle, and a sudden target may appear on the zebra crossing at any time.

[0083] The term "suddenly appearing target" is defined as a traffic participant that suddenly appears and poses a significant threat to the safe operation of vehicles, including but not limited to pedestrians, animals, and non-motorized vehicles that suddenly appear.

[0084] Figure 2 This is a diagram showing the installation locations of the onboard sensors in an intelligent vehicle according to an embodiment of the present invention. Figure 2 As can be seen, the intelligent vehicle is equipped with onboard sensors, which are installed at the center of the front bumper.

[0085] The visual blind spot environment is caused by stationary obstacles obstructing the field of view of the vehicle's onboard sensors, including but not limited to visual blind spots caused by stationary obstacles on the right side of the road and visual blind spots caused by stationary obstacles on the left side of the road.

[0086] Figure 3 This is a flowchart of a driving risk monitoring method for sudden targets in an embodiment of the present invention. As shown in the flowchart, the driving risk monitoring method includes establishing a sudden target impulsive crossing model, predicting the probability state of a sudden target impulsively crossing ahead using the sudden target impulsive crossing model, constructing a driving risk field based on the probability state of the sudden target impulsive crossing, judging the driving risk based on the driving risk field value, and taking corresponding early warning strategies. The specific steps are as follows:

[0087] Step 1: Obtain historical environmental data of the possible scenarios where the sudden target may appear, including visibility A1, time period A2, pedestrian flow A3, precipitation A4, traffic flow A5, and ambient temperature A6, and generate a sample set D.

[0088] Step 2: Build a decision tree model and use the sample set D as the decision tree training set. Perform classification training based on the C4.5 decision tree algorithm to obtain the impulsive traversal model of sudden target.

[0089] Step 3: Obtain real-time environmental data including visibility A1, time period A2, pedestrian flow A3, precipitation A4, traffic flow A5, and ambient temperature A6, and input them into the impulsive crossing model of sudden targets obtained in Step 2. The output is the probability state of impulsive crossing of sudden targets.

[0090] The probability states of the sudden target impulsive crossing include three states, which are denoted as high probability state, medium probability state and low probability state, respectively.

[0091] Step 4: Given the probability δ of a sudden target impulsively crossing, construct a driving risk field based on the probability δ of the sudden target impulsively crossing, judge the driving risk based on the field value of the driving risk field, and take corresponding early warning strategies.

[0092] Step 4.1: The onboard sensors acquire vehicle status information and real-time environmental information;

[0093] The on-board sensor is installed at the center of the front bumper of the intelligent vehicle; the vehicle status information includes the speed and position of the intelligent vehicle.

[0094] Step 4.2, give the value of the probability δ: high probability state, δ = 0.8; medium probability state, δ = 0.5; low probability state, δ = 0.2.

[0095] Step 4.3: Denote the current time of the intelligent vehicle's driving as time t, and establish the driving risk field at time t;

[0096] Step 4.3.1: Record the time when the front of the intelligent vehicle reaches the conflict area at time t as t_t. c1 (t), whose expression is as follows:

[0097]

[0098] The conflict area refers to the line segment in the path of the emerging target that coincides with the path of the intelligent vehicle. l Let v be the straight-line distance along the vehicle's direction of travel from the front of the intelligent vehicle to the conflict area at time t. c (t) represents the speed of the intelligent vehicle at time t;

[0099] Step 4.3.2: Determine whether the vehicle's onboard sensors detect a sudden target at time t. If no sudden target is detected, proceed to step 4.3.3. If a sudden target is detected, proceed to step 4.3.4.

[0100] Step 4.3.3: Establish the first driving risk field of the intelligent vehicle in the visual blind spot environment at time t, and denote it as the first driving risk field E(t)1, the expression of which is as follows:

[0101]

[0102] Where θ(t) is the blind spot detection angle of the vehicle sensor, and d1×tanθ(t) is the width of the lateral field of view of the intelligent vehicle.

[0103] Proceed to step 5.1;

[0104] Step 4.3.4: Establish the second driving risk field of the intelligent vehicle in the visual blind spot environment at time t, and denote it as the second driving risk field E(t)2;

[0105] Let θ be the angle at which the onboard sensor of an intelligent vehicle detects a suddenly appearing target. r Choose from the following:

[0106] When the onboard sensors of the intelligent vehicle first detect a sudden target, take θ r =θ(t);

[0107] When the onboard sensors of an intelligent vehicle detect a sudden target for the first time, take θ. r =θ′(t), where θ′(t) is the detection angle when the on-board sensor of the intelligent vehicle detects a sudden target;

[0108] Calculate the time t it takes for the suddenly appearing target to reach the conflict zone at time t. r1 (θ r The calculation formula is as follows:

[0109]

[0110] In the formula, W c For the width of the intelligent vehicle, v p To highlight the speed of the target;

[0111] Let the time difference of the conflict be Δt, and its expression is:

[0112] Δt=t c1 (t)-t r1 (θ r )

[0113] The expression for the second driving risk field E(t)2 is as follows:

[0114]

[0115] Proceed to step 5.2.

[0116] Step 5, Risk Warning;

[0117] Step 5.1, let the probability δ of the sudden target impulsively crossing be 0.5, and the vehicle speed v c The risk value generated when (t) equals the preset vehicle speed and the lateral field of vision equals the preset width is the first risk threshold E. thre (t)1, the following methods shall be used for judgment and risk warning:

[0118] When E(t) ≥ E thre At time (t)1, the intelligent vehicle travels at a speed v c (t) Driving is considered high-risk; a high-risk warning is issued.

[0119] When E(t)1 <E thre At time (t)1, the intelligent vehicle travels at a speed v c (t) The driving is of low risk and no warning is required.

[0120] The preset vehicle speed is set to 10 m / s, and the preset width is set to 4 m.

[0121] Step 5.2, let the vehicle speed v c The risk value generated when (t) equals the preset vehicle speed and the conflict time difference Δt equals the preset time difference is the second risk threshold E. thre (t)2, the following methods shall be used for judgment and risk warning:

[0122] When E(t)2≥E thre At time (t)2, the intelligent vehicle travels at a speed v c (t) Driving is considered high-risk; a high-risk warning is issued.

[0123] When E(t)2 <E thre At time (t)2, the intelligent vehicle travels at a speed v c (t) The driving is of low risk and no warning is required.

[0124] The preset vehicle speed is set to 10 m / s, and the preset time difference is set to 1.2 s.

[0125] Figure 4 This is a scene diagram showing that the vehicle-mounted sensor did not detect a sudden target in a blind spot environment at time t in this embodiment of the invention. Figure 4 In the diagram, line segment CD represents the path of the suddenly appearing target, and line segment AB represents the conflict zone. Figure 4 In the diagram, point 0 is the position of the vehicle-mounted sensor. The connection between point 0 and the front left corner of the static obstacle is extended to intersect line segment CD at point C. The angle θ(t) between line segment OC and the velocity direction of the controlled vehicle is the blind spot detection angle of the vehicle-mounted sensor at time t. Figure 5 This is a scene diagram of the vehicle-mounted sensor detecting a sudden target at time t in an embodiment of the present invention;

[0126] In this embodiment, the implementation process of step 2 is as follows:

[0127] Step 2.1: Define visibility A1, time period A2, pedestrian flow A3, precipitation A4, vehicle flow A5, and ambient temperature A6 as nodes, and denote them as node A. i Let i = 1, 2, 3, 4, 5, 6. Construct a set of node attributes A for the decision tree, A = {A1, A2, A3, A4, A5, A6}.

[0128] Based on node attributes, assign each node A iEach is divided into V probability fluctuation threshold intervals, as detailed below:

[0129] Visibility A1 (km) is divided into 4 probability fluctuation threshold intervals, namely [0, 0.1], [0.1, 1], [1, 10], and [10, +∞].

[0130] The time period A2 is divided into 6 probability fluctuation threshold intervals, namely [0:00, 7:00], [7:00, 9:00], [9:00, 12:00], [12:00, 17:00], [17:00, 19:00],

[0131] [19:00, 24:00];

[0132] The pedestrian flow A3 (people / minute) is divided into 5 probability fluctuation threshold intervals, namely [0, 1], [1, 5], [5, 20], [20, 50], and [50, +∞].

[0133] The precipitation A4 (mm) is divided into 5 probability fluctuation threshold intervals, namely [0, 10], [10, 20], [20, 50], [50, 150], and [150, +∞].

[0134] Traffic flow A5 (vehicles / hour) is divided into 4 probability fluctuation threshold intervals, namely [0, 50], [50, 300], [300, 500], and [500, +∞].

[0135] The ambient temperature A6 (°C) is divided into 5 probability fluctuation threshold intervals, namely [-∞, 0], [0, 10], [10, 20], [20, 30], and [30, +∞].

[0136] Step 2.2: Denote the probability of a sudden target appearing in samples within the same probability fluctuation threshold interval as ó, and divide the V probability fluctuation threshold intervals into high-probability intervals, medium-probability intervals, and low-probability intervals based on the value of ó, as follows:

[0137] When ó > 0.1, it is a high probability interval;

[0138] When 0.1 ≥ 0.01, the probability interval is medium.

[0139] When ó < 0.01, it is a low probability interval.

[0140] Step 2.3: Define the probability classification rules for sudden target appearances during intelligent vehicle operation, as follows:

[0141] If one node attribute is in a high probability range, and the other node attributes are in high, medium, or low probability ranges, then the probability of a target impulsively crossing the road during the intelligent vehicle's operation is high.

[0142] If one of the nodes has a probability fluctuation threshold range that is in the medium probability range, and the other nodes have probability fluctuation threshold ranges that are in the low probability or medium probability range, then the probability of a target impulsively crossing the road during the intelligent vehicle's operation is medium probability.

[0143] If the probability fluctuation threshold range of all nodes is in the low probability range, then the probability of a target impulsively crossing the road during the intelligent vehicle's operation is low.

[0144] Step 2.4: Use the sample set D as the decision tree training set, perform classification training based on the C4.5 decision tree algorithm, and use the generated decision tree model as the impulsive traversal model of sudden target.

[0145] The classification training based on the C4.5 decision tree algorithm includes:

[0146] Calculate the information gain ratio Gr(D, A) of the node attributes respectively. i The node feature with the highest information gain ratio is selected as the root node. In the branch nodes generated after dividing based on the root node, the information gain ratio is also used to divide the branch nodes. When the branch node and the parent node are of the same category, the division is stopped and the node is taken as a leaf node.

[0147] The information gain ratio Gr(D, A) of the node attributes i The calculation is as follows:

[0148] Let any one of the high probability interval, medium probability interval and low probability interval defined in step 2.2 be denoted as the k-th probability interval, from high to low, k = 1, 2, 3.

[0149] The information entropy E(D) of a node attribute is calculated using the following expression:

[0150]

[0151] Where, p k This represents the proportion of samples in the sample set D that fall within different probability intervals.

[0152] Calculate the information gain G(D, A) of node attributes. i ) and the dynamic gain I(A) of node attributes i The expressions are as follows:

[0153]

[0154] Where v is any one of the V probability fluctuation threshold intervals, v = 1, 2, ..., V; |D v | represents the number of samples in the sample set D that fall within the v-th probability fluctuation threshold interval of the node attribute; E(D v ) is |D v |Information entropy.

[0155] Calculate the information gain ratio Gr(D, A) of node attributes. i The formula for its calculation is:

[0156]

Claims

1. A method for monitoring driving risks of suddenly appearing targets, wherein the suddenly appearing target is defined as a traffic participant who suddenly appears and poses a significant threat to the safe driving of vehicles, including but not limited to pedestrians, animals, and non-motorized vehicles that suddenly appear; characterized in that, The driving risk monitoring method includes establishing a sudden target impulsive crossing model, predicting the probability of a sudden target impulsive crossing ahead using the model, constructing a driving risk field based on the probability of the sudden target impulsive crossing, judging the driving risk based on the field value of the driving risk field, and taking corresponding early warning strategies. The specific steps are as follows: Step 1: Obtain historical environmental data of the possible scenarios where the sudden target may appear, including visibility A1, time period A2, pedestrian flow A3, precipitation A4, traffic flow A5, and ambient temperature A6, and generate a sample set D; Step 2: Build a decision tree model and use the sample set D as the decision tree training set. Perform classification training based on the C4.5 decision tree algorithm to obtain the impulsive traversal model of sudden target. Step 3: Obtain real-time environmental data including visibility A1, time period A2, pedestrian flow A3, precipitation A4, traffic flow A5, and ambient temperature A6, and input them into the impulsive crossing model of sudden targets obtained in Step 2. The output is the probability state of impulsive crossing of sudden targets. The probability states of the sudden target impulsive crossing include three states, which are respectively denoted as high probability state, medium probability state and low probability state. Step 4: Given the probability δ of impulsive crossing of a sudden target, construct a driving risk field based on the probability δ of impulsive crossing of a sudden target, judge the driving risk based on the field value of the driving risk field, and take corresponding early warning strategies. Step 4.1: The onboard sensors acquire vehicle status information and real-time environmental information; The on-board sensor is installed at the center of the front bumper of the intelligent vehicle; the vehicle status information includes the speed and position of the intelligent vehicle. Step 4.2, given the value of the probability δ: high probability state, δ = 0.8; For medium probability states, δ = 0.5; for low probability states, δ = 0.

2. Step 4.3: Denote the current time of the intelligent vehicle's driving as time t, and establish the driving risk field at time t; Step 4.3.1: Record the time when the front of the intelligent vehicle reaches the conflict area at time t as t_t. c1 (t), whose expression is as follows: The conflict area refers to the line segment in the path of the emerging target that coincides with the path of the intelligent vehicle. l Let v be the distance along the vehicle's direction of travel between the front of the intelligent vehicle and the conflict area at time t. c (t) represents the speed of the intelligent vehicle at time t; Step 4.3.2: Determine whether the vehicle's onboard sensors detect a sudden target at time t. If no sudden target is detected, proceed to step 4.3.

3. If a sudden target is detected, proceed to step 4.3.

4. Step 4.3.3: Establish the first driving risk field of the intelligent vehicle in the visual blind spot environment at time t, and denote it as the first driving risk field E(t)1, the expression of which is as follows: Where θ(t) is the blind spot detection angle of the vehicle sensor, and d1×tanθ(t) is the width of the lateral field of view of the intelligent vehicle; Proceed to step 5.1; Step 4.3.4: Establish the second driving risk field of the intelligent vehicle in the visual blind spot environment at time t, and denote it as the second driving risk field E(t)2; Let θ be the angle at which the onboard sensor of an intelligent vehicle detects a suddenly appearing target. r Choose from the following: When the onboard sensors of the intelligent vehicle first detect a sudden target, take θ r =θ(t); When the onboard sensors of an intelligent vehicle detect a sudden target for the first time, take θ. r =θ′(t), where θ′(t) is the detection angle when the on-board sensor of the intelligent vehicle detects a sudden target; Calculate the time t it takes for the suddenly appearing target to reach the conflict zone at time t. r1 (θ r The calculation formula is as follows: In the formula, W c For the width of the intelligent vehicle, v p To highlight the speed of the target; Let the time difference of the conflict be Δt, and its expression is: Δt=t c1 (t)-t r1 (θ r ) The expression for the second driving risk field E(t)2 is as follows: Proceed to step 5.2; Step 5, Risk Warning; Step 5.1, let the probability δ of the sudden target impulsively crossing be 0.5, and the vehicle speed v c The risk value generated when (t) equals the preset vehicle speed and the lateral field of vision equals the preset width is the first risk threshold E. thre (t)1, the following methods shall be used for judgment and risk warning: When E(t) ≥ E thre At time (t)1, the intelligent vehicle travels at a speed v c (t) Driving is considered high-risk; a high-risk warning is issued. When E(t)1 <E thre At time (t)1, the intelligent vehicle travels at a speed v c (t) The driving is low-risk and requires no warning; Step 5.2, let the vehicle speed v c The risk value generated when (t) equals the preset vehicle speed and the conflict time difference Δt equals the preset time difference is the second risk threshold E. thre (t)2, the following methods shall be used for judgment and risk warning: When E(t)2≥E thre At time (t)2, the intelligent vehicle travels at a speed v c (t) Driving is considered high-risk; a high-risk warning is issued. When E(t)2 <E thre At time (t)2, the intelligent vehicle travels at a speed v c (t) The driving is of low risk and no warning is required.

2. The method for monitoring driving risks against suddenly appearing targets according to claim 1, characterized in that, The implementation process of step 2 is as follows: Step 2.1: Define visibility A1, time period A2, pedestrian flow A3, precipitation A4, vehicle flow A5, and ambient temperature A6 as nodes, and denote them as node A. i Let i = 1, 2, 3, 4, 5, 6, and construct the node attribute set A of the decision tree, A = {A1, A2, A3, A4, A5, A6}; Based on node attributes, assign each node A i Each is divided into V probability fluctuation threshold intervals; Step 2.2: Denote the probability of a sudden target appearing in samples within the same probability fluctuation threshold interval as ó, and divide the V probability fluctuation threshold intervals into high-probability intervals, medium-probability intervals, and low-probability intervals based on the value of ó, as follows: When ó > 0.1, it is a high probability interval; When 0.1 ≥ 0.01, the probability interval is medium. When ó < 0.01, it is a low probability interval; Step 2.3: Define the probability classification rules for sudden target appearances during intelligent vehicle operation, as follows: If one node attribute is in a high probability range, and the other node attributes are in high, medium, or low probability ranges, then the probability of a target impulsively crossing the road during the intelligent vehicle's operation is high. If one of the nodes has a probability fluctuation threshold range that is in the medium probability range, and the other nodes have probability fluctuation threshold ranges that are in the low probability or medium probability range, then the probability of a target impulsively crossing the road during the intelligent vehicle's operation is medium probability. If the probability fluctuation threshold range of all nodes' node attributes is in the low probability range, then the probability of a target impulsively crossing during the intelligent vehicle's operation is low. Step 2.4: Use the sample set D as the decision tree training set, perform classification training based on the C4.5 decision tree algorithm, and use the generated decision tree model as the impulsive traversal model of sudden target.

3. The method for monitoring driving risks against suddenly appearing targets according to claim 2, characterized in that, Step 2.1 describes classifying node A based on its node attributes. i The specific details of dividing the probability fluctuation threshold into V intervals are as follows: Visibility A1 (km) is divided into 4 probability fluctuation threshold intervals, namely [0, 0.1], [0.1, 1], [1, 10], and [10, +∞]. The time period A2 is divided into 6 probability fluctuation threshold intervals, namely [0:00, 7:00], [7:00, 9:00], [9:00, 12:00], [12:00, 17:00], [17:00, 19:00], and [19:00, 24:00]; The pedestrian flow A3 (people / minute) is divided into 5 probability fluctuation threshold intervals, namely [0, 1], [1, 5], [5, 20], [20, 50], and [50, +∞]. The precipitation A4 (mm) is divided into 5 probability fluctuation threshold intervals, namely [0, 10], [10, 20], [20, 50], [50, 150], and [150, +∞]. Traffic flow A5 (vehicles / hour) is divided into 4 probability fluctuation threshold intervals, namely [0, 50], [50, 300], [300, 500], and [500, +∞]. The ambient temperature A6 (°C) is divided into 5 probability fluctuation threshold intervals, namely [-∞, 0], [0, 10], [10, 20], [20, 30], and [30, +∞].

4. The method for monitoring driving risks against suddenly appearing targets according to claim 2, characterized in that, Step 2.4, the classification training based on the C4.5 decision tree algorithm, includes: Calculate the information gain ratio Gr(D,A) of the node attributes respectively. i The node feature with the highest information gain ratio is selected as the root node. In the branch nodes generated after dividing based on the root node, the information gain ratio is also used to divide the branch nodes. When the branch node and the parent node are of the same category, the division is stopped and the node is taken as a leaf node.

5. The method for monitoring driving risks against suddenly appearing targets according to claim 4, characterized in that, The information gain ratio Gr(D, A) of the node attributes i The calculation is as follows: Let any one of the high probability interval, medium probability interval and low probability interval defined in step 2.2 be denoted as the k-th probability interval, from high to low, k = 1, 2, 3; The information entropy E(D) of a node attribute is calculated using the following expression: Where, p k This represents the proportion of samples in the sample set D that fall within different probability intervals; Calculate the information gain G(D, A) of node attributes. i ) and the dynamic gain I(A) of node attributes i The expressions are as follows: Where v is any one of the V probability fluctuation threshold intervals, v = 1, 2, ..., V; |D v | represents the number of samples in the sample set D that fall within the v-th probability fluctuation threshold interval of the node attribute; E(D v ) is |D v |Information entropy; Calculate the information gain ratio Gr(D, A) of node attributes. i The formula for its calculation is:

6. The method for monitoring driving risks against suddenly appearing targets according to claim 1, characterized in that, The scenarios in which the sudden target may appear include intelligent vehicles, the current lane, zebra crossings, and the right side of the road. The intelligent vehicle is driving in the current lane, and there is a stationary obstacle parked on the right side of the road in front of the intelligent vehicle. There is a zebra crossing in front of the stationary obstacle, and the sudden target may appear on the zebra crossing at any time.

Images