Method for predicting dangerous situation of vehicle-pedestrian collision based on spatio-temporal urgency

Abstract

The present invention provides a method for predicting the dangerous situation of vehicle-pedestrian collision based on spatio-temporal urgency, belonging to the field of vehicle-pedestrian collision risk prediction for autonomous driving. Predict the multi-modal trajectories of pedestrians based on on-vehicle perspective data; use the bounding box dynamic detection algorithm to perform intersection tests on pedestrian trajectories and the planned path of the host vehicle to identify potential collision trajectories; calculate the conflict parameters under potential collision trajectories and non-potential collision trajectories, establish a model to quantify the spatio-temporal urgency, output the comprehensive danger level of pedestrians, and divide the safe driving area according to the comprehensive danger level and the safe distance between vehicles and pedestrians. The present invention fully considers all possible trajectories of pedestrians, improves the limitations of the existing vehicle-pedestrian collision risk prediction, such as incomplete consideration factors and relatively subjective evaluation mechanisms, and is more adaptable to autonomous driving and complex traffic environments.

Description

Technical Field

[0001] The present invention belongs to the technology of predicting the risk of vehicle - pedestrian collision in autonomous driving, and specifically relates to a method for predicting the dangerous situation of vehicle - pedestrian collision based on spatio - temporal urgency. Background Art

[0002] Pedestrian safety is an issue that cannot be ignored during the driving of autonomous vehicles. Accurately identifying the dangerous situation of collision between a pedestrian in front and an autonomous vehicle is of great significance for the trajectory planning and driving decision - making of autonomous vehicles. In the existing research on vehicle - pedestrian collision risks for autonomous driving, most studies use conflict techniques to compare indicators with thresholds based on simple kinematic models such as the Constant Velocity (CV) model and the Constant Acceleration (CA) model, and mainly focus on single methods. Although it has a certain degree of efficiency, due to the relatively simple motion model and traditional evaluation criteria, the uncertainty of vehicle - pedestrian motion is less considered, which directly affects the accuracy of evaluation and cannot meet the requirements of vehicle - pedestrian collision safety evaluation in complex scenarios. With the development of computer vision, trajectory extraction and prediction algorithms based on data - driven methods have begun to be applied in the research of vehicle - pedestrian collision risks. Although there is a certain improvement in the pedestrian behavior model, the selection of risk evaluation indicators is still relatively single, and the accuracy of the evaluation model cannot match the high - accuracy pedestrian trajectory prediction. It is difficult to comprehensively evaluate the vehicle - pedestrian collision risk in the future for a period of time and is not well - suited for the autonomous driving system architecture. Therefore, for the fine - grained dangerous perception requirements of autonomous driving, a risk assessment model that integrates multiple methods will be the focus of future research. Summary of the Invention

[0003] In order to solve the above - mentioned technical defects in the prior art, the present invention proposes a method for predicting the dangerous situation of vehicle - pedestrian collision based on spatio - temporal urgency. A new framework is proposed based on the prediction of pedestrian multi - modal trajectories, fully considering all possible trajectories of pedestrians, quantifying the urgency of vehicle - pedestrian collision in terms of time and space, and comprehensively obtaining the pedestrian danger level by combining the collision probability of pedestrians to study the dangerous situation of collision.

[0004] The technical solution for achieving the object of the present invention is: A method for predicting the dangerous situation of vehicle - pedestrian collision based on spatio - temporal urgency, and the specific steps are as follows:

[0005] Step 1: Obtain pedestrian motion parameters, and predict pedestrian multi - modal trajectories through a pedestrian decision - making model and a GRU network;

[0006] Step 2: Use the bounding box dynamic detection algorithm to perform intersection tests between each predicted trajectory of the pedestrian and the planned path of the host vehicle, and identify potential collision pedestrian trajectories;

[0007] Step 3: Calculate the specific parameters of the pedestrian-vehicle conflict when the pedestrian moves along the potential collision trajectory and when there is no potential collision trajectory respectively;

[0008] Step 4: Use the bivariate extreme value theory of exceeding threshold extreme values to model the time urgency of potential collision occurrence, and use the catastrophe theory to model the space urgency of collision occurrence, and calculate the time urgency and space urgency of pedestrian-vehicle collision;

[0009] Step 5: Calculate the comprehensive risk degree of the pedestrian, and divide the safe driving area according to the comprehensive risk degree and the pedestrian-vehicle safety distance.

[0010] Preferably, the specific method for obtaining pedestrian motion information to predict the pedestrian multi-modal trajectory is as follows:

[0011] Step 101: Obtain the pose information of each surrounding pedestrian through an in-vehicle vision sensor Specifically:

[0012]

[0013] Among them, is the pose information of pedestrian i at time t, is the position coordinate of pedestrian i at time t, is the speed of pedestrian i at time t, is the acceleration of pedestrian i at time t, is the orientation angle of pedestrian i at time t;

[0014] Step 102: Encode the historical pose information of the pedestrian using a GRU network, extract the interaction factors between pedestrians through the GRU network, and input the encoded vector into the Softmax function to calculate the probabilities P k (m i (m k ) of different decisions of the pedestrian, where k = 1, 2, 3, m1 is not crossing the street, m2 is crossing the street perpendicular to the vehicle driving direction, and m3 is crossing the street at an acute angle to the vehicle driving direction;

[0015] Step 103: Use the GRU decoder to generate the probability distribution P i (t') of the pedestrian position at time t', the future trajectory coordinate sequence of the pedestrian, and the motion characteristics

[0016]

[0017]

[0018] Among them is the predicted position coordinate of pedestrian i at time t', is the predicted speed of pedestrian i at time t', is the predicted acceleration of pedestrian i at time t', is the predicted orientation angle of pedestrian i at time t';

[0019] Step 104: Obtain the coordinate sequence Traj v (t) of the host vehicle traveling along the current trajectory and the vehicle motion characteristics Motion v (t):

[0020]

[0021]

[0022] Among them, is the position coordinate of the host vehicle at time t, is the speed of the host vehicle at time t, is the acceleration of the host vehicle at time t, is the heading angle of the host vehicle at time t.

[0023] Preferably, the specific method for discriminating the trajectory of a potential collision pedestrian is as follows:

[0024] Step 201: Model the collision volumes of pedestrians and autonomous vehicles;

[0025] Step 202: Calculate the bounding boxes of pedestrians and vehicles, and obtain the maximum and minimum parameter values of the bounding boxes of pedestrians and vehicles;

[0026] Step 203: Check the overlapping situation of the projections of the two bounding boxes. If it is determined to intersect, return the current time, the j-th trajectory of the i-th pedestrian where the center coordinates of the pedestrian bounding box and the vehicle bounding box are located, and the probability of this trajectory, and mark this trajectory as a potential collision trajectory. If it is determined not to intersect, proceed to the detection of the next time step;

[0027] Step 204: Update the coordinate point information of pedestrians and vehicles at the next time step in combination with the predicted pedestrian trajectory and the autonomous driving planned path, update the bounding boxes accordingly according to the transformation relationship, continuously perform collision detection until all trajectory points at all times are traversed, screen out all potential collision trajectories with intersections, obtain the set of potential collision trajectories and the set of trajectories without collision risks, and sum up the probabilities of all potential collision trajectories returned by Step 203 to obtain the probability of the potential collision set.

[0028] Preferably, the specific method for modeling the collision volumes of pedestrians and autonomous vehicles is as follows:

[0029] A circular collision volume with the pedestrian's center of gravity as the center and a radius of 0.3 meters is established; a rectangular collision volume with the vehicle's centroid as the center, a length of 5 meters, and a width of 2 meters is established. Then, the circular collision volume of the pedestrian is defined as the pedestrian bounding box, and the center of the bounding box of the coordinate point of the j-th trajectory of the i-th pedestrian at time t is the pedestrian's center of gravity. with a radius of 0.3 m, and a rectangular bounding box defined with the vehicle's centroid as the center, a length of 5 meters, and a width of 2 meters.

[0030] Preferably, in step 3, the specific parameters of the pedestrian-vehicle conflict are calculated for two types of pedestrian trajectories respectively. The specific steps include:

[0031] Calculate the time difference of collision TDTC and the time to collision TTC j of pedestrian i moving along the potential collision trajectory j :

[0032]

[0033]

[0034] In the formula: l v is the distance of the motor vehicle reaching the potential collision area; is the distance of pedestrian i moving along the j-th trajectory to the potential collision area; L is the length of the rectangular vehicle collision volume; v v is the instantaneous speed of the motor vehicle; is the instantaneous speed of the pedestrian;

[0035] Calculate the minimum encounter distance min d and the angle change rate j of pedestrian i moving along the trajectory without potential collision danger

[0036] The coordinate of the vehicle's trajectory point at time t is The coordinate of pedestrian i on the j-th trajectory is The Euclidean distance O v O ij is:

[0037]

[0038] The angle between the pedestrian and the vehicle can be represented by the arctangent value of the angle between the line connecting the center coordinates of the pedestrian and the vehicle collision volume O v O ij and the y-axis. The calculation formula is:

[0039]

[0040] The calculation formula of the angle change rate is:

[0041]

[0042] is the included angle between the vehicle and the pedestrian at time t - Δt;

[0043] The rectangular collision volume of the vehicle is divided into four regions A1, A2, A3, and A4. Among them, A1 is the triangular region enclosed by the connection lines between the rectangular centroid and the two top corners on the side close to the pedestrian, A2 is the triangular region enclosed by the connection lines between the rectangular centroid and the two top corners on the side far from the pedestrian, A3 is the triangular region enclosed by the connection lines between the rectangular centroid and the two top corners at the front end of the vehicle's forward direction, and A4 is the triangular region enclosed by the connection lines between the rectangular centroid and the two top corners at the rear end of the vehicle's forward direction. The region where the line segment O v d v is located is determined by the included angle tanθ between the vehicle and the pedestrian. The line segment O v d v specifically refers to the line segment connecting the central coordinates of the vehicle - pedestrian collision volume O v O ij inside the rectangular vehicle collision volume. The actual distance d between the pedestrian and the vehicle is determined in different cases as follows: j (t) is as follows:

[0044] ① When the line segment O v d v is located in A1 or A2, O v O ij intersects with the long side of the rectangular collision volume. At this time then calculate O v d v according to similar triangles:

[0045]

[0046] W is the width of the rectangular vehicle collision volume. In this article, W = 2.

[0047] At this time, the calculation formula for the actual distance d j (t) between the pedestrian and the vehicle is:

[0048]

[0049] r is the radius of the circular pedestrian collision volume. In this article, r = 0.3.

[0050] ② When the line segment O v d v is located in A3 or A4, O v O ij intersects with the short side of the rectangular collision volume. At this time then calculate O v d v according to similar triangles:

[0051]

[0052] L is the length of the rectangular vehicle collision volume;

[0053] At this time, the true distance d between the pedestrian and the vehicle j (t) is calculated by the formula:

[0054]

[0055] Traverse the multi-modal pedestrian trajectories to obtain the minimum vehicle-pedestrian encounter distance min d under each pedestrian trajectory j .

[0056] Preferably, the bivariate extreme value theory of exceedance threshold extreme values is used to model the time urgency of potential collisions, and the catastrophe theory is used to model the space urgency of collisions. The specific steps for calculating the time urgency and space urgency of vehicle-pedestrian collisions are as follows:

[0057] Step 401: Let {(x1,y1),(x2,y2),…,(x n ,y n )} be a set of independently observed bivariate vectors (X,Y) with a cumulative distribution function F(x,y), and the variables (x,y) exceeding the boundary thresholds u x and u y approximate the generalized Pareto distribution of the tail:

[0058]

[0059]

[0060] where ξ x ,ξ y is the shape parameter, σ x ,σ y is the scale parameter, ζ x =P r {x>u x},ζ y =P r {y>u y}.

[0061] The variables (X,Y) are transformed by exceedance threshold to:

[0062]

[0063]

[0064] The joint distribution function has marginal distributions that approximate the standard Frechet. The joint distribution of exceedance threshold extreme values is obtained from the following formula:

[0065]

[0066]

[0067] Among them, H is the mean distribution function of [0, 1], satisfying

[0068] Step 402: The specific parameters of the pedestrian-vehicle conflict when the pedestrian is moving on the potential collision trajectory Take the negative mapping, and determine the threshold u according to the super-threshold expectation map of the sample x and u y . The super-threshold expectation function e(u) is defined as:

[0069]

[0070] Among them, X j is the negative or negative value of the j-th trajectory of pedestrian i; G' is the total number of potential collision trajectories of pedestrian i;

[0071] Use maximum likelihood estimation to estimate the parameters of the generalized Pareto distribution.

[0072] Step 403: The time urgency index R of the collision i is the probability of predicting and being less than the threshold, that is:

[0073] R i = P r (x > u x ∩ y > u y )

[0074] Step 406: For the trajectories without collision risk Construct a cusp catastrophe model to predict the mutation of the pedestrian's motion state. The potential function of the cusp catastrophe is constructed as follows:

[0075] V(x) = x 4 + u1x 2 + u2x

[0076] Among them, V(z) is the potential function of the trajectory collision, x is the system state variable, and u1, u2 are the control variables of the system.

[0077] Step 407: Calculate the mutation flow form and the singularity set of the equilibrium surface of the cusp catastrophe:

[0078] V'(x) = 4x 3 + 2u1x + u2 = 0

[0079] V”(x) = 12x 2 + 2u1 = 0

[0080] Step 408: Normalize the relationships between the control variables u1, u2 and the state variable x to obtain:

[0081]

[0082] Step 409: Calculate the vehicle-pedestrian collision space urgency index C for the j-th pedestrian trajectory with no potential collision hazard i,j :

[0083]

[0084] Calculate the space urgency index C of the trajectory set : i :

[0085] C i = C i,1 + (1 - C i,1 )C i,2 + … + {1 - C i,1 - (1 - C i,1 )C i,2 - … - [1 - C i,1 - (1 - C i,1 )C i,2 - …]C i,(n-1) )}C i,n .

[0086] Preferably, calculate the comprehensive risk degree RPC of pedestrian i i , and divide the safe driving area according to RPC i and the vehicle-pedestrian safety distance. The specific steps include:

[0087] Step 501: Establish a collision risk degree model for pedestrians:

[0088] RPC i = αP ic + βR i + γC i

[0089] In the formula, P ic is the potential collision probability, R i is the time urgency index, C i is the space urgency index, and α, β, γ are index weights;

[0090] Step 502: Establish a vehicle-pedestrian safety distance model and calculate as follows:

[0091]

[0092]

[0093] Among them, v pi-x is the lateral speed of the pedestrian, is the time required for pedestrian i to walk to the potential collision area, B i is the lateral distance between the vehicle and pedestrian i, W is the vehicle width, D i-safe is the safety distance between the vehicle and pedestrian i, v v is the current vehicle speed, τ is the time delay, and d0 is the longitudinal distance between the vehicle and the pedestrian after the vehicle stops;

[0094] Step 503: According to the RPC model obtained in step 501 and the vehicle - pedestrian safety model obtained in step 502, calculate the risk degree RPC of all pedestrians at the current moment i and the safety distance D i-safe , and use visualization technology to assign the safety distance corresponding to pedestrian i to the corresponding RPC i value projection to the drivable area in front of the autonomous driving, and after superimposing all pedestrian RPC i and the safety distance, use colors to distinguish different risk - degree distances to form a visualized safe driving area.

[0095] Compared with the prior art, the significant advantages of the present invention are as follows:

[0096] (1) The present invention proposes a new framework for evaluating the dangerous situation of vehicle - pedestrian collision by integrating multiple methods, taking into account the sudden changes and extremes in vehicle - pedestrian collisions, and being more adaptable to complex traffic environments.

[0097] (2) Based on the multi - modal trajectories of pedestrians, fully considering all possible trajectories of pedestrians, the present invention improves the limitations of the existing vehicle - pedestrian collision risk prediction, such as incomplete consideration of factors and relatively subjective evaluation mechanisms, and is more adaptable to autonomous driving.

[0098] (3) The formed driving safety area of the present invention has guiding significance for the scientific decision - making and planning of autonomous driving.

[0099] Other features and advantages of the present invention will be described in the subsequent specification, and some of them will become obvious from the specification, or be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in the written specification, claims, and drawings. Brief Description of the Drawings

[0100] The drawings are only for the purpose of showing specific embodiments, and are not considered as limitations of the present invention. Throughout the drawings, the same reference signs represent the same components.

[0101] Figure 1 is a schematic diagram of the steps of a vehicle - pedestrian collision risk prediction method based on spatio - temporal urgency according to the present invention.

[0102] Figure 2 Schematic diagram for modeling the volume of vehicle - pedestrian collision.

[0103] Figure 3 Schematic diagram for discriminating the predicted trajectory of pedestrians.

[0104] Figure 4 Thermal diagram of pedestrian trajectory distribution

[0105] Figure 5 Schematic diagram for calculating the vehicle - pedestrian conflict parameter index.

[0106] Figure 6 Schematic diagram of the driving safety area. Detailed implementation manners

[0107] It is easy to understand that according to the technical solution of the present invention, without changing the essence of the present invention, those of ordinary skill in the art can imagine various implementation manners of the present invention. Therefore, the following detailed implementation manners and drawings are only exemplary descriptions of the technical solution of the present invention, and should not be regarded as all of the present invention or as a limitation or restriction on the technical solution of the present invention. On the contrary, the purpose of providing these embodiments is to enable those skilled in the art to more thoroughly understand the present invention. The preferred embodiments of the present invention will be specifically described below with reference to the drawings, in which the drawings form a part of this application and are used together with the embodiments of the present invention to illustrate the innovative concept of the present invention.

[0108] The concept of the present invention is a vehicle - pedestrian collision risk prediction method considering spatio - temporal urgency. According to the predicted multi - modal trajectories of pedestrians, it discriminates the pedestrian trajectories with potential collision risks and those without potential collision risks, and proposes a collaborative quantification of spatio - temporal urgency. As Figure 1 shown, the specific steps are as follows:

[0109] Step 1: Obtain pedestrian motion information and predict the multi - modal trajectories of pedestrians;

[0110] Step 2: Discriminate the pedestrian trajectories with potential collisions according to the multi - modal trajectories of pedestrians obtained in Step 1;

[0111] Step 3: Calculate the vehicle - pedestrian conflict parameters according to the data obtained in Step 1;

[0112] Step 4: Calculate the time urgency and space urgency of vehicle - pedestrian collisions according to the data obtained in Step 3;

[0113] Step 5: Calculate the pedestrian risk level according to the results obtained in Step 4 to obtain the safety driving classification area.

[0114] Furthermore, the pedestrian motion parameters are obtained through in - vehicle sensors, and the multi - modal trajectories of pedestrians are predicted through a pedestrian decision - making model and a GRU network. The specific steps include:

[0115] Step 101: Obtain the pose information of each surrounding pedestrian through an in-vehicle vision sensor Specifically:

[0116]

[0117] Among them, is the pose information of pedestrian i at time t, is the position coordinate of pedestrian i at time t, is the speed of pedestrian i at time t, is the acceleration of pedestrian i at time t, is the orientation angle of pedestrian i at time t.

[0118] Step 102: Encode the historical pose information of pedestrians using a GRU network, extract the interaction factors between pedestrians through the GRU network, and input the encoded vector into the Softmax function to calculate the probabilities P k (m i (m k ), k = 1, 2, 3, where m1 is not crossing the street, m2 is crossing the street perpendicular to the vehicle driving direction, and m3 is crossing the street at an acute angle to the vehicle driving direction.

[0119] Step 103: Use a GRU decoder to generate the probability distribution P i (t'), the future trajectory coordinate sequence of the pedestrian and the motion characteristics

[0120]

[0121]

[0122] Among them is the predicted position coordinate of pedestrian i at time t', is the predicted speed of pedestrian i at time t', is the predicted acceleration of pedestrian i at time t', is the predicted orientation angle of pedestrian i at time t'.

[0123] Step 104: Obtain the coordinate sequence Traj v (t) of the ego vehicle driving along the current trajectory and the vehicle motion characteristics Motion v (t):

[0124]

[0125]

[0126] Among them, is the position coordinate of the host vehicle at time t, is the speed of the host vehicle at time t, is the acceleration of the host vehicle at time t, is the heading angle of the host vehicle at time t.

[0127] Furthermore, as shown in Figure 3 , an intersection test between each predicted trajectory of the pedestrian and the planned path of the host vehicle is performed using the bounding box dynamic detection algorithm to identify potential collision pedestrian trajectories. The specific steps include:

[0128] Step 201: Model the collision volumes of the pedestrian and the autonomous vehicle. Based on the study of pedestrian traffic body dimensions, considering the hand swing and foot movement of the pedestrian's movement, a circular collision volume with a radius of 0.3 meters centered on the pedestrian's center of gravity is established; a rectangular collision volume with a length of 5 meters and a width of 2 meters centered on the vehicle's centroid is established, as shown in Figure 2 .

[0129] Step 202: Calculate the bounding boxes of the pedestrian and the vehicle: Since the established pedestrian collision volume is circular, the circular collision volume of the pedestrian is defined as the pedestrian bounding box box1, and the center of the bounding box of the j-th trajectory of the i-th pedestrian at time t is the pedestrian's center of gravity with a radius of 0.3m. The circular bounding box can reduce the calculation increased by the pedestrian's turning movement and reflect its excellent characteristics; since the vehicle collision volume is rectangular and it is assumed that the current planned path of the autonomous vehicle is a uniform straight line without considering vehicle turning, a rectangular AABB bounding box box2 with a length of 5 meters and a width of 2 meters centered on the vehicle's centroid is defined. Obtain the maximum and minimum parameter values of the pedestrian and vehicle bounding boxes:

[0130] box 1-min =(x 1-min ,y 1-min )

[0131] box 1-max =(x 1-max ,y 1-max )

[0132] box 2-min =(x 2-min ,y 2-min )

[0133] box 2-max =(x 2-max ,y 2-max )

[0134] Step 203: Check the overlapping situation of the projections of the two bounding boxes in the x and y axis directions:

[0135] If x 1-max> x 2-max and y 1-max > y 2-max , then box1 is in the upper right front of box2. At this time, let

[0136] x minMax = x 2-max , y minMax = y 2-max , x Maxmin = x 1-max , y Maxmin = y 1-max ,

[0137] Then if x minMax > x Maxmin and y minMax > y Maxmin , it is determined that the two intersect. The same reasoning can be applied to other orientation cases and will not be elaborated here. If it is determined to intersect, return the current time, the center coordinates of box1 and box2 (defining the collision area), the j-th trajectory of the i-th pedestrian where the center coordinate of box1 is located and the probability of this trajectory and mark this trajectory as a potential collision trajectory. If it is determined not to intersect, enter the detection of the next time step.

[0138] Step 204: Combine the predicted pedestrian trajectories and the autonomous driving planned path to update the pedestrian and vehicle coordinate point information at the next time step, and update the bounding box accordingly. Continuously perform collision detection until all trajectory points at all times are traversed, filter out all potential collision trajectories with intersections, and obtain the set of potential collision trajectories and the set of trajectories with no collision risk Sum the probabilities of all potential collision trajectories returned in step 203 to obtain the probability P of the potential collision set ic , and generate a heat map of the pedestrian trajectory distribution, as shown in Figure 4 .

[0139] Furthermore, step 3 calculates the specific parameters of the pedestrian-vehicle conflict under two types of pedestrian trajectories respectively. The specific steps include:

[0140] Step 301: Calculate the collision time difference TDTC and the time to collision TTC j when pedestrian i moves along the potential collision trajectory j :

[0141]

[0142]

[0143] In the formula: l vis the distance of the motor vehicle to the potential collision area; is the distance of pedestrian i moving along the j-th trajectory to the potential collision area; L is the vehicle length of the motor vehicle; v v is the instantaneous speed of the motor vehicle; is the instantaneous speed of the pedestrian.

[0144] Step 302, calculate the minimum encounter distance mind when the pedestrian i moves along the trajectory with no potential collision danger j , the angle change rate Figure 5 is the schematic diagram for index calculation.

[0145] At time t, the coordinate of the vehicle's trajectory point is The coordinate of pedestrian i on the j-th trajectory is The Euclidean distance O v O ij is:

[0146]

[0147] The angle between the pedestrian and the vehicle can be represented by the arctangent value of the angle between the line connecting the center coordinates of the pedestrian-vehicle collision volume O v O ij and the y-axis. The calculation formula is:

[0148]

[0149] The calculation formula for the angle change rate is:

[0150]

[0151] where is the angle between the pedestrian and the vehicle at time t - Δt.

[0152] Divide the rectangular collision volume of the vehicle into four regions A1, A2, A3, and A4. Among them, A1 is the triangular region enclosed by the line connecting the centroid of the rectangle and the two top corners on the side of the rectangle close to the pedestrian, A2 is the triangular region enclosed by the line connecting the centroid of the rectangle and the two top corners on the side of the rectangle far from the pedestrian, A3 is the triangular region enclosed by the line connecting the centroid of the rectangle and the two top corners at the front end of the forward direction of the rectangle, and A4 is the triangular region enclosed by the line connecting the centroid of the rectangle and the two top corners at the rear end of the forward direction of the rectangle. Determine the region where the line segment O v d v is located through the angle between the pedestrian and the vehicle tanθ. The line segment O v d v specifically refers to the line segment of the connection line of the center coordinates of the pedestrian-vehicle collision volume O v O ij inside the rectangular vehicle collision volume. Discuss the actual distance d between the pedestrian and the vehicle in different cases j(t) is as follows:

[0153] ① When the line segment O v d v is located in A1 or A2, O v O ij intersects the long side of the rectangular collision volume. At this time then calculate O according to similar triangles v d v :

[0154]

[0155] W is the width of the rectangular vehicle collision volume. In this paper, W = 2 is taken.

[0156] At this time, the true distance d between the pedestrian and the vehicle j (t) is calculated by the formula:

[0157]

[0158] r is the radius of the circular pedestrian collision volume. In this paper, r = 0.3 is taken.

[0159] ② When the line segment O v d v is located in A3 or A4, O v O ij intersects the short side of the rectangular collision volume. At this time then calculate O according to similar triangles v d v :

[0160]

[0161] L is the length of the rectangular vehicle collision volume. In this paper, L = 5 is taken.

[0162] At this time, the true distance d between the pedestrian and the vehicle j (t) is calculated by the formula:

[0163]

[0164] Traverse the multi-modal pedestrian trajectories to obtain the minimum vehicle-pedestrian encounter distance min d under each pedestrian trajectory j .

[0165] Furthermore, step 4 specifically includes using the bivariate extreme value theory of exceeding thresholds to model the time urgency of potential collisions; using catastrophe theory to model the space urgency of collisions. The specific steps are as follows:

[0166] Step 401. Let {(x1,y1),(x2,y2),…,(x n ,yn )} is a set of independent observed bivariate vectors (X, Y) with a cumulative distribution function F(x, y), exceeding the boundary thresholds u x and u y The variables (x, y) approximately follow a generalized Pareto distribution in the tail:

[0167]

[0168]

[0169] where ξ x , ξ y is the shape parameter, σ x , σ y is the scale parameter, ζ x = P r {x > u x}, ζ y = P r {y > u y}.

[0170] The variables (X, Y) are transformed by peaks over threshold to:

[0171]

[0172]

[0173] The joint distribution function has marginal distributions that are approximately standard Frechet. The joint distribution of peaks over threshold extremes is given by:

[0174]

[0175]

[0176] where H is the mean distribution function on [0, 1], satisfying

[0177] Step 402, Take the negative mapping of and determine the thresholds u x and u y according to the exceedance expectation plot of the samples. The exceedance expectation function e(u) is defined as:

[0178]

[0179] where X j is the negative or negative value of the j-th trajectory of pedestrian i, and G' is the total number of potential collision trajectories of pedestrian i.

[0180] Estimate the parameters of the generalized Pareto distribution using maximum likelihood estimation.

[0181] Step 403, the time urgency index R of the collision i To predict and The probability of being less than the threshold, i.e.:

[0182] R i = P r (x > u x ∩ y > u y )

[0183] Step 406, for the trajectory without collision risk Construct a cusp catastrophe model to predict the mutation of the pedestrian's motion state. The potential function of the cusp catastrophe is constructed as follows:

[0184] V(x) = x 4 + u1x 2 + u2x

[0185] Among them, V(x) is the potential function of the trajectory collision, x is the system state variable, and u1, u2 are the system control variables.

[0186] Step 407, calculate the mutation flow form and the equilibrium surface singularity set of the cusp catastrophe:

[0187] V'(x) = 4x 3 + 2u1x + u2 = 0

[0188] V”(x) = 12x 2 + 2u1 = 0

[0189] Step 408, normalize the relationship between the control variables u1, u2 and the state variable x to obtain:

[0190]

[0191] Step 409, calculate the spatial urgency index of the vehicle-pedestrian collision under the j-th pedestrian trajectory without potential collision risk:

[0192] Based on this, calculate the spatial urgency index C of the trajectory set : i :

[0193] C i = C i,1 +(1 - C i,1 )C i,2 +…+{1 - C i,1 -(1 - C i,1 )C i,2 -…-[1 - Ci,1 -(1 - C i,1 )C i,2 -…]C i,n-1) )}C i,n

[0194] Furthermore, in step 5, the comprehensive risk degree RPC of pedestrian i is calculated i , and the safe driving area is divided according to RPC i and the safe distance between the vehicle and the pedestrian. The specific steps are as follows:

[0195] Step 501: Establish a risk of pedestrian collision model (RPC):

[0196] RPC i = αP ic + βR i + βC i

[0197] where P ic is the potential collision probability, R i is the time urgency index, C i is the space urgency index, and α, β, γ are index weights, and their values are determined by the entropy weight method.

[0198] Step 502: Establish a safe distance model between the vehicle and the pedestrian, and the calculation is as follows:

[0199]

[0200]

[0201] where v pi-x is the lateral speed of the pedestrian, is the time required for pedestrian i to walk to the potential collision area, B i is the lateral distance between the vehicle and pedestrian i, W is the vehicle width, D i-safe is the safe distance between the vehicle and pedestrian i, v v is the current vehicle speed, τ is the time delay, and d0 is the longitudinal distance between the vehicle and the pedestrian after the vehicle stops.

[0202] Step 503: According to the RPC model obtained in step 501 and the vehicle - pedestrian safety model obtained in step 502, calculate the risk degree RPC i of all pedestrians at the current moment and the safe distance D i-safe , and use visualization technology to assign the safe distance corresponding to pedestrian i to the corresponding RPC i value and project it on the drivable area in front of the autonomous vehicle, and superimpose all pedestrians' RPC iAfter the safe distance, different risk distances are distinguished by colors to form a visual safe driving area, such as Figure 6 shown.

[0203] A prediction method for the risk situation of pedestrian-vehicle collision based on spatio-temporal urgency provides a basis for the safety decision-making theory of autonomous vehicles. Based on the multi-modal trajectories of pedestrians, the trajectories with potential collision risks are discriminated. Aiming at the mutation and extreme phenomena of pedestrian-vehicle collisions, the risk of pedestrian-vehicle collisions is quantitatively analyzed synergistically from two dimensions of time and space, and a risk situation analysis model with multiple dimensions and multiple indicators is proposed. The driving intervals are divided according to the pedestrian risk index and the safe distance index, and the safe driving area for autonomous driving is generated to improve the safety performance of autonomous vehicles.

[0204] As described above, only the preferred specific embodiments of the present invention are provided, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

[0205] It should be understood that, in order to streamline the present invention and help those skilled in the art understand various aspects of the present invention, in the above description of the exemplary embodiments of the present invention, the various features of the present invention are sometimes described in a single embodiment or with reference to a single figure. However, the present invention should not be construed as including the features included in the exemplary embodiments as essential technical features of the claims of this patent.

[0206] It should be understood that the modules, units, components, etc. included in the device of an embodiment of the present invention can be adaptively changed to be arranged in a device different from that of this embodiment. The different modules, units or components included in the device of the embodiment can be combined into one module, unit or component, or they can be divided into multiple sub-modules, sub-units or sub-components.

Claims

1. A method for predicting the dangerous situation of vehicle - pedestrian collision based on spatio - temporal urgency, characterized in that, The specific steps are as follows: Step 1: Obtain pedestrian motion parameters, and predict the multi-modal trajectory of pedestrians through a pedestrian decision-making model and a GRU network. Specifically: Step 101: Obtain the pose information of each surrounding pedestrian through an in-vehicle vision sensor; Step 102: Encode the historical pose information of pedestrians using a GRU network, extract the interaction factors between pedestrians through the GRU network, and input the encoded vector into the Softmax function to calculate the probabilities P k (m i (m k ), k = 1, 2, 3, where m1 is not crossing the street, m2 is crossing the street perpendicular to the vehicle driving direction, and m3 is crossing the street at an acute angle to the vehicle driving direction; Step 103: Use a GRU decoder to generate the probability distribution of the pedestrian position, the coordinate sequence of the pedestrian's future trajectory, and the motion characteristics at time t'; Step 104: Obtain the coordinate sequence of the ego vehicle driving along the current trajectory and the vehicle motion characteristics; Step 2: Adopt a bounding box dynamic detection algorithm to perform an intersection test between each predicted trajectory of the pedestrian and the planned path of the ego vehicle, and identify the pedestrian trajectories with potential collisions; Step 3: Calculate the specific parameters of the vehicle-pedestrian conflict when the pedestrian moves along the potential collision trajectory and when there is no potential collision trajectory respectively; Step 4: Use the bivariate extreme value theory of exceedance thresholds to model the time urgency of potential collisions, and use catastrophe theory to model the space urgency of collisions, and calculate the time urgency and space urgency of vehicle-pedestrian collisions; Step 5: Calculate the comprehensive risk degree of the pedestrian, and divide the safe driving area according to the comprehensive risk degree and the vehicle-pedestrian safety distance.

2. The method for predicting the dangerous situation of vehicle-pedestrian collision based on spatio-temporal urgency according to claim 1, wherein Obtain the pose information of each surrounding pedestrian through an in-vehicle vision sensor Specifically: Among them, is the pose information of pedestrian i at time t, is the position coordinate of pedestrian i at time t, is the speed of pedestrian i at time t, is the acceleration of pedestrian i at time t, is the orientation angle of pedestrian i at time t; The probability distribution P of the pedestrian's position at time t' generated by the GRU decoder i (t'), the sequence of future trajectory coordinates of the pedestrian and the motion characteristics are respectively as follows: where is the predicted position coordinates of pedestrian i at time t', is the predicted speed of pedestrian i at time t', is the predicted acceleration of pedestrian i at time t', is the predicted orientation angle of pedestrian i at time t'; The coordinate sequence Traj v of the host vehicle traveling along the current trajectory v and the vehicle motion feature Motion (t) are respectively: Among them, is the position coordinate of the vehicle itself at time t, is the speed of the vehicle itself at time t, is the acceleration of the vehicle itself at time t, is the heading angle of the vehicle itself at time t.

3. The method for predicting the dangerous situation of vehicle-pedestrian collision based on spatio-temporal urgency according to claim 2, wherein The specific method for identifying the pedestrian trajectories with potential collisions is as follows: Step 201: Model the collision volumes of the pedestrian and the autonomous driving vehicle; Step 202: Calculate the bounding boxes of the pedestrian and the vehicle, and obtain the maximum and minimum parameter values of the bounding boxes of the pedestrian and the vehicle; Step 203: Check the overlapping condition of the projections of the two bounding boxes. If it is determined to be intersecting, return the current time, the center coordinates of the pedestrian bounding box and the vehicle bounding box, the j-th trajectory of the i-th pedestrian, and the probability of this trajectory, and mark this trajectory as a potential collision trajectory. If it is determined not to intersect, proceed to the detection of the next time step; Step 204: Combine the pedestrian predicted trajectory and the autonomous driving planned path to update the coordinate point information of the pedestrian and the vehicle at the next time step, and update the bounding box accordingly according to the transformation relationship, and continuously perform collision detection until all trajectory points at all times are traversed, filter out all potential collision trajectories with intersections, obtain the set of potential collision trajectories and the set of trajectories without collision risks, and sum the probabilities of all potential collision trajectories returned by Step 203 to obtain the probability of the potential collision set.

4. The method for predicting the dangerous situation of vehicle-pedestrian collision based on spatio-temporal urgency according to claim 3, characterized in that The specific method for modeling the collision volumes of the pedestrian and the autonomous driving vehicle is as follows: A circular collision volume with the pedestrian's center of gravity as the center and a radius of 0.3 meters is established; a rectangular collision volume with the vehicle's centroid as the center, 5 meters in length and 2 meters in width is established. Then, the circular collision volume of the pedestrian is defined as the pedestrian bounding box, and the center of the bounding box of the coordinate point of the j-th trajectory of the i-th pedestrian at time t is the pedestrian's center of gravity The radius is 0.3 m, and a rectangular bounding box with the vehicle's centroid as the center, 5 meters in length and 2 meters in width is defined, which is the coordinate of pedestrian i on the j-th trajectory.

5. The method for predicting the dangerous situation of vehicle-pedestrian collision based on spatio-temporal urgency according to claim 4, wherein, Step 3: Calculate the specific parameters of the vehicle-pedestrian conflict under two types of pedestrian trajectories respectively. The specific steps include: Calculate the collision time difference TDTC and the time to collision TTC of pedestrian i moving along the potential collision trajectory j and the time to collision TTC j :​ Where: l v is the distance of the motor vehicle reaching the potential collision area; is the distance of pedestrian i moving along the j-th trajectory to the potential collision area; L is the length of the rectangular vehicle collision volume; v v is the instantaneous speed of the motor vehicle; is the instantaneous speed of the pedestrian; Calculate the minimum encounter distance mind of pedestrian i moving along the trajectory with no potential collision risk for the time being and the angular rate of change j ​ The trajectory point coordinates of the vehicle at time t are The coordinates of pedestrian i on the j-th trajectory are The Euclidean distance O between two trajectory points v O ij is: The included angle between the pedestrian and the vehicle can be represented by the arctangent value of the included angle between the line connecting the central coordinates of the collision volume of the pedestrian and the vehicle O v O ij and the y-axis, and the calculation formula is: The calculation formula for the angle change rate is: is the angle between the vehicle and the person at time t-Δt; Divide the rectangular collision volume of the vehicle into four regions A1, A2, A3, and A4. Among them, A1 is the triangular region enclosed by the connection lines between the rectangular centroid and the two top corners on the near-pedestrian side of the rectangle; A2 is the triangular region enclosed by the connection lines between the rectangular centroid and the two top corners on the far-pedestrian side of the rectangle; A3 is the triangular region enclosed by the connection lines between the rectangular centroid and the two top corners at the front end in the forward direction of the rectangle; A4 is the triangular region enclosed by the connection lines between the rectangular centroid and the two top corners at the rear end in the forward direction of the rectangle. Determine the region where the line segment O v d v is located. The line segment O v d v specifically refers to the line segment connecting the central coordinates of the pedestrian-vehicle collision volume O v O ij inside the rectangular vehicle collision volume. Determine the true distance d j (t) between the pedestrian and the vehicle in different cases as follows: ①When the line segment O v d v is located in A1 or A2, O v O ij intersects the long side of the rectangular collision volume. At this time then calculate O v d v according to similar triangles: W is the width of the rectangular vehicle collision volume; The true distance d between the pedestrian and the vehicle at this time j (t) is calculated by the formula: r is the radius of the circular pedestrian collision volume; ②When the line segment O v d v is located in A3 or A4, O v O ij intersects the short side of the rectangular collision volume. At this time then calculate O v d v : L is the length of the rectangular vehicle collision volume; The true distance d between the pedestrian and the vehicle at this time j (t) is calculated by the formula: Traverse the multi-modal pedestrian trajectories to obtain the minimum vehicle-pedestrian encounter distance mind under each pedestrian trajectory j .

6. The method for predicting the dangerous situation of vehicle-pedestrian collision based on spatio-temporal urgency according to claim 5, characterized in that The specific steps for using the bivariate extreme value theory of exceedance thresholds to model the time urgency of potential collisions and using catastrophe theory to model the space urgency of collisions, and calculating the time urgency and space urgency of vehicle-pedestrian collisions are as follows: Step 401. Let \(\{(x_1,y_1),(x_2,y_2),\ldots,(x n ,y n )\}\) be a set of independently observed bivariate vectors \((X,Y)\) with a cumulative distribution function \(F(x,y)\). The variables \((x,y)\) exceeding the boundary thresholds \(u x \) and \(u y \) approximately follow a generalized Pareto distribution in the tail: where ξ x , ξ y is the shape parameter, σ x , σ y is the scale parameter, ζ x = P r {x > u x}, ζ y = P r {y > u y}; Perform an exceedance transformation on the variables (X, Y) to: Joint distribution function With marginal distributions approximated by the standard Frechet, the joint distribution of the exceedance extreme values is given by: Among them, H is the mean distribution function of [0, 1], satisfying Step 402: Specific parameters of the vehicle-pedestrian conflict when the pedestrian moves along the potential collision trajectory Take the negative mapping, and determine the threshold u according to the super-threshold expectation map of the samples x and u y , and the super-threshold expectation function e(u) is defined as: Among them, X j is the negative or negative value of the j-th trajectory of pedestrian i, and G' is the total number of potential collision trajectories of pedestrian i; Use maximum likelihood estimation to estimate the parameters of the generalized Pareto distribution; Step 403, Collision Time Urgency Index R i For predicting and The probability of being less than the threshold, i.e.: R i = P r (x > u x ∩ y > u y ) Step 406: For the trajectory with no collision risk for the time being Construct a cusp catastrophe model to predict the mutation of the pedestrian's motion state. The potential function of the cusp catastrophe is constructed as follows: V(x) = x 4 + u1x 2 + u2x Among them, V(z) is the potential function for the trajectory to collide, x is the system state variable, and u1, u2 are the control variables of the system; Step 407: Calculate the catastrophe flow form and the singularity set of the equilibrium surface of the cusp catastrophe: V'(x) = 4x 3 + 2u1x + u2 = 0 V”(x) = 12x 2 + 2u1 = 0 Step 408: Normalize the relationship between the control variables u1, u2 and the state variable x to obtain: Step 409: Calculate the vehicle-pedestrian collision space urgency index C for the j-th pedestrian trajectory with no potential collision risk i,j : Calculate the spatial tightness index C of the trajectory set i :​ C i = C i,1 +(1 - C i,1 )C i,2 +…+{1 - C i,1 -(1 - C i,1 )C i,2 -…-[1 - C i,1 -(1 - C i,1 )C i,2 -…]C i,(n-1) )}C i,n 。 7. The method for predicting the dangerous situation of vehicle - pedestrian collision based on spatio - temporal urgency according to claim 1, wherein Calculate the comprehensive risk degree RPC of pedestrian i i and divide the safe driving area according to RPC i and the safe distance between the vehicle and the pedestrian. The specific steps are as follows: Step 501: Establish a collision risk model for pedestrians: RPC i = αP ic + βR i + γC i where P ic is the potential collision probability, R i is the time urgency index, C i is the space urgency index, and α, β, γ are the index weights; Step 502: Establish a safe distance model for vehicles and pedestrians, and calculate as follows: Among them, is the pedestrian's lateral speed, is the time required for pedestrian i to walk to the potential collision area, B i is the lateral distance between the vehicle and pedestrian i, W is the vehicle width, D i-safe is the safety distance between the vehicle and pedestrian i, v v is the current vehicle speed, τ is the time delay, and d0 is the longitudinal distance between the vehicle and the pedestrian after the vehicle stops; Step 503: Calculate the danger level RPC of all pedestrians at the current moment based on the RPC model obtained in Step 501 and the pedestrian-vehicle safety model obtained in Step 502 i and the safe distance D i-safe , and use visualization technology to assign the safe distance corresponding to pedestrian i to the corresponding RPC i The value projection is made on the drivable area in front of the autonomous vehicle, and all pedestrian RPCs i After adding the safe distance, different danger level distances are distinguished by colors to form a visualized safe driving area.

Images