Vehicle behavior prediction method based on dual-channel Gaussian mixture hidden Markov model

By using a dual-channel Gaussian mixture hidden Markov model to predict vehicle motion behavior, this method solves the problem of low prediction accuracy caused by the failure to fully utilize the motion state information of surrounding vehicles in existing technologies, and achieves more accurate prediction of vehicle lane-changing behavior.

CN114771575BActive Publication Date: 2024-11-15TIANJIN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210529642.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-05-16
Publication Date
2024-11-15
Estimated Expiration
2042-05-16

Smart Images

  • Figure CN114771575B_ABST
    Figure CN114771575B_ABST
Patent Text Reader

Abstract

The present disclosure provides a vehicle behavior prediction method based on a dual-channel Gaussian mixture hidden Markov model, which can be applied to the fields of autonomous driving technology and vehicle assisted driving technology. The method includes: obtaining driving environment data of a driving vehicle within a preset time period; determining the motion characteristics of a target vehicle according to the driving environment data; determining the probability of the motion feature sequence through a dual-channel Gaussian mixture hidden Markov model according to a motion feature sequence composed of the motion features, wherein the motion feature sequence includes a first motion feature sequence and a second motion feature sequence, and the dual-channel Gaussian mixture hidden Markov model recursively obtains the probability of the motion feature sequence according to the probability of the motion features in the motion feature sequence; predicting the motion behavior of the target vehicle according to the probability of the motion feature sequence. The present disclosure also provides a vehicle behavior prediction device and electronic device based on a dual-channel Gaussian mixture hidden Markov model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the fields of autonomous driving technology and vehicle assisted driving technology, and more specifically, to a vehicle behavior prediction method, device and electronic equipment based on a dual-channel Gaussian mixture hidden Markov model. Background Art

[0002] During the driving process, intelligent decision-making is important for ensuring the driving safety of the vehicle. One aspect of this is to predict the possible lane-changing behavior of the vehicle based on the driving status of surrounding vehicles, thereby reminding the vehicle driver or the autonomous driving vehicle to respond in advance to prevent collisions with merging vehicles.

[0003] In the process of realizing the concept of the present disclosure, the inventors found that there are at least the following problems in the related art: when predicting the lane-merging behavior of a vehicle, the motion state information of the surrounding vehicles cannot be fully utilized, resulting in reduced accuracy of the prediction. Summary of the invention

[0004] In view of this, the present disclosure provides a vehicle behavior prediction method, device and electronic device based on a dual-channel Gaussian mixture hidden Markov model.

[0005] One aspect of the present disclosure provides a vehicle behavior prediction method based on a dual-channel Gaussian mixture hidden Markov model, comprising:

[0006] Acquire driving environment data of the driving vehicle within a preset time period, wherein the driving environment data includes motion state information of the driving vehicle, the target vehicle, and associated vehicles, the target vehicle is located in an adjacent lane ahead of the driving vehicle in the direction of travel, and the associated vehicles represent vehicles that affect the lane change of the target vehicle;

[0007] Determine the motion characteristics of the target vehicle according to the driving environment data, wherein the motion characteristics include a first motion characteristic and a second motion characteristic, the first motion characteristic characterizes the relative distance and relative speed of the target vehicle in the direction of the road, and the second motion characteristic characterizes the inverse of the time when the target vehicle collides with the driving vehicle and the associated vehicle, the position change of the target vehicle perpendicular to the direction of the road at adjacent moments, and the angle between the driving direction of the target vehicle and the direction of the road;

[0008] According to the motion feature sequence composed of motion features, the probability of the motion feature sequence is determined by a dual-channel Gaussian mixture hidden Markov model, wherein the motion feature sequence includes a first motion feature sequence and a second motion feature sequence, and the dual-channel Gaussian mixture hidden Markov model recursively obtains the probability of the motion feature sequence according to the probability of the motion features in the motion feature sequence;

[0009] The motion behavior of the target vehicle is predicted based on the probability of the motion feature sequence.

[0010] According to an embodiment of the present disclosure, obtaining driving environment data of a vehicle within a preset time period includes:

[0011] The driving environment data of the vehicle driven within a preset time period is read from the driving environment data stored in the queue memory of a fixed length.

[0012] According to an embodiment of the present disclosure, the motion state information includes position information and speed information, and the motion characteristics of the target vehicle are determined according to the driving environment data, including:

[0013] Determine, based on the position information and speed information of the driving vehicle, the target vehicle, and the associated vehicle, a first relative distance and a first relative speed between the target vehicle and the driving vehicle in the direction of the road, and a second relative distance and a second relative speed between the target vehicle and the associated vehicle in the direction of the road;

[0014] A first motion feature and a second motion feature are determined according to the first relative distance, the second relative distance, the first relative speed, and the second relative speed.

[0015] According to an embodiment of the present disclosure, the first motion feature and the second motion feature are determined according to the first relative distance, the second relative distance, the first relative speed, and the second relative speed, and are expressed by the following formula:

[0016]

[0017]

[0018] in, is the first motion feature, It is the second motion characteristic;

[0019] Δx o-e (t) are respectively the first relative distance, the first relative speed and the inverse of the collision time between the target vehicle and the driving vehicle;

[0020] Δx o-fe (t) are respectively the second relative distance, the second relative speed and the inverse of the collision time between the target vehicle and the associated vehicle in front of the driving vehicle in the forward direction of the road;

[0021] Δx o-fo (t) are respectively a second relative distance, a second relative speed and a reciprocal of the collision time between the target vehicle and the associated vehicle in front of the target vehicle in the forward direction of the road;

[0022] Δxo-ro (t) are respectively a second relative distance, a second relative speed and a reciprocal of the collision time between the target vehicle and the associated vehicle behind the target vehicle in the forward direction of the road;

[0023] Δy o (t) is the change in the position of the target vehicle perpendicular to the road direction at adjacent moments, ψ o (t) is the angle between the target vehicle’s direction of travel and the road’s direction of travel.

[0024] According to an embodiment of the present disclosure, the probability of the motion feature sequence is determined by a dual-channel Gaussian mixture hidden Markov model according to the motion feature sequence composed of motion features, including:

[0025] Determine the probability of the motion feature according to the hidden state corresponding to the motion feature;

[0026] According to the probability of the motion features, the probability of the motion feature sequence is determined.

[0027] According to an embodiment of the present disclosure, the motion behavior of the target vehicle is predicted based on the probability of the motion feature sequence, which is expressed as follows:

[0028]

[0029]

[0030]

[0031] in, is the first motion feature sequence, is the second motion feature sequence,

[0032] and In the first motion feature sequence In the channel, a Gaussian mixture hidden Markov model for predicting the target vehicle's merging behavior and a Gaussian mixture hidden Markov model for predicting the target vehicle's lane keeping behavior;

[0033] and In the second motion feature sequence In the channel, a Gaussian mixture hidden Markov model for predicting the target vehicle's merging behavior and a Gaussian mixture hidden Markov model for predicting the target vehicle's lane keeping behavior;

[0034] and Use and Predict the first motion feature sequence The probability of and Use and Predicting the second motion feature sequence The probability of , lg() is the logarithmic function with base 10;

[0035] and The first motion feature sequence The prediction score and the second motion feature sequence The prediction score, l lon , l mix and l0 are linear comparator parameters;

[0036] B P It is the behavior prediction result. LC indicates that the target vehicle performs lane merging behavior, and LK indicates that the target vehicle performs lane keeping behavior.

[0037] According to an embodiment of the present disclosure, the above method further includes:

[0038] According to the evaluation index, the evaluation index value of the dual-channel Gaussian mixture hidden Markov model is determined, wherein the evaluation index includes the warning recall rate, the average warning recall rate and the F1-average warning recall rate, and the warning recall rate represents the probability that the predicted result of the target vehicle behavior is consistent with the actual behavior of the target vehicle;

[0039] Evaluate Gaussian mixture hidden Markov model according to the evaluation index value.

[0040] According to an embodiment of the present disclosure, wherein evaluating a dual-channel Gaussian mixture hidden Markov model according to an evaluation index value includes:

[0041] The predictive performance of the dual-channel Gaussian mixture hidden Markov model is evaluated based on the warning recall rate and the average warning recall rate. The predictive performance represents the time interval between the moment when the target vehicle performs lane merging or lane keeping behavior and the moment when the target vehicle behavior is correctly predicted.

[0042] Another aspect of the present disclosure provides a vehicle behavior prediction device based on a dual-channel Gaussian mixture hidden Markov model, comprising:

[0043] An acquisition module is used to acquire driving environment data of the driving vehicle within a preset time period, wherein the driving environment data includes motion state information of the driving vehicle, the target vehicle, and the associated vehicles, the target vehicle is located in an adjacent lane ahead of the driving vehicle in the direction of the road, and the associated vehicles represent vehicles that affect the lane change of the target vehicle;

[0044] A first determination module is used to determine the motion characteristics of the target vehicle according to the driving environment data, wherein the motion characteristics include a first motion characteristic and a second motion characteristic, the first motion characteristic represents the relative distance and relative speed of the target vehicle in the direction of the road, and the second motion characteristic represents the inverse of the time when the target vehicle collides with the driving vehicle and the associated vehicle, the position change of the target vehicle perpendicular to the direction of the road at adjacent moments, and the angle between the driving direction of the target vehicle and the direction of the road;

[0045] A second determination module is used to determine the probability of the motion feature sequence through a dual-channel Gaussian mixture hidden Markov model according to the motion feature sequence composed of motion features, wherein the motion feature sequence includes a first motion feature sequence and a second motion feature sequence, and the Gaussian mixture hidden Markov model recursively obtains the probability of the motion feature sequence according to the probability of the motion features in the motion feature sequence;

[0046] The prediction module is used to predict the motion behavior of the target vehicle based on the probability of the motion feature sequence.

[0047] Another aspect of the present disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more instructions, wherein when the one or more instructions are executed by the one or more processors, the one or more processors implement the above method.

[0048] According to an embodiment of the present disclosure, by acquiring driving environment data including motion state information of a driving vehicle, a target vehicle, and associated vehicles within a preset time period, the motion features of the target vehicle including a first motion feature and a second motion feature are determined, wherein the first motion feature represents the relative distance and relative speed between the target vehicle and the driving vehicle and the associated vehicle in the forward direction of the road, and the second motion feature represents the inverse time of the collision between the target vehicle and the driving vehicle and the associated vehicle, and then the probability of the motion feature sequence is recursively obtained according to the probability of the motion features in the motion feature sequence through a dual-channel Gaussian mixture hidden Markov model, and the motion behavior of the target vehicle is predicted. By acquiring driving environment data including the target vehicle and the associated vehicle to determine the motion features through a driving vehicle, and then predicting through a dual-channel Gaussian mixture hidden Markov model, the motion state information of the surrounding vehicles can be fully utilized to calculate the prediction results through two different motion features, at least partially overcoming the technical problem that the motion state information of the surrounding vehicles cannot be fully utilized when predicting the merging behavior of the vehicle, thereby reducing the accuracy of the prediction, and improving the accuracy of the vehicle behavior prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The above and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0050] Figure 1 A flowchart of a vehicle behavior prediction method based on a dual-channel Gaussian mixture hidden Markov model according to an embodiment of the present disclosure is schematically shown;

[0051] Figure 2 A schematic diagram of a driving environment of a vehicle obtained according to an embodiment of the present disclosure is schematically shown;

[0052] Figure 3 A schematic diagram schematically shows a single Gaussian mixture hidden Markov model structure according to an embodiment of the present disclosure;

[0053] Figure 4 The flowchart of the vehicle behavior prediction method based on the dual-channel Gaussian mixture hidden Markov model according to the embodiment of the present disclosure is schematically shown;

[0054] Figure 5 Schematic diagrams of early warning recall rate curves of three different behavior prediction methods according to embodiments of the present disclosure are schematically shown;

[0055] Figure 6 A schematic diagram of a ROC curve when the queue memory length takes different values ​​according to an embodiment of the present disclosure is schematically shown;

[0056] Figure 7 Schematically shows the ROC curve diagrams of the prediction method according to the embodiment of the present disclosure and four comparative methods;

[0057] Figure 8 A block diagram schematically shows a vehicle behavior prediction device based on a dual-channel Gaussian mixture hidden Markov model according to an embodiment of the present disclosure; and

[0058] Fig. 9 A block diagram of an electronic device suitable for implementing the method described above according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0059] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0060] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0061] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0062] In the case of using expressions such as "at least one of A, B, and C, etc.", it should generally be interpreted in accordance with the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.). In the case of using expressions such as "at least one of A, B, or C, etc.", it should generally be interpreted in accordance with the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B, or C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0063] In the related technology, the possible merging behavior of a vehicle can be predicted based on the driving status of the surrounding vehicles. However, in many cases, the merging vehicles do not strictly follow the traffic rules and use the turn signals. Therefore, it is not reliable to judge whether the merging behavior will occur based on the turn signals. Instead, it is necessary to predict the possible merging behavior in advance based on the motion status of the merging vehicles. There are two main problems with the current related prediction methods: on the one hand, they fail to fully utilize the motion status information of the surrounding vehicles; on the other hand, the accuracy of predicting vehicle behavior is low, and the time interval between the prediction time and the actual lane change behavior of the vehicle is short. Therefore, it is necessary to design a behavior prediction method to accurately predict the possible merging behavior in advance.

[0064] In view of this, an embodiment of the present disclosure provides a vehicle behavior prediction method based on a dual-channel Gaussian mixture hidden Markov model. The method includes obtaining driving environment data of a driving vehicle within a preset time period; determining the motion characteristics of a target vehicle according to the driving environment data; determining the probability of the motion characteristic sequence through a dual-channel Gaussian mixture hidden Markov model according to a motion characteristic sequence composed of the motion characteristics, wherein the motion characteristic sequence includes a first motion characteristic sequence and a second motion characteristic sequence, and the dual-channel Gaussian mixture hidden Markov model recursively obtains the probability of the motion characteristic sequence according to the probability of the motion characteristics in the motion characteristic sequence; and predicting the motion behavior of the target vehicle according to the probability of the motion characteristic sequence.

[0065] Figure 1 The flowchart of the vehicle behavior prediction method based on the dual-channel Gaussian mixture hidden Markov model according to the embodiment of the present disclosure is schematically shown.

[0066] like Figure 1 As shown, the method includes operations S101 to S104.

[0067] In operation S101, driving environment data of the driving vehicle within a preset time period is obtained, wherein the driving environment data includes motion state information of the driving vehicle, the target vehicle and the associated vehicles, the target vehicle is located in an adjacent lane ahead of the driving vehicle in the direction of travel, and the associated vehicles represent vehicles that affect the lane change of the target vehicle.

[0068] According to an embodiment of the present disclosure, the preset time period may be a period of time before the predicted moment, for example, 2 seconds or 5 seconds, which is not limited here.

[0069] According to the embodiments of the present disclosure, the driving vehicle may be a vehicle driven by a driver or a vehicle with an automatic driving function, the target vehicle may be a vehicle predicted to change lanes, or may be in an adjacent lane ahead of the driving vehicle in the direction of travel, and the associated vehicle may be a vehicle that affects the lane change of the target vehicle. For example, the driving vehicle is in the second lane on the left side of the direction of travel, the target vehicle may be a vehicle traveling in the first lane on the left side in front of the driving vehicle, and the associated vehicle may be a vehicle traveling in front of or behind the target vehicle in the first lane on the left side, or a vehicle traveling in front of the driving vehicle in the second lane on the left side.

[0070] According to an embodiment of the present disclosure, the motion state information may be speed information and position information of a driving vehicle, a target vehicle, or an associated vehicle on the road, and may be detected by a detection unit such as a sensor of the driving vehicle.

[0071] In operation S102, motion characteristics of the target vehicle are determined based on the driving environment data, wherein the motion characteristics include a first motion characteristic and a second motion characteristic, the first motion characteristic characterizing the relative distance and relative speed of the target vehicle in the direction of the road, and the second motion characteristic characterizing the inverse time of the collision between the target vehicle and the driving vehicle and the associated vehicle, the position change of the target vehicle perpendicular to the direction of the road at adjacent moments, and the angle between the driving direction of the target vehicle and the direction of the road.

[0072] According to an embodiment of the present disclosure, the relative distance and relative speed of the target vehicle in the direction of the road can be calculated from the motion state information of the driving vehicle, the target vehicle or the associated vehicle, thereby determining the first motion feature. Then, the inverse time of the collision between the target vehicle and the driving vehicle and the associated vehicle can be calculated from the relative distance and the relative speed. The second motion feature is determined from the inverse time, the change in position of the target vehicle perpendicular to the direction of the road at adjacent moments, and the angle between the driving direction of the target vehicle and the direction of the road.

[0073] In operation S103, the probability of the motion feature sequence is determined by a dual-channel Gaussian mixture hidden Markov model based on a motion feature sequence composed of motion features, wherein the motion feature sequence includes a first motion feature sequence and a second motion feature sequence, and the dual-channel Gaussian mixture hidden Markov model recursively obtains the probability of the motion feature sequence based on the probabilities of the motion features in the motion feature sequence.

[0074] According to an embodiment of the present disclosure, a first motion feature sequence may be composed of the first motion features, and a second motion feature sequence may be composed of the second motion features.

[0075] According to an embodiment of the present disclosure, the dual-channel Gaussian mixture hidden Markov model can be a trained model, which can infer the current behavior of the target vehicle based on the motion status of the target vehicle, the driving vehicle, and the associated vehicles; it can predict the target vehicle's lane merging behavior or lane keeping behavior, and the lane merging behavior can be regarded as a positive sample, and the lane keeping behavior can be regarded as a negative sample. The probability of the motion feature is calculated through the trained model, and then the probability of the motion feature is obtained through recursion.

[0076] In operation S104, the motion behavior of the target vehicle is predicted according to the probability of the motion feature sequence.

[0077] According to the embodiments of the present disclosure, the probability of the motion feature sequence can be calculated, and then the score of the target vehicle's lane merging behavior or lane keeping behavior can be obtained, and then the motion behavior of the target vehicle can be judged based on the score.

[0078] According to an embodiment of the present disclosure, by acquiring driving environment data including motion state information of a driving vehicle, a target vehicle, and associated vehicles within a preset time period, the motion features of the target vehicle including a first motion feature and a second motion feature are determined, wherein the first motion feature represents the relative distance and relative speed between the target vehicle and the driving vehicle and the associated vehicle in the forward direction of the road, and the second motion feature represents the inverse time of the collision between the target vehicle and the driving vehicle and the associated vehicle, and then the probability of the motion feature sequence is recursively obtained according to the probability of the motion features in the motion feature sequence through a dual-channel Gaussian mixture hidden Markov model, and the motion behavior of the target vehicle is predicted. By acquiring driving environment data including the target vehicle and the associated vehicle to determine the motion features through a driving vehicle, and then predicting through a dual-channel Gaussian mixture hidden Markov model, the motion state information of the surrounding vehicles can be fully utilized to calculate the prediction results through two different motion features, at least partially overcoming the technical problem that the motion state information of the surrounding vehicles cannot be fully utilized when predicting the merging behavior of the vehicle, thereby reducing the accuracy of the prediction, and improving the accuracy of the vehicle behavior prediction.

[0079] According to an embodiment of the present disclosure, obtaining driving environment data of a vehicle within a preset time period includes:

[0080] The driving environment data of the vehicle driven within a preset time period is read from the driving environment data stored in the queue memory of a fixed length.

[0081] According to an embodiment of the present disclosure, a queue memory Γ with a fixed length of τ may be used. τ (t) stores all motion states from time t-τ+1 to time t. When t>τ, the motion state earlier than time t-τ+1 is discarded; when t≤τ, the queue memory will store all historical motion states without discarding data. Queue memory Γ τ The implementation process of (t) can be expressed as:

[0082] Γ τ (t)={ξ(t-k+1),...,ξ(t-1),ξ(t)},k=min{τ,t} (6)

[0083] Figure 2 A schematic diagram of a driving environment of a vehicle obtained according to an embodiment of the present disclosure is schematically shown.

[0084] like Figure 2 As shown, the coordinate system can be defined as follows: the positive direction of the x-axis of the coordinate system is along the direction of the road, and the positive direction of the y-axis is perpendicular to the direction of the road and points to the left side of the road. from and Lane L to The lane line is the origin of the y-axis, that is, y = 0. At time t, the driving vehicle Ve Driving in lane L to Inside, the speed is v e (t), the coordinates are (x e (t),y e (t)); target vehicle V o Driving in the left adjacent lane L fr om, the speed is v o (t), the coordinates are (x o (t),y o (t)); If the target vehicle performs a lane-changing action, it will move from lane L from Merge into lane L to , T m is the time of merging; there are 3 related vehicles around, namely, the vehicle V in front of the driving vehicle fe , the speed is v fe (t), the coordinates are (x fe (t),y fe (t)), the vehicle V traveling in front of the target vehicle fo , the speed is v fo (t), the coordinates are (x fo (t),y fo (t)), the vehicle V behind the target vehicle ro , the speed is v ro (t), the coordinates are (x ro (t),y ro (t)).

[0085] The motion state ξ(t) contains the position and speed information of the driving vehicle, the target vehicle, and the associated vehicles:

[0086] ξ(t)=[x e (t),y e (t),v e (t),x o (t),y o (t),v o (t),x fe (t),y fe (t),v fe (t),x fo (t),y fo (t),v fo (t),x ro (t),y ro (t),v ro (t)] T (7)

[0087] In order to suppress the influence of measurement noise on the motion state, a sliding average filter is used to filter all the motion states in the queue memory. The working principle of the sliding average filter can be expressed as:

[0088]

[0089]

[0090]

[0091] Among them, 2w+1 is the window width of the sliding average filter, and all variables with “^” superscripts represent variables processed by the sliding average filter.

[0092] According to an embodiment of the present disclosure, the motion state information includes position information and speed information, and the motion characteristics of the target vehicle are determined according to the driving environment data, including:

[0093] Determine, based on the position information and speed information of the driving vehicle, the target vehicle, and the associated vehicle, a first relative distance and a first relative speed between the target vehicle and the driving vehicle in the direction of the road, and a second relative distance and a second relative speed between the target vehicle and the associated vehicle in the direction of the road;

[0094] A first motion feature and a second motion feature are determined according to the first relative distance, the second relative distance, the first relative speed, and the second relative speed.

[0095] According to an embodiment of the present disclosure, the first relative distance and the first relative speed between the target vehicle and the driving vehicle in the direction of the road are calculated using the following formulas:

[0096]

[0097] According to an embodiment of the present disclosure, when the associated vehicle is a vehicle traveling in front of the driving vehicle, the second relative distance and the second relative speed between the target vehicle and the associated vehicle in the direction of the road are calculated using the following formulas:

[0098]

[0099] According to an embodiment of the present disclosure, when the associated vehicle is a vehicle traveling in front of the target vehicle, the second relative distance and the second relative speed between the target vehicle and the associated vehicle in the direction of the road are calculated using the following formulas:

[0100]

[0101] According to an embodiment of the present disclosure, when the associated vehicle is a vehicle traveling behind the target vehicle, the second relative distance and the second relative speed between the target vehicle and the associated vehicle in the forward direction of the road are calculated using the following formulas:

[0102]

[0103] According to an embodiment of the present disclosure, the first motion feature and the second motion feature are determined according to the first relative distance, the second relative distance, the first relative speed, and the second relative speed, and are expressed by the following formula:

[0104]

[0105]

[0106] in, is the first motion feature, It is the second motion characteristic;

[0107] Δx o-e (t) are respectively the first relative distance, the first relative speed and the inverse of the collision time between the target vehicle and the driving vehicle;

[0108] Δx o-fe (t) are respectively the second relative distance, the second relative speed and the inverse of the collision time between the target vehicle and the associated vehicle in front of the driving vehicle in the forward direction of the road;

[0109] Δx o-fo (t) are respectively a second relative distance, a second relative speed and a reciprocal of the collision time between the target vehicle and the associated vehicle in front of the target vehicle in the forward direction of the road;

[0110] Δx o-ro (t) are respectively a second relative distance, a second relative speed and a reciprocal of the collision time between the target vehicle and the associated vehicle behind the target vehicle in the forward direction of the road;

[0111] Δy o (t) is the change in the position of the target vehicle perpendicular to the road direction at adjacent moments, ψ o (t) is the angle between the target vehicle’s direction of travel and the road’s direction of travel.

[0112] Δy o (t) and ψ o The specific calculation method of (t) is as follows:

[0113]

[0114]

[0115]

[0116]

[0117]

[0118]

[0119] According to an embodiment of the present disclosure, the probability of the motion feature sequence is determined by a dual-channel Gaussian mixture hidden Markov model according to the motion feature sequence composed of motion features, including:

[0120] Determine the probability of the motion feature according to the hidden state corresponding to the motion feature;

[0121] According to the probability of the motion features, the probability of the motion feature sequence is determined.

[0122] According to an embodiment of the present disclosure, given a motion feature sequence Ω with a length of τ τ (t):

[0123]

[0124] in, is the first motion feature in formula (1) or the second motion feature in formula (2).

[0125] The corresponding hidden state sequence Ξ τ (t) is:

[0126] Ξ τ (t)={H(t-τ+1),...,H(t-1),H(t)} (22)

[0127] Among them, the hidden state H(t) belongs to the hidden state set N H is the total number of hidden states.

[0128] Figure 3 A schematic diagram of a single Gaussian mixture hidden Markov model structure according to an embodiment of the present disclosure is schematically shown.

[0129] like Figure 3 As shown, according to an embodiment of the present disclosure, the Gaussian mixture hidden Markov model M can be expressed by formulas (23)-(26):

[0130] M=M(α0,A,B) (23)

[0131]

[0132]

[0133]

[0134] Among them, α0 is the length of N H The column vector is called the initial distribution probability of the hidden state; each element α 0i =p(H(t-τ+1)=H i ) indicates that the hidden state H(t-τ+1) at the initial moment is H i probability;

[0135] A is N H ×N H The matrix is ​​called the transfer matrix; each element a ij =p(H(t)=H j |H(t-1)=H i ) indicates that the hidden state H(t-1) at the previous moment is H i , the next hidden state H(t) is H j probability;

[0136] B is N H ×1 matrix is ​​called the emission matrix; each element It means that when the hidden state H(t) is H i When , the output motion state is The probability of i Depends on the hidden state H i The corresponding Gaussian mixture model Different hidden states correspond to different Gaussian mixture models.

[0137] Formulas (27)-(30) give the equations with N G Gaussian mixture model with centers The general representation of:

[0138]

[0139]

[0140]

[0141]

[0142] Where c is the length of N G The column vector is called the weight vector of the Gaussian mixture model; each element c i Represents the weight of the multidimensional Gaussian distribution corresponding to the i-th center;

[0143] μ i is of length NF The column vector of is called the mean vector of the multidimensional Gaussian distribution corresponding to the i-th center;

[0144] Σ i YesN F ×N F The matrix is ​​called the covariance matrix of the multidimensional Gaussian distribution corresponding to the i-th center;

[0145] N F It is a state of motion The dimension of When N is the first motion feature in formula (1), F =8; when When N is the second motion feature in formula (2), F =6.

[0146] According to the description of Gaussian mixture model, in formula (31), the emission matrix element b i Calculated by the following formula:

[0147]

[0148] in, Represents a Gaussian mixture model The probability density function of the multidimensional Gaussian distribution corresponding to the kth center, the mean of the multidimensional Gaussian distribution is μ k , the covariance is Σ k .

[0149] Gaussian mixture hidden Markov model M predicts motion feature sequence Ω τ The probability of (t) can be calculated using the following formulas (32)-(34):

[0150]

[0151]

[0152]

[0153] Among them, p(Ω τ (t)|M) is the predicted probability to be sought;

[0154] is the motion feature sequence Ω τ (t), motion characteristics is the hidden state H j The output probability is recursively solved in formula (33);

[0155] is the Gaussian mixture model calculated according to formula (31) Output probability;

[0156] a ij is the element in the Gaussian mixture hidden Markov model M transfer matrix;

[0157] α 0j is an element in the initial distribution probability of the hidden state of the Gaussian mixture hidden Markov model M.

[0158] According to an embodiment of the present disclosure, the motion behavior of the target vehicle is predicted based on the probability of the motion feature sequence, which is expressed as follows:

[0159]

[0160]

[0161]

[0162] in, is the first motion feature sequence, is the second motion feature sequence,

[0163] and In the first motion feature sequence In the channel, the Gaussian mixture hidden Markov model predicting the target vehicle's merging behavior and the Gaussian mixture hidden Markov model predicting the target vehicle's lane keeping behavior are used.

[0164] and In the second motion feature sequence In the channel, a Gaussian mixture hidden Markov model for predicting the target vehicle's merging behavior and a Gaussian mixture hidden Markov model for predicting the target vehicle's lane keeping behavior;

[0165] and Use and Predict the first motion feature sequence The probability of and Use and Predicting the second motion feature sequence The probability of , lg() is the logarithmic function with base 10;

[0166] and The first motion feature sequence The prediction score and the second motion feature sequence The prediction score, l lon, l mix and l0 are linear comparator parameters;

[0167] B P It is the behavior prediction result. LC indicates that the target vehicle performs lane merging behavior, and LK indicates that the target vehicle performs lane keeping behavior.

[0168] Figure 4 The flowchart schematically shows a vehicle behavior prediction method based on a dual-channel Gaussian mixture hidden Markov model according to an embodiment of the present disclosure.

[0169] like Figure 4 As shown, the first motion feature sequence The inputs are respectively into the Gaussian mixture hidden Markov model for predicting the target vehicle's merging behavior and the Gaussian mixture hidden Markov model for predicting the target vehicle's lane keeping behavior. and Get the first motion feature sequence The predicted score The second motion feature sequence The inputs are respectively into the Gaussian mixture hidden Markov model for predicting the target vehicle's merging behavior and the Gaussian mixture hidden Markov model for predicting the target vehicle's lane keeping behavior. and Get the second motion feature sequence The predicted score The linear comparator then predicts the behavior of the target vehicle based on the score.

[0170] According to an embodiment of the present disclosure, the above method further includes:

[0171] According to the evaluation index, the evaluation index value of the dual-channel Gaussian mixture hidden Markov model is determined, wherein the evaluation index includes the Recall of Forewarning (ROF), the Average Recall of Forewarning (ARoF) and the F1-Average Recall of Forewarning (F1-ARoF), the Recall of Forewarning represents the probability that the predicted result of the target vehicle behavior is consistent with the actual behavior of the target vehicle;

[0172] Evaluate Gaussian mixture hidden Markov model according to the evaluation index value.

[0173] According to an embodiment of the present disclosure, it is assumed that b is the merging time T m At any previous time, t b The warning recall rate RoF(t b ) is defined as:

[0174]

[0175] Among them, t r is the moment when the target vehicle’s lane-merging behavior is correctly predicted for the first time;

[0176] The duration of the trajectory is greater than |t b |The number of target vehicles;

[0177] TP(t r <t b ) indicates that at t b The total number of true positive samples (correctly predicted merging behavior) before time.

[0178] According to an embodiment of the present disclosure, the average warning recall rate is calculated as follows:

[0179]

[0180] Among them, T bmax is the maximum duration of the target vehicle’s trajectory.

[0181] Since the definite integral in formula (36) cannot be calculated directly in practical applications, the following approximate integral is used to calculate ARoF:

[0182]

[0183] Where L represents the number of sampling moments for approximate integration;

[0184] Represents the "asymptotically equal" symbol, and as L → ∞, the approximate integral converges asymptotically to the true value.

[0185] According to an embodiment of the present disclosure, F1-ARoF may be calculated by the following formula:

[0186]

[0187]

[0188] Among them, Precision is the accuracy;

[0189] TP represents the total number of true positive samples (correctly predicted merging behavior);

[0190] FP represents the total number of false positive samples (failed to correctly predict lane keeping behavior).

[0191] According to an embodiment of the present disclosure, the dual-channel Gaussian mixture hidden Markov model is evaluated according to the evaluation index value, including:

[0192] The predictive performance of the dual-channel Gaussian mixture hidden Markov model is evaluated based on the warning recall rate and the average warning recall rate. The predictive performance represents the time interval between the moment when the target vehicle performs lane merging or lane keeping behavior and the moment when the target vehicle behavior is correctly predicted.

[0193] According to the embodiments of the present disclosure, the RoF values ​​at different times are plotted in a coordinate system to obtain a RoF curve. The RoF curve intuitively describes the change of the probability of correctly predicting the merging behavior over time; the trend of the RoF curve can qualitatively describe the predictive performance of the behavior prediction method.

[0194] For example, in a specific embodiment, there are three different behavior prediction methods. The three prediction methods are all based on the Gaussian mixture hidden Markov model described by formulas (23)-(26), but consider different physical quantities described in formulas (11)-(20). The specific physical quantities considered by each method are shown in Table 1 below:

[0195] Table 1 Physical quantities considered by the three prediction methods

[0196] Prediction Methods Physical quantities considered 1 Relative longitudinal distance and relative longitudinal velocity, formula (11)-(14) 2 Lateral change and heading angle, formula (19)-(20) 3 The inverse of the collision time, the lateral change and the heading angle, formulas (15)-(20)

[0197] Figure 5 A schematic diagram of a warning recall rate curve of three different behavior prediction methods according to an embodiment of the present disclosure is schematically shown.

[0198] like Figure 5 As shown, the merging time can be set as the zero reference time, that is, T m = 0, then any time t before the merging time b <0. The RoF curves of the three prediction methods are as follows Figure 5 As shown, it can be seen that: at the merging time T m At the time of merging, the predictability of method 2 is better than that of method 1 and method 3. At the time before the merging moment, method 1 has the best predictability. Method 3 has a more stable predictability in the whole period. At the time before the merging moment, the performance of method 3 is close to that of method 1 and better than that of method 2. Near the merging moment, the performance of method 3 is better than that of method 1 and method 2.

[0199] According to the embodiments of the present disclosure, AROF can be used as a quantitative indicator for evaluating the predictive performance of the behavior prediction method: because ARoF is obtained by normalizing the RoF curve after integration, it can be regarded as the normalized value of the area under the RoF curve, which is essentially the overall expression of the prediction results within the entire trajectory duration period, so it is also used to evaluate the predictive performance; and because the value of ARoF is a real number in the interval [0, 1], it can be used as a quantitative indicator for evaluating the predictive performance.

[0200] For example, in a specific embodiment, for Figure 5 The behavior prediction method 3 shown in Figure 3 takes T bmax =7.0s, L=14, ARoF can be regarded as the result of normalization of the gray area under the RoF curve in the figure.

[0201] According to the embodiments of the present disclosure, F1-ARoF can be used as a quantitative indicator for evaluating the comprehensive performance of the behavior prediction method, because Precision is a quantitative indicator that describes the accuracy performance of the prediction result, and the accuracy performance refers to the performance that the merging behavior or lane keeping behavior implemented by the target vehicle can be correctly predicted. ARoF is a quantitative indicator that describes the predictability performance of the prediction result. F1-ARoF is the harmonic mean of the two, so it can be used as a quantitative indicator for evaluating the comprehensive performance of the behavior prediction method.

[0202] For example, in a specific embodiment, for Figure 5 The three different behavior prediction methods shown in the figure are used to calculate the three indicators of accuracy, ARoF and F1-ARoF and compare them respectively; among them, accuracy is used to evaluate the accuracy performance, ARoF is used to evaluate the predictive performance, and F1-ARoF is used to evaluate the comprehensive performance; the higher the indicator value, the better the performance. The performance comparison of the three different methods is shown in Table 2. It can be seen from the table that method 2 has the best accuracy performance; method 1 has the best predictive performance; method 3 lags behind method 2 and method 1 in the two separate indicators of accuracy performance and predictive performance, but the gap is not large; method 3 has the best comprehensive performance.

[0203] Table 2 Comprehensive performance of three prediction methods

[0204] Prediction Methods Accuracy ARoF <![CDATA[F1-ARoF]]> 1 0.757 0.511 0.610 2 0.932 0.283 0.434 3 0.912 0.476 0.625

[0205] According to the embodiments of the present disclosure, the vehicle behavior prediction method based on the dual-channel Gaussian mixture hidden Markov model proposed in the embodiments of the present disclosure can also be verified by evaluation indicators such as sensitivity, specificity, F1-score, ROC, AUC, etc. The following is a brief description of sensitivity, specificity, F1-score, ROC and AUC.

[0206] Sensitivity is defined as the percentage of correctly classified positive samples among all test samples. Sensitivity is also called the true positive rate (TPR), which is used to evaluate the performance of the behavior prediction method in correctly predicting the merging behavior. The higher the sensitivity, the better the prediction performance of the behavior prediction method. TP is the total number of true positive samples (correctly predicted merging behavior), and FN is the total number of false negative samples (failed to correctly predict merging behavior). The sensitivity calculation method is:

[0207]

[0208] Specificity is defined as the percentage of negative samples that are correctly classified among all test samples. Specificity is also called the true negative rate (TNR), which is used to evaluate the performance of the behavior prediction method in correctly predicting the lane keeping behavior; the higher the specificity, the better the lane keeping performance prediction performance of the behavior prediction method. TN is the total number of true negative (correctly predicted lane keeping behavior) samples, FP is the total number of false positive (failed to correctly predict lane keeping behavior) samples, and the specificity calculation method is:

[0209]

[0210] The F1-score is defined as the harmonic mean of recall and precision. The F1-score is used to evaluate the correct prediction performance of the behavior prediction method. The higher the F1-score, the better the behavior prediction method has in predicting all behaviors. The calculation method of the F1-score is:

[0211]

[0212] The receiver operating characteristic (ROC) curve is defined as the relationship between the true positive rate TPR and the false positive rate FPR = 1-TNR when different thresholds are selected. The closer the ROC curve is to the upper left corner of the coordinate axis (0, 1), the less the behavior prediction method is affected by the threshold selection, and the better the prediction performance.

[0213] The area under the ROC curve (AUC) is defined as the area under the ROC. The closer the ROC curve is to the upper left corner (0, 1) of the coordinate axis, the closer the AUC value is to 1.0, indicating that the prediction performance of the behavior prediction method is better.

[0214] The validation data uses data from the public dataset NGSIM on roads marked I-80 and US-101. The validation data contains 2,341 lane-merging behavior samples and 3,570 lane-keeping behavior samples. 60% of the samples are used as training sets for behavior prediction method training, and 40% of the samples are used as test sets for behavior prediction method testing.

[0215] First, the performance of the vehicle behavior prediction method based on the dual-channel Gaussian mixture hidden Markov model provided by the embodiment of the present disclosure is verified. In order to obtain more general results, tests are performed under conditions of queue memory lengths of different lengths; the queue memory length τ takes values ​​of 10, 30, 50, and 70. The performance test results are shown in Table 3:

[0216] Table 3 Evaluation index values ​​under queue memories of different lengths

[0217]

[0218] Figure 6 The schematic diagram of the ROC curve when the queue memory length τ takes different values ​​according to an embodiment of the present disclosure is schematically shown.

[0219] As shown in the figure, when the queue memory length becomes longer, the accuracy performance of the prediction method provided by the embodiment of the present disclosure decreases, and the predictability performance increases; when the queue memory length τ=30, the prediction method has the best overall performance.

[0220] Secondly, the queue memories of different lengths provided by the embodiments of the present disclosure are compared with the homologous methods based on Gaussian mixture hidden Markov models. The four methods compared are:

[0221] Comparison method 1: The behavior prediction method based on discrete-continuous hidden Markov model proposed in the reference "Highway discretionary lane changing behavior recognition using continuous and discrete hidden Markov model (ITSC 2021)".

[0222] Comparison method 2: The behavior prediction method based on Gaussian mixture hidden Markov model proposed in the reference “Gauss mixture hidden Markov model to characterise and model discretionary lane-change behaviours for autonomous vehicles (IET-ITS, 2020)”.

[0223] Comparative method 3: A behavior prediction method is designed by combining comparative method 2 with the first motion feature sequence described by formula (1).

[0224] Comparative method 4: A behavior prediction method is designed by combining comparative method 2 with the second motion feature sequence described by formula (2).

[0225] When performing comparative experiments, the queue memory length τ is uniformly taken as 30. The performance test results are shown in Table 4: The behavior prediction method provided by the embodiment of the present disclosure obtains the best results in both accuracy performance and comprehensive performance.

[0226] Table 4 Evaluation index values ​​of the method of the present disclosure and four comparative methods

[0227] method Sensitivity Specificity <![CDATA[F1 - fraction]]> AUC ARoF <![CDATA[F1-ARoF]]> The disclosed method 0.919 0.945 0.917 0.984 0.512 0.657 Comparison method 1 0.533 0.727 0.546 0.685 0.330 0.416 Comparison Method 2 0.721 0.805 0.717 0.874 0.516 0.597 Comparison method 3 0.711 0.826 0.720 0.848 0.463 0.566 Comparison method 4 0.908 0.937 0.906 0.973 0.477 0.625

[0228] Figure 7 The ROC curve diagrams of the prediction method according to the embodiment of the present disclosure and four comparative methods are schematically shown.

[0229] like Figure 7 As shown, the predictive performance of the prediction method of the embodiment of the present disclosure is higher than that of the four comparison methods.

[0230] Figure 8 The block diagram of a vehicle behavior prediction device 800 based on a dual-channel Gaussian mixture hidden Markov model according to an embodiment of the present disclosure is schematically shown.

[0231] like Figure 8 As shown, the vehicle behavior prediction device 800 based on the dual-channel Gaussian mixture hidden Markov model includes an acquisition module 810 , a first determination module 820 , a second determination module 830 , and a prediction module 840 .

[0232] The acquisition module 810 acquires the driving environment data of the driving vehicle within a preset time period, wherein the driving environment data includes the motion state information of the driving vehicle, the target vehicle and the associated vehicles, the target vehicle is located in the adjacent lane in front of the driving vehicle in the forward direction of the road, and the associated vehicles represent the vehicles that affect the lane change of the target vehicle;

[0233] The first determination module 820 is used to determine the motion characteristics of the target vehicle according to the driving environment data, wherein the motion characteristics include a first motion characteristic and a second motion characteristic, the first motion characteristic represents the relative distance and relative speed between the target vehicle and the driving vehicle and the associated vehicle in the road forward direction, and the second motion characteristic represents the inverse time of the collision between the target vehicle and the driving vehicle and the associated vehicle, the position change of the target vehicle perpendicular to the road forward direction at adjacent moments, and the angle between the driving direction of the target vehicle and the road forward direction;

[0234] A second determination module 830 is used to determine the probability of the motion feature sequence through a dual-channel Gaussian mixture hidden Markov model according to the motion feature sequence composed of motion features, wherein the motion feature sequence includes a first motion feature sequence and a second motion feature sequence, and the Gaussian mixture hidden Markov model recursively obtains the probability of the motion feature sequence according to the probability of the motion features in the motion feature sequence;

[0235] The prediction module 840 is used to predict the movement behavior of the target vehicle according to the probability of the movement feature sequence.

[0236] According to an embodiment of the present disclosure, by acquiring driving environment data including motion state information of a driving vehicle, a target vehicle, and associated vehicles within a preset time period, the motion features of the target vehicle including a first motion feature and a second motion feature are determined, wherein the first motion feature represents the relative distance and relative speed between the target vehicle and the driving vehicle and the associated vehicle in the forward direction of the road, and the second motion feature represents the inverse time of the collision between the target vehicle and the driving vehicle and the associated vehicle, and then the probability of the motion feature sequence is recursively obtained according to the probability of the motion features in the motion feature sequence through a dual-channel Gaussian mixture hidden Markov model, and the motion behavior of the target vehicle is predicted. By acquiring driving environment data including the target vehicle and the associated vehicle to determine the motion features through a driving vehicle, and then predicting through a dual-channel Gaussian mixture hidden Markov model, the motion state information of the surrounding vehicles can be fully utilized to calculate the prediction results through two different motion features, at least partially overcoming the technical problem that the motion state information of the surrounding vehicles cannot be fully utilized when predicting the merging behavior of the vehicle, thereby reducing the accuracy of the prediction, and improving the accuracy of the vehicle behavior prediction.

[0237] According to an embodiment of the present disclosure, the acquisition module 810 includes:

[0238] The acquisition unit is used to read the driving environment data of the vehicle within a preset time period from the driving environment data stored in the queue memory of a fixed length.

[0239] According to an embodiment of the present disclosure, the motion state information includes position information and speed information, and the first determination module 820 includes:

[0240] A first determination unit is used to determine a first relative distance and a first relative speed between the target vehicle and the driving vehicle in the direction of the road, and a second relative distance and a second relative speed between the target vehicle and the associated vehicle in the direction of the road according to the position information and speed information of the driving vehicle, the target vehicle and the associated vehicle;

[0241] The second determining unit is used to determine the first motion feature and the second motion feature according to the first relative distance, the second relative distance, the first relative speed and the second relative speed.

[0242] According to an embodiment of the present disclosure, the second determining module 830 includes:

[0243] A third determining unit, configured to determine a probability of the motion feature according to a hidden state corresponding to the motion feature;

[0244] The fourth determining unit is used to determine the probability of the motion feature sequence according to the probability of the motion feature.

[0245] According to an embodiment of the present disclosure, the above-mentioned device 800 further includes:

[0246] The third determination module is used to determine the evaluation index value of the dual-channel Gaussian mixture hidden Markov model according to the evaluation index, wherein the evaluation index includes the warning recall rate, the average warning recall rate and the F1-average warning recall rate, and the warning recall rate represents the probability that the prediction result of the target vehicle behavior is consistent with the actual behavior of the target vehicle;

[0247] The evaluation module is used to evaluate the Gaussian mixture hidden Markov model according to the evaluation index value.

[0248] According to an embodiment of the present disclosure, the evaluation module includes:

[0249] The evaluation unit is used to evaluate the predictive performance of the dual-channel Gaussian mixture hidden Markov model according to the warning recall rate and the average warning recall rate, wherein the predictive performance represents the time interval between the moment when the target vehicle performs a lane merging behavior or a lane keeping behavior and the moment when the target vehicle behavior is correctly predicted.

[0250] According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits, or at least part of the functions of any one of them can be implemented in one module. According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits can be split into multiple modules for implementation. According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits can be at least partially implemented as hardware circuits, such as field programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems on chips, systems on substrates, systems on packages, application specific integrated circuits (ASICs), or can be implemented by hardware or firmware in any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in any appropriate combination of any of them. Alternatively, according to the embodiments of the present invention, one or more of the modules, submodules, units, and subunits can be at least partially implemented as computer program modules, and when the computer program modules are run, the corresponding functions can be performed.

[0251] For example, any multiple of the acquisition module 810, the first determination module 820, the second determination module 830, and the prediction module 840 can be combined in one module / unit / subunit for implementation, or any one of the modules / units / subunits can be split into multiple modules / units / subunits. Alternatively, at least part of the functions of one or more modules / units / subunits in these modules / units / subunits can be combined with at least part of the functions of other modules / units / subunits and implemented in one module / unit / subunit. According to an embodiment of the present disclosure, at least one of the acquisition module 810, the first determination module 820, the second determination module 830, and the prediction module 840 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or by any one of the three implementation methods of software, hardware, and firmware or by a suitable combination of any of them. Alternatively, at least one of the acquisition module 810, the first determination module 820, the second determination module 830, and the prediction module 840 may be at least partially implemented as a computer program module, and when the computer program module is executed, a corresponding function may be performed.

[0252] It should be noted that the vehicle behavior prediction device part based on the dual-channel Gaussian mixture hidden Markov model in the embodiments of the present disclosure corresponds to the vehicle behavior prediction method part based on the dual-channel Gaussian mixture hidden Markov model in the embodiments of the present disclosure. The description of the vehicle behavior prediction device part based on the dual-channel Gaussian mixture hidden Markov model specifically refers to the vehicle behavior prediction method part based on the dual-channel Gaussian mixture hidden Markov model, which will not be repeated here.

[0253] Fig. 9 A block diagram of an electronic device suitable for implementing the method described above according to an embodiment of the present disclosure is schematically shown. Fig. 9 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0254] like Fig. 9As shown, the electronic device 900 according to an embodiment of the present disclosure includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage part 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include an onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0255] In RAM 903, various programs and data required for the operation of electronic device 900 are stored. Processor 901, ROM 902 and RAM 903 are connected to each other via bus 904. Processor 901 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 902 and / or RAM 903. It should be noted that the program can also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in the one or more memories.

[0256] According to an embodiment of the present disclosure, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to the bus 904. The system 900 may further include one or more of the following components connected to the I / O interface 905: an input section 906 including a keyboard, a mouse, etc.; an output section 907 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, a modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the I / O interface 905 as needed. A removable medium 911, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 910 as needed, so that a computer program read therefrom is installed into the storage section 908 as needed.

[0257] According to an embodiment of the present disclosure, the method flow according to an embodiment of the present disclosure can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program contains a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, the above-mentioned functions defined in the system of the embodiment of the present disclosure are executed. According to an embodiment of the present disclosure, the system, equipment, device, module, unit, etc. described above can be implemented by a computer program module.

[0258] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the apparatus, method and electronic device according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram may represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box may also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions. It can be understood by those skilled in the art that the features recorded in the various embodiments and / or claims of the present disclosure can be combined and / or combined in a variety of ways, even if such a combination or combination is not explicitly recorded in the present disclosure. In particular, without departing from the spirit and teaching of the present disclosure, the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0259] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above separately, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. The scope of the present disclosure is defined by the attached claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A vehicle behavior prediction method based on a dual-channel Gaussian mixture hidden Markov model, comprising: Acquiring driving environment data of the driving vehicle within a preset time period, wherein the driving environment data includes motion state information of the driving vehicle, a target vehicle, and associated vehicles, the target vehicle is located in an adjacent lane ahead of the driving vehicle in the road forward direction, and the associated vehicles represent vehicles that affect the lane change of the target vehicle; Determine the motion characteristics of the target vehicle according to the driving environment data, wherein the motion characteristics include a first motion characteristic and a second motion characteristic, the first motion characteristic characterizing the relative distance and relative speed of the target vehicle in the road advancing direction and the driving vehicle and the associated vehicle, and the second motion characteristic characterizing the inverse of the time when the target vehicle collides with the driving vehicle and the associated vehicle, the position change of the target vehicle perpendicular to the road advancing direction at adjacent moments, and the angle between the driving direction of the target vehicle and the road advancing direction; According to the motion feature sequence composed of the motion features, the probability of the motion feature sequence is determined by a dual-channel Gaussian mixture hidden Markov model, wherein the motion feature sequence includes a first motion feature sequence and a second motion feature sequence, and the dual-channel Gaussian mixture hidden Markov model recursively obtains the probability of the motion feature sequence according to the probabilities of the motion features in the motion feature sequence; Predicting the motion behavior of the target vehicle according to the probability of the motion feature sequence; The prediction of the motion behavior of the target vehicle based on the probability of the motion feature sequence is expressed as follows: (3) (4) (5) in, is the first motion feature sequence, is the second motion feature sequence, and In the first motion feature sequence a Gaussian mixture hidden Markov model predicting the target vehicle to have a merging behavior and a Gaussian mixture hidden Markov model predicting the target vehicle to have a lane keeping behavior in a channel; and In the second motion feature sequence a Gaussian mixture hidden Markov model predicting the target vehicle to have a merging behavior and a Gaussian mixture hidden Markov model predicting the target vehicle to have a lane keeping behavior in a channel; and Use and Predicting the first motion feature sequence The probability of and Use and Predict the second motion feature sequence The probability of It is a logarithmic function with base 10; and The first motion feature sequences are The prediction score and the second motion feature sequence The prediction score of , and is the linear comparator parameter; is the behavior prediction result, LC indicates that the target vehicle performs a lane merging behavior, and LK indicates that the target vehicle performs a lane keeping behavior.

2. The method according to claim 1, wherein: The step of obtaining the driving environment data of the vehicle within a preset time period includes: The driving environment data of the vehicle during the preset time period is read from the driving environment data stored in the queue memory of the fixed length.

3. The method according to claim 1, wherein: The motion state information includes position information and speed information, and determining the motion characteristics of the target vehicle according to the driving environment data includes: Determine, according to the position information and speed information of the driving vehicle, the target vehicle and the associated vehicle, a first relative distance and a first relative speed between the target vehicle and the driving vehicle in the direction of the road, and a second relative distance and a second relative speed between the target vehicle and the associated vehicle in the direction of the road; The first motion feature and the second motion feature are determined according to the first relative distance, the second relative distance, the first relative speed, and the second relative speed.

4. The method according to claim 3, wherein: The first motion feature and the second motion feature are determined according to the first relative distance, the second relative distance, the first relative speed, and the second relative speed, and are expressed by the following formula: (1) (2) in, is the first motion feature, It is the second motion characteristic; , , are respectively the first relative distance between the target vehicle and the driving vehicle, the first relative speed and the inverse of the collision time; , , are respectively a second relative distance, a second relative speed and a reciprocal of the collision time between the target vehicle and the associated vehicle in front of the driving vehicle in the direction of the road; , , are respectively a second relative distance, a second relative speed and a reciprocal of the collision time between the target vehicle and an associated vehicle ahead of the target vehicle in the direction of the road; , , are respectively a second relative distance, a second relative speed and a reciprocal of the collision time between the target vehicle and an associated vehicle behind the target vehicle in the forward direction of the road; is the change in position of the target vehicle perpendicular to the direction of the road at adjacent moments, It is the angle between the driving direction of the target vehicle and the forward direction of the road.

5. The method according to claim 3, wherein: The method of determining the probability of the motion feature sequence composed of the motion features by using a dual-channel Gaussian mixture hidden Markov model includes: Determining a probability of the motion feature according to a hidden state corresponding to the motion feature; The probability of the motion feature sequence is determined according to the probability of the motion feature.

6. The method according to claim 1, further comprising: Determine the evaluation index value of the dual-channel Gaussian mixture hidden Markov model according to the evaluation index, wherein the evaluation index includes a warning recall rate, an average warning recall rate and an F1-average warning recall rate, and the warning recall rate represents the probability that the prediction result of the target vehicle behavior is consistent with the actual behavior of the target vehicle; Evaluate the Gaussian mixture hidden Markov model according to the evaluation index value; Wherein, the warning recall rate is expressed as: in, express The warning recall rate at each moment, Merge time At any time before, is the moment when the target vehicle’s lane-merging behavior is correctly predicted for the first time. Is the trajectory duration greater than The target number of vehicles, Indicated in The total number of true positive samples before time; The average warning recall rate is expressed as: Among them, ARoF represents the average warning recall rate, is the maximum duration of the target vehicle’s trajectory; The F1-average warning recall rate is expressed as: Among them, F1-ARoF means F1-average warning recall rate, and Precision means precision.

7. The method according to claim 6, wherein: The step of evaluating the dual-channel Gaussian mixture hidden Markov model according to the evaluation index value includes: The predictive performance of the dual-channel Gaussian mixture hidden Markov model is evaluated based on the warning recall rate and the average warning recall rate, wherein the predictive performance characterizes the time interval between the moment when the target vehicle performs lane merging behavior or lane keeping behavior and the moment when the target vehicle behavior is correctly predicted.

8. A vehicle behavior prediction device based on a dual-channel Gaussian mixture hidden Markov model, comprising: An acquisition module is used to acquire driving environment data of the driving vehicle within a preset time period, wherein the driving environment data includes motion state information of the driving vehicle, a target vehicle, and associated vehicles, the target vehicle is located in an adjacent lane ahead of the driving vehicle in the direction of the road, and the associated vehicles represent vehicles that affect the lane change of the target vehicle; A first determination module is used to determine the motion characteristics of the target vehicle according to the driving environment data, wherein the motion characteristics include a first motion characteristic and a second motion characteristic, the first motion characteristic represents the relative distance and relative speed of the target vehicle in the road advancing direction and the driving vehicle and the associated vehicle, and the second motion characteristic represents the inverse time of the collision between the target vehicle and the driving vehicle and the associated vehicle, the position change of the target vehicle perpendicular to the road advancing direction at adjacent moments, and the angle between the driving direction of the target vehicle and the road advancing direction; A second determination module is used to determine the probability of the motion feature sequence through a dual-channel Gaussian mixture hidden Markov model according to the motion feature sequence composed of the motion features, wherein the motion feature sequence includes a first motion feature sequence and a second motion feature sequence, and the Gaussian mixture hidden Markov model recursively obtains the probability of the motion feature sequence according to the probabilities of the motion features in the motion feature sequence; A prediction module, used for predicting the motion behavior of the target vehicle according to the probability of the motion feature sequence; The motion behavior of the target vehicle is predicted based on the probability of the motion feature sequence, which is expressed by the following formula: (3) (4) (5) in, is the first motion feature sequence, is the second motion feature sequence, and In the first motion feature sequence a Gaussian mixture hidden Markov model predicting the target vehicle to have a merging behavior and a Gaussian mixture hidden Markov model predicting the target vehicle to have a lane keeping behavior in a channel; and In the second motion feature sequence a Gaussian mixture hidden Markov model predicting the target vehicle to have a merging behavior and a Gaussian mixture hidden Markov model predicting the target vehicle to have a lane keeping behavior in a channel; and Use and Predicting the first motion feature sequence The probability of and Use and Predict the second motion feature sequence The probability of It is a logarithmic function with base 10; and The first motion feature sequences are The prediction score and the second motion feature sequence The prediction score of , and is the linear comparator parameter; is the behavior prediction result, LC indicates that the target vehicle performs a lane merging behavior, and LK indicates that the target vehicle performs a lane keeping behavior.

9. An electronic device, comprising: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.