Automatic shunting method and system for traffic vulnerable participant group facing automatic driving game

Abstract

The application provides an automatic shunting method and system for a traffic weak participant group in automatic driving game, comprising the following steps: S1, acquiring all traffic weak participant group information of a current frame through sensing information; S2, judging a lane or a pedestrian crossing where the traffic weak participant group is located, and preliminarily classifying obstacles; S3, shunting the traffic weak participant group by using rules; and S4, constructing attributes of a new flow after shunting. The application is obviously clustered in behavior of the traffic weak participant group, can save a large amount of calculation resources such as collision judgment and trajectory prediction, reasonably avoids driving risks, and enhances stability, safety and comfort of automatic driving.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving, and more specifically, to an automatic diversion method and system for vulnerable traffic participants in autonomous driving game theory. Background Technology

[0002] Autonomous driving technology is the future trend of transportation development, and intersection control methods for autonomous vehicles are currently a research hotspot and challenge. Existing research mostly focuses on the interaction between autonomous vehicles and other motor vehicles. However, in reality, urban roads contain a large number of vulnerable traffic participants such as non-motorized vehicles and pedestrians. Their behavior is difficult to predict and their perception is difficult to be accurate, making them the most significant factors affecting the smooth passage of autonomous vehicles through intersections.

[0003] Currently, in most autonomous driving solutions, braking measures are taken to avoid collisions when an autonomous vehicle is about to cross a pedestrian crossing and there are pedestrians crossing the crossing. However, due to inaccurate predictions and fluctuations in perception, pedestrian behavior may lead to misjudgments, unreasonable trajectory planning, and scenarios where the vehicle stops and starts or misses the opportunity to cross.

[0004] Patent document CN114299456A discloses a method for assessing pedestrian crossing risk at intersections based on real-time trajectory detection, including: 1. Assessing the distance between pedestrians and vehicles at the intersection one by one; 2. Obtaining the behavioral trajectories of each target based on a multi-target tracking algorithm; using target detection box information and the corresponding target's behavioral trajectory to detect conflicts between pedestrians and vehicles; 3. Calculating the TTC value sequence of pedestrians and vehicles, and obtaining a pedestrian-vehicle conflict risk index based on the TTC value sequence; 4. Establishing an indicator feature matrix based on intersection information, calculating the comprehensive weight coefficient of the intersection environmental indicators; and obtaining a comprehensive evaluation index of pedestrian crossing risk based on the pedestrian-vehicle conflict risk index and the comprehensive weight coefficient of the environmental indicators. The disadvantages of this technology are that assessing risk by calculating pedestrian-vehicle distance is time-consuming, requiring separate comparison of the safe distance between each pedestrian and vehicle, and the continuity of pedestrian prediction and the stability of perception cannot be guaranteed across multiple frames. Furthermore, it does not treat multiple vulnerable traffic participants as obstacles through traffic diversion, thus failing to reduce computation time while ensuring continuity and stability across multiple frames.

[0005] Patent document CN114312836A discloses a method, device, equipment, and storage medium for an autonomous vehicle to yield to pedestrians. This invention determines the starting position and influence range of a pedestrian crossing in a planned driving path; based on the influence range of the pedestrian crossing, it obtains a set of pedestrians intending to cross the pedestrian crossing and the movement information of each pedestrian in the set, thus obtaining pedestrian reference information; based on the pedestrian reference information and the vehicle's driving information, it determines a risk assessment value for each pedestrian colliding with the vehicle; if the risk assessment value for any pedestrian colliding with the vehicle is greater than a preset risk threshold, a virtual wall is generated at the starting position of the pedestrian crossing. However, this invention has the following drawbacks: 1. It is not flexible enough. Although the virtual wall can ensure the stability of traffic decisions, it completely occupies the planning space, making it difficult to find a feasible plan; 2. It relies too much on pedestrian trajectory prediction. If the pedestrian trajectory prediction is inaccurate, it will cause the virtual wall to be unstable or completely impassable; 3. If a virtual wall is set up for a pedestrian who suddenly appears, it will cause the vehicle to brake suddenly, which is an unsafe behavior and greatly affects driving comfort. However, a slight reduction in speed or detour is sufficient for normal passage.

[0006] Patent document CN112487905B discloses a method and system for predicting the danger level of pedestrians around vehicles. This includes predicting pedestrian trajectories based on vehicle first-person perspective data, using data-driven temporal network modeling to achieve long-term trajectory prediction, reducing computational costs and shortening prediction time; a danger level identifier based on clustering analysis and classifiers, capable of identifying pedestrian danger levels based on feature parameters, avoiding the uncertainty caused by manually dividing parameter ranges to determine danger levels; predicting pedestrian movement trajectories based on a trained pedestrian trajectory fitter, extracting a set of pedestrian feature parameters, and inputting the parameter set into the trained pedestrian danger level identifier to predict the pedestrian danger level. However, this invention relies too heavily on predicted pedestrian trajectories. If the predicted pedestrian trajectory is inaccurate, it will lead to instability of the virtual wall or complete impassability. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the purpose of this invention is to provide an automatic diversion method and system for vulnerable traffic participants in autonomous driving game theory.

[0008] An automatic traffic diversion method for vulnerable participants in autonomous driving game theory, provided by the present invention, includes:

[0009] Step S1: Obtain information on all vulnerable traffic participants in the current frame through sensory information;

[0010] Step S2: Determine the lane or pedestrian crossing where the vulnerable traffic participants are located, and perform preliminary classification of obstacles;

[0011] Step S3: Use rules to divert traffic-disadvantaged groups;

[0012] Step S4: Construct the properties of the new stream after splitting.

[0013] Preferably, in step S1:

[0014] Information on all vulnerable traffic participants in the current frame is obtained through sensory information, including speed, location, and angle; vulnerable traffic participants include pedestrians and non-motorized vehicles.

[0015] Preferably, in step S2:

[0016] Step S2.1: Use obstacle location points to find all lanes and pedestrian crossings within a preset distance.

[0017] Step S2.2: Calculate the total score by weighting the angle and distance to the lane or pedestrian crossing.

[0018] l_score = Lane type distance parameter / Distance of vulnerable traffic participants from the lane centerline

[0019] heading_score = Lane type angle parameter ^(Angle between the speed direction of the vulnerable traffic participant and the orientation of the corresponding point in the lane / (PI / 12)).

[0020] Among them, the lane type angle parameter is less than 1;

[0021] total_score=min(l_score,1.0)*min(heading_score,1.0);

[0022] Where PI is the mathematical constant pi, l_score is the distance cost, heading_score is the angle cost, and total_score is the total cost;

[0023] Step S2.3: Select the lane with the lowest total_score as the lane where the vulnerable traffic participants are located;

[0024] Step S2.4: Using lane information, assess the relationship between vulnerable traffic participants and their own vehicles, including parallel, opposite, and intersecting relationships.

[0025] Preferably, in step S3:

[0026] Step S3.1: Divert traffic according to the type of vulnerable participants in transportation;

[0027] Step S3.2: Classify the vulnerable traffic participants for reverse, intersection, and parallel traffic respectively. Add obstacles that do not belong to reverse, intersection, and parallel traffic to each relationship and divert traffic in reverse, intersection, and parallel traffic in sequence.

[0028] Step S3.3: Before diverting traffic, sort the vulnerable traffic participants according to their distance from the vehicle, establish a new flow, and select the nearest vulnerable traffic participant not found on the map as the center;

[0029] Step S3.4: Find other vulnerable traffic participants that are closest to the central vulnerable traffic participant group and mark them as found on the map;

[0030] Step S3.5: Calculate the angle and distance parameters, total_score.

[0031] l_score = Distance parameter between different traffic disadvantaged groups / Distance between the central traffic disadvantaged group and other traffic disadvantaged groups.

[0032] heading_score = angle parameter of traffic disadvantaged group type ^ (angle between the speed direction of the traffic disadvantaged group in this center and the speed direction of other traffic disadvantaged groups / (PI / 12))

[0033] total_score=min(l_score,1.0)*min(heading_score,1.0);

[0034] Step S3.6: Construct an obstacle map and traverse all obstacles using a breadth-first search method;

[0035] Breadth-first search refers to starting from a certain point, visiting all unvisited neighboring nodes at once, and then visiting them layer by layer from these visited neighboring nodes. Breadth-first traversal is to traverse the graph in the manner of breadth-first search.

[0036] Step S3.7: If the total_score of the new other vulnerable traffic participants is less than or equal to the threshold_value, the new other vulnerable traffic participants are the central traffic participants. Return to step S3.4 and continue execution.

[0037] Where threshold_value is a hyperparameter that is set;

[0038] Step S3.8: If the total_score of the new other vulnerable traffic participants is greater than the threshold_value, the new other vulnerable traffic participants cannot be the central traffic participants, and return to step S3.3;

[0039] The process ends after all vulnerable groups in the transportation sector have been visited.

[0040] Preferably, in step S4:

[0041] Step S4.1: Speed ​​allocation: Select the speed of the traffic disadvantaged group closest to the vehicle as the speed of the entire flow;

[0042] Step S4.2: Size allocation: Draw the minimum convex envelope of all traffic disadvantaged participants corresponding to this flow, and use it as the occupied space of the traffic disadvantaged participant flow;

[0043] Step S4.3: Calculate the distance between the minimum envelope and the vehicle. If the minimum envelope contains the vehicle or the distance is less than the threshold, perform flow segmentation.

[0044] Step S4.4: Stream segmentation:

[0045] Using the vehicle as the center of the coordinate axis, the average of the positions of the nearest vulnerable traffic participants on the left and the nearest vulnerable traffic participants on the right is taken as the split vector, ((x left + x right) / 2, (y left + y right) / 2);

[0046] Where x_left is the x-coordinate of the center point of the left-side vulnerable traffic participant, y_left is the y-coordinate of the center point of the left-side vulnerable traffic participant, x_right is the x-coordinate of the center point of the right-side vulnerable traffic participant, and y_right is the y-coordinate of the center point of the right-side vulnerable traffic participant.

[0047] Find the angle between the vectors corresponding to all vulnerable traffic participants in the flow and the segmentation vector;

[0048] Grouping those with positive angles into one group and those with negative angles into another group, the division produces two different flows.

[0049] An automatic traffic diversion system for vulnerable participants in autonomous driving game theory, provided by the present invention, includes:

[0050] Module M1: Obtains information on all vulnerable traffic participants in the current frame through sensing information;

[0051] Module M2: Determines the lane or pedestrian crossing where the vulnerable traffic participants are located, and performs preliminary classification of obstacles;

[0052] Module M3: A rule-based diversion method for disadvantaged groups in transportation;

[0053] Module M4: Constructs the properties of the new stream after splitting.

[0054] Preferably, in module M1:

[0055] Information on all vulnerable traffic participants in the current frame is obtained through sensory information, including speed, location, and angle; vulnerable traffic participants include pedestrians and non-motorized vehicles.

[0056] Preferably, in module M2:

[0057] Module M2.1: Uses obstacle location points to find all lanes and pedestrian crossings within a preset distance.

[0058] Module M2.2: Calculates the total score by weighting the angle and distance to the lane or pedestrian crossing.

[0059] l_score = Lane type distance parameter / Distance of vulnerable traffic participants from the lane centerline

[0060] heading_score = Lane type angle parameter ^(Angle between the speed direction of the vulnerable traffic participant and the orientation of the corresponding point in the lane / (PI / 12)).

[0061] Among them, the lane type angle parameter is less than 1;

[0062] total_score=min(l_score,1.0)*min(heading_score,1.0);

[0063] Where PI is the mathematical constant pi, l_score is the distance cost, heading_score is the angle cost, and total_score is the total cost;

[0064] Module M2.3: Select the lane with the lowest total_score as the lane where the vulnerable traffic participants reside;

[0065] Module M2.4: Using lane information, assess the relationship between vulnerable traffic participants and their own vehicles, including parallel, oncoming, and intersecting relationships.

[0066] Preferably, in module M3:

[0067] Module M3.1: Diverts traffic based on the type of vulnerable participants in transportation;

[0068] Module M3.2: Classifies vulnerable traffic participants into reverse, intersection, and parallel traffic categories respectively. Obstacles that do not belong to reverse, intersection, and parallel traffic categories are added to each category and traffic is diverted in reverse, intersection, and parallel traffic order.

[0069] Module M3.3: Before traffic splitting, sort the vulnerable traffic participants according to their distance from the vehicle, establish a new flow, and select the nearest vulnerable traffic participant not found on the map as the center;

[0070] Module M3.4: Find other vulnerable traffic participants that are closest to the center's vulnerable traffic participant group and mark them as found on the map;

[0071] Module M3.5: Calculates angle and distance parameters. total_score:

[0072] l_score = Distance parameter between different traffic disadvantaged groups / Distance between the central traffic disadvantaged group and other traffic disadvantaged groups.

[0073] heading_score = angle parameter of traffic disadvantaged group type ^ (angle between the speed direction of the traffic disadvantaged group in this center and the speed direction of other traffic disadvantaged groups / (PI / 12))

[0074] total_score=min(l_score,1.0)*min(heading_score,1.0);

[0075] Module M3.6: Constructs an obstacle map and uses breadth-first traversal to iterate through all obstacles;

[0076] Breadth-first search refers to starting from a certain point, visiting all unvisited neighboring nodes at once, and then visiting them layer by layer from these visited neighboring nodes. Breadth-first traversal is to traverse the graph in the manner of breadth-first search.

[0077] Module M3.7: If the total_score of the new other vulnerable traffic participants is less than or equal to the threshold_value, the new other vulnerable traffic participants are the central traffic participants, return to module M3.4, and continue execution;

[0078] Where threshold_value is a hyperparameter that is set;

[0079] Module M3.8: If the total_score of the new other vulnerable traffic participants is greater than the threshold_value, the new other vulnerable traffic participants cannot be the central traffic participants, and return to Module M3.3;

[0080] The process ends after all vulnerable groups in the transportation sector have been visited.

[0081] Preferably, in module M4:

[0082] Module M4.1: Speed ​​Allocation: Select the speed of the traffic disadvantaged group closest to the vehicle as the speed of the entire flow;

[0083] Module M4.2: Size Allocation: Draw the minimum convex envelope of all vulnerable traffic participants corresponding to this flow, and use it as the occupied space of the vulnerable traffic participant flow;

[0084] Module M4.3: Calculate the distance between the minimum envelope and the vehicle. If the minimum envelope contains the vehicle or the distance is less than the threshold, perform a stream segmentation operation.

[0085] Module M4.4: Stream Splitting:

[0086] Using the vehicle as the center of the coordinate axis, the average of the positions of the nearest vulnerable traffic participants on the left and the nearest vulnerable traffic participants on the right is taken as the split vector, ((x left + x right) / 2, (y left + y right) / 2);

[0087] Where x_left is the x-coordinate of the center point of the left-side vulnerable traffic participant, y_left is the y-coordinate of the center point of the left-side vulnerable traffic participant, x_right is the x-coordinate of the center point of the right-side vulnerable traffic participant, and y_right is the y-coordinate of the center point of the right-side vulnerable traffic participant.

[0088] Find the angle between the vectors corresponding to all vulnerable traffic participants in the flow and the segmentation vector;

[0089] Grouping those with positive angles into one group and those with negative angles into another group, the division produces two different flows.

[0090] Compared with the prior art, the present invention has the following beneficial effects:

[0091] 1. This invention shows that the behavior of vulnerable traffic participants is clearly clustered, and the diversion process can save a lot of computing resources for collision judgment and trajectory prediction;

[0092] 2. The forms of vulnerable groups in transportation vary, but the flow can move stably. The flow splitting of this invention can ensure stability between multiple frames, and perceive the changes of vulnerable groups in transportation through the stability of the flow.

[0093] 3. The present invention will divert traffic to vulnerable groups, making the behavior of autonomous vehicles more stable. They will pass through when they can, and stop or slow down to give way when they cannot, thus reasonably avoiding driving risks and improving the safety and comfort of autonomous driving. Attached Figure Description

[0094] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0095] Figure 1 A schematic diagram of the automatic diversion method for vulnerable traffic participants in autonomous driving game theory.

[0096] Figure 2A schematic diagram of the system flow for an automatic diversion method for vulnerable traffic participants in autonomous driving game theory.

[0097] Figure 3 A diagram illustrating the diversion of vulnerable traffic participants when vehicles pass through an intersection, allowing them to proceed.

[0098] Figure 4 This diagram illustrates the result of traffic diversion for vulnerable groups when a vehicle attempts to pass through an intersection, preventing them from doing so. Detailed Implementation

[0099] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0100] Example 1:

[0101] This invention relates to autonomous driving for traffic flow within intersections, and more specifically, to the diversion of vulnerable traffic participants through rules and value assessments.

[0102] The technical problem this patent aims to solve is:

[0103] 1. There are many pedestrians at intersections, and autonomous driving must handle pedestrians crossing the road and non-motorized vehicles crossing the intersection. However, there are currently few autonomous driving methods specifically designed for these vulnerable traffic participants.

[0104] 2. The behavior of vulnerable groups in transportation is highly clustered, and processing them individually would consume a lot of computing resources.

[0105] 3. Vulnerable traffic participants have diverse physical characteristics and their perception is prone to instability and abrupt changes. Traffic diversion can ensure stability across multiple frames, facilitating the planning of more reasonable trajectories.

[0106] 4. The behavior of vulnerable groups in transportation is difficult to predict. Diversion can provide more accurate speed and direction of movement, thus ensuring safety.

[0107] 5. Diverting traffic to vulnerable groups will make the behavior of autonomous vehicles more stable. They can cross when they can and stop or slow down when they cannot. At the same time, they will not completely occupy the planned feasible space. The actual operational flexibility and upper limit are higher than the simple and crude setting of parking or rushing strategies.

[0108] According to the present invention, an automatic traffic diversion method for vulnerable participants in autonomous driving game theory is provided, such as... Figure 1 As shown, it includes:

[0109] Step S1: Obtain information on all vulnerable traffic participants in the current frame through sensory information;

[0110] Specifically, in step S1:

[0111] Information on all vulnerable traffic participants in the current frame is obtained through sensory information, including speed, location, and angle; vulnerable traffic participants include pedestrians and non-motorized vehicles.

[0112] Step S2: Determine the lane or pedestrian crossing where the vulnerable traffic participants are located, and perform preliminary classification of obstacles;

[0113] Specifically, in step S2:

[0114] Step S2.1: Use obstacle location points to find all lanes and pedestrian crossings within a preset distance.

[0115] Step S2.2: Calculate the total score by weighting the angle and distance to the lane or pedestrian crossing.

[0116] l_score = Lane type distance parameter / Distance of vulnerable traffic participants from the lane centerline

[0117] heading_score = Lane type angle parameter ^(Angle between the speed direction of the vulnerable traffic participant and the orientation of the corresponding point in the lane / (PI / 12)).

[0118] Among them, the lane type angle parameter is less than 1;

[0119] total_score=min(l_score,1.0)*min(heading_score,1.0);

[0120] Where PI is the mathematical constant pi, l_score is the distance cost, heading_score is the angle cost, and total_score is the total cost;

[0121] Step S2.3: Select the lane with the lowest total_score as the lane where the vulnerable traffic participants are located;

[0122] Step S2.4: Using lane information, assess the relationship between vulnerable traffic participants and their own vehicles, including parallel, opposite, and intersecting relationships.

[0123] Step S3: Use rules to divert traffic-disadvantaged groups;

[0124] Specifically, in step S3:

[0125] Step S3.1: Divert traffic according to the type of vulnerable participants in transportation;

[0126] Step S3.2: Classify the vulnerable traffic participants for reverse, intersection, and parallel traffic respectively. Add obstacles that do not belong to reverse, intersection, and parallel traffic to each relationship and divert traffic in reverse, intersection, and parallel traffic in sequence.

[0127] Step S3.3: Before diverting traffic, sort the vulnerable traffic participants according to their distance from the vehicle, establish a new flow, and select the nearest vulnerable traffic participant not found on the map as the center;

[0128] Step S3.4: Find other vulnerable traffic participants that are closest to the central vulnerable traffic participant group and mark them as found on the map;

[0129] Step S3.5: Calculate the angle and distance parameters, total_score.

[0130] l_score = Distance parameter between different traffic disadvantaged groups / Distance between the central traffic disadvantaged group and other traffic disadvantaged groups.

[0131] heading_score = angle parameter of traffic disadvantaged group type ^ (angle between the speed direction of the traffic disadvantaged group in this center and the speed direction of other traffic disadvantaged groups / (PI / 12))

[0132] total_score=min(l_score,1.0)*min(heading_score,1.0);

[0133] Step S3.6: Construct an obstacle map and traverse all obstacles using a breadth-first search method;

[0134] Breadth-first search refers to starting from a certain point, visiting all unvisited neighboring nodes at once, and then visiting them layer by layer from these visited neighboring nodes. Breadth-first traversal is to traverse the graph in the manner of breadth-first search.

[0135] Step S3.7: If the total_score of the new other vulnerable traffic participants is less than or equal to the threshold_value, the new other vulnerable traffic participants are the central traffic participants. Return to step S3.4 and continue execution.

[0136] Where threshold_value is a hyperparameter that is set;

[0137] Step S3.8: If the total_score of the new other vulnerable traffic participants is greater than the threshold_value, the new other vulnerable traffic participants cannot be the central traffic participants, and return to step S3.3;

[0138] The process ends after all vulnerable groups in the transportation sector have been visited.

[0139] Step S4: Construct the properties of the new stream after splitting.

[0140] Specifically, in step S4:

[0141] Step S4.1: Speed ​​allocation: Select the speed of the traffic disadvantaged group closest to the vehicle as the speed of the entire flow;

[0142] Step S4.2: Size allocation: Draw the minimum convex envelope of all traffic disadvantaged participants corresponding to this flow, and use it as the occupied space of the traffic disadvantaged participant flow;

[0143] Step S4.3: Calculate the distance between the minimum envelope and the vehicle. If the minimum envelope contains the vehicle or the distance is less than the threshold, perform flow segmentation.

[0144] Step S4.4: Stream segmentation:

[0145] Using the vehicle as the center of the coordinate axis, the average of the positions of the nearest vulnerable traffic participants on the left and the nearest vulnerable traffic participants on the right is taken as the split vector, ((x left + x right) / 2, (y left + y right) / 2);

[0146] Where x_left is the x-coordinate of the center point of the left-side vulnerable traffic participant, y_left is the y-coordinate of the center point of the left-side vulnerable traffic participant, x_right is the x-coordinate of the center point of the right-side vulnerable traffic participant, and y_right is the y-coordinate of the center point of the right-side vulnerable traffic participant.

[0147] Find the angle between the vectors corresponding to all vulnerable traffic participants in the flow and the segmentation vector;

[0148] Grouping those with positive angles into one group and those with negative angles into another group, the division produces two different flows.

[0149] Example 2:

[0150] Example 2 is a preferred embodiment of Example 1, and is used to illustrate the present invention in more detail.

[0151] This invention also provides an automatic traffic diversion system for disadvantaged traffic participants in autonomous driving games. The automatic traffic diversion system for disadvantaged traffic participants in autonomous driving games can be implemented by executing the process steps of the automatic traffic diversion method for disadvantaged traffic participants in autonomous driving games. That is, those skilled in the art can understand the automatic traffic diversion method for disadvantaged traffic participants in autonomous driving games as a preferred embodiment of the automatic traffic diversion system for disadvantaged traffic participants in autonomous driving games.

[0152] According to the present invention, an automatic traffic diversion system for vulnerable participants in autonomous driving game theory is provided, such as... Figure 2 As shown, it includes:

[0153] Module M1: Obtains information on all vulnerable traffic participants in the current frame through sensing information;

[0154] Specifically, in module M1:

[0155] Information on all vulnerable traffic participants in the current frame is obtained through sensory information, including speed, location, and angle; vulnerable traffic participants include pedestrians and non-motorized vehicles.

[0156] Module M2: Determines the lane or pedestrian crossing where the vulnerable traffic participants are located, and performs preliminary classification of obstacles;

[0157] Specifically, in module M2:

[0158] Module M2.1: Uses obstacle location points to find all lanes and pedestrian crossings within a preset distance.

[0159] Module M2.2: Calculates total_score by weighting the angle and distance to the lane or pedestrian crossing.

[0160] l_score = Lane type distance parameter / Distance of vulnerable traffic participants from the lane centerline

[0161] heading_score = Lane type angle parameter ^(Angle between the speed direction of the vulnerable traffic participant and the orientation of the corresponding point in the lane / (PI / 12)).

[0162] Among them, the lane type angle parameter is less than 1;

[0163] total_score=min(l_score,1.0)*min(heading_score,1.0);

[0164] Where PI is the mathematical constant pi, l_score is the distance cost, heading_score is the angle cost, and total_score is the total cost;

[0165] Module M2.3: Select the lane with the lowest total_score as the lane where the vulnerable traffic participants reside;

[0166] Module M2.4: Using lane information, assess the relationship between vulnerable traffic participants and their own vehicles, including parallel, oncoming, and intersecting relationships.

[0167] Module M3: A rule-based diversion method for disadvantaged groups in transportation;

[0168] Specifically, in module M3:

[0169] Module M3.1: Diverts traffic based on the type of vulnerable participants in transportation;

[0170] Module M3.2: Classifies vulnerable traffic participants into reverse, intersection, and parallel traffic categories respectively. Obstacles that do not belong to reverse, intersection, and parallel traffic categories are added to each category and traffic is diverted in reverse, intersection, and parallel traffic order.

[0171] Module M3.3: Before traffic splitting, sort the vulnerable traffic participants according to their distance from the vehicle, establish a new flow, and select the nearest vulnerable traffic participant not found on the map as the center;

[0172] Module M3.4: Find other vulnerable traffic participants that are closest to the center's vulnerable traffic participant group and mark them as found on the map;

[0173] Module M3.5: Calculates angle and distance parameters. total_score:

[0174] l_score = Distance parameter between different traffic disadvantaged groups / Distance between the central traffic disadvantaged group and other traffic disadvantaged groups.

[0175] heading_score = angle parameter of traffic disadvantaged group type ^ (angle between the speed direction of the traffic disadvantaged group in this center and the speed direction of other traffic disadvantaged groups / (PI / 12))

[0176] total_score=min(l_score,1.0)*min(heading_score,1.0);

[0177] Module M3.6: Constructs an obstacle map and uses breadth-first traversal to iterate through all obstacles;

[0178] Breadth-first search refers to starting from a certain point, visiting all unvisited neighboring nodes at once, and then visiting them layer by layer from these visited neighboring nodes. Breadth-first traversal is to traverse the graph in the manner of breadth-first search.

[0179] Module M3.7: If the total_score of the new other vulnerable traffic participants is less than or equal to the threshold_value, the new other vulnerable traffic participants are the central traffic participants, return to module M3.4, and continue execution;

[0180] Where threshold_value is a hyperparameter that is set;

[0181] Module M3.8: If the total_score of the new other vulnerable traffic participants is greater than the threshold_value, the new other vulnerable traffic participants cannot be the central traffic participants, and return to Module M3.3;

[0182] The process ends after all vulnerable groups in the transportation sector have been visited.

[0183] Module M4: Constructs the properties of the new stream after splitting.

[0184] Specifically, in module M4:

[0185] Module M4.1: Speed ​​Allocation: Select the speed of the traffic disadvantaged group closest to the vehicle as the speed of the entire flow;

[0186] Module M4.2: Size Allocation: Draw the minimum convex envelope of all vulnerable traffic participants corresponding to this flow, and use it as the occupied space of the vulnerable traffic participant flow;

[0187] Module M4.3: Calculate the distance between the minimum envelope and the vehicle. If the minimum envelope contains the vehicle or the distance is less than the threshold, perform a stream segmentation operation.

[0188] Module M4.4: Stream Splitting:

[0189] Using the vehicle as the center of the coordinate axis, the average of the positions of the nearest vulnerable traffic participants on the left and the nearest vulnerable traffic participants on the right is taken as the split vector, ((x left + x right) / 2, (y left + y right) / 2);

[0190] Where x_left is the x-coordinate of the center point of the left-side vulnerable traffic participant, y_left is the y-coordinate of the center point of the left-side vulnerable traffic participant, x_right is the x-coordinate of the center point of the right-side vulnerable traffic participant, and y_right is the y-coordinate of the center point of the right-side vulnerable traffic participant.

[0191] Find the angle between the vectors corresponding to all vulnerable traffic participants in the flow and the segmentation vector;

[0192] Grouping those with positive angles into one group and those with negative angles into another group, the division produces two different flows.

[0193] Example 3:

[0194] Example 3 is a preferred example of Example 1, and is used to illustrate the present invention in more detail.

[0195] Specific process of this invention:

[0196] 1. Obtain information (speed, position, angle, etc.) of all vulnerable traffic participants (dynamic obstacles: pedestrians, non-motorized vehicles) in the current frame through sensory information.

[0197] 2. Using obstacle location information, determine the lane or pedestrian crossing where vulnerable traffic participants are located, and perform preliminary obstacle classification:

[0198] a. Use obstacle location points to find all lanes and pedestrian crossings within 2 meters;

[0199] b. Calculate the total score by weighting the angle and distance to the lane or pedestrian crossing:

[0200] l_score = Lane type distance parameter / Distance of the vulnerable traffic participant from the lane centerline

[0201] heading_score = lane type angle parameter ^ (angle between the speed direction of the vulnerable traffic participant and the orientation of the corresponding point in the lane / (PI / 12)), where lane type angle parameter < 1.

[0202] total_score=min(l_score,1.0)*min(heading_score,1.0);

[0203] Where PI is the mathematical constant pi, l_score is the distance cost, heading_score is the angle cost, and total_score is the total cost.

[0204] c. Select the lane with the lowest total_score as the lane where the vulnerable traffic participants reside;

[0205] d. Using lane information, assess the relationship between the vulnerable traffic participant group and the vehicle (parallel, opposite, intersecting, unknown).

[0206] 3. Utilizing rules to divert traffic from disadvantaged groups.

[0207] a. Traffic flow will be differentiated based on the type of vulnerable participants in traffic (pedestrians, non-motorized vehicles);

[0208] b. Classify the vulnerable traffic participants according to three relationships (reverse, intersecting, parallel). Obstacles with unknown relationships are added to each relationship and diverted in the order of reverse, intersecting, and parallel relationships.

[0209] c. Before diverting traffic, sort the vulnerable traffic participants according to their distance from the vehicle, establish a new flow, and select the nearest vulnerable traffic participant not found on the map as the center;

[0210] d. Identify other vulnerable transportation participants closest to the center and mark them as found on the map;

[0211] e. Calculate the angle and distance parameters, total_score:

[0212] l_score = Distance parameter between different traffic disadvantaged groups / Distance between the central traffic disadvantaged group and other traffic disadvantaged groups.

[0213] heading_score = angle parameter of traffic disadvantaged group type ^ (angle between the speed direction of the traffic disadvantaged group in this center and the speed direction of other traffic disadvantaged groups / (PI / 12))

[0214] total_score=min(l_score,1.0)*min(heading_score,1.0);

[0215] f. Construct an obstacle map and traverse all obstacles using a breadth-first search method:

[0216] Breadth-first search refers to starting from a certain point, visiting all unvisited neighboring nodes at once, and then visiting them layer by layer from these visited neighboring nodes. Breadth-first traversal is a graph traversal performed in the manner of breadth-first search.

[0217] g. If the total_score of the new other vulnerable traffic participants is less than or equal to the threshold_value, then the new other vulnerable traffic participants are considered as the central traffic participants, and we return to step d; threshold_value is a hyperparameter that is set, with an empirical value of 0.4.

[0218] h. If the total_score of the new other vulnerable traffic participants is greater than the threshold_value, then the new other vulnerable traffic participants are not considered as central traffic participants, and return to step c;

[0219] The process ends after all vulnerable groups in the transportation sector have been visited.

[0220] 4. Attribute construction of the new stream after splitting (speed allocation, size allocation)

[0221] a. Speed ​​allocation: Select the speed of the traffic disadvantaged group closest to the vehicle as the speed of the entire flow;

[0222] b. Size allocation: Draw the minimum convex envelope of all vulnerable traffic participants corresponding to the flow, and use it as the occupied space of the vulnerable traffic participant flow;

[0223] c. Calculate the distance between the minimum envelope and the vehicle. If the minimum envelope contains the vehicle or the distance is less than the threshold, then a flow segmentation operation is required.

[0224] d. Stream splitting:

[0225] Using the vehicle as the center of the coordinate axis, the average position of the nearest vulnerable traffic participants on the left and the nearest vulnerable traffic participants on the right is selected as the split vector, i.e., ((x_left + x_right) / 2, (y_left + y_right) / 2).

[0226] Where x_left is the x-coordinate of the center point of the left-side vulnerable traffic participant, y_left is the y-coordinate of the center point of the left-side vulnerable traffic participant, x_right is the x-coordinate of the center point of the right-side vulnerable traffic participant, and y_right is the y-coordinate of the center point of the right-side vulnerable traffic participant.

[0227] Find the angle between the vectors corresponding to all vulnerable traffic participants in the flow and the segmentation vector;

[0228] Grouping those with positive angles into one group and those with negative angles into another group, the division produces two different flows.

[0229] Figure 3 and Figure 4 This is a diagram illustrating the automatic diversion of vulnerable road users when a vehicle passes through an intersection.

[0230] Example 4:

[0231] Example 4 is a preferred example of Example 1, which is used to illustrate the present invention in more detail.

[0232] Just as rules can be used to divert traffic from disadvantaged groups, clustering methods can also be used to divert traffic from disadvantaged groups.

[0233] Clustering is performed on vulnerable traffic participants using four dimensions: location (x, y), speed magnitude, and angle (angle, velocity).

[0234] a. Given a value of K and K initial cluster centroids, assign each point (i.e., data record) to the cluster represented by the nearest cluster centroid. After all points are assigned, recalculate the centroid of a cluster based on all points within that cluster (taking the average value). Then iteratively perform the steps of assigning points and updating cluster centroids until the change in cluster centroids is very small or the specified number of iterations is reached.

[0235] b. Use the final cluster center of the previous frame as the initial cluster center of the next frame to ensure stability;

[0236] The ck value is determined by factors such as the total number of vehicles and the complexity of the intersection.

[0237] Those skilled in the art will understand that, in addition to implementing the system, apparatus, and their modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and their modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.

[0238] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. An automatic traffic diversion method for disadvantaged participants in autonomous driving game theory, characterized in that, include: Step S1: Obtain information on all vulnerable traffic participants in the current frame through sensory information; Step S2: Determine the lane or pedestrian crossing where the vulnerable traffic participants are located, and perform preliminary classification of obstacles; Step S3: Use rules to divert traffic-disadvantaged groups; Step S4: Construct the properties of the new stream after splitting; In step S2: Step S2.1: Use obstacle location points to find all lanes and pedestrian crossings within a preset distance. Step S2.2: Calculate the total score by weighting the angle and distance to the lane or pedestrian crossing. l_score = Lane type distance parameter / Distance of vulnerable traffic participants from the lane centerline heading_score = lane type angle parameter ^(angle between the speed direction of the vulnerable traffic participant and the orientation of the corresponding point in the lane / (PI / 12)). Among them, the lane type angle parameter is less than 1; total_score=min(l_score,1.0)*min(heading_score,1.0); Where PI is the mathematical constant pi, l_score is the distance cost, heading_score is the angle cost, and total_score is the total cost; Step S2.3: Select the lane with the lowest total_score as the lane where the vulnerable traffic participants are located; Step S2.4: Using lane information, assess the relationship between vulnerable traffic participants and their own vehicles, including parallel, opposite, and intersecting relationships; In step S3: Step S3.1: Divert traffic according to the type of vulnerable participants in transportation; Step S3.2: Classify the vulnerable traffic participants for reverse, intersection, and parallel traffic respectively. Add obstacles that do not belong to reverse, intersection, and parallel traffic to each relationship and divert traffic in reverse, intersection, and parallel traffic in sequence. Step S3.3: Before diverting traffic, sort the vulnerable traffic participants according to their distance from the vehicle, establish a new flow, and select the nearest vulnerable traffic participant not found on the map as the center; Step S3.4: Find other vulnerable traffic participants that are closest to the central vulnerable traffic participant group and mark them as found on the map; Step S3.5: Calculate the angle and distance parameters, total_score. l_score = Distance parameter between the types of vulnerable traffic participants / Distance between the central vulnerable traffic participant and other vulnerable traffic participants. heading_score = angle parameter of traffic disadvantaged group type ^ (angle between the speed direction of the traffic disadvantaged group in this center and the speed direction of other traffic disadvantaged groups / (PI / 12)) total_score=min(l_score,1.0)*min(heading_score,1.0); Step S3.6: Construct an obstacle map and traverse all obstacles using a breadth-first search method; Breadth-first search refers to starting from a certain point, visiting all unvisited neighboring nodes at once, and then visiting them layer by layer from these visited neighboring nodes. Breadth-first traversal is to traverse the graph in the manner of breadth-first search. Step S3.7: If the total_score of the new other vulnerable traffic participants is less than or equal to the threshold_value, the new other vulnerable traffic participants are the central traffic participants. Return to step S3.4 and continue execution. Where threshold_value is a hyperparameter that is set; Step S3.8: If the total_score of the new other vulnerable traffic participants is greater than the threshold_value, the new other vulnerable traffic participants cannot be the central traffic participants, and return to step S3.3; The process ends after all vulnerable groups in transportation have been visited; In step S4: Step S4.1: Speed ​​allocation: Select the speed of the traffic disadvantaged group closest to the vehicle as the speed of the entire flow; Step S4.2: Size allocation: Draw the minimum convex envelope of all traffic disadvantaged participants corresponding to this flow, and use it as the occupied space of the traffic disadvantaged participant flow; Step S4.3: Calculate the distance between the minimum envelope and the vehicle. If the minimum envelope contains the vehicle or the distance is less than the threshold, perform flow segmentation. Step S4.4: Stream segmentation: Using the vehicle as the center of the coordinate axis, the average of the positions of the nearest vulnerable traffic participants on the left and the nearest vulnerable traffic participants on the right is taken as the split vector, ((x_left + x_right) / 2, (y_left + y_right) / 2). Where x_left is the x-coordinate of the center point of the left-side vulnerable traffic participant, y_left is the y-coordinate of the center point of the left-side vulnerable traffic participant, x_right is the x-coordinate of the center point of the right-side vulnerable traffic participant, and y_right is the y-coordinate of the center point of the right-side vulnerable traffic participant. Find the angle between the vectors corresponding to all vulnerable traffic participants in the flow and the segmentation vector; Grouping those with positive angles into one group and those with negative angles into another group, the division produces two different flows.

2. The automatic diversion method for vulnerable traffic participants in autonomous driving game theory according to claim 1, characterized in that, In step S1: Information on all vulnerable traffic participants in the current frame is obtained through sensory information, including speed, position, and angle; Vulnerable groups in transportation include pedestrians and non-motorized vehicles.

3. An automatic traffic diversion system for disadvantaged participants in autonomous driving game theory, characterized in that, include: Module M1: Obtains information on all vulnerable traffic participants in the current frame through sensing information; Module M2: Determines the lane or pedestrian crossing where the vulnerable traffic participants are located, and performs preliminary classification of obstacles; Module M3: A rule-based diversion method for disadvantaged groups in transportation; Module M4: Constructs the properties of the new stream after splitting; In module M2: Module M2.1: Uses obstacle location points to find all lanes and pedestrian crossings within a preset distance. Module M2.2: Calculates the total score by weighting the angle and distance to the lane or pedestrian crossing. l_score = Lane type distance parameter / Distance of vulnerable traffic participants from the lane centerline heading_score = lane type angle parameter ^(angle between the speed direction of the vulnerable traffic participant and the orientation of the corresponding point in the lane / (PI / 12)). Among them, the lane type angle parameter is less than 1; total_score=min(l_score,1.0)*min(heading_score,1.0); Where PI is the mathematical constant pi, l_score is the distance cost, heading_score is the angle cost, and total_score is the total cost; Module M2.3: Select the lane with the lowest total_score as the lane where the vulnerable traffic participants reside; Module M2.4: Utilizes lane information to assess the relationship between vulnerable traffic participants and their own vehicles, including parallel, oncoming, and intersecting relationships; In module M3: Module M3.1: Diverts traffic based on the type of vulnerable participants in transportation; Module M3.2: Classifies vulnerable traffic participants into reverse, intersection, and parallel traffic categories respectively. Obstacles that do not belong to reverse, intersection, and parallel traffic categories are added to each category and traffic is diverted in reverse, intersection, and parallel traffic order. Module M3.3: Before traffic splitting, sort the vulnerable traffic participants according to their distance from the vehicle, establish a new flow, and select the nearest vulnerable traffic participant not found on the map as the center; Module M3.4: Find other vulnerable traffic participants that are closest to the center's vulnerable traffic participant group and mark them as found on the map; Module M3.5: Calculates angle and distance parameters. total_score: l_score = Distance parameter between the types of vulnerable traffic participants / Distance between the central vulnerable traffic participant and other vulnerable traffic participants. heading_score = angle parameter of traffic disadvantaged group type ^ (angle between the speed direction of the traffic disadvantaged group in this center and the speed direction of other traffic disadvantaged groups / (PI / 12)) total_score=min(l_score,1.0)*min(heading_score,1.0); Module M3.6: Constructs an obstacle map and uses breadth-first traversal to iterate through all obstacles; Breadth-first search refers to starting from a certain point, visiting all unvisited neighboring nodes at once, and then visiting them layer by layer from these visited neighboring nodes. Breadth-first traversal is to traverse the graph in the manner of breadth-first search. Module M3.7: If the total_score of the new other vulnerable traffic participants is less than or equal to the threshold_value, the new other vulnerable traffic participants are the central traffic participants. Return to module M3.4 and continue execution. Where threshold_value is a hyperparameter that is set; Module M3.8: If the total_score of the new other vulnerable traffic participants is greater than the threshold_value, the new other vulnerable traffic participants cannot be the central traffic participants, and return to Module M3.3; The process ends after all vulnerable groups in transportation have been visited; In module M4: Module M4.1: Speed ​​Allocation: Select the speed of the traffic disadvantaged group closest to the vehicle as the speed of the entire flow; Module M4.2: Size Allocation: Draw the minimum convex envelope of all vulnerable traffic participants corresponding to this flow, and use it as the occupied space of the vulnerable traffic participant flow; Module M4.3: Calculate the distance between the minimum envelope and the vehicle. If the minimum envelope contains the vehicle or the distance is less than the threshold, perform a stream segmentation operation. Module M4.4: Stream Splitting: Using the vehicle as the center of the coordinate axis, the average of the positions of the nearest vulnerable traffic participants on the left and the nearest vulnerable traffic participants on the right is taken as the split vector, ((x_left + x_right) / 2, (y_left + y_right) / 2). Where x_left is the x-coordinate of the center point of the left-side vulnerable traffic participant, y_left is the y-coordinate of the center point of the left-side vulnerable traffic participant, x_right is the x-coordinate of the center point of the right-side vulnerable traffic participant, and y_right is the y-coordinate of the center point of the right-side vulnerable traffic participant. Find the angle between the vectors corresponding to all vulnerable traffic participants in the flow and the segmentation vector; Grouping those with positive angles into one group and those with negative angles into another group, the division produces two different flows.

4. The automatic diversion system for vulnerable traffic participants in autonomous driving game theory according to claim 3, characterized in that, In module M1: Information on all vulnerable traffic participants in the current frame is obtained through sensory information, including speed, position, and angle; Vulnerable groups in transportation include pedestrians and non-motorized vehicles.

Images