A method for predicting the trajectory of side-lane vehicles in car-following scenarios

Abstract

The present invention relates to the technical field of vehicle trajectory prediction, and specifically to a method for predicting the trajectory of a side lane vehicle in a following scenario, comprising: obtaining environmental historical state parameters and future state parameters of a main vehicle; the environmental historical state parameters include historical state parameters of the main vehicle, historical state parameters of a target vehicle, and historical state parameters of an interactive vehicle; and obtaining the future trajectory of the target vehicle according to the environmental historical state parameters and the future state parameters of the main vehicle. The future trajectory of the target vehicle is obtained by using the historical state parameters of the main vehicle, the historical state parameters of the target vehicle, the historical state parameters of the interactive vehicle, and the future state parameters of the main vehicle, which can reflect the interaction mechanism of multiple vehicles in the environment and improve the accuracy of the prediction of the trajectory of the target vehicle. At the same time, since the influence of the future motion state of the main vehicle on the future motion trajectory of the target vehicle is taken into account, a more reliable input reference condition is provided for the main vehicle decision, which effectively improves the safety of the vehicle's autonomous following and reduces the accident rate.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle trajectory prediction, and in particular to a method for predicting the trajectory of a side-lane vehicle in a car-following scenario. Background Art

[0002] During the long-term operation of the vehicle, there are many traffic participants around the vehicle. Especially during the process of autonomous following, the vehicle often faces the scene of vehicles in the side lane suddenly cutting in, which can easily cause the control decision of the following vehicle to be untimely and cause a collision. Therefore, accurate prediction of the future trajectory of the vehicle in the side lane can effectively improve the safety of autonomous following and reduce the accident rate.

[0003] Existing side-lane vehicle trajectory prediction methods include trajectory prediction methods based on physical models and trajectory prediction methods based on data models. Trajectory prediction methods based on physical models usually use state estimation algorithms to estimate the future motion state of the vehicle based on the vehicle's dynamic characteristics. This trajectory prediction method based on physical models is difficult to achieve accurate long-term prediction of the target vehicle's future trajectory. With the rapid development of artificial intelligence and deep learning, and the increasingly complete natural driving database, data-driven trajectory prediction methods have begun to be widely studied. Data-driven trajectory prediction methods obtain the future trajectory of the target vehicle based on the historical motion state of the target vehicle, and have high accuracy in predicting the long-term trajectory of the target vehicle.

[0004] However, the trajectory prediction methods based on data models in the prior art usually only consider the historical motion state of the target vehicle. In the actual process, the decision-making behavior of the driver of the target vehicle is not only continuously related to its own historical motion state, but also affected by the behavior of other vehicles in the surrounding environment. Therefore, predicting the future operation trajectory of the target vehicle based only on the historical motion state of the target vehicle is difficult to reflect the interaction mechanism of multiple vehicles in the environment, resulting in low accuracy in trajectory prediction of the target vehicle in a complex dynamic environment. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention proposes a method for predicting the trajectory of a side lane vehicle in a car-following scenario, which improves the accuracy of predicting the trajectory of a target vehicle.

[0006] The technical solution adopted by the present invention is a method for predicting the trajectory of a side-lane vehicle in a car-following scenario.

[0007] In a first possible implementation, a method for predicting the trajectory of a side lane vehicle in a car-following scenario includes: obtaining environmental historical state parameters and future state parameters of a main vehicle; the environmental historical state parameters include historical state parameters of the main vehicle, historical state parameters of a target vehicle, and historical state parameters of an interacting vehicle; and obtaining the future trajectory of the target vehicle based on the environmental historical state parameters and the future state parameters of the main vehicle.

[0008] In combination with the first feasible method, in the second feasible method, the historical state parameters of the environment and the future state parameters of the main vehicle are obtained, including: using the environmental perception sensor and differential global positioning system of the main vehicle to obtain the historical state parameters of the main vehicle; the state parameters include longitudinal coordinates, lateral coordinates, longitudinal speed and lateral speed; obtaining the future state parameters of the main vehicle according to the historical state parameters of the main vehicle; using the environmental perception sensor of the main vehicle to respectively obtain the historical state parameters of the target vehicle and the historical state parameters of the interactive vehicle.

[0009] In combination with the first feasible method, in a third feasible method, the future trajectory of the target vehicle is obtained according to the environmental historical state parameters and the future state parameters of the main vehicle, including: obtaining each first environmental hidden state according to each environmental historical state parameter; obtaining an environmental joint hidden state according to each first environmental hidden state; obtaining a second environmental hidden state according to the environmental joint hidden state; obtaining the main vehicle hidden state according to the future state parameters of the main vehicle; obtaining a final hidden state according to the second environmental hidden state and the main vehicle hidden state; and obtaining the future trajectory of the target vehicle according to the final hidden state.

[0010] In combination with the third feasible method, in a fourth feasible method, each first environment hidden state is obtained respectively according to each environment historical state parameter, including: each environment historical state parameter is input into each bidirectional long short-term memory neural network respectively to obtain each first environment hidden state.

[0011] In combination with the third feasible method, in a fifth feasible method, an environmental joint hidden state is obtained according to each first environmental hidden state, including: obtaining an attention coefficient of the target vehicle according to each first environmental hidden state; weighting each first environmental hidden state according to the attention coefficient of the target vehicle to obtain an environmental joint hidden state.

[0012] Combined with the fifth implementable method, in the sixth implementable method, the environment joint hidden state is obtained by the following formula:

[0013]

[0014] In the above formula, exp represents the exponential function with the natural constant e as the base, || represents vector concatenation, and α m is the attention coefficient of the mth vehicle to the target vehicle, LeakyReLU is the activation function, and a T is the transposed matrix, W is the linear transformation matrix, H o is the first environment hidden state corresponding to the target vehicle, H m is the first hidden state of the environment corresponding to the mth vehicle, N represents the nodes around the target vehicle, and the nodes around the target vehicle are composed of the main vehicle and the interactive vehicle, H krepresents the first environment hidden state corresponding to the kth node, N, m and k are all positive integers, m≤N;

[0015] H c =σ(∑ m∈N α m H m );

[0016] In the above formula, σ represents the nonlinear activation function, H c is the joint hidden state of the environment.

[0017] In combination with the third feasible method, in the seventh feasible method, a second environmental hidden state is obtained according to the environmental joint hidden state, including: using a fully connected network to splice the environmental joint hidden state and the first environmental hidden state corresponding to the target vehicle to obtain the second environmental hidden state.

[0018] In combination with the seventh implementable manner, in an eighth implementable manner, the second environment hidden state is obtained by the following formula:

[0019] H s =FC(concat(H o ,H c );W g );

[0020] In the above formula, H s is the hidden state of the second environment, FC(·) represents the fully connected network, W g is the fully connected network parameter, H o is the first environment hidden state corresponding to the target vehicle, H c is the joint hidden state of the environment.

[0021] In combination with the third feasible manner, in a ninth feasible manner, a final hidden state is obtained according to the hidden state of the second environment and the hidden state of the main vehicle, including: using a fully connected network to splice the hidden state of the second environment and the hidden state of the main vehicle to obtain the final hidden state.

[0022] In combination with the third feasible method, in the tenth feasible method, the future trajectory of the target vehicle is obtained according to the final hidden state, including: using a logic gate neural network to decode the final hidden state to obtain the future trajectory and motion state of the target vehicle.

[0023] It can be seen from the above technical scheme that the beneficial technical effects of the present invention are as follows: the future trajectory of the target vehicle is obtained through the historical state parameters of the main vehicle, the historical state parameters of the target vehicle, the historical state parameters of the interactive vehicle and the future state parameters of the main vehicle, which can reflect the interaction mechanism of multiple vehicles in the environment and improve the accuracy of trajectory prediction of the target vehicle in a complex dynamic environment. At the same time, since the influence of the future motion state of the main vehicle on the future motion trajectory of the target vehicle is taken into account, a more reliable input reference condition is provided for the main vehicle decision, which effectively improves the safety of the vehicle's autonomous following and reduces the accident rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the prior art description. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0025] Figure 1 A schematic diagram of a method for predicting the trajectory of a vehicle in a lane following scenario provided by an embodiment of the present invention;

[0026] Figure 2 A schematic diagram of an application environment and a coordinate system provided by an embodiment of the present invention;

[0027] Figure 3 A schematic diagram of a decoding layer provided by an embodiment of the present invention;

[0028] Figure 4 A schematic diagram of another method for predicting the trajectory of a lane-side vehicle in a car-following scenario provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The following embodiments of the technical solution of the present invention are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore only used as examples, and cannot be used to limit the protection scope of the present invention.

[0030] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application should have the common meanings understood by those skilled in the art to which the present invention belongs.

[0031] Combination Figure 1 As shown, this embodiment provides a method for predicting the trajectory of a vehicle in a lane following scenario, including:

[0032] Step S01, obtaining environmental historical state parameters and host vehicle future state parameters; the environmental historical state parameters include host vehicle historical state parameters, target vehicle historical state parameters, and interactive vehicle historical state parameters;

[0033] Step S02: Obtain the future trajectory of the target vehicle according to the historical state parameters of the environment and the future state parameters of the host vehicle.

[0034] In some embodiments, the main vehicle is the vehicle in which the user rides, the target vehicle is the selected side-lane vehicle, and the interactive vehicle is other vehicles in the environment except the main vehicle and the target vehicle.

[0035] Combination Figure 2 As shown, in some embodiments, in a two-lane straight-following environment, the main vehicle is traveling on the right road, the target vehicle is the vehicle closest to the main vehicle on the left road, and the other three vehicles in the environment are interactive vehicles. At time t, the horizontal direction of the centerline of the lane where the main vehicle is located is the x-axis, the driving direction of the main vehicle is the direction of the x-axis, the projection position of the center of mass of the main vehicle on the x-axis is the origin o, the straight line perpendicular to the x-axis and passing through the origin o is the y-axis, and the direction from the lane where the main vehicle is located to the next lane is the direction of the y-axis, and an xoy rectangular coordinate system is established.

[0036] Optionally, obtaining environmental historical state parameters and host vehicle future state parameters includes: establishing an xoy rectangular coordinate system; obtaining host vehicle historical state parameters using the host vehicle's environmental perception sensor and a differential global positioning system; the state parameters include longitudinal coordinates, lateral coordinates, longitudinal speed, and lateral speed; obtaining host vehicle future state parameters based on the host vehicle's historical state parameters; and obtaining target vehicle historical state parameters and interactive vehicle historical state parameters respectively through the host vehicle's environmental perception sensor.

[0037] Optionally, the environment perception sensor includes a laser radar, a millimeter wave radar, and a camera. In some embodiments, a differential global positioning system is used to obtain longitudinal coordinates and lateral coordinates, and an environment perception sensor is used to obtain longitudinal speed and lateral speed.

[0038] In some embodiments, within 5 seconds before time t, the historical states of the host vehicle, the target vehicle, and the interactive vehicle are observed. The sampling frequency is 10 Hz, and the observation sequence length is 50. obs =50. That is, 5 seconds are divided into 50 moments, and the historical states of the main vehicle, the target vehicle, and the interactive vehicle at each moment are sampled to obtain the historical state parameters of the main vehicle, the target vehicle, and the interactive vehicle.

[0039] In some embodiments, the historical state parameters of the main vehicle include the longitudinal coordinate, transverse coordinate, longitudinal speed and transverse speed of the main vehicle at several historical moments. The historical state parameters of the main vehicle are expressed by the following formula:

[0040]

[0041]

[0042] In the above formula, Xe The historical status parameters of the main vehicle. is the state parameter of the main vehicle at the i-th moment, T obs is the length of the observation sequence, is the longitudinal coordinate of the main vehicle at the i-th moment, is the lateral coordinate of the main vehicle at the i-th moment, is the longitudinal velocity of the main vehicle at the i-th moment, is the lateral velocity of the main vehicle at the i-th moment.

[0043] In some embodiments, the target vehicle historical state parameters include the longitudinal coordinate, lateral coordinate, longitudinal speed and lateral speed of the target vehicle at several historical moments. The target vehicle historical state parameters are expressed by the following formula:

[0044]

[0045]

[0046] In the above formula, X o is the historical state parameter of the target vehicle, is the state parameter of the target vehicle at the i-th moment, T obs is the length of the observation sequence, is the longitudinal coordinate of the target vehicle at the i-th moment, is the lateral coordinate of the target vehicle at the i-th moment, is the longitudinal velocity of the target vehicle at the i-th moment, is the lateral velocity of the target vehicle at the i-th moment.

[0047] In some embodiments, the interactive vehicle historical state parameters include the longitudinal coordinates, lateral coordinates, longitudinal speed and lateral speed of the interactive vehicle at several historical moments. The interactive vehicle historical state parameters are expressed by the following formula:

[0048]

[0049]

[0050] In the above formula, X n is the historical state parameter of the interactive vehicle, is the state parameter of the interactive vehicle at the i-th moment, T obs is the length of the observation sequence, is the longitudinal coordinate of the interacting vehicle at the i-th moment, is the lateral coordinate of the interacting vehicle at the i-th moment, is the longitudinal velocity of the interacting vehicle at the i-th moment, is the lateral velocity of the interacting vehicle at the i-th moment.

[0051] In some embodiments, the prediction duration is 3 seconds, the prediction frequency is 10 Hz, and the prediction sequence length is 30. pre =30. That is, the next 3 seconds are divided into 30 moments, the historical state parameters of the main vehicle are calculated, and the future state parameters of the main vehicle at each future moment are obtained.

[0052] In some embodiments, the future state parameters of the host vehicle include the longitudinal coordinate, transverse coordinate, longitudinal speed and transverse speed of the host vehicle at several future moments. The future state parameters of the host vehicle are expressed by the following formula:

[0053]

[0054] In the above formula, is the future state parameter of the main vehicle, T pre To predict the sequence length, The main vehicle will be in the future pre State parameters at the moment.

[0055] Optionally, the future trajectory of the target vehicle is obtained according to the environmental historical state parameters and the future state parameters of the main vehicle, including: obtaining each first environmental hidden state according to each environmental historical state parameter; obtaining an environmental joint hidden state according to each first environmental hidden state; obtaining a second environmental hidden state according to the environmental joint hidden state; obtaining the main vehicle hidden state according to the future state parameters of the main vehicle; obtaining a final hidden state according to the second environmental hidden state and the main vehicle hidden state; and obtaining the future trajectory of the target vehicle according to the final hidden state.

[0056] Optionally, obtaining each first environment hidden state according to each environment historical state parameter includes: inputting each environment historical state parameter into each bidirectional long short-term memory neural network to obtain each first environment hidden state.

[0057] Optionally, the historical state parameters of the main vehicle, the historical state parameters of the target vehicle and the historical state parameters of the interactive vehicle are respectively input into the corresponding bidirectional long short-term memory neural network to obtain the first environmental hidden state corresponding to the main vehicle, the first environmental hidden state corresponding to the target vehicle and the first environmental hidden state corresponding to the interactive vehicle.

[0058] In some embodiments, the first environment hidden state corresponding to the main vehicle is expressed by the following formula:

[0059]

[0060] In the above formula, H e is the first environment hidden state corresponding to the main vehicle, T obs is the length of the observation sequence, i=1,2,...,T obs .

[0061]

[0062]

[0063]

[0064] In the above formula, concat(·) represents the concatenation operation, LSTM(·) is the long short-term memory neural network structure, and W h is the first network parameter, W f is the second network parameter, is the state vector output by the long short-term memory neural network corresponding to the main vehicle, is the state parameter of the main vehicle at the i-th moment, Main vehicle No. T obs - Status parameters at time i, is the first splicing parameter of the main vehicle at the i-th moment, is the second splicing parameter of the main vehicle at the i-th moment.

[0065] In some embodiments, the first environment hidden state corresponding to the target vehicle is expressed by the following formula:

[0066]

[0067] In the above formula, H o is the first environment hidden state corresponding to the target vehicle, T obs is the length of the observation sequence.

[0068]

[0069]

[0070]

[0071] In the above formula, concat(·) represents the concatenation operation, LSTM(·) is the long short-term memory neural network structure, and W h is the first network parameter, is the state vector output by the long short-term memory neural network corresponding to the target vehicle, is the state parameter of the target vehicle at the i-th moment, The target vehicle is T obs - Status parameters at time i, is the first splicing parameter of the target vehicle at the i-th moment, is the second splicing parameter of the target vehicle at the i-th moment.

[0072] In some embodiments, the first environment hidden state corresponding to the interactive vehicle is expressed by the following formula:

[0073]

[0074] In the above formula, H n is the first environment hidden state corresponding to the interactive vehicle, T obs is the length of the observation sequence.

[0075]

[0076]

[0077]

[0078] In the above formula, concat(·) represents the concatenation operation, LSTM(·) is the long short-term memory neural network structure, and W h is the first network parameter, is the state vector output by the long short-term memory neural network corresponding to the interactive vehicle, is the state parameter of the interactive vehicle at the i-th moment, For the interactive vehicle T obs - Status parameters at time i, is the first splicing parameter of the interactive vehicle at the i-th moment, is the second splicing parameter of the interacting vehicle at the i-th moment.

[0079] Optionally, the future state parameters of the main vehicle are input into a corresponding bidirectional long short-term memory neural network to obtain the hidden state of the main vehicle.

[0080] In some embodiments, the hidden state of the host vehicle is expressed by the following formula:

[0081]

[0082] In the above formula, H f The main vehicle is hidden, T pre is the predicted sequence length.

[0083]

[0084]

[0085]

[0086] In the above formula, concat(·) represents the concatenation operation, LSTM(·) is the long short-term memory neural network structure, and W f is the second network parameter, is the state vector output by the long short-term memory neural network of the main vehicle at the nth moment in the future, is the state parameter of the main vehicle at the jth moment, Main vehicle No. T pre- Status parameters at time j, is the first splicing parameter of the main vehicle at the future nth moment, The second splicing parameter of the main vehicle at the nth moment in the future.

[0087] Optionally, obtaining an environmental joint hidden state according to each first environmental hidden state includes: obtaining an attention coefficient of the target vehicle according to each first environmental hidden state; and weighting each first environmental hidden state according to the attention coefficient of the target vehicle to obtain an environmental joint hidden state.

[0088] Optionally, the environment joint hidden state is obtained by the following formula:

[0089]

[0090] In the above formula, exp represents the exponential function with the natural constant e as the base, || represents vector concatenation, and α m is the attention coefficient of the mth vehicle to the target vehicle, LeakyReLU is the activation function, and a T is the transposed matrix, W is the linear transformation matrix, H o is the first environment hidden state corresponding to the target vehicle, H m is the first hidden state of the environment corresponding to the mth vehicle, N represents the nodes around the target vehicle, and the nodes around the target vehicle are composed of the main vehicle and the interactive vehicle, H k represents the first environment hidden state corresponding to the kth node, N, m and k are all positive integers, m≤N;

[0091] H c =σ(∑ m∈N α m H m );

[0092] In the above formula, σ represents the nonlinear activation function, H c is the joint hidden state of the environment.

[0093] Optionally, each first environment hidden state is input into a graph attention mechanism to obtain a joint hidden state of the environment. The graph attention mechanism includes: obtaining an attention coefficient of the target vehicle according to each first environment hidden state; weighting each first environment hidden state according to the attention coefficient of the target vehicle to obtain a joint hidden state of the environment. By introducing the graph attention mechanism, the spatial correlation of different vehicle motions in the environment can be obtained, thereby obtaining the influence of surrounding vehicles on the expected motion trajectory of the target vehicle in a complex environment, thereby improving the accuracy of trajectory prediction of the target vehicle.

[0094] Optionally, obtaining a second environment hidden state according to the environment joint hidden state includes: using a fully connected network to concatenate the environment joint hidden state and a first environment hidden state corresponding to the target vehicle to obtain the second environment hidden state.

[0095] Optionally, the second environment hidden state is obtained by the following formula:

[0096] H s =FC(concat(H o ,H c );W g );

[0097] In the above formula, H s is the hidden state of the second environment, FC(·) represents the fully connected network, W g is the fully connected network parameter, H o is the first environment hidden state corresponding to the target vehicle, H c is the joint hidden state of the environment.

[0098] Optionally, obtaining a final hidden state according to the second environment hidden state and the main vehicle hidden state includes: using a fully connected network to splice the second environment hidden state and the main vehicle hidden state to obtain the final hidden state.

[0099] Optionally, the final hidden state is obtained by the following formula:

[0100] H=FC(concat(H s ,H f );W g )

[0101] In the above formula, H is the final hidden state, H s is the hidden state of the second environment, H f The main vehicle is hidden, W g are the parameters of the fully connected network.

[0102] Optionally, obtaining the future trajectory of the target vehicle according to the final hidden state includes: decoding the final hidden state using a logic gate neural network to obtain the future trajectory and motion state of the target vehicle.

[0103] Optionally, the final hidden state is input into the logic gate neural network for decoding to obtain the trajectory and motion state of the target vehicle at the next moment; the trajectory and motion state of the target at the next moment are re-input into the logic gate neural network for iteration to obtain the trajectory and motion state of the target vehicle in the predicted time domain.

[0104] In some embodiments, the trajectory and motion state of the target vehicle at the next moment are obtained by the following formula:

[0105]

[0106]

[0107] In the above formula, is the state parameter of the target vehicle at the next moment after time t, is the longitudinal coordinate of the target vehicle at the next moment after time t, is the lateral coordinate of the target vehicle at the next moment after time t, is the longitudinal velocity of the target vehicle at the next moment after time t, is the lateral velocity of the target vehicle at the next moment after time t, σ(r 1 ) is the trajectory and motion state of the target vehicle at the next moment, r 1 is the hidden state of the logic gate neural network, GRU(·) represents the logic gate neural network structure, W G is the neural network parameter of logic gates, is the state parameter of the target vehicle at time t, and H is the final hidden state.

[0108] In some embodiments, the trajectory and motion state of the target vehicle in the prediction time domain are obtained by the following formula:

[0109]

[0110]

[0111] In the above formula, t+n represents the nth moment after moment t. is the state parameter of the target vehicle at the nth moment after time t, T pre is the predicted sequence length, r n-1 is the hidden state of the logic gate neural network at the n-1th moment, r n is the hidden state of the logic gate neural network at the nth moment, is the state parameter of the target vehicle at the nth moment, GRU(·) represents the logic gate neural network structure, W G is the parameter of the logic gate neural network, and σ represents the activation function.

[0112] In some embodiments, a side lane vehicle trajectory prediction system in a car-following scenario includes a coding layer and a decoding layer. After obtaining the environment historical state parameters and the main vehicle future state parameters, the environment historical state parameters and the main vehicle future state parameters are input into the coding layer. In the coding layer, each environment historical state parameter is input into each bidirectional long short-term memory neural network to obtain each first environment hidden state; the main vehicle future state parameter is input into the corresponding bidirectional long short-term memory neural network to obtain the main vehicle hidden state; each first environment hidden state is input into the graph attention mechanism to obtain the environment joint hidden state; the environment joint hidden state and the first environment hidden state corresponding to the target vehicle are spliced ​​using a fully connected network to obtain a second environment hidden state; the second environment hidden state and the main vehicle hidden state are spliced ​​using a fully connected network to obtain a final hidden state. The final hidden state is input into the decoding layer for decoding.

[0113] In some embodiments, Figure 3 is a schematic diagram of the decoding layer, such as Figure 3 As shown, the final hidden state H is input into the logic gate neural network (Gated Recurrent Unit, GRU) to obtain the hidden state r of the logic gate neural network at the first moment in the future 1 , r 1 Input the activation function σ to obtain the state parameters of the target vehicle at the first moment after time t Will and r 1 Input GRU for iteration to obtain the hidden state r of the logic gate neural network at the second moment in the future 2 , r 2 Input the activation function σ to obtain the state parameters of the target vehicle at the second moment after time t Continue iterating in the above manner until and Input GRU for iteration to obtain the future Tth pre The hidden state of the logic gate neural network at time Will Input the activation function σ to obtain the Tth time after time t for the target vehicle pre State parameters at the moment Then the final prediction result of the state parameters of the target vehicle at each moment in the future is:

[0114]

[0115] In the above formula, is the future state parameter of the target vehicle, that is, the future trajectory and running state of the target vehicle. t+n represents the state parameter of the target vehicle at the nth moment after time t, that is, the state parameter of the target vehicle at the nth moment after time t. is the prediction result of the target vehicle at the first moment, is the prediction result of the target vehicle at the second moment, is the prediction result of the target vehicle at the nth moment.

[0116] Combination Figure 4 As shown, in some embodiments, the historical state parameter X of the main vehicle in the environment is obtained. e 、The historical state parameter X of the target vehicle o 、The historical state parameters of several interactive vehicles X1···X n and the future state parameters of the host vehicle The main vehicle historical state parameter X e 、The historical state parameter X of the target vehicle o 、The historical state parameters of several interactive vehicles X1···X n and the future state parameters of the host vehicle Input each bidirectional long short-term memory neural network (Long Short-Term Memory, Bi-LSTM) respectively to obtain the first environment hidden state H corresponding to the main vehicle e , the first environment hidden state H corresponding to the target vehicle o , the first environment hidden state H1 corresponding to the first interactive vehicle... the first environment hidden state H corresponding to the nth interactive vehicle n , Main vehicle hidden state H f The main vehicle, target vehicle and interactive vehicle in the environment are regarded as nodes, and the first hidden state H corresponding to the main vehicle is e , the first environment hidden state H corresponding to the target vehicle o The first environment hidden state H1···H corresponding to the interacting vehicle n Input graph attention mechanism to extract the joint hidden state H of the environment c . Combine the environment with the hidden state H c The first environment hidden state H corresponding to the target vehicle o Input the fully connected network for splicing to obtain the second environment hidden state H s . Hide the second environment state H s and the hidden state of the main vehicle H f Input the fully connected network for splicing to obtain the final hidden state H. Input the final hidden state H into the logic gate neural network (Gated Recurrent Unit, GRU) for decoding to obtain the future trajectory and motion state of the target vehicle, that is, the future state parameters of the target vehicle

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A method for predicting the trajectory of a vehicle in a lane following scenario, characterized in that: include: Obtaining environmental historical state parameters and host vehicle future state parameters; the environmental historical state parameters include host vehicle historical state parameters, target vehicle historical state parameters, and interactive vehicle historical state parameters; Acquire the future trajectory of the target vehicle according to the historical state parameters of the environment and the future state parameters of the host vehicle; Obtain the historical state parameters of the environment and the future state parameters of the main vehicle, including: The historical state parameters of the main vehicle are obtained by using the environment perception sensor and differential global positioning system of the main vehicle; the state parameters include longitudinal coordinates, lateral coordinates, longitudinal speed and lateral speed; Acquire the future state parameters of the main vehicle according to the historical state parameters of the main vehicle; The target vehicle's historical state parameters and the interaction vehicle's historical state parameters are respectively obtained by using the host vehicle's environmental perception sensor; Acquiring the future trajectory of the target vehicle according to the historical state parameters of the environment and the future state parameters of the host vehicle includes: Acquire each first environment hidden state respectively according to each of the environment historical state parameters; Acquire an environment joint hidden state according to each of the first environment hidden states; Acquire a second environment hidden state according to the environment joint hidden state; Acquire the hidden state of the main vehicle according to the future state parameter of the main vehicle; Acquire a final hidden state according to the second environment hidden state and the main vehicle hidden state; Obtaining a future trajectory of the target vehicle according to the final hidden state; Acquiring each first environment hidden state according to each of the environment historical state parameters respectively includes: Inputting each of the environmental historical state parameters into each bidirectional long short-term memory neural network to obtain each first environmental hidden state; Acquiring an environment joint hidden state according to each of the first environment hidden states includes: Acquire an attention coefficient of the target vehicle according to each first environment hidden state; Weighting each of the first environment hidden states according to the attention coefficient of the target vehicle to obtain an environment joint hidden state; The environment joint hidden state is obtained by the following formula: In the above formula, exp represents the exponential function with the natural constant e as the base, || represents vector concatenation, and α m is the attention coefficient of the mth vehicle to the target vehicle, LeakyReLU is the activation function, and a T is the transposed matrix, W is the linear transformation matrix, H o is the first environment hidden state corresponding to the target vehicle, H m is the first hidden state of the environment corresponding to the mth vehicle, N represents the nodes around the target vehicle, and the nodes around the target vehicle are composed of the main vehicle and the interactive vehicle, H k represents the first environment hidden state corresponding to the k-th node, N, m and k are all positive integers, m≤N; H c =σ(∑ m∈N a m H m ); In the above formula, σ represents the nonlinear activation function, H c is the joint hidden state of the environment.

2. The method according to claim 1, characterized in that Acquiring a second environment hidden state according to the environment joint hidden state includes: The environment joint hidden state and the first environment hidden state corresponding to the target vehicle are concatenated using a fully connected network to obtain a second environment hidden state.

3. The method according to claim 2, characterized in that The second environment hidden state is obtained by the following formula: H s =FC(concat(H o ,H c );W g ); In the above formula, H s is the hidden state of the second environment, FC(·) represents the fully connected network, W g is the fully connected network parameter, H o is the first environment hidden state corresponding to the target vehicle, H c is the joint hidden state of the environment.

4. The method according to claim 1, characterized in that: Acquiring a final hidden state according to the second environment hidden state and the main vehicle hidden state, including: The second environment hidden state and the main vehicle hidden state are concatenated using a fully connected network to obtain a final hidden state.

5. The method according to claim 1, characterized in that Obtaining the future trajectory of the target vehicle according to the final hidden state includes: The final hidden state is decoded using a logic gate neural network to obtain the future trajectory and motion state of the target vehicle.

Images