CN114872718A

CN114872718A - Vehicle trajectory prediction method, vehicle trajectory prediction device, computer equipment and storage medium

Info

Publication number: CN114872718A
Application number: CN202210374183.8A
Authority: CN
Inventors: 张博维; 黄晋; 于伟光; 江昆; 杨殿阁
Original assignee: Tsinghua University
Current assignee: Tsinghua University
Priority date: 2022-04-11
Filing date: 2022-04-11
Publication date: 2022-08-09
Anticipated expiration: 2042-04-11
Also published as: CN114872718B

Abstract

The application relates to a vehicle trajectory prediction method, a vehicle trajectory prediction device, a computer device and a storage medium. The method comprises the following steps: acquiring historical track information of a target vehicle and historical track information of surrounding vehicles; the historical track information is track information of the vehicle at each moment in the historical period; inputting the track information of the target vehicle at the current moment and the historical track information of all surrounding vehicles into a track prediction model to obtain an initial predicted track vector of the target vehicle; determining each influence characteristic vector between the target vehicle and each surrounding vehicle according to the spatial position information of the target vehicle at each moment, the spatial position information of each surrounding vehicle at each moment and the influence characteristic extraction strategy; and determining the predicted track of the target vehicle according to the initial predicted track vector of the target vehicle and each influence characteristic vector between the target vehicle and each surrounding vehicle. By adopting the method, the accuracy of the predicted target vehicle running track can be improved.

Description

Vehicle trajectory prediction method, vehicle trajectory prediction device, computer equipment and storage medium

Technical Field

The present application relates to the field of automatic driving technologies, and in particular, to a vehicle trajectory prediction method, apparatus, computer device, and storage medium.

Background

With the development of the automatic driving technology, the automatic driving vehicle predicts the driving track of the surrounding vehicle within a period of time in the future by using the perception information, which is helpful for improving the reliability of route planning and decision making of the automatic driving vehicle, reducing the risk of accidents and improving the safety of automatic driving of the vehicle.

The traditional vehicle track prediction method is to input information such as the position, the speed, the headway and the like of a target vehicle and surrounding vehicles into a deep neural network, and predict the future driving track of the surrounding vehicles through the neural network. However, this method does not take into account the mutual influence between the target vehicle and the surrounding vehicle (i.e., the situation in which the predicted travel locus of the target vehicle is influenced by the relative distance between the surrounding vehicle and the target during the history period and the change in the spatial position), resulting in a low accuracy of the predicted travel locus of the target vehicle.

Disclosure of Invention

In view of the above, it is necessary to provide a vehicle trajectory prediction method, apparatus, computer device, computer readable storage medium and computer program product for solving the above technical problems.

In a first aspect, the present application provides a vehicle trajectory prediction method. The method comprises the following steps:

acquiring historical track information of a target vehicle and historical track information of surrounding vehicles; the peripheral vehicles are vehicles around the target vehicle, the historical track information is track information of the vehicles at all moments in a historical period, and the track information comprises spatial position information of the vehicles;

inputting the track information of the target vehicle at the current moment and the historical track information of each surrounding vehicle into a track prediction model to obtain an initial predicted track vector of the target vehicle;

determining each influence feature vector between the target vehicle and each surrounding vehicle according to the spatial position information of the target vehicle at each moment, the spatial position information of each surrounding vehicle at each moment and an influence feature extraction strategy;

and determining the predicted track of the target vehicle according to the initial predicted track vector of the target vehicle and each influence characteristic vector between the target vehicle and each surrounding vehicle.

Optionally, the trajectory information includes spatial position information, and the trajectory prediction model includes a coding network and a convolutional network; inputting the track information of the target vehicle at the current moment and the historical track information of each surrounding vehicle into a track prediction model to obtain an initial predicted track vector of the target vehicle, wherein the method comprises the following steps:

inputting the track information of the target vehicle at the current moment and the historical track information of each surrounding vehicle into a coding network to obtain a target track vector of each surrounding vehicle; the target track vector is a historical track vector of the surrounding vehicles based on track information of the target vehicle at the current moment;

and determining an initial predicted track vector of the target vehicle according to the target track vector of each surrounding vehicle, the track information of the target vehicle at the current moment and the convolution network.

Optionally, the historical track vector is a track vector of the vehicle at each time in a historical period; determining an initial predicted trajectory vector of the target vehicle according to the historical trajectory vector of each of the surrounding vehicles, the trajectory information of the target vehicle at the current time, and the convolutional network, including:

determining a spatial trajectory vector of each surrounding vehicle according to the trajectory vector of each surrounding vehicle at the current moment and the spatial position information of each surrounding vehicle;

and determining an initial predicted track vector of the target vehicle according to the space track vector of each surrounding vehicle, the track information of the target vehicle at the current moment and the convolution network.

Optionally, the determining, according to the spatial location information of the target vehicle at each time, the spatial location information of each of the surrounding vehicles at each time, and the influence feature extraction policy, each influence feature vector between the target vehicle and each of the surrounding vehicles includes:

establishing a vehicle directed graph model at each moment according to the spatial position information of the target vehicle at each moment and the spatial position information of each surrounding vehicle at each moment;

determining each influence coefficient between the target vehicle and each surrounding vehicle at each moment according to the vehicle directed graph model and the influence coefficient algorithm at each moment;

and determining each influence characteristic vector between the target vehicle and each surrounding vehicle according to each influence coefficient between the target vehicle and each surrounding vehicle at each moment and the coding network.

Optionally, the determining, according to each influence coefficient between the target vehicle and each of the surrounding vehicles at each time and the coding network, each influence feature vector between the target vehicle and each of the surrounding vehicles includes:

aiming at each moment, establishing an influence coefficient matrix of the moment according to each influence coefficient of the moment;

carrying out compression dimensionality reduction on the influence coefficient matrix at the moment to obtain a low-dimensional influence coefficient matrix at the moment;

and inputting the low-dimensional influence coefficient matrix of each moment into a coding network to obtain each influence characteristic vector between the target vehicle and each surrounding vehicle.

Optionally, the determining the predicted trajectory of the target vehicle according to the initial predicted trajectory vector of the target vehicle and each influence feature vector between the target vehicle and each of the surrounding vehicles includes:

determining a target predicted trajectory vector of the target vehicle according to the initial predicted trajectory vector of the target vehicle and the influence characteristic vectors between the target vehicle and the surrounding vehicles;

and inputting the target predicted track vector of the target vehicle into a decoding network to obtain the predicted track of the target vehicle.

In a second aspect, the present application further provides a vehicle trajectory prediction device. The device comprises:

the acquisition module is used for acquiring historical track information of the target vehicle and historical track information of all surrounding vehicles; the peripheral vehicles are vehicles around the target vehicle, the historical track information is track information of the vehicles at all moments in a historical period, and the track information comprises spatial position information of the vehicles;

the prediction module is used for inputting the track information of the target vehicle at the current moment and the historical track information of each surrounding vehicle into a track prediction model to obtain an initial predicted track vector of the target vehicle;

the determining module is used for determining each influence characteristic vector between the target vehicle and each surrounding vehicle according to the spatial position information of the target vehicle at each moment, the spatial position information of each surrounding vehicle at each moment and an influence characteristic extracting strategy;

and the decoding module is used for determining the predicted track of the target vehicle according to the initial predicted track vector of the target vehicle, the influence characteristic vectors between the target vehicle and the surrounding vehicles and a decoding network.

Optionally, the trajectory information includes spatial position information, and the prediction module is specifically configured to:

Optionally, the historical track vector is a track vector of the vehicle at each time in a historical period; the prediction module is specifically configured to:

Optionally, the determining module is specifically configured to:

compressing and reducing the dimension of the influence coefficient matrix at the moment to obtain a low-dimensional influence coefficient matrix at the moment;

Optionally, the decoding module is specifically configured to:

In a third aspect, the present application provides a computer device. The computer device comprises a memory storing a computer program and a processor implementing the steps of the method of any of the first aspects when the processor executes the computer program.

In a fourth aspect, the present application provides a computer-readable storage medium. On which a computer program is stored which, when being executed by a processor, carries out the steps of the method of any one of the first aspects.

In a fifth aspect, the present application provides a computer program product. The computer program product comprises a computer program which, when executed by a processor, performs the steps of the method of any one of the first aspects.

The vehicle track prediction method, the vehicle track prediction device, the computer equipment and the storage medium are used for predicting the vehicle track by acquiring the historical track information of the target vehicle and the historical track information of each surrounding vehicle; the peripheral vehicles are vehicles around the target vehicle, the historical track information is track information of the vehicles at all moments in a historical period, and the track information comprises spatial position information of the vehicles; inputting the track information of the target vehicle at the current moment and the historical track information of each surrounding vehicle into a track prediction model to obtain an initial predicted track vector of the target vehicle; determining each influence feature vector between the target vehicle and each surrounding vehicle according to the spatial position information of the target vehicle at each moment, the spatial position information of each surrounding vehicle at each moment and an influence feature extraction strategy; and determining the predicted track of each peripheral vehicle according to the initial predicted track vector of the target vehicle and each influence characteristic vector between the target vehicle and each peripheral vehicle. Based on the influence feature extraction strategy, when the running track of the target vehicle is predicted, the predicted running track of the target vehicle is adjusted through the mutual influence condition between the target vehicle and each surrounding vehicle, so that the accuracy of the predicted running track of the target vehicle is improved.

Drawings

FIG. 1 is a flow diagram of a vehicle trajectory prediction method in one embodiment;

FIG. 1a is a schematic diagram of a measurement range of an on-board infrared range finder in one embodiment;

FIG. 2 is a flow diagram illustrating the steps for determining an initial predicted trajectory vector in one embodiment;

FIG. 3 is a flow chart illustrating the determination of the influence feature vector in one embodiment;

FIG. 3a is a diagram of a vehicle directed graph model in one embodiment;

FIG. 4 is a flow chart illustrating the determining step of the influence feature vector in another embodiment;

FIG. 5 is a schematic flow chart illustrating an example of vehicle trajectory prediction in one embodiment;

FIG. 6 is a block diagram showing the construction of a vehicle trajectory predicting device according to an embodiment;

FIG. 7 is a diagram illustrating an internal structure of a computer device according to an embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

The vehicle track prediction method provided by the embodiment of the application can be applied to a terminal, a server and a system comprising the terminal and the server, and is realized through interaction of the terminal and the server. The terminal may include, but is not limited to, various personal computers, notebook computers, tablet computers, and the like. The terminal is used for acquiring historical track information of the target vehicle and each surrounding vehicle, and adjusting the predicted running track of the target vehicle according to the mutual influence condition between the target vehicle and each surrounding vehicle when the running track of the target vehicle is predicted based on the influence feature extraction strategy, so that the accuracy of the predicted running track of the target vehicle is improved.

In one embodiment, as shown in fig. 1, a vehicle trajectory prediction method is provided, which is described by taking an example of the method applied to a terminal, and includes the following steps:

in step S101, the history track information of the target vehicle and the history track information of each surrounding vehicle are acquired.

The peripheral vehicles are vehicles around the target vehicle, the historical track information is track information of the vehicles at all times in the historical period, and the track information comprises spatial position information of the vehicles.

In the present embodiment, the terminal acquires the trajectory information of the target vehicle at each time within the history period and the trajectory information of the surrounding vehicles of the target vehicle (i.e., the surrounding vehicles) at each time within the history period, by the on-vehicle infrared distance meters provided around the predicted vehicle. The prediction vehicle executes the above steps and predicts a vehicle of the travel locus of the target vehicle, which may be, but is not limited to, an autonomous vehicle. The target vehicle is any one of the vehicles around the predicted vehicle. The number of the on-vehicle infrared distance meters may be, but is not limited to, one, and the trajectory information is data including spatial position information, velocity, and acceleration of the vehicle.

The measuring range of the vehicle-mounted infrared distance measuring instrument arranged around the vehicle is predicted to be a rectangle, the width of the rectangle is the distance which takes the lane where the target vehicle is located as the center, the widths of two sides of the rectangle are respectively the width of one lane, and the length of the rectangle is the farthest straight line distance which can be measured by the vehicle-mounted infrared distance measuring instrument. The farthest straight distance is not limited. For example, as shown in fig. 1a, the rectangle R is a grid of 3 × k, the width b of a single grid is a lane width, and the length a is l/k (k is a rectangular length) determined according to the distance measurement radius of the infrared distance meter, each grid can accommodate one vehicle, and the target vehicle is located in the central grid.

The spatial position of the target vehicle, and the spatial positions of the surrounding vehicles of the target vehicle are established in a two-dimensional scenet coordinate system with the lane line as a reference line. The vehicle speed includes a lateral speed of the vehicle, and a longitudinal speed of the vehicle, and the vehicle acceleration includes a lateral acceleration of the vehicle, and a longitudinal acceleration of the vehicle. The history period is a time period in a fixed time range before the current time, and the time period can be 3S, 4S or 5S.

In one embodiment, the expression for the trajectory information of the target vehicle over the historical period may be:

wherein

The position coordinates, the speed and the acceleration in the X direction and the Y direction of the vehicle at the moment of the ith frame in the time sequence. m is the number of frames (i.e., the number of times included in the history period).

And step S102, inputting the track information of the target vehicle at the current moment and the historical track information of all surrounding vehicles into a track prediction model to obtain an initial predicted track vector of the target vehicle.

In this embodiment, the trajectory prediction model includes an encoder and a trajectory prediction network, the terminal encodes the trajectory information of the target vehicle at the current time and the historical trajectory information of each peripheral vehicle through the encoder to obtain a historical trajectory vector of each peripheral vehicle relative to the target vehicle, which can be processed by the trajectory prediction network, and processes the historical trajectory vector of each peripheral vehicle relative to the target vehicle through the trajectory prediction network, so as to process the future travel trajectory of the target vehicle and obtain an initial predicted trajectory vector of the target vehicle.

The encoder may be, but is not limited to, a Long Short-term Memory network (LSTM) that extracts spatial position information, speed, and acceleration of the vehicle from the trajectory information and converts the information into a vector containing the driving direction of the vehicle.

The trajectory prediction network may be, but is not limited to, a convolutional neural network that obtains an initial predicted trajectory vector of the target vehicle at the current time by performing convolution and pooling operations on historical trajectory vectors of respective surrounding vehicles with respect to the target vehicle, where the initial predicted trajectory vector is a vector containing spatial position information of the vehicle and kinematic information of the vehicle (i.e., acceleration and speed of the vehicle).

Step S103, determining each influence characteristic vector between the target vehicle and each surrounding vehicle according to the space position information of the target vehicle at each moment, the space position information of each surrounding vehicle at each moment and the influence characteristic extraction strategy.

In this embodiment, the terminal establishes a vehicle directed graph model of spatial position information of the target vehicle and each surrounding vehicle at each time of the historical period by influencing the feature extraction strategy for each time. And the terminal processes the vehicle directed graph models at all moments in the historical time period and sorts the vehicle directed graph models according to the sequence of the time sequence to obtain the sorted and processed vehicle directed graph models. And the terminal inputs the sequenced vehicle directed graph models into the encoder, so that each influence characteristic vector between the target vehicle and each surrounding vehicle is obtained. The specific processing procedure will be described in detail later.

The establishment of the vehicle directed graph model is limited by the distance which can be detected by the vehicle-mounted infrared detector of the target vehicle, namely, the range of the vehicle directed graph model is the same as the range which can be detected by the vehicle-mounted infrared detector.

The method comprises the steps of establishing a vehicle directed graph model at each moment, processing the vehicle directed graph model at each moment, and inputting the processed vehicle directed graph models into an encoder to obtain each influence characteristic vector between a target vehicle and each surrounding vehicle.

The encoder may be, but is not limited to, an encoder in a Long Short-term Memory network (LSTM) for extracting an influence feature vector between the target vehicle and each surrounding vehicle in the history period, the influence feature vector being represented as a vector containing information of a vehicle traveling direction and a spatial position.

The influence feature vector includes position information between vehicles and a position change situation between vehicles in a unit time. The position information between the vehicles includes spatial position information between the vehicles and a straight-line distance between the vehicles. The position change situation between the vehicles in the unit time includes a space position change situation between the vehicles in the unit time and a straight-line distance change situation between the vehicles in the unit time. For example, the position change between the vehicle a and the target vehicle in one unit time is that the vehicle a travels from the left rear of the target vehicle to the left front of the target vehicle in one unit time; the distance between the A vehicle and the target vehicle changes from 3m to 1m in one unit time. The unit time is not limited.

And step S104, determining the predicted track of the target vehicle according to the initial predicted track vector of the target vehicle and each influence characteristic vector between the target vehicle and each surrounding vehicle.

In this embodiment, the terminal performs splicing according to the initial predicted trajectory vector of the target vehicle and the influence feature vectors between the target vehicle and each surrounding vehicle to obtain a spliced initial predicted trajectory vector. And the terminal inputs the spliced initial predicted track vector into a decoder to obtain the predicted track of the target vehicle. The predicted trajectory is represented as trajectory information including a vehicle traveling direction, a vehicle speed, and a vehicle acceleration for each prediction of the probability distribution.

For example, the position of the target at the current time is the right left lane of the predicted vehicle, and the distance is 5m from the predicted vehicle, and the predicted trajectory of the target vehicle is: firstly, a first predicted position of a target vehicle is located right ahead of a lane where the predicted vehicle is located, the straight-line distance from the predicted vehicle is 3m, the vehicle speed is 40km/h, and the vehicle acceleration is 1m/s (the probability is 50%); secondly, the second predicted position of the target vehicle is located in front of a left lane of the predicted vehicle, the linear distance from the predicted vehicle is 2m, the vehicle speed is 50km/h, and the vehicle acceleration is 1.2m/s (the probability is 30%); and thirdly, the third predicted position of the target vehicle is positioned behind the left lane of the predicted vehicle, the straight line distance from the predicted vehicle is 1m, the vehicle speed is 30km/h, and the vehicle acceleration is-1 m/s (the probability is 20%).

The decoder may be, but is not limited to, an encoder in a Long Short-term Memory (LSTM) network, and the decoder and the encoder belong to the same Long Short-term Memory network.

Based on the scheme, the predicted running track of the target vehicle is adjusted through the mutual influence condition between the target vehicle and each surrounding vehicle when the running track of the target vehicle is predicted based on the influence feature extraction strategy, so that the accuracy of the predicted running track of the target vehicle is improved.

Optionally, the trajectory information includes spatial position information, and the trajectory prediction model includes a coding network and a convolutional network; equivalently, as shown in fig. 2, inputting the trajectory information of the target vehicle at the current time and the historical trajectory information of each surrounding vehicle into the trajectory prediction model to obtain an initial predicted trajectory vector of the target vehicle, includes:

step S201, inputting the track information of the target vehicle at the current moment and the historical track information of each surrounding vehicle into a coding network to obtain the target track vector of each surrounding vehicle.

The target track vector is a historical track vector of surrounding vehicles based on track information of the target vehicle at the current moment.

In this embodiment, the trajectory prediction model includes an encoder (i.e., an encoding network) and a convolutional neural network (convolutional network). And the terminal converts the spatial position information of the surrounding vehicles at each moment in the historical period according to the spatial position information of the target vehicle at the current moment to obtain new spatial position information taking the spatial position information of the target vehicle at the current moment as a coordinate origin, and marks the spatial position information of the target vehicle at the current moment as the origin. And the terminal inputs the converted historical track information of the surrounding vehicles into the encoder to obtain target track vectors of the surrounding vehicles, and the steps are repeated until the target track vectors of all the surrounding vehicles are obtained.

For example, the current time of the target vehicle is t ₀ At the time, the number of surrounding vehicles is N, and past t of the surrounding vehicles is acquired _s，his Historical track P in time _s，his The formula of (1) is:

wherein

Including spatial position information, velocity and acceleration of the ith frame. m is _i The number of the frames is determined by the historical track duration and the acquisition frequency of the week vehicle. i is an element of [1, N ∈]A virtual number for the surrounding vehicle.

Terminal P _t，his Spatial position coordinates of the middle vehicle are converted into t ₀ A coordinate system taking the position of the target vehicle at the moment as the center of a circle, and track information of each moment after conversion

Embedding into 32 dimensions through a full connection layer and activating by using a LeakyRelu function to obtain a 32 x m dimension tensor. The terminal inputs the tensor into a coder of a single-layer LSTM network with 64 hidden units, and a 64-dimensional hidden layer vector h obtained by the last time step coding is taken _t，m (i.e., the target trajectory vector of the surrounding vehicle), the vector h _t，m Contains kinematic information of the surrounding vehicle (i.e., vehicle speed and vehicle acceleration). The terminal will h _t，m Reducing dimension to 32 dimensions by using a full connection layer, and activating by using a LeakyRelu function to obtain a tensor V _t . The above process can be represented by the following formula:

V _t ＝LeakyRelu(FCN _64→32 (h _t，m ))

wherein, c _t，m Is the output tensor of the mth time step encoder.

Step S202, determining an initial predicted track vector of the target vehicle according to the target track vector of each surrounding vehicle, the track information of the target vehicle at the current moment and the convolution network.

In this embodiment, the terminal establishes a multilayer vector according to the converted new spatial position information of each surrounding vehicle at the current time, the target trajectory vector of each surrounding vehicle, and the trajectory information of the target vehicle at the current time, and performs convolution operation on the multilayer vector twice through a convolutional neural network, and performs pooling operation once, so as to obtain an initial predicted trajectory vector of the target vehicle. The initial predicted trajectory vector includes spatial position information of the target vehicle predicted by the terminal, and kinematic information (i.e., speed, acceleration) of the predicted target vehicle.

Based on the scheme, the track information of the target vehicle at the current moment and the historical track information of all surrounding vehicles are processed and predicted through the encoder and the convolutional neural network, so that the initial predicted track vector of the target vehicle is obtained, and the accuracy of predicting the initial track vector of the target vehicle is improved.

Optionally, the historical track vector is a track vector of the vehicle at each moment in the historical period; determining an initial predicted track vector of a target vehicle according to a historical track vector of each surrounding vehicle, track information of the target vehicle at the current moment and a convolution network, wherein the method comprises the following steps: determining a space track vector of each surrounding vehicle according to the track vector of each surrounding vehicle at the current moment and the space position information of each surrounding vehicle; and determining an initial predicted track vector of the target vehicle according to the space track vector of each surrounding vehicle, the track information of the target vehicle at the current moment and the convolution network.

In this embodiment, the terminal establishes a rectangular grid with the spatial position information of the target vehicle at the current time as a center of a circle according to the converted new spatial position information of each surrounding vehicle at the current time, places the target trajectory vector of each surrounding vehicle at a corresponding position of each surrounding vehicle in the rectangular grid to obtain a multilayer vector (i.e., the spatial trajectory vector of each surrounding vehicle), and performs convolution operation on the multilayer vector twice through a convolutional neural network, and performs pooling operation once to obtain an initial predicted trajectory vector of the target vehicle.

In one embodiment, the rectangular grid is a rectangle R, and the terminal compares the target trajectory vector of each surrounding vehicle obtained in step S201

And the track information of the target vehicle at the current moment is placed in the grid, and a multilayer vector T containing the kinematic information of each surrounding vehicle and the spatial position information thereof and the track information of the target vehicle at the current moment is constructed ₀ The dimensions of the multilayer vector layer are 3 × k × 64 (the single-layer dimension is 3 × k, k is the rectangular length). For the hyper-parameters of each vector layer, the terminal activates using the LeakyRelu function for each vector layer as shown in Table 1.

TABLE 1 convolutional neural network hyper-parameters

For the activated vector layer, the terminal performs convolution and pooling operations through a convolution neural network, and the specific operation formula is as follows:

T ₁ ＝conv1(LeakyRelu(T ₀ ))#(5)

T ₂ ＝conv2(LeakyRelu(T ₁ ))#(6)

T ₃ ＝maxpooling(LeakyRelu(T ₂ ))#(7)

T＝flatten(T ₃ )#(8)

in the above formula, conv1, conv2 and MaxPooling represent two convolution operations and one maximum pooling operation, and flatten represents a dimension transformation operation.

Through the operation, the terminal obtains an initial predicted track vector T of the target vehicle, wherein the initial predicted track vector T comprises the future spatial position information of the target vehicle and the future kinematic information of the target vehicle

Based on the scheme, the spatial track vectors of the surrounding vehicles in the spatial structure and the track information of the target vehicle at the current moment are convolved and pooled, so that the accuracy of the predicted initial track information is improved.

Optionally, as shown in fig. 3, determining each influence feature vector between the target vehicle and each surrounding vehicle according to the spatial location information of the target vehicle at each time, the spatial location information of each surrounding vehicle at each time, and the influence feature extraction policy includes:

in step S301, a vehicle directed graph model at each time is created based on the spatial position information of the target vehicle at each time and the spatial position information of each surrounding vehicle at each time.

In this embodiment, for each time in the history period, the terminal establishes a vehicle directed graph model at the time based on the vehicle spatial position information of the target vehicle at the time and the spatial position information of each surrounding vehicle at the time. And the terminal repeats the steps until the vehicle directed graph model at all the time is built.

The formula of the vehicle directed graph model is as follows:

G _i (V，E)(i∈[1，m])

the model is represented as a grid, i is the virtual number of the vehicle, V is the spatial position information of the vehicle, and E is the distance information between the surrounding vehicle and the target vehicle. As shown in FIG. 3a, each sub-grid represents a node, e.g., G, at frame i, where N surrounding vehicles are included in the grid _i The medium target vehicle and each of the surrounding vehicles occupy (N +1) nodes corresponding to the grid, and nodes corresponding to grids not occupied by vehicles are empty.

Step S302, determining each influence coefficient between the target vehicle and each surrounding vehicle at each moment according to the vehicle directed graph model and the influence coefficient algorithm at each moment.

In this embodiment, the terminal determines, for each vehicle directed graph model at each time, an influence coefficient between the vehicles at the time based on spatial position information between the vehicles in the vehicle directed graph model at the time and an influence coefficient algorithm, and selects each influence coefficient between the target vehicle and each surrounding vehicle from the influence coefficients between the vehicles.

Specifically, when the spatial position information of the S car and the T car is V respectively _s And V _t Then, the formula of the influence coefficient algorithm is as follows:

wherein d is _st ＝d _ts Is a V _s And V _t The Euclidean distance between the centers of the two corresponding vehicles, lambda is a scaling constant and can be set according to the length of the rectangle R; mu.s _st And mu _ts Is the azimuth influence coefficient.

Mu since the driver is more concerned about the vehicle ahead in his field of view and less concerned about the vehicle behind during driving _st And mu _ts For distinguishing the influence of the orientation influence coefficient of the vehicle. If the S car is in front of the T car (i.e. y) _s ＞y _t ) Then, μ is specified _st 0.8, and μ _ts 0.2. Negative exponential form maps the distance into the (0, 1) weight interval, at d _st ＝d _ts Or 0. The formula of the influence coefficient algorithm is expressed as follows:

In step S303, influence feature vectors between the target vehicle and the surrounding vehicles are determined based on the influence coefficients between the target vehicle and the surrounding vehicles at each time and the coding network.

In this embodiment, for each time, the terminal processes the influence coefficient between the target vehicle and each surrounding vehicle at that time, and performs encoding by using an encoder to obtain each influence feature vector between the target vehicle and each surrounding vehicle. The specific processing procedure will be described in detail later.

Based on the scheme, each influence characteristic vector between the target vehicle and each surrounding vehicle is obtained through the influence characteristic extraction strategy, and therefore the accuracy of the predicted track is improved.

Optionally, as shown in fig. 4, determining each influence feature vector between the target vehicle and each surrounding vehicle according to each influence coefficient between the target vehicle and each surrounding vehicle at each time and the coding network includes:

in step S401, for each time, an influence coefficient matrix at the time is established based on each influence coefficient at the time.

In this embodiment, for each time of the historical period, the terminal establishes an influence coefficient map model of the time according to each influence coefficient of the time and a vehicle directed map model of the time, and performs matrixing processing on the influence time map model to obtain an influence coefficient matrix of the time. The coefficient matrix may be, but is not limited to, an adjacency matrix.

And S402, compressing and reducing the dimension of the influence coefficient matrix at the moment to obtain a low-dimensional influence coefficient matrix at the moment.

In this embodiment, the terminal performs compression and dimension reduction on the influence coefficient matrix at the time to obtain a low-dimensional influence coefficient matrix at the time. Compressing and reducing dimension of the influence coefficient matrix to a low-dimensional influence coefficient matrix A _i The dimension reduction formula of (2) is as follows:

∑ _i is a diagonal matrix with diagonal elements of A _i The singular value of (a). Taking three singular values with the maximum value as diagonal elements to form a matrix sigma _sub，i ，U _i And with

The corresponding singular vectors in the matrix U _sub，i And

the terminals use the full connection layer to transmit sigma respectively _sub，i ，U _sub，i And V _sub，i Mapping to 32 dimensions, and connecting in series to obtain vector

(Vector)

A low-dimensional influence coefficient matrix A representing the time _i 。

And step S403, inputting the low-dimensional influence coefficient matrix at each moment into a coding network to obtain each influence characteristic vector between the target vehicle and each surrounding vehicle.

In this embodiment, the terminal activates the vector at each time using the LeakyRelu function

(i.e., a low-dimensional influence coefficient matrix) and arranging the vectors at each time in time order

And (namely, a low-dimensional influence coefficient matrix) is input into an encoder for encoding to obtain each influence characteristic vector between the target vehicle and each surrounding vehicle.

Specifically, the vector of the terminal at a certain moment

(i.e. theLow-dimensional influence coefficient matrix) is input into the encoder for encoding, and the terminal takes the 64-dimensional hidden layer tensor h obtained by the last time step encoding _g，m And extracting an influence characteristic h of the target vehicle and each surrounding vehicle _g，m . The terminal will influence the characteristics h _g，m Reducing the dimension to 16 dimensions through a full connection layer to obtain each influence characteristic vector G between the target vehicle and each surrounding vehicle _v . The coding formula and the dimension reduction formula are as follows:

h _g，m ，c _g，m ＝LSTM(LeakyRelu(FCN _3k→16 (V _G )))

G _v ＝FCN _64→16 (h _g，m )

in the above formula, c _t，m The output tensor of the LSTM network for the mth time step.

Based on the scheme, the terminal determines each influence characteristic vector between the target vehicle and each surrounding vehicle through the influence coefficient between the target vehicle and each surrounding vehicle and the coding network, so that the influence characteristic vectors between the target vehicle and each surrounding vehicle are extracted, and preparation is made for optimizing the initial predicted track vector later.

Optionally, determining the predicted trajectory of the target vehicle according to the initial predicted trajectory vector of the target vehicle and each influence feature vector between the target vehicle and each surrounding vehicle, including: determining a target predicted track vector of the target vehicle according to the initial predicted track vector of the target vehicle and the influence characteristic vectors between the target vehicle and all surrounding vehicles; and inputting the target predicted track vector of the target vehicle into a decoding network to obtain the predicted track of the target vehicle.

In this embodiment, the terminal merges and concatenates the initial predicted trajectory vector of the target vehicle and the influence feature vectors between the target vehicle and each of the surrounding vehicles to obtain the target predicted trajectory vector of the target vehicle. And the terminal inputs the target track prediction vector of the target vehicle into a decoding network to obtain the predicted track of each surrounding vehicle.

Specifically, the terminal converts the V of the target vehicle _t ，T，G _v Concatenation of three vectors gives 1 × (16+32+8(k-3))And (3) maintaining the vector omega, inputting the omega vector of the target vehicle into an LSTM decoder for decoding, and obtaining the predicted track of each surrounding vehicle.

At each future moment, the track vector corresponding to the omega of the target vehicle and the track information at the previous moment is input into an LSTM decoder with 128 hidden units, and dimension reduction is carried out from a full connection layer to an output quantity dimension. The predicted trajectory represents information for each predicted trajectory in the form of probability distribution parameters. Assuming that the coordinates of each sampling point on the future track are subjected to binary Gaussian distribution, the probability density function expression is as follows:

in the above equation, five parameters of the binary Gaussian distribution are [ mu ] _X ，μ _Y ，σ _X ，σ _Y ρ is corresponding to the mean value of the abscissa (μ) of the surrounding vehicle at that time _X ) Mean value of ordinate (. mu.) _Y ) Abscissa variance (σ) _X ) Variance of ordinate (σ) _Y ) And a correlation coefficient (ρ).

Based on the scheme, the predicted trajectory of the rage vehicle is determined through the initial predicted trajectory vector of the target vehicle and the influence characteristic vectors between the target vehicle and the surrounding vehicles, so that the accuracy of the predicted trajectory of the target vehicle is improved.

The present application further provides a trajectory prediction example, as shown in fig. 5, a specific processing procedure includes the following steps:

in step S501, history track information of the target vehicle and history track information of each surrounding vehicle are acquired.

Step S502, inputting the track information of the target vehicle at the current moment and the historical track information of each surrounding vehicle into a coding network to obtain the target track vector of each surrounding vehicle.

In step S503, the spatial trajectory vector of each surrounding vehicle is determined based on the trajectory vector of each surrounding vehicle at the current time and the spatial position information of each surrounding vehicle.

Step S504, the initial predicted track vector of the target vehicle is determined according to the space track vector of each surrounding vehicle, the track information of the target vehicle at the current moment and the convolution network.

In step S505, a vehicle directed graph model at each time is created based on the spatial position information of the target vehicle at each time and the spatial position information of each surrounding vehicle at each time.

Step S506, determining each influence coefficient between the target vehicle and each surrounding vehicle at each moment according to the vehicle directed graph model and the influence coefficient algorithm at each moment.

Step S507, for each time, a time influence coefficient matrix is established according to each time influence coefficient.

And step S508, compressing and dimension reducing the moment influence coefficient matrix to obtain a moment low-dimensional influence coefficient matrix.

Step S509, the low-dimensional influence coefficient matrix at each time is input to the coding network, and each influence feature vector between the target vehicle and each surrounding vehicle is obtained.

Step S510 determines a target predicted trajectory vector of the target vehicle according to the initial predicted trajectory vector of the target vehicle and the influence feature vectors between the target vehicle and the surrounding vehicles.

Step S511, the target predicted trajectory vector of the target vehicle is input to the decoding network, and the predicted trajectory of the target vehicle is obtained.

It should be understood that, although the steps in the flowcharts related to the embodiments as described above are sequentially displayed as indicated by arrows, the steps are not necessarily performed sequentially as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least a part of the steps in the flowcharts related to the embodiments described above may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, and the execution order of the steps or stages is not necessarily sequential, but may be rotated or alternated with other steps or at least a part of the steps or stages in other steps.

Based on the same inventive concept, the embodiment of the application also provides a vehicle track prediction device for implementing the vehicle track prediction method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the method, so that specific limitations in one or more embodiments of the vehicle trajectory prediction device provided below can be referred to the limitations of the vehicle trajectory prediction method in the foregoing, and details are not repeated herein.

In one embodiment, as shown in fig. 6, there is provided a vehicle trajectory prediction apparatus including: an obtaining module 610, a predicting module 620, a determining module 630 and a decoding module 640, wherein:

an obtaining module 610, configured to obtain historical track information of a target vehicle and historical track information of each surrounding vehicle; the peripheral vehicles are vehicles around the target vehicle, the historical track information is track information of the vehicles at all times in the historical period, and the track information comprises spatial position information of the vehicles;

the prediction module 620 is used for inputting the track information of the target vehicle at the current moment and the historical track information of all surrounding vehicles into a track prediction model to obtain an initial predicted track vector of the target vehicle;

a determining module 630, configured to determine, according to the spatial location information of the target vehicle at each time, the spatial location information of each surrounding vehicle at each time, and the influence feature extraction policy, each influence feature vector between the target vehicle and each surrounding vehicle;

and the decoding module 640 is used for determining the predicted track of the target vehicle according to the initial predicted track vector of the target vehicle, each influence characteristic vector between the target vehicle and each surrounding vehicle and the decoding network.

Optionally, the trajectory information includes spatial location information, and the prediction module 620 is specifically configured to:

inputting the track information of the target vehicle at the current moment and the historical track information of all surrounding vehicles into a coding network to obtain target track vectors of all surrounding vehicles; the target track vector is a historical track vector of surrounding vehicles based on track information of the target vehicle at the current moment;

and determining an initial predicted track vector of the target vehicle according to the target track vector of each surrounding vehicle, the track information of the target vehicle at the current moment and the convolutional network.

Optionally, the historical track vector is a track vector of the vehicle at each moment in the historical period; the prediction module 620 is specifically configured to:

determining the space track vector of each surrounding vehicle according to the track vector of each surrounding vehicle at the current moment and the space position information of each surrounding vehicle;

Optionally, the determining module 630 is specifically configured to:

compressing and dimensionality reducing the influence coefficient matrix at the moment to obtain a low-dimensional influence coefficient matrix at the moment;

and inputting the low-dimensional influence coefficient matrix at each moment into the coding network to obtain each influence characteristic vector between the target vehicle and each surrounding vehicle.

Optionally, the decoding module 640 is specifically configured to:

determining a target predicted track vector of the target vehicle according to the initial predicted track vector of the target vehicle and the influence characteristic vectors between the target vehicle and all surrounding vehicles;

The respective modules in the vehicle trajectory prediction apparatus described above may be implemented in whole or in part by software, hardware, and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as shown in fig. 7. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used for carrying out wired or wireless communication with an external terminal, and the wireless communication can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. The computer program is executed by a processor to implement a vehicle trajectory prediction method. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on a shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.

Those skilled in the art will appreciate that the architecture shown in fig. 7 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.

In an embodiment, a computer device is provided, comprising a memory and a processor, the memory having stored therein a computer program, the processor implementing the steps of the above method embodiments when executing the computer program.

In an embodiment, a computer-readable storage medium is provided, on which a computer program is stored which, when being executed by a processor, carries out the steps of the above-mentioned method embodiments.

In an embodiment, a computer program product is provided, comprising a computer program which, when being executed by a processor, carries out the steps of the above-mentioned method embodiments.

It should be noted that, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, presented data, etc.) referred to in the present application are information and data authorized by the user or sufficiently authorized by each party.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, database, or other medium used in the embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high-density embedded nonvolatile Memory, resistive Random Access Memory (ReRAM), Magnetic Random Access Memory (MRAM), Ferroelectric Random Access Memory (FRAM), Phase Change Memory (PCM), graphene Memory, and the like. Volatile Memory can include Random Access Memory (RAM), external cache Memory, and the like. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM), among others. The databases referred to in various embodiments provided herein may include at least one of relational and non-relational databases. The non-relational database may include, but is not limited to, a block chain based distributed database, and the like. The processors referred to in the embodiments provided herein may be general purpose processors, central processing units, graphics processors, digital signal processors, programmable logic devices, quantum computing based data processing logic devices, etc., without limitation.

The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A vehicle trajectory prediction method, characterized in that the method comprises:

2. The method of claim 1, wherein the trajectory information comprises spatial location information, and wherein the trajectory prediction model comprises a coding network and a convolutional network; inputting the track information of the target vehicle at the current moment and the historical track information of each surrounding vehicle into a track prediction model to obtain an initial predicted track vector of the target vehicle, wherein the method comprises the following steps:

3. The method of claim 2, wherein the historical trajectory vector is a trajectory vector of the vehicle at each time in a historical period; determining an initial predicted trajectory vector of the target vehicle according to the historical trajectory vector of each of the surrounding vehicles, the trajectory information of the target vehicle at the current time, and the convolutional network, including:

4. The method of claim 1, wherein determining respective impact feature vectors between the target vehicle and the respective surrounding vehicles based on the spatial location information of the target vehicle at each time instant, the spatial location information of the respective surrounding vehicles at each time instant, and an impact feature extraction strategy comprises:

5. The method of claim 4, wherein determining the respective influence eigenvectors between the target vehicle and the respective surrounding vehicles based on the respective influence coefficients between the target vehicle and the respective surrounding vehicles at each time instant and the coding network comprises:

6. The method of claim 1, wherein determining the predicted trajectory of the target vehicle based on the initial predicted trajectory vector of the target vehicle and the respective influencing feature vectors between the target vehicle and the respective surrounding vehicles comprises:

7. A vehicle trajectory prediction apparatus, characterized by comprising:

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method of any of claims 1 to 6.

9. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that the computer program realizes the steps of the method of any one of claims 1 to 6 when executed by a processor.